Biochemical Journal

Review article

Signalling to translation: how signal transduction pathways control the protein synthetic machinery

Christopher G. Proud


Recent advances in our understanding of both the regulation of components of the translational machinery and the upstream signalling pathways that modulate them have provided important new insights into the mechanisms by which hormones, growth factors, nutrients and cellular energy status control protein synthesis in mammalian cells. The importance of proper control of mRNA translation is strikingly illustrated by the fact that defects in this process or its control are implicated in a number of disease states, such as cancer, tissue hypertrophy and neurodegeneration. Signalling pathways such as those involving mTOR (mammalian target of rapamycin) and mitogen-activated protein kinases modulate the phosphorylation of translation factors, the activities of the protein kinases that act upon them and the association of RNA-binding proteins with specific mRNAs. These effects contribute both to the overall control of protein synthesis (which is linked to cell growth) and to the modulation of the translation or stability of specific mRNAs. However, important questions remain about both the contributions of individual regulatory events to the control of general protein synthesis and the mechanisms by which the translation of specific mRNAs is controlled.

  • mammalian target of rapamycin (mTOR)
  • mRNA
  • mRNA translation
  • ribosome
  • signal transduction
  • translation factor


Regulating the overall rate of protein synthesis is important for modulating cell and tissue metabolism. One important example of this is the anabolic effect of insulin is to activate protein synthesis in tissues such as fat and skeletal muscle. Secondly, cell and tissue growth require enhanced rates of protein synthesis. On the other hand, excessive protein accumulation may lead to tissue hypertrophy, e.g. in the case of cardiac muscle [1] and certain types of benign tumours (such as hamartomas) [2]. Cell proliferation also depends upon maintaining an adequate rate of protein synthesis to allow cells to become large enough to generate two daughter cells, although the links between mammalian cell size and division are not clear-cut (see, e.g., [3]). Thirdly, control of the translational machinery also plays key roles in controlling gene expression. It is therefore not surprising that specific translation factors and the signalling pathways that control them are strongly implicated in the dysregulation of cell function that leads to transformation and cancer. In particular, in many cases, the synthesis of specific proteins involves or depends upon tight control over the translation of their mRNAs. There are now numerous examples of this kind of qualitative control of protein synthesis, a few selected examples of which are discussed here. Conversely, and fourthly, protein synthesis places heavy demands upon the cell in terms of requirements for both (essential) amino acids as precursors and metabolic energy, because protein synthesis consumes a high proportion of cellular ATP and GTP. In view of the above considerations, it is clear that overall protein synthesis must be subject to tight control to match the requirements of the cell/tissue with the supply of metabolic energy and amino acids.

These regulatory mechanisms primarily involve rapid (minutes) changes in the activity or association of components of the translational machinery, which are primarily mediated by changes in the states of phosphorylation, e.g. of translation factors and specific RNA-binding proteins (see Figure 1 for a simplified overview). Over the longer term (hours to days), the control of protein synthesis involves changes in the cellular levels of translation factors and ribosomes, i.e. in the cells' capacity for protein synthesis. The pathways that control translational capacity mesh with those that control of translation factor activity, presumably to ensure co-ordinated control of protein synthesis.

Figure 1 Overview of the control of protein synthesis by signal transduction pathways

Signalling pathways activated by receptors at the plasma membrane regulate translation factors required for general translation and thereby modulate overall protein synthesis. They may also regulate certain translation factors and mRNA-specific binding proteins to alter the rates of translation (or the stability) of specific mRNAs and thus control translation in an mRNA-specific manner. The availabilities of metabolic energy and of amino acids also provide important inputs to the signalling processes that control mRNA translation. This depiction is purely schematic and not exhaustive. A more detailed scheme is presented in Figure 8.

Recently, there have been substantial advances in our understanding of both the control of the translational apparatus and the signal transduction pathways involved. The aim of this article is to review these topics and to highlight some key questions that remain to be answered. A complete picture of the control of protein synthesis is critical for our understanding of the regulation of cell growth and proliferation.


The process of protein synthesis or mRNA translation is conventionally divided into three main stages: initiation, elongation and termination. Each stage requires translation factors that transiently associate with the ribosome. This review is concerned exclusively with the regulation of translation in mammalian cells, where these factors are termed eukaryotic initiation, elongation and termination (release) factors (eIFs, eEFs and eRFs). The purpose of this review is not to describe in detail the functions of individual translation factors (for reviews on this subject, see, e.g., [4,5]), but rather to describe how intracellular signalling pathways control them to modulate specific steps in mRNA translation.

This article thus focuses primarily on the translation factors that are known to be subject to direct regulation, and a number of aspects of these proteins are summarized in Table 1. In structuring this article, I will start by providing an overview of the roles of the relevant translation factors and other translational components. I will then outline the main signalling pathways that are known to regulate them, and subsequently try to integrate these two strands to provide an overall picture of how signalling pathways can operate to modulate mRNA translation in mammalian cells.

View this table:
Table 1 Regulatory phosphorylation events that impinge upon mammalian translation factors and related proteins

Listed are phosphorylation sites in components of the mammalian translational machinery that are known to be regulated and/or to affect the activity of the corresponding proteins. The Table is not intended to be an exhaustive list of all the phosphorylation that have identified in translation factors.

eIF2/2B: recruitment of Met-tRNAi (initiator methionyl-tRNA)

The initiation stage of translation involves the recruitment of the 40 S ribosomal subunit to the mRNA and the identification of the start codon. Initiation is a major site for control of mRNA translation. Several major events are involved in this process, including the recruitment of Met-tRNAi to the 40 S subunit (Figures 2i and 2ii). This requires eIF2, a heterotrimeric GTP-binding protein which binds Met-tRNAi only in its GTP-bound state. The GTP is hydrolysed during the initiation process and eIF2 leaves the 40 S subunit as inactive eIF2•GDP. Regeneration of active eIF2•GTP requires a GEF (guanine-nucleotide-exchange factor) called eIF2B (Figure 3), a heteropentamer (α–ϵ) [4]. eIF2Bϵ contains the catalytic domain towards its C-terminus [6] and undergoes phosphorylation at a number of sites [7,8]. The extreme C-terminus of eIF2Bϵ binds eIF2 and contains two phosphorylation sites for protein kinase CK2 that are required for this interaction [7].

Figure 2 Initiation pathway for mRNA translation

This scheme only indicates the major steps, i.e. (i) binding of eIF4F and other factors to the cap, mediated by eIF4E and its scaffold partner, eIF4G, which also binds the eIF4E kinases, the Mnks, (ii) recruitment of Met-tRNAi, mediated by eIF2, (iii) scanning and subsequent start codon recognition, which is followed (iv) by release of initiation factors, (v) the addition of the 60 S subunit, and (vi) the commencement of elongation. This Figure is not intended to be detailed or complete. m7GTP, 7-methylguanosine triphosphate. To see an animated version of this Figure, go to

Figure 3 Targets for PI3K signalling include eIF2B and TSC2

eIF2B catalyses GDP/GTP exchange on eIF2, a heterotrimer of which the γ-subunit is the guanine-nucleotide-binding component. eIF2B is a heteropentamer (α–ϵ). For simplicity, only the catalytic, ϵ, subunit is depicted here. eIF2Bϵ is a phosphoprotein. Only two of the six known sites are indicated: the GSK3 site (phosphorylation of which inhibits the activity of eIF2B) and the priming site required by GSK3, which is phosphorylated by members of the DYRK family, at least in vitro. Residue numbering for eIF2Bϵ is based on the human protein. TSC2 function appears to be inhibited following its phosphorylation by PKB, and may be activated by GSK3. See also Figure 5.

eIF2α is subject to regulatory phosphorylation (at Ser51; Table 1). eIF2(αP) acts as a competitive inhibitor of eIF2B, thus blocking recycling of eIF2 and inhibiting translation initiation (Figure 3). Four eIF2α kinases are known in mammalian cells. They share strong similarity in their catalytic domains, but possess distinct regulatory domains which confer control generally under ‘stressful’ conditions [9]. As these kinases are not known to be regulated by signalling pathways, they will not be discussed in detail here (for recent reviews, see [9,10]).

The fact that all four kinases ultimately control the activity of eIF2B suggests that regulating eIF2B is of key importance for the control of translation. This is borne out by the findings: (i) that overexpression of the catalytic subunit of eIF2B (ϵ) leads to the growth of neonatal cardiomyocytes [11], a process that is driven by increased rates of protein synthesis [1], and (ii) that eIF2B is activated by a diverse range of stimuli that turn on mRNA translation, including insulin and growth factors [1214]. Such activation occurs without discernible decreases in the phosphorylation of eIF2α, indicating that additional mechanisms control eIF2B activity. As mentioned above, eIF2Bϵ is a phosphoprotein [7,15], and phosphorylation modulates its activity [see also under PI3K (phosphoinositide 3-kinase) below] (Figure 3).

An interesting feature of eIF2B is that the severe inherited neurodegenerative disease VWM (leucoencephalopathy with vanishing white matter), a demyelinating disorder also termed childhood ataxia with CNS (central nervous system) hypomyelination, is caused by mutations in the genes for eIF2B [16,17]. Mutations in any of the five genes (α–ϵ; also termed EIF2B1EIF2B5) can cause VWM [18]. The phenotypes observed in VWM are highly variable in terms of severity, age of onset and the range of lesions observed [17,18]. Although it is now clear that VWM mutations impair eIF2B function in a variety of ways [19,20], the actual basis of the disease is unclear. These findings emphasize the crucial importance of the normal function or control of eIF2B for normal cell physiology.

Recruitment of the ribosome to the mRNA

eIF4E binds the 5′-cap structure of the mRNA, which includes a 7-methylguanosine moiety (Figure 2i). eIF4E contains a single site of phosphorylation (Ser209 in mammals; Table 1). Early work suggested that phosphorylation of eIF4E enhanced its binding to cap analogues or capped RNA [21]. Later reports, using fully defined components, showed that phosphorylation actually decreases its affinity for 7-methylguanosine or capped RNA [22,23].

eIF4E also binds a number of protein partners including eIF4G, a multidomain scaffold protein which interacts with other components of the translational machinery (Figures 2 and 4A). These include eIF4A, an RNA helicase that is thought to unwind regions of secondary structure in the 5′-UTRs (untranslated regions) of mRNAs. As such features can inhibit normal cap-dependent translation initiation (depicted in Figure 2iii), increased binding of eIF4A to eIF4G/eIF4E should stimulate this process, especially for mRNAs whose 5′-UTRs contain significant secondary structure. The helicase function of eIF4A is enhanced by eIF4B, which interacts with both eIF4A and eIF3 [24]. eIF4B and eIF4G are both phosphoproteins, as discussed below. The eIF4A–eIF4E–eIF4G complex is often referred to as eIF4F [24] (Figure 4A).

Figure 4 Targets for mTOR signalling

(A) Control of the eIF4F initiation factor complex. Binding of 4E-BPs (e.g. 4E-BP1) to eIF4E occludes their binding site for eIF4E, thus preventing the formation of productive eIF4F initiation factor complexes. mTOR-regulated phosphorylation of 4E-BP1, which occurs at multiple sites (see the text and B) results in the release of 4E-BP1 from eIF4E, allowing eIF4G to bind and recruit other translation factors, such as the helicase eIF4A, to the 5′-end of the mRNA. eIF4G also binds PABP and the eIF4E kinases, the Mnks. (B) 4E-BP1 undergoes phosphorylation at multiple sites in vivo. At least four of them are regulated by mTORC1 in response to amino acids and/or stimuli such as insulin. The N-terminal sites depend on the N-terminal RAIP motif, whereas Ser65 and Thr70 depend both on the phosphorylation of these sites and on the TOS motif. These latter sites are adjacent to the eIF4E-binding motif and control binding to eIF4E (see the text), which occurs through the indicated consensus motif. Less is known about the function or regulation of the other sites. (C) eEF2 kinase. The CaM-binding domain and the unusual catalytic (kinase) domain are shown, as well as the C-terminal region that is required for efficient phosphorylation of eEF2. eEF2 kinase undergoes phosphorylation in vivo at multiple sites, a selection of which is shown. Where known, the effect of these phosphorylation events on eEF2 kinase activity is indicated (orange circles, inhibition; green hexagons, activation) as is the identity of the kinase responsible. In some cases (???), the corresponding kinase is not known or additional kinases to that shown are involved. (D) S6Ks: the overall structural organiziation of S6K 1 and 2 is similar. Each exists as two splice variants, the longer of which contains an N-terminal extension that includes a nuclear- or nucleolar-targeting signal. C-terminal to the kinase domain lie a number of sites whose phosphorylation is essential for full kinase activity: one of these (Thr412) can be phosphorylated by mTOR in vitro. Residue numbering is based upon the longer form of human S6K1, since S6K1 has been studied in much more detail than S6K2. Almost all of the regulatory phosphorylation sites identified in S6K1 are conserved in S6K2.

The N-terminus of eIF4G binds the PABP [poly(A)-binding protein], leading to circularization of the mRNA and enhancement of its translation. eIF4G also binds to eIF3 [26]. Mammalian eIF3 possesses 12 subunits and binds to the ribosome's 40 S subunit. It thus provides a key link between the mRNA (which binds via eIF4E, PABP and other interactions to eIF4G and its partners) and the 40S subunit.

The biological activity of eIF4E can be regulated by a set of 4E-BPs (eIF4E-binding proteins). Three 4E-BPs exist in mammals, although only 4E-BP1 has been studied in much detail. The 4E-BPs interact with eIF4E [27] via a binding motif which resembles those found in eIF4GI and eIF4GII, i.e. with the sequence ΦXXXXΦΦ, where Φ is a hydrophobic residue and X is any amino acid. By binding eIF4E, the 4E-BPs block its interaction with eIF4G, thereby making eIF4E unavailable for initiation complex formation [24,27]. The 4E-BPs are phosphoproteins. 4E-BP1 contains at least seven sites of phosphorylation (Figure 4B and Table 1), of which four are known to be regulated via signalling pathways (Thr37, Thr46, Ser65 and Thr70 in human 4E-BP1).

Elongation factors

In mammalian cells, peptide-chain elongation requires two eEFs. eEF1 binds GTP and recruits aminoacyl-tRNAs to the A-site of the ribosome to match the codon located in that site. It is actually eEF1A that does this, whereas eEF1B (a heterotrimer) acts as the GEF for eEF1A, as eIF2B does for eIF2. eEF1A and subunits of eEF1B are targets for phosphorylation by several kinases [28]. These include PKA (protein kinase A), PKC (protein kinase C) and protein kinases involved in cell cycle control [notably CDK1 (cyclin-dependent kinase 1)/cyclin B], as described in detail in [28] (Table 1). It remains to be established what effects these modifications have on elongation rates, although there is evidence that elongation is controlled during mitosis and that eEF1B localization changes during the cell cycle [28].

eEF2 is a monomer that binds GTP. It mediates the ‘translocation step’ of elongation where the ribosome moves by one codon relative to the mRNA and the peptidyl-tRNA moves from the A- to the P-site [29]. eEF2 is a phosphoprotein. In vivo, phosphorylation occurs at a single site (Thr56 [30]) and this impairs the interaction of eEF2 with the ribosome, thus inactivating it [31]. Phosphorylation is catalysed by an unusual and highly specific enzyme, eEF2 kinase (Figure 4C). Its catalytic domain shows no sequence similarity to the main protein kinase superfamily and it belongs to a distinct small family of protein kinases have been termed ‘α-kinases’ [32,33]. The three-dimensional structure of one member of this family [the cytoplasmic domain of the TRP (transient receptor potential) channel Chak1] has been solved crystallographically [34]. Despite the lack of sequence similarity between Chak1 and other protein kinases, the three-dimensional structure revealed that the binding between Chak1 and the nucleotide substrate probably includes similar interactions to those seen in conventional protein kinases.

eEF2 kinase is normally completely dependent upon Ca2+ and CaM (calmodulin) for activity (reviewed in [35]). The CaM-binding site is situated immediately N-terminal to the catalytic domain. The C-terminal region of eEF2 kinase contains several phosphorylation sites, as discussed below. Only one will be dealt with in this section, as it is phosphorylated by PKA (Ser500 in human eEF2 kinase [36,37]; Figure 4C), which will not be discussed anywhere else in this review. Phosphorylation of eEF2 kinase at Ser500 leads to acquisition of Ca2+-independent activity [36,37]. At resting levels of Ca2+, this is likely to correspond to kinase activation, leading to increased phosphorylation and inactivation of eEF2 [38,39]. Bearing in mind that the overwhelming majority of the energy used in translation (>99%) is consumed during elongation, it is tempting to speculate that this may allow agents that activate PKA to slow down elongation and conserve energy which can be diverted for other requirements, e.g. in adipocytes stimulated by β-adrenergic agonists [38].

The ability of Ca2+ ions to activate eEF2 kinase and inhibit eEF2 may allow neural signals to inhibit the rate of elongation to decrease energy utilization in contracting muscle and to complement their ability to turn on contraction, which requires energy. The activation of eEF2 kinase by Ca2+ is also proposed to play a role in ‘reprogramming’ translation in neuronal cells in response to neurotransmitter stimulation [40,41]. It is argued that, by slowing elongation and making this, rather than initiation, limiting for translation, mRNAs that normally initiate translation inefficiently would be more easily able to enter polysomes and be translated. Similar effects are seen with cycloheximide, a reagent that blocks elongation [42]. Other neuromodulatory stimuli such as 5-hydroxytryptamine (serotonin) and BDNF (brain-derived neurotrophic factor) decrease eEF2 phosphorylation [43,44]. Given the growing body of evidence that controlling translation is important in learning, memory and behaviour [45,46], it will be important to study the role of regulating elongation in these processes.

Ribosomal protein phosphorylation

Certain ribosomal proteins are also phosphorylated within cells. The best-known of these is S6, a component of the 40 S subunit. In mammals, there are two genes for S6Ks (S6 kinases), S6K1 and S6K2 (Figure 4D). Each gives rise to two transcripts and two distinct polypeptides, which differ at their N-termini by a sequence containing a potential nuclear localization signal [47]. The activation of the S6Ks involves their phosphorylation at multiple sites (Figure 4D and Table 1).

In addition to S6, which is discussed in detail below, other substrates for the S6Ks have also been identified. Within the realm of mRNA translation, these include eIF4B [48,49] and eEF2 kinase [50]. In both cases, the effect of S6K-mediated phosphorylation favours activation of translation: for eIF4B, it increases its association with eIF3 [48,51] and it causes inactivation of eEF2 kinase [50], providing a link between mTOR (mammalian target of rapamycin; see below) and the control of eEF2.

Regulation of the translation of specific mRNAs

There are far too many examples of the translational control of specific mRNAs to provide comprehensive coverage here. Such control allows cells rapidly to adjust the rate of synthesis of specific proteins in response to changing conditions. Enhancing the translation of a specific mRNA rapidly steps up production of the corresponding protein, without the need to turn on transcription, process the mRNA and export it to the cytoplasm. There are, moreover, many examples of translational control being exploited to achieve spatial or temporal control of protein expression in early development [52]. Within this review, only two classes of mRNAs that are subject to specific translational control will be discussed: i.e. the 5′-TOP (5′-tract of oligopyrimidines) mRNAs and mRNAs containing AREs (AU-rich elements) in their 3′-UTRs.



The lipid kinase PI3K plays a key role in signalling downstream of a wide variety of receptors, including those for insulin and the insulin-like growth factors. Activation of PI3K leads to increased production of phosphatidylinositol 3,4-bisphosphate, which activates effectors such as PKB (protein kinase B, also termed Akt) [53] (Figure 3). The action of PI3K is opposed by the lipid phosphatase PTEN (phosphatase and tensin homologue deleted on chromosome 10), a tumour suppressor. Mutations in PTEN occur in many human tumours [54], leading to constitutive activation of PKB and, consequently, of mTOR signalling (see below).

Regulation of eIF2B via PKB and GSK3 (glycogen synthase kinase 3)

One substrate for PKB is GSK3, which is named for its role in controlling the rate-limiting enzyme of glycogen deposition, but which also plays diverse roles in other important cellular processes, including development, gene transcription and the organization of the microtubular cytoskeleton [55,56]. PKB phosphorylates both GSK3 isoforms (α and β) at their N-termini, inhibiting their activity against certain substrates, including eIF2Bϵ [8,55] (Ser540 in human eIF2Bϵ; Figure 3 and Table 1). Such substrates are ones whose phosphorylation by GSK3 requires a ‘priming’ phosphorylation event at a serine residue four residues C-terminal to the GSK3 target site. In the case of eIF2B, the ‘priming kinases’ may be members of the DYRK (dual-specificity tyrosine-phosphorylated and -regulated kinase) group of enzymes [57] (Figure 3). Both the GSK3 and priming sites are found in Drosophila eIF2Bϵ, but not in the orthologues from yeast [58]. The phosphorylation of eIF2B by GSK3 inhibits its activity [15]. The inactivation of GSK3 by insulin thus leads to dephosphorylation of the inhibitory GSK3 site on eIF2Bϵ and to activation of eIF2B. This provides a mechanism whereby agents such as insulin can activate eIF2B, a key component for general translation (Figure 3). It has also been reported that eIF2B may play a role in the anti- and pro-apoptotic effects of signalling through PKB/Akt and GSK3 respectively [59]. However, the mechanisms remain unclear; it is tempting to speculate that inhibition of eIF2B may promote the translation of proteins that play a pro-apoptotic role (or vice versa). Inhibition of eIF2B activity via phosphorylation of eIF2α does lead to the specific up-regulation of the translation of certain mRNAs, such as those for GCN4 (general control non-derepressible 4) in yeast and for ATF4 (activating transcription factor 4) in mammals [4,60].

Although, such as by insulin, activating PKB and inhibiting GSK3 elicits the dephosphorylation of the GSK3 site in eIF2Bϵ, the dephosphorylation of this site seems to be insufficient, by itself, to evoke activation of eIF2B [61]. This implies that other regulatory inputs may be important. There are reports of other inputs to eIF2B from amino acids [62] and from ERK (extracellular-signal-regulated kinase) signalling [63,64]. The input from ERK to the control of eIF2B appears to be mediated via activation of PP1 (protein phosphatase 1), although it is not known which phosphorylation site in eIF2B is involved [64,65]. Further work is necessary both to determine the molecular basis of these inputs and to define the importance of the control of eIF2B for the overall activation of protein synthesis.

The N-terminal sites in GSK3 can also be phosphorylated by p90RSK which lies downstream of ERK signalling (see below). Although this could provide a mechanism by which Ras/ERK signalling could activate eIF2B, the available data suggest that the activation of eIF2B via this pathway does not involve changes in the phosphorylation of the GSK3 site in eIF2Bϵ [63] and seems, instead, to be mediated by changes in the activity of a form of PP1 [65].

mTOR, a ‘master regulator’ of the translational machinery

mTOR plays a key role in regulating the phosphorylation of a number of components of the translational machinery. mTOR is a multidomain protein which interacts with a number of other proteins (Figure 5). Its N-terminus contains several HEAT (Huntingtin, elongation factor 3, the PR65/A subunit of protein phosphatase 2A and the lipid kinase Tor) domains (modules for protein–protein interactions) and its C-terminus includes a kinase domain similar to the PI3Ks, and thus to other PI3K-related kinases, such as ATM (ataxia telangiectasia mutated), ATR (ATM- and Rad3-related) and the DNA-activated protein kinase [66]. Immediately N-terminal to its kinase domain is a region that binds the immunophilin FKBP12 (FK506-binding protein 12) when that protein is bound to rapamycin [67]. Rapamycin/FKBP12 inhibits some, but by no means all, of the functions of mTOR.

Figure 5 mTOR complexes

The mTORC1 and mTORC2 complexes are shown schematically. Some other proteins have been reported to bind mTOR, but are not shown here for simplicity. mTORC2 phosphorylates PKB, but little is known about its regulation. mTORC1 phosphorylates S6Ks and 4E-BPs in vitro (although additional kinases may be involved in their phosphorylation in vivo). The kinase activity of mTOR has been shown to be stimulated, in vitro, by Rheb•GTP. The hydrolysis of Rheb-bound GTP to GDP is stimulated by TSC2, which, together with TSC1, negatively regulates mTORC1. TSC2 can be phosphorylated at multiple sites by the indicated kinases. The functional effects of these phosphorylation events remain to be clarified, but they are presumed to activate (green hexagons) or inhibit (orange circles) TSC2 function. An unfilled circle indicates uncertainty about the direction of the effect. Residue numbers are based on human TSC2; owing to the existence of distinct splice variants, numbering systems vary, so that numbers used here do not correspond to those used by certain authors. Signalling through mTORC1 is also activated by leucine, by an unknown mechanism. This does not require TSC2.

In vitro, mTOR exhibits protein kinase activity. This activity is higher in the presence of high (non-physiological) concentrations of Mn2+ than in the presence of physiological levels of Mg2+. In vitro, mTOR can phosphorylate proteins such as 4E-BP1 and S6K1 at sites that are phosphorylated in an mTOR-dependent manner in vivo [68,69]. Interestingly, the phosphorylation of Thr37/Thr46 in 4E-BP1 is relatively insensitive to rapamycin in living cells [70] and is also only partially (<30%) inhibited by rapamycin in vitro (in kinase reactions using immunoprecipitated mTOR) [71].

The phosphorylation of 4E-BP1 and S6K1 is stimulated by serum, insulin and growth factors. In many cell types, this requires that cells are provided with amino acids; amino acid starvation causes a swift impairment of mTOR signalling, as indicated by dephosphorylation of 4E-BP1 and S6K1, and agents such as insulin can no longer (fully) induce their phosphorylation. The most effective single amino acid in most cases is leucine, an essential branched-chain amino acid [72]. It is likely that leucine acts within the cell and it seems that its metabolism is not required. However, it is not yet clear how it activates mTOR signalling (Figure 5).

mTOR forms two types of multisubunit complex

A major advance in recent years has been the discovery that mTOR (and its orthologues in species as evolutionarily distant as budding yeast) form at least two types of complex involving distinct partner proteins (Figure 5). This topic has been the subject of several recent articles [67,73], and will not be dealt with in detail here. The key point is that at least two distinct kinds of complex exist in mammals. One complex, mTORC (mTOR complex) 1 contains the partner raptor (regulatory associated protein of mTOR; termed KOG1 in yeast); the second one, mTORC2, contains a different partner, rictor (rapamycin-insensitive companion of mTOR; mAVO3). Both mTORC1 and mTORC2 contain the protein GβL (also termed mLst8). mTORC1 mediates effects of mTOR that are sensitive to rapamycin, whereas effects mediated through mTORC2 are insensitive to this compound. The reasons for this difference, and indeed the mechanisms by which rapamycin impairs functions of mTORC1, are still not completely clear. Some studies show that treatment of cells with rapamycin destabilizes the mTOR–raptor (mTORC1) complex (see, e.g., [74]), but other investigators have found that raptor could be purified with mTOR as a complex with FKBP12/rapamycin [75], which runs contrary to the idea that rapamycin causes dissociation of this complex.

Raptor interacts with the TOS (target of rapamycin signalling) motifs in targets of mTOR that are regulated in a rapamycin-sensitive manner. This presence of the TOS motif and thus, one assumes, its binding to raptor enhances the ability of mTORC1 to phosphorylate proteins 4E-BP1 and S6K1 in vitro [76,77]. Indeed, the presence of raptor in mTOR immunoprecipitates greatly enhances the phosphorylation of 4E-BP1 by mTOR in vitro [76].

Effects due to mTORC2 include the phosphorylation of the ‘hydrophobic motif’ site in PKB [7880] [which is equivalent to the mTORC1 site (Thr389) in S6K1]. Since it is mTORC1, not mTORC2, that is linked to the control of components of the translational machinery, the remaining discussion of mTOR will focus on mTORC1.

Upstream control of mTORC1

How do agents such as insulin activate mTORC1 signalling? The generally accepted model for this is depicted in Figures 3 and 4. By activating PKB, insulin elicits the phosphorylation of TSC2 (tuberous sclerosis complex 2), one component of a dimeric complex also containing TSC1 (Figure 5). TSC2 acts as a GAP (GTPase-activating protein) for the small G-protein Rheb [81,82]. Rheb interacts with mTOR [via the N-terminal (ATP-binding region) of its kinase domain]. Unusually for a G-protein, Rheb interacts with mTOR irrespective of its guanine-nucleotide-binding status [83,84]; however, only Rheb•GTP was found to stimulate the in vitro kinase activity of mTOR [83]. These findings have led to the concept (Figures 3 and 5) whereby the PKB-mediated phosphorylation of TSC2 leads to the inhibition of its GAP activity towards Rheb, allowing Rheb to accumulate in its GTP-bound state, leading to the activation of mTORC1 (Figure 5). However, there is a lack of direct evidence that phosphorylation of TSC2 inhibits its GAP activity. Alternatively, phosphorylation of TSC2 may alter its stability, its association with TSC1 or its intracellular localization [85,86].

As described above, loss of PTEN function causes activation of PKB and thus stimulation of mTORC1 signalling. Interestingly, the proliferation of some tumour-derived cell lines (e.g. ones lacking the tumour suppressor PTEN) is extremely sensitive to inhibition by rapamycin, implying a key role for signalling through mTORC1 [87]. Thus activation of mTORC1 resulting from constitutive PKB activation leads to a transformed phenotype. This contrasts with the effect of loss of TSC1–TSC2 function, which also results in activation of mTOR, but (usually) leads to non-malignant tumours characterized by very large cell size, termed hamartomas [88]. This difference probably reflects additional roles that PKB/Akt plays in cell survival and other processes [89]. Activation of mTORC1 is also linked to a range of other oncogenes or proto-oncogenes, including PI3K, Akt, Ras, NF (nuclear factor) and LKB1 (reviewed in [87,9092]).

In amino-acid-starved cells, the overexpression of Rheb restores mTOR signalling (see, e.g., [93]). This gave rise to the idea that amino acids may, similarly to insulin, enhance the proportion of Rheb in its GTP-bound state, perhaps by interfering with TSC2 function. However, amino acid starvation has, at most, only modest effects on the proportion of Rheb bound to GTP [84,94], and amino acid starvation still impairs mTOR signalling in TSC2-null cells [84]. This indicates that amino acids can work in a TSC2-independent way to regulate mTOR. The results presented by Long et al. [83] indicate that amino acid starvation somehow impairs the interaction between Rheb and mTOR. It is an important priority to discover how amino acids do this.

mTORC1 regulates the phosphorylation of 4E-BP1 and S6K1 at multiple sites

The 4E-BPs and S6Ks each contain a TOS motif which binds the mTORC1 component, raptor. Each of these proteins is subject to phosphorylation at multiple sites, many of which undergo dephosphorylation in response to starvation of cells for amino acids. Phosphorylation of both proteins follows a hierarchy: in the case of 4E-BP1, phosphorylation of Thr37/Thr46 appears to be required for modification of Thr70, following which Ser65 undergoes phosphorylation; phosphorylation at Ser101 is also required for modification at Ser65 [9597]. Ser65 is generally the site that is mostly strongly increased in response to agents such as insulin. Phosphorylation of Thr37/Thr46 does not regulate the binding of 4E-BP1 to eIF4E directly, whereas phosphorylation at Ser65 and Thr70 does, although their individual contributions to the control of eIF4E binding is controversial (see, e.g., [70,97,98]).

Two short amino acid motifs play key roles in modulating 4E-BP1 phosphorylation (Figure 4B). The TOS motif in the extreme C-terminus is required for phosphorylation of Ser65/Thr70. In all three mammalian 4E-BPs, the sequence of the motif is the same, FEMDI (Phe-Glu-Met-Asp-Ile). The S6Ks (see below) also contain TOS motifs, their N-termini, with the related sequences FDIDL (Phe-Asp-Ile-Asp-Leu) (S6K1) or FDLDL (Phe-Asp-Leu-Asp-Leu) (S6K2) [77,99]. The TOS motif is thought to bind directly to the protein raptor, a partner of mTOR (see below also).

Phosphorylation of Thr37/Thr46 does not require the TOS motif, but depends instead on an N-terminal RAIP (Arg-Ala-Ile-Pro) motif [70,100] (Figure 4B). Their phosphorylation is maintained by amino acids (which activate mTOR signalling) and is affected only modestly by agents such as insulin, which generally stimulates phosphorylation at Ser65, and, more variably, Thr70 [70]. Phosphorylation of endogenous 4E-BP1 at Thr37/Thr46 is also rather insensitive to treatment of cells with rapamycin [70], but the evidence available strongly indicates that these sites are, nonetheless, regulated by mTOR [70]. The relative rapamycin-insensitivity of the N-terminal sites in 4E-BP1 (Thr37/Thr46) could indicate that are also targets for mTORC2. However, the available evidence suggests this is not the case: 4E-BP1 is an extremely poor substrate for phosphorylation by mTORC2 in vitro (B. D. Fonseca and C. G. Proud, unpublished work).

The complex hierarchical phosphorylation of 4E-BP1 and its dependence on two distinct motifs leads one to question whether all the mTOR-regulated sites in 4E-BP1 and S6K1 are really direct targets for phosphorylation by mTOR itself.

Other known targets of mTOR signalling, e.g. the S6Ks and eEF2 kinase (see below), lack recognizable RAIP motifs. S6Ks also undergo hierarchical phosphorylation at multiple sites, which leads to their activation [47]. Interestingly, the main site that is phosphorylated by mTOR in vitro, Thr389, appears to be phosphorylated late in the hierarchy. It is the phosphorylation of this site that correlates best with S6K activity [101]. However, the phosphorylation of several other sites is also blocked by rapamycin, raising important questions about the identity of the mTORC1-regulated mechanisms (kinases) that control these modifications in vivo.

What are the physiological roles of the 4E-BPs and S6Ks? Findings from 4E-BP-knockout mice

By binding to eIF4E, 4E-BPs prevent it from interacting with the related scaffold proteins eIF4GI and eIF4GII [27,102] to form initiation factor complexes competent to mediate cap-dependent translation. However, their physiological roles in the normal control of protein synthesis still remain to be fully established. At most, only modest phenotypes have been reported for 4E-BP1-knockout mice [103,104]. The latter paper reported alterations in white and brown adipose tissue and metabolic rate, which were associated with increased expression of PGC-1 (peroxisome-proliferator-activated receptor γ co-activator 1), a transcriptional regulator involved in mitochondrial biogenesis and thus in adaptive thermogenesis. Data from mice lacking 4E-BP2 indicate a role for this protein in synaptic plasticity in the hippocampus; such animals exhibit deficits in learning and memory [105]. Knocking out 4E-BPs affects the normal regulation of eIF4F formation (e.g. knocking out 4E-BP2 increases basal eIF4F levels [104]), implying that proper control of eIF4F is important for normal cellular control, but the mechanisms involved remain obscure. As many tissues express both 4E-BP1 and 4E-BP2 [104], it will be important to study the phenotypes of animals in which both have been knocked out, and ultimately to extend this to 4E-BP3-knockout animals (see the Note added in proof at the end of the Conclusions and perspectives section). The possible mechanisms by which 4E-BP3 is controlled remain to be investigated.

Links between eIF4E, 4E-BPs, mTOR and cancer

There is strong evidence that eIF4E plays an important role in cell transformation and in human cancers [106,107]. For example, overexpressing eIF4E in cell lines can cause their transformation. eIF4E is expressed at high levels in many human cancers, levels of expression correlating positively with the ‘aggression’ of the tumour. Transgenic animal models have confirmed a role for eIF4E in transformation and tumour development [108,109]. eIF4E represses apoptosis [110,111] and targeted inhibition of eIF4E induces rapid cell death [112]. How does eIF4E contribute to transformation and tumour progression? This may, of course, involve its anti-apoptotic properties, but how does it exert these effects? The most obvious way would be by promoting the translation of the mRNAs for proteins with roles in cell-cycle progression (e.g. cyclin D1 [113]), cell transformation (e.g. ornithine decarboxylase [114,115]), tumour vascularization {e.g. VEGF (vascular endothelial growth factor) [116]} or metastasis (e.g. matrix metalloproteinase-9 [117]) [107,118]. Such increased translation could perhaps be due to the increased formation of eIF4F complexes by ‘out-titrating’ the inhibitory 4E-BPs (see below). This is consistent with the ability of overexpression of eIF4G to transform cells [119].

A further link to cell transformation may be the ability of eIF4E to increase the cytoplasmic levels of the mRNA for cyclin D1, a key regulator at the G1/S boundary of the cell cycle [120]. This apparently involves a feature in the 3′-UTR of the cyclin D1 mRNA [121], rather than the cap-binding function of eIF4E, which is thought to allow eIF4E to promote transport of the cyclin D1 mRNA from the nucleus to the cytoplasm. Such a function would require eIF4E to spend some time in the nucleus, and indeed a proportion of the total cellular eIF4E is found in the nuclear fraction [122]. It remains to be established whether cyclin D1 mRNA transport also involves eIF4E in cells where it is expressed at normal levels. Priorities here are (i) to identify additional mRNAs whose translation is up-regulated when eIF4E is overexpressed, and (ii) to identify other messages whose nucleocytoplasmic transport may be mediated by eIF4E. This will, one hopes, cast much-needed new light on the mechanisms that underlie the role of eIF4E in tumorigenesis.

There is currently a high level of interest in the role of mTOR in cancer [87,91,107]. The fact that the availability of eIF4E is controlled by signalling through mTOR provides one probable link between mTOR signalling and tumorigenesis/cell proliferation.

The role of S6 phosphorylation remains unclear

The functional significance of the S6Ks is less clear. No clear phenotypes related to translation have been reported for knockout mice, even from mice lacking S6K1 and S6K2 [123]. The fact that knocking out murine S6K1 [124] or its Drosophila orthologue [125] leads to reduced animal size appears consistent with a role in the positive control of protein synthesis and is consonant with the effects of mutating the phosphorylation sites in mouse S6 [126].

Although S6 phosphorylation and the S6Ks were identified a relatively long time ago, the physiological function of S6 phosphorylation remains obscure. It was thought previously to be important in controlling the translation of a specific set of mRNAs (the 5′-TOP mRNAs). However, it is now clear that this is not the case [123]. Meyuhas and collegues created knock-in mice and cells in which all five regulated sites of phosphorylation within S6 were altered to alanines (S6[5A]) [126]. These studies demonstrated that S6 phosphorylation is not required for control of 5′-TOP mRNA translation. S6[5A] MEFs (mouse embryonic fibroblasts) actually showed higher rates of protein synthesis, but were also smaller in size than wild-type MEFs [126]. Rapamycin treatment decreased the size of wild-type MEFs, but did not affect the size of S6[5A] MEFs [126], indicating that phosphorylation of S6 is likely to play a role in cell size control, although how it does so is unclear.

The major phenotypes reported in S6[5A] mice were decreased circulating insulin levels and enhanced peripheral insulin sensitivity [126]. This appears to result from decreased overall insulin production rather than reduced total β-cell mass. S6[5A] mice show relative glucose intolerance, whereas S6K1−/− mice show enhanced insulin sensitivity [127], apparently due to feedback inhibition by S6K1 at the level of IRS1 (insulin receptor substrate 1)/PI3K [128,129]. The observation that S6[5A] mice also show improved insulin sensitivity suggests that direct phosphorylation of IRS1 by S6K1 is not the primary, or only, mechanism involved here (see [126] for further discussion).

Cardiac hypertrophy involves increased cell (myocyte) size, and is associated with elevated rates of protein synthesis [1]. Rapamycin inhibits, and even reverses, the development of cardiac hypertrophy in vivo, indicating a key role for mTORC1 signalling in this process [130,131]. Perhaps surprisingly, given the data linking them to the control of cell size, knocking out S6K1 and S6K2 in mice did not affect the development of physiological (exercise-induced) or pathological hypertrophy [132].

eEF2 kinase is also regulated by mTOR

Insulin and a number of other agents rapidly induce the dephosphorylation of eEF2 [35]. The insulin-induced dephosphorylation of eEF2 is associated with an accelerated rate of elongation [133]. Insulin causes the inactivation of eEF2 kinase and this involves a decrease in its ability to bind CaM [133,134]. These effects are blocked by rapamycin, identifying eEF2 kinase as a target for regulation by mTOR. However, unlike the S6Ks and 4E-BPs, eEF2 kinase does not contain an identifiable TOS motif, does not interact with raptor and is not a substrate for phosphorylation by mTOR in vitro (E. M. Smith and C. G. Proud, unpublished work). This suggests that other signalling components act as intermediaries to link mTORC1 to the control of eEF2 kinase. Since the control of eEF2 has been the subject of other recent reviews [29,35], discussion will focus on the links between mTORC1 and eEF2 kinase and on recent developments in understanding the control of eEF2 kinase.

Three mTOR-regulated sites of phosphorylation have been identified in eEF2 kinase (Ser78, Ser359 and Ser366; Figure 4C). Ser366 is phosphorylated by S6K1 and by the ERK-activated kinase p90RSK [50,135], and this impairs the activity of eEF2 kinase at low Ca2+ ion concentrations. Phosphorylation at Ser359 also inhibits eEF2 kinase activity (at saturating Ca2+ ion concentrations), and is regulated by amino acids and rapamycin [134,136], suggesting that it is a target for an unidentified mTOR-regulated protein kinase. Most recently, phosphorylation at Ser78 has been shown to inhibit eEF2 kinase by interfering with its ability to bind CaM [134]. This site lies immediately next to the CaM-binding site in eEF2 kinase (Figure 4C). Its phosphorylation is strongly promoted by insulin and blocked by rapamycin or amino acid starvation. Again, it is not yet known which kinase phosphorylates Ser78.

Major goals is this area must be to determine the roles of these three sites (and perhaps others) in the control of eEF2 kinase by mTORC1 and amino acids; to identify the protein kinases that phosphorylate Ser78 and Ser359 in eEF2 kinase, which may have other roles in signalling downstream of mTOR; and to elucidate the importance of the control of eEF2 to overall regulation of protein synthesis under different conditions. An important step toward the last goal is the production of animals and cells that are devoid of eEF2 kinase activity [32].

Stress-activated kinase signalling also impinges on eEF2 kinase [137]. SAPK4 (stress-activated protein kinase 4) [also called p38 MAPK (mitogen-activated protein kinase) δ] phosphorylates Ser359 (which inactivates eEF2 kinase). Ser377 is a potential target for several protein kinases, although phosphorylation here was reported not to affect eEF2 kinase activity under the conditions tested. It is possible that, like phosphorylation at Ser366, it affects calcium sensitivity, not maximal activity. Ser396 is a substrate for several stress-activated kinases in vitro, and this phosphorylation slightly inhibits eEF2 kinase activity.

5′-TOP mRNAs

The mRNAs encoding the cytoplasmic ribosomal proteins (r-proteins) in mammalian cells are characterized by the presence at their extreme 5′-end of a tract of pyrimidine nucleotides [138]. The translation of these mRNAs is subject to rapid regulation. In serum-starved cells, these mRNAs (for example the mRNA for eEF1A) are not associated with large polysomes and a proportion actually runs in non-polysomal fractions on a sucrose density gradient, whereas the remainder appears to be associated with monosomes or disomes [139]. Following stimulation of cells with serum, the mRNA (e.g. eEF1A) moves into large polysomes, indicative of increased efficiency of translation of this message [138,140]. This group of mRNAs also includes certain other components of the translational machinery such as elongation factors eEF1A and eEF2 [141], and PABP [142].

This shift (translational activation) is inhibited by rapamycin [138,140], implying a role for mTORC1 signalling, although the effect is only partial. Amino acid starvation appears to have a greater inhibitory effect on the polysomal association of 5′-TOP mRNAs (such as that for the 60 S subunit protein, L32) [143]. Since amino acids positively regulate mTOR, these data are consistent with a role for mTORC1 in controlling 5′-TOP mRNA translation (Figure 5).

The fact that 5′-TOP mRNA is sensitive to rapamycin had led to the suggestion that their regulation might involve the S6Ks and S6 phosphorylation. With the demise of this model [123,126], it is now a high priority to establish how the translation of the 5′-TOP mRNAs is controlled, and how mTORC1 signalling feeds into this.

It is important to note that mTORC1 signalling also positively regulates rDNA (ribosomal DNA) transcription (reviewed in [144]), suggesting the possibility of co-ordinated regulation of synthesis of the protein and RNA components of ribosomes, and thus of ribosome biogenesis. A role for mTORC1 makes excellent physiological sense, since mTORC1 co-ordinates multiple inputs including those from amino acids (crucial for both ribosome biogenesis and mRNA translation), cellular energy and anabolic, mitogenic and hypertrophic stimuli (all three of these require increased ribosome numbers to promote protein accumulation, cell proliferation or cell growth). Fully elucidating the links between mTORC1 and rDNA transcription/ribosome biogenesis is a major goal in this area. This topic was recently discussed in detail [144].

Control of the translational machinery by cellular energy status

As alluded to already, protein synthesis is an energy-hungry process. The addition of each amino acid requires hydrolysis of the equivalent of four ATP molecules. Estimates of the proportion of the total cellular energy economy devoted to translation vary, and of course this will differ a lot between different cell types, with dividing cells presumably requiring more protein synthesis than post-mitotic primary cells. A high proportion of cellular metabolic energy is used in translation [145]. Almost all will be consumed in the elongation phase. Actually making the components required for translation, e.g. ribosomes, also requires substantial resources.

A number of connections between cellular energy status or hypoxia, which leads to depletion of ATP, have been identified, as recently reviewed in detail [146]. Dennis et al. [147] have argued that the high Km of mTOR for ATP would make it able to ‘sense’ falling ATP levels, resulting in decreased mTOR kinase activity and impairment of translation. However, such a mechanism could only come into play under very severe (non-physiological) conditions of energy depletion where ATP levels fall catastrophically. In contrast, AMPK (AMP-activated protein kinase) is a much more sensitive system for reacting to changing adenine nucleotide levels. AMPK is activated by AMP (normal 5′-AMP, not cAMP) which is produced when ATP levels fall due to the ‘salvaging’ action of adenylate kinase: a small fall in ATP levels actually causes a proportionally much larger increase in AMP, so this energy-sensing mechanism is very sensitive to changes in NTP levels. AMPK is a key component in the cellular ‘energy- management’ machinery [148]. It is therefore not surprising that AMPK is linked to the control of the translational machinery, although these links were only elucidated quite recently.

AMPK regulates mTORC1

AMPK phosphorylates TSC2 and turns off mTORC1 signalling, presumably because AMPK-mediated phosphorylation of TSC2 stimulates its GAP activity towards Rheb [149]. Other potential negative regulators of mTORC1 in response to hypoxia include the stress-induced proteins RTP801 and RTP801L (also termed Redd1 and Redd2), which inhibit mTORC1 signalling by acting upstream of TSC1–TSC2 [150,151].

Recent data point to convergence between Wnt signalling via GSK3 and AMPK in the control of TSC2/mTORC1 [152,153]. Phosphorylation of TSC2 by AMPK provides the priming phosphorylation event that is required to allow GSK3 to phosphorylate TSC2 at two sites. Mutation of these sites to residues that cannot be phosphorylated sensitized cells to apoptosis induced by glucose (i.e. energy) deprivation. AMPK would thus provide an essential input to allow control of TSC2 by GSK3. As described above, insulin and other agents inhibit GSK3 via phosphorylation mediated, for example, by PKB. It remains to be established what role the inactivation of GSK3 by, for example, insulin plays in the control of mTORC1 signalling alongside the direct phosphorylation of TSC2 by PKB. GSK3 is also inactivated by Wnt signalling, albeit by a different mechanism than its control by insulin. Wnt signalling plays a key role in many cellular processes, including survival, proliferation and, above all, development. The recent data show that Wnt signalling activates mTORC1 through GSK3 and TSC2. The input from GSK3 will require sufficient AMPK activity to ensure that the priming site is phosphorylated. Given the strong links between Wnt signalling and cancer, these data suggest that intervening in mTORC1 signalling may be beneficial in these types of cancers, in addition to others mentioned above.

AMPK also phosphorylates eEF2 kinase

A second link involves the direct phosphorylation of eEF2 kinase by AMPK [154,155] at Ser398 (Figure 4C) which appears to activate eEF2 kinase. Conditions where AMPK is activated are associated with inhibition of elongation and increased phosphorylation of eEF2 [155157]. Since most of the energy required for translation is used during chain elongation, it makes sense to slow down this process when ATP is in short supply. This would also preserve polysomes, allowing translation to resume quickly once cellular energy status has improved. Inhibiting initiation, in contrast, would cause a loss of polysomes, making the resumption of translation more complicated as ribosomes would have to be recruited once more to the mRNA. The ability of AMPK, via TSC2, to impair mTORC1 signalling [149,158] could also lead to increased phosphorylation of eEF2, as mTORC1 negatively regulates eEF2 kinase. The fact that treatment of TSC2-null cells (TSC2−/− MEFs) with agents that activate AMPK still causes increased phosphorylation of eEF2 [84] indicates that the direct control of eEF2 kinase by AMPK is important in controlling the elongation machinery in the absence of a functional TSC2–Rheb–mTORC1 axis. Recent studies suggest that control of eEF2 by cellular energy status plays a role in shutting down protein synthesis under conditions of hypoxia [158160], although other mechanisms, including eIF2α phosphorylation, also contribute to this [161,162]. Data from experiments where eEF2 kinase expression was ablated by RNA interference indicate that it plays an important role in the shut-off of protein synthesis in hypoxic breast epithelial cells [159] and in the cytoprotective effect of AMPK activation in neonatal cardiomyocytes [158].

MK2 (MAPK-activated protein kinase 2), a kinase that is activated by p38 MAPK α/β, also phosphorylates TSC2 [163]. Since MK2 is switched on by cellular stresses and by cytokines, this could provide an input to the control of mTOR, although this remains to be clarified.

ERK and p38 MAPK signalling

Mammalian cells contain several MAPK signalling modules of which the best understood are the classical MAPK (ERK), p38 MAPK α/β and JNK (c-Jun N-terminal kinase) pathways. There are clear links between the first two and the control of the translational machinery. Each involves downstream kinases that phosphorylate components of the translational machinery and/or other proteins that regulate mRNA translation (Figure 6A). ERK activates members of the p90RSK group of kinases (RSK1–RSK4) which, as described below, phosphorylate several translation factors or their regulators. p38 MAPK α/β activate MK-2, which regulates the stability of mRNAs, such as that for TNFα (tumour necrosis factor α), probably via the phosphorylation of ARE-binding proteins (see, e.g., [164167]). Detailed discussion of the regulation of mRNA stability is outside the remit of the present review. However, it is likely that controlling the translation of ARE-containing mRNAs is also important for regulating the expression of proteins that they encode. Recent data suggest that the Mnks (MAPK signal-integrating kinases or MAPK-interacting kinases) may be important in this [168,169]. Controlling mRNA function at post-transcriptional levels, such as mRNA stability or translation (rather than at transcription), allows much faster responses. This is probably important under a wide range of conditions, such as, in the case of TNFα, combating infection.

Figure 6 Links to MAPK signalling modules

(A) There are connections from ERK and p38 MAPK α/β signalling to the translational machinery via the p90RSK group of kinases, which are activated by ERKs; MK2, which is activated by p38 MAPKs; and the Mnks, which are activated by both. ‘MKKKs’ indicates that multiple upstream MAPK kinase kinases may activate MKK3 (MAPK kinase 3) and/or MKK6. (B) Schematic illustration of the Mnks. The two human genes each generate two isoforms, yielding four proteins which differ mainly at their C-termini. The N-termini of Mnk2a/b are longer than those of Mnk1a/b. Major features are shown, including the threonine residues in the activation loop which are phosphorylated by ERK/p38 MAPKs. The C-termini of Mnk1a and Mnk2a are similar (but contain distinct MAPK-binding sites) and differ markedly from those of Mnk1b and Mnk2b. Numbering is based on the human Mnks.


Mnk1 and 2 are activated by phosphorylation by either the classical MAPK, ERK1/2, or by p38 MAPKs α or β [170172] (Figure 6A). They bind through a region in the N-termini to the C-terminus of eIF4G and mediate eIF4E phosphorylation in vivo [173,174].

In human cells, expression of each Mnk gene yields two different mRNAs by alternative splicing. Each pair is identical through the N-terminal and catalytic domains, but encodes distinct C-termini [175177], giving rise to proteins termed Mnk1a/b and Mnk2a/b, where the ‘a’ form is longer (Figure 6B). Mnk1a and Mnk2a each contain a canonical MAPK-binding motif, but this and certain other features are absent from the ‘b’ forms, giving them distinct properties. For example, whereas Mnk1a/2a are almost exclusively cytoplasmic, Mnk1b/2b are also found in the nucleus. Mnk2b has very low activity compared with Mnk2b under all conditions so far tested. Mnk1a and Mnk2a also show very distinct characteristics: Mnk1a has low basal activity, which is markedly activated by stimulation of ERK or p38 MAPK signalling, whereas Mnk2a has high basal activity, which is hardly increased by activation of these pathways and is rather insensitive to their blockade [178,179]. One probable reason for this is that Mnk2a, but not Mnk1a, stably binds the active phosphorylated form of ERK [178] which can thus continuously activate Mnk2a, keeping it in an active form. In fact, when complexed with Mnk2a, phospho-ERK is protected from dephosphorylation by MAPK phosphatases [178]; this may explain why Mnk2a activity is high even in cells where ERK has not been acutely activated or which have been treated with inhibitors of the ERK pathway. In cells which express Mnk2a, the activation of mTOR signalling may promote eIF4E phosphorylation by increasing the association of eIF4E with eIF4G/Mnk2, even in response to agents that do not turn on ERK or p38 MAPK α/β.

The physiological significance of eIF4E phosphorylation in mammals remains unclear, although several possibilities have been discussed [180]. In Drosophila, mutation of the residue corresponding to Ser209 (Ser251 in deIF4E) leads to decreased viability and impaired growth and development [181]. For example, by weakening the interaction of eIF4E with the cap, phosphorylation of eIF4E may allow the eIF4F complex (and associated factors, such as eIF3 and the 40 S subunit) to detach from the 5′-cap during scanning, perhaps accelerating this process or allowing a second initiation complex to bind to the mRNA even before the previous one has located the start codon. This could be especially important for the efficient translation of mRNAs with long 5′-UTRs.

Alternatively, phosphorylation of eIF4E may allow it to diseng-age from mRNAs that are already being translated in order to bind to others that have arrived recently in the cytoplasm or have perhaps been translationally derepressed to allow their translation. It is notable that eIF4E phosphorylation is generally mediated by mitogen- or stress- and cytokine-activated signalling rather than by anabolic signals, such as insulin. The former type of signals may elicit the activation of the translation of new subsets of mRNAs, and, as suggested, eIF4E phosphorylation may contribute to this.

Data from Aplysia californica indicate that the dephosphorylation of eIF4E triggers a switch from cap-dependent to cap-independent translation, which involves an IRES (internal ribosome-entry site) [182]. IRES elements are also found in mRNAs from many viruses that infect mammalian cells, but no role for eIF4E (de)phosphorylation in modulating their translation has yet been reported. However, Mnks have been linked to modulation of the translation of certain viral RNAs, such as HSV-1 (herpes simplex virus-1) (where the use of the Mnk inhibitor CGP57380 revealed a positive role for the Mnks in HSV-1 mRNA translation and viral replication [183]) and adenovirus (where an adenovirus-encoded protein of 100 kDa displaces Mnks from eIF4G, thus impairing eIF4E phosphorylation). This is associated with impairment of host cell translation without inhibiting adenoviral mRNA translation, which is initiated via a cap-independent mechanism (but not via an IRES [184]).

In addition to eIF4E, Mnks also phosphorylate eIF4G ([171], and J. L. Parra-Palau, M. Buxadé and C. G. Proud, unpublished work) (Figure 6A) and a few other proteins, including at least one that binds to the regulatory region of cytokine mRNAs [hnRNP (heterogeneous nuclear ribonucleoprotein) A1] [169].

Regulation of ARE-containing mRNAs by MAPK signalling

Recent data indicate that the Mnks play a key role in the production of the inflammatory cytokine, TNFα [168,169] (Figures 6A and 7). Overexpression of TNFα leads to inflammatory diseases [185], and mice expressing a TNFα mRNA that lacks the AREs develop chronic arthritis and bowel inflammation similar to Crohn's disease [185]. Accordingly, TNFα production is under very tight control which is exerted at multiple levels: transcription, splicing, mRNA stability and mRNA translation [186,187]. Post-transcriptional control of TNFα synthesis is largely mediated through AREs in the 3′-UTR of its mRNA [188]. Indeed, mRNAs for many cytokines, proteins that modulate cell proliferation or growth, and products of immediate early genes contain AREs that modulate mRNA stability or translation.

Figure 7 Control of ARE-containing mRNAs

The coding region (flanked by AUG and stop codons), the 5′-cap (star) and the poly(A) tail are indicated, as are the AREs in the 3′-UTR. ARE-BPs bind to the AREs, inducing destabilization and probably also impairing translation of the mRNA. Certain ARE-BPs are substrates for kinases in downstream of the ERK and especially the p38 MAPK pathways. Phosphorylation of the ARE-BPs may alter their affinity for the mRNA or otherwise remodel the mRNA–protein complexes, facilitating the stabilization of the mRNAs or their translation, thereby promoting production of the encoded protein. Probably the most widely studied such mRNA is that for TNFα.

Signalling through the ERK and p38 MAPK α/β pathways promotes TNFα expression, suggesting a role for these enzymes, or for kinases that are activated by them, in this regulation. Indeed, data from overexpression studies, use of siRNA (small interfering RNA) and of a selective Mnk inhibitor point to a key positive role for Mnks in the production of TNFα in Jurkat T-cells [169]. For example, the Mnk inhibitor CGP57380 or siRNA-mediated knockdown of Mnk1 greatly inhibits TNFα synthesis [168,169]. These inhibitory effects do not seem to be exerted through decreases in the level of the TNFα mRNA, suggesting they may reflect impairment of the translation of the mRNA.

It is now well established that MK2, an enzyme that is activated by p38 MAPK α/β, plays a key role in TNFα production at a post-transcriptional level [189] (Figure 8). Mice lacking MK2 show, for example, a reduced response to endotoxin [189] and resistance to collagen-induced arthritis, a process in which TNFα plays a critical role [190]. The AREs of the 3′-UTRs of the mRNAs for cytokines such as TNFα bind a number of specific proteins [called ARE-BPs (ARE-binding proteins)] [188]. It is now well-established that one of these, TTP (tristetraprolin), is phosphorylated by MK2 [191]. Recent data indicate that, via TTP, MK2 regulates both the stability and the translation of the TNFα mRNA [192]. MK2 also phosphorylates hnRNP A0 (another ARE-BP) [166] and PABP-1 [193].

Figure 8 Detailed overview of the control of protein synthesis by signal transduction pathways

This diagram provides a fuller overview of the links between signalling pathways and the control of the translational machinery, extending the simplified scheme shown in Figure 1. For details, see the text. Steps which are unclear or poorly defined are indicated by question marks. Key regulators of the translational machinery or of the translation of specific mRNAs are boxed. RNA-BP, RNA-binding protein.

The available data suggest that several ARE-BPs may be Mnk substrates, including hnRNP A1, which has been shown to be a physiological Mnk substrate [169,194]. Phosphorylation of hnRNP A1 by Mnks decreases its ability to bind the TNFα mRNA AREs in vitro and the TNFα mRNA in cells [169]. This suggests that hnRNP A1 may impair the expression of TNFα by inhibiting the translation of the TNFα mRNA and/or by mediating its sequestration into stress granules, sites for storage/triage of mRNAs that are not actively being translated [195,196].

Although use of CGP57380 fails to distinguish between Mnk1 and Mnk2, the observations that Mnk1 activity is tightly regulated, whereas that of Mnk2 is not [179], and that siRNA-mediated knockdown of Mnk1 alone markedly impairs TNFα synthesis [169], suggest that Mnk1 is likely to be the more important regulator here. The Mnks and MK2 may be promising new targets for the development of novel anti-inflammatory drugs. In view of the highly unusual nature of the ATP-binding pocket in the Mnks, there may be scope for developing ATP-competitive agents that are Mnk-specific [197].

The RSKs (p90RSKs) and translational control

The protein kinases termed RSKs (or p90RSKs) are also activated by ERK signalling (but not by p38 MAPKs) [198]. They may provide a link between Ras/Raf/MEK (MAPK/ERK kinase)/ERK signalling and the activation of mTOR, thus providing further examples of the mechanisms by which oncogenes may control mTOR. mTORC1 signalling is activated by signalling through MEK/ERK. RSKs can phosphorylate TSC2 at a C-terminal site, Ser1798 in the most widely used numbering system [199,200] (Figures 5 and 6). As for the phosphorylation of TSC2 by PKB, there is a paucity of direct evidence that RSK-mediated phosphorylation actually impairs TSC2's Rheb-GAP. However, the circumstantial data can be interpreted as indicating that RSK-mediated phosphorylation of TSC2 leads to activation of mTOR. This would allow growth factors and mitogens, and, in cardiomyocytes, hypertrophic α1-adrenergic agonists, to turn on mTORC1 signalling (to the S6Ks, 4E-BPs and eEF2 kinase). This link could of course be mediated by other kinases in the pathway, including ERKs themselves, as suggested by Ma et al. [201]. These authors identified Ser664 as the major site in TSC2 that is phosphorylated by ERK in vitro, and showed that mutation of this site (together with another, Ser540) to alanine increased the ability of TSC2 to suppress mTORC1 signalling in MEFs, Ras-induced transformation and tumorigenesis in a model of TSC-linked tumours. These and other data suggest that phosphorylation of TSC2 by ERK links Ras/Raf/MEK signalling to the control of mTORC1.

However, Rolfe et al. [199] were unable to observe any phosphorylation of TSC2 by ERK2 in vitro. Furthermore, in their quantitative (mass spectrometric) phosphorylation study, Ballif et al. [202] only observed an increase in phosphorylation of one site in TSC2 that matches the minimal consensus for phosphorylation by ERK (Ser/Thr-Pro; Ser664), and the changes here were both modest in magnitude and slow in onset. It thus remains to be firmly established whether activation of mTORC1 signalling via the ERK pathway is mediated through the phosphorylation of TSC2 directly by ERK or by a downstream kinase such as RSK.

p90RSK also phosphorylates at least two other proteins that are involved in translational control. It phosphorylates eEF2 kinase at Ser366, the same site as S6K1, which inhibits eEF2 kinase activity [50]. This may underlie, at least in part, the ability of agents that activate ERK signalling to induce the dephosphorylation of eEF2 [203]. Similarly, p90RSK and S6K1 phosphorylate eIF4B at the same site, Ser422, promoting its association with eIF3 [51].

Bearing in mind that ERK/p90RSK signalling also leads to activation of mTORC1, PI3K/PKB and Ras/ERK signalling clearly converge on a number of components of the translational machinery to promote the assembly of initiation factor complexes and the activation of the elongation machinery. This is likely to be a key element of the abilities of agents that stimulate these pathways to promote cell growth and proliferation. Both pathways lie downstream of proteins that are implicated in tumorigenesis or tumour suppression, e.g. PI3K and PTEN, in one case, and Ras and Raf, in the other.

Other stress conditions

Other stressful conditions impair translation and affect multiple components of the translational machinery [204]. The conditions that are often used as ‘cell stresses’, such as treatment with arsenite or H2O2 (‘oxidative stress’), heat shock or hyperosmolarity probably exert multiple effects. These can include the activation of stress-activated MAPKs, especially p38 MAPK. Stresses can also elicit the stimulation of certain eIF2α kinases, e.g. PERK [PKR (protein kinase R)-like endoplasmic reticulum kinase] and HRI (haem-regulated inhibitor), but that topic lies beyond the scope of the present review.

As described above, p38 MAPKs α/β activate the Mnks (mainly Mnk1a, [176,178,179]). p38 MAPKs are also activated by endotoxin, cytokines and other immune modulators. However, the significance of this for the control of translation is unclear: they may reflect roles in the control of protein synthesis by such stimuli, but this requires detailed study.


This review has described a range of regulatory inputs that impinge on the basal translational machinery, or on the translation of specific messages, as summarized schematically in Figure 8. Indeed, the last few years have seen major advances in our understanding of the control of the translational apparatus and of the signalling pathways, such as mTOR, that modulate it.

It is clear that the initiation and elongation stages of translation are controlled by many regulatory inputs, but their quantitative importance for the overall regulation of protein synthesis remains to be established. This may vary between cell types and between stimuli, and be overlaid with the effects of nutrients and cellular energy status. More complete information on the roles of individual components come from studies on mice in which individual factors have been knocked out (for components that are not essential, e.g. the 4E-BPs) or mutated (knock-ins, for key proteins such as S6 [126] and eIF2α [205]). A particularly exciting aspect is the realization that dysfunction or dysregulation of the translational machinery leads to human diseases as diverse as cancer, tissue hypertrophy, neurodegeneration and inflammation. One hopes that further discoveries related to the mechanisms and control of mRNA translation will help our understanding of these conditions and ultimately open up new therapeutic approaches for them.

Note added in proof (received 26 February 2007)

While this paper was under review, Le Bacquer et al. [216] reported that mice lacking 4E-BP1 and 4E-BP2 show a phenotype involving diet-induced obesity and insulin resistance.


Recent work in the my laboratory on the control of mRNA translation has been supported by the Wellcome Trust, the U.K. Medical Research Council (MRC), the U.K. Biotechnology and Biological Sciences Research Council (BBSRC), the British Heart Foundation (BHF), the European Union (EU), the Signal Transduction Therapy Consortium at the University of Dundee, the Heart and Stroke Foundation of Canada, the Canadian Institutes for Health Research, and Ajinomoto Amino Acid Research Program. I apologize to the authors of those research papers whose original contributions I was unable to cite, owing to space considerations. In many cases, recent review articles have been cited instead.

Abbreviations: AMPK, AMP-activated protein kinase; ARE, AU-rich element; ARE-BP, ARE-binding protein; ATM, ataxia telangiectasia mutated; CaM, calmodulin; CDK, cyclin-dependent kinase; DYRK, dual-specificity tyrosine-phosphorylated and -regulated kinase; eEF, eukaryotic elongation factor; eIF, eukaryotic initiation factor; 4E-BP, eIF4E-binding protein; ERK, extracellular-signal-regulated kinase; FKBP12, FK506-binding protein 12; GAP, GTPase-activating protein; GEF, guanine-nucleotide-exchange factor; GSK3, glycogen synthase kinase 3; hnRNP, heterogeneous nuclear ribonucleoprotein; HRI, haem-regulated inhibitor; IRES, internal ribosome-entry site; IRS1, insulin receptor substrate 1; MAPK, mitogen-activated protein kinase; MEF, mouse embryonic fibroblast; MEK, MAPK/ERK kinase; Met-tRNAi, initiator methionyl-tRNA; MK2, MAPK-activated protein kinase 2; Mnk, MAPK signal-integrating kinase or MAPK-interacting kinase; mTOR, mammalian target of rapamycin; mTORC, mTOR complex; PABP, poly(A)-binding protein; PERK, protein kinase R-like endoplasmic reticulum kinase; PI3K, phosphoinositide 3-kinase; PKA, protein kinase A; PKB, protein kinase B; PKC, protein kinase C; PKR, protein kinase R; PP1, protein phosphatase 1; PTEN, phosphatase and tensin homologue deleted on chromosome 10; raptor, regulatory associated protein of mTOR; rDNA, ribosomal DNA; siRNA, small interfering RNA; S6K, S6 kinase; 5′-TOP, 5′-tract of oligopyrimidines; TNFα, tumour necrosis factor α; TOS, target of rapamycin signalling; TSC, tuberous sclerosis complex; TTP, tristetraprolin; UTR, untranslated region; VWM, leucoencephalopathy with vanishing white matter


View Abstract