eIF2-dependent Translation Initiation: Memory Consolidation and Disruption in Alzheimer’s Disease

Abstract

Memory storage is a conserved survivability feature, present in virtually any complex species . During the last few decades, much effort has been devoted to understanding how memories are formed and which molecular switches define whether a memory should be stored for a short or a long period of time. Among these, de novo protein synthesis is known to be required for the conversion of short- to long-term memory. There are a number translational control pathways involved in synaptic plasticity and memory consolidation, including the phosphorylation of the eukaryotic initiation factor 2 alpha (eIF2α), which has emerged as a critical molecular switch for long-term memory consolidation. In this review, we discuss findings pertaining to the requirement of de novo protein synthesis to memory formation, how local dendritic and axonal translation is regulated in neurons, and how these can influence memory consolidation. We also highlight the importance of eIF2α-dependent translation initiation to synaptic plasticity and memory formation. Finally, we contextualize how aberrant phosphorylation of eIF2α contributes to Alzheimer’s disease (AD) pathology and how preventing disruption of eIF2-dependent translation may be a therapeutic avenue for preventing and/or restoring memory loss in AD.

1. Introduction

The long-term storage of memories is an evolutionary adaptation across species that allows them to increase their survivability to the environment. In humans, the ability to communicate and form memories throughout life permits rapid adaptation to environmental changes, as a single individual can learn not only from an experience, but also from the experience of others. It is only natural that such abilities would stimulate human curiosity, which has made memory formation a subject of intense investigation throughout the last century.

The formation of memory is a multi-step process, and each type of memory involves different structures of the encephalon. In general, the mnemonic process is driven by structural components of the limbic system and specific cortical subregions. The most well studied regions involved in memory formation are the dorsal hippocampus and the amygdala. Classical denomination placed the former as an essential structure for spatial memory and the latter as an essential structure for emotional memory. However, these structures receive numerous inputs from different brain regions, such as the dorsal striatum and prefrontal and anterior cingulate cortices. These axonal inputs can either act to activate neurons in the hippocampus or amygdala, or they can act as inhibitory feedforward/feedback systems. It is clear that the initial encoding of certain types of memory takes place in these regions, as well as the conversion of short- to long-term memory (for a complete review on systems consolidation of memory, please refer to [1]).

For many years, most studies focused on memory formation and storage were based on lesions in specific structures of the brain or the modulation of neurotransmitter activity using local drug infusion followed by behavioral assessment. These studies generated valuable information that has served as the foundation for modern neuroscience. The last two decades were marked by dramatic technical advances in the tools used to investigate brain function, allowing for precise spatial and temporal modulation of specific cell types (for complete reviews on the tools available, please refer to [2–5]). With these remarkable tools in hand, the neuroscience community has been able to investigate systems neuroscience under the lens of cellular and molecular mechanisms, adding a new level of understanding of how brain connectivity and memory is acquired, consolidated, and stored throughout life.

One of the molecular mechanisms that is critical for memory consolidation is de novo protein synthesis. For many years it has been know that long-lasting synaptic plasticity in the hippocampus and long-term memory consolidation requires mRNA translation [6–9] and may play a role even earlier in the process of memory formation [10, 11]. Moreover, de novo protein synthesis also has been shown to be involved in memory reconsolidation [12] and extinction [13]. Thus, understanding the role played by master regulators of protein synthesis (also referred to as translational control) in the brain is key to fully comprehending how memory works.

Here, we review recent advances of our understanding of how de novo protein synthesis is involved in long-term memory. We have attempted to integrate information that ranges from systems neuroscience to intricate molecular changes that occur in response to synaptic stimulation. We have focused mainly on how control of translation initiation by eukaryotic initiation factor 2 (eIF2) is key to synaptic plasticity and long-term memory. Finally, we discuss evidence that the disturbance of protein synthesis regulated by eIF2 participates in the memory defects that arise in neurodegenerative disorders.

2. Protein synthesis and its role in memory

When a cell receives a stimulus, the innate response of the cell is to alter its characteristics to adapt to environmental changes. These modifications happen at every level, from epigenetic changes to synthesis of new proteins to post-translational modifications of existing proteins. Notably, the pool of proteins being synthesized by the cell can be impacted by two distinct mechanisms: (1) the modification of the pool of genes being transcribed into mRNAs and (2) rapid regulatory events targeting translation factors, mRNAs and ribosomes that can change the pool of mRNAs that are being loaded onto ribosomes and their subsequent rate of being translated into proteins.

With respect to regulation of protein synthesis, neurons are quite similar to other cells. After postsynaptic activation by neurotransmitters, neurons rapidly respond to adapt. Among the events that orchestrate long-term changes in neuronal function, protein synthesis was extensively shown to participate in synaptic plasticity and memory consolidation. Since the first evidence collected by Flexner (1969), an increasing collection of data support the notion that de novo protein synthesis is crucial for the transition of short-term to long-term memory. For example, the infusion of the translation elongation inhibitor anisomycin in the lateral amygdala immediately after Pavlovian auditory threat conditioning was shown to impair associative memory retention [10]. This effect was not reproduced when the drug was infused six hours after conditioning, which indicates that de novo protein synthesis is necessary during the early stages of memory consolidation. Remarkably, de novo protein synthesis appears to be triggered following other types of synaptic activation, as evidence by the observation that the infusion of anisomycin into the basolateral amygdala after memory retrieval induced the loss of an otherwise consolidated long-term memory [12, 14]. This strongly suggests that de novo protein synthesis has a role not only during memory consolidation, but also during memory reconsolidation.

Despite the advances in memory research achieved using pharmacological approaches with translation inhibitors, it is noteworthy that the infusion of drugs does not permit precise temporal or spatial control of protein synthesis, but instead relies on the half-life of the drug and its diffusion capacity, as well as on its target specificity. Furthermore, translation inhibitors typically target a whole brain region, disregarding the types of cells being affected by the drug. Cell type-specific modulation of translation with genetic deletion of translational control molecules partially overcame this issue, but these approaches still lacked spatial and temporal control, as well as increased risk of abnormalities due to the absence of the protein of interest during brain development. To circumvent these problems, a genetically-engineered mouse model was developed recently that allowed for both spatial and temporal control of the initiation of protein synthesis [11]. This chemogenetic system, termed cell-type-specific drug-inducible Protein Synthesis Inhibition (ciPSI), consists of the catalytic fragment of the double-stranded RNA activated protein kinase (PKR) fused to the NS5A-5B binding site, a site that is recognized by the protease NS3/4, which degrades the iPKR. However, NS3/4 can be rapidly inhibited by the drug asunaprevir, which allows for sustained expression of iPKR. Once expressed and activated, this kinase can block protein synthesis by triggering the integrated stress response via phosphorylation of the alpha subunit of eIF2 (ISR, see section 3.2 for more details). Due to the transient kinetics of the drug [15], the inducible PKR is active for up to one hour, giving the system temporal precision. Using this system, it was shown that rapid protein synthesis in excitatory neurons of the lateral amygdala is necessary during consolidation for long-term threat memory formation [11]. Notably, the ciPSI-induced memory impairment is in line with a recent report demonstrating that another genetically encodable protein synthesis inhibitor blocked structural synaptic plasticity, a process widely known to be part of memory consolidation [16].

The development of technological approaches such as ciPSI have now permitted the interrogation of the role played by protein synthesis in different cell types during memory consolidation. The induction of long-term potentiation (LTP) in excitatory postsynapses of interneurons depends on translation, similar to that in excitatory neurons [17]. In fact, learning-induced hippocampal activity increased de novo protein synthesis in both excitatory and parvalbumin-expressing inhibitory neurons from the hippocampus, indicating that translation in inhibitory neurons might play a role in memory consolidation [18]. Consistent with this notion, chemogenetic blockade of translation initiation in somatostatin and PKCδ-expressing inhibitory neurons in the centrolateral amygdala resulted in impaired threat- and safety-conditioned memory consolidation, respectively. Remarkably, these inhibitory neuron subpopulations likely rely on cell type-specific translation initiation programs to trigger activity-dependent protein synthesis [11]. Similarly, a conditional knock in approach to selectively increase translation in different neuron sub-populations revealed enhanced mnemonic capacity of mice that had elevated protein synthesis in either CamK2α- and somatostatin-expressing neurons, but not in parvalbumin-expressing neurons in the hippocampus [19].

Curiously, enhancing protein synthesis appears to have a pro-mnemonic effect, presumably by reducing the threshold for neuronal activation in the hippocampus and increasing the strength of the memory being consolidated [11, 19–21]. However, the beneficial effect of memory enhancement comes with a cost. Blockade of protein synthesis during consolidation followed by reactivation of excitatory neurons prior to retrieval leads to recovery of auditory threat conditioning memory. However, this fear is generalized, i.e. the mice cannot differentiate between the threat-announcing tone and a non-related sound [11]. Furthermore, different transgenic mouse lines with increased translation in forebrain excitatory neurons present memory extinction defects [22, 23]. Finally, chronic increases in de novo protein synthesis are linked to non-cognitive abnormal behavior, such as higher anxiety, repetitive behavior, and social behavior impairment [9]. Although there are currently no specific mechanisms that explain why increases in global protein synthesis would lead to such aberrant behaviors, one possibility is that the imbalance in translation leads to alterations in the pool of mRNAs being incorporated in ribosomes, leading to differences in the proteome and subsequently cellular function [24–28]. Altogether, these studies indicate that protein synthesis must be well balanced and tightly controlled in neurons to guarantee proper cognitive function and behavior.

2.1. Local translation in dendrites

A closer examination of how neurons are modified in order for a memory to be consolidated reveals crucial molecular changes that strengthen synaptic connections. Historically, synaptic plasticity is tightly associated with the amount of GluA1-containing AMPARs inserted in the postsynaptic density and a broadening or shrinkage of the postsynaptic density [29, 30]. Because inhibition of protein synthesis can disrupt long-lasting synaptic plasticity, there has been great interest in understanding how de novo protein synthesis regulates local synaptic changes. Increasing neuronal protein synthesis leads to increased neuronal excitability as well as elevated amounts of AMPARs at the postsynaptic density [14, 17, 22]. In fact, translation is necessary for the rapid synthesis, as well as the insertion and removal of AMPAR subunits (Figure 1, left inset) [31, 32]. Furthermore, blocking protein synthesis prevents activity-induced increases in the size of the postsynaptic density [16]. These studies indicate that de novo protein synthesis plays a major role in synaptic adaptation in response to activity.

Figure 1. Local protein synthesis alters synapse function.

Postsynaptic protein synthesis (left inset) can be regulated by both ionotropic and metabotropic glutamate receptors, represented here as NMDAR and mGluRs, at excitatory synapses. Translation also can be stimulated by activation of the tyrosine kinase (TyK) receptor TrkB, via brain-derived neurotrophic factor (BDNF). The stimulation of protein synthesis by activation of these receptors is thought to mediate de novo production of AMPAR subunits that assemble into functional receptors that can then be expressed in the postsynaptic density, thereby strengthening that synapse. Notably, it has been suggested that the process of receptor exocytosis is also dependent on protein synthesis. On the other side of the synapse, local protein synthesis in the presynaptic terminal (right inset) may be regulated by retrograde signals such as endocannabinoids that are produced postsynaptically. It is not known whether action potentials can increase protein synthesis. Once translation is enhanced presynaptically, it is believed to play a role in controlling the recycling pool and fusion of neurotransmitter vesicles to the presynaptic membrane. Figure generated with Biorender.

One remarkable feature of neurons is that they are multi-compartmentalized cells. This means that neurons accomplish key adaptive cellular processes in different compartments in an independent way, and one of these processes is local mRNA translation. Dendritic shafts and spines have their own pool of actively translating ribosomes referred to as polyribosomes, as well as selectively localized mRNAs that are important for rapid synaptic responses to presynaptic stimuli that may mediate long-lasting changes in the synaptic contact [33, 34]. These mRNAs can travel from the nucleus to neuronal compartments by being encapsulated in granules [35, 36]. However, it was still unclear how dynamically these mRNAs are translated in dendrites. The use of a single-molecule imaging technology for nascent peptides has revealed that distal dendrites translate mRNAs intermittently, whereas proximal dendrites produce a more steady level of protein synthesis [37]. This opens the possibility that mRNA granules might serve as temporary storage compartments for silent mRNAs, and burst into being actively translated upon postsynaptic depolarization [38].

2.2. Local translation in axons

Historically, it was thought that neuronal protein synthesis was largely restricted to soma and dendrites [39]. This view was challenged by early evidence showing that axons could, similar to somatic and dendritic compartments, could incorporate isotope-labeled amino acids [40, 41]. Moreover, more recent evidence has confirmed that local protein synthesis is also a feature of axons. Electron microscopy coupled with immunolabeling of nascent proteins indicated that presynaptic terminals have robust protein synthesis, even in the absence of stimuli [42]. Further EM investigations revealed that presynaptic boutons harbor translation factors and ribosomal proteins [43]. A deeper investigation using translating ribosome affinity profiling (TRAP) sequencing revealed that mRNAs are locally translated presynaptically and mostly encode for proteins related to translation and synaptic communication [42]. This pool of mRNAs being translated is flexible, and can be shaped by induction of neuronal activity during learning [43]. Thus, it is clear that axonal translation, similar to dendritic translation, is a fundamental process involved in synaptic communication.

Why is local translation so important for axonal function? The answer to this question remains unclear. It was shown that mature axons from visual circuits locally translate mRNAs that encode proteins related to synaptic transmission, specifically GABAergic communication [44]. Remarkably, the activation of the presynaptic endocannabinoid receptor CB1 prevents the release of GABA neurotransmitter in hippocampal synaptic boutons via local protein synthesis [45]. At excitatory connections, the blockade of presynaptic protein synthesis with anisomycin enhances the probability of spontaneous neurotransmitter release in the synaptic cleft and accelerates the recycling of the vesicle pool, indicating that axonal protein synthesis might play a role in conserving energy and preventing excitotoxicity [46]. Finally, the fact that the presynaptic transcriptome of excitatory connections contains numerous mRNAs that encode transmembrane receptors indicate that, similar to what is observed postsynaptically, the presynaptic terminal can rapidly change its protein composition in response to stimuli [42, 43]. However, it is not known whether changes in protein synthesis in response to neuronal activity are directly mediated by the action potential or whether this is a consequence of synaptic activity (Figure 1, right inset).

3. Translation initiation control plays a key role in long-term memory formation

Given the evidence presented above, it is clear that de novo protein synthesis plays a major role in synaptic activation and communication, and memory consolidation. However, the lack of detailed molecular mechanisms that can rapidly stimulate or inhibit protein synthesis must be resolved. For example, what are the checkpoints that control local translation both presynaptically and postsynaptically? If these translational control mechanisms are different, can they be used to modulate the specific synaptic compartments? How are these regulators of protein synthesis integrated with neuronal function? Moreover, how do they contribute to long-lasting synaptic plasticity and long-term memory consolidation? In the next section, we summarize current evidence that the regulation of translation initiation plays a fundamental role in promoting memory consolidation.

3.1. Translation initiation

To better understand how the control of translation initiation contributes to neuronal function, one must first understand how translation initiation occurs. First, it should be noted that mRNA translation is tightly regulated by a number of parallel and converging signaling pathways. It is likely that the complexity of translational control is necessary to ensure that the neuronal response will be appropriate to the stimulus received. These translation control pathways consist of a number of checkpoints that are usually regulated by post-translational modifications, typically protein phosphorylation, that can rapidly increase or decrease translation. One of the major translation initiation checkpoints is the phosphorylation of the eukaryotic initiation factor 2 (eIF2) on its alpha subunit (eIF2α), which we will focus on for the remainder of this review. For detailed reviews on the other translation initiation regulatory events that are involved in synaptic activity and memory consolidation, we would like to refer the reader to [47–49].

Translation initiation relies on the formation of the ternary initiation complex (TIC), which is composed of a molecule of GTP, a ^MettRNA-Met, and eIF2. The assembly of the TIC depends on the interaction between eIF2 and the guanosine exchange factor (GEF) eIF2B, which promotes the exchange of guanosine diphosphate (GDP) bound to eIF2 for guanosine triphosphate (GTP). When the TIC is formed, it is directed to the 40S subunit of the ribosome, docking in the E site, generating the pre-initiation complex. The fully functional initiation complex is assembled when eIF4F, which is bound to the 5’ terminal of the mRNA, docks to the ribosome. The 5’ terminal of the mRNA contains a methylated cap, hence the name cap-dependent translation. Once the scanning ribosome finds the start codon in the mRNA, the GTP is hydrolyzed and eIF2 is released from the complex and sent for recycling where it is reloaded with another GTP and another ^MettRNA-Met (Figure 2, black arrows; for a full overview of the translation initiation process please see [50]).

Figure 2. The Integrated Stress Response.

In the figure, the ISR is depicted by red arrows, while normal translation initiation via eIF2 is shown in black arrows. The eIF2α kinases PERK (localized in the endoplasmic reticulum), PKR, GCN2 and HRI sense specific cellular stress signals and dimerize, thereby promoting their kinase activity to phosphorylate eIF2 on its alpha subunit. Once phosphorylated, eIF2α inhibits the GDP-GTP exchange by the eIF2B complex and thus the ternary initiation complex (TIC) formation, reducing the formation of the 40S preinitiation complex, ultimately blocking protein synthesis. Importantly, ISRIB is capable of boosting protein synthesis independently of eIF2α phosphorylation, since it promotes the GEF activity of eIF2B. However, the downregulation of translation results in the increased translation of a subset of mRNAs with specific 5’UTR elements, including ATF4. Once translated, ATF4 migrates to the nucleus and transcribes a subset of genes that are responsible for responding to cellular stress and for feedback control of the ISR. Amongst these genes, PPP1R15A is transcribed and its protein product, GADD34, dephosphorylates eIF2α. When unphosphorylated, eIF2 binds to eIF2B, which exchanges a GDP for a GTP. The newly formed TIC binds to the 40S preinitiation complex, providing the ^MettRNA-Met necessary for the scanning ribosome to initiate translation. Once the start codon is found, eIF2 is released, and the 80S complex is formed. Figure generated with Biorender.

3.2. The integrated stress response

eIF2α phosphorylation (eIF2α-P) is a critical component of a complex, adaptive signaling cascade known as the Integrated Stress Response (ISR). As stated in its name, this pathway is necessary for fast cellular responses to harmful events that signal the necessity for downregulating translation. eIF2α can be phosphorylated by four different kinases, which act as sensors for various stress signals: the general control non-derepressible 2 (GCN2), activated by amino acid starvation; the double-stranded RNA activated protein kinase (PKR), responsive to viral infections; PKR-like endoplasmic reticulum (ER) resident protein kinase (PERK), stimulated by misfolded protein accumulation in the ER lumen and ER stress; and heme-regulated inhibitor (HRI), activated during heme deficiency, oxidative stress, heat shock, and proteasome inhibition. When one of these kinases is activated, eIF2α-P levels rapidly increase. The phosphorylation of eIF2α dramatically changes the physical interaction between eIF2 and eIF2B, resulting in both decreased GEF activity and ternary complex formation. Without the ternary complex, there is no possibility of loading ribosomes with methionine and thus, translation initiation is blocked.

It is important to note, however, that the activation of the ISR is typically an acute event, with the purpose of resolving an incident of cellular stress. Thus, it is not surprising that the ISR possesses an intricate internal negative feedback control mechanism, which restores protein synthesis to its normal levels after cellular physiology is restored. This is guaranteed by a transcription factor that is preferentially translated when eIF2-dependent initiation is blocked: the activated transcription factor 4 (ATF4). Once translated, ATF4 migrates to the nucleus and transcribes a number of stress-related genes that include chaperones, heat-shock proteins, and proteins involved in translation restoration, including growth arrest and DNA damage protein 34, or GADD34 (PPP1R15A). GADD34 is a scaffolding protein that simultaneously binds protein phosphatase 1 (PP1) and eIF2α-P, and the physical approximation between these two proteins results in eIF2α dephosphorylation [51, 52]. In addition to GADD34, which acts as a rapid feedback response, cells also constitutively express the constitutive repressor of eIF2α phosphorylation, or CReP (PPP1R15B), another scaffolding protein responsible for housekeeping maintenance of translation initiation control (Figure 2, red arrows; for a full review on the ISR, please refer to [53]). Thus, it is clear that the ISR must be tightly regulated in order to rescue the cell from a stressful condition without permanently damaging its physiology.

3.3. eIF2α phosphorylation and memory

The translation initiation machinery can be found in every neuronal compartment, including the soma, dendritic spines, and axons, and it drives the synthesis of new proteins once the neuron receives a local stimulus [42, 54]. Indeed, the regulatory events that control eIF2α-P levels are involved with highly specialized neuronal functions, such as synaptic plasticity and memory consolidation. The phosphorylation level of eIF2α is decreased in the dorsal hippocampus during the early consolidation phase of auditory threat conditioning [20]. Blocking eIF2α dephosphorylation through the infusion of Sal003 in the hippocampus reduced fear memory retention, which was accompanied by impaired hippocampal synaptic plasticity [20, 55]. Remarkably, the eIF2α-driven synaptic plasticity impairment in the hippocampus seems to rely on the expression of ATF4, although it remains elusive how this transcription factor contributes to block long-term potentiation in of the face of increased eIF2α phosphorylation (ATF4 role in neuronal plasticity is discussed more in depth below) [20]. In fact, despite the increase in eIF2α-P being canonically associated to a decrease in protein synthesis, one report suggests that the memory consolidation impairment induced by eIF2α-P accumulation in the hippocampus might not rely on protein synthesis downregulation [56]. Importantly, accumulation of eIF2α-P can induce a persistent long-term depression (LTD) in hippocampal slices, which results in less AMPAR exposed on the surface of neurons in vitro [57]. On the other hand, the genetic substitution of eIF2α by a phospho-dead variant (eIF2α-S51A) resulted in stronger long-term memory consolidation in both contextual and auditory threat conditioning tasks [20]. Collectively, these findings demonstrate that the phosphorylation status of eIF2α is critical for long-term memory formation.

One bottleneck for the study of how translation initiation modulates memory was the lack of reliable pharmacological approaches to modulate this type of translational control downstream of the four eIF2α kinases. The development of the compound ISRIB resulted in the first pharmacological evidence showing that blocking the consequences of eIF2α-P can enhance memory [21]. ISRIB is a small, synthetic molecule that targets eIF2B and increases its GEF activity [58, 59] and more efficient ternary complex formation. An important feature of ISRIB is that it can cross the blood-brain barrier, which allows peripheral delivery of the drug to modulate translation initiation in the brain [21]. Remarkably, the success of ISRIB led to an effort to repurpose already FDA-approved drugs as modulators of this specific pathway [60]. In this groundbreaking work, Halliday and colleagues screened over 1000 compounds from a NINDS custom collection and found that trazodone hydrochloride and dibenzoylmethane could counteract eIF2α-P-driven stress responses. This represents a major clinical advance, as trazodone hydrochloride is already an FDA-approved anti-depressant, which allows for immediate clinical trials to treat other diseases driven by chronic excess eIF2α phosphorylation [60]. It is still unknown, though, whether these compounds share the memory enhancing feature delivered by ISRIB.

As mentioned earlier, one of the most important molecular consequences of accumulation of eIF2α-P is an increase in ATF4 mRNA translation. ATF4, also known as cAMP-responsive element (CRE) binding protein 2 (CREB2), is a transcription factor that coordinates the ISR, transcribing genes involved in responding to cellular stress. However, in the neuroscience field, ATF4/CREB2 also has a prominent role in memory consolidation as it directly competes with CREB1 for its activating substrates. Thus, one might predict that an increase in ATF4 protein levels would result in impaired CREB1-dependent memory consolidation. Indeed, in Aplysia the overexpression of ATF4/CREB2 blocks serotonin-dependent long-term facilitation [61]. Conversely, ATF4/CREB2 expression must be downregulated in the presynaptic sensory neuron to generate 5-HT-dependent long-term facilitation [62]. Furthermore, a single-trial induced learning protocol increased the neuronal expression of a micro-RNA that targets ATF4/CREB2 in Lymnaea, inducing its degradation by Dicer [63]. Curiously, the downregulation of ATF4 mRNA in the hippocampus of mice resulted in impaired synaptic plasticity due to impaired glutamatergic function, which ultimately damaged long-term memory formation [64]. In line with these findings, Pasini and colleagues [65] found that ATF4/CREB2 might modulate the expression of postsynaptic ion channels, such as AMPAR, through the stabilization of Cdc42, a member of the Rho GTPase family. Moreover, hippocampal neurons lacking ATF4/CREB2 express less GluA1 (an AMPAR subunit) on the surface of dendritic spines, which was correlated with less mushroom-shaped (mature) and more filopodia-like (immature) spines [66]. Similarly, the downregulation of hippocampal ATF4/CREB2 resulted in decreased GABA_B inhibitory postsynaptic currents, as well as decreased surface expression of this receptor [67]. These reports are in line with findings by Hu and colleagues [68], which demonstrated that an increase in ATF4 expression in the post-synaptic neuron is necessary to induce persistent long-term facilitation in Aplysia neurons. Collectively, this large body of evidence show that ATF4/CREB2 has a prominent role in synaptic plasticity and memory formation.

The studies described above place the suppression of the ISR (and eIF2α-P) as a definitive modulator of synaptic plasticity and memory consolidation. However, large gaps of how this pathway is physiologically regulated during memory formation remain to be filled. For instance, it is not known whether the reduction of eIF2α phosphorylation both in the hippocampus and in the amygdala after learning is due to inhibition of eIF2α kinases or an increase in the transcription/translation of the PP1 scaffolding proteins GADD34/CReP). Substantial evidence supports the notion that the eIF2α kinases act as a brake for the establishment of long-lasting synaptic plasticity, as well as memory formation. For example, the genetic ablation of PKR facilitates long-term potentiation and enhances memory formation in mice [69]. Examination of another branch of the ISR demonstrated that the knockdown of PERK in pyramidal cells of hippocampal area CA1 enhanced neuronal excitability and threat conditioning memory [70]. However, memory enhancement by knockdown of PERK may come with a cost as ablation of PERK in forebrain excitatory neurons results impairments in reversal learning and extinction memory [23]. Finally, mice with a constitutive deletion of GCN2 display a unique behavioral phenotype: although they have enhanced memory during a task designed to be difficult, their memory is impaired when mice are exposed to a standard training paradigm [71].

Much less is known about the role of GADD34/CReP in memory consolidation. A rapid analysis of the single-cell RNA sequencing database from the Allen Institute reveals that both GADD34 and CReP have limited, but still functional amounts of mRNA expression in different cell types of the mouse brain. However, it is noteworthy that GADD34 is known to have its expression increased in response to activation of the ISR (reviewed by [72]). In a neuroblastoma cell line, the activation of the CB1 receptor induces CREB-dependent transcription of GADD34, which then mediates activity-dependent dephosphorylation of eIF2α [73]. Thus, it is possible that synaptic activation triggers downstream effectors of the ISR, without modulating eIF2α kinases that typically sense cellular stress. However, to obtain definitive answers on how these proteins participate in long-lasting synaptic plasticity and memory consolidation, more detailed molecular mechanistic studies are necessary.

4. Chronic increases in eIF2α phosphorylation in Alzheimer’s disease

The demonstration of the central role played by eIF2α-P in memory formation led researchers to investigate whether translation initiation control is disrupted in pathophysiological conditions, including neurological disorders. Indeed, accumulation of eIF2α-P has been reported in numerous neurodegenerative conditions, including Alzheimer’s disease, prion disease, Parkinson’s disease, amyotrophic lateral sclerosis, and vanishing white matter disease [55, 74–82]. Furthermore, a recent report supports the notion that the ISR might as well be involved in age-dependent cognitive loss [83]. In the next section we will summarize the most recent advances in Alzheimer’s disease.

The first evidence of dysregulation of eIF2 activity in Alzheimer’s disease came in 2002, when Chang and colleagues [74, 75] showed that neurons in AD brains exhibited a marked accumulation of eIF2α-P and PKR-P (Thr466). Higher levels of PKR-P correlate with progressive AD-related neurodegeneration [76]. Notably, the exposure of rat primary neurons to soluble oligomeric forms of the Aβ peptide (AβOs) was shown to trigger the phosphorylation of PKR, which in turn increases eIF2α-P [55]. Additional findings showed that genetic deletion of PKR prevents cognitive deficits induced by the infusion of AβOs in the lateral ventricle of mouse brains [55]. Aβ also can trigger accumulation of eIF2α-P that persists even after high frequency-stimulation (HFS) of hippocampal slices, thus blocking increases in hippocampal protein synthesis following induction of long-term potentiation [78]. Notably, conditional genetic ablation of PERK from forebrain excitatory neurons and constitutive genetic deletion of GCN2 in APP/PS1 mice prevented age-dependent memory loss, possibly due to recovery of postsynaptic proteostasis [78]. In addition to the participation of PKR, PERK and GCN2 in disease progression, there is also a possibility that HRI might be involved in proteostasis disruption in AD as it is known to be activated by well-known AD-related cell stressors, such as oxidative stress and proteasome inhibition [84–87]. However, there is currently no direct evidence of HRI participation in the increased eIF2α-P in AD. These results showed for the first time that targeting defective eIF2-dependent regulation of protein synthesis might be an attractive therapeutic route to counteract AD progression that results in memory loss.

Very recent studies have shown that the sub-chronic administration of ISRIB restores long-lasting synaptic plasticity and memory of both mice infused with AβOs in the lateral ventricle and aged APP/PS1 mice, even without restoration of synaptic proteins or reduction of the hippocampal Aβ load [82]. This report further suggests that therapeutic agents that boost translation in AD brains may be able to improve memory impairments. However, a different regimen of ISRIB administration did not improve the memory of either APP/PS1 or APP J20 models [88, 89]. In addition, ISRIB has been administered only after cognitive defects were already established in AD model mice. This approach disregards the observation that protein synthesis dysregulation contributes to amyloidogenic deposition in AD brains (discussed more in depth below). Furthermore, de novo proteomics has revealed that there is altered synthesis of a subset of proteins in the hippocampus even before the onset of memory impairments in APP/PS1 mice [90]. Finally, the quest for compounds better suited for human treatment remains, as ISRIB is highly insoluble. One alternative approach is the re-purposing of FDA-approved drugs and redirecting them to the treatment of AD. As mentioned earlier, trazodone hydrochloride already has proven effective in preventing prion-related neurodegeneration and cognitive defects in mice [60], suggesting that it may be a suitable candidate for clinical trials with AD patients.

One of the current critical questions concerning dysregulated translational control in neurodegenerative diseases is whether it is a cause or consequence of the disease. The activation of PKR in neuroblastoma cells induces the translation of BACE1, a secretase that is involved in the amyloidogenic processing of the amyloid precursor protein (APP) into amyloid β (Aβ) peptide [91]. The knockout of PKR in another transgenic mouse model of AD, the 5xFAD mouse, alleviated the amyloidogenic processing of APP by restoring the levels of BACE1 in the brain [92]. The sustained activation of PERK may be involved in AD-like pathology development in the temporal lobe of epileptic patients [93]. Similarly, Down syndrome cases that progressed to AD pathology displayed elevated accumulation of eIF2α-P and other ISR markers, when compared to Down syndrome cases without apparent AD pathology [94]. Consistent with this, it was shown that either pharmacological or genetic disruption of the ISR in Down syndrome mice rescues synaptic plasticity and memory deficits [95]. These studies, in addition to others mentioned above, points towards an intriguing hypothesis in which the amyloidogenic branch of the disease promotes itself through increased activation of the ISR, ultimately amplifying AD pathogenesis (Figure 3) [96]. However, despite these encouraging findings, it remains to be determined whether early dysregulation of neuronal protein synthesis triggers AD pathology and the specific roles played by each of the eIF2α kinases in the progression of the disease.

Figure 3. The reciprocal relationship between the amyloidogenic processing of APP and disrupted translation in Alzheimer’s disease.

The amyloid cascade is triggered by the sequential cleavage of amyloid precursor protein (APP) by its secretases. This generates amyloid beta (Aβ) peptides that quickly start to aggregate, generating first soluble Aβ oligomers (AβOs) and then fibrils that will accumulate in the brain parenchyma and form amyloid plaques. Notably, AβOs are known to be powerful neurotoxic molecules that bind to different neuronal receptors and induce a series of harmful events. Among these, AβOs enhance the neuronal levels of eIF2α-P through the activation of PKR. Furthermore, evidence suggests that PERK and GCN2 may also play a role in the chronic elevation of eIF2α-P. Chronic increases in eIF2α-P results in impaired protein synthesis, leading to the translation of specific mRNAs with specialized 5’UTRs, including BACE1, the first secretase that cleaves APP. This sequence of events could trigger a cycle where the amyloidogenic processing enhances eIF2α-P, which then blocks general protein synthesis and promotes amyloidogenic processing by increasing the translation of BACE1. Figure generated with Biorender.

5. Concluding remarks

In this review, we have summarized key findings demonstrating that de novo protein synthesis is a fundamental process required for memory consolidation. Detailed understanding of how translation can contribute to long-lasting plasticity and memory consolidation was long hampered by the lack of state-of-the-art tools to manipulate protein synthesis with spatiotemporal precision. However, the advancement of molecular tools that control translation initiation in specific cells has allowed glimpses of the precise roles played by protein synthesis in cognitive processes.

Even with the recent advances in our understanding of the cell types and temporal windows in which de novo protein synthesis is required for memory consolidation, a number of challenging questions have emerged. Information concerning the role that de novo protein synthesis plays in other memory-related process, such as reconsolidation, retrieval, and extinction, remains embryonic. In addition, there is currently very little evidence of how specific patterns of neuronal activity generated during learning are able to trigger the activation of translational control pathways and whether these are global, neuron-wide changes in translation or whether they are specific to activated synapses. Moreover, the importance of proper regulation of eIF2α phosphorylation during synaptic plasticity and memory consolidation raises the question of how different translation control mechanisms contribute to the orchestrate the new synthesis of proteins that drive long-term memory formation. Do different translational programs play different roles depending on the neuronal subtype? Are these translational control pathways shaping the activity of cell types other than neurons in the brain during memory consolidation? Answers to these types of questions will not only illuminate how memories are formed and stored, but also will contribute to elucidate how the brain reacts to insults that cause neurodegenerative disease and their progression. The observation that pharmacological activation of translation initiation boosts memory in AD models [82] raises fundamental questions regarding which types of cells have disrupted translation and how it participates in disease progression. Elucidating the role played by protein synthesis in specific cell types in AD could provide pave new avenues for treatment of memory impairments of it and other neurodegenerative diseases.