Discussion 6

Decay happens: the role of active forgetting in memory Oliver Hardt1, Karim Nader1, and Lynn Nadel2

1 McGill University, Department of Psychology, Montré al, Qué bec, H3A 1B1, Canada 2 Department of Psychology, The University of Arizona, Tucson, AZ, 85701, USA

Opinion

Glossary

Active forgetting: the idea that, instead of passively disintegrating, memories

are actively removed, on the basis of, for example, relevance or recency.

AMPA receptor: an ionotropic glutamate receptor responsible for most

excitatory fast neurotransmission in the central nervous system.

Catastrophic interference: neural networks store memories as patterns of

activity, such that representations consist of a set of nodes and the weights of

their connections. Because neural networks have a finite number of nodes and

connections, networks eventually reach saturation and the addition of another

activity pattern will disrupt existing memories, leading to catastrophic

interference.

Consolidation: cellular (or synaptic) consolidation refers to the processes that

stabilize the learning-induced changes in synaptic morphology that represent

the biological substrate of memory. Disrupting these processes before

completion causes partial or full memory loss. Systems consolidation refers

to a reorganization process of the brain systems that support memory,

specifically, the hypothesis that some memories that initially require

hippocampal involvement no longer do so after some time.

Decay: forgetting due to a gradual loss of the substrate of memory. In his law

of disuse, Thorndike posited that, unless regularly used, all memory decays,

akin to a muscle that will atrophy if it is not exercised [11]. It has generally been

assumed that decay is a passive process.

Episodic memory: memory for what, when, and where. It is a matter of debate

whether this form of event memory is uniquely human or not.

Explicit memory: unlike implicit memory, explicit memory is consciously and

intentionally retrieved. Explicit memory is either episodic (event memory) or

semantic (factual knowledge).

Forgetting: forgetting refers to the absence of expression of previously

properly acquired memory in a situation that normally would cause such

expression. This can reflect actual memory loss or a failure to retrieve existing

memory.

Interference: according to interference accounts of forgetting, mental activity

can impact memory by affecting actual memory content or its retrieval.

Acquiring a new memory, for example, can retroactively impair existing

memory, or existing memory can proactively impair memory acquisition.

Reconsolidation: use (retrieval or reactivation) can induce a transient state of

heightened plasticity in long-term memories that resembles the unstable state

of new, not yet consolidated memories. In this state, reactivated memories are

malleable and can be modified and modulated.

Rapid eye movement (REM) sleep: a sleep state characterized by saccadic eye

movements typically occurring in rapid bursts, low to absent muscle tone, and

EEG activity consisting mainly of theta and beta waves.

Slow-wave sleep (SWS): also called deep sleep, a sleep state characterized by

little to no rapid eye movement and EEG activity consisting mainly of slow,

large delta waves.

Long-term potentiation: a long-lasting increase in synaptic potentiation (or

synaptic strength) that can be induced with high-frequency (tetanic) stimula-

tion pulses or correlated firing of the post-synaptic and pre-synaptic neuron.

Although the biological bases of forgetting remain ob- scure, the consensus among cognitive psychologists emphasizes interference processes, rejecting decay in accounting for memory loss. In contrast to this view, recent advances in understanding the neurobiology of long-term memory maintenance lead us to propose that a brain-wide well-regulated decay process, occurring mostly during sleep, systematically removes selected memories. Down-regulation of this decay process can increase the life expectancy of a memory and may even- tually prevent its loss. Memory interference usually occurs during certain active processing phases, such as encoding and retrieval, and will be stronger in brain areas with minimal sensory integration and less pattern separation. In areas with efficient pattern separation, such as the hippocampus, interference-driven forgetting will be minimal, and, consequently, decay will cause most forgetting.

Current thinking on forgetting Forgetting of established long-term memory (see Glossary) may indicate that memory is either physically unavailable (that is, memory is lost) or that it is (temporarily) inacces- sible. With some exceptions, theories proposed within the domains of experimental and cognitive psychology often emphasize one type of forgetting over the other [1]. Two explanations for actual, non-pathological memory loss have been proposed, one involving decay of aspects of the memory trace, the other involving interference with it.

Current consensus favors the latter of these two explana- tions for actual memory loss (see Supplementary Material for an abbreviated history of decay theory). It is supposed that interference processes are responsible for much of everyday forgetting and the decay hypothesis has been generally rejected as an explanation for forgetting of long- term memories [1,2]. Interference manifests in two principal ways. First, shortly after initial learning, task-related or task-unrelated mental activity can impair memory, proba- bly by disrupting cellular consolidation processes [3,4]. Sec- ond, the expression of established, fully consolidated long- term memory can suffer from interference at the retrieval stage [5]. For example, during retrieval, competing memo- ries may interfere with the recall process. Although it was thought that this type of reproductive or output interference mainly determined whether or not a memory was retrieved [6], recent research on post-retrieval memory plasticity

Corresponding author: Hardt, O. ([email protected]).

1364-6613/$ – see front matter � 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.101

suggests that it could also affect the content of memory [7]. Because retrieval of consolidated memories induces plasticity in the relevant traces, subsequent exposure to new material can then affect the restabilization, or recon- solidation, of the reactivated memory, akin to what can happen after initial learning [8]. This can lead to the inci- dental incorporation of new material into the reactivated

Long-term depression: a long-lasting decrease in synaptic potentiation. In the

hippocampus, it can be induced with low-frequency stimulation.

6/j.tics.2013.01.001 Trends in Cognitive Sciences, March 2013, Vol. 17, No. 3 111

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

memory [9] or can in some circumstances decrease memory retention [10].

Able to explain many experimental results, interference theories have pushed aside alternative accounts of for- getting. It was once widely assumed that, unless periodi- cally recalled, long-term memory may ‘simply’ vanish and fade away over time due to some unspecified biological process [11]. It has long been known that the converse is true: regular use supports long-term memory mainte- nance. The recently well-documented beneficial effects of testing on retention [12] show that the act of recall pro- motes long-term memory preservation. It should be noted that frequent recall can also distort and impair memories, with the timing of recall after learning determining wheth- er memory distortions or improvements will occur [13]. Memory in animals also benefits from repeated use [14]. When tested two days after learning to fear a certain spatial context, rats will express fear only towards the training context, but not towards other contexts. Three weeks after training, however, rats fear familiar and novel contexts alike. This change in memory for what place to fear can be prevented by reactivating the fear memory, that is, by re-exposing rats briefly to the training context several times during the three weeks between training and memory test. Rats reminded in this way will fear the spatial context in which training was carried out more so than they fear other contexts, whereas rats that have not been regularly re-exposed to the training context will fear the trained as well as other contexts equally [14]. However, these demonstrations of the effects of use provide only indirect support for the original notion of forgetting by decay and can be interpreted as supporting interference theory. It has not been easy to provide direct evidence of memory loss through disuse.

Notwithstanding the success of interference-based the- ories to describe the factors that promote forgetting, the truth is that we do not know why or how the brain actually forgets [15]. Our goal in this article is to discuss this age-old debate in the context of recent findings in the study of memory at both the cellular and systems levels, and to put forward a neurobiologically-based framework for memory and forgetting. Recent advances in the study of the cellular/ molecular underpinnings of long-term memory persis- tence, to be discussed in detail below, suggest memory decay as a major forgetting process. They allow us to assign organized memory removal a central role in the everyday forgetting of consolidated memories and in memory orga- nization.

It is generally assumed that forgetting is more a vice (i.e., dysfunction) than a virtue (i.e., constitutive process); however, the idea that forgetting might be beneficial for memory has been frequently expressed [3,4,16–19] and Jorge Luis Borges illustrated its essential role for the human experience in his short story about Funes [20]. As Funes could not forget anything, he could not live a normal life because a sea of unimportant details swamped every moment of awareness. We agree that, without con- stitutive forgetting, efficient memory would not be possible in the first place.

In our view, decay-driven forgetting is a direct conse- quence of a memory system that engages in promiscuous

112

encoding. The benefit of such promiscuity is access to a lot of information, so that ‘choices’ about what to keep and what to delete can be made off-line, mostly during certain sleep phases. The cost is the need for a dedicated forgetting mechanism that removes unwanted information. We pro- pose decay as an active, well-regulated process, in contrast to the standard notion of decay as a passive process akin to radioactive decay. In our view, a well-regulated, dedicated process that systematically removes memories not only is more efficient, but can also be better controlled (up- or down-regulated), depending on specific demands and metaplastic constraints, which allows for greater flexibility and adaptability of the memory system.

The circuit architecture of a given brain system, in particular the nature of its pattern separation capacities (i.e., the degree to which neural representations overlap, with orthogonal patterns being maximally separated) will determine whether interference or decay presents as the predominant forgetting mechanism. In systems with effi- cient pattern separation, such as the hippocampus, inter- ference will be low or even absent. In systems with little pattern separation, encoding of new traces will necessarily cause interference. We propose that during certain sleep phases, such as slow-wave sleep or rapid eye movement (REM) sleep, when interference by new learning is not a factor, decay happens in all brain systems. In brain areas that exhibit low interference at all times, such as the hippocampus, decay will be the primary mechanism to prevent extensive interference, that is, a state of system failure induced by pattern overload.

Forgetting in multiple memory systems: hippocampus and neocortex In agreement with most current views (for a review, see [21]), we suppose that what are generally referred to as explicit memories initially consist of two components: con- tent representations largely dependent on neocortical net- works and a spatial-contextual representation dependent on the hippocampus that indirectly links, and serves to index, the dispersed neocortical representations. The list- or story-learning tasks usually employed to study for- getting (predominantly interference) in humans are exam- ples of such explicit, episodic memories.

At the heart of our approach is the notion that the circuit architecture of the hippocampus diminishes interference between hippocampal memory traces [22]. Together with CA3, the dentate gyrus allows for orthogonalization of representations, or activity patterns, even in light of very similar inputs as, for example, in the case of highly similar spatial environments [23]. It seems that, in the dentate gyrus specifically, young adult-born granule cells promote pattern separation [24]. Taken together, these mecha- nisms reduce the overlap of representations in CA1 and CA3 [25]. This being the case, interference cannot account for most non-pathological forms of forgetting of hippocam- pal memory (Figure. 1a). Instead, we propose that for- getting of the hippocampal component of explicit memory takes the form of loss, or reversal, of learning- induced changes in synaptic potentiation [26], the biologi- cal substrate of the decay process envisioned by Thorndike in 1913 [11].

(a) Hippocampus orthogonal coding prevents interference in neocortex (b) Interference in amnesic pa�ents

Hi pp

oc am

pu s

N eo

co rt

ex

N o

Hi pp

oc am

pu s

N eo

co rt

ex

Hi pp

oc am

pu s

N eo

co rt

ex

(c) Decay-like forge�ng leads to context-free neocor�cal memory traces over �me

TRENDS in Cognitive Sciences

Figure 1. Systems effects of decay-like forgetting in hippocampus. We assume that most memories have hippocampal and extra-hippocampal components. (a) The effects

of the successive encoding of four memory representations. In the hippocampus, because of efficient pattern separation, even similar inputs will result in non-overlapping

representations. This is not the case for most neocortical networks, where the risk of interference by new learning is therefore high. The hippocampal representations,

linking to extra-hippocampal memory components, thus allow for reinstantiation of the original pattern despite overlapping extra-hippocampal representations. (b) The

same input leads to inseparable patterns in neocortical areas when the hippocampal component is missing, as in amnesic patients. (c) Once memories are encoded, they

are subject to constitutive decay-like forgetting. We show here an example of memory traces that survive this type of forgetting. After encoding, cellular consolidation

processes stabilize neocortical memories (this can take from minutes to some hours). During sleep, hippocampal replay reactivates these new memories, which can

strengthen extra-hippocampal traces. At the same time, decay-like processes begin to remove hippocampal memories (either during the same or a different sleep phase).

Once decay processes remove the hippocampal component, only neocortical components remain, provided that they had been sufficiently strengthened. These memories

are thus no longer bound to a spatial-contextual representation hosted by the hippocampus.

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

We assume that explicit memories always include both a hippocampal and neocortical component, even when memo- ry tasks can in principle be performed without the involve- ment of the hippocampus. In plastic memory phases, typically termed cellular consolidation and reconsolidation, this hippocampal trace can be modulated, that is, strength- ened or weakened, thus increasing or decreasing the speed of its decay. In many, perhaps most, cases of everyday memory, the hippocampal spatial-contextual trace decays rather quickly, that is, within days and weeks, whereas the dispersed neocortical traces can persist, despite the loss of the hippocampal representation of the context initially linking these contents and their neocortical traces.

The regions that represent the varied contents of an everyday memory are dispersed across the neocortex and are only sparsely linked, if at all. This fact creates a problem for the associative learning of arbitrary multimod- al associations, which form the basis of episodic or episodic- like memories. This problem led to the suggestion that the hippocampus serves as a hub that links dispersed neocor- tical representations, allowing for arbitrary associations

and, hence, for the formation of episodic memories [27,28]. Given the nature of neocortical coding, the laying down of new neocortical representations can interfere with existing representations, which may result in what has been re- ferred to as catastrophic interference [29,30]. Absent the powerful pattern separation mechanism instantiated in hippocampal circuits, neocortical circuitry suffers from a relative inability to separate one memory from another – which can manifest as memory loss on the behavioral level.

By contrast, the pattern separation properties of the dentate gyrus [24,25] greatly minimize the probability of interference between memories in the hippocampus, even closely related ones [23]. Because the sensory/perceptual content of memory is represented in neocortical structures from the outset, namely, during encoding [31], one way to minimize confusion between similar memories in neocor- tex is by linking the contents of a specific episodic memory to its unique spatio-temporal context. These contextual memory representations serve as indices that point direct- ly and indirectly to discrete synaptic subpopulations in neocortical areas. Thus, although the neocortex likely

113

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

represents memories using overlapping neuronal popula- tions, discrete subsets that represent specific memory contents can be activated by virtue of the hippocampal contextual index, that is, the hippocampal projections to these neurons (Figure 1a). The hippocampal contribution is, therefore, critical for the protection of recently acquired memories, which are presumed to be relatively weak and especially prone to disruption by interference caused by new learning [32]. It is important to bear in mind, of course, that there are multiple synapses separating the hippocam- pal contextual representation from the highly specific sen- sory-perceptual details that define a specific episode, and that the indexing process must work its way through many levels.

We assume that the more ‘perceptual’ (and less ‘concep- tual’) the representations a brain area supports, the stron- ger the retroactive interference caused by the encoding of new memory traces. That is, in early sensory processing areas, interference will be more pronounced than in areas further along the visual processing stream. For example, recently formed memories will suffer extensive interfer- ence from new visual encoding in V1, whereas in perirhinal cortex interference levels will be moderate to low, and in hippocampus interference will be low or generally absent. As sensory signals become more integrated, the risk of interference from new input diminishes because the recruited neuronal subpopulations overlap less and less. This means that perceptual details should be more readily lost and that, as a consequence, forgetting processes are intimately involved in the development of ‘schemas’, or ‘concepts’, or ‘abstract’ knowledge.

Our proposal resembles in some respects the hierarchi- cal-representational perspective on memory organization [33]. Briefly, this view assumes that visual representations are distributed along the visual processing stream, in that early (in humans, posterior) regions, such as V1, represent concrete object features, whereas later (in humans, anterior) regions, such as perirhinal cortex, represent the conjunc- tions of these. Feature conjunction representations reduce interference probability because they bind together a rela- tively unique set of distributed representations that togeth- er constitute the representation of an object. Rats with perirhinal lesions often present with deficits in object rec- ognition, which are exacerbated by visual experience during the retention interval. Similar to findings with amnesic patients [34], such deficits can be overcome if visual stimu- lation is reduced in the time between learning and memory testing [35].

Over time, neocortical representations may, with the help of the hippocampus, become sparser and more efficient, the network contents becoming better integrated into exist- ing memory ensembles by adjusting and optimizing synap- tic potentiations, probably by virtue of offline hippocampal reactivations (during certain sleep phases or periods of quiet wakefulness). In a sense, this hippocampus-mediated pat- tern optimization is similar to the idea that the hippocam- pus functions as a ‘teacher’ that permits the neocortex to gradually integrate new memories into existing representa- tions, but the proposed mechanisms differ significantly [29]. Once these new representations have been integrated into existing network representations, the hippocampus is no

114

longer essential to retrieval of some aspects of a memory. Although expression of the memory would still benefit from a hippocampal contribution, in that this would, for example, allow situation-specific retrieval, expression in a generic sense might be possible without it (hence the limited effect of hippocampal lesions on expression of gist-like older mem- ories, which occurs in a context-generalized fashion).

Like others, we propose that during learning the hip- pocampus indirectly provides a kind of pattern separa- tion for potentially overlapping neocortical memory representations (Figure 1a), thereby reducing the proba- bility that interference will impair these not-yet-consoli- dated new memory patterns [36]. This account finds some support in the finding that less forgetting of existing memories is observed when during learning of new, similar material, these old memories are reactivated, as indicated by hippocampal activity patterns associated with these memories [37]. Indirect support for this as- sumption comes from studies showing that interference can be reduced in rats when the first and the second list of items are encoded in different spatial contexts [38]. Strong support, however, comes from recent findings on the role of adult neurogenesis in rodents. The dentate gyrus supports continuous neurogenesis in the adult brain [39]. Suppression of neurogenesis in the dentate gyrus impairs spatial learning [40], and the removal of newly generated dentate gyrus neurons disrupts estab- lished hippocampus-dependent memories, but has no impact on memories that do not require the hippocampus [41]. Increasing evidence links dentate gyrus neurogen- esis to pattern separation [22,42]. In accord with the notion that pattern separation mediated by the dentate gyrus reduces interference, suppression of neurogenesis promotes interference in a hippocampus-dependent ol- factory discrimination task [43].

Our proposal accounts for the oft-reported observation that amnesic patients, who suffer from compromised me- dial-temporal and hippocampal function, present with ex- tensive interference [44]. Instead of viewing this increased interference as the cause of amnesia, we view it as an indirect consequence of the inability to separate overlap- ping neocortical ensembles (Figure 1a, b). Research in recent years has convincingly demonstrated that memory retention in amnesic patients can be dramatically en- hanced when periods of rest and reduced sensory stimula- tion follow memory encoding; when similar material is learned or when other activity follows memory encoding, amnesic patients forget (and much more so than healthy control subjects). Amnesic patients forget word lists or prose text within minutes after learning, but if learning is followed by up to one hour of inactivity or rest, forgetting is greatly reduced [34,45], and memory acquired in this manner can then even persist for at least a week [4]. These findings strongly suggest that in brains with damaged or compromised hippocampal function interference from new learning can cause extensive forgetting during wake, or active states, possibly by disrupting memory consolidation processes. Importantly, this type of interference mainly affects memories in plastic states, such as those after encoding (consolidation) and after retrieval (reconsolida- tion). If these memories are allowed to consolidate without

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

interference, they can last as long as they would in healthy individuals.

A role for decay in everyday forgetting The findings from these studies with amnesic patients also show that new learning of similar material or interpolated activity does not cause complete amnesia in healthy control subjects, and, depending on the material, forgetting due to interference may even be minimal. For example, in one study [34], retention of a word list dropped from 41% immediately after learning to 19% when new learning was followed by ten minutes of exacting cognitive tasks. When memory for a story was tested, however, one hour of potentially interfering activity only led to a drop from 62% to 50%, whereas retention after one hour without interfer- ing activity was 57%. Interference thus does not always cause meaningful loss for recently acquired memories and it may not be the main factor that causes forgetting in fully consolidated memories [34].

In everyday life, mental activity follows almost all learning, including encoding of similar memories, yet many memories are retained at the end of the day and some even survive for a lifetime. Some memories will be compromised during the day as a consequence of interfer- ence caused by new memory encoding or, more generally speaking, stimulus processing, which somehow impairs ongoing cellular consolidation. It remains to be deter- mined, however, how the consolidated remains of the day may be lost, especially for memories that stay dormant or are rarely used, such that they do not enter retrieval- induced states of plasticity, during which they would be vulnerable again to interference processes.

We propose that for many consolidated memories decay- like processes, rather than interference, lead to the active removal of their neurobiological substrate. What is lost in the hippocampus likely includes the spatial-contextual com- ponent (Figure 1c), which itself seems to serve as the memo- ry trace that permits the retrieval of memory contents stored elsewhere [46]. Decay-like forgetting in the hippocampus may remove the cues necessary to retrieve extra-hippocam- pal content. Because this component of a memory also protects the extra-hippocampal content component from interference (Figure 1a), its removal will render extra-hip- pocampal memory representations more vulnerable to dis- ruption by new learning. In this way, decay of traces in the hippocampus can indirectly promote interference in neocor- tical sites. As studies with amnesic patients have demon- strated, interference by mental activity exerts its strongest influence on subsequent retention only if it occurs shortly after encoding [4]. Thus, once a memory is consolidated, this type of interference may not be effective in permanently removing it should it turn out to be irrelevant. The type of memory decay described in our proposal provides a mecha- nism that can remove obsolete consolidated memory traces in many brain areas; in the hippocampus it seems to be the main forgetting mechanism.

We suppose that such decay is an integral feature of memory systems, providing a solution to an adaptive problem that the brain must solve. Often organisms cannot immediately determine the lasting importance of an arbi- trary conjunction of events. Prior experience, and thus

available knowledge, may guide attention, but frequently an organism cannot know a priori whether a long-lasting memory should be formed, and if so, which aspects of an event should be encoded and which aspects ignored. The ability to quickly acquire as much information as possible is highly adaptive, increasing the probability of having captured knowledge that might prove important later. Since significance often becomes evident only after the fact, preserving as much detailed information as possible is important. Having stored this information, some form of post-hoc significance/relevance signaling, such as stress or other emotional reactions, can then lead to the selective strengthening of some of the recently acquired memoires [47]. A good example of the dynamics displayed by such a system is memory for the preceding day’s lunch. Most likely, this memory will still be available the following day. However, a week later, or even just a few days later, it will have faded out. Nonetheless, should malaise set in few hours after dinner, memory for what had been eaten might be available longer, and, in the case of severe aversive reactions, this memory may persist for years or even a lifetime, and often in great episodic detail.

Increases in the levels of certain stimulants, such as epinephrine and glucocorticoids, accompany these emo- tional reactions, which can enhance memory retention, and it has been shown that these substances modulate cellular consolidation processes in several brain regions, such as the hippocampus and amygdala [48]. Affecting expression of AMPA receptors and thereby synaptic poten- tiation, this modulation can regulate synaptic strength [49]. These post-acquisition modulatory signals can also regulate metaplasticity, affecting the extent to which syn- apses will undergo plastic alterations, such as increases or reductions of synaptic strength, in the future [50]. Similar regulatory processes can also follow memory use, as has been demonstrated in several studies on the reconsolida- tion effect [8]. This kind of promiscuous memory formation, left unchecked, would saturate the system fairly quickly, compromising memory function. Automated forgetting in the form of potentiation reduction would be a simple yet efficient method to remove mnemonic waste in a system in which interference cannot effectively and systematically eliminate memories deemed less relevant. Those memories that were strengthened after formation will better with- stand the forces of forgetting (i.e., decay, as well as inter- ference), and thus, eventually, only those memories will remain. In addition, metaplastic modulations can render memories less likely to lose potentiation by promoting processes that increase potentiation relatively more than processes that decrease it [51].

It seems that the best time for this form of well-orga- nized memory removal of consolidated memories is during sleep, when the brain is not engaged in the encoding of new memories. The notion that systematic forgetting is an important function of sleep is not new. It has been proposed that reverse learning occurs during REM sleep in order to remove unnecessary memories acquired during the day [17]. According to the homeostatic synaptic scaling account of sleep [18], synaptic potentiation (stemming from day- time learning) is down-regulated brain-wide during slow- wave sleep. This rescaling process, which is supposed to

115

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

reduce overall energy requirements, preserves relative syn- aptic weight differences and may eventually lead to for- getting because downscaling may effectively silence, or even remove, synapses that are already weakly potentiated.

The suggestion that forgetting occurs during sleep does not run counter to the many demonstrations of memory enhancing effects of sleep [52]. First, these beneficial mne- monic effects could reflect the selective protection of retained memories against the effects of systematic for- getting, whereas other, unprotected, memories are lost. This view finds some support in recent studies that indi- rectly suggest the occurrence of forgetting in sleep. Two studies have shown that sleep can strengthen certain, relevant memories, whereas no such benefits are observed for memories deemed irrelevant [53,54]. A study in human infants provides stronger evidence, showing that a nap promotes rule generalization in artificial language learn- ing [55]. This finding suggests that contextual elements were forgotten during sleep, permitting abstraction. Sec- ond, it is possible that different sleep phases will selec- tively promote specific types of plasticity. This idea finds some support in recent findings which suggest that synap- tic downscaling occurs preferentially in REM sleep [56], whereas selective synaptic potentiation characterizes slow-wave sleep instead [57]. Thus, the numerous demon- strations of the beneficial effects of sleep on memory reten- tion in species as diverse as honeybees [58], rats [59], and humans [60] may reflect the net benefit of processes that eliminate memories and processes that strengthen them. Memory-specific plasticity parameters, set by the type of post-acquisition signaling discussed above, determine how much decay and how much strengthening will take place during certain sleep phases.

Possible molecular pathways of decay-like forgetting Recent studies on long-term memory maintenance support the notion that hippocampal traces can decay over time. These studies established the constitutively active, atypi- cal protein kinase C isoform M-zeta (PKMz) as both neces- sary and sufficient for maintaining long-term potentiation and for sustaining long-term memory in various tasks and brain regions [61]. PKMz is synthesized upon induction of LTP or during formation of memory [62]. After translation, PKMz is phosphorylated and then stays constitutively active [63]. Transiently inhibiting PKMz activity impairs and often abolishes fully established memories [64], even those that are several months old [65]. On the other hand, overexpression of the kinase can enhance weak long-term memories [66].

Two of these studies on the role of PKMz in long-term memory maintenance are of particular importance for our proposal regarding forgetting by decay. In these, rats ac- quired non-reinforced long-term object location memory, which is known to depend on the hippocampus [67,68]. The first study established that maintaining long-term memory of object location requires continuous activity of PKMz, at both recent (1 day) and more remote (6 days) time points. Despite dependency on PKMz, however, object loca- tion memory is lost sometime between 7 and 35 days after training [69]. The second study asked how PKMz maintains these object location memories and showed that it does so by

116

regulating the trafficking of GluA2-containing AMPA recep- tors in the dorsal hippocampus [70]. These data support the conclusion that PKMz prevents the internalization of these receptors. In that paper, it was also shown that, for auditory fear conditioning, the strength of the freezing response correlates linearly with the amount of GluA2-AMPARs in post-synaptic densities in the basolateral nucleus of the amygdala. The smaller the amount of GluA2-AMPARs in these post-synaptic densities, the less memory was expressed. Taken together, these two studies suggest that the loss of object location memory observed between 7 and 35 days after acquisition likely reflects the loss of GluA2-con- taining AMPA receptors in the relevant synapses in the dorsal hippocampus. This loss of the substrate that supports memory over time is reminiscent of the idea articulated in Thorndike’s law of disuse – that synaptic connections weak- en over time when memory is not exercised [11].

Recent findings demonstrating preserved LTP and long- term memory formation in developmental, as well as in- ducible, PKMz knockout mice [71,72] seem to question whether PKMz is essential for long-term memory mainte- nance, as we claim here. These findings are quite interest- ing, as they likely provide evidence for evolutionarily conserved compensatory mechanisms [73], similar to what has been observed in mutant mice in which aCaMKII autophosphorylation was deficient [74]. Although aCaM- KII autophosphorylation is necessary in wild type animals to form memories, mutant mice were able to compensate by using different mechanisms. In the case of PKMz, another atypical PKC isoform, PKCi/l, can functionally compen- sate when PKMz is absent [75,76]. This is not surprising, as the nucleotide sequences of the full-length isoforms, such as PKCz and PKCi/l, are almost identical: both have the same pseudosubstrate sequence, such that ZIP, the peptide used in many studies to inhibit PKMz, also inhibits PKCi/l [77]. Proteolysis can transform activated PKC iso- forms into constitutively active PKM forms [78]. In addi- tion, like PKMz, PKCi/l is also activated during LTP [79]. Thus, we suggest that, in the absence of PKMz, PKCi/l may assume the essential functional roles of PKMz, such as maintaining memory by regulating GluA2-dependent AMPA receptor trafficking.

There are several possible mechanisms that naturally implement decay [80,81], although it remains to be seen which of these in fact occur. We mention here only two of these. First, it is known that learning can lead to synthesis of PKMz within minutes after the event [82]. Memory reactivation or strengthening (by post-acquisition modula- tions) could maintain PKMz upregulation, thereby sustain- ing the AMPA receptors critical to the maintenance of the memory. If the memory is not reactivated or strengthened, no new PKMz would be supplied, and when existing PKMz

is degraded, internalization of GluA2-containing AMPA receptors will set in, leading to the loss of potentiation and thus memory. Second, reductions of synaptic strength in the form of LTD and depotentation require GluN2b-con- taining NMDA receptors [83]. A signal to internalize GluA2-containing AMPA receptors might thus come in the form of an LTD-like stimulus from GluN2B-containing NMDA receptors. LTD leads to degradation of PKMz [84], and PKMz loss leads to the loss of memory. In order to

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

activate NMDA receptors, action potentials are not neces- sary [85]. Glutamate is released in small amounts even during basal states and, when it binds to NMDA receptors, a small amount of calcium can enter the cell. For this type of calcium channeling, the magnesium block does not need to be removed from the receptor [86]. This calcium influx might resemble an LTD- or depotentiation-like signal, leading to the internalization of GluA2-containing AMPA receptors. The finding that blocking NMDA receptors dur- ing a five-day retention interval prevents forgetting of spatial memory in rats [87] lends support to this account of the role of NMDA receptors in forgetting. Owing to the role of the NMDA receptor in learning and memory forma- tion, this result has been interpreted as supportive of interference-based accounts of forgetting [16] because blocking NMDA receptors during the retention interval likely blocks new learning. However, because the drugs used in this study (CPP and AP5) block all NMDA recep- tors, irrespective of subtype composition, it is also possible that memory was preserved because inactivating GluN2b- NMDA receptors reduced LTD or depotentiation, which prevented memory decay.

Other mechanisms that implement decay are certainly possible [81] and most likely several will be involved; we mention these two to illustrate some possible ways the form of decay we propose here could emerge. We assume, however, that decay-like forgetting will depend on actively regulating GluA2-AMPAR contents at post-synaptic sites and that the NMDA receptor will play a metaplastic role in

Box 1. Predictions

In our model of forgetting, the hippocampal component of most

memories is ultimately lost, such that long-term human memories

will generally be of a semantic rather than an episodic nature

(Figure 1c). This episodic-to-semantic shift over time is very similar, if

not identical, to the context generalization phenomenon, that is, the

tendency to express behavior that once was specific to the learning

context in other contexts over time. In animals, such generalization is

often studied using contextual fear conditioning, as briefly alluded to

above. Approximately 10-15 days after learning, animals that shortly

after training only feared the training context now fear any context

remotely resembling the original one. According to our model, this

generalization results from forgetting of the hippocampal contextual

component of the memory, such that only elemental stimuli

represented outside the hippocampus remain to trigger the fear

response, and these elemental stimuli (e.g., the metal grid on the

floor, the shape of the box, etc.) are common to many contexts. Along

with others [93], we thus predict that such generalization would not

be observed when the hippocampal trace is artificially maintained

during the retention interval by interfering with normal decay

processes.

This approach can be extended to other memory phenomena. For

example, conditioned responses, such as fear to a tone acquired

during auditory fear conditioning, can be extinguished by repeatedly

presenting the conditioned stimulus alone (i.e., presenting the tone

without the reinforcer). As a consequence of this extinction training,

the conditioned stimulus no longer elicits the conditioned response.

However, the conditioned response can return some days after the

extinction training. It remains unclear what processes mediate this

spontaneous recovery, but one possibility is that the extinction

memory could be lost over time through the action of decay-like

forgetting processes as outlined above. Because extinction training

does not literally erase the conditioning memory it inhibits [94], the

conditioned response would return when the extinction event is

forgotten. Consequently, spontaneous recovery could be prevented

determining the degree to which memories will be subject to decay. Such regulation is necessary, as those memories deemed significant will need to be protected from the constant force of forgetting. This might include, for exam- ple, memories accompanied by strong emotional reactions, novelty signals, or memories that reflect recurring events. GluN2b-containing NMDARs may be involved in these regulatory mechanisms, having been linked in previous research to metaplastic processes. For example, they are necessary to induce reconsolidation after retrieval [88]. More importantly, it seems that strong memories, that is, memories that are not quickly forgotten, are character- ized by reduced GluN2b-NMDAR expression [89,90]. It is thus possible that resistance to forgetting could be inverse- ly related to the presence of GluN2b-NMDAR at postsyn- aptic sites. This suggestion is similar to the notion that the ratio of GluN2A- to GluN2b-containing NMDA receptors determines the direction of synaptic plasticity [91] – a stronger GluN2b than GluN2A expression thus favors LTD over LTP processes and vice versa.

Our proposal views forgetting of consolidated long-term memory as an active process that systematically removes learning-induced changes in synaptic potentiation over time. Recent data from Drosophila melanogaster provide strong support for these suggestions [92]. Here, olfactory memory requires the dDa1 dopamine receptor in the mush- room body. Surprisingly, another type of dopamine recep- tor, DAMB, is necessary for forgetting. In mutant flies that did not express the DAMB receptor, forgetting of labile

by artificially maintaining the extinction memory. This possibility

remains to be tested.

If this type of forgetting represents a well-regulated feature of the

brain, then perhaps it could play a part in certain pathologies. One

obvious candidate would be Alzheimer’s disease (AD). Whereas later

stages of AD are characterized not only by devastating memory loss

but also by the inability to form new memories, in earlier stages of the

disease memories can still be formed, even though they are rather

quickly forgotten. Accelerated forgetting has also been found in a

mouse model of AD [95]. Such findings are generally seen as

indicating impaired memory consolidation processes, but it is

possible that deregulated forgetting plays a role, as well. Findings

showing that beta-amyloid promotes removal of postsynaptic AMPA

receptors by engaging pathways involved in long-term depression

[96] suggest that forgetting processes as described above may be

involved in AD pathology. If forgetting is indeed a well-regulated

process, and deregulated forgetting plays a part in AD, then the

pathways involving forgetting may offer novel pharmacological

targets for clinical interventions to alleviate some symptoms of the

disease.

Our model assumes that hippocampal circuit architecture prevents

interference by virtue of pattern separation. Several lines of evidence

suggest that pattern separation is largely a function of the dentate

gyrus. The dentate gyrus is one of the two currently identified areas in

which neurogenesis occurs in the adult brain [97]. As others before

us, we propose that neurogenesis is involved in pattern separation

and thus predict that the rate of neurogenesis in the dentate gyrus will

be inversely correlated with the degree of interference in the

hippocampus [98–100]. Thus, the less neurogenesis, the more new

learning will interfere with memory traces in the hippocampus. We

further predict that reduction of neurogenesis will lead to interference

in both hippocampus-dependent and hippocampus-independent

tasks, based on our assumption that the hippocampus provides

pattern separation for extra-hippocampal areas.

117

Box 2. Questions for future research

� What distinguishes strong, long-lasting consolidated memories

from weak, short-lasting consolidated memories on the molecular

and the systems levels? Are strong memories distinguished by

down-regulation of the molecular pathways involved in decay?

� Is decay of the same speed in all brain areas?

� Does suppression of sleep decelerate decay-dependent forgetting,

as predicted by our model?

� Do forgetting and memory consolidation occur during different

sleep phases or could both processes occur at the same time?

� Does suppression of decay in the hippocampus lead to eventual

catastrophic interference?

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

long-term memories was absent; stronger memories re- quired additional activation of the dopaminergic pathway. In Drosophila, forgetting appears to be an organized pro- cess that depends on activation of a specific pathway that removes long-term memories. Although it remains to be demonstrated to what extent these processes found in an invertebrate organism translate to mammalian brains, it is worth noting (i) that the same receptor class involved in learning is also involved in forgetting, which is similar to the role we propose for the NMDA receptor, and (ii) that the mushroom body, at least functionally, has many similari- ties to the mammalian hippocampus.

Concluding remarks In this article, we have suggested that decay-like forgetting is a well-organized neuronal process that systematically removes memories from the hippocampus over time, per- haps preferentially during sleep. This type of forgetting is essential to maintain overall system functionality. Be- cause most of the memories automatically formed during the day are irrelevant, such forgetting will ensure that most of these unwanted and unneeded memories are re- moved. Understanding decay-like forgetting as a normal and regulated component of memory offers alternative, simpler, and testable explanations for several memory phenomena, and perhaps even contributes to a better understanding of some disorders, such as Alzheimer’s Disease (Box 1). Recent advances in discovering the molec- ular mechanisms involved in long-term memory mainte- nance will provide efficient tools to study these predictions (see also Box 2).

Acknowledgments We would like to thank our colleagues Paola V. Migues, Todd Sacktor, and Almut Hupbach for commenting on earlier drafts of our paper. We also would like to thank the two anonymous reviewers of this paper for their detailed and constructive criticism. O.H. and K.N. were supported by CIHR and NSERC; L.N. was supported by NSF, Down Syndrome Research and Treatment Foundation, Research Down Syndrome, Fonda- tion Lejeune, and the Thrasher Research Fund.

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.tics.2013.01.001.

References 1 Roediger, H.L., III et al. (2010) Forgetting: preliminary

considerations. In Forgetting (Della Salla, S., ed.), pp. 1–22, Psychology Press

118

2 Neath, I. and Brown, G.D.A. (2012) Arguments against memory trace decay: a SIMPLE account of Baddeley and Scott. Front. Psychol. 3, 35

3 Wixted, J.T. (2005) A theory about why we forget what we once knew. Curr. Dir. Psychol. Sci. 14, 6–9

4 Dewar, M. et al. (2012) Brief wakeful resting boosts new memories over the long term. Psychol. Sci. 23, 955–960

5 Skaggs, E.B. (1933) A discussion on the temporal point of interpolation and degree of retroactive inhibition. J. Comp. Psychol. 16, 411–414

6 Dewar, M.T. et al. (2007) Forgetting due to retroactive interference: a fusion of Mü ller and Pilzecker’s (1900) early insights into everyday forgetting and recent research on anterograde amnesia. Cortex 43, 616–634

7 Hardt, O. et al. (2010) A bridge over troubled water: reconsolidation as a link between cognitive and neuroscientific memory research traditions. Annu. Rev. Psychol. 61, 141–167

8 Nader, K. and Hardt, O. (2009) A single standard for memory: the case for reconsolidation. Nat. Rev. Neurosci. 10, 224–234

9 Hupbach, A. et al. (2007) Reconsolidation of episodic memories: a subtle reminder triggers integration of new information. Learn. Mem. 14, 47–53

10 Walker, M.P. et al. (2003) Dissociable stages of human memory consolidation and reconsolidation. Nature 425, 616–620

11 Thorndike, E.L. (1913) Educational Psychology, Teachers College, Columbia University

12 Roediger, H.L.H. and Karpicke, J.D.J. (2006) Test-enhanced learning: taking memory tests improves long-term retention. Psychol. Sci. 17, 249–255

13 Wheeler, M.A. and Roediger, H.L. (1992) Disparate effects of repeated testing: Reconciling Ballard’s (1913) and Bartlett’s (1932) results. Psychol. Sci. 3, 240–245

14 de Oliveira Alvares, L. et al. (2011) Periodically reactivated context memory retains its precision and dependence on the hippocampus. Hippocampus 22, 1092–1095

15 Wixted, J.T. (2004) The psychology and neuroscience of forgetting. Annu. Rev. Psychol. 55, 235–269

16 Rosenzweig, E.S. et al. (2002) Making room for new memories. Nat. Neurosci. 5, 6–8

17 Crick, F. and Mitchison, G. (1983) The function of dream sleep. Nature 304, 111–114

18 Tononi, G. and Cirelli, C. (2003) Sleep and synaptic homeostasis: a hypothesis. Brain Res. Bull. 62, 143–150

19 James, W. (1892) Textbook of Psychology, Macmillan 20 Borges, J.L. (1998) Collected Fictions, Penguin 21 Nadel, L. and Hardt, O. (2011) Update on memory systems and

processes. Neuropsychopharmacology 36, 251–273 22 Yassa, M.A. and Stark, C.E.L. (2011) Pattern separation in the

hippocampus. Trends Neurosci. 34, 515–525 23 Leutgeb, J.K. et al. (2007) Pattern separation in the dentate gyrus and

CA3 of the hippocampus. Science 315, 961–966 24 Nakashiba, T. et al. (2012) Young dentate granule cells mediate

pattern separation, whereas old granule cells facilitate pattern completion. Cell 149, 188–201

25 McHugh, T.J. et al. (2007) Dentate gyrus NMDA receptors mediate rapid pattern separation in the hippocampal network. Science 317, 94–99

26 Tsumoto, T. (1993) Long-term depression in cerebral cortex: a possible substrate of ‘forgetting’ that should not be forgotten. Neurosci. Res. 16, 263–270

27 Marr, D. (1971) Simple memory: a theory for archicortex. Philos. Trans. R. Soc. Lond. B: Biol. Sci. 262, 23–81

28 Teyler, T.J. and Rudy, J.W. (2007) The hippocampal indexing theory and episodic memory: Updating the index. Hippocampus 17, 1158–1169

29 McClelland, J.L. et al. (1995) Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol. Rev. 102, 419–457

30 McCloskey, M. and Cohen, N.J. (1989) Catastrophic interference in connectionist networks: the sequential learning problem. In The Psychology of Learning and Motivation (Bower, G.H., ed.), pp. 109– 165, Academic Press

31 Danker, J.F. and Anderson, J.R. (2010) The ghosts of brain states past: Remembering reactivates the brain regions engaged during encoding. Psychol. Bull. 136, 87–102

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

32 Dudai, Y. (2004) The neurobiology of consolidations, or, how stable is the engram? Annu. Rev. Psychol. 55, 51–86

33 Saksida, L.M. and Bussey, T.J. (2010) The representationa- hierarchical view of amnesia: translation from animal to human. Neuropsychologia 48, 2370–2384

34 Cowan, N. et al. (2004) Verbal recall in amnesiacs under conditions of diminished retroactive interference. Brain 127, 825–834

35 McTighe, S.M. et al. (2010) Paradoxical false memory for objects after brain damage. Science 330, 1408–1410

36 Becker, S. and Wojtowicz, J.M. (2007) A model of hippocampal neurogenesis in memory and mood disorders. Trends Cogn. Sci. 11, 70–76

37 Kuhl, B.A. et al. (2010) Resistance to forgetting associated with hippocampus-mediated reactivation during new learning. Nat. Neurosci. 13, 501–506

38 Butterly, D.A. et al. (2012) Hippocampal context processing is critical for interference free recall of odor memories in rats. Hippocampus 22, 906–913

39 Lledo, P-M. et al. (2006) Adult neurogenesis and functional plasticity in neuronal circuits. Nat. Rev. Neurosci. 7, 179–193

40 Zhang, C-L. et al. (2008) A role for adult TLX-positive neural stem cells in learning and behaviour. Nature 451, 1004–1007

41 Arruda-Carvalho, M. et al. (2011) Posttraining ablation of adult- generated neurons degrades previously acquired memories. J. Neurosci. 31, 15113–15127

42 Sahay, A. et al. (2011) Increasing adult hippocampal neurogenesis is sufficient to improve pattern separation. Nature 472, 466–470

43 Luu, P. et al. (2012) The role of adult hippocampal neurogenesis in reducing interference. Behav. Neurosci. 126, 381–391

44 Warrington, E.K. and Weiskrantz, L. (1978) Further analysis of the prior learning effect in amnesic patients. Neuropsychologia 16, 169–177

45 Dewar, M. et al. (2009) Delaying interference enhances memory consolidation in amnesic patients. Neuropsychology 23, 627–634

46 Hardt, O. et al. (2009) Storage or retrieval deficit: the yin and yang of amnesia. Learn. Mem. 16, 224–230

47 Roozendaal, B. and McGaugh, J.L. (2011) Memory modulation. Behav. Neurosci. 125, 797–824

48 McGaugh, J.L. and Roozendaal, B. (2009) Drug enhancement of memory consolidation: historical perspective and neurobiological implications. Psychopharmacology 202, 3–14

49 Krugers, H.J. et al. (2010) Stress hormones and AMPA receptor trafficking in synaptic plasticity and memory. Nat. Rev. Neurosci. 11, 675–681

50 Kim, J.J. and Yoon, K.S. (1998) Stress: metaplastic effects in the hippocampus. Trends Neurosci. 21, 505–509

51 Pérez-Otaño, I. and Ehlers, M.D. (2005) Homeostatic plasticity and NMDA receptor trafficking. Trends Neurosci. 28, 229–238

52 Diekelmann, S. and Born, J. (2010) The memory function of sleep. Nat. Rev. Neurosci. 11, 114–126

53 Rauchs, G. et al. (2011) Sleep contributes to the strengthening of some memories over others, depending on hippocampal activity at learning. J. Neurosci. 31, 2563–2568

54 Wilhelm, I. et al. (2011) Sleep selectively enhances memory expected to be of future relevance. J. Neurosci. 31, 1563–1569

55 Gómez, R.L. et al. (2006) Naps promote abstraction in language- learning infants. Psychol. Sci. 17, 670–674

56 Grosmark, A.D.A. et al. (2012) REM sleep reorganizes hippocampal excitability. Neuron 75, 1001–1007

57 Chauvette, S.S. et al. (2012) Sleep oscillations in the thalamocortical system induce long-term neuronal plasticity. Neuron 75, 1105–1113

58 Beyaert, L. et al. (2012) Honeybees consolidate navigation memory during sleep. J. Exp. Biol. 215, 3981–3988

59 Inostroza, M. et al. (2013) Sleep-dependency of episodic-like memory consolidation in rats. Behav. Brain Res. 237, 15–22

60 Axmacher, N. et al. (2008) Ripples in the medial temporal lobe are relevant for human memory consolidation. Brain 131, 1806–1817

61 Sacktor, T.C. (2011) How does PKMz maintain long-term memory? Nat. Rev. Neurosci. 12, 9–15

62 Ling, D.S.F. et al. (2002) Protein kinase Mz is necessary and sufficient for LTP maintenance. Nat. Neurosci. 5, 295–296

63 Yao, Y. et al. (2013) Matching biochemical and functional efficacies confirm ZIP as a potent competitive inhibitor of PKMz in neurons. Neuropharmacology 64, 37–44

64 Serrano, P. et al. (2008) PKMz maintains spatial, instrumental, and classically conditioned long-term memories. PLoS Biol. 6, e318

65 Shema, R. et al. (2009) Boundary conditions for the maintenance of memory by PKM in neocortex. Learn. Mem. 16, 122–128

66 Shema, R. et al. (2011) Enhancement of consolidated long-term memory by overexpression of protein kinase Mzeta in the neocortex. Science 331, 1207–1210

67 Hawk, J.D. et al. (2011) Post-training intrahippocampal inhibition of class I histone deacetylases enhances long-term object-location memory. Learn. Mem. 18, 367–370

68 Roozendaal, B. et al. (2010) Membrane-associated glucocorticoid activity is necessary for modulation of long-term memory via chromatin modification. J. Neurosci. 30, 5037–5046

69 Hardt, O. et al. (2010) PKMzeta maintains 1-day- and 6-day-old long- term object location but not object identity memory in dorsal hippocampus. Hippocampus 20, 691–695

70 Migues, P.V. et al. (2010) PKMzeta maintains memories by regulating GluR2-dependent AMPA receptor trafficking. Nature 13, 630–634

71 Volk, L.J. et al. (2013) PKM-z is not required for hippocampal synaptic plasticity, learning and memory. Nature 493, 420–423

72 Lee, A.M. et al. (2013) Prkcz null mice show normal learning and memory. Nature 493, 416–419

73 Frankland, P.W. and Josselyn, S.A. (2013) Neuroscience: Memory and the single molecule. Nature 493, 312–313

74 Radwanska, K. et al. (2011) Mechanism for long-term memory formation when synaptic strengthening is impaired. Proc. Natl. Acad. Sci. U.S.A. 108, 18471–18475

75 Kishikawa, M. et al. (2008) aPKC enables development of zonula adherens by antagonizing centripetal contraction of the circumferential actomyosin cables. J. Cell Sci. 121, 2481–2492

76 Rodriguez, E.M. et al. (2009) Atypical protein kinase C activity is required for extracellular matrix degradation and invasion by Src- transformed cells. J. Cell. Physiol. 221, 171–182

77 Sacktor, T.C. (2012) Memory maintenance by PKMz - an evolutionary perspective. Mol. Brain 5, 31

78 Inoue, M. et al. (1977) Studies on a cyclic nucleotide-independent protein kinase and its proenzyme in mammalian tissues. II. Proenzyme and its activation by calcium-dependent protease from rat brain. J. Biol. Chem. 252, 7610–7616

79 Kelly, M.T. et al. (2007) Regulation of protein kinase Mzeta synthesis by multiple kinases in long-term potentiation. J. Neurosci. 27, 3439– 3444

80 Silva, A.J. and Josselyn, S.A. (2002) The molecules of forgetfulness. Nature 418, 929–930

81 Genoux, D.D. et al. (2002) Protein phosphatase 1 is a molecular constraint on learning and memory. Nature 418, 970–975

82 Osten, P. et al. (1996) Protein synthesis-dependent formation of protein kinase Mzeta in long-term potentiation. J. Neurosci. 16, 2444–2451

83 Dalton, G.L. et al. (2012) NMDA GluN2A and GluN2B receptors play separate roles in the induction of LTP and LTD in the amygdala and in the acquisition and extinction of conditioned fear. Neuropharmacology 62, 797–806

84 Hrabetova, S. and Sacktor, T.C. (2001) Transient translocation of conventional protein kinase C isoforms and persistent downregulation of atypical protein kinase Mzeta in long-term depression. Brain Res. Mol. Brain Res. 95, 146–152

85 Espinosa, F. and Kavalali, E.T. (2008) NMDA Receptor Activation by Spontaneous Glutamatergic Neurotransmission. J. Neurophysiol. 101, 2290–2296

86 Lü scher, C. and Malenka, R.C. (2012) NMDA Receptor-Dependent Long-Term Potentiation and Long-Term Depression (LTP/LTD). Cold Spring Harb. Perspect. Biol. 4, http://dx.doi.org/10.1101/ cshperspect.a005710

87 Villarreal, D.M. et al. (2002) NMDA receptor antagonists sustain LTP and spatial memory: active processes mediate LTP decay. Nat. Neurosci. 5, 48–52

88 Lee, J.L.C. and Everitt, B.J. (2008) Appetitive memory reconsolidation depends upon NMDA receptor-mediated neurotransmission. Neurobiol. Learn. Mem. 90, 147–154

89 Zinebi, F. et al. (2003) NMDA currents and receptor protein are downregulated in the amygdala during maintenance of fear memory. J. Neurosci. 23, 10283–10291

119

Opinion Trends in Cognitive Sciences March 2013, Vol. 17, No. 3

90 Wang, S-H. et al. (2009) Cellular and systems mechanisms of memory strength as a constraint on auditory fear reconsolidation. Nat. Neurosci. 12, 905–912

91 Yashiro, K. and Philpot, B.D. (2008) Regulation of NMDA receptor subunit expression and its implications for LTD, LTP, and metaplasticity. Neuropharmacology 55, 1081–1094

92 Berry, J.A.J. et al. (2012) Dopamine is required for learning and forgetting in Drosophila. Neuron 74, 530–542

93 Biedenkapp, J.C. and Rudy, J.W. (2007) Context preexposure prevents forgetting of a contextual fear memory: Implication for regional changes in brain activation patterns associated with recent and remote memory tests. Learn. Mem. 14, 200–203

94 Bouton, M.E. et al. (2006) Contextual and temporal modulation of extinction: behavioral and biological mechanisms. Biol. Psychiatry 60, 352–360

120

95 Daumas, S. et al. (2008) Faster forgetting contributes to impaired spatial memory in the PDAPP mouse: deficit in memory retrieval associated with increased sensitivity to interference? Learn. Mem. 15, 625–632

96 Hsieh, H. et al. (2006) AMPAR removal underlies ab-induced synaptic depression and dendritic spine loss. Neuron 52, 831–843

97 Zhao, C. et al. (2008) Mechanisms and functional implications of adult neurogenesis. Cell 132, 645–660

98 Winocur, G. et al. (2012) Adult hippocampal neurogenesis and memory interference. Behav. Brain Res. 227, 464–469

99 Appleby, P.A. and Wiskott, L. (2009) Additive neurogenesis as a strategy for avoiding interference in a sparsely-coding dentate gyrus. Network 20, 137–161

100 Becker, S. (2005) A computational principle for hippocampal learning and neurogenesis. Hippocampus 15, 722–738