A unifying model for mTORC1-mediated regulation of mRNA translation.

The mTOR complex 1 (mTORC1) kinase nucleates a pathway that promotes cell growth and proliferation and is the target of rapamycin, a drug with many clinical uses. mTORC1 regulates messenger RNA translation, but the overall translational program is poorly defined and no unifying model exists to explain how mTORC1 differentially controls the translation of specific mRNAs. Here we use high-resolution transcriptome-scale ribosome profiling to monitor translation in mouse cells acutely treated with the mTOR inhibitor Torin 1, which, unlike rapamycin, fully inhibits mTORC1 (ref. 2). Our data reveal a surprisingly simple model of the mRNA features and mechanisms that confer mTORC1-dependent translation control. The subset of mRNAs that are specifically regulated by mTORC1 consists almost entirely of transcripts with established 5' terminal oligopyrimidine (TOP) motifs, or, like Hsp90ab1 and Ybx1, with previously unrecognized TOP or related TOP-like motifs that we identified. We find no evidence to support proposals that mTORC1 preferentially regulates mRNAs with increased 5' untranslated region length or complexity. mTORC1 phosphorylates a myriad of translational regulators, but how it controls TOP mRNA translation is unknown. Remarkably, loss of just the 4E-BP family of translational repressors, arguably the best characterized mTORC1 substrates, is sufficient to render TOP and TOP-like mRNA translation resistant to Torin 1. The 4E-BPs inhibit translation initiation by interfering with the interaction between the cap-binding protein eIF4E and eIF4G1. Loss of this interaction diminishes the capacity of eIF4E to bind TOP and TOP-like mRNAs much more than other mRNAs, explaining why mTOR inhibition selectively suppresses their translation. Our results clarify the translational program controlled by mTORC1 and identify 4E-BPs and eIF4G1 as its master effectors.

The mTOR kinase is the catalytic subunit of two complexes, mTOR Complex 1 and 2 (mTORC1/2), that regulate growth and are often deregulated in disease (reviewed in [1]). mTORC1 is the allosteric target of the well-known drug rapamycin, which has clinical uses in organ transplantation, cardiology, and oncology. A major function of mTORC1 is to regulate protein synthesis, which it is thought to control through several substrates, including the S6 kinases, the inhibitory eIF4E-binding proteins (4E-BPs), and the eIF4G initiation factors. ATP-competitive inhibitors of mTOR such as Torin1 impair protein synthesis and proliferation to much greater degrees than rapamycin[1,2], largely due to their inhibition of rapamycin-resistant functions of mTORC1. Because earlier efforts to identify mRNAs translationally regulated by mTORC1 relied on rapamycin[5-7], it is likely that the mTORC1-regulated translational program is not fully defined.

As a step towards defining this program, we examined the effects of Torin1 on protein synthesis in mouse embryonic fibroblasts (MEFs). To focus on the direct translational outputs of mTORC1 and avoid secondary effects, we treated cells with Torin1 for only 2 h. Torin1 blocked canonical mTORC1-dependent events, such as the phosphorylation of S6K1 and 4E-BP1, but did not increase the phosphorylation of eIF2α, which represses translation and is induced by stresses like amino acid deprivation (Fig. 1a). In wild-type (WT) MEFs, Torin1 suppressed [35]S-Cys/Met incorporation into protein by ~65% and shifted ribosomes out of polysomes, indicating that mTOR inhibition causes a severe defect in translation initiation (Fig. 1b, c).

Figure 1

Profile of mTOR-regulated translation

(a) WT MEFs were treated with vehicle (DMSO), 250 nM rapamycin or Torin1, or starved for amino acids for 2 h and analyzed for protein levels. (b) WT MEFs were treated for 2 h with vehicle (DMSO), 250 nM rapamycin or Torin1, or 10 ug/ml cycloheximide, pulsed for 30 min with [35]S-Cys/Met and [35]S incorporation into protein quantified and normalized to the total protein. Data are mean +/− s.d. (n=3). (c) Polysome profiles of WT MEFs treated with DMSO or 250 nM Torin1 for 2 h. (d) Distributions of ribosome footprint (RF) frequency in vehicle- or Torin1-treated cells. RF libraries from cells treated as in (c) were used to determine RF frequencies (reads per million, RPM) for 4840 mRNAs. (e) β-actin mRNA abundance in fractions from (c) were quantified by qPCR, and calculated as a percentage of the total in all fractions. Data are means +/− s.e.m. (n=2). (f) Distribution of changes in translational efficiency from vehicle- or Torin1-treated cells. RF frequencies from (d) were normalized to transcript levels to calculate translational efficiencies. Ribosome densities (reads per kilobase per million, RPKM) from vehicle- and Torin1-treated cells are inset. mRNAs with suppressed (z-score < −1.5) or resistant (z-score > 1.5) translational efficiencies are indicated. (g) Torin1-dependent changes in translational efficiency for indicated mRNA classes. For histone mRNAs, results reflect changes in ribosome density only. Significance determined by two-tailed Mann-Whitney U test.

To systematically monitor the translation of individual mRNAs, we analyzed vehicle- and Torin1-treated MEFs using transcriptome-scale ribosome profiling[8]. Ribosome profiling provides a precise measurement of mRNA translation by quantifying ribosome-protected mRNA fragments (ribosome footprints or RFs) using deep sequencing. In proliferating MEFs, we detected 3.9 million exon-mapped RFs that corresponded to 12,856 actively translated Refseq mRNAs. 4840 could be monitored at levels sufficient for robust measurements of Torin1-induced translational changes (Supplementary Table 1). The frequency of RFs that map to each mRNA (gene-specific reads per million total exon-mapped reads, or RPM) reflects the proportion of ribosomes engaged in the translation of that transcript. In vehicle- and Torin1-treated cells, the distributions of RF frequencies were largely super-imposable (median log2(change in RF frequency) = 0.08), arguing that mTOR inhibition has similar effects on the translation of most mRNAs (Fig. 1d). Given this and the [35]S-Cys/Met incorporation results (Fig. 1b), we determined (see methods) that mTOR inhibition suppresses the translation of nearly all (99.8%) mRNAs to some degree, with a mean reduction in translation of 61% (median = 60.5%). Consistent with this conclusion, β-actin mRNA, which, like most mRNAs, underwent little change in RF frequency upon Torin1 treatment (log2(Δ RPM) = −0.08), was nevertheless partially but significantly depleted from polysomes in Torin1-treated cells (Fig. 1e). Thus, acute mTOR inhibition has the unappreciated capacity to moderately suppress the translation of nearly all mRNAs.

To identify the mRNAs most regulated by mTOR at the translational level, we calculated the Torin1-induced change in the translational efficiency of each mRNA (Fig. 1f). This measurement normalizes RF frequency to the abundance of the corresponding transcript and so decouples translational and transcriptional regulation. Using a z-score cutoff of ±1.5, we selected 253 suppressed and 198 resistant mRNAs for further analysis. Gene ontology (GO) analyses of Torin1-suppressed mRNAs showed enrichment for those involved in various steps in protein synthesis (Supplementary Fig. 1a), albeit with differences amongst components of the translational machinery (Fig. 1g; Supplementary Table 2). For instance, Torin1 suppressed the translation of eIF4B but not other eIF4F complex components, and of nearly all cytoplasmic ribosomal proteins, except Rps27a, which has extra-ribosomal functions[9]. Torin1-resistant mRNAs are enriched for transcription factors (Supplementary Fig. 1a), such as Stra13, Myc, Paf1, and Foxo1. Additionally, the translation of mRNAs with putative internal ribosomal entry sites (IRES)[10,11] and, unexpectedly, those encoding histones were also clearly resistant to Torin1 (Fig. 1g), indicating that these mRNAs use modes of initiation that do not depend on mTOR activity[12].

We considered the features that define the mRNAs that are most translationally suppressed upon mTOR inhibition. Two types of mRNAs are thought to be highly mTOR-dependent: (1) those with long and complex 5′ UTRs that are reported to be regulated through a 4E-BP-dependent mechanism[3] and (2) mRNAs with 5′ terminal oligopyrimidine (TOP) motifs that are regulated through an unknown mechanism[13]. Surprisingly, the translational efficiency of commonly cited examples of mRNAs with long, complex UTRs, such as cyclin D1 (log2(Δ) = −0.07), cyclin D3 (log2(Δ) = 0.09), Myc (log2(Δ) = 0.92) and Vegfa (log2(Δ) = 0.79)[14], was not significantly suppressed in our dataset. We found no evidence that 5′ UTR length or complexity correlated positively with sensitivity to mTOR inhibition and, if anything, mRNAs with shorter and less complex 5′ and 3′ UTRs tended to be more sensitive (Supplementary Fig. 1b-d). However, UTR length per se does not determine mTOR-dependency because mRNAs with similarly short CDS and UTR lengths, like those for cytoplasmic and mitochondrial ribosomal proteins (Supplementary Fig. 1b), were differentially sensitive to mTOR inhibition (Fig. 1g). Although it is puzzling that we find little evidence for the selective regulation of mRNAs with complex 5′ UTRs, these mRNAs may be affected, upon prolonged mTOR inhibition, by secondary consequences of the acute changes described here. Consistent with this possibility, 24-48 hours of mTOR inhibition are required to maximally exclude the cyclin D1 mRNA from polysomes[15,16].

Torin1 suppressed the translational efficiencies of all known TOP mRNAs in our dataset (mean log2(Δ) = −1.49) (Fig. 2a; Supplementary Table 2). TOP mRNAs are defined as those with a C immediately after the 5′-cap, followed by an uninterrupted stretch of 4-14 pyrimidines[13,17], and tend to encode proteins associated with translation[13,18]. When averaged across known TOP mRNAs, Torin1 depleted ribosome footprint density throughout the CDS (Supplementary Fig. 2a) and shifted known TOP mRNAs (eEF2, Rps20) out of polysomes (Fig. 2e). RNAi-mediated depletion of raptor, an essential mTORC1 component, also selectively inhibited the translation of TOP mRNAs (Supplementary Fig. 3).

Figure 2

Translation of TOP and TOP-like mRNAs is hyper-sensitive to mTOR inhibition

(a) Torin1-induced changes in translational efficiencies of 65 known TOP mRNAs in WT MEFs (outlined bars) compared to changes in all 4840 mRNAs (solid bars). Significance determined by the Mann-Whitney U test. (b) The pyrimidine content of the 10 nt surrounding the TSS for 3025 mRNAs where the TSS could confidently identified, excluding 65 known TOP mRNAs (expected frequency = 0.518). Boxplots indicate the TSS pyrimidine content for mRNAs binned according to Torin1-dependent change in translational efficiency. Significance determined by binomial test. (c) Numbers of indicated mRNA classes. (d) TSS annotations for selected TOP and TOP-like mRNAs. Primary and secondary TSS locations from dbTSS (purple) are indicated, as are annotations from Refseq (gray), Ensemble (blue), and UCSC (green). (e) Polysome analyses of selected TOP (eEF2, Rps20), unrecognized TOP (Hsp90ab1) and TOP-like (Vim, Ybx1) mRNAs. Data are means +/− s.e.m. (n=2).

Torin1 also suppressed the translation of many mRNAs not previously defined as TOP mRNAs. After excluding known TOP mRNAs from analysis, we found that the 10 nucleotides surrounding the predominant transcriptional start site (TSS) in the mRNAs most suppressed by mTOR inhibition were still highly enriched for pyrimidines (Fig. 2b). This enrichment could reflect the presence of previously undocumented TOP motifs and/or of similar motifs that do not meet the TOP definition. We used the database of transcriptional start sites (dbTSS)[19] as well as the Refseq, Ensembl, and UCSC resources to examine the transcriptional start sites (TSS) of the 100 mRNAs most translationally suppressed by mTOR inhibition. 57 of these were known TOP mRNAs, and, of the remaining 43, 15 had previously unrecognized TOP motifs while 13 contained a stretch of pyrimidines that was near but did not begin at the most frequent TSS. As this suggested that the established TOP motif definition might be too conservative we defined a relaxed TOP-like motif consisting of a stretch of at least 5 pyrimidines within 4 nt of the most frequent TSS. Although this motif was relatively common amongst all TSSs (freq = 0.16), it was highly enriched amongst the most suppressed mRNAs and significantly depleted amongst mRNAs with a greater than average increase in translational efficiency following mTOR inhibition (Fisher’s exact test p-value = 3.1 × 10−8; Supplementary Fig. 2b). Remarkably, we found that 85 of the 100 mRNAs most sensitive to mTOR inhibition are either known TOP mRNAs or contain an unrecognized TOP or TOP-like motif (Fig. 2c,d; Supplementary Table 3). SeveralmRNAs that failed to meet our criteria contain pyrimidine sequences interrupted by a single purine (e.g., Hspa8), suggesting that even our TOP-like definition may be too conservative.

Like established TOP mRNAs, many previously unrecognized TOP and TOP-like mRNAs encode proteins with roles in protein synthesis (Supplementary Table 3) while others point to new effectors of the mTORC1 pathway (Fig. 2d). For instance, vimentin and Ybx1 participate in the epithelial-mesenchymal transition, a process known to be affected by mTOR inhibition[20,21]. By analyzing polysome profiles prepared from Torin1-treated cells, we confirmed that several unrecognized TOP (Hsp90ab1) or TOP-like mRNAs (Vim, Ybx1) were depleted from polysome fractions as strongly as established TOP mRNAs (Rps20, eEF2) (Fig. 2e). Furthermore, TOP-like and TOP motifs conferred similar degrees of mTOR-dependent translation control when placed upstream of a luciferase reporter (Supplementary Fig. 4a, b, d). Because some TOP-like mRNAs may be mis-annotated and actually contain canonical TOP motifs, we in vitro transcribed capped mRNA beginning with a single purine followed by a pyrimidine sequence and found that, like TOP mRNAs, it was translated less efficiently than an mRNA lacking this motif when mTOR was inhibited (Supplementary Fig. 4e, f). Thus, TOP or TOP-like motifs are more numerous than previously recognized and define the vast majority of mRNAs highly dependent on mTOR for translation.

How mTOR regulates TOP mRNA translation has been a persistent mystery. The S6Ks were originally considered key mediators, but later studies did not support this possibility[22,23]. Because TOP mRNA translation is less inhibited by rapamycin than dual-mTOR/PI3K inhibitors and RNAi-mediated mTOR suppression[4], we suspected that it might be regulated through the 4E-BPs, which mTORC1 phosphorylates in a largely rapamycin-resistant fashion[2,24,25]. In 4E-BP1/2 double-knockout MEFs (DKO), Torin1 had no effect on the interaction of eIF4E with eIF4G1 (Fig. 3a, b). Furthermore, in DKO cells, Torin1 had a minimal effect on [35]S-Cys/Met incorporation and did not perceptibly shift ribosomes out of polysomes (Fig. 3c, d), indicating that the 4E-BPs mediate a large part of mTOR-dependent control of general translation. Moreover, ribosome profiling of vehicle- and Torin1-treated DKO cells revealed that the distribution of Torin1-induced changes in translational efficiency was much narrower in DKO (σ = 0.225) than in WT (σ = 0.401) cells (Fig. 3e), arguing that the 4E-BPs are also required for the largesttranslational effects caused by mTOR inhibition. Indeed, as monitored by ribosome profiling, established TOP mRNAs were barely inhibited by Torin1 in DKO cells (Fig. 3f), which we confirmed by polysome analysis of individual mRNAs in MEFs (Fig. 3g) and in HeLa cells with RNAi-mediated knockdown of 4E-BP1 (Supplementary Fig. 5). Expression of a dominant negative 4EBP1-4A mutant, as well as RNAi-mediated depletion of eIF4E, were sufficient to selectively inhibit TOP mRNA translation in actively growing cells (Supplementary Fig. 6). Expression of the 4EBP1-4A mutant suppressed the translation of TOP reporter constructs as well (Supplementary Fig. 4c). We found no evidence that previously identified pyrimidine-binding proteins, such as Tia1/R or La, play a role in the selective regulation of TOP mRNAs by mTORC1 (Supplementary Fig. 7). However, we cannot rule out a role for these proteins in the amino acid regulation of TOP mRNA translation, which is maintained in DKO cells likely through the Gcn2 pathway (Supplementary Fig. 8). These results argue that the translation of mRNAs with TOP and TOP-like motifs is highly sensitive to 4E-BP phosphorylation, and that this is the basis of their regulation by mTORC1.

Figure 3

mTOR regulates general protein synthesis and TOP mRNA translation through the 4E-BPs

(a) WT and 4EBP1/2 double-knockout (DKO) MEFs were treated with DMSO, 250 nM rapamycin or Torin1 for 2 h, lysates subjected to m[7]GTP pull-downs, and analyzed for levels of indicated proteins. (b) WT and DKO MEFs expressing FLAG-GFP or FLAG-eIF4E were treated as in (a), and immunoprecipitates analyzed for indicated proteins. (c) DKO MEFs were treated for 2 h with vehicle (DMSO), 250 nM rapamycin or Torin1, or 10 μg/ml cycloheximide were analyzed as in Figure 1b. Data are mean +/− s.d. (n=3). (d) Polysome profiles of DKO MEFs treated with DMSO or Torin1 for 2 h (e) Torin1-dependent changes in translational efficiency in DKO (gray bars) and WT MEFs (blue bars). (f) Torin1-dependent translational suppression of 65 TOP mRNAs in WT and DKO MEFs. Significance determined by Mann-Whitney U test. (g) Polysome analyses of selected non-TOP (β-actin), known TOP (eEF2, Rps20), unrecognized TOP (Hsp90ab1) and TOP-like (Vim, Ybx1) mRNAs in DKO cells. Data are means +/− s.e.m. (n=2).

To understand why the translation of TOP and TOP-like mRNAs has a 4E-BP mediated hyper-dependence on mTOR, we considered the established functions of the 4E-BPs[3]. A key step in eIF4E-dependent initiation is the cooperative binding of eIF4E and eIF4G1 to mRNA, which nucleates the eIF4F complex[26]. eIF4G1 also interacts with eIF3, which orchestrates assembly of the 43S pre-initiation complex on the mRNA. When mTORC1 is inactive, dephosphorylated 4E-BPs bind to eIF4E and thereby prevent its association with eIF4G1 (Fig. 3a, b). mTOR inhibition also prevents the association of eIF4G1 with eIF3 in WT but, unexpectedly, not in DKO cells (Fig. 4a). Expression of the 4EBP1-4A mutant similarly disrupted the eIF4G1-eIF3 interaction (Supplementary Fig. 6b). Because destabilization of the eIF4F complex weakens the affinity of eIF4E for the mRNA cap[26], we hypothesized that mTOR inhibition might selectively impair the binding of eIF4E to TOP and TOP-like mRNAs. Indeed, Torin1 treatment of cells caused a selective loss of TOP and TOP-like mRNAs from eIF4E, which strongly correlated with their degree of translational suppression (Fig. 4b). Consistent with a special role for eIF4G1 in TOP mRNA translation, RNAi-mediated depletion of eIF4G1 in WT cells, which mimicked the effects of Torin1 on overall protein synthesis and polysome profiles, selectively suppressed the translation of TOP mRNAs, without affecting mTORC1 activity (Fig. 4c-f). Importantly, in the DKO cells, eIF4G1 depletion also selectively repressed TOP mRNA translation (Fig. 4c,e,f), consistent with eIF4G1 acting downstream of the 4E-BPs. A functionally redundant eIF4G1 homolog, eIF4G3, is not well expressed in the MEFs (Fig. 4g) and its loss had little effect on translation in HeLa cells (Supplementary Fig. 9). MEFs do express a distinct eIF4G1 homolog, DAP5/eIF4G2 (Fig. 4g), which does not bind eIF4E but still mediates a substantial fraction of protein synthesis[27,28]. While DAP5/eIF4G2 depletion significantly suppressed overall protein synthesis, it did not have selective effects on the translation of TOP mRNAs (Fig. 4a,d,e,h,i). Therefore, unlike other mRNAs, TOP mRNAs require eIF4G1 to anchor eIF4E to the cap, and this underlies their selective translational regulation by the 4E-BPs and mTORC1.

Figure 4

Destabilization of the eIF4E/eIF4G1 interaction dissociates TOP mRNAs from eIF4E and inhibits their translation

(a) WT and DKO MEFs were treated for 2 h with DMSO or 250 nM Torin1 and eIF3b immunoprecipitates analyzed for indicated proteins. (b) FLAG-eIF4E was immunoprecipitated from WT MEFs treated with DMSO or 250 nM Torin1 for 2 h. RNA was extracted, and abundance of TOP and TOP-like (TOP/L) (Eef2, Rps20, Hsp90ab1, Pabpc1, Ybx1, Vim) and non-TOP (Actb, Mrpl22, Ccnd1, Slc2a1, Gabarapl1, Myc) mRNAs quantified by QPCR. Changes in eIF4E binding of mRNAs were plotted against changes in translational efficiency from Fig. 1f. eIF4E binding data are means +/− s.e.m. (n=4). (c) Levels of indicated proteins in cells expressing indicated shRNAs. (d) Cells expressing indicated shRNAs were pulsed for 30 min with [35]S-labeled Cys/Met and analyzed as in Figure 1b. Data are mean +/− s.d. (n=3). Significance determined by t-test. (e) Polysome profiles for WT or DKO cells expressing indicated shRNAs. (f) RNA isolated from gradients in (e) was analyzed by qPCR for the indicated mRNAs as in Fig. 1e. Data are means +/− s.e.m. (n=2). (g) Abundance of indicated transcripts from RNA-seq analysis. Data are means +/− s.e.m. (n=3). (h) Lysates from cells expressing shGFP or eIF4G2-specific shRNAs were analyzed by immunoblotting. (i) Fractions from shEIF4G2-2 gradients in (e) were analyzed as in (f). (j) mTORC1 regulates the selective translation of TOP and TOP-like mRNAs through the 4EBP-dependent control of eIF4G1-mediated initiation.

We find that the effects of acute mTOR inhibition on mRNA translation are largely mediated by the 4E-BPs, including the moderate suppression of the translation of all mRNAs and the more striking inhibition of TOP and TOP-like mRNA translation. As the 4E-BPs are required for the mTORC1-dependent regulation of proliferation[16], the translational control of TOP mRNAs may play a fundamental role in this process (Fig. 4j) as well as in cancers associated with hyperactive mTOR signaling. We focused on suppressed mRNAs, but many other transcripts are translated with increased efficiency, and may be important for cellular survival under conditions of impaired mTORC1 signaling.

Methods Summary

To generate ribosome and mRNA profiling libraries, WT MEFs (4EBP1/2+/+; p53−/−) or DKO MEFS (4EBP1/2−/−; p53−/−) were treated with vehicle or 250 nM Torin1 for 2 h. Cellular extracts were partitioned for either ribosome profiling or mRNA profiling. Small RNA libraries were prepared according to established protocols[8] with some modifications, and analyzed by high-throughput sequencing. Transcript abundance was determined through an iterative alignment and mapping strategy to a non-redundant library of mouse transcripts based on Refseq definitions.