MNKs act as a regulatory switch for eIF4E1 and eIF4E3 driven mRNA translation in DLBCL

The phosphorylation of eIF4E1 at serine 209 by MNK1 or MNK2 has been shown to initiate oncogenic mRNA translation, a process that favours cancer development and maintenance. Here, we interrogate the MNK-eIF4E axis in diffuse large B-cell lymphoma (DLBCL) and show a distinct distribution of MNK1 and MNK2 in germinal centre B-cell (GCB) and activated B-cell (ABC) DLBCL. Despite displaying a differential distribution in GCB and ABC, both MNKs functionally complement each other to sustain cell survival. MNK inhibition ablates eIF4E1 phosphorylation and concurrently enhances eIF4E3 expression. Loss of MNK protein itself downregulates total eIF4E1 protein level by reducing eIF4E1 mRNA polysomal loading without affecting total mRNA level or stability. Enhanced eIF4E3 expression marginally suppresses eIF4E1-driven translation but exhibits a unique translatome that unveils a novel role for eIF4E3 in translation initiation. We propose that MNKs can modulate oncogenic translation by regulating eIF4E1-eIF4E3 levels and activity in DLBCL.

T here are multiple aetiologies to cancer development and maintenance, which essentially exert a selection pressure for enhancing oncogenic gene expression and/or reducing tumour suppressor activity. The transformation process leading up to malignancy is a result of a biased cellular homeostasis that favours increased and uncontrolled growth and proliferation. One of the least explored yet fundamentally important cellular processes that controls oncogenic transcript selection and expression is mRNA translation. Dysregulation of the translation process can alter the cellular landscape that can lead to cancer initiation, maintenance, progression, invasion and metastasis 1,2 . Cap-dependent translation is the primary mechanism of mRNA translation in eukaryotic cells. The most common member of the cap-dependent translation machinery that is often upregulated in cancer is eukaryotic translation initiation factor 4E 1 (eIF4E1), a 25-kDa protein that serves to initiate cap-dependent translation via mRNA cap binding, a highly regulated rate-limiting step in translation initiation. eIF4E1 functions to bridge mRNA to the ribosome via the eIF4F complex assembly 1,3-7 . The oncogenic potential of eIF4E1 has been characterized in many model systems and is emerging as an attractive therapeutic target for cancer, giving rise to eIF4E/eIF4E-cap inhibitors like ISIS183750 and ribavirin in clinical trials 8 .
The sole upstream regulators of eIF4E1 phosphoactivation are mitogen-activated protein kinase (MAPK) interacting kinases 1 and 2 (MNK1 and MNK2), which operate by phosphorylating eIF4E1 at serine 209 (S209) when both eIF4E1 and MNK are positioned in close proximity to each other on binding to the scaffolding protein, eIF4G 9 . The regulation of MNKs, in turn, is modulated via ERK and p38 MAPKs, as predicted by the presence of a MAPK-binding domain in the C terminus of the longer alternate splice form of both MNKs [10][11][12] .
Although most work has proposed that ERK and p38 MAPKs function upstream of MNKs, a very recent work by Maimon et al. 13 proposed a downstream activation of p38 MAPKs by the longer isoform of MNK2 (that is, MNK2a). These authors described a tumour suppressive role for this MNK2a-p38 MAPKs activity and a pro-oncogenic role for p38-MNK2b axis 13 .
MNK2 exhibits higher basal activity, poor upstream control by ERK or p38 MAPKs, and is emerging as a strong chemoresistant candidate to rapalogues and gemcitabine treatments 12,[14][15][16][17] . A number of studies have established the importance of the MNK-eIF4E1 axis in several human malignancies. Surprisingly, mice lacking both MNKs exhibited normal survival with no obvious phenotype 18 . Although this finding highlighted the dispensable nature of MNK activity under normal physiological conditions, other studies employing both in vitro and in vivo approaches have provided substantial evidence that downregulation of MNK or eIF4E1 phosphorylation in cancer is favourable for tumour regression 15,19,20 .
There are three members in the eIF4E family, where two members, eIF4E1 and eIF4E2, have been shown to bind the 7-methyl-guanosine (m 7 G)-cap using the classical aromatic sandwich model. Owing to the weak cap-binding ability and lack of significant eIF4G association reported in earlier studies, eIF4E2 was not believed to initiate translation in normal cells 7,21 .
However, recent studies have demonstrated eIF4E2-directed translation under low-oxygen conditions, giving rise to a new perspective for eIF4E2-modulated protein synthesis in tumour hypoxia 7,[21][22][23] . The third member of this family, eIF4E3, was not believed to have cap-binding ability as its primary structure lacks one of the two aromatic residues needed to bind m 7 G-cap. However, a recent intriguing finding by Osborne et al. 7 using biophysical approaches showed that eIF4E3 is indeed able to bind m 7 G-cap in an atypical manner and exerts tumour suppressive effects in cells.
Here, we comprehensively interrogate the MNK-eIF4E axis in DLBCL, a highly aggressive and heterogeneous non-Hodgkin's lymphoma. We show that both MNKs complement each other for cell survival despite exhibiting a differential distribution in DLBCL subtypes. MNKs, via its kinase-dependent or independent roles, alter the ability for eIF4E1 and eIF4E3 to bind the mRNA cap structure, thus, displaying a capacity to 'switch' the cellular translatome.

Results
MNK1 and MNK2 are differentially expressed in DLBCL subtypes. The status of MNKs, their distribution and activity in DLBCL have not been described to date. Data mining from previously reported gene expression analyses demonstrate that DLBCL samples can exhibit a varied range of MNK1 and MNK2 expressions. However, there are no comparisons made to evaluate the relative abundance of MNK1 or MNK2 in any individual sample in a given study population (Supplementary Table 1, www.oncomine.org). We analysed a panel of DLBCL cell lines that are characterized as either ABC or GCB origin for MNK1 and MNK2 expression levels via semi-quantitative real-time PCR (RT-qPCR) 24 . MNK1 was expressed in both ABC-and GCB-DLBCL; however, a significantly stronger expression of MNK1 was evident in all GCB-DLBCL cell lines tested (Fig. 1a). In contrast, ABC-DLBCL exhibited a stronger MNK2 expression. Western blot analysis of these cell lines further confirmed that the discrete MNK distribution pattern was also consistent at the protein level (Fig. 1b). We extracted RNA from paraffinembedded primary lymphoma samples for RT-qPCR analysis 25 . These lymphoma samples were previously scored as ABC-or GCB-DLBCL by an independent pathologist based on immunohistochemical characterization 26 . Primary patient samples of GCB-DLBCL exhibited enhanced MNK1 expression consistent with cell line data, and, notably, undetectable MNK2 levels ( Fig. 1c,d). MNK2 was only detectable at measureable levels in primary ABC-DLBCL samples, in addition to MNK1 (Fig. 1d). Given the distinct MNK1 and MNK2 distribution pattern in ABC and GCB, we next asked whether both MNKs were important for cellular viability and one could complement the other in promoting survival. We found that reducing the expression of either MNK1 or MNK2 significantly reduced overall cell survival as reported previously 27 and, in the same experiments, when one MNK member is overexpressed while the other is knocked down (that is, when MNK2 is overexpressed and MNK1 is knocked down simultaneously, or vice versa), cell survival as well as phosphorylation of eIF4E1 in either case were significantly rescued ( Fig. 1e-h). The differential expression in ABC-and GCB-DLBCL notwithstanding, we find that MNK1 and MNK2 can complement each other functionally to maintain cellular survival. Similarly, enforced expression of MNK1-AA mutant (MNK1-phosphonull) that cannot be phosphorylated for downstream activity caused significant cell death in GM02184, a non-malignant B-cell line, and to a much higher degree in DLBCL cell lines Pfeiffer (GCB-DLBCL) and HLY-1 (ABC-DLBCL) (Fig. 1i), suggesting that expression of an inactive MNK (MNK1-AA) acts as a dominant negative mutant in both MNK1-or MNK2-expressing cells, reaffirming a common or competitive role for both MNKs. Immunohistochemistry staining of a primary human lymphoma tissue microarray revealed significant elevation of p-eIF4E1 staining in DLBCL in comparison with normal tissues, regardless of the ABC or GCB classifications (normal LN ¼ 20, ABC ¼ 35 and GCB ¼ 11 samples) (Supplementary Fig. 1d).
MNKs are regulated by p38 and not ERK in DLBCL. Two isoforms of each MNK have been identified in human; the longer isoforms, termed MNK1a and MNK2a, contain the MAPKbinding site in the C terminus, which allows binding and activation of MNKs by either ERK or p38 MAPKs. Although MNK1 and MNK2 show differences at the C terminus, both MNKs share high homology in the central catalytic domain and contain phosphorylation sites for the activation of its kinase function. Given that MNK activation is associated with either ERK or p38-MAPK as upstream regulators, we probed these two targets as potential regulators of MNK activity in an array of DLBCL cell lines and in GM02184. We find that MEK inhibition (using AZD6244), which reduced ERK phosphorylation, had no impact on MNK kinase activity in DLBCL, evidenced by no change in eIF4E1-S209 phosphorylation at any time point tested, ranging from 45 min to 72 h ( Fig. 2a and Supplementary Figs 1a,2). We concurrently tested the ERK2 inhibitor, no.76 (3-(2-aminoethyl)-5-((4-ethoxyphenyl) methylene)-2,4-thiazolidine-dione, HCl) and found corroborative results showing no impact of ERK on MNK activity in DLBCL ( Supplementary Fig. 2). Others have demonstrated protein phosphatase 2 (PP2A) to dephosphorylate MNK and eIF4E1 (ref. 28). We explored PP2A activation using FTY720 as a potential MNK inhibitor and found no impact on eIF4E1-S209 phosphorylation ( Supplementary Fig. 3).
We next examined the impact of p38 inhibition on MNK activity. Inhibition of p38 using VX702 showed a significant reduction of MNK and eIF4E1 phosphorylation within 1 h post treatment ( Fig. 2b and Supplementary Fig. 1b) and sustained eIF4E1 phosphorylation reduction 4 h after treatment (Fig. 2c,d). p38 inhibition also resulted in reduced MCL-1, further confirming the effective disruption of eIF4E1-driven translation via manipulation of the p38-MNK-eIF4E1 axis (Fig. 2c,e). GM02184   showed a moderate but significant p-eIF4E1 reduction at 1 h treatment ( Fig. 2b and S1b), but rapidly regained eIF4E1 phosphorylation by 4 h with no significant changes in MCL-1.
p38 inhibitor treatment of DLBCL cells resulted in growth inhibition, showing reduced viable cell number by trypan exclusion assay, without significant cell death (Fig. 2f).  We performed a CFSE-based assay to evaluate cell proliferative response. As illustrated in Fig. 2g (and Supplementary Fig. 4a), p38 inhibition resulted in a slower rate of cell proliferation, trailing at least one cell cycle behind vehicle-treated HLY-1 cells after 72 h. We concurrently performed cell cycle analysis of cells treated with vehicle or VX702 and found a marginal accumulation in the S-phase ( Supplementary Fig. 4b), however, with a significantly reduced BrdU incorporation ( Supplementary  Fig. 4c), demonstrating a reduced S-phase efficiency and prolonged cell cycle length following p38 inhibition in DLBCL.
To affirm the direct effect of p38 on the MNK-eIF4E1 axis, we generated cell lines stably expressing wild-type MNK1 and MNK2, as well as MNK1-TD (MNK1 phosphomimetic) and MNK1-AA (MNK1-phosphonull) mutants in HLY-1 parent line. Inhibition of p38 exerted significant effects on eIF4E1-S209 phosphorylation in cell lines expressing wild-type MNK1 or MNK2, consistent with previous reports 29 . However, in the MNK1-TD and MNK1-AA mutants, where MNK phosphorylation cannot be manipulated by p38, eIF4E1-S209 phosphorylation was not affected by p38 inhibitor, further confirming that p38 activity on MNKs and subsequent eIF4E1-S209 phosphorylation is dependent on its ability to phosphorylate MNK, and that p38 activity is upstream to MNKs in DLBCL (Fig. 2h,i). In these experiments, we also probed for MCL-1 as a readout for eIF4E1 activity as previously reported 19 . We found that MCL-1 was reduced in control and MNK2 overexpression HLY-1 cells treated with VX702. However, MNK-TD and MNK-AA mutants that have constitutively activated or inactivated phosphorylation site that bypass regulation of p38 did not exhibit a reduction in MCL-1, consistent with unchanged eIF4E1 phosphorylation level (Fig. 2i,j). Both MNK1 and MNK2 cells exhibited a significant reduction in p-eIF4E1 levels; however, only MNK2 cells showed a robust reduction in MCL-1. Both MNK1 and MNK2 cells exhibit higher basal levels of MNKs and p-eIF4E1; thus, one can argue that the effects of p38 inhibition were not sufficient for efficient reduction in MCL-1 readout. Although the fact that MNK2 cells showed significant MCL-1 reduction may, at first, appear to counter the argument, this reduction could be an effect of recently reported MNK2 upstream regulation of p38 (ref. 13). p38 inhibition using VX702 causes MNK2 inhibition, which in turn exerts a looped positive feedback of p38 inhibition to further accentuate the cycle of inhibition (Fig. 2k).
The mechanistic basis for the preferential upstream regulation of MNKs by p38 MAPKs and not ERK in DLBCL is still unknown. The regulation of MNKs could be tissue specific and be subjected to co-expression of other factors in the cell. Consistently, we find our data to be in agreement with murine B cells that are also regulated primarily by p38 MAPKs 30 .
MNKs differentially regulate eIF4E1 and eIF4E3. There have previously been a number of conflicting views on the importance of eIF4E1 phosphorylation status in initiating translation under basal conditions [18][19][20] . To explore how MNK activity and variable protein expression affect eIF4E1 and mRNA translation in DLBCL, we performed the following experiments. First, we established a serial knockdown of MNK1 and MNK2 proteins in the HLY-1 cell line. Unexpectedly, we found that eIF4E1 protein level decreased proportionally to MNK knockdown in a dose-dependent manner. These data were consistent across all three shRNAs designed and validated for MNK1 and MNK2 ( Fig. 3a-j). To attest against an off-target shRNA effect, we transduced MNK1 or MNK2 shRNA in HLY-1 cell lines that were overexpressing MNK1 and MNK2 to introduce the same amount of shRNA viral particles in the cells. The amount of shRNA introduced could not sufficiently reduce the abundant MNK levels. We found that the decrease in eIF4E1 levels associated with MNK1/2 knockdowns was completely abolished, further assuring that the effects of the shRNAs are not directly on eIF4E1 and are mediated via protein levels of MNKs (Fig. 3k,l, Supplementary Fig. 5a,b). In addition to confirming MNKdependent eIF4E1 protein level, these data also strengthen the point that MNK1 and MNK2 could complement each other in maintaining eIF4E1 levels in vivo (Figs 1e-h and 3k,l). We compared our findings with the previously reported MNK knockout mouse data 18 . Ueda et al. 18 reported neither a phenotype nor changes in total eIF4E1 level in these mice. However, when these mice were crossed with PTEN-deficient T-lymphoma-forming mice (tPTEN), the resultant MNK knockout/tPTEN mice exhibited delayed lymphomagenesis. These data, presented more recently by Ueda et al. 20 , revealed a possible explanation for the discrepancy observed with our findings 20 . Basal eIF4E1 level and phosphorylation in the tPTEN/MNK WT (MNK1 þ / À , MNK2 þ / À ) tumour were significantly higher than the WT thymus samples, possibly showing a higher dependency on the MNK-eIF4E1 pathway in these tumours. However, in the same tumours, when MNK1, MNK2 or both were eliminated, 50% (1/2), 100% (1/1) and 66% (2/3) of tumours, respectively, exhibited reduced total eIF4E1 protein level in comparison with WT MNKs in the tPTEN background. This raised the possibility that eIF4E1 regulation by MNK protein is potentially enhanced and is more prominent when the MNK-eIF4E1 axis becomes crucial for survival. We interrogated a series of transformed cell lines to determine whether this phenomenon was preserved across different tissues. As illustrated in Fig. 3m (and Supplementary Fig. 5c), we find that the MNK-dependent eIF4E1 protein level was consistent in two other B-lymphoid origin cell lines; however, in T-leukaemia (Jurkat) and colorectal carcinoma (HCT116) cells, the reduction in MNK level did not affect the eIF4E1 levels.
To investigate the mechanism of MNK-dependent eIF4E1 expression, we measured eIF4E1 protein half-life with and without MNK knockdown. Cells were first transduced with either MNK2 shRNA or non-target (NT) control viruses. Forty-eight hours post transduction, cells were treated with cycloheximide to halt translation/protein synthesis and then samples were collected at various time points to measure eIF4E1 protein levels. Following immunoblotting and probing for eIF4E1, we performed densitometry on triplicate experiments to calculate protein half-life. We found that eIF4E1 protein half-life was B4.5 h, as reported previously, and the half-life was not affected by MNK knockdown (Fig. 3n). These data suggest that MNK-dependent eIF4E1 depletion was not related to accelerated protein degradation or reduced stability determined by MNK. However, the reduction of eIF4E1 with MNK knockdown was eliminated after inhibition of mRNA translation with cycloheximide, suggesting that MNK-dependent eIF4E1 depletion was associated with mRNA translation or stability. Total mRNA levels of eIF4E1 in untreated, non-target (NT) control and MNK2 knockdown were subsequently investigated. We treated cells with actinomycin-D to inhibit transcription of new mRNA templates and collected samples at various time points post treatment. We found no significant reduction in eIF4E1 mRNA levels on MNK2 knockdown, strongly arguing against mRNA instability as a reason for the eIF4E1 protein depletion seen with MNK knockdown (Fig. 3o, Supplementary Fig. 5d). In fact, we saw greater mRNA stability with MNK knockdown. Taken together, these data showed that the reduction of eIF4E1 protein expression with MNK knockdown was not mediated via altered protein or mRNA stability. We analysed the heavy polysomal fraction from sucrose density gradient separation containing highly translated polysome-bound mRNA transcripts. Intriguingly, we found that while no significant difference in total mRNA was detected, eIF4E1 mRNA level in the polysomal fraction was significantly reduced (Fig. 3o-p). These experimental findings are consistent with MNKs playing a novel role in eIF4E1 translation and protein expression in addition to its well-recognized kinase activity in phosphorylating eIF4E1. We investigated MNK kinase inhibition in DLBCL using two well-characterized MNK inhibitors: cercosporamide and CGP57380. Cercosporamide has been shown to be a more selective and potent inhibitor of MNKs in comparison with CGP57380. However, both inhibitors also been shown to exhibit a broader effect on several other kinases 10,29,31 . In accordance with previous reports, both inhibitors exerted significant effect on MNK kinase activity, resulting in reduced eIF4E1-S209 phosphorylation with no changes in eIF4E1 total protein level. Thus, we find that MNK in its physical form has a kinaseindependent role in maintaining the cellular eIF4E1 level, while the kinase-dependent function of MNK is important for eIF4E1 phosphorylation. As an eIF4E1-mediated translation readout, we also found MCL-1 was reduced in both treatments. Unexpectedly, we found eIF4E3, another eIF4E family member with significant cap-binding potential, was increased following MNK inhibitor treatment. To further probe eIF4E3 upregulation, we took advantage of the differential readouts of both cercosporamide and CGP57380 on both HLY-1 and Pfeiffer cell lines. Cercosporamide showed a stronger reduction in eIF4E1-S209 phosphorylation, while CGP57380 (ref. 10) showed a stronger increase in eIF4E3 expression (Fig. 4a,b). MNK inhibition using either compound resulted in significant cell death (Fig. 4c). We performed an m 7 G-cap-binding assay in DLBCL cell lines following treatment with both MNK inhibitors. We found that cercosporamide treatment, which showed a stronger reduction in eIF4E1-S209 phosphorylation, did not affect eIF4E1 m 7 G-capbinding ability, arguing against the need for eIF4E1-S209 phosphorylation for enhanced cap binding. On the contrary, treatment with CGP57380, having shown a substantial increase in eIF4E3 expression, resulted in a significant decrease in eIF4E1 binding to m 7 G-cap, proposing a potential role for eIF4E3 in affecting eIF4E1 cap binding (Fig. 4d,e).
The increase in eIF4E3 following MNK kinase inhibitor was not due to changes in total mRNA level (Fig. 4f). To understand the mechanism of eIF4E3 upregulation, we co-treated cells with an MNK inhibitor, CGP57380, that caused eIF4E3 upregulation, with cycloheximide to halt mRNA translation. Halting translation significantly eliminated eIF4E3 upregulation, strongly suggesting that observed the eIF4E3 upregulation was mediated via an mRNA translation process (Fig. 4g,h). Here, we report that reduction in MNK protein level suppressed eIF4E1 translation, while inhibition of MNK kinase activity significantly reduced eIF4E1-S209 phosphorylation and increased eIF4E3 levels in DLBCL cell lines.
Unphosphorylated eIF4E1 enhances eIF4E3 expression. To understand eIF4E1 and eIF4E3 modulation by MNK in DLBCL, we established stable cell lines of HLY-1 (ABC-DLBCL) and Pfeiffer (GCB-DLBCL) expressing wild-type and mutant eIF4E1, eIF4E3 and MNKs. The eIF4E1-S209D is a phosphomimetic mutation while the eIF4E1-S209A is a phosphonull mutation representing constitutively phosphorylated eIF4E1 and phosphorylation-deficient eIF4E1 proteins, respectively, as previously described 19,28,31,32 . It is important to note that the commercial phospho-specific antibody of eIF4E1 does not recognize the S209D mutant 32 ; thus, in these experiments we also employed MCL-1 protein levels as a downstream effector readout (MCL-1 being tightly regulated by phospho-eIF4E1). The phosphorylation of eIF4E1 at S209 was enhanced by the enforced expression of wild-type eIF4E1, MNK1, MNK2 and MNK1-TD mutant (a constitutively activated MNK1 mutant) in both HLY-1 and Pfeiffer (Fig. 5a,b and Supplementary Fig. 6). Consistently these cells also exhibited proliferative advantage compared with vector control HLY-1 cells (Fig. 5c). Remarkably, the enforced expression of the eIF4E1-S209A mutant, which cannot be phosphorylated at S209, showed increased expression of eIF4E3 as well as reduced eIF4E1 cap binding compared with vector control and wild-type eIF4E1 (Fig. 5a,d,e). Conventional dogma suggests that the phosphorylation status of eIF4E1 at S209 is solely determined by MNKs; here we find that the abundance of unphosphorylated eIF4E1 is a trigger for eIF4E3 upregulation. By comparing the MNK kinase inhibitor study (Fig. 4a) and the enforced eIF4E1-S209A study (Fig. 5a), we postulate that MNK kinase inhibition modulating dephosphorylated eIF4E1 is the prompt for eIF4E3 upregulation via translation enhancement. To buttress these findings, we repeated these experiments in another cell line (Pfeiffer) and observed similar results ( Supplementary  Fig. 6). eIF4E3 expression is not very abundant, and its level in a series of cells are shown in Supplementary Fig. 7a. Enforced expression of the MNK1-AA mutant that cannot phosphorylate eIF4E1 also enhanced endogenous eIF4E3 level ( Supplementary  Fig. 7b).
eIF4E3 has been recently shown to be able to bind m 7 G-cap in an atypical manner, with about 40-fold lower affinity than eIF4E1, using biophysical determinants 7 . It has been hypothesized that eIF4E3 may compete for the same mRNA transcript as eIF4E1, thus displacing eIF4E1 from cap and reducing eIF4E1-cap-mediated translation. Some translational targets of eIF4E1 such as cyclin D1 and VEGF have been shown to be suppressed by enforced eIF4E3 expression 7 . First, to determine the mechanism by which eIF4E3 may reduce eIF4E1 binding to m 7 G-cap, we expressed an eIF4E3 delta199 (D199) mutant with C-terminal truncation that lacks significant capbinding activity in HLY-1 and Pfeiffer cells. The expression of wild-type eIF4E3 significantly reduced eIF4E1 cap binding, while the eIF4E3-D199 mutant expression did not exert the same effect (Fig. 5d). These findings were consistent when we repeated similar experiments to analyse eIF4E1 binding to the eIF4G complex ( Fig. 5d lower panel). Overexpression of wild-type eIF4E3 not only reduced eIF4E1 binding to cap, but also exhibited enhanced binding of eIF4E3 to cap as demonstrated in the cappull down experiments (Fig. 5f,g). Taken together, these data strongly confirm that eIF4E3 requires the cap-binding ability to inhibit eIF4E1 from binding cap, potentially via a substrate competition mechanism. As binding to cap alone may not necessarily indicate effective mRNA translation, we next carried out reciprocal immunoprecipitation of eIF4E3 and eIF4G to ask whether eIF4E3 is able to bind the scaffolding protein eIF4G. We found that eIF4E3 was indeed able to bind eIF4G, further enforcing that eIF4E3 may be a functional translation protein, in addition to being an eIF4E1 inhibitory molecule. eIF4E3 also physically associated with eIF4A, another component of the capbinding complex, necessary for translation initiation (Fig. 5h). We initially showed that eIF4E1 cap-binding was significantly reduced when cells were treated with an MNK inhibitor CGP57380. We next asked whether eIF4E3 cap-binding is enhanced under the same circumstances. We treated cells with an MNK inhibitor, performed cap-pull down assay and found that, in addition to reducing eIF4E1 cap-binding (Fig. 4d), MNK inhibitor CGP57380 enhanced eIF4E3 binding to cap (Fig. 5i). To interrogate the presence of eIF4E3 in the translation initiation complex, we performed sucrose density gradient fractionation (Fig. 5k). Similar to eIF4E1, eIF4E3 was present in the lighter sucrose fractions of the gradient, further affirming its role in translation initiation (Fig. 5l). We also find that MNK inhibitor treatment enhanced the presence of eIF4E3 in these fractions (Fig. 5l).
eIF4E1 and eIF4E3 exhibit distinct translatome pattern. Our data affirm that eIF4E3 is capable of forming a novel eIF4F capbinding complex, which we call the eIF4F-3 ('3' denoting eIF4E3) via its interaction with eIF4A and eIF4G, and plays a role in the early processes of mRNA translation initiation. We next asked whether eIF4E3 can drive active translation? As the knockdown of eIF4E1 or eIF4E3 caused significant cell death ( Supplementary   Fig. 8a,b), we overexpressed either protein in HLY-1 cells and performed sucrose density gradient fractionation. RNA extracted from the polysome fractions containing highly translated polysome-bound mRNAs (#9-11, see Fig. 5k) was used for translatome analysis using microarray-chip gene expression analysis. Simultaneously, we obtained total RNA from these cells for transcriptome analysis. Both eIF4E1 and eIF4E3 commonly regulate the translation of a high percentage of genes. Nevertheless, both eIF4E1 and eIF4E3 also displayed distinct gene expression profiles that were significantly altered in the translatome (Fig. 6a-c, Supplementary Table 2). Interestingly, we find MNK2 (MKNK2) translation to be regulated by eIF4E1 and translation of targets like PCNA and CDK2 are regulated by eIF4E3. Next, we performed a principal component analysis on the translatome and transcriptome data. Both eIF4E1 and eIF4E3 displayed a mostly overlapping translatome; however, these smaller-by-scale changes at the translatome level resulted in a larger distinction at the transcription level, evidenced by the distinct clustering of each transcriptome data set (Fig. 6d).   ARTICLE By using Ingenuity Pathway Analysis (IPA), we analysed the differentially expressed genes in eIF4E1 and eIF4E3 complete translatomes. IPA Core Analysis revealed NF-kB complex activation as the primary molecular network enriched in eIF4E1 translatome (Fig. 6e). This gives eIF4E1 a prominent role in oncogenic transformation via an NF-kB-dependent transcriptional upregulation in addition to its known role for selecting weak messages (mRNAs that contain long and highly structured untranslatable regions at their 5 0 -end) for translation 33 . Analysis of the eIF4E3 translatome revealed an important role for eIF4E3 in modulating microRNA maturation via the regulation of dicer. eIF4E3 also affects ADAR (adenosine   504  1,489  345   633   H1FX  RAB33A  GPBAR1  LRCH4  LOC642502  ADAM15  MRPL20  TMEM93  MTP18  SNRPN  LOC728809  HIST3H2A  HIST1H2AC  NUDT18  HIST2H2AA3  TMEN41A  SLC25A26  LOC650826  C9orf89  ENTPD6  LOC100132564   TIMM23  CUTA  TEAD4  ISOC2  ATF5 GPR114 C21orf70   LOC729816  HIST1H4C  RECQL4  ADRM1  PARP12  TNFRSF8  FABP5  TMSB4X  LOC642956  CDT1  SLC29A1  C7orf40  TUBG1  PPAN-P2RY11  POLA2  EIF3I  CDC25A  DHRS2  RPP40  ID1  EBNA1BP2   UNG  CHAF1A  METT11D1  MR11  PRDX1  LOC648392  LRRC50  ITM2C  TUBA3D  POLR1C  KIAA0101  SLC38A10  TUBA3C  IL6 PCNA  Supplementary Fig. 10. deaminase, RNA-specific), which is known to regulate RNA editing and transcript stability [34][35][36] , as well as transcription factors like n-Myc, HMGA1, CDX2 and TWIST1 (Fig. 6f). We corroborated our translatome analysis by western blotting. eIF4E1 but not eIF4E3 cells enhanced c-Myc, a known target of NF-kB transcription activation 37 . Similarly, eIF4E3 cells but not eIF4E1 cells exhibited enhanced n-Myc expression and reduced DICER1 expression (Fig. 6g). To further explore the potential for eIF4E1 driven NF-kB complex activation, we selected three known targets of NF-kB, that is, BTK, YY1 and CDK6, which showed clear correlation with eIF4E1 expression or knockdown in HLY-1 cells (Fig. 6h), and probed in eIF4E1 expressing Pfeiffer and GM02184. We observed a modest increase in CDK6 expression in GM02184, otherwise all other targets were not upregulated by eIF4E1 expression (Supplementary Fig. 9a). Next, to explore whether this phenomenon was restricted to the ABC-DLBCL cells, we knockdown eIF4E1 in another ABC-DLBCL cell line, SUDHL-2, and probed for the same targets. We found that the   Fig. 9b). We identified two most significantly enriched 5 0 -untranslated region (UTR) motifs in eIF4E1 and eIF4E3 translatome (Fig. 7). The locations of the motifs in their corresponding targets are illustrated in the location map charts. We adapted a luciferasenormalized RT-qPCR approach employed by the Sabatini lab 38 to measure the percentage of target mRNA in each polysome fraction to validate the selection of motif-containing transcripts by eIF4E1 or eIF4E3 (Fig. 7). Transcripts containing eIF4E3driven motifs, that is, POLA2 and DDX49, were most abundant in the heavier fractions of eIF4E3 cells, while the same transcripts were displaced to the lighter fractions in the eIF4E1 cells. Similarly, transcripts containing the eIF4E1-driven motifs, that is, DGCR6 and DTD1, were displaced to the lighter fractions in the eIF4E3 cells. In eIF4E1 cells, these transcripts were more abundant in the heavier fractions in comparison with eIF4E3 cells; however, more transcripts accumulated in the monosomal and early polysomal fractions. These findings are consistent with the presence of excessive eIF4E1 initiating mRNA translation; however, the cells may still require other factors that may be motif specific and rate-limiting for eIF4E1-driven translation, presenting a potential avenue for future studies. The complete lists of eIF4E1 and eIF4E3-driven targets displaying respective motifs and the nucleotide frequency statistics are shown in Supplementary Tables 3-5. Our analysis comparing previously published mTOR-driven motifs revealed that eIF4E1-and eIF4E3-driven motifs are novel and independent from previously published mTOR-driven motifs 38,39 , although mTOR can affect eIF4E1 activity via 4E-BP1 phosphorylation (Supplementary Table 6).

Discussion
Translation dysregulation can be extended to almost all types of human malignancies and serves as an attractive target for potential therapeutic intervention. Here, we have presented data to support a unique role for MNKs in DLBCL, which serve as a 'switch' that modulate eIF4E1-and eIF4E3-driven translation in cells. We have demonstrated that MNK1 and MNK2 are differentially distributed in the GCB and ABC subsets of DLBCL, and that the preferential expression of MNK2 in ABC-DLBCL can be a contributing factor to the aggressive nature of this subtype. MNK2, even at low levels, exhibits high basal activity compared with MNK1 and is not easily manipulated by upstream regulators 14 .
The novel regulation of eIF4E1 level by MNKs illustrates a reciprocal correlation between the two proteins. The regulation of eIF4E1 level by MNK in a kinase-independent fashion is mediated via down regulation of eIF4E1 mRNA translation by lack of MNKs. However, enforced expression of MNKs did not significantly upregulate eIF4E1 expression, suggesting the possibility of other cellular surveillance mechanisms that may work to maintain eIF4E1 at a physiological level.
We have shown that the absence of MNK kinase activity on eIF4E1 results in upregulation of eIF4E3, which can attenuate eIF4E1 cap binding (Fig. 8). While inhibition of eIF4E1 is emerging as an attractive target in malignancies, there are potential caveats. As eIF4E1 inhibition may reduce the prooncogenic cellular translatome, the compensatory translation by eIF4E3 may, in addition to promoting growth inhibition, also provide a mechanism to sustain cell viability without rapid proliferation. By regulating the translation of crucial transcription factors like n-Myc, eIF4E3 can also potentially upregulate the transcription of many pro-proliferative targets, which may contribute to the relapse of tumours treated with MNK or eIF4E1 inhibitors.
We first identified upregulation of eIF4E3 when we inhibited MNK kinase activity. Our subsequent investigation revealed that the lack of phosphorylation in eIF4E1 at S209, the effector of MNK kinase inhibition, was the trigger for eIF4E3 upregulation. It was not previously known that eIF4E3 could initiate effective mRNA translation, although its in vitro cap-binding ability has been recently described 7 . Here, we show that eIF4E3 binds to cap and forms a novel eIF4F cap-binding complex. eIF4E3 not only exerts inhibitory effects on eIF4E1-mediated translation, but also facilitates translation of select messages. We find eIF4E1mediated translation enhances the NF-kB pathway, a wellrecognized pathway essential for lymphocyte development, proliferation and survival, which if deregulated is linked to T-and B-cell lymphomagenesis 40,41 . Hariri et al. 42 have reported that eIF4E1 is a direct transcriptional target of NF-kB and aberrant expression of NF-kB-driven eIF4E1 was evident in a subset of acute myeloid leukaemia. In this study, we find that the inverse is also true, where NF-kB complex activation is indirectly enhanced via eIF4E1-mediated translation. eIF4E1 and eIF4E3 are able to recognize and bind the m 7 G-cap structure on mRNA. It is unclear how the two proteins can facilitate the selection of mRNA to be translated. One hypothesis is that eIF4E1 and eIF4E3 select for interacting factors, via altering conformational changes of the cap-binding complex or by causing steric hindrance to the assembly of other components of the translation machineries. Nonetheless, both eIF4E1 and eIF4E3 display preferential 5 0 -UTR motifs in their highly regulated transcripts.
We conclude that MNKs play a pivotal role in the control of eIF4E1 and eIF4E3 levels and function in vivo, and this fine balance is crucial to maintain a normal cellular phenotype. Aberrant disruption to this balance, perturbing cap-binding Increased abundance of eIF4E3 in a cellular context enhances the ability for eIF4E3 to bind cap. The relative abundance of either eIF4E1 or eIF4E3 is determined by MNKs. The accessibility to mRNA cap structure by both eIF4Es mandates a distinct cellular translatome that dictates pro-or anti-oncogenic phenotype.
preference for both eIF4E1 or eIF4E3 proteins, enables a switch, causing subsequent alterations in the cellular translatome that translates a transformed phenotype.  ), 100 unit ml À 1 penicillin and 0.1 mg ml À 1 streptomycin (Sigma-Aldrich) at 37°C with 5% CO 2 . All other cells were grown in RPMI-1640 media supplemented with 10% FBS, 100 unit ml À 1 penicillin and 0.1 mg ml À 1 streptomycin at 37°C with 5% CO 2. In experiments where cells were treated with select inhibitors, exponentially growing cells were plated at 2.5 Â 10 5 cells per ml density before treatment and incubated for variable time points at 37°C with 5% CO 2 , before harvesting for further analysis. Stable cell lines of HLY-1 and Pfeiffer overexpressing eIF4E1, eIF4E1-S209D, eIF4E1-S209A, eIF4E3, MNK1, MNK2, MNK1-TD and MNK1-AA proteins were generated by retroviral transduction and were selected and maintained with puromycin (2 mg ml À 1 ).

Methods
Plasmids, lentiviral production and transduction. Expression plasmids for eIF4E1-S209D and eIF4E1-S209A were a gracious gift from Dr Hans-Guido Wendel (Memorial Sloan Kettering Cancer Center, NY). Expression plasmids for eIF4E1, eIF4E3 and eIF4E3-D199 were cloned from previously described constructs 7 . MNK1, MNK2, MNK1-TD and MNK1-AA were a gracious gift from Dr Herman Gramm (Novartis Institutes for BioMedical Research, Switzerland). All expression plasmids were cloned into a lentiviral-vector system CD513B (System Biosciences, CA). For lentiviral packaging, HEK293T/17 cells were seeded at 40% confluence and transfected the following day with psPAX2 and pMD2.G (Addgene plasmids 12,260 and 12,259; deposited by Dr Didier Trono). Viruscontaining medium was concentrated using Amicon Ultra-15 100 kDA centrifugal filters (EMD Millipore) as per the manufacturer's instruction. HLY-1 and Pfeiffer cells (2x10 5 cells per ml density in a 10 ml volume) were treated with 4 mg ml À 1 of polybrene (American Bioanalytical) and centrifuged for 35 mins at 800 r.c.f. at 30°C. Cell pellets were resuspended in virus-containing medium for 12 h. Cells were selected and maintained with puromycin (2 mg ml À 1 ).
Gene knockdown using shRNA. Lentiviral-shRNA plasmids against MNK1 and MNK2 were purchased from Sigma-Aldrich (MNK1: TRCN0000314869, MNK2: TRCN0000342226 and TRCN0000006098, designed and validated by The RNAi Consortium shRNA library, Broad Institute of Harvard and MIT). Transduction viruses were prepared as described above and the multiplicity of infection (MOI) of all shRNA determined in HEK293T/17. Cells were transduced via spinoculation as described above. Following transduction, cells were harvested at 24 or 48 h for analysis. All shRNA sequences are listed in Supplementary Table 7.
Chemicals. MNK inhibitors CGP57380 and cercosporamide were purchased from Tocris and Sigma-Aldrich, respectively. VX702 (p38 inhibitor) was purchased from Cayman Chemical, and MEK inhibitor AZD6244 was purchased from CalBiochem. All chemicals were dissolved in DMSO before treating cells in culture to a final DMSO concentration not exceeding 0.1%. Solutions were prepared fresh on the day of experiment.
Flow cytometry. For cell cycle analysis, transduced cells were collected (4 Â 10 5 cells) by centrifugation (500 r.c.f., 5 min, 4°C), and washed twice with cold PBS, followed by overnight fixation in 75% ethanol at 4°C. Cells were stained with propidium iodide (PI) at 50 mg ml À 1 in 0.1% BSA and analyzed within 1 h of staining on BD FACSCanto. For cell proliferation assay, 10 million cells were washed twice in PBS and resuspended in 0.5 ml PBS. Carboxyfluorescein succinimidyl ester (CFSE) was prepared at 10 mM (2 Â ) concentration in PBS. CFSE 2 Â solution (0.5 ml) was added drop wise into the cell suspension while rapidly mixing the cells on vortex. Cells were incubated at 37°C for 10 mins in the dark with occasional mixing. Cold RPMI medium with 10% FBS was added to stop CFSE incorporation. Cells were pelleted and resuspended in medium with or without drug treatment. After 48 and 72 h, cells were pelleted and resuspended in 0.5% BSA/PBS solution. TO-PRO was used to gate out dead cells. All flow cytometry analysis was run on BD FACSCanto and data analysed with FlowJo software (Tree Star Inc., OR).
RNA extraction and real-time PCR. Exponentially growing cells were harvested by centrifugation. Total RNA was extracted using Trizol and reverse transcribed using Superscript III (Invitrogen) for standard real-time PCR (RT-qPCR) analysis. For polysomal fraction RT-qPCR analysis, RNA was extracted from each sucrose fraction by standard Trizol method. Five nanograms of polyA þ synthetic luciferase mRNA (Promega) was spiked into the RNA of each fraction for normalization (following methods established by Thoreen et al. 38 ). All RNA samples were reverse transcribed using Superscript III. RNA from tissue microarray (TMA) slides containing paraffin sections pre-classified as ABC-and GCB-DLBCL by an independent pathologist 26 was isolated using RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Life Technologies) according to the manufacturer's protocol 25 . RT-qPCR reactions were carried out using SYBR green (Quanta) detection system on a Bio-Rad CFX Connect equipment. All primer sequences used are tabulated in Supplementary Table 7.
Bromodeoxyuridine (BrdU) ELISA. We performed BrdU incorporation assay using BrdU Cell Proliferation Assay kit (Cell Signaling). In brief, cells were incubated with labelling medium containing BrdU for 4 h at 37°C and 5% CO 2 . Cells were then fixed and processed for BrdU detection following the manufacturer's protocol.
Immunohistochemistry staining. We performed immunohistochemistry (IHC) staining of primary lymphoma tissues on a tissue microarray (TMA) slide (US Biomax, LM801) for p-eIF4E1 staining at Mass Histology Services (Boston, MA). In brief, TMA slide was de-waxed and hydrated. Slide was then pre-treated with citrate buffer and rinsed in water. Slide was treated with 3% H 2 O 2 followed by another rinse in water and two washes in PBS. Sample was then blocked in 2% horse serum and incubated in a primary antibody at 1:250 dilution for 1 h at room temperature. After three washes, the slide was incubated with an ImmPRESS HRP conjugated secondary antibody (Vector Labs) for 45 min at room tenperature. Following two more washes in PBS, slide was treated with DAB substrate and rinsed in water. Slide was then counterstained with haematoxylin and differentiated in acid alcohol (2 quick dips). Finally, the slide was placed under running water for 10 mins, dehydrated and mounted with coverslip.
Microarray analysis. RNA from total (for transcriptome analysis) or polysomal fractions (#9,10 and 11, see Fig. 5k; for translatome analysis) from empty vector and eIF4E1-or eIF4E3-expressing cells from three independent experiments was extracted and labelled using Illumina Total Prep RNA Amplification Kit (Ambion was performed to exclude genes with large variance. Genes that were significantly altered with a Z-ratio of above 1.5 (fold increase) or below -1.5 (fold decrease) display a false discovery rate, FDRr0.3 (using empirical Bayes approach) and P-value o0.05 in comparison with control samples were considered statistically significant. To filter gene sets that are solely altered in the translatome without changes in transcriptome, genes that were significantly altered in the translatome were screened and eliminated if concurrent changes in transcriptome were observed. Overlapping genes between two groups were analysed and presented in a Venn diagram. Genes that showed statistically significant differential expression in the translatome of each study group (eIF4E1 and eIF4E3) versus control group were subsequently analysed using IPA (www.analysis.ingenuity.com; Ingenuity Systems). IPA core analysis was performed to identify top molecular functions involved in eIF4E1 and eIF4E3 translatome. Network analysis uses a curated knowledge based on known functional interactions and protein functions to algorithmically infer biochemical interactions.
Motif enrichment analysis. To discover the enrichment of motifs in the 5 0 -UTR regions of the targets of eIF4E1 and eIF4E3, we utilized the AMADEUS motif discovery platform 43 . The 5 0 -UTR sequences for the human gene set were downloaded from the resources available on the AMADEUS website. The genes that were found to be part of the eIF4E1 and eIF4E3 translatomes were used as the target gene sets and compared with the human gene set. The background and target gene sets were tested for GC and length bias. Motif discovery was performed to identify 10 bp motifs localized within a range 200 bp downstream of the transcription start site (TSS). The enrichment factor and P-values were computed using the hypergeometric (HG) over-representation scores for each identified motif. The motifs were also tested for strand bias, chromosomal preference and localization. The position-weighted matrix of candidate motifs were further compared with that of the TOP/TOP-like motif described by Thoreen et al. and Eliseeva et al. 38,44 Statistical analysis. Where applicable, a two-tailed Student's t-test was performed; values were considered statistically significant at P-value o0.05. All experiments were performed in at least three biological replicates for statistical analysis.