mTORC1 and CK2 coordinate ternary and eIF4F complex assembly

Ternary complex (TC) and eIF4F complex assembly are the two major rate-limiting steps in translation initiation regulated by eIF2α phosphorylation and the mTOR/4E-BP pathway, respectively. How TC and eIF4F assembly are coordinated, however, remains largely unknown. We show that mTOR suppresses translation of mRNAs activated under short-term stress wherein TC recycling is attenuated by eIF2α phosphorylation. During acute nutrient or growth factor stimulation, mTORC1 induces eIF2β phosphorylation and recruitment of NCK1 to eIF2, decreases eIF2α phosphorylation and bolsters TC recycling. Accordingly, eIF2β mediates the effect of mTORC1 on protein synthesis and proliferation. In addition, we demonstrate a formerly undocumented role for CK2 in regulation of translation initiation, whereby CK2 stimulates phosphorylation of eIF2β and simultaneously bolsters eIF4F complex assembly via the mTORC1/4E-BP pathway. These findings imply a previously unrecognized mode of translation regulation, whereby mTORC1 and CK2 coordinate TC and eIF4F complex assembly to stimulate cell proliferation.

M essenger RNA (mRNA) translation plays a major role in homeostasis, whereas its dysregulation underpins a variety of pathological states including cancer, metabolic syndrome and neurological disorders 1 . Activation of mRNA translation requires rapid and highly coordinated assembly of the eukaryotic translation initiation factor 4F (eIF4F) complex composed of cap-binding subunit eIF4E, large scaffolding protein eIF4G and DEAD box helicase eIF4A, and the ternary complex (TC) comprised of eIF2, GTP and initiator tRNA (tRNA i Met ) 2 . eIF4F recruits mRNA to the ribosome, whereas TC delivers tRNA i Met (ref. 2). Mammalian/mechanistic target of rapamycin complex 1 (mTORC1) integrates a number of stimuli including nutrients, growth factors and hormones to bolster protein synthesis 3 . mTORC1 phosphorylates and inactivates the eIF4E-binding proteins (4E-BPs), which leads to their dissociation from eIF4E, thereby allowing eIF4E:eIF4G interaction and eIF4F complex assembly 1 . How mTORC1-dependent stimulation of eIF4F assembly is coordinated with TC recycling, however, remains largely underexplored.
eIF2 is a heterotrimer that comprises eIF2a, b and g subunits 2 . After recognition of the start codon by tRNA i Met , eIF2-bound GTP is hydrolyzed to GDP and the TC complex is recycled by the guanine nucleotide exchange factor (GEF) eIF2B, which converts eIF2:GDP to eIF2:GTP 2 . eIF2a phosphorylation, which is induced by eIF2a kinases (protein kinase RNA-like endoplasmic reticulum kinase (PERK), protein kinase RNA-activated (PKR), general control nonderepressible 2 (GCN2) and haem-regulated inhibitor kinase) in response to various types of stress including endoplasmic reticulum stress, amino-acid unavailability, haem deficiency and viral infection, inhibits GEF function of eIF2B, thereby suppressing TC recycling and limiting TC levels 1,2 . This leads to suppression of global protein synthesis, with concomitant increase in translation of mRNAs harbouring inhibitory upstream open reading frames (uORF mRNAs) that encode stress-induced transcriptional regulators such as activating transcription factor 4 (ATF4) and CCAAT-enhancer-binding protein homologous protein (CHOP) 4 . Persistent mTORC1 activation is thought to induce chronic endoplasmic reticulum stress and perturbs AKT signalling, resulting in secondary elevation in eIF2 kinase activity and eIF2a phosphorylation 4 . However, it is largely unknown how mTORC1 affects eIF2a phosphorylation during acute activation of the translational machinery by nutrients, growth factors or hormones (for example, insulin).

mTOR decreases phospho-eIF2a-stimulated translation.
Recently, a transcriptome-wide catalogue of mRNAs whose translation is upregulated after induction of eIF2a phosphorylation by acute endoplasmic reticulum stress (hereafter referred to as 'eIF2a-sensitive' mRNAs) was determined 5 . To investigate the effects of changes in mTOR signalling on translation of 'eIF2asensitive' mRNAs 5 , we used the polysome profiling technique, wherein mRNAs are separated based on the numbers of ribosomes they bind using a sucrose gradient and ultracentrifugation, followed by analysis of the changes in translation and cytosolic mRNA levels on a transcriptome-wide scale 6 . Transcriptome-wide polysome profiling in MCF7 cells revealed that induction of mTOR signalling by 4 h insulin treatment coincides with translational suppression of 'eIF2asensitive' mRNAs 5 , as compared with those whose translation was determined to be independent of eIF2a phosphorylation 5 (background mRNAs; Fig. 1a,b; Po5.2e À 12). In turn, addition of the active-site mTOR inhibitor torin1 (ref. 7) to insulin-treated cells selectively induced translation of 'eIF2a-sensitive' mRNAs ( Fig. 1a,b; Po3.7e À 16). These effects of modulation of mTOR signalling on translation of 'eIF2a-sensitive' mRNAs were not caused by the changes in cytoplasmic mRNA levels (Fig. 1b).
mTOR affects phospho-eIF2b and phospho-eIF2a levels. Insulin induced mTORC1 signalling as illustrated by elevated phosphorylation of 4E-BPs and the S6 kinase (S6K) substrate ribosomal protein S6 (rpS6) as compared with control serum-starved cells, which was reverted by the allosteric mTOR inhibitor rapamycin or active-site mTOR inhibitor (KU-0063794 and torin1) (Fig. 1c; compare lanes 2 and 7 with lanes 3-5 and 8-10, respectively). In addition, insulin decreased phospho-eIF2a levels as compared with control serum-starved cells, and cells stimulated with insulin in the presence of mTOR inhibitors ( Fig. 1c; compare lanes 2 and 7 with lanes 3-5 and 8-10, respectively). In stark contrast to effects of treatments on eIF2a phosphorylation, eIF2b phosphorylation on serine 2 (Ser2) was enhanced by insulin relative to the control, and diminished when cells were stimulated with insulin in the presence of mTOR inhibitors (Fig. 1c; compare lanes 2 and 7 with lanes 3-5 and 8-10, respectively). Phosphorylation of eIF2a stimulates ATF4 mRNA translation and consequently upregulates ATF4 protein levels 8 . Although the effects of insulin and mTOR inhibitors on eIF2a phosphorylation were detected after 30 min, neither insulin nor mTOR inhibitors exerted a major impact on ATF4 protein expression at this time point (Fig. 1c; compare lane 1 with lanes 2-5). Even though these results appear surprising, they are consistent with several studies showing that under a number of experimental conditions that favour eIF2a phosphorylation including ultraviolet irradiation, endoplasmic reticulum and osmotic stress, induction of ATF4 protein expression is either delayed or absent [9][10][11] . Accordingly, insulin downregulated ATF4 protein levels relative to the control, which was reversed by mTOR inhibitors after 4 h ( Fig. 1c; lanes [6][7][8][9][10]. Under these conditions, insulin inhibited ATF4 mRNA translation as illustrated by the shift of ATF4 mRNA towards lighter polysomes, as compared with control serum-starved cells or cells that were stimulated with insulin in the presence of torin1 (Fig. 1d, left panel). In contrast, distribution of b-actin mRNA across the polysomes was essentially unchanged between the treatments (Fig. 1d, right panel). These findings show that under these conditions, mTOR activation leads to inhibition of ATF4, but not b-actin mRNA translation, which is consistent with results obtained from the transcriptome-wide polysome-profiling study (Fig. 1b). Therefore, similarly to the establish selectivity of the stimulatory effects of mTOR on translation of terminal oligopyrimidine (TOP) and long and structured 5 0 -untranslated region (UTR) harbouring mRNAs 12 , mTOR also appears to selectively inhibit translation of mRNAs that are upregulated under conditions wherein eIF2a phosphorylation is increased. Although insulin slightly induced steady-state ATF4 mRNA levels compared with torin1, these effects were not significant (Fig. 1e), thereby indicating that the changes in ATF4 protein expression observed between treatments are largely caused by the changes in translation and not by the upstream steps in the gene expression pathway (for example, transcription or mRNA stability). Therefore, short-term (4 h) inhibition of mTOR signalling correlates with reduction in phospho-eIF2b levels, increase in phospho-eIF2a levels and concomitant translational activation of 'eIF2a-sensitive' mRNAs including ATF4. Since the latter findings were based on a single-cell line, we tested the effects of serum withdrawal and repletion on phosphorylation of eIF2b and eIF2a in the presence of mTOR inhibitors in parallel in HEK293E and MCF7 cells. A 30-min serum stimulation induced mTOR activity in HEK293E and MCF7 cells as demonstrated by increased S6K phosphorylation, which coincided with increased eIF2b phosphorylation and reduced phospho-eIF2a levels relative to the control (Fig. 2a; compare lanes 1 and 2). These effects were largely diminished when cells were stimulated with serum in the presence of rapamycin or torin1 ( Fig. 2a; compare lane 2 versus lanes 3-4). Therefore, observed correlation between changes in mTOR activity and phosphorylation status of eIF2a and eIF2b is not limited to a single-cell line.    (a) Absorbance profiles (254 nm) of cytosolic extracts loaded onto 5-50% sucrose gradients and sedimented by ultracentrifugation. mRNA was isolated from fractions containing 43 ribosomes (box) and analysed in parallel with cytoplasmic mRNA using microarrays (see Methods). Positions of small (40S) and large (60S) ribosome subunits, monosomes (80S) and polysomes in the gradients are indicated. AU, absorbance units. (b) Transcriptome-wide effects of 4-h insulin and insulin þ torin1 treatments on the translatome (polysome associated; upper panel) or steady-state cytoplasmic mRNA levels (lower panel). Presented are mRNAs whose translation is upregulated by eIF2a phosphorylation ('eIF2a sensitive', red curve) and those that are not affected under conditions where eIF2a phosphorylation is stimulated (background; black curve) according to the study of Baird et al. 5 .
Wilcoxon P values contrasting fold changes for eIF2a-regulated to background mRNAs are indicated. The experiment was carried out in four independent replicates. (c) MCF7 cells were treated as in b for the indicated time periods. In addition to torin1, allosteric mTOR inhibitor rapamycin (RAP; 50 nM) and active-site mTOR inhibitor KU-0063794 (KU; 3 mM) were used. Phosphorylation and expression levels of indicated proteins were monitored by western blotting. b-Actin served as a loading control. Experiments were repeated in at least two independent replicates and quantified by densitometry ( Supplementary Fig. 9). (d,e) MCF7 cells were serum starved for 16 h (Starved) and then treated and fractionated as in b. Relative amounts of ATF4 and b-actin mRNA in polysome fractions (d) or cytosolic extracts (for steady-state mRNA measurements) (e) were determined by reverse transcriptionquantitative PCR (RT-qPCR). Position of monosome (80) and polysomal fractions are shown. (d,e) S.d.'s and interaction (treatment and fraction) P values from a two-way analysis of variance (ANOVA) using means of two independent experiments each consisting of technical replicates are indicated.
CK2 phosphorylates eIF2b and stimulates eIF4F assembly. It has been reported that Casein kinase 2 (CK2), which is stimulated by insulin and serum 13,14 phosphorylates eIF2b on serines 2 and 67 (Ser2 and Ser67), and eIF2b phosphorylation at these sites may stimulate global mRNA translation and cell proliferation 15 . We therefore investigated whether CK2 plays a role in the apparent correlation between induction of eIF2b phosphorylation and stimulation of mTOR signalling by serum and insulin. Starved HEK293E cells were serum stimulated in the presence of a vehicle (dimethylsulphoxide, DMSO), torin1 or CK2 inhibitor CX-4945 for 30 min. Both torin1 and CX-4945 reduced serum-induced phosphorylation of eIF2b in HEK293E cells, which was paralleled by increased eIF2a phosphorylation, relative to the control ( Fig. 2b;   . Levels and phosphorylation status of indicated proteins in the pulled down material (25%) and inputs (10%) were determined by western blotting. b-Actin served to exclude contamination of m 7 GTP cap pull downs (for example, non-specific binding to the agarose beads) and as a loading control for inputs. Parallel results were obtained in MCF7 and HCT116 cells ( Supplementary Fig. 1a,b). (d) Expression of HA-CK2a in U2OS cells was induced by tetracycline withdrawal (doxycycline was used at 1.5 mg ml À 1 to suppress expression of CK2 subunits in control cells). Cells were starved for 16 h after which cells were stimulated by 10% serum (FBS) in the presence of a vehicle (DMSO), 250 nM torin1, 50 mM CX-4945 or a combination thereof for the indicated time period. (e) HCT116 PTEN þ / þ or PTEN À / À cells were treated with the indicated concentration of CX-4945 for 1 h in the presence of 10% serum (FBS). (f) Cells described in e were incubated with DMSO or torin1 (250 nM) for 1 h. (d-f) Expression levels and phosphorylation status of indicated proteins were monitored by western blotting. b-Actin served as a loading control. Experiments in this panel were repeated at least two times independently and the representative results are shown. Where appropriate, quantification was performed using densitometry ( Supplementary Fig. 9).
inhibitors reduced 4E-BP1 phosphorylation ( Fig. 2b; compare lane 1 with lanes 2-4), although higher concentration of CX-4945 (50 mM) was required for strong inhibition of eIF2b and 4E-BP1 phosphorylation and induction of eIF2a phosphorylation as compared with rpS6 phosphorylation (15 mM) (Fig. 2b; lane 3 versus 4). Consistent with inhibiting 4E-BP1 phosphorylation, CX-4945 impaired eIF4F assembly as illustrated by reduced eIF4E:eIF4G1 association and increased eIF4E:4E-BP1 binding relative to the control, and this effect was more pronounced when a higher concentration of CX-4945 (that is, 50 mM) was used (Fig. 2c, compare lane 1 with lanes 2-4). Equivalent results were observed in MCF7 cells, whereby torin1 and CX-4945 (15 and 50 mM) reduced eIF2b phosphorylation and disrupted eIF4F complex ( Supplementary Fig. 1a). Moreover, similar effects of 50 mM CX-4945 on the mTORC1 signalling and eIF4F complex assembly were observed in HCT116 cells ( Supplementary  Fig. 1b). Collectively, these findings demonstrate that CK2 regulates not only eIF2b phosphorylation but also eIF4F complex assembly. Next, we sought to determine the relationship between CK2 and mTOR in regulating eIF2b phosphorylation and eIF4F complex assembly. We first employed U2OS cells in which CK2a-HA/Myc-b expression and consequently CK2 activity are increased by removal of doxycycline from the growth media 16 (Fig. 2d). Induction of CK2a-HA/Myc-b expression increased eIF2b phosphorylation, while exhibiting minimal effects on 4E-BP1 and rpS6 phosphorylation, which was expected as these experiments were carried out under full serum conditions wherein mTORC1 activity is high ( Fig. 2d; compare lane 1 versus 2 and 6 versus 7). In turn, CX-4945 decreased phosphorylation of eIF2b as well as 4E-BP1 and rpS6 after 2 and 7 h ( Fig. 2d; compare lane 2 versus 3 and 7 versus 8). Increase in CK2 activity engendered by overexpression of its subunits did not conspicuously affect the ability of torin1 to inhibit phosphorylation of 4E-BP1 and rpS6 ( Fig. 2d; compare lane 2 versus 4 and 7 versus 9). In contrast, the effects of torin1 on eIF2b phosphorylation were attenuated in cells in which CK2a-HA/Myc-b overexpression was induced, whereby torin1 inhibited eIF2b phosphorylation after 7 h but not after 2 h (Fig. 2d; compare lane 2 versus 4 and 7 versus 9). These findings suggest that CK2 directly phosphorylates eIF2b, while stimulating 4E-BP1 phosphorylation via mTORC1, whereby the effects of mTOR inhibitors on eIF2b phosphorylation are mitigated by CK2a-HA/Myc-b overexpression. mTORC1 phosphorylates eIF2b independently of CK2. CK2 phosphorylates PTEN resulting in its inactivation and consequent upregulation of mTOR activity 17,18 . Therefore, to further assess the hierarchy of mTORC1 and CK2 in regulation of eIF2b phosphorylation, isogenic HCT116 PTEN þ / þ and PTEN À / À cells 19 were treated with CX-4945 and torin1. Although the primary CK2 phospho-acceptor sites on PTEN were mapped to Ser370/385 (ref. 17), due to limited availability of corresponding antibodies, we monitored the effects of CX-4945 on phosphorylation of Ser380/Thr382/383, which are also CK2 responsive 20 . In PTEN þ / þ cells, CX-4945 strongly reduced PTEN phosphorylation and mTOR signalling as judged by reduction in rpS6 and 4E-BP1 phosphorylation ( Fig. 2e; compare lane 1 with lanes 2-4), whereas in PTEN À / À cells, there was no apparent effect of CX-4945 on mTOR signalling ( Fig. 2e; compare lane 5 with lanes 6-8). Whereas 1-h CX-4945 treatment reduced eIF2b phosphorylation and induced phospho-eIF2a levels in a dose-dependent manner in PTEN þ / þ cells, these effects of CX-4945 were largely absent in PTEN À / À cells ( Fig. 2e; compare lanes 2-4 with 6-8). CX-4945 however suppressed eIF2b phosphorylation and increased phospho-eIF2a levels after 7 h, which was not accompanied by a reduction in mTOR signalling ( Supplementary Fig. 1c). In contrast to eIF2b phosphorylation, PTEN status did not have a major impact on the effects of CX-4945 on another bona fide CK2 substrate eEF1d inasmuch as CX-4945 inhibited phosphorylation of eEF1d to a comparable extent in both PTEN À / À and PTEN þ / þ cells after 1 h ( Fig. 2e; compare lanes 2-4 to 6-8). Conversely, torin1, which inhibits mTORC1 downstream of PTEN 21 , reduced phosphorylation of 4E-BP1 and rpS6, irrespective of the PTEN status in the cell ( Fig. 2f; lanes 2 and 4 versus lanes 1 and 3). Moreover, in both cell lines, eIF2b phosphorylation was suppressed to a comparable extent and this coincided with increased eIF2a phosphorylation ( Fig. 2f; lanes 2 and 4 versus lanes 1 and 3). In turn, torin1 did not elicit a major effect on the phosphorylation of eEF1d ( Fig. 2f; lanes 1-4). These findings indicate that in the absence of mTOR inhibition, the effects of CK2 on eIF2b are delayed. To further establish the role of mTOR signalling in phosphorylation of eIF2b, we used TSC2 À / À mouse embryonic fibroblasts (MEFs) wherein mTOR is activated by the loss of a negative upstream regulator TSC1/2 (ref. 22). Similarly to cells lacking PTEN, torin1, but not CX-4945, inhibited eIF2b phosphorylation on Ser2 ( Supplementary Fig. 1d,e). These data suggest that analogous to the attenuation of the effects of mTOR inhibition on eIF2b phosphorylation observed in U2OS cells overexpressing CK2 (Fig. 2d), lack of ability of CX-4945 to suppress mTOR mitigates its inhibitory effects on eIF2b phosphorylation ( Fig. 2e; Supplementary Fig. 1e). These results suggest that CK2 and mTOR may collaboratively regulate eIF2b phosphorylation on Ser2. Inability of CK2 inhibitors to downregulate mTOR signalling in PTEN À / À and TSC À / À cells appears to coincide with delayed inhibition of eIF2b phosphorylation. Moreover, in TSC À / À MEFs, in which serum starvation does not result in a marked mTOR inhibition, removal of serum had much lesser effect on eIF2b phosphorylation as compared with TSC þ / þ MEFs ( Supplementary Fig. 1f). This shows that efficient suppression of eIF2b phosphorylation in response to CK2 inhibitors and serum withdrawal requires downregulation of mTOR signalling, thereby suggesting that at least under certain conditions mTOR may phosphorylate eIF2b independently of CK2. Although sequences surrounding Ser2 and Ser67 in eIF2b do not correspond to a presumed mTOR consensus 23 , eIF2b contains a putative TOS motif ( Supplementary Fig. 2a,b). The TOS motif is required for binding to raptor and recruitment of mTORC1 substrates 24 . Indeed, mTORC1 phosphorylated eIF2b in in vitro kinase assay, and this effect was reduced when mTORC1 was inhibited (by serum starvation or torin1 treatment of cells from which mTORC1 was isolated), when Ser2 and Ser67 were substituted with alanines (eIF2b S(2,67)A), or when TOS motif was disrupted (phenylalanine 89 substituted with alanine; TOS mutant) (Supplementary Fig. 2c-e). Reduced phosphorylation of eIF2b mutants did not stem from the inadvertent effects of mutations on the secondary structure of the protein as illustrated by indistinguishable far-ultraviolet circular dichroism spectra of wild-type (WT) and mutant eIF2b proteins ( Supplementary Fig. 2f). CK2 was not detected in the mTORC1 complex used in in vitro kinase assay ( Supplementary  Fig. 2g). This excluded the possibility that eIF2b was phosphorylated due to the contamination of mTORC1 in vitro kinase assay reactions with CK2. In HEK293E cells, eIF2b co-immunoprecipitated with raptor when mTOR was activated by serum stimulation (Supplementary Fig. 2h). Disruption of putative eIF2b TOS motif abolished raptor:eIF2b interaction ( Supplementary Fig. 2i), but not eIF2b:CK2 association ( Supplementary Fig. 2j). Notably, although the indistinguishable amounts of CK2a were immunoprecipitated by TOS mutant and other eIF2b variants, TOS mutant was not phosphorylated in HEK293E cells (see below and Fig. 3g; lane 4, Supplementary  Fig. 4c). Taken together, these findings indicate that at least under certain conditions mTORC1 phosphorylates eIF2b independently of CK2. mTOR exists in two complexes, mTORC1 and mTORC2 that differ in their function and composition 3 . mTORC1 integrates signals from nutrients, hormones and growth factors (including insulin) to induce anabolic processes including protein synthesis, whereas mTORC2 regulates AGC kinases including AKT and has been implicated in cytoskeleton maintenance and co-translational protein degradation 3 . To confirm that eIF2b phosphorylation is conveyed via mTORC1, but not mTORC2, HEK293E cells were depleted of mTORC1-specific component raptor or mTORC2-specific factor rictor, respectively. While raptor depletion abolished eIF2b phosphorylation, decrease in a FBS:  0.0 rictor expression did not exert a significant effect on phospho-eIF2b levels, but as expected decreased phosphorylation of the mTORC2 target AKT as compared with control ( Fig. 3a,b). Either raptor or rictor depletion induced phospho-eIF2a (Fig. 3a,b), thus indicating that mTORC1 and mTORC2 influence eIF2a phosphorylation by distinct mechanisms likely as a response to different stimuli (see discussion).
Phospho-eIF2b suppresses translation of ATF4 mRNA. We next determined whether eIF2b phosphorylation plays a role in translation regulation. Although initial reports suggested that eIF2b may be dispensable for translation 25,26 , subsequent studies demonstrated that eIF2b facilitates tRNA binding to eIF2 (ref. 27). Overexpression of non-phosphorylatable S(2,67)A and TOS eIF2b mutants increased eIF2a phosphorylation relative to WT eIF2b even in serum-fed cells and this effect was comparable with endoplasmic reticulum stress inducer thapsigargin ( Fig. 3c; compare lanes 2, 4 and 5 with lanes 1 and 3). Induction of ATF4 protein levels by serum starvation was mitigated in S(2,67)D eIF2b mutant-expressing cells, as compared with control cells and cells expressing WT, S(2,67)A or TOS eIF2b mutants ( Fig. 3d; compare lane 5 with lanes 1-4). Similarly to thapsigargin, S(2,67)A or TOS eIF2b mutants induced expression of a luciferase reporter mRNA bearing the 5 0 UTR of human ATF4 (p5 0 UTR ATF4-firefly luciferase) 28 (Fig. 3e). Moreover, addition of recombinant eIF2b WT, but not S(2,67)A or TOS mutants, increased mRNA translation in rabbit reticulocyte lysate (RRL) until amounts of eIF2b exceeding levels of eIF2a and eIF2g were reached ( Supplementary Fig. 3a). To confirm these results, RRL was depleted of eIF2 and reconstituted by adding equimolar amounts of eIF2a/eIF2g and increasing amounts of recombinant eIF2b WT, S(2,67)A or TOS mutants ( Supplementary Fig. 3b). Although all three eIF2b variants were capable of supporting basal levels of mRNA translation as illustrated by higher translational activity in all reconstituted RRLs, as compared with eIF2-depleted RRL, only WT eIF2b increased mRNA translation in a concentration-dependent manner (Supplementary Fig. 3c). Importantly, recombinant WT, but not S(2,67)A or TOS eIF2b mutants, were phosphorylated on Ser2 in RRL ( Supplementary Fig. 3d), which is consistent with previous reports showing CK2 activity in RRL 29 . Cells expressing S(2,67)A or TOS eIF2b mutants exhibited lower global protein synthesis relative to WT or phosphomimetic S(2,67)D eIF2b mutant, as monitored by 35 S-labelling ( Fig. 3f). Monitoring absorbance profiles (254 nm) of cytosolic HEK293E extracts separated by ultracentrifugation on a 15-35% sucrose gradient further confirmed that global translation is reduced under conditions where eIF2b cannot be phosphorylated, as evidenced by an increase in 80S monosome peak in S(2,67)A and TOS eIF2b mutant-versus WT-expressing cells (Supplementary Fig. 3e). These data show that eIF2b phosphorylation correlates with decreased translation of mRNAs harbouring uORFs (for example, ATF4) while bolstering global protein synthesis.
Phospho-eIF2b bolsters TC recycling and protein synthesis. We next investigated whether eIF2b mediates effects of mTORC1 on eIF2a phosphorylation and mRNA translation. To avoid interference of the endogenous eIF2b, and since depletion of eIF2b reduced cell viability 30 (Supplementary Fig. 4a), we first overexpressed FLAG-tagged WT, S(2,67)A, TOS and S(2,67)D eIF2b mutants in HEK293E cells and then depleted endogenous eIF2b by short hairpin RNA (shRNA; Supplementary Fig. 4b).
In serum-stimulated cells, eIF2b phosphorylation was detected in vector/scrambled shRNA-infected control cells, which still express endogenous eIF2b and cells expressing exogenous WT eIF2b, but not in cells expressing exogenous mutant forms of eIF2b with mutated phospho-acceptor sites to generate non-phosphorylatable (S(2,67)A) or phosphomimetic (S(2,67)D) mutants or disrupted TOS motif (TOS; Fig. 3g; compare lanes 1-2 versus 3-5; Supplementary Fig. 4c). Phospho-eIF2a and ATF4 protein levels were higher in non-phosphorylatable S(2,67)A and TOS eIF2b mutants, as compared with control, WT and phosphomimetic S(2,67)D eIF2b mutant-expressing cells ( Fig. 3g; compare lanes 3-4 versus 1-2 and 5). Although mTOR activity was comparably reduced by torin1 across all cell lines as monitored by inhibition of S6K phosphorylation ( Fig. 3g; compare lanes 2-5 with lanes 6-10), only S(2,67)D eIF2b mutant attenuated induction of eIF2a phosphorylation and dampened increase in ATF4 expression ( Fig. 3g; compare lanes 6-9 with 10). Neither serum-stimulated nor torin1-treated cells (4 h) exhibited major differences in steady-state ATF4 mRNA levels (Fig. 3h), thereby suggesting that the observed effects on ATF4 protein levels after 4 h treatments most likely occur at the level of translation. These results demonstrate that eIF2b phosphorylation plays a major role in mediating mTOR-dependent downregulation of ATF4 protein synthesis and stimulation of global mRNA translation. Since we observed that phosphorylation status of eIF2b appears to affect eIF2a phosphorylation, we set out to determine the role of eIF2b phosphorylation in TC recycling using HEK293E cells, which express eIF2b variants and are depleted of endogenous eIF2b ( Supplementary Fig. 4b). After 30-min serum stimulation, S(2,67)A and TOS eIF2b mutants immunoprecipitated drastically lower amounts of tRNA i Met as compared with WT and S(2,67)D eIF2b mutant ( Fig. 4a; compare lanes 10-11 with lanes 9 and 12; Supplementary Fig. 4d). In turn, rapamycin and  Supplementary Fig. 4e). eIF2a phosphorylation inhibits GEF activity of eIF2B and eIF2:eIF2B dissociation 2 . We therefore investigated whether eIF2b phosphorylation affects eIF2:eIF2B association by monitoring the amount of eIF2Bd, which mediates recruitment of eIF2B to eIF2 (ref. 31) by immunoprecipitating cell lysates with an anti-FLAG antibody. In serum-stimulated cells, the amount of eIF2Bd was higher in immunoprecipitated material from S(2,67)A and TOS eIF2b mutant-expressing cells, as compared with those expressing WT and S(2,67)D eIF2b mutant (Fig. 4c). In turn, torin1 increased WT eIF2b:eIF2Bd, but not S(2,67)D eIF2b:eIF2Bd association, to the levels observed in S(2,67)A and TOS eIF2b mutant-expressing cells (Fig. 4c; compare lane 2 with lane 6 versus lane 5 with lane 9). Expression of exogenous eIF2b variants did not conspicuously affect mTORC1 signalling (for example, Fig. 3g), thereby indicating that these effects were mediated by eIF2b and not other mTORC1 substrates. Altogether, these findings show that eIF2b phosphorylation decreases eIF2a phosphorylation and increases eIF2:tRNA i Met binding while stimulating dissociation of eIF2B from eIF2. This suggests that phospho-eIF2b stimulates TC recycling. Phosphorylation status of eIF2a is determined via the action of eIF2a kinases (for example, PKR and PERK) and phosphatases (for example, protein phosphatase 1 (PP1)) 4 . Notwithstanding that our results show that phosphorylation status of eIF2b affects phospho-eIF2a levels (Fig. 3c,d,g; Supplementary Figs 4c and 5a); PERK and PKR activation status as monitored by Thr981 and Thr446 appeared to be largely unaffected by eIF2b WT or mutant proteins ( Supplementary  Fig. 5a). Moreover, although concentration-dependent induction of PERK phosphorylation by thapsigargin was comparable between WT and S(2,67)D eIF2b mutant, corresponding induction of ATF4 protein expression was seemingly reduced in S(2,67)D eIF2b mutant, relative to WT eIF2b-expressing cells (Supplementary Fig. 5b). These observations suggest that the effects of eIF2b on eIF2a phosphorylation are not likely to be mediated by PERK or PKR. Phospho-eIF2b recruits NCK1 to eIF2. eIF2b binds the noncatalytic region of tyrosine kinase adaptor protein 1 (NCK1), which has been implicated in recruitment of PP1 to eIF2 leading to eIF2a dephosphorylation 32 . Consistently, we observed that the levels of NCK1 are reduced in eIF2-depleted RRL ( Supplementary  Fig. 3b, left panel), and although it has been suggested that eIF2b phosphorylation may negatively regulate its binding to NCK1 using in vitro binding assay 33 , insulin, which we show strongly increases phosphorylation of eIF2b, stimulates recruitment of NCK1 to ribosomes 34 . To address the potential role of NCK1 in mediating the effects of phospho-eIF2b on eIF2a phosphorylation, immunoprecipitations were carried out in parental HEK293E cells or in HEK293E cells expressing WT or eIF2b mutants. Endogenous eIF2b and NCK1 were co-immunoprecipitated in serum-stimulated cells, but not in serum-starved cells (Fig. 5a; compare lanes 1 and 2), wherein eIF2b:NCK1 interaction was disrupted by mTOR inhibitors (Fig. 5a; lanes 3-4). In serum-stimulated cells, the amount of NCK1 in the material immunoprecipitated with eIF2b WT and S(2,67)D mutants was higher relative to the amounts of NCK1 co-immunoprecipitated with S(2,67)A and TOS eIF2b mutants ( Fig. 5b; lanes 2 and 5 versus lanes 3 and 4). Torin1 decreased NCK1:WT eIF2b but not NCK1:S(2,67)D eIF2b co-immunoprecipitation ( Fig. 5c; lane 2 versus 5). We showed that the loss of PTEN attenuates the effects of CK2 inhibition on eIF2b phosphorylation (Fig. 2e).  Fig. 5c,d). In turn, 7-h CX-4945 treatment reduced NCK1:WT eIF2b association that was paralleled by decreased exogenous WT eIF2b phosphorylation ( Supplementary Fig. 5d). These findings show that, at least in part, the effects of phospho-eIF2b on phospho-eIF2a status are mediated via recruitment of NCK1 to eIF2. We observed that S(2,67)D eIF2b mutant antagonizes torin1-induced upregulation in phopsho-eIF2a levels compared with all other eIF2b variants (Fig. 3g, lane 10 versus lanes 6-9). Therefore, to further confirm that NCK1 mediates the effects of phopsho-eIF2b on eIF2a phosphorylation, we investigated whether downregulation of NCK1 will abolish the ability of S(2,67)D eIF2b mutant to antagonize increase in eIF2a phosphorylation induced by torin1. Indeed, although expression of S(2,67)D eIF2b was sufficient to antagonize induction of eIF2a phosphorylation by torin1 in control, scrambled short interfering RNA (siRNA)transfected cells, this effect was lost in cells in which NCK1 levels were reduced by siRNA (Fig. 5d; lane 6 versus 11). Collectively, these data suggest a model, whereby eIF2b phosphorylation stimulates NCK1 recruitment to eIF2, which leads to decrease in phopsho-eIF2a levels.
eIF2b mediates effects of mTOR and CK2 on proliferation. Our results indicate that eIF2b phosphorylation mediates the effects of mTORC1 and CK2 on TC formation and global protein synthesis.
Since global protein synthesis rates closely correlate with proliferation rates 35 , we investigated whether eIF2b acts as a mediator of the effects of mTORC1 and CK2 on proliferation of cells expressing WT, S(2,67)A, S(2,67)D or TOS eIF2b mutants, wherein endogenous eIF2b was depleted ( Supplementary Fig. 4b).
In 10% serum, cells expressing S(2,67)A and TOS eIF2b mutants proliferated slower than those expressing WT eIF2b, whereas S(2,67)D eIF2b mutant-expressing cell exhibited slightly increased proliferation relative to WT eIF2b (Fig. 5e). Anti-proliferative effects of torin1, serum depletion or amino-acid deprivation were markedly attenuated by S(2,67)D eIF2b mutant as compared with WT eIF2b-expressing cells ( Fig. 5e; Supplementary Fig. 6a-c). In addition, CX-4945 strongly reduced proliferation of WT eIF2bexpressing cells while having much lesser effect on the proliferation of cells expressing the S(2,67)D eIF2b mutant ( Supplementary  Fig. 7a). These findings demonstrate that eIF2b phosphorylation plays a major role in mediating the effects of mTORC1 and CK2 on cell proliferation. NCK1 is thought to stimulate dephosphorylation of eIF2a via recruitment of PP1 (ref. 32). Thus, we treated cells with salubrinal, which inhibits eIF2a dephosphorylation by PP1 (ref. 36). In contrast to inhibition of mTORC1 or CK2 signalling, S(2,67)D eIF2b mutant failed to attenuate anti-proliferative effects of salubrinal ( Supplementary Fig. 7b). This was paralleled by the inability of S(2,67)D eIF2b mutant to impede salubrinal-induced eIF2a phosphorylation (Supplementary Fig. 7c). Moreover, induction of expression of non-phosporylatable S51A eIF2a mutant in HT1080 cells 37 abolished the stimulatory effects of S(2,67)D eIF2b on proliferation (Supplementary Fig. 8a-c). Taken together, these findings show that the effects of eIF2b on cellular proliferation are mediated by eIF2a phosphorylation and put forward a model whereby CK2 and mTORC1 bolster TC formation via eIF2b phosphorylation, followed by the NCK1dependent dephosphorylation of eIF2a. However, additional roles of NCK1 that lead to the inhibition of eIF2a phosphorylation, such as NCK1-dependent inhibition of PERK 38 may also play a role in this process. Moreover, eIF2b associates with tRNA i Met (ref. 39), as well as the additional translation initiation factors including eIF1 (ref. 40), eIF1A 39 , eIF2B 41 and eIF5 (ref. 42). Therefore, it is plausible that at least some of the effects of eIF2b phosphorylation on TC recycling, protein synthesis and proliferation are mediated via eIF2a-independent mechanisms including direct modulation of eIF2b:tRNA i Met interaction or alteration of eIF2b interaction with translation initiation factors other than eIF2 subunits.

Discussion
It has been shown that in yeast, inhibition of TOR signalling leads to eIF2a phosphorylation via GCN2 kinase [43][44][45] . In mammals, the PI3K/AKT/mTOR and eIF2a signalling have been linked by multiple mechanisms that appear to be dependent on a stressor, length of exposure to stress or proliferation status of the cell 4 . These mechanisms include inhibition of PERK and GCN2 via the mTORC2/AKT pathway and PKR via PTEN [46][47][48] . eIF2a phosphorylation and mTORC1 have also been linked via the catalytic subunit of protein phosphatase 6 (PP6C) 49 . Our study unravels a hitherto unprecedented mechanism of translational regulation, whereby acute activation of mTORC1 bolsters TC formation by inducing phosphorylation of eIF2b, which in turn stimulates the recruitment of NCK1 to eIF2, thereby leading to eIF2a dephosphorylation (Fig. 6). In turn, constitutive activation of mTORC1 leads to a decrease in AKT activity via activation of S6K1-dependent negative feedback mechanism and induction of endoplasmic reticulum stress due to chronically elevated protein synthesis that results in induction of eIF2a kinases and eIF2a phosphorylation 50,51 . Collectively, these findings suggest a  Figure 6 | Schematic representation of the proposed model for the coordination of ternary complex (TC) and eIF4F assembly by mTOR and CK2. On acute stimulation with nutrients, growth factors and insulin, CK2 appears to bolster mTORC1 activity, which is likely mediated via inhibition of PTEN. This leads to phosphorylation and inactivation of 4E-BPs, thereby facilitating eIF4F complex assembly. Simultaneously, CK2 bolsters TC formation by phosphorylating eIF2b. Phosphorylation of eIF2b results in the recruitment of NCK1 to eIF2, which correlates with eIF2a dephosphorylation, likely mediated by PP1. mTORC1 also stimulates eIF2b phosphorylation independently of CK2, which along with its established role in inducing eIF4F levels, demonstrates that mTORC1 also may coordinate TC and eIF4F assembly. In response to chronic increase in protein synthesis or stress (for example, thapsigargin (TG); shown in purple), activation of eIF2a kinases overcomes the effects of eIF2b phosphorylation on TC recycling, thereby allowing cells to fine-tune protein synthesis levels and energy consumption and/or adapt to stress. Broken lines represent uncertainties, such as precise hierarchy of the effects of CK2 and mTORC1 on eIF2b (Ser2) phosphorylation.
fine-tuning mechanism, whereby stimulation of mTORC1 initially leads to upregulation in TC levels to allow induction of translation, whereas prolonged mTOR-dependent translational activation is compensated by downregulation of TC recycling, thereby preserving protein and energy homeostasis. Moreover, although induction of ATF4 expression by thapsigargin was attenuated by phosphomimetic eIF2b mutant as compared with those expressing WT eIF2b, higher thapsigargin concentrations were sufficient to increased ATF4 levels even in phosphomimetic S(2,67)D eIF2b mutant-expressing cells (Supplementary Fig. 5b). This suggests that induction of stress above a certain threshold is likely to override the inhibitory effects of eIF2b phosphorylation on ATF4 expression, thereby antagonizing stimulatory effects of the mTORC1/eIF2b axis on translation and proliferation and allowing cells to rapidly adapt to stress.
We show that acute inhibition of mTORC1 in addition to reducing translation of TOP and 'eIF4E-sensitive' mRNAs via LARP1-and 4E-BP-dependent mechanisms, respectively 52-56 , also upregulates translation of 'eIF2a-sensitive' mRNAs (for example, ATF4) that appears to be mediated via reduction of eIF2b phosphorylation. In contrast to endoplasmic reticulum stressors such as thapsigargin 57 , acute inhibition of the mTORC1/4E-BP/ eIF4E axis has a lesser effect on global protein synthesis and is therefore expected to only marginally effect on translation of housekeeping mRNAs (for example, b-actin) 58 . Therefore, although mTOR inhibitors and inducers of eIF2a kinases increase eIF2a phosphorylation, their impact on the translatome appears to be different. Notwithstanding that mTOR and eIF2a kinase employ a range of distinct effectors, these findings further corroborate a tenet that translation regulation under acute stress or stress recovery is achieved via the cross-talk between the mTOR and eIF2a kinase pathways. Moreover, the complexity of cross-talk between mTOR and eIF2a kinases is further illustrated by studies showing that although we observe translational activation of ATF4 mRNA after 4 h, it is known that prolonged treatment with mTOR inhibitors (412 h) suppresses ATF4 expression at the level of transcription 59 . Future studies are however required to decipher molecular underpinnings of mTOR/eIF2a kinase cross-talk and its role in stress response.
CK2 and mTORC1 appear to phosphorylate overlapping sites on human eIF2b, namely, Ser2 and Ser67. Strikingly, the effects of mTORC1 on phosphorylation of eIF2b on Ser2 are affected by the changes in CK2 activity and vice versa CK2 stimulation of Ser2 eIF2b phosphorylation is attenuated in cells wherein CK2 signalling is uncoupled from mTOR. These observations suggest a complex relationship between CK2 and mTOR in regulating eIF2b phosphorylation. Intriguingly, inspection of publicly available mass spectrometry data indicates the existence of additional phospho-sites on eIF2b that appear to be targeted by mTORC1 and likely CK2 (Ser105 and Thr111) [60][61][62] . These additional phospho-acceptor sites on eIF2b may therefore act as priming sites for Ser2 phosphorylation. This may explain delayed effects of mTOR and CK2 inhibitors on Ser2 eIF2b phosphorylation in cells in which CK2 is overexpressed, or wherein CK2 is uncoupled from mTORC1, respectively. However, future studies are required to determine the precise hierarchy of mTORC1-and CK2-dependent eIF2b phosphorylation.
Our results show that eIF2b phosphorylation-dependent stimulation of TC formation is synchronized with eIF4F complex assembly, which is mediated by CK2 and mTORC1. Inactivation of 4E-BPs by mTORC1-dependent phosphorylation stimulates eIF4F complex assembly 1 , whereas mTORC1-mediated phosphorylation of eIF2b appears to simultaneously facilitate TC recycling. CK2 also appears to bolster TC assembly via direct phosphorylation of eIF2b, which is consistent with previous findings showing that decreased CK2 activity coincides with increased eIF2a phosphorylation and elevated CHOP expression 63 . At the same time, CK2 bolsters 4E-BP phosphorylation by activating mTORC1, which at least in part is mediated via inhibition of PTEN. Since dysregulated mTORC1 and CK2 signalling underpin a number of diseases characterized by aberrant protein synthesis and proliferation including cancer, our findings suggest that eIF2b phosphorylation may also play a prominent role in these pathologies.
TSC2 WT and knockout MEFs were generated by Kwiatkowski's group and obtained from Sonenberg's lab.
Constructs and recombinant proteins. Human eIF2b was subcloned from pOTB7 vector (provided by Dr Kimchi) in pEF-FLAG vector (generously provided by Dr Ronai) using SalI and BamHI enzymes. Ser2 and Ser67 were mutated into an alanine (A) or aspartate (D) and phenylalanine 89 into an alanine (TOS mutant) using Quick-Change Site-Directed Mutagenesis (Agilent Technologies). For stable expression in mammalian cells, FLAG-eIF2b variants were subcloned in pWPI-GFP vector (Addgene) using PmeI enzyme. To produce the recombinant proteins, eIF2b was subcloned in pGEX-6p-1 vector (GE Healthcare Life Sciences) using BamHI and EcoRI. Recombinant proteins were purified as follows: 50 ml of BL21 (DE3) competent E. coli transformed by heat shock with B300 ng of pGEX-6p-1-eIF2b constructs (encoding WT or S(2,67)A, TOS and S(2,67)D eIF2b mutants) was induced with isopropylthiogalactoside for 3 h and then lysed in lysis buffer (50 mM Tris HCl (pH 7.5), 300 mM NaCl, 10 mM TCEP (tris(2-carboxyethyl)phosphine; Sigma-Aldrich)). Lysates were incubated with 20 ml of Glutathione Sepharose 4B (GE Healthcare Life Sciences) for 1 h at 4°C and then washed twice with lysis buffer supplemented with 5% glycerol and 0.1% NP-40, and 700 mM NaCl. Recombinant proteins were then washed once in buffer B (20 mM Tris HCl pH 7.5, 100 mM KCl, 0.1 mM EDTA, 5% glycerol). For in vitro translation and circular dichroism experiments, glutathione S-transferase (GST) tag was removed with PreScission protease (GE Healthcare Life Sciences). GST tag was not removed for the in vitro kinase assay.
siRNA, lentiviral shRNA and generation of cell lines expressing eIF2b variants depleted of endogenous eIF2b. Scrambled control (NS1) and siRNA targeting human NCK1 (5 0 -AACAUCCAUUACAUCUCCUUUCUCGAA-3 0 ) were obtained from Integrated DNA Technologies. siRNA were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions at the final concentration of 10 nM. Cells were lysed 72 h post transfection in RIPA buffer and analysed by western blotting. Lentiviral vectors carrying shRNA targeting human raptor (plasmid 1857), human rictor (plasmid 1854) and the non-target shRNA control (plasmid 1864) were from Addgene. shRNAs targeting human eIF2b (TRCN0000291996) and scrambled control (SHC002) were from Sigma-Aldrich (Mission collection). Vectors encoding shRNAs (7 mg) were co-transfected into 5 Â 10 6 HEK293T cells with 7 mg of each lentivirus packaging plasmids PLP1, PLP2 and PLP-VSVG (Invitrogen). Viral supernatant was collected 48 h post transfection, filtered through a Mixed Cellulose Ester filter (0.45 mm, Fisher Scientific), mixed in 1:1 ratio with growth media and added to cells for 24 h. Infection was carried out in the presence of 8 mg ml À 1 of polybrene (Sigma-Aldrich). Forty-eight hours post infection, cells were selected and maintained in full growth media supplemented with 5 mg ml À 1 puromycin. To generate cells lines expressing FLAG-eIF2b variants, viruses were generated and HEK293E cells were infected as described above. Expression of shRNA-insensitive eIF2b variants was monitored 72 h post infection by western blotting and cells were then infected with viruses carrying shRNA targeting eIF2b or scrambled control. Forty-eight hours post infection, cells were selected with 2 mg ml À 1 puromycin as described above and expression of indicated exogenous eIF2b variants as well as depletion of endogenous eIF2b was determined by western blotting (Supplementary Fig. 4b).
Antibodies and western blotting. For western blotting, cells were scraped in 1 Â PBS (pH 7.4), centrifuged and lysed in RIPA buffer (10 mM Tris (pH 7.3), 1% (w/v) Na-deoxycholate, 1% (v/v) Triton, 150 mM NaCl, 1 mM EDTA, 50 mM NaF, 10 mM b-glycerophosphate and protease inhibitors) supplemented with complete protease inhibitors (Roche). Whole-cell protein extracts (10-60 mg) were analysed by SDS-polyacrylamide gel electrophoresis (SDS-PAGE; 6-15% polyacrylamide gels were used depending on molecular weight of analysed proteins) and transferred to nitrocellulose membrane (Bio-Rad) using wet transfer apparatus (Cleaver). Following antibodies were diluted as indicated in 5% (w/v) bovine serum albumin (Sigma-Aldrich) in 1  were generated in Litchfield's lab. Secondary antibodies (Amersham) were used at 1:10,000, and signals were revealed by chemiluminescence (ECL, GE Healthcare). Where possible, membranes were stripped and reprobed with indicated antibodies. In the cases where this was not possible (for example, wherein the phospho and total antibodies or antibodies recognizing same proteins were used and significant signal remained on the membrane after striping), same lysates were ran simultaneously on duplicate gels, and probed with phospho and total antibodies. Each experiment was performed at least twice independently and the representative data are shown. X-ray films and/or ECL scans of whole membranes are shown in Supplementary Fig. 10. Although we trust that western blotting should be used for qualitative rather than quantitative measurements, as requested by reviewers, we performed densitometric analysis using ImageJ (W. S. Rasband, ImageJ; National Institutes of Health, Bethesda, MD). The resulting data were log2 transformed, normalized per replicate and to the mean of the control, and analysed using analysis of variance in R (r-project.org) (Supplementary Fig. 9).
In vitro kinase assay. For in vitro kinase assay, HEK293E cells were transfected in a 10-cm Petri dish with 7 mg of HA-Raptor using Lipofectamine 2000 (Invitrogen) according to manufacturer's instruction. HA-immunoprecipitation was performed in CHAPS buffer (40 mM HEPES KOH (pH 7.4), 2 mM EDTA, 10 mM sodium pyrophosphate, 0.1 M NaCl, 0.3% CHAPS), washed twice with CHAPS buffer and once in CHAPS buffer supplemented with 0.4 M NaCl. Beads were then washed twice in kinase buffer (25 mM HEPES KOH (pH 7.5), 50 mM KCl, 10 mM MgCl 2 ) and resuspended in 40 ml of kinase buffer. One-sixth of the immunoporecipitated material and 1 mg of recombinant GST-eIF2b variants were used per reaction. Reaction was carried out at 30°C for 30 min in a final volume of 50 ml, in the presence of 100 mM ATP, 0.5 ml of 10 mCi ml À 1 of 32 P-gATP (Perkin Elmer) and 1 Â of kinase buffer. Reactions were stopped with 4 Â Laemmli Sample Buffer and samples were loaded on a 10% SDS-PAGE gel. SDS-PAGE gel was then stained with Coomassie brilliant blue R-250, dried for 2 h at 80°C and the phosphorylation was measured by 32 P incorporation using Storm 860 Molecular Imager. Each experiment was carried out at least in two independent replicates.
Reporter translation assays. In vitro translation assay and in vitro transcription of the reporter renilla mRNA were performed by depleting eIF2 using anti-eIF2b antibody. eIF2b was first linked to 10 ml of protein G agarose beads in 1 Â at 4°C for 2 h. Beads and antibody were then washed twice in PBS and twice in buffer D (25 mM HEPES KOH (pH 7.3), 50 mM KCl, 75 mM KOAc, 2 mM MgCl 2 ). After the last wash, buffer D was removed and the antibody-conjugated beads were incubated with nuclease-treated RRL for 2 h at 4°C. Purified eIF2a and eIF2g lacking the eIF2b subunit were kindly provided by Yuri Svitkin from Sonenberg's laboratory. For translation reaction, 50 ng of the reporter mRNA was added to 7 ml of RLL in a final volume of 10 ml. Titration experiments were performed adding increasing amount of eIF2 b recombinant proteins (40, 80, 160 and 270 ng). In eIF2-depleted RRL, eIF2 was reconstituted by adding 37.5 or 75 ng of purified eIF2a/g subunits and 37.5 or 75 ng of recombinant eIF2b. Translation reaction was carried out at 30°C for 30 min and stopped adding equal volume of 2 Â PBS. Renilla expression was measured in 3 ml of RRL with 50 ml of Renilla Luciferase Assay system (Promega).
Translation of the reporter mRNAs harbouring 5 0 UTR of human ATF4 (ref. 28). was monitored by dual-luciferase assay. HEK293E cells expressing eIF2b variants were seeded in a 10-cm Petri dish and transfected with a mixture of 600 ng of pRenilla (provided by Dr Sonenberg) and 1.2 mg of p5 0 UTR ATF4-firefly luciferase reporter vectors 28 using Lipofectamin 2000 (Invitrogen) according to manufacturer's instruction. Twenty-four hours post transfection, 9 Â 10 5 cells were seeded in a six-well plate in triplicate and the luciferase assay was performed at 48 h post transfection with cells at 80% confluency. Cells were collected in Passive Lysis Buffer, and firefly and renilla luciferase activity was measured using the Dual-luciferase Reporter Assay kit (Promega) according to the manufacturer's instructions. Experiments were repeated three times independently (n ¼ 3), whereby each biological replicate consisted of a technical duplicate.
Genome-wide polysome profiling and RT-qPCR. Polysome profiling 6 was performed in four independent biological replicates. MCF7 cells were seeded in a 15-cm Petri dish, serum starved for 16 h ('control') and then treated with 4.2 nM of recombinant human insulin (Sigma-Aldrich) for 4 h alone ('insulin') or in combination with 250 nM torin1 ('insulin þ torin1'). Cells were collected at 80% confluency and lysed in hypotonic lysis buffer (5 mM Tris HCl pH (7.5), 2.5 mM MgCl 2 , 1.5 mM KCl, 100 mg ml À 1 cycloheximide, 2 mM dithiothreitol (DTT), 0.5% Triton, 0.5% sodium deoxycholate). Ten per cent of the lysates was saved to isolate cytoplasmic mRNA. The amount of RNA in each lysate was measured at 254 nm and 12 ODs were loaded on 5-50% sucrose gradients generated using Gradient Master (Biocomp) and subjected to ultracentrifugation (SW41 rotor; Beckman 36,000, 2 h and 4°C). Sucrose gradients were fractionated by displacement by 60% sucrose/ 0.01% bromphenol blue, using ISCO Foxy fraction collector (35 s for each fraction ¼ 750 ml per fraction) equipped with a ultraviolet lamp for continuous absorbance monitoring. Fractions were flash-frozen immediately after fractionation and stored at À 80°C. RNA was isolated with Trizol (Thermo Fisher Scientific) according to the manufacturer's instruction. For microarray analysis, fractions corresponding to heavy polysomes (more than three ribsomes) were pooled (Fig. 1a). RNA was submitted to the Bioinformatics and Expression analysis core facility at Karolinska Institutet. RNA quality was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies) and complementary DNA was generated and hybridized onto the Affymetrix Human Gene 1.1 ST Array. The oligo package version 1.30.0 was used to summarize and normalize expression data using robust multiarray average (rma) in R version 3.1.1 (www.r-project.org) and bioconductor version 3.0. We used updated probe set definitions 66 as these showed improved precision and accuracy 67 . We assessed the reproducibility using principal component analysis. Samples clustered according to RNA (cytosolic or polysome associated) and experimental group (control, insulin or insulin þ torin1) indicating good reproducibility. We obtained mean expression levels for each treatment and RNA combination. These were then used to calculate mean log 2 fold changes between insulin þ torin1 versus insulin and insulin versus control for each RNA separately (polysome associated or cytosolic). Two-sided Wilcoxon rank-sum test was used to compare differences in fold changes between mRNAs whose translation was stimulated by phosphorylation of eIF2a 5 to those that were not using data from polysome-associated mRNA or cytoplasmic mRNA separately. Data are displayed in Fig. 1. Raw and processed data are available at the Gene Expression Omnibus (GSE76766). For reverse transcriptase-quantitative PCRs (RT-qPCRs), RNA was extracted using Trizol according to the manufacturer's instructions. RT-qPCRs were performed using SuperScript III Reverse Transcriptase, followed by Fast SYBR Green Mastermix (both from Invitrogen), according to the manufacturer's instructions. Analyses were carried out using relative standard curve method as described in http://www3.appliedbiosystems.com/cms/groups/mcb_support/documents/ generaldocuments/cms_040980.pdf. Experiments were performed at least in independent duplicates (n ¼ 2), whereby every sample was analysed in a technical triplicate. Primers were designed using NCBI Primer-BLAST (http://www.ncbi. nlm.nih.gov/tools/primer-blast/) such that T m was between 57 and 63°C, T m difference was o3°C and that primer pairs were separated by at least one intron. Primers were obtained from Integrated DNA Technologies, and their sequences and size of the amplicons are listed below: ATF4 (NM_001675.4, NM_182810.2); Amplicon ¼ 226 nt Human ATF4-Forward 5 0 -TCAAACCTCATGGGTTCTCC-3 0 Human ATF4-Reverse 5 0 -GTGTCATCCAACGTGGTCAG-3 0 Note: these primers recognize Homo sapiens ATF4, transcript variant 1, mRNA (NM_001675.4) and Homo sapiens ATF4, transcript variant 2, mRNA (NM_182810.2), both of which contain inhibitory uORFs and are translationally activated when eIF2a phosphorylation is induced and encode identical protein 68,69 . PCR product is of same size irrespective which variant is amplified.
b-Actin (ACTB; NM_001101.3); Amplicon ¼ 163 nt Actin HF 5 0 -ACCACACCTTCTACAATGAGC-3 0 Actin HR 5 0 -GATAGCACAGCCTGGATAGC-3 0 Metabolic ( 35 S-methionine/cysteine) labelling. For 35 S-methionine/cysteine labelling, HEK293E cells were seeded in six-well plates, serum starved for 16 h and then deprived of methionine and cysteine for an additional 2 h using DMEM without the above amino acids (Gibco). Cells were then treated with 10% dialysed serum containing 10 mCi ml À 1 of 35 S-Met/Cys (Perkin Elmer) and 250 nM of torin1 were indicated. Cells were lysed in RIPA buffer without SDS and 10 ml of the protein extract measured by LS6500 Multi Purpose Scintillation Counter (Beckman Coulter). Experiments were performed in independent triplicates (n ¼ 3) each of which was performed in a technical duplicate.
Cap (m 7 GTP) pull-down assay. MCF7 and HCT116 cells were treated as described in Supplementary Fig. 1a, and lysed in two volumes of buffer B (50 mM MOPS KOH (pH 7.4), 100 mM NaCl, 50 mM NaF 2 mM EDTA, 2 mM EGTA, 1% NP-40, 1% sodium deoxycholate, 7 mM b-mercaptoethanol, protease inhibitors and phosphatase inhibitor cocktail 1 (Sigma-Aldrich)) on ice for 15 min with sporadic vortexing. Extracts were cleared by centrifugation (16,100g for 10 min at 4°C). m 7 GTP-Agarose beads (g-aminohexyl-m7GTP-agarose; Jena Biosciences, Jena, Germany) were equilibrated in buffer C (50 mM MOPS KOH (pH 7.4), 100 mM NaCl, 50 mM NaF, 0.5 mM EDTA, 0.5 mM EGTA, 7 mM b-mercaptoethanol, 2 mM benzamidine or 0.5 mM PMSF, 1 mM Na 3 VO 4 and 0.1 mM GTP (Sigma-Aldrich)). After equilibration, lysates were diluted (B500 mg of total cell protein) to 1 ml with buffer C (in a 2-ml tube) and incubated with equilibrated m 7 GTP-Agarose beads (B50 ml of 50% slurry) for 20 min at 4°C end-over-end rotation. Ten per cent of the lysate was used as the input. The beads were collected by centrifugation (500g for 5 min at 4°C) and washed four times with 1.5 ml of buffer C. Bound proteins were eluted with 0.2 mM m 7 GTP, resuspended in SDS-PAGE loading buffer and analysed by western blotting along with the inputs. Experiments were performed in independent duplicate.
Co-immunoprecipitations, sqRT-PCR and qRT-PCR analyses. Raptor immunopreciptations were carried out using an anti-raptor antibody (Millipore). HEK293E cells were serum starved overnight and then treated as described in figure legends. Cells were collected in CHAPS buffer (0.2% CHAPS, 40 mM HEPES (pH 7.4), 120 mM NaCl, 1 mM EDTA, 10 mM pyrophosphate, 10 mM b-glycerophosphate, 50 mM NaF, 1 mM DTT). Immunoprecipitation was performed in the presence of 0.5 mg ml À 1 reversible crosslinker 3,3 0 -dithiobis (sulfosuccinimidylpropionate) DTTSP (Life Technologies). Two microlitres of anti-raptor antibody (Millipore, 1:500) and 30 ml of protein G-sepharose 50% slurry (Millipore) were equilibrated in CHAPS buffer and incubated in 100 ml of CHAPS buffer for 30 min at 4°C, washed followed by incubation with 1 mg of the lysates for 4 h at 4°C, with end-to-end rotation. Beads were washed four times with 1 ml of ice-cold CHAPS buffer and collected by centrifugation (500g for 2 min at 4°C). Fifty microlitres of beads were resuspended in the sample buffer, boiled and analysed by western blotting. For endogenous eIF2b immunoprecipitations, cells were collected by scraping, washed three times in ice-cold PBS (1,200 r.p.m. for 5 min at 4°C) and lysed in NET-2 buffer (50 mM Tris (pH 7.4), 150 mM NaCl, 2 mM MgOAc, 0.1% NP-40, 1 mM DTT, 1 Â EDTA-free protease inhibitors (Roche)) by 3 Â 10-s bursts using microtip at power 6 on ice. The lysates were spun for 10 min/16,100g at 4°C. An amount of 1 mg ml À 1 of the protein was incubated with 40 ml of the NET-2 buffer equilibrated protein A sepharose beads (Millipore; 30 min at 4°C; end-over-end rotation) on which the supernatant was split in two, set at 500 mg ml À 1 and incubated for 2 h at 4°C (end-over-end rotation) with either eIF2b antibody (P-3) #sc-9978 from Santa Cruz Biotechnologies or the appropriate IgG1 control (#M5284; Sigma; 2 mg of the antibody/IgG per 100-500 mg of total cell protein). On the incubation with the antibody, protein A sepharose beads were added (B30 ml mg À 1 of antibody), and the incubation was carried out for the additional 2 h at 4°C. After the immunoprecipitated beads were washed one time with NET-2 buffer containing 300 mM NaCl and five times with NET-2 buffer containing 150 mM NaCl. Beads were eluted in Laemmli buffer by boiling and eluates were analysed by western blotting.
Proliferation assay. HEK293E cells (1 Â 10 4 ) expressing eIF2b variants, in which endogenous eIF2b was depleted by shRNA ( Supplementary Fig. 4b), were seeded in a 96-well plate and treated as indicated in figure legends. For amino-acid deprivation experiments, 5 Â 10 4 cells were seeded in a 96-well plate and maintained in DMEM without amino acids (Gibco) supplemented with 10% dialysed serum for 6 and 18 h (Invitrogen). Proliferation rate was determined using Cell Proliferation ELISA BrdU kit (Roche). Absorbance at 370 nm (reference wavelength 492 nm) was measured using a Benchmark Plus microplate reader (Bio-Rad). The experiments were performed in independent triplicate (n ¼ 3), with four technical replicates per biological replicate.
Circular dichroism. Far-ultraviolet circular dichroism spectra of the eIF2b WT and S(2,67)A, TOS and S(2,67)D eIF2b mutants (4 mg in 100 ml of 10 mM phosphate buffer (pH 7.2), 50 mM NaCl) purified as described in the 'Constructs and recombinant proteins' section were continuously detected from 197 to 250 nm using Chirascan CD Spectrometer (Applied Photophysics). The measurements were carried out in using a 1-mm path-length cuvette (Hellma) at room temperature using a 1-nm bandwidth. For each sample, two spectra were collected and averaged. The spectral contribution of the buffer was corrected for by subtraction.
Identification of TOS motifs. TOS motifs (F E/D/V M/I/L D/E/V I/E/L) were identified in RefSeq proteins using pattern matching 70 in both mouse and human proteins. For genes encoding for proteins with multiple isoforms containing a TOS motif, only one isoform was reported. Human proteins with TOS motifs were reported with a comparison with TOS motifs in mouse proteins (human and mouse proteins were linked using gene symbols).