nanoCAGE reveals 5′ UTR features that define specific modes of translation of functionally related MTOR-sensitive mRNAs

1Genome-wide gene expression studies have mostly focused on measuring “steady-state” mRNA abundance that reflects alterations in transcription and mRNA stability. Changes in steady-state mRNA abundance do not, however, completely mirror those occurring in the corresponding proteome. Although still debated, these studies implicate post-transcriptional mechanisms, including mRNA translation, as key influencers over the proteome.

2Control of translation largely occurs at the initiation step, during which the mRNA is recruited to the ribosome by the eIF4F complex. This complex consists of the mRNA cap-binding protein EIF4E, a large scaffolding protein EIF4G, and DEAD box helicase EIF4A. eIF4F complex assembly is regulated by the mechanistic/mammalian target of rapamycin (mTOR) complex 1 (mTORC1), which phosphorylates and inactivates 4E-binding proteins (EIF4EBP1, 2, and 3). EIF4EBPs bind to EIF4E, thereby preventing EIF4E:EIF4G association. Phosphorylation of EIF4EBPs leads to their dissociation from EIF4E, allowing EIF4E:EIF4G interaction and eIF4F assembly. Although EIF4E is required for cap-dependent translation of all nuclear-encoded mRNAs, a subset of mRNAs characterized by long and complex 5′ UTRs that encode proliferation- (e.g., cyclins), survival- (e.g., BCL2 family members), and tumor-promoting proteins (e.g., MYC) are thought to be particularly sensitive to EIF4E.

3UTR features dictate translation efficiency. In mammals, mRNAs with long, complex 5′ UTRs exhibit a high EIF4A helicase requirement for scanning of 43S complex toward the initiation codon. EIF4A processivity, as a single protein, is low. EIF4A:EIF4G interaction dramatically increases EIF4A processivity. EIF4E recruits EIF4A to the eIF4F complex, thereby bolstering EIF4A activity. Recent ribosome-profiling studies indicate that the mTORC1/EIF4EBP/EIF4E pathway almost exclusively regulates translation of mRNAs harboring 5′ terminal oligopyrimidine (5′ TOP) and related motifs. The 5′ TOP motif consists of a C immediately after the mRNA cap, followed by 4–15 pyrimidines. TOP mRNAs encode translational machinery components, including ribosomal proteins and elongation factors.

4The generality of the conclusions that EIF4EBPs are major regulators of TOP mRNA translation has been questioned, because under a number of conditions, translation of TOP mRNAs, albeit MTOR-dependent, is EIF4EBP-independent. Consistent with this view, the effects of serum deprivation on TOP mRNA translation also appeared to be EIF4E-independent. In turn, LARP1 and TIA1/TIAL1 (also known as TIAR) were reported to mediate effects of mTORC1 on TOP mRNAs. Polysome-profiling showed that the mTORC1/EIF4EBP axis also regulates translation of non-TOP mRNAs, including those encoding mitochondrial and survival- and proliferation-promoting proteins. In fact, it appears that ribosome- and polysome-profiling studies provided strikingly different sets of MTOR-sensitive mRNAs.

5Elucidating the precise mechanisms that control mRNA translation also requires accurate knowledge of transcription start sites (TSSs). Despite substantial efforts to define TSSs using CAGE/nanoCAGE or TSS-seq, there are no resources that provide TSS information for the most commonly used model cell lines. Databases such as RefSeq or UCSC are thus routinely used to infer the link between 5′ UTR features and translation. These databases are thought to contain numerous errors, and for a more accurate understanding of the relationship between the 5′ UTR features and translational control, both TSSs and translation efficiency should be determined in the same cell.

6We compared ribosome- and polysome-profiling for their ability to identify mRNAs undergoing different shifts in translational efficiency. In polysome-profiling, translation efficiency is directly determined by separating efficiently (commonly defined as mRNAs associated with more than three ribosomes) and not efficiently translated mRNAs (defined as mRNAs associated with three or less ribosomes). In contrast, with ribosome-profiling translational efficiency is inferred based on the number of RNA sequencing reads corresponding to ribosome-protected fragments. TOP mRNAs are highly expressed (Fig. 1A,B), engage a considerable proportion of cellular ribosomes when MTOR is active, and show large shifts in translational efficiency upon MTOR inhibition, as compared to mRNAs that do not contain a 5′ TOP motif (non-TOP mRNAs) (Supplemental Fig. 1;; ).

7We therefore determined performance of ribosome- vs. polysome-profiling in identifying differentially translated mRNAs as a function of the size of shifts in translational efficiency using a simulation approach. Suitable parameters for simulation were obtained by comparisons of modeled normal distributions of ribosome association converted to sedimentation distances (Fig. 1C) to experimentally determined ribosome associations in insulin- vs. insulin + torin1-treated MCF7 cells (Supplemental Fig. 2A–C). Using these parameters, we then modeled a range of shifts in control vs. MTOR inhibition conditions (Fig. 1D) and estimated the mean number of ribosomes per mRNA (i.e., ribosome-profiling) or the proportion of the mRNAs that are associated with more than three ribosomes (i.e., polysome-profiling). We calculated fold-change differences and their p-values (control vs. MTOR inhibition) for each shift, for both polysome- and ribosome-profiling simulations. Because equal proportions of all shifts were represented in the simulation, the proportion of each shift among the smallest P-values or largest fold changes is expected to be the same. Using P-value-based ranking, polysome-profiling appeared not to be affected by the size of the shift in polysome association, while ribosome-profiling was biased toward identifying mRNAs that exhibit larger shifts (Fig. 1E–G). Similar bias was observed for polysome-profiling when fold-change analysis was employed, but polysome-profiling still outperformed ribosome-profiling (Fig. 1H,I). These analyses were stable inasmuch as modifying the simulation parameters did not alter the outcome (Supplemental Fig. 3).

8Using the same approach, we compared performance of ribosome- vs. polysome-profiling in identifying changes of non-TOP vs. TOP mRNA translation as a function of MTOR activity. Parameters were determined from experimentally tested TOP and non-TOP mRNAs (Supplemental Figs. 1, 2). Shifts to sub-80S fractions cannot be modeled using normal distributions, and the mean of the distribution under MTOR inhibition was therefore set to one ribosome. While polysome-profiling identified expected equal proportions of TOP and non-TOP mRNAs among genes with the lowest P-values, ribosome-profiling again favored identification of TOP mRNAs as differentially translated (Supplemental Fig. 4A). Analysis based on fold changes showed bias for identification of TOP mRNAs over non-TOP mRNAs as differentially translated for both polysome- and ribosome-profiling, but the bias toward identifying TOP mRNAs as differentially translated was still less pronounced for polysome-profiling (Supplemental Fig. 4B,C). Thus, ribosome-profiling shows a bias which favors identification of mRNAs that exhibit larger shifts in polysomes as translationally regulated, which is the case for TOP mRNAs after modulation of MTOR activity.

9Gene expression is commonly quantified by relative rather than absolute measurements. Accordingly, changes in each specific mRNA translation are measured in relation to the mean effect on translation of all mRNAs. To account for this, we performed simulations using relative translation efficacies and included invariant mRNAs (i.e., those mRNAs that did not appear to change significantly their translation in response to MTOR inhibition in both ribosome- and polysome-profiling studies, hereafter referred to as “nonmovers”). Simulation parameters for ribosome association for nonmovers were obtained by fitting a normal distribution to the polysome-profile (Supplemental Fig. 2A). Changes in translation were then simulated as described above for TOP and non-TOP mRNAs while including nonmovers. We assessed how increasing the proportion of TOP mRNAs that changed their translation impacts predicted fold changes for TOP- and non-TOP mRNAs. This was done before and after normalization by global centering (which is equivalent to an absolute and relative assessment of effects on translation, respectively). For relative measurements, when an increased proportion of TOP mRNAs were differentially translated, there was an underestimation of the fold changes for both TOP and non-TOP mRNAs (Fig. 1J). This effect was more pronounced for ribosome-profiling, e.g., if ∼3.5% of all mRNAs are TOP mRNAs and they changed translation, non-TOP mRNAs will show a less than twofold difference, which is a commonly used threshold.

10As MTOR inhibition also involves a global down-regulation of protein synthesis, we assessed how this would impact on TOP and non-TOP mRNA translation. For this simulation we used the parameters identified for abundance and proportions of TOP mRNAs, non-TOP mRNAs, and nonmovers (Fig. 1A), and gradually introduced a reduction in translation for nonmover mRNAs to simulate reduction in global protein synthesis caused by MTOR inhibition (Fig. 1K). In this case, as above, non-TOP mRNAs exhibited a less than twofold change when ribosome-profiling was simulated. Altogether, methodological features of ribosome-profiling introduce a bias that obscures identification of changes in non-TOP mRNA translation. This is because non-TOP mRNAs show smaller shifts and lower abundance and are therefore more sensitive to global changes in protein synthesis as compared to TOP mRNAs. This is likely a key factor underpinning the discrepancies between the repertoires of MTOR-regulated mRNAs observed in ribosome- vs. polysome-profiling studies.

11Optimal RNA-seq depth under control conditions (i.e., vehicle- or mock-treated cells) is a prerequisite to identify a reduction in translation induced by a treatment. We observed that <2000 genes consistently had >256 ribosome-profiling counts in the control condition (Supplemental Fig. 5A) across the two replicates reported in. Indeed, low read counts strongly influenced the signal to variance relationship, and the variance strongly increased at <64 reads (Supplemental Figs. 5B, 6A). Using edgeR analysis (see below), we observed that the vast majority of MTOR-sensitive mRNAs did not shift below the <64-read threshold upon MTOR inhibition (Supplemental Fig. 5B). This suggests that only abundant mRNAs (e.g., TOP mRNAs) (Fig. 1A,B) would be identified as being translationally suppressed by MTOR inhibitors (MTOR-i) in this study. Indeed, <10% of non-TOP mRNAs, identified as being MTOR-sensitive in a polysome-profiling study, consistently obtained >256 ribosome-profiling counts in the control condition in. Alternative algorithms corroborated these findings (Supplemental Fig. 6B). A second ribosome-profiling data set investigating MTOR-sensitive translation also suffered from this type of low sensitivity and could not identify non-TOP mRNAs as being differentially translated (Supplemental Figs. 5C,D, 7A,B). Thus, despite the differences in cell lines and conditions that were used in respective studies, discrepancies in the observed effects of MTOR-i on the translatome between ribosome- and polysome-profiling studies appear to stem from technical biases and the low sensitivity in ribosome-profiling studies, and our simulation analysis indicates that, in addition to TOP mRNAs, MTOR regulates translation of non-TOP mRNAs.

12The above analysis indicates that other 5′ UTR features beyond 5′ TOP motifs must render mRNA translation MTOR-sensitive. We employed nanoCAGE sequencing to map TSSs and 5′ UTRs to investigate this hypothesis (Fig. 2A). When we reached a sequencing “plateau,” where few additional genes or TSSs were identified upon additional sequencing (Fig. 2B), we had obtained 5′ UTR sequence information for ∼12,000 RefSeq transcripts, corresponding to ∼6500 genes in MCF7 cells maintained under the control, mock-treatment condition used when examining MTOR-sensitive translation by polysome-profiling (Supplemental Table S1; ). Examination of a selection of TOP mRNAs showed that nanoCAGE accurately captured TSSs giving rise to mRNAs with TOP motifs also when these were located downstream from the TSS suggested by RefSeq (Fig. 2C). TSSs identified by nanoCAGE confirmed that most of the MTOR-regulated mRNAs identified by polysome-profiling (PP242-sensitive) do not contain a 5′ TOP motif (∼90%) (Fig. 2D).

13Determination of 5′ UTR lengths using the most prevalent peak length (i.e., the length suggested by the position with the most RNA-seq reads) or the weighted mean length (allowing all RNA-seq reads to contribute to a mean) revealed that ∼30% 5′ UTRs identified by nanoCAGE were half the length or shorter than those described in the RefSeq database (Fig. 2E,F; Supplemental Fig. 8A,B). Among the MTOR-sensitive mRNAs identified by polysome-profiling, the TSSs determined by nanoCAGE, but not RefSeq, indicated a set of transcripts with extremely short 5′ UTRs (<30 nucleotides [nt]). This set was enriched in mRNAs encoding proteins with mitochondrial functions (e.g., ATP5O, ATP5G1) (Fig. 3A,B), with one exception, RDH13, which had a long 5′ UTR. These were in stark contrast to cell cycle- and survival-promoting mRNAs with long 5′ UTRs (e.g., CCND3, ODC1, MCL1, BIRC5 [survival] and MYC) (Fig. 3C). mRNAs with short 5′ UTRs, encoding proteins with mitochondrial function, were enriched (4.7-fold enrichment, P = 0.03) for translation initiator of short 5′ UTR (TISU) elements (Fig. 3D,E; ). Thus, in addition to TOP mRNAs, MTOR-sensitive mRNAs include non-TOP mRNAs with long 5′ UTRs, which encode proliferation- and survival-promoting proteins, and a subset of mRNAs with short (<30 nt) 5′ UTRs that were enriched with TISU elements (Fig. 3A–E).

14TISU elements render mRNA translation sensitive to MTOR but not to EIF4A inhibition. We therefore depleted EIF4E or EIF4A1 in MCF7 and HEK293E cells and monitored expression of proteins encoded by MTOR-sensitive mRNAs with either long (CCND3, MCL1, BIRC5, and BCL2) or short (ATP5O, ATP5G1, NDUFS6, and UQCC2) 5′ UTRs. Whereas EIF4E depletion decreased expression of both subsets of proteins, reduction in EIF4A1 selectively affected proteins encoded by long 5′ UTR mRNAs (Fig. 4A; Supplemental Fig. 10A). We replicated these results using the active-site MTOR-i torin1 and two chemically distinct EIF4A inhibitors (EIF4A-i), hippuristanol and silvestrol. As expected, treatment with torin1 abolished mTORC1 signaling, whereas hippuristanol slightly induced mTORC1 activity as illustrated by changes in EIF4EBP1 and RPS6 phosphorylation (Supplemental Fig. 10B). Torin1 increased EIF4E:EIF4EBP1 and reduced EIF4E:EIF4G1 binding. In contrast, EIF4E:EIF4G1 association was unaffected by hippuristanol (Supplemental Fig. 10C). Thus, as noted following depletion of EIF4E, torin1 reduced expression of proteins encoded by both short and long 5′ UTR MTOR-sensitive mRNAs, whereas EIF4A-i selectively down-regulated proteins encoded by long 5′ UTR mRNAs (Supplemental Fig. 10B). These effects appeared to be mediated by the EIF4A1 isoform, as EIF4A2 depletion failed to decrease expression of CCND3, which is encoded by a prototypical long 5′ UTR mRNA (Supplemental Fig. 10D). As observed previously, EIF4A1 depletion led to an increase in EIF4A2 levels (Supplemental Fig. 10D; ). Collectively, these results demonstrate that EIF4A1 regulates expression of proteins encoded by a subset of MTOR-sensitive mRNAs harboring long, but not short, 5′ UTRs.

15Depletion of EIF4E and EIF4A1 did not affect steady-state levels of mRNAs with either long (CCND3, MCL1, BIRC5, and BCL2) or short (ATP5O, ATP5G1, NDUFS6, and UQCC2) 5′ UTRs (Fig. 4B). Under these conditions, the effects of EIF4E or EIF4A1 depletion on global mRNA translation were comparable (Fig. 4C). Whereas EIF4E down-regulation decreased translation of both long and short 5′ UTR mRNAs, EIF4A1 depletion decreased translation of mRNAs with long 5′ UTRs but only marginally altered translation of mRNAs with short 5′ UTRs (Fig. 4D; Supplemental Fig. 11A). Neither EIF4A1 nor EIF4E depletion altered distribution of MTOR-insensitive mRNAs, e.g., ACTB (beta actin) and GAPDH mRNAs (Fig. 4D; Supplemental Fig. 11A). Parallel results were obtained using torin1 and hippuristanol (Supplemental Fig. 11B,C).

16To further identify 5′ UTR features that render MTOR-sensitive mRNAs largely insensitive to changes in EIF4A1 activity, HEK293E cells were cotransfected with a Renilla (control) and a variety of firefly (FF) luciferase reporters. FF reporters included IRF7 5′ UTR (IRF7[5′UTR]-FF), which is long and translated in an EIF4E-dependent manner (Fig. 5A; ), ATP5O 5′ UTR (ATP5O[5′UTR]-FF), which contains a 4-nt portion of the TISU element upstream of the initiation codon (Fig. 5B), and ATP5O with a complete TISU element, including a 5-nt portion distal to the initiation codon (ATP5O[TISU]-FF) (Fig. 5C). ATP5O(5′UTR)-FF therefore harbors a disrupted TISU element whereby the TISU element sequence is substituted with GAGAA (the first 5 nt of the FF luciferase ORF) downstream from the initiation codon. EIF4E down-regulation strongly decreased expression of all three FF reporters. In contrast, while EIF4A1 depletion strongly repressed luciferase activity of IRF7-FF, it only had a moderate effect on ATP5O(5′UTR)-FF (1.2-fold reduction vs. 6.6-fold observed in EIF4E-depleted cells) and did not significantly affect luciferase activity driven by ATP5O(TISU)-FF (Fig. 5A–C). Similar results were obtained with torin1 and hippuristanol (Supplemental Fig. 12A–C). As TISU element integrity did not appear to exert a major influence on EIF4A1- or EIF4E-sensitive translation (Fig. 5B,C; Supplemental Fig. 12B,C), we also monitored expression of FF luciferase driven by constructs harboring the UQCC2 5′ UTR that does not contain a TISU and NDUFS6 5′ UTR, but which, similar to ATP5O(5′UTR)-FF, only contains a portion of the TISU element upstream of the initiation codon. In both cases, results paralleled those observed using the ATP5O reporters (Fig. 5D,E). Increasing secondary structure and length by adding a stable stem–loop structure (ATP5O[5′UTR]-SL-FF) rendered the reporter both EIF4E- and EIF4A1-sensitive (Fig. 5F). Importantly, under all conditions listed above, control Renilla luciferase expression was largely insensitive to both EIF4E and EIF4A1 (Supplemental Figs. 13, 14). These data demonstrate that translation of MTOR-sensitive mRNAs with short 5′ UTRs is strongly sensitive to EIF4E, but not EIF4A1, while there is no obligatory requirement for an intact TISU element.

17mTORC1/EIF4EBP1/EIF4E axis bolsters mitochondrial biogenesis and ATP production by stimulating translation of nuclear-encoded mRNAs with mitochondrial functions, including respiratory chain components, e.g., ATP5O. Here, nanoCAGE identified a subset of MTOR-sensitive mRNAs encoding mitochondria-related proteins characterized by short 5′ UTRs (Fig. 3B–D), which is consistent with the findings that mRNAs encoding mitochondrial proteins frequently contain TISU elements. As translation of mRNAs with short 5′ UTRs is largely EIF4A-independent, we postulated that short-term EIF4A1 inhibition would have a lesser impact on mitochondrial number and function as compared to MTOR inhibition. To directly compare the effects of MTOR-i vs. EIF4A-i, we selected concentrations of torin1 and silvestrol that comparably suppressed HEK293E cell proliferation (∼60% of control) (Fig. 6A, left panel). After 16 h (a time period we previously found to be suitable for detecting the impact of MTOR inhibition on cell mitochondrial content and respiration), torin1, but not silvestrol, reduced mitochondrial content (Fig. 6A, right panel).

18In addition, although both torin1 (1.9-fold) and silvestrol (1.4-fold) reduced mitochondrial respiration, the effect of torin1 was more substantial (Fig. 6B). Similar findings were obtained using siRNA directed against EIF4E or EIF4A1 (Supplemental Fig. 15A). Furthermore, whereas EIF4A1 down-regulation slightly elevated mitochondrial respiration relative to the control, mitochondrial respiration was dramatically inhibited by EIF4E depletion (Supplemental Fig. 15B). Therefore, short-term treatment with MTOR-i suppresses mitochondrial respiration more dramatically than EIF4A-i, which is consistent with the disparity of their effects on translation of mRNAs harboring short 5′ UTRs, i.e., those encoding proteins involved in mitochondrial respiration and functions.

19Inhibition of the mTORC1/EIF4EBP1/EIF4E axis coincides with down-regulation of proteins implicated in mitochondrial respiratory function (e.g., ATP5O, ATP5G1, UQCC2, NDUFS6) as well as in the maintenance of the integrity of the mitochondrial outer membrane (e.g., BCL2, MCL1, and BIRC5) (Fig. 4A). In contrast, suppression of EIF4A1 activity selectively affects proteins that maintain mitochondrial outer membrane integrity (Fig. 4A). This explains why MTOR inhibitors mostly have a cytostatic effect and suggests that prolonged EIF4A1 inhibition may trigger mitochondrial-mediated apoptosis by disrupting the coordinated expression of proteins with mitochondrial metabolic functions from those regulating mitochondrial structural integrity. We therefore treated HEK293E cells with torin1 or silvestrol for 72 h and measured mitochondrial membrane potential using TMRE (tetramethylrhodamine, ethyl ester), a cell-permeant dye that accumulates in active but not depolarized mitochondria. As a positive control, cells were treated with carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP), a compound that depolarizes mitochondria (Fig. 7A). Exposure to silvestrol (72 h) markedly induced depolarization of mitochondria (Fig. 7A), and this correlated with induction of apoptosis as compared to control or torin1-treated cells (Fig. 7B). Similar results were observed in nutrient-deprived cells depleted of EIF4E or EIF4A1 (Supplemental Fig. 15C,D). These observations indicate that suppression of translation of both long and short 5′ UTR MTOR-sensitive mRNAs by torin1 yields a predominantly cytostatic outcome, whereas suppression of EIF4A1 function induces apoptosis by exclusively inhibiting translation of mRNAs with long 5′ UTRs encoding proteins that maintain mitochondrial structural integrity.

20MTOR inhibition increases autophagy, which eliminates depolarized mitochondria. To determine whether EIF4A1 inhibition affects autophagy, we monitored autophagic flux by treating HEK293E cells with torin1 or silvestrol in the presence of chloroquine. Chloroquine prevents autophagosome-lysosome fusion causing accumulation of the autophagic markers MAP1LC3A-II (LC3A-II) and MAP1LC3B-II (LC3B-II). While torin1 induced autophagy, as evidenced by elevated levels of both MAP1LC3A-II and MAP1LC3B-II (Fig. 7C), silvestrol prevented accumulation of MAP1LC3A-II and MAP1LC3B-II induced by chloroquine (Fig. 7C). By inhibiting the lysosomal pH gradient, chloroquine impairs mTORC1 activation, as illustrated by the decrease in the phosphorylation of EIF4EBP1 and RPS6 after 16 h (Fig. 7C). mTORC1 inhibition by chloroquine was partially reversed by silvestrol (Fig. 7C). This is consistent with induction of mTORC1 in response to EIF4A1 inhibition (Supplemental Fig. 10B; ). Therefore, in addition to inducing mitochondrial depolarization by altering mRNA translation, EIF4A-i prevents induction of autophagy and elimination of depolarized mitochondria.

21The advantage of ribosome- over polysome-profiling is that ribosome-profiling allows determination of ribosome positioning on mRNA at single-nucleotide resolution. However, there are a number of challenges regarding interpretation of ribosome-profiling. The relative performance of ribosome- vs. polysome-profiling to study changes in translation efficiency also remains unclear. mRNAs exhibit different translation properties, in part conferred by their 5′ UTR features. We show that ribosome-profiling studies should be interpreted with caution when simultaneously studying subclasses of mRNAs with disparate translation properties. Ribosome-profiling efficiently identifies changes in translation of highly abundant mRNAs that show large shifts in polysome association (“class A mRNAs”) but is far less sensitive in detecting less abundant mRNAs that shift less (“class B mRNAs”). TOP mRNAs, upon MTOR inhibition, shift from heavy polysomes to subpolysomal fractions (i.e., “class A mRNAs”), whereas the majority of non-TOP mRNAs transit from intermediate-heavy to light polysomes (i.e., “class B mRNAs”) (Supplemental Fig. 1). In contrast, polysome-profiling captures the translational changes in both classes of mRNAs. This may explain discrepancies between the catalogs of MTOR-regulated mRNAs between ribosome- vs. polysome-profiling studies. Consistently, while ribosome-profiling led to the conclusion that the mTOR/EIF4EBP1/EIF4E axis exclusively controls TOP and TOP-like mRNA translation, the same group found that EIF4E predominantly affects translation of non-TOP mRNAs implicated in ROS clearance using polysome-profiling.

22Length, and plausibly other 5′ UTR features, distinguishes two separate subgroups of MTOR-sensitive non-TOP mRNAs that are functionally distinct. First, there are those mRNAs harboring long 5′ UTRs that encode survival- and proliferation-promoting factors and are strongly affected by EIF4E and EIF4A1. Second, there are mRNAs that harbor short 5′ UTRs/TISU elements, translated in an EIF4E- but not an EIF4A1-sensitive manner, enriched in those encoding mitochondria-related proteins. In mammals, EIF4A1 selectively stimulates translation of mRNAs that encode survival-promoting proteins (e.g., BCL2, MCL1, and BIRC5) without altering translation of mitochondria-related short 5′ UTR mRNAs. In yeast, Ded1 (the ortholog of DDX3X in mammals), rather than EIF4A, seems to act as a major helicase that resolves secondary 5′ UTR structures. Accordingly, in yeast, changes in EIF4A activity uniformly affect translation of most mRNAs, while modulation of Ded1 primarily affects those with complex 5′ UTRs. Several features that distinguish yeast and mammalian EIF4A may explain these distinctions. This includes differences in the apparent stability of the eIF4F complex, as EIF4E:EIF4G but not EIF4E:EIF4G:EIF4A complexes can be purified from yeast. Moreover, MTOR-regulated auxiliary factor EIF4B significantly bolsters RNA-dependent ATPase activity of mammalian but not yeast EIF4A. It is also plausible that helicases thought to be required to unwind exceptionally strong 5′ UTR secondary structures (e.g., DHX29 and DDX3X) provide an additional level of diversity to MTOR-dependent translational regulation. However, the signaling mechanisms that regulate DHX29 and DDX3X function remain to be established.

23Translation of TISU mRNAs is 5′ cap-dependent, resistant to glucose deprivation but not to rapamycin, and requires EIF1:EIF4G1 interaction. We demonstrate that MTOR-i or EIF4E depletion, but not inhibition of EIF4A1, suppresses translation of short 5′ UTR mRNAs (<30 nt). This suggests that EIF4E is key for translation of mRNAs containing short 5′ UTRs and TISU elements, which is postulated to occur by facilitating recruitment of EIF4G1 and which does not appear to be affected by EIF4A-i (Supplemental Fig. 10C). Moreover, MTOR and EIF4E seem to regulate translation of ATP5O mRNAs independently of the TISU element downstream from the initiation codon, as well as translation of UQCC2 mRNA that comprises a short 5′ UTR without a TISU (Fig. 5B–D). This suggests that the length of the 5′ UTR, rather than the presence of a TISU, is the primary factor that determines EIF4E- versus EIF4A1-sensitivity.

24mTORC1 modulates translation of TOP mRNAs via LARP1, whereas synthesis of mitochondrial, proliferation- and survival-promoting proteins is controlled by the EIF4EBP/EIF4E pathway. Notably, 5′ UTR features further separate non-TOP MTOR-sensitive mRNAs into two distinctive groups that differ in their EIF4A1-sensitivity. MTOR-i prevent translation of short and long 5′ UTR mRNAs diminishing mitochondrial functions, accompanied by metabolic quiescence and cytostasis, whereas selective suppression of translation of a subset of long but not short 5′ UTR mRNAs by EIF4A-i leads to mitochondrial depolarization and apoptosis (Fig. 7D). These findings suggest that mTORC1 deploys distinct effectors to modulate translation of a variety of functional groups of mRNAs far beyond TOP mRNAs to regulate cellular homeostasis in response to a variety of extracellular stimuli and intracellular cues.

25In addition to the differential effects of MTOR-i and EIF4A-i on translation, it is likely that these compounds exert opposing metabolic effects. MTOR-i reduce mitochondrial ATP production while suppressing ATP consumption by anabolic processes and stimulating autophagy, which facilitates the replenishment of a variety of cellular building blocks. Although EIF4A-i may also decrease ATP consumption by the translational machinery, they are likely to stimulate other anabolic processes via increasing mTORC1 activity (Supplemental Fig. 10B; ). Contrary to previous reports, the effects of silvestrol on tumor cell viability cannot be accounted for by modulation of p-EIF4E status. EIF4A-i may therefore avoid compensatory effects seen with MTOR-i, resulting in a more effective precipitation of mitochondrial dysfunction and cell death. Indeed, unlike MTOR-i, silvestrol attenuated autophagy, which is considered a major mechanism for nutrient replenishment and clearance of depolarized mitochondria.

26In conclusion, important 5′ UTR mRNA variations determine MTOR-sensitivity. Current databases do not always accurately reflect the 5′ UTR features, while certain analytical biases currently limit the accuracy of ribosome-profiling techniques for monitoring changes in translational efficiency. Finally, two functionally distinct subgroups of non-TOP MTOR-sensitive mRNAs provide the molecular basis for the superior anti-neoplastic effects of EIF4A-i over MTOR-i.

27Distributions of ribosome associations were obtained by sampling from a normal distribution. These were then converted to mean number of ribosomes (equivalent to ribosome-profiling) or proportions of mRNAs associated with more than three ribosomes (equivalent to polysome-profiling), followed by calculation of fold changes and P-values that were compared across polysome- and ribosome-profiling. Detailed protocols for simulations are provided in the Supplemental Material.

28NanoCAGE libraries were prepared and analyzed as described but with a number of modifications that are described in detail in the Supplemental Material.

29HEK293E and MCF7 cells (ATCC) were maintained in Dulbecco's modified Eagle's medium (DMEM; Wisent) and RPMI-1640 (Wisent), respectively, supplemented with 10% fetal bovine serum (Wisent), 1% penicillin/streptomycin (Wisent), and 1% L-glutamine (Wisent). Where indicated, cells were treated with 250 nM torin1 (Tocris), 1 μM hippuristanol, and 25 nM silvestrol or the same volume of vehicle (DMSO). Lists of siRNAs, shRNAs antibodies, and dilutions that were used in Western blots are provided in the Supplemental Material.

30MCF7 cells were seeded in a 15-cm Petri dish, harvested at 80% confluency, and lysed in hypotonic lysis buffer (5 mM Tris HCl, pH 7.5, 2.5 mM MgCl2, 1.5 mM KCl, 100 μg/mL cycloheximide, 2 mM DTT, 0.5% Triton, 0.5% sodium deoxycholate). Polysome-profiling was carried out as described. RT-qPCR was performed as previously described. Fractions were collected as described in, and RNA was extracted using TRIzol according to the manufacturer's instructions. RT and qPCR were performed by SuperScript III Reverse Transcriptase and Fast SYBR Green Mastermix (Invitrogen), respectively. Experiments were done at least in independent duplicates (n = 2) whereby every sample was analyzed in a technical triplicate. Analyses were carried out using a relative standard curve method as described in http://www3.appliedbiosystems.com/cms/groups/mcb_support/documents/generaldocuments/cms_040980.pdf. Primers are listed in the Supplemental Material.

31We performed one-way or two-way ANOVA in R as indicated in the figure legends.

32nanoCAGE data from this study have been submitted to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE77033.