Biochemical mechanisms for translational regulation in synaptic plasticity

Key Points

• Translation factors and translational control mechanisms are downstream targets of several signalling pathways and are crucial in synaptic plasticity. Some forms of translational control alter general protein synthesis, whereas others regulate translation of specific messenger RNAs (mRNAs).

• Translation initiation refers to the assembly of a translation-competent ribosome at the AUG start codon on an mRNA. The first step involves the binding of the translation-initiation factor eIF2, which is a G protein, to methionyl-transfer RNA in a GTP-dependent manner.

• eIF2 has three subunits (α, β and γ), and the conversion of inactive eIF2·GDP to active eIF2·GTP by eIF2B is blocked by phosphorylation of eIF2α. Four kinases that are present in the brain — PKR, HRI, PERK and GCN2 — phosphorylate eIF2α on Ser51.

• The eIF2B enzyme complex consists of five polypeptides (α–ε), with eIF2Bε catalysing guanine nucleotide exchange on eIF2. The importance of eIF2B function in the brain is highlighted by the fact that mutations in each eIF2B subunit can cause leukoencephalopathy with vanishing white matter.

• The integrity of the eIF4F cap-binding complex and, therefore, translation efficiency, is regulated by 4E-BPs. Phosphorylation of 4E-BPs by the extracellular signal-regulated kinase (ERK), phosphoinositide 3-kinase (PI3K) and mammalian target of rapamycin (mTOR) signalling pathways is crucial for protein synthesis-dependent synaptic plasticity and memory.

• The cap-binding protein eIF4E is phosphorylated by the protein kinases Mnk1 and Mnk2, which are phosphorylated and activated by ERK and p38. eIF4E phosphorylation is associated with synaptic plasticity and memory.

• In addition to regulating 4E-BP phosphorylation and function, mTOR directly phosphorylates and activates the S6 kinase (S6K), which phosphorylates the ribosomal protein S6, an essential component of the small, 40S ribosomal subunit. S6K and S6 phosphorylation have been implicated in synaptic plasticity and memory.

• The cytoplasmic polyadenylation element (CPE) in the 3′-untranslated region is important in extension of the poly(A) tail and translation activation. Recent evidence indicates a crucial role for CPE-binding protein in long-term facilitation in Aplysia californica and in hippocampal synaptic plasticity.

Abstract

Changes in gene expression are required for long-lasting synaptic plasticity and long-term memory in both invertebrates and vertebrates. Regulation of local protein synthesis allows synapses to control synaptic strength independently of messenger RNA synthesis in the cell body. Recent reports indicate that several biochemical signalling cascades couple neurotransmitter and neurotrophin receptors to translational regulatory factors in protein synthesis-dependent forms of synaptic plasticity and memory. In this review, we highlight these translational regulatory mechanisms and the signalling pathways that govern the expression of synaptic plasticity in response to specific types of neuronal stimulation.

Main

The consolidation and storage of long-term memories, but not short-term memories, require macromolecular synthesis. At the cellular level, the synthesis of new macromolecules is required for long-lasting forms of synaptic plasticity, including LONG-TERM FACILITATION (LTF) in sensory neurons of the mollusc Aplysia californica and a type of LONG-TERM POTENTIATION (LTP) known as LATE-PHASE LTP (L-LTP) in hippocampal neurons (see Ref. 1 for a review). By contrast, shorter-lasting forms of synaptic plasticity, including SHORT-TERM FACILITATION (STF) in Aplysia and EARLY-PHASE LTP (E-LTP), require post-translational modifications of existing proteins, but not the synthesis of new macromolecules. Although significant progress has been made in understanding transcriptional regulation in long-lasting synaptic plasticity and long-term memory, this is not the only biochemical mechanism that controls gene expression. The regulation of messenger RNA (mRNA) transport, mRNA stability and protein synthesis also contributes to the control of gene expression. The regulation of protein synthesis is particularly intriguing because it allows cells to rapidly modulate the production of proteins without involving new transcription or mRNA transport. To understand the regulation of gene expression during long-lasting forms of synaptic plasticity and long-term memory, it is important to examine the regulation of mRNA translation. The reader is directed to other recent reviews for details about the regulation of transcription and mRNA transport and localization in neurons during periods of neuronal activity^2,3.

Dendrites and dendritic spines contain polyribosomes, translation factors and mRNA^2,4,5,6,7,8, indicating that mRNA is translated in synaptic locations. Consistent with this idea, dendrites can translate mRNA⁹, and new protein synthesis is induced in dendrites after exposure of cultured neurons to either brain-derived neurotrophic factor (BDNF)¹⁰ or activity blockade^11,12. In addition, membrane depolarization, BDNF and glutamate receptor agonists induce the synthesis of new proteins in isolated dendrites and synaptoneurosomes^13,14,15. There is also evidence that local protein synthesis is required for synaptic plasticity. In hippocampal area CA1, both BDNF-induced potentiation¹⁶ and metabotropic glutamate receptor (mGluR)-dependent LONG-TERM DEPRESSION (LTD)¹⁷ are blocked by protein-synthesis inhibitors, even when the pre- and postsynaptic pyramidal cell somas are excised from their dendrites. Presynaptic local protein synthesis is required for long-term facilitation (LTF) in Aplysia sensory neurons¹⁸ and for activity-dependent synaptic plasticity in Xenopus laevis nerve–muscle cultures¹⁹. Furthermore, LTP-inducing stimulation causes polyribosomes to move from dendritic shafts to dendritic spines with enlarged synapses²⁰, which indicates that local protein synthesis might be involved in morphological changes that are associated with LTP. These findings indicate that protein synthesis is triggered at synaptic locations and that local translation is required for several forms of synaptic plasticity. However, until recently, little was known about the biochemical signalling mechanisms that couple neurotransmitter and neurotrophin receptors to translational regulation in synaptic plasticity and memory. In this review, we describe several biochemical mechanisms that regulate translation initiation and discuss examples of this type of regulation in synaptic plasticity and memory.

Regulation of translation initiation

Protein synthesis is divided into three steps: initiation, elongation and termination. Initiation refers to the assembly of a translation-competent ribosome at the AUG start codon on an mRNA; translation elongation is the codon-dependent assembly of a polypeptide; and termination involves release of the completed protein when the ribosome reaches a termination codon. These three steps are facilitated by trans-acting proteins that are referred to, respectively, as eukaryotic initiation factors (eIF) (Box 1), elongation factors (eEF) and release factors (eRF). The initiation phase of protein synthesis (Fig. 1) can be further divided into three steps: first, the specific initiator methionyl–transfer RNA (Met–tRNA_i^Met) binds to the small (40S) ribosomal subunit to form a 43S pre-initiation complex; second, the 43S complex binds to an mRNA to form a 48S pre-initiation complex that scans the AUG start codon; and third, the large (60S) ribosomal subunit joins this complex to form an 80S ribosome. As might be expected for a complex biochemical process, the initiation phase of translation is a common target for regulation.

Figure 1: Pathway of translation initiation in eukaryotes.

A binary complex of eukaryotic translation initiation factor 2 (eIF2) and GTP binds to methionyl–transfer RNA (Met–tRNA_i^Met), and the ternary complex associates with the 40S ribosomal subunit. The association of additional factors, such as eIF3 and eIF1A (1A), with the 40S subunit promotes ternary complex binding and generates a 43S pre-initiation complex. The cap-binding complex, which consists of eIF4E (4E), eIF4G and eIF4A (4A), binds to the 7-methyl-GTP (m⁷GTP) cap structure at the 5′ end of a messenger RNA (mRNA). eIF4G also binds to the poly(A)-binding protein (PABP), thereby bridging the 5′ and 3′ ends of the mRNA. This mRNA circularization and the ATP-dependent helicase activity of eIF4A are thought to promote the binding of the 43S pre-initiation complex to the mRNA, which produces a 48S pre-initiation complex. Following scanning of the ribosome to the AUG start codon, GTP is hydrolysed by eIF2, which triggers the dissociation of factors from the 48S complex and allows the eIF5B- and GTP-dependent binding of the large, 60S ribosomal subunit. Although the precise timing and requirements for the release of factors from the pre-initiation complexes are not clear, the 80S product of the pathway is competent for translation elongation and protein synthesis.

External cues are transduced to regulate protein synthesis by several diverse mechanisms, including post-translational modification of translation factors or binding of regulatory proteins to the 5′- or 3′-untranslated regions (UTRs) of specific mRNAs. To help to understand translational regulatory strategies, it is useful to review the translation pathway and the roles of the translation factors^21,22. The translation factor eIF2, which is a G protein, binds to Met–tRNA_i^Met in a GTP-dependent manner. This ternary complex associates with the 40S subunit and other initiation factors, including eIF1, eIF1A, eIF3 and possibly eIF5, to form the 43S pre-initiation complex. The binding of the 43S complex to an mRNA is facilitated by the eIF4 family of translation factors. In addition to the AUG start codon, the 5′-7-methyl-GTP (m⁷GTP) cap and the 3′-poly(A) tail on mRNAs are also important in translation initiation. The cap-binding complex eIF4F consists of the m⁷GTP cap-binding protein eIF4E, the DEAD-box RNA helicase eIF4A and the scaffolding protein eIF4G. Interestingly, eIF4G also binds to the poly(A)-binding protein (PABP) and the 43S component eIF3. Through its interactions with both eIF4E (binding to the mRNA cap) and PABP (binding to the mRNA poly(A) tail), eIF4G effectively circularizes mRNAs. This mRNA circularization might similarly enhance eIF4F binding and 48S-complex formation, providing a mechanism for the synergistic enhancement of translation by these modifications^23,24. The eIF4A component of eIF4F, with the initiation factor eIF4B, is thought to unwind secondary structures at the 5′ end of mRNAs. These mRNA-remodelling activities, in combination with the eIF4G–eIF3 interaction, are thought to promote the binding of the 43S complex to the mRNA, which forms the 48S pre-initiation complex.

After binding close to the 5′ end of the mRNA, the ribosomal complex migrates or scans down the mRNA and usually stops at the first AUG codon it encounters. Stable secondary RNA structures in the 5′ UTR or proteins that are bound to the 5′ UTR impede ribosome binding and scanning, thereby inhibiting translation^23,25,26. eIF4A or other helicase activities might remodel the 5′ UTR to facilitate ribosomal binding and scanning. Recognition of the start codon triggers hydrolysis of GTP by eIF2, which leads to the release of factors from the 48S complex and allows the eIF5B- and GTP-dependent joining of the large ribosomal subunit. eIF2, which is now bound to GDP, is then released from the 48S complex. Similar to many G proteins, eIF2 has a higher affinity for GDP than for GTP, and the guanine-nucleotide exchange factor (GEF) eIF2B catalyses the exchange of GTP for GDP on eIF2, thereby recycling eIF2 for further rounds of translation initiation.

Translation factors and translational control mechanisms are downstream targets of several signalling pathways and are crucial during development and cellular stress responses. Many forms of translational control are homeostatic responses that alter general protein synthesis. However, reduced nutrient availability, oxidative stress, viral infection and misfolded proteins trigger inhibition of general protein synthesis, but stimulate translation of specific mRNAs. This gene-specific translational control depends on regulatory elements in the mRNA, such as upstream open reading frames (uORFs), RNA secondary structures or regulatory protein-binding sites²². So, mRNA specificity in translational control can be achieved through mRNA-specific binding proteins or through alterations to the general translational machinery (Box 2). Downregulation of general protein synthesis has been linked to two forms of mRNA-specific translational control: first, mRNAs that compete poorly for initiation factors and ribosomes are hypersensitive to small reductions in translational activity and their translation is preferentially inhibited; and second, translation of a specific class of mRNAs, including the uORF-containing GENERAL CONTROL NON-DEREPRESSIBLE 4 (GCN4) mRNA in yeast and the ACTIVATING TRANSCRIPTION FACTOR 4 (ATF4) mRNA in mammalian cells, is paradoxically increased when general translation is inhibited²². In the following sections, we review several mechanisms of translational control and evidence that they are linked to synaptic plasticity.

Phosphorylation of eIF2α

The conversion of inactive eIF2·GDP to active eIF2·GTP by eIF2B is regulated by phosphorylation (Fig. 2). eIF2 has three subunits (α, β and γ), and phosphorylation of eIF2α on serine at residue 51 (Ser51) converts eIF2 from a substrate to a competitive inhibitor of eIF2B²⁷. Phosphorylation of eIF2α does not inhibit the general function of eIF2 — to deliver Met–tRNA_i^Met to the 40S subunit — but renders the protein defective in recycling. Most cells express more eIF2 than eIF2B, and phosphorylation of a fraction of the eIF2 is sufficient to inhibit eIF2B and block protein synthesis. The relative abundance of eIF2 and eIF2B in the nervous system has not been reported.

Figure 2: Recycling of eIF2 by eIF2B and regulation by the eIF2α kinases.

The eukaryotic translation initiation factor 2 (eIF2)–GTP binary complex binds to methionyl–transfer RNA (Met–tRNA_i^Met) and forms a ternary complex that then associates with the 40S ribosomal subunit. After start-codon recognition, GTP is hydrolysed by eIF2 and the binary eIF2–GDP complex is then released. The guanine nucleotide-exchange factor (GEF) eIF2B converts inactive eIF2–GDP to active eIF2–GTP, a process that is inhibited by phosphorylation (P) of the α-subunit of eIF2 on serine 51 by one of the four known eIF2α kinases. Phosphorylation of eIF2α converts eIF2 to a competitive inhibitor of eIF2B, and inhibition of eIF2B results in lowered levels of ternary complexes, which reduces general translation but increases translation of a specific class of messenger RNAs (mRNAs) with upstream open reading frames (uORFs)²². ATF4, activating transcription factor 4; C/EBP, CCAAT/enhancer-binding protein; GCN, general control non-derepressible; HRI, haem-regulated initiation factor 2α kinase; m⁷G, 7-methyl-GTP; PERK, eIF2α kinase 3; PKR, protein kinase-RNA regulated, interferon-inducible double-stranded RNA dependent.

Four kinases — PKR (protein kinase-RNA regulated), HRI (haem-regulated initiation factor 2α kinase), PERK (eIF2α kinase 3) and GCN2 (general control non-derepressible 2) — phosphorylate eIF2α on Ser51 (Table 1). PKR is activated by double-stranded RNA (dsRNA), HRI by low haem levels, PERK by endoplasmic reticulum (ER) stress and unfolded proteins in the ER, and GCN2 by amino-acid limitation. All four kinases are present in the brain. GCN2 mRNA is abundant in the mouse brain^28,29, consistent with the restriction of GCN2 to the CNS during Drosophila melanogaster embryogenesis³⁰. Phosphorylation of eIF2α is observed in the brain and increases significantly in neurons on reperfusion after ischaemia³¹. This eIF2α phosphorylation is probably mediated by PERK³², which is consistent with the finding that downregulation of eIF2α protects neurons during oxidative stress, perhaps by enhancing glutathione synthesis³³. Surprisingly, mice that lack any one of the eIF2α kinases have only limited phenotypes, which indicates that the kinases perform redundant functions by responding to overlapping signals. PKR-knockout mice show elevated sensitivity to some viruses^34,35; HRI-knockout mice have dysregulated globin protein synthesis in red blood cells³⁶; PERK-knockout mice develop diabetes^37,38, which is consistent with the ability of PERK mutations to cause insulin-dependent diabetes in patients with Wolcott–Rallison syndrome³⁹; and GCN2-knockout mice fail to coordinate protein synthesis during amino-acid starvation and show increased lethality when they are deprived of the essential amino acid leucine^40,41. Intriguingly, GCN2-knockout mice also have impaired regulation of phosphorylation of eIF4E-binding proteins (4E-BPs) and S6 kinase 1 (S6K1) (Ref. 41) — two other regulators of protein synthesis that will be discussed later.

Table 1 Protein kinases that regulate translation

Although the eIF2α kinases are generally described as stress-responsive regulators of general protein synthesis, activation of GCN2 in yeast by amino-acid starvation specifically stimulates the expression of GCN4 through regulated re-initiation at uORFs in the GCN4 mRNA⁴². Similarly, in mammalian cells, eIF2α phosphorylation stimulates translation of the mRNA that encodes the transcription factor ATF4 (Refs 43,44). So, eIF2α phosphorylation regulates both general and gene-specific protein synthesis. Phosphorylation of eIF2α during synaptic plasticity and memory has not been directly investigated, but treatment of cultured neurons with BDNF, which induces lasting plasticity, decreases eIF2α phosphorylation and enhances protein synthesis⁴⁵. In addition, GCN2-knockout mice have altered synaptic plasticity and memory (N. Sonenberg, personal communication). These findings indicate that proper regulation of eIF2α phosphorylation is probably required for normal synaptic plasticity and memory.

Regulation by eIF2B

eIF2 function is modulated by the regulation of eIF2B, as well as being regulated by phosphorylation of eIF2α. The eIF2B complex consists of five polypeptides (α–ε), with eIF2Bε catalysing guanine nucleotide exchange on eIF2 (Ref. 27). The importance of eIF2B function in the brain is highlighted by the fact that mutations in each eIF2B subunit can cause leukoencephalopathy with vanishing white matter⁴⁶. Phosphorylation of eIF2Bε by glycogen synthase kinase 3β (GSK3β) in vitro inhibits its activity⁴⁷, and eIF2Bε is dephosphorylated in vivo after inactivation of GSK3β, which leads to increased eIF2B activity^47,48. Although regulation of GSK3β has not been reported to be important in synaptic plasticity, treatment of neuronal cultures with BDNF results in decreased GSK3β activity due to increased phosphorylation, and this is correlated with an increase in eIF2B activity⁴⁵. GSK3β also seems to be important in learning and memory. Mice that overexpress GSK3β show spatial learning deficits in the Morris water maze⁴⁹ that have been attributed to the downstream effects of tau hyperphosphorylation. However, the learning deficits could also result from enhanced phosphorylation and inhibition of eIF2B, and the accompanying decrease in the levels of active, GTP-bound eIF2. Further studies are necessary to determine whether GSK3β regulates eIF2 activity during either synaptic plasticity or learning and memory.

Regulation by 4E-BPs

The integrity of the eIF4F cap-binding complex is modulated by 4E-BPs²⁴. When not bound to 4E-BP, eIF4E readily associates with eIF4G to form the eIF4F complex and promote translation initiation. The 4E-BPs block eIF4F formation and inhibit protein synthesis by competing with eIF4G for binding to eIF4E. The binding of 4E-BPs to eIF4E is regulated by phosphorylation²⁴: hypophosphorylated 4E-BPs bind to eIF4E and inhibit translation, whereas multi-site phosphorylation of 4E-BPs prevents their binding to eIF4E and allows eIF4F formation. Three 4E-BPs have been identified in mammals: 4E-BP1 is most prominent in adipose tissues and the pancreas, and 4E-BP1-knockout mice have less white adipose tissue⁵⁰; 4E-BP3 is most abundant in the liver; and 4E-BP2 is the most abundant isoform in the brain, which contains little or no 4E-BP1 and 4E-BP3 (Ref. 50). Phosphorylation of 4E-BPs occurs in an ordered, hierarchical fashion. Threonine residues 37 and 46 (Thr37 and Thr46) are phosphorylated first, and this primes phosphorylation of Thr70 and Ser65 (Ref. 51). Phosphorylation of all these residues is required to block eIF4E binding. 4E-BP phosphorylation is regulated by the extracellular signal-regulated kinase (ERK), phosphoinositide 3-kinase (PI3K) and mammalian target of rapamycin (mTOR) signalling pathways²⁴. Although phosphorylation of 4E-BP on the regulatory sites is sensitive to inhibitors of ERK, PI3K and mTOR, the identity of the kinase that phosphorylates each site has not been firmly established.

ERK is required for most forms of synaptic plasticity and memory^52,53. Similarly, PI3K is required for several protein synthesis-dependent forms of synaptic plasticity and memory^54,55,56,57. Rapamycin, an inhibitor of mTOR that decreases 4E-BP phosphorylation, inhibits synaptic plasticity in invertebrates, including LTF in Aplysia sensory neurons⁵⁸ and at the crayfish neuromuscular junction⁵⁵. In the hippocampus, mGluR-induced LTD⁵⁹, mGluR-dependent depotentiation of LTP⁶⁰, insulin-induced LTD⁶¹, BDNF-induced potentiation⁷ and L-LTP⁷ are all blocked by rapamycin. So, several forms of synaptic plasticity and memory require ERK, PI3K, mTOR and, presumably, phosphorylation of 4E-BP (Fig. 3). Consistent with this idea, emerging evidence indicates that proper regulation of 4E-BP is required for normal synaptic plasticity and memory. Treatment of hippocampal cultures with BDNF increases the phosphorylation of 4E-BP⁶², L-LTP-inducing stimulation is associated with increased phosphorylation of 4E-BP and an increase in eIF4F-complex formation, and 4E-BP2-knockout mice have altered synaptic plasticity and memory (J. Banko and E.K., unpublished observations). Therefore, proper regulation of 4E-BP by PI3K, ERK and mTOR is likely to be crucial for protein synthesis-dependent synaptic plasticity and memory.

Figure 3: Signalling pathways that are involved in translational regulation in L-LTP and mGluR-dependent LTD.

Activation of group I metabotropic glutamate receptors (mGluRs) and NMDA (N-methyl-D-asparate) receptors (NMDARs) activate the mitogen-activated protein kinase kinase (MEK)/extracellular signal-regulated kinase (ERK) and phosphatidylinositol 3-kinase (PI3K) signalling pathways. Sequential activation of PI3K, phosphoinositide-dependent kinase 1 or 2 (PDK1/2), Akt and mammalian target of rapamycin (mTOR) results in activation of S6 kinase 1 (S6K1). Phosphorylation (P) of eIF4E-binding proteins (4E-BPs) by mTOR and other kinases induces the release of 4E-BP from eukaryotic translation initiation factor 4E (eIF4E) and results in the association of eIF4E with eIF4G and the formation of the active eIF4F (eIF4E·eIF4A·eIF4G) complex. eIF4F promotes messenger RNA (mRNA) binding to the 43S pre-initiation complex to form the 48S initiation complex. The eIF4F complex and the poly(A) tail act synergistically to stimulate mRNA translation. The ERK-dependent phosphorylation of both MAPK-interacting serine/threonine kinase 1 (Mnk1), which can phosphorylate eIF4E, and S6K1, which can phosphorylate ribosomal protein S6, is correlated with enhanced translation initiation. Of note, the signal transduction cascades depicted here are also activated by brain-derived neurotrophic factor (BDNF) in the hippocampus and cultured cortical neurons, and by serotonin in Aplysia californica sensory neurons. m⁷G, 7-methyl-GTP; PIP₂, phosphatidylinositol-4,5-bisphosphate; PIP₃, phosphatidylinositol-3,4,5-trisphosphate.

Phosphorylation of eIF4E

In addition to the regulation of eIF4F assembly by 4E-BP phosphorylation, the cap-binding protein eIF4E is a target for direct phosphorylation. Phosphorylation of eIF4E is stimulated by the ERK and p38 mitogen-activated protein kinase pathways, and correlates with increased translation rates in serum-stimulated cells⁶³. The protein kinases MAPK-interacting serine/threonine kinase 1 and 2 (Mnk1 and Mnk2), which are phosphorylated and activated by ERK and p38, phosphorylate eIF4E on Ser209 (Refs 64–66). In mice that are deficient in both Mnk1 and Mnk2, phosphorylation of eIF4E is abolished⁶⁷. Interestingly, both Mnk1 and Mnk2 bind to the C terminus of eIF4G, and phosphorylation of eIF4E by Mnk1 or Mnk2 seems to depend on the binding of both proteins to the scaffold protein eIF4G^24,64. So, eIF4E phosphorylation by Mnk1 or Mnk2 is an indirect measure of eIF4F assembly. Phosphorylation of eIF4E reduces its cap-binding affinity^68,69, which indicates that phosphorylation promotes eIF4E recycling after the ribosome has bound to an mRNA⁶³. According to this model, dephosphorylated eIF4E binds to an mRNA cap, which then recruits eIF4F and the ribosome. Subsequent phosphorylation of eIF4E releases it from the cap and allows the ribosome to begin scanning. Despite the compelling nature of this model and the correlation of eIF4E phosphorylation with increased translation rates, mice that lack both Mnk1 and Mnk2 are viable and apparently normal⁶⁷. However, synaptic plasticity and memory have not been analysed in these mice. In addition, and at odds with the lack of phenotype in the Mnk1/Mnk2-knockout mice, Drosophila that express a non-phosphorylatable eIF4E–S251A mutant, which corresponds to a Ser209 mutant in mammalian eIF4E, show impaired viability and growth, indicating that eIF4E phosphorylation is important for normal development⁷⁰.

eIF4E phosphorylation might be important for synaptic plasticity and memory (Fig. 3). For example, Aplysia neurons treated with serotonin, which is required for LTF, have a p38-dependent increase in eIF4E phosphorylation⁷¹. In addition, ERK-dependent increases in the phosphorylation of eIF4E are induced in cultured neurons that are treated with BDNF⁶² and in the CA1 region of hippocampal slices treated with NMDA (N-methyl-D-aspartate)⁷². More recently, using dominant-negative MAPK/ERK kinase (MEK) mice that have decreased ERK activity, Kelleher and colleagues showed that L-LTP is associated with an ERK-dependent increase in the phosphorylation of eIF4E⁶². In further studies with the dominant-negative MEK mice, this same group showed that an ERK-dependent increase in the phosphorylation of eIF4E accompanies ASSOCIATIVE FEAR CONDITIONING, indicating that phosphorylation of eIF4E might also be involved in hippocampus-dependent memory formation⁶². Although there is no direct evidence that eIF4E phosphorylation is required for protein synthesis-dependent synaptic plasticity, these studies provide intriguing correlative data that link phosphorylation of eIF4E by Mnk1 through activation of either ERK or p38 to the regulation of translation initiation in synaptic plasticity and memory.

S6 and the S6 kinases

In addition to regulating 4E-BP phosphorylation and function, mTOR activation has been linked to translation elongation (Box 3) and mTOR directly phosphorylates and activates the p70 S6K⁷³. S6K is also activated through the PI3K, phosphoinositide-dependent kinase 1 (PDK1) and ERK pathways⁷⁴ (Fig. 3). S6K phosphorylates the ribosomal protein S6, an essential component of the small, 40S ribosomal subunit. The role of S6 and S6 phosphorylation in translational regulation is not well understood. S6 is located close to the mRNA- and tRNA-binding sites on the 40S subunit, and conditional deletion of S6 in the mouse liver impairs ribosome biogenesis and cell proliferation but not cell growth⁷⁵. The role of S6 phosphorylation has been difficult to study owing to the presence of two S6Ks (S6K1 and S6K2) and additional kinases in the ERK pathway that phosphorylate S6 (Ref. 76). In general, S6 phosphorylation is enhanced in growth factor- and nutrient-stimulated cells and correlates with increased levels of translation^73,74. The S6Ks and phosphorylation of S6 were previously implicated in the translational regulation of a specific class of mRNAs that encode components of the translational machinery, including eEF2, eEF1A, PABP and S6 (Ref. 77). These TERMINAL OLIGOPYRIMIDINE (TOP) mRNAs contain polypyrimidine tracts at the 5′ end, and their translation is upregulated by many of the conditions that stimulate S6 phosphorylation⁷⁷. However, in cells that are deficient in both S6K1 and S6K2, the regulation of TOP mRNA translation is maintained⁷⁶ and S6 is still phosphorylated, leaving open the possibility that S6 phosphorylation stimulates TOP mRNA translation⁷⁶. Another substrate of the S6Ks is the translation initiation factor eIF4B⁷⁸, a stimulator of the helicase eIF4A and of mRNA binding during translation initiation. It is not clear how, or whether, S6K phosphorylation modulates eIF4B function in protein synthesis.

Although the roles of S6K and S6 phosphorylation in translational regulation are not clear, the kinases have been implicated in synaptic plasticity and memory. For example, Aplysia synaptosomes treated with serotonin — a treatment that results in LTF in Aplysia neurons — show an increase in S6K1 activity, S6 phosphorylation and in levels of S6 (Ref. 79) and eEF2 (Ref. 80). In the CA1 region of the hippocampus, L-LTP-inducing stimulation results in an mTOR-dependent increase in S6K phosphorylation⁸¹. In addition, injection of the mouse hippocampus with a synthetic phospho-peptide to stimulate PI3K results in increased S6K phosphorylation and improved hippocampus-dependent memory⁸². Both L-LTP and associative fear conditioning are associated with an ERK-dependent increase in S6 phosphorylation in CA1 (Ref. 62). These findings indicate that the regulation of S6K and S6 is important for synaptic plasticity and memory. It will be of interest to determine whether the protein synthesis-dependent forms of synaptic plasticity and memory require an increase in the synthesis of ribosomal proteins and other components of the translation apparatus.

Regulation by the CPE-binding protein

In Xenopus oocytes, translation of a specific class of maternally inherited mRNAs is impaired, partly owing to a short poly(A) tail⁸³. Translation of these dormant mRNAs is activated during the maturation of the oocyte or during early embryogenesis. The activation requires extension of the poly(A) tail, which depends on two RNA sequence elements in the 3′ UTR: the cytoplasmic polyadenylation element (CPE), a U-rich sequence that serves as a binding site for the CPE-binding protein (CPEB), and the sequence 5′-AAUAAA-3′, which binds to the cleavage and polyadenylation specificity factor (CPSF). In the oocyte, where translation of CPE-containing mRNAs is repressed (masked), CPEB binds to the 3′ UTR and recruits the protein Maskin⁸³ (Fig. 4). Maskin contains an eIF4E-binding site and might act as a 4E-BP, blocking eIF4F formation (or binding) on the masked mRNA and thereby preventing its translation⁸⁴. In this way, the CPEB–Maskin–eIF4E interaction could circularize a target mRNA and keep it in a translationally dormant state owing to its short poly(A) tail and impaired eIF4F binding (Fig. 4).

Figure 4: Regulation of translation initiation by polyadenylation after NMDA receptor activation.

The 3′-untranslated region of a specific class of messenger RNA (mRNA) contains sequences that allow the binding of cytoplasmic polyadenylation element-binding protein (CPEB) and polyadenylation specificity factor (CPSF). Translation of transcripts that are bound to CPEB and its binding partner Maskin is inhibited, but can be de-repressed by extension of the poly (A) tail. NMDA (N-methyl-D-aspartate) receptor (NMDAR) activation results in calcium influx and activation of Aurora and/or calcium/calmodulin-dependent protein kinase II (CaMKII), which then phosphorylates (P) CPEB. This leads to the interaction between CPEB and CPSF and the subsequent recruitment of poly(A) polymerase (PAP) to lengthen the poly(A) tail. Poly(A)-binding protein (PABP) then binds to the extended poly(A) tail, interacts with eIF4G to circularize the mRNA, which releases eIF4E from Maskin, and results in enhanced translation of the CPE-containing mRNAs. m⁷G, 7-methyl-GTP; uORF, upstream open reading frame.

At the onset of maturation, CPEB is phosphorylated by a member of the aurora family of protein kinases⁸³. Phosphorylation of CPEB increases its affinity for CPSF, thereby promoting binding of CPSF to the 3′ UTR of the dormant mRNAs (Fig. 4). CPSF then recruits poly(A) polymerase (PAP), which elongates the poly(A) tail on the mRNAs and creates additional binding sites for PABP. According to current models, PABP directly recruits eIF4G, which competes with Maskin for binding to eIF4E⁸³. Enhanced eIF4G binding causes Maskin to dissociate and relieves translational repression of the dormant mRNAs. At the same time, formation of the PABP–eIF4G–eIF4E complex circularizes the mRNAs and promotes 43S-complex binding. So, translation of the targeted mRNAs is promoted by disruption of the CPEB–Maskin–eIF4E translation-silencing (masking) complex through enhanced binding of PABP to the elongated poly(A) tails and subsequent eIF4G recruitment (Fig. 4).

Recent studies indicate a crucial role for CPEB in synaptic plasticity. In Aplysia sensory neurons, LTF is correlated with a PI3K- and mTOR-dependent increase in CPEB, whereas downregulation of CPEB expression by antisense oligonucleotides blocks LTF⁸⁵, which indicates that CPEB is required for this type of plasticity. LTF in Aplysia is also associated with an increase in actin mRNA polyadenylation that depends on cyclic AMP (cAMP)-dependent protein kinase⁸⁶. CPEB is also required for hippocampal synaptic plasticity. LTP that is induced by theta-burst stimulation (TBS), E-LTP and captured LTP are reduced in CPEB1-knockout mice⁸⁷. Interestingly, L-LTP that is induced with stronger stimulus protocols (for example, multiple trains of TBS and HFS) is unaltered in the CPEB1-knockout mice⁸⁷. These findings indicate that CPEB1 is involved in LTP that is induced by weaker stimulation protocols but not in L-LTP. Other CPEB family members in the brain⁸⁸ might be involved in L-LTP that is induced by strong stimulation protocols. Consistent with the proposed role of CPEB in synaptic plasticity, stimulation of cultured neurons with NMDA results in polyadenylation and translation of CPE-containing mRNAs in dendrites⁸⁹. Although CPEB governs the translation of only a select group of mRNAs, it is noteworthy that one of these target mRNAs in the mouse brain encodes the α-subunit of calcium/calmodulin-dependent protein kinase II (α-CaMKII). Interestingly, α-CaMKII is rapidly synthesized after LTP-inducing stimulation^90,91 and is necessary for LTP and memory⁹² (Box 1). One caveat to this model of CPEB-dependent translational regulation in synaptic plasticity is that no Maskin homologues have been identified in mammals. It is possible that distinct eIF4E-binding proteins might be functional homologues of Maskin, which are recruited by CPEB to compete with the eIF4E–eIF4G interaction, silence translation and generate dormant mRNAs.

Two extra features of CPEB indicate that CPEB-mediated translational regulation might be important in long-lasting synaptic plasticity and long-term memory. First, an amino-terminal extension of neuronal CPEB from Aplysia can be converted into a PRION-like state⁹³. This autocatalytic mechanism in the mammalian hippocampus would allow CPEB to self-perpetuate at synapses, which provides a mechanism for the maintenance of long-lasting increases in protein synthesis during synaptic plasticity and long-term memory. Second, CPEB is regulated by CaMKII, a kinase that is crucial for both LTP and memory function⁹². CPEB is phosphorylated by CaMKII in vitro, and CPE-dependent translation that is induced by membrane depolarization also partially depends on CaMKII (Ref. 94). So, activation of CaMKII after LTP-inducing stimulation could stimulate CPE-dependent translation, including translation of α-CaMKII mRNA (Box 1), and promote a feedforward mechanism for maintaining increased protein synthesis during LTP and memory. Further studies are necessary to test the exciting hypothesis that persistent CPE-dependent translation is a molecular mechanism for long-lasting synaptic plasticity and long-term memory.

Concluding remarks

Studies in several model systems, including invertebrate and mammalian cell cultures, hippocampal slices and intact behaving animals, indicate that translational regulatory mechanisms are crucial for many forms of synaptic plasticity and memory. Although progress has been made in identifying signalling cascades that couple neurotransmitter and neurotrophin receptors to translation initiation regulatory factors, several important issues and questions remain to be considered.

One important question is whether protein synthesis-dependent synaptic plasticity in invertebrates and vertebrates involves the same translational regulatory mechanisms. Protein synthesis-dependent synaptic plasticity in vertebrates seems to be localized postsynaptically, whereas in invertebrates such as Aplysia it seems to be localized presynaptically. Presynaptic protein synthesis has not been convincingly shown in adult vertebrate systems; however, presynaptic protein synthesis is involved in growth-cone navigation in response to environmental stimuli in developing axons⁹⁵. Similarly, within a particular neuron, it is not known whether the translational regulatory mechanisms are identical between axons, dendrites and the soma. So, although translational regulatory strategies seem to be conserved across species, key regulators such as the eIF2α and 4E-BP kinases might be spatially restricted, leading to the localized control of protein synthesis within the neuron.

Another important question concerns the spatial, temporal and magnitude differences in translational regulation during protein synthesis-dependent forms of synaptic plasticity and memory. The signalling pathways that couple various cell-surface receptors to translation initiation have several points of convergence and divergence, which would allow neurons to fine-tune translational regulation in response to a particular type of synaptic input. Careful characterization of the differences in the localization, duration and magnitude of the biochemical changes in translation initiation factors should provide important clues to the identity of the mRNAs that are translated in response to various types of input.

Traditionally, both LTP in the hippocampus and LTF in Aplysia have been described as having two distinct phases: a transient phase that involves post-translational modification of pre-existing proteins, and an enduring phase that requires de novo macromolecular synthesis. This enduring phase has distinct kinetic patterns of sensitivity to transcription and translation inhibitors: the former producing a delayed inhibition, and the latter impeding the initiation of the enduring plasticity^{62,96,97,98,99}. These findings are consistent with the interesting possibility that local translation might be part of a retrograde signal that recruits nuclear transcription¹⁰⁰. So, long-lasting synaptic plasticity might require initial induction of protein synthesis followed by induction of transcription to generate additional proteins. There are several examples of translation activation by particular patterns of stimulation that normally induce protein synthesis-independent, short-lasting plasticity^58,101. What then is the function of translation activation? An intriguing possibility is that translation activation is a component of a tagging mechanism used by the activated synapse, whereby the synapse is marked for the consolidation of enduring synaptic plasticity. Although it is clear that translational activation does not account for all aspects of synaptic tagging⁹⁶, it is required to mark the synapse for persistent functional^58,101 and structural⁵⁸ changes that underlie enduring plasticity. Finally, there are examples of translation-dependent, transcription-independent forms of synaptic plasticity. In both BDNF-induced potentiation¹⁶ and mGluR-dependent LTD¹⁷, local translation is necessary for the expression of synaptic plasticity. Despite these advances, it still is not known how the mRNAs are selected or how the encoded proteins are used to sustain the different types of synaptic plasticity (Box 4).

Much progress has been made in delineating biochemical signalling mechanisms that regulate translation initiation during synaptic plasticity and memory. Ongoing studies with diverse model systems — most notably knockout mice that are deficient in key translational regulatory factors — are revealing interesting links among the biochemical activities of translation factors, the electrophysiology of neurons and mouse behaviour. We anticipate that these and future studies will shed light on the molecular mechanisms that underlie protein synthesis-dependent synaptic plasticity and memory.

Box 1 | Roles of translation factors, complexes and regulatory proteins

• Eukaryotic initiation factor 1 (eIF1): promotes the binding of eIF2·GTP·Met–tRNA_i^Met (methionyl–transfer RNA) and messenger RNA (mRNA) to the small (40S) ribosomal subunit.

• eIF1A: promotes the binding of eIF2·GTP·Met–tRNA_i^Met and mRNA to 40S.

• eIF2: binds GTP and Met–tRNA_i^Met to form a ternary complex; delivers Met–tRNA_i^Met to 40S.

• eIF2α: the regulatory subunit of eIF2; phosphorylated on Ser51 by general control non-derepressible 2 kinase (GCN2).

• eIF2B: guaninine nucleotide exchange factor (GEF) for eIF2; mutated in leukoencephalopathy with vanishing white matter disease (VWM).

• eIF3: promotes the binding of eIF2·GTP·Met–tRNA_i^Met and mRNA to 40S.

• eIF4A: an RNA helicase and ATPase; a component of eIF4F.

• eIF4B: a stimulator of eIF4A.

• eIF4E: 7-methyl-GTP (m⁷GTP) cap-binding protein; a component of eIF4F.

• eIF4F: the cap-binding complex, which consists of eIF4A, eIF4E and eIF4G, promotes mRNA binding to the 43S complex.

• eIF4G: the scaffolding component of eIF4F. eIF4G binds to eIF4E, eIF4A, poly(A)-binding protein (PABP), eIF3, and MAPK-interacting serine/threonine kinase 1 and 2 (Mnk1 and Mnk2).

• eIF5: the GTPase-activiting protein (GAP) for eIF2.

• eIF5B: a GTPase; promotes ribosome-subunit joining.

Other proteins and complexes

• PABP: poly(A)-binding protein, which also binds to eIF4G.

• 4E-BP: a translation-regulatory protein; binds to eIF4E and inhibits eIF4F formation.

• CPEB: an RNA-binding protein; recognizes the cytoplasmic polyadenylation element (CPE) in the 3′-untranslated region of target mRNAs.

• Maskin: a protein that binds to both CPEB and eIF4E, and blocks translation initiation by preventing eIF4F formation on the masked (repressed) mRNA.

• 43S pre-initiation complex: a translation complex that consists of the 40S subunit, eIF2·GTP·Met–tRNA_i^Met, eIF1, eIF1A, eIF3 and possibly eIF5.

• 48S pre-initiation complex: a translation complex that consists of the 43S complex and mRNA.

Box 2 | Specificity of mRNA translation in LTP and mGluR-LTD

Long-term potentiation (LTP), induced by either high-frequency stimulation or brain-derived neurotrophic factor (BDNF), and metabotropic glutamate receptor (mGluR)-dependent long-term depression (LTD) seem to use similar signalling pathways to regulate translation initiation in the hippocampus (Fig. 3). How does activation of the same signalling pathways result in LTP on one hand and mGluR-LTD on the other? There are several possibilities. First, differences in the magnitude or duration of the activated pathways might result in either LTP or LTD. As noted in the main text, phosphorylation of eukaryotic initiation factor 2α (eIF2α) downregulates general protein synthesis while stimulating translation of a specific class of messenger RNAs (mRNAs) that contain upstream open reading frames (uORFs). Therefore, differences in the extent of inhibition of general protein synthesis and/or the threshold of such inhibition that is required for the translation of specific uORF-containing mRNAs might determine whether LTP or LTD is induced. Second, there might be spatial differences in the activation of the signalling pathways, which result in the activation of discrete signalling modules that translate mRNAs in that particular location. Third, cap-independent translational regulation might be important in dictating whether LTP or LTD is induced. In internal initiation, ribosomes access the translational start site by binding to an mRNA sequence called an INTERNAL RIBOSOMAL ENTRY SITE (IRES) that is remote from the 7-methyl-GTP (m⁷GTP) cap of the mRNA¹⁰². Several conditions that inhibit general translation activate IRES-mediated translation¹⁰², and in Aplysia californica a form of neuronal plasticity is associated with enhanced IRES-directed translation¹⁰³. Therefore, redirection of the translational program of a neuron through the activation of an alternative initiation mechanism would be a simple, yet elegant, method to regulate specific mRNA translation in the synapse. Fourth, different receptors might regulate the translation of specific mRNAs. For example, phosphorylation of cytoplasmic polyadenylation element-binding protein (CPEB) seems to be triggered by the NMDA (N-methyl-D-aspartate) receptor but not by mGluR⁸⁹. So, CPE-containing mRNAs, which are transported to dendrites by CPEB¹⁰⁴, are more likely to be translated during NMDA receptor-dependent LTP than during mGluR-LTD.

Is there a translational regulatory protein that is controlled specifically during mGluR-LTD? An attractive candidate is the fragile X mental retardation protein (FMRP). FMRP is an mRNA-binding protein that is thought to function as a translational repressor^105,106,107 and to be involved in the transport of dendritic mRNAs^108,109. FMRP and its mRNA are present in dendrites^13,110, and stimulation of mGluRs can result in the rapid synthesis of FMRP in synaptosomes¹³ and in cultured neurons¹¹¹. Dendritic localization of FMRP and its mRNA is altered following stimulation of group I mGluRs but not other glutamate receptors¹⁰⁹. In the hippocampus, FMRP-knockout mice have enhanced mGluR-induced LTD¹¹² but have normal LTP^113,114, indicating that the functions of FMRP are modified specifically in response to mGluR activation. Similar to CPEB, FMRP seems to bind specific mRNAs, many of which encode proteins that are important for synaptic function^{111,115,116,117}.

Box 3 | Elongation: another mechanism for translation regulation during synaptic plasticity

Translation is also regulated at the elongation step. One way to regulate peptide elongation is through the phosphorylation of eukaryotic elongation factor 2 (eEF2). eEF2 is a GTP-binding protein that mediates the translocation of peptidyl transfer RNA (tRNA) from the A site to the P site on the ribosome¹¹⁸. Phosphorylation of eEF2 by eEF2 kinase, a Ca²⁺/calmodulin-dependent enzyme, inhibits eEF2 activity and reduces peptide elongation^119,120,121. The eEF2 kinase is regulated by activation of mammalian target of rapamycin (mTOR), which results in phosphorylation near the calmodulin-binding site of the kinase, thereby decreasing both the calmodulin-binding and kinase activity¹²². This could explain the recent observation that serotonin stimulates rapamycin-sensitive dephosphorylation of eEF2 in Aplysia californica synaptosomes⁸⁰. So, activation of mTOR during protein synthesis-dependent synaptic plasticity might increase not only translation initiation, but also translation elongation.

Is elongation regulated during synaptic plasticity? In intact tadpole tecta¹²³ and cultured cortical neurons¹²⁴, NMDA (N-methyl-D-aspartate) receptor activation increases the phosphorylation of eEF2. Similarly, treatment of rat synaptoneurosomes with NMDA results in increased eEF2 phosphorylation, which is correlated with increased synthesis of α-calcium/calmodulin-dependent protein kinase II (α-CaMKII) but a decrease in overall protein synthesis¹²⁵. Similarly, chemically-induced LTP is associated with an increase in eEF2 phosphorylation, a decrease in overall protein synthesis and a concomitant increase in Arc and Fos protein levels¹²⁶. So, phosphorylation of eEF2 might decrease overall protein synthesis while increasing the translation of specific messenger RNAs (mRNAs). These findings indicate that, although mTOR is necessary for translation initiation in numerous forms of synaptic plasticity, it is not required to increase the rate of peptide-chain elongation. Whether this is true for other forms of protein synthesis-dependent synaptic plasticity remains to be determined, as it has been suggested that an increase in general translation might underlie long-lasting synaptic plasticity in the hippocampus⁶². As discussed in the main text, the increase in specific protein production when general protein synthesis is inhibited could reflect the presence of mRNA-specific regulatory sequences such as upstream open reading frames, the involvement of mRNA regulatory factors or perhaps differences in mRNA localization.

Box 4 | mRNA localization in development and memory

Although this review focuses primarily on translational regulatory mechanisms, synaptic plasticity also depends on the appropriate localization of specific messenger RNAs (mRNAs) to the synapses where signals are recieved. In Drosophila melanogaster, the RNA-binding protein Staufen is necessary for localizing the Oskar mRNA to the posterior pole of the oocyte and for anchoring the bicoid mRNA at the anterior pole of eggs²⁶. Staufen is also required for mRNA localization during development of the Drosophila CNS. staufen mRNA is upregulated in adult flies in memory formation, and staufen mutants have defects in long-term memory¹²⁷. A Staufen homologue in mammalian hippocampal neurons^2,128,129 is a component of the RNA granules that are thought to transport specific mRNAs from the cell body to the dendrite^2,130. Other RNA-binding proteins, including translation factors, have also been identified in these RNA granules¹³⁰, and RNA interference that targets these proteins impairs transport of a calcium/calmodulin-dependent protein kinase II (CaMKII) reporter mRNA in cultured mouse neurons¹³⁰. In flies, RNA-binding proteins such as Staufen are thought to promote translation of properly localized mRNA and to repress translation of those that are 'mislocalized'. So, RNA granules in neurons might repress translation of the target mRNAs during transport from the cell body to the dendrite and specifically enhance their translation at the activated synapse. If this model is correct, it would be important to determine whether staufen mutant mice, like the mutant flies, have deficits in synaptic plasticity and memory and, if so, whether these defects can be linked to premature or mislocalized translation of Staufen target mRNAs.

Similar to Staufen, the Pumilio protein in Drosophila binds to regulatory elements in the 3′-untranslated region of target mRNAs²⁶. Pumilio might recruit extra factors, such as Nanos, to promote deadenylation and repress translation of target mRNAs. The Pumilio mRNA is induced during memory formation in adult flies and pumilio mutants have impaired long-term memory¹²⁷. Although not all the factors that are regulated by Staufen and Pumilio in development have been implicated in long-term memory in flies, the mRNA encoding Orb, a Drosophila homologue of cytoplasmic polyadenylation element-binding protein (CPEB) that cooperates with Staufen to regulate translation of oskar mRNA^26,83, is also upregulated in flies during memory formation¹²⁷. So, CPEB-like proteins might have similar functions in long-term memory formation in both flies and mammals. The defined roles of Staufen, Pumilio and Orb in both axis formation and memory function in Drosophila, and the identification of homologues for all three proteins in mammals, raise the possibility that signalling cascades that govern body pattern formation and memory in flies are conserved in mammals and regulate synaptic plasticity and memory.