eIF4E3 forms an active eIF4F complex during stresses (eIF4F^S) targeting mTOR and re-programs the translatome

Cell culture and treatment

HEK293T, N2a, NIH 3T3 and MEF cells were grown at 37°C in a humidified 5% CO₂ chamber. The cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented by 1% penicillin/streptomycin (Gibco) and 10% fœtal bovine serum (Gibco) for HEK293T, N2a and MEF cells or 5% fœtal bovine serum for NIH 3T3 cells. The DMEM used for N2a cells was pyruvate free.

Viral particles production and transduction

HEK293T cells at 80% confluence were transfected with the second-generation packaging and envelope vectors (pWPI: https://www.addgene.org/12254/). Viral particles, collected and filtered through a 0.45 μm filter at 48 h post-transfection, were used to transduce HEK293T and N2a cells. Five days later, GFP expressing cells were sorted by flow cytometry.

Western blot

Protein extracts prepared in Laemmli buffer [50 mM Tris–HCl pH 6.8, 8% (v/v) glycerol, 4% β-mercapto-ethanol, 2% SDS, 0.015% Bromophenol blue] were resolved on SDS-polyacrylamide gels and electro-transferred to PVDF. Antibodies used in this study were: anti-eIF4E1 (Cell Signaling, #9742S), anti-eIF4E3 (Protintech, #17282-1-AP), anti-eIF4G1 (Santa Cruz, sc-133155), anti-eIF4G3 (ThermoScientific PA5-31101), sc-133155), anti-eIF2α (Invitrogen, #44728G), anti-HA (Covance clone 16B12), anti-FLAG (M2 antibody, Sigma), anti-4EBP1 (Cell Signalling, #9452), anti-actin (Millipore, #MAB1501), anti-RPS6 (Cell Signaling, #2317), anti-Phospho-Akt (Ser473) (Cell Signaling, #9271), anti-AKT (Cell Signaling, #9272), anti-phospho-p70 S6 kinase (Thr389) (Cell Signaling, #9205), anti-p70 RPS6 kinase (Cell Signaling, #9202), goat anti-mouse HRP secondary antibody (Bio-Rad) and goat anti-rabbit HRP secondary antibody (Bio-Rad). The Anti-MYC tag was a gift from Prof. Dominique Soldati (University of Geneva, Switzerland) and anti-eIF4A was a gift from Prof. Michael Altmann (University of Bern, Switzerland). Immunoblots were quantitated using Image Lab (Biorad).

Cap pull down

HEK293T cells were transfected with wild type and tryptophan mutants of HA-tagged eIF4E1 and eIF4E3 using calcium phosphate. Cells were lysed in cap binding buffer [50 mM Tris–HCl pH 7.6, 150 mM NaCl, 2 mM EDTA, 1% (v/v) Triton, 0.5% (v/v) NP40]. 500 μg of protein was incubated overnight at 4°C with m7GTP agarose beads (Jena biosciences, #AC-155S). Beads were washed with cap binding buffer and suspended in 20 μl Laemmli buffer. Input, non-binding and binding fractions were resolved on a polyacrylamide gel.

Polysome profiling

Polysome profiling was performed as previously described (35). Briefly, 20–60% sucrose (Sigma) gradients were prepared manually in 100 mM KCl, 5 mM MgCl₂, 20 mM HEPES and 2 mM DTT. N2a cells were treated for 5 min with 50 μg/ml of cycloheximide (Sigma) and then collected in cold PBS containing 100 μg/ml cycloheximide. Cells were pelleted and lysed in polysome lysis buffer [100 mM KCl, 50 mM Tris–HCl pH 7.6, 1.5 mM MgCl₂, 1 mM DTT, 1 mg/ml Heparin (Sigma), 1.5% (v/v) NP-40, 100 μg/ml cycloheximide] supplemented with protease cocktail inhibitor EDTA-free (Roche) on ice for 20 min. Lysates were cleared by centrifugation (14 000g) and supernatants were loaded onto the gradients. These were centrifuged in a SW41 rotor for three and a half hours at 35 000 rpm at 4°C. After centrifugation, the gradients were analysed through an UV-lamp and an Absorbance detector while being collected in 1 ml fractions using a Foxi Junior Fraction Collector (Isco).

RNA-seq

N2a cell total RNA was purified using the TRIzol Reagent following the manufacturers’ instructions. Libraries were prepared with the TruSeq stranded RNA Library Prep (Illumina) and sequenced at the iGE3 Genomic Platform (UNIGE) on a Hiseq 4000.

Ribosome profiling

N2a cells were treated for 2 h with DMSO or Torin1 and then for 5 min with 100 μg/ml of cycloheximide (Sigma). Cells were collected in cold PBS supplemented with 100 μg/ml of cycloheximide and then lysed on ice for 15 min in mammalian lysis buffer [100 mM KCl, 50 mM HEPES pH 7, 1.5 mM MgCl₂, 1 mM DTT, 1% (v/v) Triton, 5U DNase I (Ambion), 100 μg/ml cycloheximide] supplemented with protease cocktail inhibitor EDTA-free (Roche). Lysates were clarified by centrifugation (10 000 g) and then digested with 7.5 U of RNase I (Ambion). RNA was pelleted and purified on a 1 M sucrose cushion by centrifugation in a S45A rotor at 40 000 rpm for 4 h at 4°C. Pellets were resuspended in TRIzol Reagent (Invitrogen) and RNA was extracted according to the manufacturers instructions. Ribosomal RNA was depleted from the extract (RiboMinus v2 Eukaryote Kit, Invitrogen). RNA fragments with a size ranging from 28 to 34 nt were extracted from a polyacrylamide urea gel and purified by precipitation. RNA was dephosphorylated using T4 Polynucleotide Kinase (NEB). The reverse transcriptase linker was then ligated with T4 RNA Ligase 2 truncated (NEB) and reverse transcribed using M-MLV RNase H minus (Promega). cDNA products were purified on gel and amplified by PCR using the Phusion polymerase (NEB). Libraries were sequenced at the iGE3 Genomic Platform (UNIGE) on a Hiseq 4000.

RiboSeq and RNASeq mapping

For the Ribo-Seq samples, all fastq files were adaptor stripped using cutadapt. Only trimmed reads were retained, with a minimum length of 15 and a quality cutoff of 2 (parameters: -a CTGTAGGCACCATCAAT – trimmed-only –minimum-length = 15 –quality-cutoff = 2). Histograms were produced of ribosome footprint lengths and reads were retained if the trimmed size was between 25 and 35. For all Ribo-Seq and RNA-Seq samples, reads were mapped, using default parameters, with HISAT2 (36), using their pre-prepared UCSC mm10 genome and index. Only primary alignments were retained and reads were removed if they mapped to rRNA, tRNA and pseudogenes according to mm10 RepeatMasker definitions from UCSC. A full set of transcript and CDS sequences for Ensembl release 84 was then established. Only canonical transcripts [defined by mm10 knownCanonical table, downloaded from UCSC] were retained with their corresponding CDS. Reads were then mapped to the canonical transcriptome with bowtie2 (37) using default parameters.

Ribo-Seq analysis

The P-site position of each read was predicted by riboWaltz (38) and confirmed by inspection. Counts were made by aggregating P-sites overlapping with the CDS and P-sites Per Kilobase Million (PPKMs) were then generated through normalizing by CDS length and total counts for the sample. Differential expression was performed pairwise between Ctrl or KO triplicates in the presence of Torin1 or DMSO using edgeR (39) on default settings. Transcripts were only kept in the analysis if they had a CPM >1 in all triplicates for either at least one of the conditions in the pairwise comparison. For further analysis transcripts were filtered if their CDS length was not a multiple of three and if they did not begin with a standard start codon (40) and end with a standard stop codon (UAG, UGA, UAA). This left 20351 transcripts. Scaled plots summarizing the p-site depth profile over all relevant genes for the whole CDS were plotted by splitting every CDS in the gene group into 100 equal bins and aggregating the number of p-sites falling in each. They were normalized by counts for each CDS and total counts genome-wide.

RNA-Seq analysis

Counts were made by aggregating any reads overlapping with the CDS and RPKMs were then generated through normalizing by CDS length and total counts for the sample. Differential expression was performed as with the Ribo-Seq.

Translational efficiency (TE) analysis

TE was assessed using RiboDiff (41) with default parameters with the same Ribo-Seq and RNA-Seq samples as input, using the same expression pre-filters as the edgeR differential expression analysis.

Bimolecular fluorescent complementation assay

Yeast vectors containing Venus fragments were a generous gift from Prof. Martine Collart (University of Geneva, Switzerland). The Venus fragments were transferred into a pcDNA3 backbone. EIF4E3 was fused upstream of Venus fragment 1 at the SalI site. The partner was fused to the NotI site upstream of Venus Fragment 2. HEK293T cells were transfected with both plasmids using calcium phosphate at 24 h post-seeding. At 48 h post transfection, cells were treated with DMSO or Torin1 for 2 h. Cells were fixed in methanol at –20°C, stained by DAPI (Sigma) and analyzed by confocal microscopy (Zeiss).

ATP and NADH assay

5 × 10⁴ control and eIF4E3 KO N2a cells were seeded into a 96-well dish and treated with DMSO or Torin1 for the duration of the assay. At 4 h post seeding and then every subsequent 24 h, ATP and NADH were independently measured using the Cell Glo Titer (Promega) and MTT reagent (Promega) following the manufacturers instructions.

Seahorse analysis

N2a ctrl or KO cells were pretreated for three days with DMSO or Torin1. 3 × 10⁴ cells were seeded into 96-well plates suitable for a 96-well metabolic analyser (Seahorse XF96, Agilent Technologies). Twenty-four hours later and prior to the assay, cells were washed once with Seahorse XF Base Medium (Agilent Technologies) and incubated with 180 μl of Seahorse XF Base Medium (Agilent Technologies) supplemented with 10 mM glucose and 2 mM glutamine. Cell metabolism was probed using a Mito Stress Test kit (Agilent Technologies). The measurement cycles (mix, wait and measure) were performed according to the standard settings. Three measurements of basal respiration were performed followed by three measurement cycles after the serial addition of 2 μM oligomycin, 0.5 μM carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP) and 0.5 μM rotenone and antimycin A. The experiment was repeated twice, each time measuring more than 20 technical replicates. The ATP linked respiration rate, basal respiration rate and extracellular acidification rate were analysed using Wave (Agilent Technologies).

Dopamine quantification

N2a ctrl and KO cells were plated in six-well plates and treated with DMSO or with Torin1 for the duration of the assay. Four days after treatment, cells were collected in cold PBS, lysed in 0.1% (v/v) Triton X-100 and sonicated using a Branson Sonifier 450 (Branson, Danbury, CT, USA) at full power for 30 s. Cell extracts were then centrifuged at 3000 g for 1 min and the supernatant was used to measure intracellular dopamine through extraction in activated alumina and quantified by ultraperformance liquid chromatography–tandem mass spectrometry (UPLC–MS/MS) (42). Values were normalized to the protein amounts.

Glycerol gradients

Glycerol gradients were prepared as previously described (43). HEK293T cells were lysed in polysome lysis buffer without cycloheximide. Cell extracts were loaded onto a 5–20% linear glycerol gradient in 100 mM KCl, 5 mM MgCl₂, and 20 mM HEPES prepared in an SW60 tube. Gradients were centrifuged for 22 h at 40 000 rpm in a SW60 rotor at 4°C. After centrifugation, 10 fractions of 400 μl were collected from the bottom of the tube. Proteins were precipitated by methanol/chloroform precipitation and analysed by western blotting. The pellet was resuspended directly in 40 μl of Laemmli buffer.

Yeast-two-hybrid

Y2H was performed by Hybrigenics Services (https://www.hybrigenics-services.com/) against the Human Cancer Prostate_RP1 library.

Co-immunoprecipitation

DNA cloning

The pcDNA3 HA eIF4GI (1-1599) plasmid was a gift from Prof. Nahum Sonenberg (http://n2t.net/addgene) (44). The pcDNA5/FRT/TO V5 HSPA8 plasmid was a gift from Prof. Harm Kampinga (http://n2t.net/addgene:19514) (45). eIF4G3 (Biocat #BC094683-seq-TCHS1003-GVO-TRI) and Th (Biocat #BC156668-seq-TOMS6004-GVO-TRI) were purchased from BioCat. The pcDNA3 eIF4E1-FLAG S53A clone was a generous gift from Prof. David Sabatini (USA).

eIF4E3 binds the cap and is not regulated by 4EBP1

eIF4E3 has cap binding activity (20). However, the interaction involves only a single tryptophan residue (W115 in the human protein), whereas eIF4E1 traps the cap between two tryptophans creating an aromatic sandwich (W56 and W110). To confirm the role of W115, HA-tagged versions of wild type and mutant murine eIF4E1 and human eIF4E3 were transiently overexpressed. Cap binding proteins were selected and characterized by western blotting (Figure 1A and B). While we could pull-down WT forms of eIF4E1^HA and eIF4E3^HA, none of the eIF4E1^HA single tryptophan mutants (W56A and W110A) or eIF4E3^HA W115A were retained.

The interaction between eIF4E3 and 4EBP1 was probed by transient co-overexpression of a rat ^MYC/HIS4EBP1, previously reported to interfere with eIF4F assembly in HEK293 cells (46), with eIF4E3^FLAG or eIF4E1^FLAG. eIF4E1^FLAG but not eIF4E3^FLAG was pulled down with ^MYC/HIS4EBP1 (Figure 1C), confirming previous observations that had employed purified proteins (18). We also noted that 4EBP1 remained largely hyperphosphorylated (the β/γ forms) in cells expressing eIF4E3^HA, whereas the hypophosphorylated α form was more evident in cells transduced with eIF4E1^HA, as previously reported (10,47) (Figure 1D: see also Supplementary Figure S5A). The changes in 4EBP1 phosphorylation status were not affected by any of the mutants. These results demonstrate that eIF4E3 is not being regulated by 4EBP1.

In normal growth conditions, the eIF4F complex carries mainly eIF4E1

Over-expression/co-IP studies indicated that eIF4E3 could form an eIF4F complex that was named eIF4F(3) (18,31,32). However, we were unable to reproduce these results with endogenous proteins. Therefore, to follow recruitment of the endogenous eIF4E3 into eIF4F complexes we employed glycerol gradients a technique that we had previously employed to analyse TC assembly (35,48). Before loading, the cell extracts were divided into two equal fractions. One was heated in denaturing conditions (+SDS) to disrupt complexes. Gradient analysis of native cell extracts revealed that most of the endogenous eIF4E1 co-sedimented in the lower end of the gradient, namely fractions 3/4, with eIF4G1 and eIF4A1 (Figure 2A). A minor fraction co-sedimented with hypophosphorylated 4EBP1 (fraction 7) but not with hyperphosphorylated 4EBP1 (fractions 8–10). eIF4A1, which is a very abundant cellular protein (49), sedimented in multiple fractions. With regards to eIF4E3, the vast majority of the protein sedimented in the upper half of the gradient from fractions 5–9, with only a minor amount co-sedimenting with eIF4G1 in fraction 4. Our eIF4E3 antibody also detected a faster migrating protein band (indicated as * in the lower panel of Figure 2A) co-sedimenting with the full-length protein in fractions 6–8. Denaturation disrupted all the complexes in the initiation process as confirmed by the shift of eIF4E1/eIF4A (the eIF4F complex) and eIF2α (the TC) to the upper fractions (Supplementary Figure S1A and S1B) (48). These results suggest that under normal growth conditions eIF4F is composed of eIF4E1, eIF4G1 and eIF4A1, with eIF4E3 remaining in the light pool. We next explored if this eIF4F complex is associated with ribosomes by examining the sedimentation profile of ribosomal protein RPS6. We did not find RPS6 within the gradients from the native extracts (Figure 2B). However, it was detected in the SDS-denatured extracts suggesting that under native conditions it was associated only with the 40S subunit that had pelleted i.e. there is no free RPS6 pool in the cytoplasm. This we confirmed by re-suspending the pellets in Laemmli buffer. Western blotting revealed the presence of RPS6 and eIF4E1, eIF4G1 eIF4A1 but not eIF4E3 (Figure 2C). Curiously, despite its small size (∼ 29 kDa for the human), the RPS6 sedimented into the gradient suggesting that it may aggregate or assemble on RNA (the gradient did not contain SDS). Therefore, our sedimentation analysis monitored the formation of complexes not associated with the 40S.

eIF4E3 sediments with the eIF4F complex when eIF4E1 is sequestered by 4EBP1

Our results confirm that eIF4E3 does not bind 4EBP1. It also does not co-sediment with eIF4G1 in normal growth conditions. We hypothesized that eIF4E3 would assemble an eIF4F complex under conditions in which eIF4E1 is sequestered by 4EBP1. We therefore treated cells with the mTOR inhibitor Torin1 (50). This drug targets both mTORC1 and mTORC2 as confirmed by the dephosphorylation of 4EBP1, RPS6K and AKT-Ser⁴⁷³ (Figure 2D). It also does not change the intracellular levels of the eIF4E3 protein (Supplementary Figure S1C). We repeated the sedimentation analysis on extracts prepared from cells treated with Torin1 or with DMSO for 2 h. As expected, a major fraction of eIF4E1 moved from the bottom to the top of the gradient and co-sedimented with hypophosphorylated 4EBP1 (Figure 2E, fractions 7–9). eIF4G1 remained in fractions 2–4 whereas a significant amount of eIF4A1 moved into the lighter fractions 7/8. The presence of eIF4A1 in the fractions spanning 2–8 is consistent with its role in diverse aspects of RNA biology that extend beyond translation initiation (51). A major fraction of eIF4E3 now co-sedimented with eIF4G1 in fractions 3 and 4. Curiously, the sedimentation of the faster migrating band observed on the eIF4E3 blots (*) was unaltered by Torin1 treatment, indicating that it was not recruited into an eIF4F complex. This behaviour suggests that it might be the N-terminally truncated isoform B that was reported not to interact with eIF4G (the eIF4E3 polyclonal antibody used in these studies was raised against the C-terminal region of the protein) (52,53). Therefore, upon mTOR inhibition the sedimentation profiles of eIF4E1 and eIF4E3 are inversed (Figure 2F and G). This indicates that eIF4E3 replaces eIF4E1 in the eIF4F complex when the latter is sequestered by hypophosphorylated 4EBP1.

These results suggest that the PI3K/AKT/mTOR pathway regulates eIF4E3 recruitment into an eIF4F complex. To confirm this, we examined the impact of upstream inhibitors of the mTOR pathway on the sedimentation profiles of eIF4E1, eIF4E3 and eIF4G1. We targeted PI3K activity using LY294002 (54,55). LY294002 caused a loss of AKT-Ser⁴⁷³, 4EBP1 and S6K1 phosphorylation (Figure 2H) mimicking Torin1 treatment (Figure 2I). The majority of eIF4E1 moved to the top of the gradient in fractions 8 and 9, and eIF4E3 was now found in the same fractions as eIF4G1. This confirms that eIF4E3 co-sediments with the eIF4F complex when the PI3K/AKT pathway is compromised. We then tried to mimic the effect in a physiological setting. The activation of mTORC1 is nutrient dependent with its Raptor and mLst8 subunits acting as nutrient sensors (56,57). We starved cells of nutrients and a carbon source by removing serum and glucose from the media. After 2 h of starvation, AKT and S6K1 remained phosphorylated whereas 4EBP1 was hypophosphorylated (Figure 2J). A major part of eIF4E1 still co-sedimented with eIF4G1 (Figure 2K). However, some eIF4E3 was also detected in the eIF4G1 fraction (fraction 4). Even though the effect was less marked, physiological conditions such as glucose starvation appear to promote the eIF4E3 interaction with eIF4G1.

The eIF4E3 interactome

To validate that eIF4E3 could assemble a bona fide eIF4F complex we performed a yeast-2-hybrid (Y2H) screen to determine its partners. eIF4E3 was used as a prey and screened against a peptide library originating from human prostate cancer cell lines. This background was selected because eIF4E3 has been implicated in androgen-independent prostate cancer progression (33). About 50 million interactions were tested and 360 clones screened positive. Data mining revealed ten potential partners of eIF4E3 whose functions are listed in Table 1. A selected interaction domain (SID) for each prey sequence is depicted in Supplementary Figure S2A. Among the ten partners were eIF4G1 and eIF4G3, consistent with previous observations that eIF4E3 could form an eIF4F(3) complex with eIF4G1 (31,58) and with eIF4G3 (32).

The SIDs on eIF4G1 and eIF4G3 were overlapping and corresponded to the reported binding site for eIF4E1 (Supplementary Figure S2A), suggesting that eIF4E1 and eIF4E3 compete for the same surface on the scaffolding protein. To confirm the interaction, we performed co-IP. Cells transduced with human eIF4E3^HA were treated for 2 h with Torin1 or DMSO, lysed and incubated with beads carrying covalently cross-linked HA antibody. The presence of eIF4G1 and eIF4G3 on the beads was then analysed by Western blot (Figure 3A). Both eIF4G1 and eIF4G3 co-IPed with eIF4E3^HA in Torin1 conditions confirming that eIF4E3 assembles into an eIF4F complex during stress (hereafter referred to as eIF4F^S) with either eIF4G1 (eIF4F^S1) or eIF4G3 (eIF4F^S3).

The seven other partners are not linked to translation and show no common motif with the exception of Angel1 with eIF4G1/3 (Supplementary Figure S2B). Angel1 is a member of the CCR4 family (59) and interacts with eIF4E1 via the YxxxxLΦ 4E binding motif. However, it plays no role in translational control (60). Curiously, the same motif is also found on 4EBP. It has been proposed that differences within (S/T)VxxFW surface found on both the eIF4E1 and eIF4E3 proteins may permit the latter to discriminate between 4EBP and eIF4G, and presumably Angel1 (61,62). The other partners are CDC5L that is involved in the G2/M transition during the cell cycle; three heat shock proteins of the HSP70 family that share more than 80% homology (63); the HIF1α inhibitor HIF1AN and the X-linked inhibitor of Apoptosis XIAP. To confirm the interactome, we employed the Split Venus fragment technique that is a bimolecular fluorescent complementation assay (64). The C-term of eIF4E3 was fused to the N-term of Venus (VN210) and the putative partner protein to its C-term (VC210). Cells were co-transfected with eIF4E3-VN and either empty-VC or partner-VC. To test if the interaction responded to Torin1, cells were treated for 2 h 2 days post-transfection. YFP was not observed with eIF4E3-VN alone or in the presence of empty-VC (Figure 3B). The eIF4G1-VC gave an enhanced close proximity signal with eIF4E3-VN during Torin1 treatment although a signal could be detected in the DMSO control (Figure 3C). Even under these conditions of overexpression, a signal with eIF4G3-VC was only observed in Torin1 treated cells, consistent with our co-IP results (Figure 3D). However, HSPA8-VC, XIAP-VC, HIF1AN-VC and CDC5L-VC gave signals independently of drug treatment (Figure 3E–H and Supplementary Figure S3). Interestingly, the signal localization was dependent on the partner tested. For eIF4G1-VC, eIF4G3-VC, HIF1AN-VC, HSPA8-VC and XIAP-VC, the signals were detected exclusively in the cytosol (Figure 3E–G and Supplementary Figure S3). Complexes with CDC5L were observed in the nucleus (Figure 3H, top panel) with a staining reminiscent of nucleoli (Supplementary Figure S3, third panel) (65). We observed cytosolic signals in cells entering mitosis (Figure 3H, lower panel). Our results indicate that trace amounts of eIF4E3 may be present in the nucleus in association with CDC5L.

eIF4E3 impacts on the translatome in stress conditions: Polysome seeding based on 5′ TL length

With the confirmation that eIF4E3 forms an eIF4F^S complex, we sought to exploit a knockout (KO) approach to examine the impact of eIF4E3 loss on the translatome. As HEK293T cells express low levels of eIF4E3, we screened for a cell line in which protein expression was high (Supplementary Figure S4). While eIF4E1 was found in all cell lines tested (Supplementary Figure S4A), the expression level of eIF4E3 varied considerably (it was undetectable in the non-tumoural cell lines NIH3T3, MEF and MRC5, but highest in the mouse neuroblastoma cell line N2a) (Supplementary Figure S4B). We therefore decided to pursue our analysis in N2a cells. Neuroblastoma cells represent cancer cells that arise from chromaffin cells mostly found in the medulla of the adrenal glands (66), a tissue in which eIF4E3 levels in the mouse are also high (https://www.mousephenotype.org/data/genes/MGI:1914142). We KOed eIF4E3 using CRISPR-Cas9 (67). N2a cells were transfected with empty vector (control cells: ctrl) or with three gRNAs targeting exons 1, 2 and 4 (KO cells). Protein expression was reduced to undetectable levels (Figure 4A). We performed polysome profiling on ctrl and KO cells treated with DMSO or Torin1 for 2 h. We did not observe marked changes in the profiles between ctrl and KO cells treated with DMSO (Figure 4B, first panel), but Torin1 treatment reduced the heavy polysomal fraction in both (Figure 4B, second and third panels). However, ctrl and KO cells exhibited very similar profiles in the presence of Torin1 (Figure 4B, fourth pane). Therefore, to quantify the impact of eIF4E3 on the translatome, we turned to ribosome profiling (ribo-seq) (68). This was performed on biological triplicates of ctrl and KO cells treated with DMSO or with Torin1 for 2 h. The phosphorylation status of 4EBP1 was used to monitor drug treatment using an aliquot of the cell extract sampled before preparation of the ribo-seq library (Supplementary Figure S5A). We obtained good reproducible data with the exception of an outlier in the N2a/DMSO ctrl that was discarded for the analysis (Supplementary Figure S5B). The ribo-seq data showed clear triplicity throughout the annotated CDSs and correct footprint size distribution (Supplementary Figure S5C and D). The transcriptome was also ascertained in all conditions (Supplementary Figure S5E). This revealed that the steady state level of 36 mRNAs were significantly altered in the ctrl versus KO, with eIF4E3 being the most downregulated in the KO (Supplementary Figure S6A). We also noted that Torin1 treatment modified the transcriptome (Supplementary Figure S6A), an observation already reported (69). The absence of eIF4E3 had a significant additional impact on the transcriptome with half of the differentially expressed (DE) genes (ctrl/Torin1 versus KO/Torin1) being downregulated and the other half upregulated (Supplementary Figure S6A). To assess the impact of eIF4E3 on global translation, translation efficiency (TE) was evaluated from our ribo-seq and RNA-seq data. The TE was unchanged between ctrl and KO (Figure 4C). This is consistent with eIF4E3 playing no role in the translational readout in normal physiological conditions as it is not found in an eIF4F complex. However, Torin1 treatment had a major impact on the translatome in both cell lines, modifying the TE of a number of genes (Supplementary Figure S6A). It is established that the translation of one mRNA subpopulation is highly mTOR dependent. These are referred to as the 5′ Terminal-Oligo-Pyrimidine (TOP) mRNAs and they share a common 5′ TL signature starting with a cytidine immediately after the 5′ cap followed by a 4–14 poly-pyrimidine stretch (70–72). These mRNAs essentially code for ribosomal proteins and translation elongations factors (72,73). TOP mRNAs selected from a list taken from (72) were downregulated in both cell lines after Torin1 treatment (Supplementary Figure S6A, orange dots) confirming that the drug treatment worked. Their downregulation was similar in both cells lines, meaning that eIF4E3 does not impact TOP mRNA translation (Supplementary Figure S6B, R = 0.96). When comparing the Torin1 treated cells, a considerable number of genes exhibited TE changes in the eIF4E3 KO (Figure 4D). The majority of these genes (ctrl/Torin1 versus KO/Torin1) had a lower TE, meaning their translation under the Torin1-induced stress is eIF4E3 dependent. Nevertheless, about one third had a higher TE, suggesting that eIF4E3 has an inhibitory effect on their translation in stress conditions. The ribosome occupancy on the CDS was similar in all four conditions and it did not change between the down and up regulated populations (Supplementary Figure S6C–E). This result indicates that translation elongation was not altered in the four conditions; hence, TE changes reflect a defect at the level of initiation. As eIF4E3 is a cap binding protein, we looked for possible motif(s) or sequence signature(s) in the 5′ TL of the mRNA subpopulations responding to the KO under stress, namely, the neutral (no significant change in TE – ctrl/Torin1 versus KO/Torin1), stimulated (high TE) or inhibited (low TE). We were unable to find sequence motifs that correlated with each group, even when selecting only the most highly up and down-regulated subgroups. However, whereas the CDS, 3′ UTR and overall mRNA lengths did not vary within the three populations, 5′ TL length was significantly different (Figure 4E). The downregulated population had a significantly shorter 5′ TL compared to the neutral population, whereas the upregulated population had a significant longer 5′ TL (Figure 4E). The high TE population also exhibited a 5′ TL G/C distribution lower than that observed within both the low TE and non-responsive populations (which were essentially identical: Figure 4F). The results suggest that eIF4E3 can impact both positively and negatively on the translational readout during stress and this correlates mainly with 5′ TL length.

N2a KO cells manifest defects in cellular metabolism under Torin1-induced stress

We were then interested to look at the impact of the KO on the cell phenotype under stress conditions. As the proteome is an excellent marker for the cellular phenotype and correlates better with the translatome than with the transcriptome (74), we exploited only the ribo-seq data to determine what proteins are differentially expressed (Figure 5A). In agreement with the TE, the ribo-seq revealed that two-thirds of the mRNAs were downregulated (n = 492) and one-third upregulated (n = 278). KEGG analysis demonstrated that each population was enriched in defined pathways (Figure 5B and C) and in functions associated with different cellular compartments (Supplementary Figure S7A and B).

To validate the KEGG analysis we chose to explore two metabolic pathways. Firstly, to confirm that cellular metabolism was affected in the KO under stress, the levels of ATP and NAD(P)H were measured in ctrl and KO N2a cells. Since the impact of translatome changes on the proteome will also depend on protein stability, and mammalian proteins have half-lives ranging from hours to days, measurements were performed over several days of drug treatment. Cells were seeded and treated with DMSO or with Torin1 throughout the duration of the assay. Every 24 h, ATP and NAD(P)H levels were assayed and their values normalized to day 0 and to the cells treated with DMSO. ATP and NAD(P)H levels showed a significant decrease at day 2 and day 3 post-treatment, respectively (Figure 5D and E). To test if the reduction in intracellular ATP level was related to reduced mitochondrial respiration as suggested by the pathway enrichment analysis (Figure 5B), we probed the oxygen consumption rate using a 96-well metabolic analyser (Seahorse XF96, Agilent Technologies). Following three days of pre-treatment with DMSO or Torin1, ctrl and KO cells were seeded into 96-well plates and probed 36 h later using a Mito Stress Test kit. The ATP linked respiration rate (difference between basal respiration and respiration following oligomycin addition), was modestly reduced in ctrl cells following Torin1 treatment (20% reduction), but markedly reduced in KO cells (60% reduction), confirming a reduction in ATP production through mitochondrial respiration in KO cells upon Torin1 treatment (Figure 5F). Similarly, KO cells displayed the lowest absolute basal respiration rate (oxygen consumption rate, OCR) as well as the lowest relative OCR when normalized to the basal glycolytic rate (extracellular acidification rate, ECAR). This relative respiration rate was 2-fold reduced in WT cells upon Torin1 treatment, but 3.4-fold reduced in KO cells (Figure 5G). We conclude that mitochondrial respiration is reduced in N2a cells upon Torin1 treatment and that this effect is markedly aggravated in the KO cell line, contributing to reduced ATP levels.

One of the hallmarks of cellular stress in neuroendocrine cells such as neuroblastoma cells is the production of catecholamine (66). However, N2a cells only produce dopamine when differentiated into dopaminergic cells. The differentiation process triggers the expression of tyrosine hydroxylase (Th) an enzyme that is normally weakly express in undifferentiated N2a cells (75). Th catalyses the conversion of l-tyrosine to l-DOPA, an intermediate in dopamine synthesis. Mining of the ribo-seq data suggested that the KO under Torin1 conditions would also influence this metabolic pathway. To investigate this, both ctrl and KO cells were transduced with lentivectors carrying Th^HA (Figure 5H). This restored the production of dopamine in both backgrounds as evidenced by an increase in its intracellular (i.c.) levels (Supplementary Figure S7C). Cells were then treated for four days with Torin1 or DMSO and i.c. dopamine levels determined. While we did not observe any significant change under DMSO conditions (Figure 5I), KO cells under Torin1 had about 50% less, i.c. dopamine compared to control (Figure 5J).