Abstract

   Dysregulation of the mitogen-activated protein kinase interacting
   kinases 1/2 (MNK1/2)–eukaryotic initiation factor 4E (eIF4E) signaling
   axis promotes breast cancer progression. MNK1 is known to influence
   cancer stem cells (CSCs); self-renewing populations that support
   metastasis, recurrence, and chemotherapeutic resistance, making them a
   clinically relevant target. The precise function of MNK1 in regulating
   CSCs, however, remains unexplored. Here, we generated MNK1 knockout
   cancer cell lines, resulting in diminished CSC properties in vitro and
   slowed tumor growth in vivo. Using a multiomics approach, we
   functionally demonstrated that loss of MNK1 restricts tumor cell
   metabolic adaptation by reducing glycolysis and increasing dependence
   on oxidative phosphorylation. Furthermore, MNK1-null breast and
   pancreatic tumor cells demonstrated suppressed metastasis to the liver,
   but not the lung. Analysis of The Cancer Genome Atlas (TCGA) data from
   breast cancer patients validated the positive correlation between MNK1
   and glycolytic enzyme protein expression. This study defines metabolic
   perturbations as a previously unknown consequence of targeting MNK1/2,
   which may be therapeutically exploited.
     __________________________________________________________________

   Omics-based study and TCGA analyses reveal an unappreciated role of
   MNK1 in supporting tumor cell glycolysis and liver metastasis.

INTRODUCTION

   Breast cancer (BC) is the most commonly diagnosed cancer among women
   and the second leading cause of cancer-related death ([60]1). For
   patients with nonmetastatic disease, 5-year survival rates range
   between 86 and 99% ([61]2). However, approximately one-third of
   patients with BC will present with distant-site metastases, after which
   the 5-year survival rate is reduced to ~27% ([62]2). Prognosis is
   particularly poor for patients diagnosed with liver metastasis, a
   common secondary site for BC; even with systemic hormone and/or
   chemotherapy, median survival is only 20 months ([63]3). Stage IV BC is
   considered incurable with currently available therapies; thus, the
   ability to reduce BC metastasis remains an unmet clinical need.

   Perturbations in mRNA translation are well known to promote cellular
   transformation and malignant progression ([64]4), as exemplified by the
   mitogen-activated protein kinase interacting kinases 1/2
   (MNK1/2)–eukaryotic initiation factor 4E (eIF4E) axis ([65]5). eIF4E
   binds to the 5′ cap of mRNAs to mediate cap-dependent translation.
   eIF4E is phosphorylated on serine-209 by MNK1/2, encoded by the genes
   MKNK1 and MKNK2 ([66]6, [67]7). Although the precise function of this
   phosphorylation event remains unclear in the field, it is clear that
   MNK-mediated phosphorylation of eIF4E enhances cellular transformation
   ([68]8). Moreover, eIF4E phosphorylation has been well established as
   promoting disease progression, in part due to the increased translation
   of a subset of proinvasive and prosurvival mRNAs ([69]9–[70]11). Our
   team has published studies demonstrating that phospho-eIF4E–deficient
   mice are protected against developing lung metastases in both
   spontaneous and orthotopic models of murine BC ([71]10–[72]12). In
   addition, we have explored the multifaceted role of the MNK1/2-eIF4E
   axis across nontumoral components of the tumor microenvironment (TME),
   including the immune cell landscape ([73]11–[74]14) and stromal
   cell–derived extracellular matrix ([75]15). Clinically, the level of
   phospho-eIF4E has been associated with poor patient prognosis in
   multiple cancer types ([76]9, [77]16), encouraging the pharmacological
   development of MNK1/2 inhibitors that are under clinical testing
   ([78]NCT05744739 and [79]NCT04622007).

   We have previously reported that tumor-intrinsic MNK1 signaling
   promotes primary BC progression from ductal carcinoma in situ to
   invasive ductal carcinoma ([80]17). This study recognized a link
   between MNK1 and BC stem cells (BCSCs), showing that modulating MNK1
   expression altered BCSC properties in vitro and in vivo. Increasing
   evidence suggests that BCSCs, a small pool of undifferentiated and
   self-renewing tumor cells, are the powerhouse behind tumor progression,
   therapeutic resistance, and metastasis ([81]18–[82]22). There are
   several biomarkers, signaling activities, and biological features that
   define and characterize cancer stem cell (CSC) pools ([83]23). In
   particular, CSCs maintain a remarkable ability to reprogram their
   metabolic state to adapt to environmental changes ([84]24–[85]26),
   which is thought to be critical for their ability to proliferate.
   Reports from this emerging field implicate glycolysis as a crucial
   metabolic pathway for CSC populations, including BCSCs, to sustain
   stemness and self-renewal, and highlight oxidative phosphorylation
   (OxPhos) and glutamine metabolism as functionally important in some
   cases ([86]26, [87]27). While MNK1 has been implicated in regulating
   stemness ([88]17, [89]28, [90]29), the precise functional role of this
   kinase in BCSCs, and how this affects metastatic potential to different
   secondary sites, remains unknown. Metabolic adaptation is especially
   interesting in the context of metastasis given recent findings that
   highlight tissue-specific metabolic dependencies of metastatic cells.
   For instance, glycolytic-dominant metabolism can be advantageous for BC
   metastasis to the liver ([91]30), whereas lung metastases rely largely
   on OxPhos ([92]31) and uptake of environmental pyruvate ([93]32), and
   brain metastases require fatty acid synthesis to fuel their growth
   ([94]33).

   In this study, we sought to expand our understanding of how MNK1
   contributes to BCSC properties. Our results suggest that MNK1 affects
   BCSC function to affect tumor growth rates in vivo. We performed
   protein and gene expression analyses, ^13C metabolite tracing, and
   bioenergetic profiling to uncover that MNK1 supports tumor cell
   glycolysis. Furthermore, we determined that genetic and pharmacologic
   inhibition of MNK1 restricts tumor cell metabolic adaptability and
   increases their energetic reliance on OxPhos, resulting in suppressed
   metastasis to the liver, but not the lung, in breast and pancreatic
   tumor models. Last, by interrogating mass spectrometry (MS)–derived
   protein data from The Cancer Genome Atlas (TCGA) PanCancer Atlas Breast
   Invasive Carcinoma dataset, we demonstrate that MNK1 expression
   positively correlates with the expression of glycolysis-related
   enzymes. Together, our findings document the metabolic consequences of
   disrupting MNK1 in BC cells and highlight the potential utility of
   MNK1/2 inhibitors as therapeutic agents in targeting tumor metabolism
   to reduce metastasis.

RESULTS

MNK1 ablation in 4T1 cells alters stemness phenotypes in vitro and in vivo

   To better understand the contribution of tumor-intrinsic MNK1 to BC
   stemness and metastasis, we chose to knock out MNK1 (MNK1 KO) in 4T1
   murine BC cells. Using CRISPR-Cas9, we generated and pursued two MNK1
   KO cell lines (herein referred to as sg-MKNK1-1 and sg-MKNK1-2)
   alongside a single guide RNA (sgRNA)–null control (herein referred to
   as Cas9CTL). Western blotting confirmed the absence of MNK1 in both KO
   cell lines, alongside an expected decrease in the phosphorylation of
   eIF4E ([95]Fig. 1A). 4T1 cell viability was not altered by the loss of
   MNK1 ([96]Fig. 1B) nor was the proliferative rate under standard
   monolayer culture conditions ([97]Fig. 1C). We previously showed that
   phosphorylation of eIF4E, the substrate of MNK1 kinase, promotes
   epithelial-to-mesenchymal transition (EMT) ([98]10). In line with these
   findings, we found that MNK1 KO 4T1 cells reacquired expression of the
   epithelial marker E-cadherin concomitantly with reduced expression of
   mesenchymal markers vimentin and matrix metalloproteinase 3 (MMP3)
   ([99]Fig. 1D). Given our interest in CSCs and the known link between
   EMT and development of CSC-like features ([100]34), we next chose to
   culture Cas9CTL and MNK1 KO cells on low-adherence plastic to assess
   tumorsphere formation from individual cells that maintain the ability
   to self-renew while avoiding anoikis. While all cell lines tested were
   able to form spheres, their morphology differed markedly, as MNK1 KO
   tumorspheres appeared smaller, rounder, and less disordered than
   Cas9CTL tumorspheres ([101]Fig. 1E). Subsequent quantification
   confirmed a significant reduction in the average size of sg-MKNK1-1 and
   sg-MKNK1-2 tumorspheres compared to Cas9CTL ([102]Fig. 1F), while the
   number of tumorspheres formed was not altered ([103]Fig. 1G). Next, we
   used flow cytometry to quantify the expression of CD44 and CD24, two
   commonly used BCSC markers, across our cell lines under monolayer
   culture. Although distinct positive and negative populations were not
   discernible, we found that the mean fluorescence intensity of CD44 was
   robustly reduced in both MNK1 KO cell lines ([104]Fig. 1H), while only
   sg-MKNK1-2 had reduced CD24 expression ([105]Fig. 1I). In addition, we
   profiled the impact of MNK1 KO on the activity of aldehyde
   dehydrogenases (ALDHs), an enzyme group whose activity is enhanced in
   CSCs ([106]35). As expected, MNK1 KO significantly reduced ALDH
   activity in 4T1 cells as measured by flow cytometry ([107]Fig. 1J). We
   next assessed whether MNK1 KO cells differ from their control
   counterparts in tumor outgrowth in vivo. Thus, we injected Cas9CTL and
   MNK1 KO 4T1 cells into the mammary fat pads of BALB/c mice.
   Immunohistochemistry (IHC) confirmed that sg-MKNK1-1 and sg-MKNK1-2
   tumors remained devoid of MNK1 expression in vivo ([108]Fig. 1K).
   Tumors derived from MNK1 KO cell lines grew significantly slower than
   those from Cas9CTL cells ([109]Fig. 1L). Last, we monitored tumor
   formation from serially diluted concentrations of Cas9CTL and MNK1 KO
   cells, a gold standard for assessing stemness. MNK1 KO cells exhibited
   similar tumor-initiation capacity compared to Cas9CTL cells, even at
   the lowest cell concentration ([110]Fig. 1M). Coupled with our in vitro
   observations, the data from this orthotopic model suggest that MNK1
   ablation may not dictate the frequency of CSCs within a population but
   rather the functional capabilities of the CSC pool.

Fig. 1. MNK1 ablation dampens 4T1 stemness properties.

   [111]Fig. 1.
   [112]Open in a new tab

   (A) Western blotting of the indicated proteins in 4T1 cells. Circled
   proteins are significantly differentially expressed between cell lines
   (see fig. S9 for quantification). (B) Percentage of live cells of
   indicated cell lines. Error bars represent SD, one-way analysis of
   variance (ANOVA) with Tukey’s multiple comparison test, n = 4. NS, not
   significant. (C) Proliferation of indicated cell lines over 96 hours
   following serum starvation. Error bars represent SD, two-way ANOVA with
   Tukey’s multiple comparison test, n = 3. (D) Western blotting of the
   indicated proteins in 4T1 cells. Circled proteins are significantly
   differentially expressed between cell lines (see fig. S9 for
   quantification). (E) Representative bright-field images of the
   indicated cell lines grown under low-adherence tumorsphere culture. 3D,
   three-dimensional. (F and G) Average tumorsphere size (F) and
   tumorsphere number (G) grown from 5000 of the indicated cells.
   Quantification performed on tumorspheres 7 days after seeding. Error
   bars represent SD, one-way ANOVA with Tukey’s multiple comparison test,
   n = 3 (25 spheres analyzed per cell line per repeat). (H and I) Average
   surface expression of CD44 (H) and CD24 (I) in the indicated cell
   lines. Error bars represent SD, one-way ANOVA with Tukey’s multiple
   comparison, n = 3. MFI, mean fluorescence intensity. (J) Quantification
   of ALDH activity in the indicated cell lines. Data represent the
   percentage of ALDH^+ cells relative to Cas9CTL; error bars represent
   SD, one-way ANOVA with Tukey’s multiple comparison test, n = 3. (K)
   Representative images of IHC staining against MNK1
   [3,3′-diaminobenzidine (DAB)] in indicated tumors. (L) Growth rates of
   tumors formed from the indicated cell lines following mammary fat pad
   injection. Error bars represent SEM, two-way ANOVA with Tukey’s
   multiple comparison test, n = 7 (Cas9CTL) and 5 (sg-MKNK1-1 and
   sg-MKNK1-2). (M) Number and percentage of mice engrafted following
   mammary fat pad injection of serially diluted Cas9CTL, sg-MKNK1-1, and
   sg-MKNK1-2 cell lines.

Multiomics analysis highlights alterations in glycolysis and OxPhos pathways
in MNK1 KO 4T1 cells

   BCSCs are regulated by several major signaling pathways and have
   multiple biological features that help maintain their activity. We used
   a multiomics approach to understand the mechanism by which MNK1
   ablation regulates 4T1 stem-like behavior. Matched RNA sequencing
   (RNA-seq) and MS-based discovery proteomics were performed on Cas9CTL,
   sg-MKNK1-1, and sg-MKNK1-2 cell lines in quadruplicate ([113]Fig. 2, A
   and B). Loss of MNK1 resulted in the differential expression of
   numerous targets at both the RNA ([114]Fig. 2C) and protein level
   ([115]Fig. 2D), with a large proportion being common to both MNK1 KO
   clones. As expected, principal component analysis (PCA) revealed
   genetic and proteomic heterogeneity between our sg-MKNK1 cell lines
   ([116]Fig. 2, E and F). At the protein level, sg-MKNK1 cell lines are
   more similar to each other than to Cas9CTL ([117]Fig. 2F), particularly
   when focusing on proteins that are down-regulated in MNK1 KO cells
   compared to Cas9CTL (see dashed box in [118]Fig. 2B). Downstream
   validation of differentially regulated targets was limited to
   expression and phenotype alterations that were consistent across both
   sg-MKNK1 cell lines. Functional enrichment analyses were performed to
   identify biological processes that were commonly differentially
   expressed between Cas9CTL and MNK1 KO cells ([119]Fig. 2, G and H).
   Gene set enrichment analysis (GSEA) of our RNA-seq suggested that Wnt
   and Sonic Hedgehog (Shh) signaling pathways, known to be critical in
   maintaining BCSC characteristics, were significantly down-regulated in
   MNK1 KO cells ([120]Fig. 2G and fig. S1A). However, we found no clear
   direction of regulation in our proteomic data, potentially due to poor
   coverage of Wnt and Shh signaling components (fig. S1B). Given that
   dampened Wnt and Shh signaling might help to explain, at least in part,
   our observations of decreased stem-like properties upon MNK1 ablation,
   we proceeded with Western blot analysis for several proteins in these
   pathways. We observed a reduced expression of DVL2 and LRP6, two Wnt
   signaling mediators, in sg-MKNK1 cells, while expression of β-catenin
   was increased (fig. S1C). MNK1 KO cells were further characterized by a
   reduced expression of smoothened, a critical receptor component of Shh
   signaling, although the level of repression varied between cell lines
   (fig. S1C). Given the lack of consistent changes in the expression of
   Wnt and Shh pathway-related proteins in our cell lines, we did not
   further pursue this line of investigation.

Fig. 2. RNA-seq and proteomic profiling of Cas9CTL and sg-MKNK1 4T1 cell
lines.

   [121]Fig. 2.
   [122]Open in a new tab

   (A) Volcano plots summarizing differential gene expression as detected
   by RNA-seq. For RNA-seq data: n = 3 (Cas9CTL and sg-MKNK1-2) and 4
   (sgMKNK1-1). (B) Heatmap expressing relative abundance values for all
   proteins identified by MS. For proteomics data, n = 4 per group. (C and
   D) Venn diagrams depicting the number of differentially expressed genes
   (DEGs) (C) and differentially expressed proteins (DEPs) (D) between
   cell lines. (E and F) PCA of RNA-seq (E) and proteomics (F) datasets.
   (G) GSEA of RNA-seq data through the Kyoto Encyclopedia of Genes and
   Genomes (KEGG) database. Signatures shown represent the top 5 up- and
   down-regulated pathways relative to Cas9CTL cells, according to
   normalized enrichment scores. Only signatures with significant
   differential regulation (adjusted P < 0.1) in both sg-MKNK1 cell lines
   are shown. Bar color denotes cell line. PPAR, peroxisome
   proliferator–activated receptor. (H) Ingenuity Pathway Analysis (IPA)
   of proteomics data in the indicated cell lines relative to Cas9CTL
   cells. Pathways shown represent the top 10 differentially expressed,
   irrespective of direction of regulation, according to enrichment
   z-score values. (I and J) Western blotting of the indicated proteins in
   4T1 cells. Circled proteins are significantly differentially expressed
   between cell lines (see fig. S9 for quantification). (K and L)
   Expression of a proliver metastasis gene signature across RNA-seq and
   proteomics datasets. Proteomics data are shown as a bubble plot (K);
   color corresponds with expression relative to Cas9CTL, and bubble size
   corresponds to P value. RNA-seq data are shown as GSEA plots (L);
   normalized enrichment score (NES) and false discovery rate (FDR) values
   are listed. (M and N) Expression of MKNK1 (M) and MKNK2 (N) in
   parental, tumor-explant, liver-, lung-, and bone-tropic 4T1 cells.
   Values acquired from publicly available RNA microarray data. Error bars
   represent SD, multiple unpaired t tests with Welch’s correction and
   Holm-Sidak multiple comparison test, n = 3. ACO2, aconitase 2; ADLO,
   aldolase A; CS, citrate synthase; ENO1, enolase 1; EMT,
   epithelial-to-mesenchymal transition; FH, fumarate hydratase; GAPDH,
   glyceraldehyde-3-phosphate dehydrogenase; HGF, hepatocyte growth
   factor; HK2, hexokinase 2; IDH2, isocitrate dehydrogenase 2; LDHA,
   lactate dehydrogenase A; MPC2, mitochondrial pyruvate carrier 2;
   NDUFS1/NDUFA10, NADH:ubiquinone oxidoreductase core subunit S1/subunit
   A10; PDH, pyruvate dehydrogenase; PPAR, peroxisome
   proliferation-activated receptor; SHDA, succinate dehydrogenase complex
   flavoprotein subunit A.

   We instead turned our attention to glycolysis and OxPhos pathways,
   which were strongly dysregulated in MNK1 KO cells ([123]Fig. 2, G and
   H). There are two main mechanisms through which cells produce adenosine
   5′-triphosphate (ATP) from glucose—glycolysis and OxPhos coupled with
   the tricarboxylic acid (TCA) cycle ([124]36). These include
   substrate-level phosphorylation during the metabolism of glucose to
   pyruvate and the generation of reducing equivalents (NADH and FADH[2])
   from glycolysis and the TCA cycle that supply electrons to the electron
   transport chain (ETC) to fuel ATP generation via OxPhos. Given that our
   multiomics profiling revealed that sg-MKNK1 cells differentially
   regulate these highly linked metabolic processes in opposite
   directions, we chose to characterize the components of these pathways
   in more detail. From our GSEA analyses, we observed that both sg-MKNK1
   cell lines significantly repress the glycolysis pathway and up-regulate
   OxPhos, while the TCA cycle is largely unaltered (fig. S1D). These
   findings were recapitulated in our proteomic dataset, where we observed
   a strongly reduced expression of almost all glycolytic enzymes in MNK1
   KO cells (fig. S1E). In contrast, the expression of most components of
   the TCA cycle and ETC remained unchanged or increased. Several subunits
   of complex I, the first enzyme in the mitochondrial respiratory chain,
   were robustly up-regulated in MNK1 KO cells (fig. S1E). Western
   blotting verified the reduced expression of several glycolysis-related
   enzymes in sg-MKNK1 cells relative to Cas9CTL ([125]Fig. 2I). In
   relation to pyruvate consumption, we found that MPC2, which transports
   pyruvate into mitochondria, was up-regulated in MNK1 KO cells
   ([126]Fig. 2I). Opposingly, the expression of pyruvate dehydrogenase
   (PDH), an enzyme that converts pyruvate into acetyl–coenzyme A
   (acetyl-CoA) for use in the TCA cycle, was reduced in cells lacking
   MNK1 ([127]Fig. 2I). TCA cycle enzyme expression was largely unchanged
   between the different cell lines, with the exception of IDH2 which was
   up-regulated in sg-MKNK1 cells ([128]Fig. 2J). Last, prompted by our
   proteomics data (fig. S1E), we validated the up-regulation of the
   complex I subunit NDUFA10 ([129]Fig. 2J), further suggesting that MNK1
   KO cells up-regulate complex I expression. These findings support the
   conclusion that 4T1 cells lacking MNK1 have rewired their metabolism to
   depend more on OxPhos. Moreover, these data support the concept that
   MNK1 KO cells have dampened stemness properties ([130]Fig. 1), as a
   glycolytic-dominant metabolism has been shown to play a functional role
   in maintaining BCSC populations ([131]37–[132]39). Previous work by
   Dupuy et al. ([133]30) had further identified a gene signature that
   underpins high glycolytic activity in 4T1 cells to provide a selective
   advantage in their ability to metastasize to the liver. We probed the
   expression of this signature in our Cas9CTL and MNK1 KO cell lines and
   found that it was significantly reduced in sg-MKNK1-1 and sg-MKNK1-2
   cells at protein ([134]Fig. 2K) and RNA levels ([135]Fig. 2L). In
   addition, we analyzed publicly available microarray data from Dupuy et
   al. ([136]30) [Gene Expression Omnibus (GEO): [137]GSE62598] to compare
   the expression of MKNK1 and MKNK2 across parental 4T1 cells, primary
   4T1 tumor–derived explants, and in vivo–selected liver-, lung-, and
   bone-tropic 4T1 cells. We found that MKNK1 expression is uniquely
   up-regulated in liver-tropic cells ([138]Fig. 2M), whereas MKNK2 is
   most highly expressed in lung-tropic 4T1 cells ([139]Fig. 2N).
   Together, these findings suggest a correlation between MNK1 expression
   in tumor cells and their maintenance of metabolic proliver metastasis
   features.

Experimental metastasis models demonstrate that MNK1 ablation significantly
dampens metastatic tumor outgrowth in the liver, but not the lung

   From our multiomics profiling ([140]Fig. 2), we postulated that MNK1 KO
   4T1 cells may have a reduced ability to metastasize to the liver. To
   address this, we performed intrasplenic injections of tumor cells to
   model experimental liver metastasis ([141]Fig. 3A), as this allows us
   to robustly replicate the later stages of the metastatic cascade
   independent of differences in primary tumor growth. We found that mice
   injected with MNK1 KO 4T1 cells had significantly fewer visible
   metastases on the liver at 21 days postinjection, compared to their
   Cas9CTL counterparts ([142]Fig. 3B), leading to a marked increase in
   mouse survival ([143]Fig. 3C). We further performed an intrasplenic
   injection of a pool of five independent sg-MKNK1 4T1 cell lines to
   confirm that the observed effect was not an artifact of clonal
   selection (i.e., individual sg-MKNK1 clones) and observed a reduction
   in liver tumor burden in mice injected with the pool of sg-MKNK1 cell
   lines (fig. S2A). Western blotting validated that this pool lacks MNK1
   expression and subsequently has repressed eIF4E phosphorylation (fig.
   S2B). To ensure that metastases visible on the surface of the liver
   were representative of total tumor burden, we performed quantification
   of hematoxylin and eosin (H&E)–stained step sections of tumor-bearing
   livers (fig. S2C). As expected, we confirmed that livers harboring MNK1
   KO 4T1 cells had a significantly lower number of total metastases
   ([144]Fig. 3D) and reduced overall percentage tumor burden ([145]Fig.
   3E).

Fig. 3. MNK1-null tumor cells have a reduced ability to form liver
metastases.

   [146]Fig. 3.
   [147]Open in a new tab

   (A) Schematic of the intrasplenic injection model. (B) Number of
   metastases visible on the liver 21 days after injection of the
   indicated cell lines. Error bars represent SEM, n = 14 (Cas9CTL), 10
   (sg-MKNK1-1), and 8 (sg-MKNK1-2). (C) Kaplan-Meier curves showing mouse
   survival following intrasplenic injection of the indicated cells. n =
   10 (Cas9CTL), 8 (sg-MKNK1-1), and 7 (sg-MKNK1-2). (D and E)
   Quantification of tumor burden in H&E-stained liver sections, showing
   the average number of metastases (D) and percentage tumor area (E).
   Error bars represent SEM, n = 10 (Cas9CTL), 5 (sg-MKNK1-1), and 8
   (sg-MKNK1-2). (F) Schematic of in vivo seeding competition assay. (G)
   Representative flow cytometry plots and quantification of Cas9CTL and
   sg-MKNK1 cells in the liver 48 hours after injection. Error bars
   represent SEM, n = 7 (Cas9CTL:sg-MKNK1-1) and 5 (Cas9CTL:sg-MKNK1-2).
   (H) Tumor burden in the lung 14 days after tail vein injection of the
   indicated cells. Error bars represent SEM, n = 10 per group. (I)
   Western blot analysis of the indicated proteins in Cas9CTL,
   sg-MKNK1-EV, and sg-MKNK1-AB 4T1 cells. Circled proteins are
   significantly differentially expressed between cell lines (see fig. S9
   for quantification). (J) Number of metastases visible on the liver 21
   days after intrasplenic injection of the indicated cell lines. Error
   bars represent SEM, unpaired t test with Welch’s correction, n = 5
   (sg-MKNK1-EV and sg-MKNK1-AB), 12 (pLKO), and 11 (shLDHA). (K) Western
   blotting of the indicated proteins in FC1199 cell lines. Circled
   proteins are significantly differentially expressed between cell lines
   (see fig. S9 for quantification). (L) Number of metastases visible on
   the liver 21 days after intrasplenic injection of the indicated cell
   lines. Error bars represent SEM, unpaired t test with Welch’s
   correction, n = 5 (Cas9CTL-1, sg-MKNK1-1, and sg-MKNK1-2) and 4
   (Cas9CTL-2). For (B), (D), (E), (G), (H), and (L), one-way ANOVA with
   Tukey’s multiple comparison test was used.

   Given this robust protection against the development of liver
   metastases ([148]Fig. 3, B, D, and E), we wanted to understand whether
   MNK1 KO cells harbor an inability to extravasate out of the vasculature
   or a failure to colonize and outgrow thereafter. To assess this, we
   developed an in vivo seeding competition assay whereby an equal ratio
   of Cas9CTL and MNK1 KO 4T1 cells, each tagged with a different color of
   CellTracker dye, were intrasplenically injected into the same mouse.
   Two days after injection, we perfused the liver to flush the
   vasculature and remove any tumor cells that had failed to extravasate
   before assessing relative cell abundance by flow cytometry ([149]Fig.
   3F). Using this protocol, we demonstrated that neither sg-MKNK1-1 nor
   sg-MKNK1-2 4T1 cells had a defect in their extravasation ability when
   compared to Cas9CTL ([150]Fig. 3G), leading us to conclude that MNK1
   plays an important role after extravasation and is required for tumor
   cell colonization in the extravascular liver microenvironment. To
   determine whether MNK1 KO cells had a general inability to metastasize,
   we performed tail vein injections of Cas9CTL and MNK1 KO 4T1 cells
   (fig. S2D, schematic). MNK1 KO resulted in a mild to nonsignificant
   reduction in the ability of 4T1 cells to form metastases in the lung,
   although tumor growth in the liver was severely diminished ([151]Fig.
   3H and fig. S2E, representative images). This suggests that the
   antimetastatic effects of tumor-intrinsic MNK1 ablation are influenced
   by tissue-specific microenvironments. Next, to further verify the role
   of glycolysis in MNK1-mediated liver metastasis, we generated a
   MNK1-addback cell line (termed sg-MKNK1-AB). We reexpressed a
   constitutively active form of MNK1 ([152]7) into our sg-MKNK1-2 cells,
   which we compared to their empty vector control counterpart
   (sg-MKNK1-EV). Western blotting confirmed that our MNK1-addback cell
   line expressed active MNK1, showing an induction of phospho-eIF4E and
   rescue in the expression of the glycolytic enzymes lactate
   dehydrogenase A (LDHA) and aldolase A (ALDOA) ([153]Fig. 3I). In vivo,
   sg-MKNK1-AB cells formed significantly more liver metastases compared
   to their vector controls ([154]Fig. 3J), helping to support the
   causative role of MNK1 in liver metastasis. If glycolysis underpins
   this process, then we hypothesized that inhibiting critical glycolytic
   enzymes, in the MNK1-addback model, would restrict liver metastasis. To
   test this hypothesis, we used short hairpin RNA (shRNA) to silence LDHA
   in our sg-MKNK1-AB cells; a strategy that has been well documented to
   reduce tumor cell glycolysis ([155]40–[156]42). Significant knockdown
   of LDHA, as confirmed through Western blotting (fig. S2F), ablated the
   ability of our MNK1-addback cells to metastasize to the liver compared
   to the pLKO vector control cell line ([157]Fig. 3J). These results
   support the conclusion that MNK1 promotes liver metastasis via
   up-regulation of glycolytic enzymes, such as LDHA.

   Last, we assessed whether ablating MNK1 would affect liver metastasis
   in other cancer types that are highly liver-tropic. To this end, we
   generated Cas9CTL and sg-MKNK1 cell lines using the FC1199 cell line; a
   well-established model of murine pancreatic ductal adenocarcinoma
   (PDAC) originally derived from PDAC tumors in KPC (Kras; Trp53;
   P48-Cre) mice ([158]43) on a C57BL/6 background. We performed Western
   blotting to confirm the MNK1 KO status of this model and observed
   reduced expression of the glycolytic enzymes ALDOA and LDHA in sg-MKNK1
   FC1199 cells ([159]Fig. 3K), as was shown in our MNK1 KO 4T1 cell lines
   ([160]Fig. 2I). Similarly, intrasplenic injection of paired Cas9CTL and
   sg-MKNK1 FC1199 cell lines phenocopied the protection against liver
   metastasis that we observed in 4T1 cells, with MNK1 KO cells resulting
   in significantly fewer metastases on the liver surface at endpoint
   ([161]Fig. 3L). These data illustrate that targeting MNK1 can reduce
   liver metastasis across multiple cancer types.

Stable isotope tracing demonstrates that MNK1 KO cells have reduced
glycolytic activity and altered glutamine metabolism

   There is a growing appreciation that the metabolic state of a tumor
   cell can dictate tissue-specific tropism and the success of metastatic
   dissemination. We therefore postulated that the reduced ability of MNK1
   KO cells to metastasize to the liver ([162]Fig. 3, B to E, J, and L)
   could be due to altered cellular metabolism. Given the results of our
   multiomics functional enrichment analyses ([163]Fig. 2, G and H) and
   the fact that the liver is a gluconeogenic tissue, we chose to assess
   functional differences in the use of glucose between MNK1 KO and
   Cas9CTL 4T1 cells. We performed stable isotope tracer analysis (SITA)
   using U-[^13C]-glucose, incubating cells with the labeled metabolite
   for 4 hours before MS analysis ([164]Fig. 4A). Results shown in
   [165]Fig. 4 represent labeled metabolite pool sizes; the full mass
   isotopomer distribution profiles for all metabolites shown can be found
   in the Supplementary Materials (figs. S3 and S4). We first assessed
   several U-[^13C]-glucose–derived glycolytic intermediates and
   end-products in Cas9CTL and MNK1 KO cells ([166]Fig. 4, B to G). While
   differences in the labeled pool sizes of early-stage glycolytic
   intermediates glucose-6-phosphate and fructose-1,6-bisphosphate were
   inconsistent between cell lines ([167]Fig. 4, B and C), sg-MKNK1 cells
   had significantly higher levels of labeled (m + 3)
   glyceraldehyde-3-phosphate (G3P) compared to Cas9CTL ([168]Fig. 4D).
   Labeled metabolites downstream of G3P, however, were found at
   significantly lower levels in sg-MKNK1 cells. These included pyruvate
   and lactate (m + 3) ([169]Fig. 4, E and F), as well as
   glycerol-3-phosphate (m + 3), which is synthesized by reducing
   dihydroxyacetone phosphate (DHAP) after its conversion from G3P
   ([170]Fig. 4G). These data highlight G3P as a potential bottleneck in
   the glycolytic chain of MNK1 KO cells ([171]Fig. 4H), in line with the
   reduced expression of glyceraldehyde-3-phosphate dehydrogenase (GAPDH),
   enolase 1 (ENO1), and LDHA ([172]Fig. 2I and fig. S9C). In contrast,
   analysis of the cell culture medium revealed no difference in
   steady-state labeling of extracellular lactate ([173]Fig. 4I). We
   hypothesize that the discrepancy between intra- and extracellular
   lactate labeling may be due to its use as a fuel by 4T1 cells, as one
   of several physiologically relevant carbon sources ([174]44), but
   future tracing studies are needed to formally test this. In addition to
   glycolytic products, we observed largely similar levels of the TCA
   cycle intermediates citrate and malate (m + 2), as well as aspartate (m
   + 2), which is derived from TCA intermediate oxaloacetate ([175]Fig. 4,
   J to L). Last, it is likely that both MNK1 KO and Cas9CTL cells also
   use pyruvate to replenish their pools of TCA intermediates principally
   through PDH, thereby contributing two labeled carbons into the TCA
   cycle per pyruvate molecule, rather than through anaplerotic pyruvate
   metabolism through pyruvate carboxylase, which contributes three
   labeled carbons per pyruvate. This is evidenced by the higher
   fractional enrichment of m + 2–labeled citrate and malate compared to m
   + 3 (fig. S3, H and I). However, note that metabolism of pyruvate
   through PDH can also yield m + 3–labeled intermediates upon a second
   turn of the TCA cycle. Overall, these data support that MNK1 ablation
   in 4T1 cells reduces glycolysis.

Fig. 4. U-[^13C]-glucose and U-[^13C]-glutamine SITA of Cas9CTL and sg-MKNK1
4T1 cells.

   [176]Fig. 4.
   [177]Open in a new tab

   (A) Stable isotope tracing diagram for U-[^13C]-glucose through
   glycolysis and into the TCA cycle via PDH and pyruvate carboxylase
   (PC). (B to G) U-[^13C]-glucose tracing into glycolysis
   (glucose-6-phosphate and fructose-1,6-bisphosphate m + 6, G3P,
   pyruvate, and lactate m + 3, 4-hour tracer) and glycerol-3-phosphate (m
   + 3, 4-hour tracer) in the indicated 4T1 cell lines expressed as
   relative size of the labeled pool. (H) Schematic summary of the
   glycolytic pathway in relation to the data in (B) to (G). (I)
   U-[^13C]-glucose tracing into extracellular lactate (m + 3, 4-hour
   tracer) in the indicated 4T1 cell lines expressed as relative size of
   the labeled pool. (J to L) U-[^13C]-glucose tracing into the TCA cycle
   via PDH (citrate, malate, and aspartate m + 2, 4-hour tracer) in the
   indicated 4T1 cell lines expressed as relative size of the labeled
   pool. (M) Stable isotope tracing diagram for U-[^13C]-glutamine into
   the TCA cycle. (N to R) U-[^13C]-glutamine tracing to glutamate
   (glutamine uptake and glutamate m + 5, 4-hour tracer) and into the TCA
   cycle (citrate, malate, and aspartate m + 4, 4-hour tracer) in the
   indicated 4T1 cell lines expressed as relative size of the labeled
   pool. For all panels, experiments were performed in technical
   triplicates and repeated three times. Graphs shown correspond to one
   representative experiment; error bars represent SD of technical
   triplicates. Statistics shown were calculated for technical repeats,
   one-way ANOVA with Tukey’s multiple comparison test. DHAP,
   dihydroxyacetone phosphate; GLS1/2, glutaminase 1/2; GLUD1, glutamate
   dehydrogenase 1; GLUL, glutamate-ammonia ligase; GOT1/2,
   glutamic-oxaloacetic transaminase 1/2; GPI, glucose-6-phosphate
   isomerase; GPT1/2, glutamic-pyruvic transaminase 1/2; MDH2, malate
   dehydrogenase 2; OGDH, oxoglutarate dehydrogenase; PC, pyruvate
   carboxylase; PFK, phosphofructokinase; PGAM, phosphoglycerate mutase;
   PGK, phosphoglycerate kinase; PKM, pyruvate kinase; SCS, succinyl CoA
   synthase.

   Up-regulated glutamine anaplerosis is another key metabolic feature of
   transformed cells ([178]45) and glutamine contributes to regulating
   redox homeostasis, nucleotide and nonessential amino acid biosynthesis,
   and epigenetic modifications in CSC populations ([179]46). We therefore
   chose to perform subsequent SITA experiments using U-[^13C]-glutamine
   ([180]Fig. 4M). We observed no significant differences in overall
   glutamine uptake ([181]Fig. 4N) or conversion to glutamate ([182]Fig.
   4O) and no consistent changes in reductive carboxylation as indicated
   by the fractional enrichment of citrate m + 5 in MNK1 KO cells compared
   to Cas9CTL (fig. S4C). Both Cas9CTL and MNK1 KO cells use glutamine as
   an anaplerotic substrate, as indicated by equally labeled pools of
   malate (m + 4) ([183]Fig. 4P). However, these glutamine carbons were
   more likely to be used for aspartate synthesis in sg-MKNK1 cells
   ([184]Fig. 4Q), while the pool size of labeled citrate was reduced
   ([185]Fig. 4R). The underlying mechanisms driving these alterations in
   glutamine metabolism, or their relevance in our model, remain unclear.

Bioenergetic assessment of MNK1 KO cells reveals an energetic dependence on
OxPhos

   Having demonstrated that MNK1 KO 4T1 cells have reduced levels of
   glycolytic end-products ([186]Fig. 4, E and F), alongside increased
   expression of enzymes related to pyruvate consumption and mitochondrial
   metabolism ([187]Fig. 2, I and J, and fig. S1E), we hypothesized that
   tumor cells lacking MNK1 may have restricted tumor cell metabolic
   adaptability. We first assessed broad mitochondrial properties and
   found that MNK1 KO 4T1 cells have (i) increased mitochondrial mass
   compared to Cas9CTL ([188]Fig. 5A), as expected by our multiomics
   analysis ([189]Fig. 2J and fig. S1E), and (ii) greater mitochondrial
   reactive oxygen species production ([190]Fig. 5B). Next, we performed
   bioenergetic profiling of Cas9CTL and MNK1 KO 4T1 cells using the
   Seahorse XF Bioanalyzer platform. We observed no change in the
   extracellular acidification rate (ECAR) between MNK1 KO and Cas9CTL
   cells ([191]Fig. 5C), which is consistent with our SITA results
   demonstrating no difference in lactate abundance in the medium
   ([192]Fig. 4I). In addition, there were no differences in oxygen
   consumption rates (OCRs) across cell lines ([193]Fig. 5D), despite
   reduced overall production of glycolytic products pyruvate and lactate
   in MNK1 KO cells. We interpret these findings to suggest that both
   Cas9CTL and MNK1 KO 4T1 cells, grown under steady-state monolayer
   culture conditions, primarily use OxPhos to meet their energetic
   demands, for which our MNK1 KO cells are not disadvantaged. This may
   explain why MNK1 KO cells (BC and PDAC) exhibited no proliferation
   defects in vitro ([194]Figs. 1C and [195]5, F and H).

Fig. 5. Bioenergetic assessment of Cas9CTL and sg-MKNK1 4T1 and FC1199 cells.

   [196]Fig. 5.
   [197]Open in a new tab

   (A and B) Average mitochondrial mass (A) and mitochondrial reactive
   oxygen species levels (B) of the indicated 4T1 cells. Error bars
   represent SD, one-way ANOVA with Tukey’s multiple comparison test, n =
   4. (C and D) Analysis of ECAR (C) and OCR (D) in indicated 4T1 cells,
   normalized to cell number. Data shown display one representative
   repeat; error bars represent SD of technical repeats. R/AA,
   rotenone/antimycin A. (E to H) Effect of oligomycin (1 μM) treatment on
   cell growth over 48 hours, relative to a solvent control, in the
   indicated cells. Oligomycin was removed after 8 hours. Error bars
   represent SD, two-way ANOVA with Tukey’s multiple comparison test, n =
   3. Representative images are shown in (F) and (H). (I and J) Relative
   ATP level (I) and proliferation (J) of the listed 4T1 cell lines over
   24 hours of oligomycin (1 μM) treatment under low-glucose conditions
   (10 mM glucose). Solid lines represent mean value, and shaded areas
   represent SD. Two-way ANOVA with Tukey’s multiple comparison test, n =
   3 per group. (K) Doubling time of the indicated 4T1 cell lines as
   calculated from data in (J). Error bars represent SD, one-way ANOVA
   with Tukey’s multiple comparison test, n = 3. (L and M) Effect of
   combining oligomycin (1 μM), SEL201 (2.5 μM), and EFT508 (1 μM) on
   Cas9CTL 4T1 (L) and Cas9CTL-2 FC1199 (M) cell growth over 48 hours,
   relative to a no treatment control. Cells were dosed with MNKi for 72
   hours total, starting 24 hours before oligomycin treatment. Oligomycin
   was removed after 8 hours. Data for oligomycin treatment without MNKi
   are the same as shown in (E). Error bars represent SD, two-way ANOVA
   with Tukey’s multiple comparison test, n = 3 per group.

   Next, we chose to assess the metabolic adaptability of Cas9CTL and MNK1
   KO cells through use of metabolic inhibitors. Several reports across
   cancer types demonstrate that metastatic tumor cells in the liver TME
   favor a glycolytic-dominant metabolism ([198]30, [199]47–[200]49).
   Thus, we first chose to assess the impact of inhibiting OxPhos in tumor
   cell cultures, thereby mimicking the metabolic stresses of the liver
   TME. We found that all MNK1 KO tumor cell lines (sg-MKNK1-1 and
   sg-MKNK1-2) were sensitive to treatment with the OxPhos inhibitor
   oligomycin, resulting in a significant reduction in cell growth, while
   Cas9CTL cells were unaffected ([201]Fig. 5, E and F). Addback of MNK1
   largely reversed this effect, as sg-MKNK1-AB cells were significantly
   less sensitive to oligomycin treatment compared to sg-MKNK1-EV cells
   ([202]Fig. 5, E and F). Encouragingly, these findings repeated in
   FC1199 MNK1 KO cells. Baseline sensitivity to oligomycin was higher in
   FC1199 compared to 4T1 cells, but OxPhos inhibition in sg-MKNK1 cells
   resulted in a greater repression of cell growth than Cas9CTL ([203]Fig.
   5, G and H). Next, we used 2-deoxy-d-glucose (2DG) to inhibit
   glycolysis and assess its impact on cell growth. We observed no clear
   effect of MNK1 ablation or addback on 2DG sensitivity, as 2DG inhibited
   growth by ~25% in all cell lines tested, although this reduction was
   only statistically significant in Cas9CTL and sg-MKNK1-AB cells (fig.
   S5, A and B). This suggests that the absence of MNK1 sensitizes tumor
   cells to OxPhos inhibition specifically, which, we propose, is due to
   the reduced metabolic adaptability of MNK1-null tumor cells. In an
   environment that selects against an OxPhos-dominant metabolism, we
   predict that MNK1-null tumor cells are less able to adapt and use
   glycolysis to generate ATP for proliferation. To investigate this, we
   performed live-cell ATP imaging in 4T1 Cas9CTL and sg-MKNK1 cells. Cell
   lines were engineered to stably express fluorescent ATP indicators that
   allow for direct measurement of relative ATP levels and treated with
   oligomycin for 24 hours. As glucose availability is often limited in
   the TME, unlike standard in vitro culture (25 mM), we also chose to
   reduce glucose concentrations to more physiologically relevant levels
   (10 mM) ([204]44). Using this approach, we found that MNK1 KO 4T1 cells
   exhibited significantly lower ATP levels ([205]Fig. 5I), slower
   proliferation ([206]Fig. 5J), and increased doubling rates ([207]Fig.
   5K) compared to their Cas9CTL counterparts. Prolonged oligomycin
   exposure is likely necessary for these effects, as we observed no
   differences in OCR between cell lines during short-term oligomycin
   treatment ([208]Fig. 5D). Combined, these data demonstrate that MNK1 KO
   4T1 cells have an impaired ability to use glucose for ATP generation
   and are particularly disadvantaged in an environment that favors
   glycolysis over OxPhos, as demonstrated by the impaired growth of
   MNK1-null tumors in the mammary fat pad and liver ([209]Figs. 1L and
   [210]3, B to E, J, and L).

   As MNK1/2 inhibitors are currently in clinical trials, we next chose to
   investigate how well our findings can translate into a therapeutic
   context. Our team has previously shown that the inhibitors SEL201 and
   EFT508 effectively block MNK1/2 activity, robustly inhibiting eIF4E
   phosphorylation ([211]12, [212]13, [213]50, [214]51). Western blotting
   validated that these MNK inhibitors (MNKis) repressed MNK1/2 activity
   in our 4T1 cells, as indicated by complete repression of phospho-eIF4E,
   without reducing MNK1 expression (fig. S6A). Inhibiting MNK1/2 kinase
   activity was insufficient to reduce LDHA expression (fig. S6A). Even at
   high drug doses, SEL201 and EFT508 were unable to reduce ALDOA or LDHA
   protein expression (fig. S6B), unlike our cell lines where MNK1 protein
   expression is ablated ([215]Fig. 2I). Nonetheless, we found that
   combination treatment of MNKi with oligomycin phenocopied the data
   obtained in both 4T1 and FC1199 genetic MNK1 KO models, resulting in a
   significant reduction in tumor cell proliferation compared to either
   MNKi or oligomycin treatment alone ([216]Fig. 5, L and M). We repeated
   these assays in three human BC cell lines: T47D, MCF7, and MDA-MB-231.
   While baseline sensitivity to oligomycin treatment was cell line
   dependent, the combination of MNKi with oligomycin consistently
   resulted in significantly repressed cell proliferation compared to
   oligomycin treatment alone (fig. S6, C to E). Combined MNKi and 2DG
   treatments further phenocopied our data in the 4T1 MNK1 KO model, as
   MNK1/2 pharmacologic inhibition did not potentiate the effects of 2DG
   treatment alone (fig. S6F). Encouraged by these results, we
   investigated the efficacy of using an MNKi to target liver metastasis
   in vivo. Mice were predosed with EFT508 immediately before intrasplenic
   injection of Cas9CTL 4T1 cells, followed by drug dosing until end point
   (fig. S6G). We observed no difference in mouse body weight changes
   between EFT508 and vehicle groups (fig. S6H). However, EFT508 treatment
   alone was insufficient to reduce tumor burden in the liver, as
   quantified by the number of surface metastases at 21 days after
   injection (fig. S6I). IHC was performed to confirm that EFT508 was
   on-target, as indicated by the significant reduction of phospho-eIF4E
   in liver metastases and the surrounding normal stroma of treated mice
   compared to vehicle controls (fig. S6J). We next attempted to model
   dual targeting of MNK1/2 and OxPhos in vivo by combining treatments of
   EFT508 and IACS-010759, a potent inhibitor of mitochondrial complex I
   ([217]52). Unexpectedly, two-thirds of the mice dosed with IACS-010759
   died, presumably because of drug-associated adverse events or toxicity,
   and the experiment was suspended. In summary, our findings highlight
   changes in tumor cell metabolism as a previously unknown consequence of
   using MNK1/2 inhibitors.

MNK1 protein expression correlates with glycolysis in TCGA BC patient data

   Our data thus far support that tumor-intrinsic MNK1 signaling regulates
   glycolysis to promote liver metastasis in preclinical mouse models. To
   further study the link between MNK1 and glycolysis in human disease, we
   explored protein data from the PanCancer Atlas Breast Invasive
   Carcinoma dataset deposited on TCGA. First, we stratified 102 patients
   by their expression of MNK1, as detected by MS. We then generated
   clustered heatmaps to compare the relative expression of MNK1 with the
   expression of proteins involved in glycolysis ([218]Fig. 6A), the TCA
   cycle ([219]Fig. 6B), and OxPhos ([220]Fig. 6C), as defined by the
   Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database. We
   observed a clear association between high expression of MNK1 and
   several glycolytic enzymes ([221]Fig. 6A), including GAPDH, ENO1, and
   LDHA, which were all observed to be significantly repressed in MNK1 KO
   4T1 cells ([222]Fig. 2I and fig. S9C). Analysis of several individual
   hits identified by our proteomics screening confirmed these
   correlations, with Pearson coefficients ranging from 0.40 to 0.65
   ([223]Fig. 6D). Notably, largely similar correlations were found
   between MNK2 expression and the same enzymes (fig. S7A). Unexpectedly,
   we observed a similar association between the expression of MNK1 and
   TCA cycle enzymes ([224]Fig. 6B), while the pattern for OxPhos was
   bimodal, with groups of genes that were both positively and negatively
   correlated with MNK1 expression ([225]Fig. 6C).

Fig. 6. Analysis of TCGA MS data from patients with BC.

   [226]Fig. 6.
   [227]Open in a new tab

   (A to C) Clustered heatmaps visualizing the relative protein expression
   of MNK1 and glycolysis-related enzymes, as listed in the KEGG
   glycolysis/gluconeogenesis (A), TCA cycle (B), and OxPhos (C) gene
   sets. Data are stratified according to MNK1 expression. Each column
   represents one patient, and each row represents one protein. (D)
   Correlation of MNK1 protein expression with the protein expression of
   listed enzymes. Pearson rank-order correlation, solid black lines show
   linear regression. P and R values are indicated. (E to G) Correlation
   of MNK1 protein expression with ssGSEA enrichment scores for the
   glycolysis/gluconeogenesis (E), TCA cycle (F), and OxPhos (G) pathways.
   Pearson rank-order correlation, solid black lines show linear
   regression, and shaded areas represent 95% confidence intervals of the
   linear regression. P and R values are indicated. For (A) to (G), n =
   102. (H) Kaplan-Meier curve showing overall survival of patient with
   high versus low MNK1 protein expression. Log-rank (Mantel-Cox) test, n
   = 25 per group.

   To quantify the relationship between the expression of MNK1 and an
   entire pathway, we performed single-sample GSEA (ssGSEA) projections
   using the same patient-derived protein data as above. This involves
   calculation of an enrichment score per patient, independent of
   phenotype labeling, which represents the relative activity level of a
   given biological process rather the expression of a subset of genes.
   Using this method, we confirmed that MNK1 expression is strongly
   positively correlated with glycolysis ssGSEA enrichment scores
   ([228]Fig. 6E) and only weakly correlated with TCA cycle ssGSEA scores
   ([229]Fig. 6F). We observed an equally strong correlation between
   glycolysis enrichment scores and the expression of PDK1 (fig. S7B), an
   enzyme that has been reported to facilitate the adaptation to a
   glycolytic-dominant metabolism in liver-tropic cells ([230]30). In
   contrast, we found no correlation between MNK1 expression and OxPhos
   scores ([231]Fig. 6G). These data suggest that MNK1 may support
   glycolytic activity in tumors from patients with BC, without affecting
   OxPhos, in line with our own preclinical observations reported herein.
   We further expanded our analysis to investigate MNK2 and found similar
   positive correlations between MNK2 expression and glycolysis (fig. S7C)
   and TCA cycle (fig. S7D) ssGSEA scores. Unlike MNK1, however, we found
   that MNK2 expression also correlates positively with OxPhos scores
   (fig. S7E), which aligns with MKNK2 being up-regulated in lung-tropic
   4T1 cells specifically ([232]Fig. 2N). Furthermore, we were intrigued
   to assess whether our earlier observations regarding Wnt and Shh
   pathways were recapitulated in patients. Unlike our MNK1 KO 4T1 model,
   both Wnt and Shh signaling were negatively correlated with MNK1
   expression, as shown by clustered heatmap (fig. S8, A and B) and ssGSEA
   analyses (fig. S8, C and D). Last, although not statistically
   significant, overall survival of patients with BC can be separated on
   the basis of MNK1 expression alone, with the high MNK1-expressing group
   having a numerically worse prognosis compared to the low MNK1 group
   ([233]Fig. 6H). Together, these results suggest the potential clinical
   relevance of targeting MNK1 to modulate tumor metabolic features and
   target tissue-specific metabolic dependencies.

DISCUSSION

   CSCs are critical components of successful tumor initiation,
   progression, and metastasis ([234]18–[235]22), but our therapeutic
   tools for targeting these populations in secondary sites remain
   limited. Defining the molecular mechanisms that underpin CSC
   populations is therefore critical to design new strategies for
   metastasis prevention in patients. Here, through our investigation of
   stemness features regulated by MNK1, we showed that genetic and
   pharmacological targeting of MNK1 limits the metabolic adaptability of
   BC and PDAC tumor cells. We found that targeting MNK1 reduced
   glycolysis, as validated using metabolomic investigation, which
   increased dependence on OxPhos to meet energetic demands. Hence, dual
   targeting of MNK1/2 and OxPhos reduced ATP levels and significantly
   impaired tumor cell proliferation. We showed that this led to strongly
   repressed primary tumor growth in vivo and almost entirely abolished
   metastasis to the liver, but not the lung. Last, we have shown that
   MNK1 expression is positively correlated with expression of the
   glycolytic pathway in samples of human patients with BC.

   This study sheds light on the signaling pathways that contribute to the
   complex changes in cell metabolism that facilitate metastasis. The
   Warburg effect predicts that most cancer cells rely on aerobic
   glycolysis to fuel their proliferation ([236]36). It is thought that
   BCSC populations are largely driven by high rates of glycolysis
   ([237]24), at least in the primary site. One study even demonstrates
   that a metabolic switch from OxPhos to aerobic glycolysis is essential
   for the function of BCSCs and enhances tumorigenicity in vitro and in
   vivo ([238]38). The data presented here support this conclusion as MNK1
   KO 4T1 cells, with reduced glycolytic activity, have significantly
   reduced rates of primary tumor growth ([239]Fig. 1L). Our knowledge of
   the metabolic processes that dictate metastatic outgrowth and tropism,
   however, is less advanced. The “seed and soil” theory, alongside more
   recent work by Schild et al. ([240]53), proposes that specific cancer
   cells evolve to have distinct metabolic signatures that predetermine
   their propensity to colonize a distant tissue. Furthermore, it has been
   suggested that only cells with metabolic adaptability, a hallmark of
   CSCs, have the intrinsic ability to survive in secondary sites
   ([241]54). Some studies have even provided evidence for organ-specific
   metastatic abilities of CSC subpopulations in multiple cancers
   ([242]49, [243]55, [244]56). Although this study does not assess the
   metabolic function of CSCs specifically, we predict that the restricted
   metabolic adaptability caused by targeting MNK1 would be
   disadvantageous to CSC and non-CSC pools alike. Moreover, we show that
   this disadvantage is not equal across metastatic sites; targeting MNK1
   in 4T1 cells more strongly reduced their abundance in the liver than in
   the lung ([245]Fig. 3, B to E and H), independent of their ability to
   extravasate from the vasculature ([246]Fig. 3G) but underpinned by
   their ability to perform glycolysis ([247]Fig. 3J). Our data support a
   model in which a tumor cell that lacks MNK1 is less able to adapt its
   metabolic program to be compatible with the microenvironment of its
   target organ. A metastatic tumor cell that seeds in the lung, an
   environment rich in oxygen and oxidative stress, will use OxPhos to
   outgrow. In contrast, a tumor cell that colonizes the liver, a tissue
   that regulates glucose consumption/storage and is more likely to be
   hypoxic, must be able to generate ATP from glucose to thrive. Thus, the
   liver microenvironment would be predicted to be disadvantageous for the
   survival of MNK1-null tumor cells. This may explain why we observed
   different effects of MNK1 ablation on tumor cell growth in
   two-dimensional versus three-dimensional cultures, as tumorspheres are
   enriched in BCSCs (highly glycolytic) compared to monolayer culture.
   Future metabolomics studies on tumorsphere cultures would be needed to
   assess this directly.

   Similarly, when expanding our findings into models of PDAC metastasis,
   we observed that targeting of MNK1 also reduced glycolytic enzyme
   expression and metastasis to the liver ([248]Fig. 3, K and L). This
   aligns well with the seminal study by Ying et al. ([249]57), which
   demonstrated how PDAC progression is driven by KRAS-mediated activation
   of anabolic glucose metabolism and that targeting this pathway led to
   the emergence of treatment-resistant CSC-like cells with an
   OxPhos-dominant metabolism ([250]58). As in BC, more recent work has
   suggested that two subsets of CSCs exist in PDAC; one forms lung
   metastases through primarily oxidative metabolism, while the other
   forms liver metastases through aerobic glycolysis and fatty acid
   oxidation ([251]49). As the liver is the primary site of PDAC
   metastasis and liver metastases are a major cause of PDAC-related
   death, understanding the molecular drivers underpinning metastatic
   outgrowth in the liver is crucial to the design of targeted
   therapeutics that can prevent metastatic disease. It is unknown whether
   the control of glycolysis by MNK1 is relevant in cancer types other
   than BC and PDAC. In contrast to the findings of this study, we have
   previously shown that MNK1 ablation in melanoma cell lines
   significantly reduced their ability to metastasize to the lung
   following tail vein injection ([252]51), suggesting that the
   contribution of MNK1 to metastasis varies between diseases. More work
   is needed to clarify the differences in metabolic programs between
   cancer types and cell lines. Nevertheless, our findings support the
   exploration of MNK1 as a therapeutically relevant target for
   populations of cells that rely on glycolytic metabolism, rather than a
   specific disease model or subtype. Beyond glycolysis, the MNK kinases
   have been shown to play roles in cellular lipid metabolism in the
   context of high-fat diet–induced obesity, including control of
   triglyceride turnover and the translational regulation of genes that
   dictate lipid transport and storage ([253]59, [254]60). Furthermore, it
   has been demonstrated that several genes associated with mitochondrial
   biogenesis are up-regulated in the adipose tissue of MNK1/2 double KO
   mice compared to wild type ([255]59), in line with our observation that
   sg-MKNK1 cells have increased mitochondrial mass compared to Cas9CTL
   cells ([256]Fig. 5A). In contrast, we observed no correlation between
   MNK1 and OxPhos in our analysis of patient proteomic data ([257]Fig. 6,
   C and G) and no significant difference in 2DG sensitivity between
   MNK1-expressing and MNK1-null cells (fig. S5A). It therefore may also
   be true that the up-regulation of ETC-related subunits in MNK1 KO 4T1
   cells ([258]Fig. 2J and fig. S1, D and E) is a necessary consequence of
   down-regulated glycolysis rather than a direct result of MNK1 ablation.
   Continued work is therefore needed to better understand what role
   MNK1/2 plays as a regulatory node of cellular metabolism.

   Our analysis of protein expression data from patients with BC
   ([259]Fig. 6) suggests that targeting MNK1 to affect tumor cell
   metabolism may translate from murine to human studies. We show that
   genetic MNK1 KO models of BC are protected against metastatic outgrowth
   in the liver ([260]Fig. 3, B to E and J), but use of EFT508 alone was
   unable to phenocopy this result (fig. S6I). We hypothesize that there
   may be unknown consequences of broadly blocking MNK activity across the
   TME using a MNK1/2 inhibitor, as opposed to the more specific
   tumor–intrinsic targeting of our genetic model. In particular, although
   a critical role for mRNA translation control in nontumor components of
   the TME is emerging ([261]61), it is yet to be determined how blocking
   MNK pharmacologically affects the cooperative metabolic networks that
   exist between cell types of the TME or how MNKi-mediated disruption of
   glycolysis affects immune cell function. Further research is required
   to better understand these interactions. However, we have shown in
   vitro that blocking MNK1 increases tumor cell sensitivity to oligomycin
   and that combined treatment with MNKi significantly reduced tumor cell
   growth in multiple cancer cell lines ([262]Fig. 5, L and M, and fig.
   S6, C to E). It may therefore be necessary to combine an MNKi with an
   OxPhos inhibitor to adequately restrict the metabolic adaptability of
   tumor cells in vivo, even in the liver microenvironment. We propose
   this drug combination as a potential strategy to target metastases by
   simultaneously affecting the two primary mechanisms of ATP production.
   This is a crucial target in metastasizing tumor cells as sufficient ATP
   generation is emerging as a critical threshold for successful cell
   seeding ([263]62). MNKi are well tolerated in patients and are
   currently under clinical investigation across cancer types and
   treatment modalities ([264]NCT04261218, [265]NCT03690141, and
   [266]NCT04622007). Several drugs targeting aspects of mitochondrial
   metabolism are either already Food and Drug Administration–approved or
   in clinical trials, including metformin ([267]63, [268]64) and
   atovaquone ([269]65). However, clinical benefit of these inhibitors is
   yet to be achieved in the context of oncology. We performed preclinical
   studies using combined treatment of EFT508 and the complex I inhibitor
   IACS-010759, which was advanced into phase I clinical trials at the
   time ([270]NCT02882321 and [271]NCT03291938). These trials were
   recently discontinued because of emerging dose-limiting toxicities,
   including neuropathy in ~50% of patients and physiological changes
   indicative of neuropathy in mice ([272]66). This agrees with our
   observation that IACS-010759 led to toxicity in our mouse models.
   Future studies are therefore needed to assess how to effectively
   combine MNKi and OxPhos inhibition and to determine at what stage of
   the metastatic cascade these drugs could be applied for clinical
   benefit.

Study limitations

   One question that remains unanswered in our study is as follows: By
   what mechanism does blocking MNK1 down-regulate glycolysis? Although
   eIF4E is the only validated substrate for MNK1/2 kinases, we do not
   assess whether the phenotypes we observe are a consequence of altered
   translational control by phospho-eIF4E. It is conceivable that there
   are eIF4E-independent effects of modulating MNK expression and/or
   activity that are yet to be identified. Moreover, a mechanism involving
   the regulation of selective mRNA translation may exist either directly
   or indirectly. Phospho-eIF4E may regulate the translation of glycolytic
   enzymes themselves or that of an upstream mediator such as
   hypoxia-inducible factor 1α (HIF1α), as was emphasized by our
   proteomics data ([273]Fig. 2H). The EMT-associated protein Snail, known
   to be under translational control of phospho-eIF4E ([274]10), has been
   shown to silence the promoter of FBP1, resulting in an induction of
   glycolysis and CSC-like properties in basal-like BC ([275]38). Future
   studies are needed to determine which of these mechanisms, if any, are
   at play in our models. Regardless, this study highlights MNK1 within an
   ever-expanding body of literature that defines translational machinery
   as a central node for regulating energetic balance and cellular
   metabolism ([276]67–[277]71).

   Our comparison of genetic KO and kinase inhibition of MNK1 was
   unsuccessful in shedding light on these mechanistic details. MNK1-null
   tumor cells are more sensitive to OxPhos inhibition than
   MNK1-expressing cells ([278]Fig. 5E), a phenotype that is recapitulated
   through combined treatment of oligomycin and pharmacologic MNK1/2
   inhibition ([279]Fig. 5L). Furthermore, ablation of MNK1 reduced
   glycolytic enzyme expression at the protein level ([280]Fig. 2I), which
   was rescued upon MNK1 addback ([281]Fig. 3I). In contrast, low- and
   high-dose treatments with MNKi were insufficient to reduce enzyme
   expression, at least LDHA and ALDOA (fig. S6, A and B). In patient
   populations, however, the correlation between MNK1 and glycolytic
   enzyme expression is clear. The reason for this discrepancy between
   genetic and therapeutic models remains unknown. One explanation is that
   without a clear mechanistic understanding of this regulation, we chose
   inappropriate readouts to address this question. Alternatively, it is
   plausible that we must target MNK1 protein expression, rather than
   kinase activity, to repress glycolytic enzyme expression. This could
   explain why EFT508, which binds competitively with ATP to block MNK1/2
   kinase activity, was unable to reduce liver metastasis in vivo despite
   reducing eIF4E phosphorylation (fig. S6, I and J). Similar conclusions
   have been drawn from studying the role of MNK2 in adipocyte
   differentiation, where genetic knockdown of MNK2 perturbed lipid
   synthesis and accumulation, while treatment with EFT508 did not
   ([282]72). It therefore seems likely that MNK1/2 have
   kinase-independent functions that remain unknown. Advances in
   technologies such as proteolysis targeting chimera ([283]73), or the
   recently identified VNLG-152R that degrades MNK1/2 ([284]74), may
   improve the viability of targeting these kinases to augment anticancer
   therapeutic outcomes.

   Last, this study is limited in its investigation of the MNK2 kinase. We
   chose to assess the impact of ablating tumor-intrinsic MNK1
   specifically, due in part to the understanding that MNK1 activity is
   induced by extracellular stimuli, whereas MNK2 activity is constitutive
   ([285]75, [286]76). We assume that the low levels of phospho-eIF4E in
   sg-MKNK1 cells are a result of MNK2-mediated phosphorylation, but a
   lack of reliable antibodies to detect MNK2 specifically makes it
   difficult to assess this directly in our model. Our RNA microarray
   analysis revealed that 4T1 cells up-regulate specific MNK isoforms
   based on the secondary site that they have been selected to metastasize
   to ([287]Fig. 2, M and N). Furthermore, patient-derived proteomic data
   highlight that MNK1 and MNK2 expressions correlate equally with
   glycolysis ssGSEA scores ([288]Fig. 6E and fig. S7C), whereas only MNK2
   significantly correlates with OxPhos ssGSEA scores (fig. S7E). These
   data may suggest that both MNK1 and MNK2 can regulate glycolysis,
   whereas regulation of OxPhos may be MNK2 specific. However, future
   studies are needed to assess this directly. Generating tumor-intrinsic
   MNK2 KO and MNK1/2 double KO cell lines will help to provide the tools
   needed to study isoform-specific contributions of the MNK1/2 kinases to
   cellular metabolism and metastasis.

   In conclusion, this study supports the utility of targeting metabolic
   pathways as strategies to reduce metastasis. We have gained an
   increased understanding of the contribution that MNK1 kinase makes in
   controlling CSC populations, while further identifying reduced
   glycolysis as a previously unknown functional consequence of targeting
   MNK1/2. Better understanding of the metabolic dependencies of
   metastatic tumor cells in tissue-specific microenvironments may prove
   useful in identifying vulnerabilities that can be exploited, through
   use of MNK-targeting therapeutics, to improve therapeutic efficacy.

MATERIALS AND METHODS

Cell culture and treatments

   4T1 cells were donated by P. Siegel (McGill University). FC1199 cells
   were gifted from the Tuveson laboratory (Cold Spring Harbor
   Laboratories) and obtained in collaboration with A. Lowy (University of
   California San Diego, La Jolla) and P. Brodt (McGill University). MCF7,
   T47D, and MDA-MB-231 cell lines were purchased from American Type
   Culture Collection. 4T1 cells were cultured in Dulbecco’s modified
   Eagle’s medium (DMEM) high glucose (Wisent Bioproducts) supplemented
   with 10% fetal bovine serum, sodium bicarbonate (1.5 g/liter), 10 mM
   Hepes, and antibiotics. All other cell lines were cultured in DMEM high
   glucose (Wisent Bioproducts) supplemented with 10% fetal bovine serum
   and antibiotics. Cells were subcultured upon reaching ~90% confluence
   by trypsinization and subsequent dilution. Cell lines were maintained
   at low passage number before use, and viability was determined via
   trypan blue exclusion. For all cell treatments, drugs were supplemented
   into complete growth medium at the indicated concentrations. SEL201 was
   provided by Selvita and used at 2.5 μM. EFT508 (tomivosertib) was
   purchased from Selleckchem (#S8275) and used at 1 μM. Oligomycin was
   purchased from MP Biomedicals (#151786) and used at 1 μM. 2DG was
   purchased from MedChemExpress (#HY-13966) and used at 1 mM. For
   tumorsphere culture, cell lines were trypsinized, dissociated into
   single cells, and seeded in six-well plates with ultralow adherence
   (Corning, #3471) at 5000 cells per well. Bright-field images were
   captured at 5 days after seeding. Tumorsphere number and size were
   quantified using ImageJ.

Generation of MNK1 KO, addback, and shRNA cell lines

   CRISPR-mediated MNK1 KO cell lines were generated as previously
   described ([289]17). In summary, parental 4T1 cells were transfected
   with Cas9–green fluorescent protein (GFP)/sgRNA plasmids. Forty-eight
   hours after transfection, GFP^+ cells (transient fluorescence) were
   single cell–sorted by fluorescence-activated cell sorting and expanded
   in culture. The resulting clones were screened for absent expression of
   MNK1 by Western blot, and selected clones were further validated by
   Sanger sequencing of genomic DNA. Two clones were selected for
   follow-up investigation in vitro and in vivo, termed sg-MKNK1-1 and
   sg-MKNK1-2 4T1. A sgRNA-null control cell line was generated using the
   same transfection protocol with the Cas9-GFP plasmid only, termed
   Cas9CTL. MNK1 KO FC1199 cell lines were generated using the same
   protocol, plasmid constructs, and sgRNA sequences. For the 4T1 model,
   two sg-MKNK1 cell lines were compared to the same Cas9CTL line. For the
   FC1199 model, pairs of Cas9CTL and sg-MKNK1 cell lines were generated
   and compared separately. Where indicated, a pool of five sg-MKNK1 4T1
   clones were combined together in equal proportions (20% of the total
   cell population per clone). Plasmids and sgRNA constructs were
   purchased from GenScript, using the pSpCas9 BB-2A-GFP (PX458) vector.
   Two CRISPR guide RNA sequences were designed to target murine MKNK1:
   TCGAAGTCGAGTGTTCCGTG and GACGTCCCGCACCACTCTAC. sg-MKNK1-2 4T1 cells
   were further engineered to express a constitutively active form of MNK1
   (MNK1^T332D) or the corresponding empty vector control. Transduced
   cells were selected for puromycin resistance to yield sg-MKNK1-AB and
   sg-MKNK1-EV cell lines. The MNK1-addback cell line was subject to
   additional genetic manipulation using shRNA. Using our previously
   described protocol ([290]77), sg-MKNK1-AB cells were transduced with a
   combination of two shRNA constructs targeting LDHA (Mus musculus
   [291]NM_010699.2; TRCN0000308704/TRCN0000308638) or a pLKO.5 control
   vector. Puromycin-selected cells were used for downstream
   investigation. The efficiency of the LDHA knockdown in the MNK1-addback
   cells was validated by Western blotting.

Mouse models and orthotopic mammary fat pad injection

   Wild-type BALB/c and C57BL/6 N mice were purchased from Charles River
   Laboratories (Saint-Constant, Canada) and subsequently bred in-house.
   All tumor cell injections were performed on mice approximately 8 to 10
   weeks of age. Animal experiments were conducted according to the
   regulations established by the Canadian Council of Animal Care, using
   protocols approved by the McGill University Animal Care and Use
   Committee (Acceptable Use Policy numbers: 6009 and 5260). For
   orthotopic injections, 2 × 10^5 4T1 cells (Cas9CTL or sg-MKNK1) were
   suspended in phosphate-buffered saline (PBS) and injected into the
   inguinal mammary fat pad of BALB/c mice. Tumor initiation was measured
   as the number of days after injection at which a palpable mass had
   formed. Tumor length (L) and width (W) were routinely measured over
   time, with tumor volume calculated using the formula V = 3.1416/6 × L ×
   W^2. At 25 days after injection, mice were euthanized, and primary
   tumors were excised, formalin-fixed, and paraffin-embedded (FFPE). For
   the limiting dilution experiment, serially diluted concentrations (2 ×
   10^5, 2 × 10^4, and 2 × 10^3) of Cas9CTL and sg-MKNK1 4T1 cells were
   suspended in PBS and injected into the inguinal mammary fat pad of
   mice. Tumors were allowed to form for 8 weeks after injection until a
   palpable mass was detected (engrafted).

Experimental metastasis mouse models: Intrasplenic and tail vein injections

   Intrasplenic injections (with splenectomy) were used to model
   experimental liver metastasis. Mice were anaesthetized, hair on the
   left flank was removed, and a small incision was made in the skin and
   peritoneum below the ribcage. A total of 1 × 10^5 (4T1) or 2 × 10^5
   (FC1199) cells (Cas9CTL, sg-MKNK1, MNK1-addback, or shRNA-manipulated)
   were suspended in PBS and injected directly into the spleen of female
   BALB/c (4T1) or male C57BL/6 (FC1199) mice using a 26-gauge needle.
   Cells travel via the splenic vein and hepatic portal vein directly to
   the liver. Following injection, vasculature connecting the spleen and
   the pancreas was ligated using nonabsorbable braided silk suture (Fine
   Science Tools, #18020-60). Vessels were subsequently cut, and the
   spleen was removed—this is necessary to prevent the formation of a
   primary splenic tumor. The peritoneum was then sutured shut and the
   skin closed using wound clips. Mouse body weight was monitored every 3
   to 4 days, before being euthanized at 21 (4T1) or 22 (FC1199) days
   after injection. At end point, the number of metastases visible on the
   liver surface was quantified before FFPE processing. Mice with
   extrahepatic metastases were excluded from the study. For survival
   cohorts, mice were euthanized upon signs of moribundity or a 20% drop
   in body weight, whichever presented first. For treatment cohorts, mice
   were randomly grouped before treatment. EFT508 was dissolved in a
   solution of 10% 1-methyl-2-pyrrolidinone and 90% propylene glycol, with
   one equivalent of HCl, for administration by oral gavage at a dose of
   10 mg/kg of body weight per mouse, five times per week. The first dose
   was administered ~1 hour before intrasplenic injection. Tail vein
   injections were used to model experimental lung metastasis; 1 × 10^5
   4T1 cells (Cas9CTL or sg-MKNK1) were injected into the tail vein of
   each mouse. At 14 days after injection, mice were euthanized, and lungs
   were excised and dissociated into single cells using the gentleMACS
   Octo Dissociator (Miltenyi Biotec, #130-096-427). Cell suspensions were
   cultured in the presence of 30 μM 6-thioguanine (Sigma-Aldrich,
   #A4660); these conditions are toxic to cells other than the naturally
   resistant 4T1. After 5 days, remaining 4T1 colonies were fixed with 4%
   paraformaldehyde (PFA) for 15 min at room temperature and stained with
   0.5% crystal violet for 30 min at room temperature, and metastatic
   burden was scored by four independent individuals.

In vivo seeding competition assay

   4T1 cells (Cas9CTL and sg-MKNK1) were labeled using CellTrace Violet
   (Invitrogen, #34571) or Far Red (Invitrogen, #[292]C34564) dyes
   according to the manufacturer’s instructions before intrasplenic
   injection. Each mouse received a total of 1 × 10^6 cells composed of
   one MNK1 KO clone and Cas9CTL cell lines at a 1:1 ratio, with each cell
   line labeled using a different color of CellTrace dye. At 48 hours
   after injection, mice were euthanized, and the liver perfused via the
   hepatic portal vein with 15 ml of cold PBS. Following this, livers were
   excised and single cell–dissociated using the gentleMACS Octo
   Dissociator (Miltenyi Biotec, #130-096-427). Cell suspensions were then
   analyzed on an LSRFortessa (BD Biosciences) for the presence of
   fluorescent tumor cells.

RNA-seq and analysis

   RNA was isolated from Cas9CTL and sg-MKNK1 4T1 cell pellets in
   quadruplicate using the Absolutely RNA Miniprep Kit (Agilent
   Technologies, #400800). RNA concentration and quality were assessed
   using a Bioanalyzer (Agilent Technologies). Note that the RNA quality
   for one sg-MKNK1-2 technical replicate was below threshold and this
   sample was therefore excluded. Library preparation was performed using
   the KAPA mRNA HyperPrep Kit (Roche). Samples were sequenced on a
   NextSeq High-Output 2× 75–base pair flow cell with a yield of ~20 M
   single-end reads per sample. FASTQ data were normalized using the
   Nextflow nf-core pipeline (DOI: 10.5281/zenodo.1400710) and expressed
   as log-transformed counts per million reads mapped (log[10]CPM). The
   DESeq2 R package was used to identify differentially expressed genes by
   calculating the fold change in log[10]CPM values between MNK1 KO
   (sg-MKNK1-1 or sg-MKNK1-2) and Cas9CTL cells using default settings.
   Results were then filtered on the basis of fold change (greater than 2
   in either direction) and adjusted P value (less than 0.05). P values
   were calculated using the Wald test with false discovery rate
   (FDR)/Benjamini-Hochberg multiple comparison test. PCA for gene
   expression data was performed through the factoextra R package. One
   Cas9CTL technical replicate was found to be an outlier and thus removed
   from the dataset. Functional enrichment analyses were performed on
   preranked fold change data using GSEA software (Broad Institute,
   version 4.1.0). GSEAPreranked was run using default settings against
   the KEGG molecular signature database on the “Mouse Gene Symbol
   Remapping Human Orthologs MSigDB.v.2023.2.Hs.chip” platform.

Preparation of cell lines for proteome analysis

   Quadruplicates of 4T1 Cas9CTL and sg-MKNK1 cell lines were lysed by the
   addition of 100 μl of lysis buffer containing 5% SDS in 100 mM tris (pH
   8.5) and heating the samples (95°C) for 10 min. Samples were
   consequently subjected to probe based ultrasonication (Thermo Fisher
   Scientific Sonic Dismembrator) and clarified by centrifugation at
   21,000g for 5 min. Five percent of the resulting extract was reserved
   for the determination of protein concentration using bicinchonic acid
   assay (Thermo Fisher Scientific/Pierce). The remainder of the cell
   lysates were reduced in 20 mM tris(2-carboxyethyl)phosphine for 30 min
   at 60°C and alkylated in 25 mM iodoacetamide for 30 min in the dark at
   room temperature. Fifty micrograms of protein was proteolytically
   digested with trypsin (Promega, sequencing grade) overnight at 37°C
   using S-Trap microcartridges (ProtiFi). Peptides were eluted from
   S-Trap microcartridges sequentially with 50 mM ammonium bicarbonate,
   0.2% formic acid, and 50% acetonitrile respectively. Peptide-containing
   eluates were frozen at −80°C and lyophilized to dryness using a vacuum
   centrifuge (Labconco). Peptide samples were reconstituted in 0.1%
   formic acid to a final concentration of 150 ng/μl.

Liquid chromatography–MS/MS data acquisition and analysis

   An equivalent of 1 μg of protein was analyzed by nanoflow liquid
   chromatography–MS/MS using an Easy-nLC 1200 system coupled to a Q
   Exactive Plus (both Thermo Fisher Scientific). Samples were first
   loaded onto a precolumn (Acclaim PepMap 100 C18, 3 μm, 75 μm in
   internal diameter × 2 cm in length; Thermo Fisher Scientific) for
   online desalting and then separated using a 100-min gradient with an
   Acclaim PepMap 100 C18 analytical column (2-μm particle, 75 μm in ID ×
   250 mm; Thermo Fisher Scientific) using water with 0.1% formic acid
   (mobile phase A) and 84% acetonitrile with 0.1% formic acid (mobile
   phase B) at 300 nl/min from 3 to 40% B. MS acquisition was conducted in
   data-dependent acquisition mode on the basis of the top 15 most intense
   precursor ions (+2 to +4 charge state). Full MS scans were acquired at
   70,000 resolution from 350 to 1500 mass/charge ratio [automatic gain
   control (AGC), 1 × 10^6; 50-ms maximum injection time], and MS2 spectra
   were collected at 17.5K resolution (AGC, 2 × 10^4; 64-ms maximum
   injection time) using a normalized collision energy of 28. Dynamic
   exclusion was set to 40 s. Acquired MS/MS data were analyzed using
   Proteome Discoverer version 2.5 (Thermo Fisher Scientific). Database
   searching was conducted using the Sequest HT node, using a reference
   mouse proteome FASTA file containing only reviewed canonical sequences
   downloaded from UniProt (downloaded 28 February 2019). Label-free
   quantitation was conducted using the Minora feature detection node.
   Proteins were quantified on the basis of unique peptides and scaled on
   the basis of total peptide abundance, and missing values were imputed
   using the low abundance resampling method. Protein expression ratios
   were calculated on the basis of protein abundance, and P values were
   calculated using analysis of variance (ANOVA) (individual proteins)
   within Proteome Discoverer version 2.5. Only proteins quantified with
   at least one protein unique peptide in three of four biological
   replicates from at least one sample group were retained. Statistical
   significance of differentially expressed proteins was based on a
   two-sigma fold-change cutoff (effectively twofold change), as well as
   having an FDR-adjusted P value of less than 0.05. PCA for gene
   expression data was performed through the factoextra R package.
   Statistically differentially expressed proteins were used an input for
   canonical pathway enrichment analysis using Ingenuity Pathway Analysis
   software under default settings (QIAGEN).

Metabolomics

   For each biological replicate, 4T1 Cas9CTL and sg-MKNK1 cells were
   seeded at 62,500 cells per well in technical triplicate in a six-well
   tissue culture treated plate (Corning) and cultured in 4T1 complete
   medium (defined above). The following day, cells were examined, and
   medium was aspirated, replaced with fresh medium, and incubated for
   another 20 hours. Next, medium for U-[^13C]-glucose (Cambridge
   Isotopes) tracing was prepared using a base medium of 1× DMEM (Wisent,
   #319-062) subsequently supplemented with sodium pyruvate, glutamine,
   Hepes, and sodium bicarbonate to concentrations equivalent to 4T1
   complete medium. Next, a 5× (125 mM) solution of U-[^13C]-glucose was
   prepared by dissolving U-[^13C]-glucose in the supplemented 1× DMEM.
   4T1 complete medium was then aspirated from all wells and replaced with
   2 ml of supplemented 1× DMEM, while 0.5 ml of 5× U-[^13C]-glucose was
   added overtop, giving a final concentration of 25 mM, and incubated (or
   traced) in 5% CO[2] at 37°C for 4 hours. Two wells of unlabeled
   controls were seeded for each cell line, which contained 2 ml of
   supplemented 1× DMEM and 0.5 ml of 5× ^12C-glucose at a final
   concentration of 25 mM and incubated as above for 4 hours. In addition,
   to normalize downstream metabolomic intensity values to cell number,
   three wells per cell line were seeded in 4T1 complete medium and
   counted at the experimental end point. To prepare the cells for
   metabolomic profiling, after 4 hours of incubation, plates were
   immediately placed on ice, and medium was aspirated. Next, wells were
   washed twice with 1 ml of 0.9% saline. Following saline aspiration,
   plates were placed over dry ice for 5 min, wrapped in aluminum foil,
   and immediately placed in −80°C. Cells were lysed and processed using
   the stable isotope labeling and metabolomics method published in Kaymak
   et al. ([293]44).

Seahorse bioenergetic profiling

   Before bioenergetic profiling using the Seahorse Mito Stress Test Kit
   (Agilent Technologies), 4T1 Cas9CTL and sg-MKNK1 cells were seeded at
   5000 cells per well in a 96-well Seahorse XFe96 FluxPak plate (Agilent
   Technologies) and incubated with 80 μl of DMEM (see conditions above)
   for 24 hours in 5% CO[2] at 37°C incubator until ~70% confluence was
   reached. In addition, 24 hours before bioenergetic profiling, the
   Seahorse Flux Assay Kit was calibrated with 200 μl of Milli-Q water
   (Millipore) and left in a non-CO[2], 37°C incubator for 24 hours. The
   following day, seeded cells were examined for appropriate confluence,
   and 80 μl DMEM culture medium was gently removed using a multichannel
   pipette and replaced with 140 μl of Seahorse XF assay medium: Seahorse
   XF DMEM base medium (Agilent Technologies), supplemented with 1 mM
   sodium pyruvate, 2 mM glutamine, and 10 mM glucose (final
   concentrations) (pH 7.4) at 37°C. In addition, water was removed from
   the flux assay kit and replaced with Seahorse XF calibrant solution
   (Agilent Technologies) and incubated at 37°C (non-CO[2]) for 2 to 3
   hours before conducting bioenergetic profiling. During the 2- to 3-hour
   incubation, Mito Stress Test Kit drugs—oligomycin, carbonyl cyanide
   p-trifluoromethoxyphenylhydrazone (FCCP), rotenone/antimycin A, and
   monensin—were dissolved in Seahorse XF assay medium to working
   concentrations of 16, 9, 5, and 220 μM, respectively. Next, 20 μl of
   each drug was pipetted into injection ports A, B, C, and D,
   respectively. Final drug concentrations of oligomycin, FCCP,
   rotenone/antimycin A, and monensin were 2, 1, 0.5, and 20.1 μM,
   respectively. Following the 2- to 3-hour incubation of the
   extracellular flux assay kit, probes were calibrated in the Seahorse
   Bioanalyzer. After, the flux assay kit was replaced with the 96-well
   plate containing Cas9CTL and MNK KO cells, while the drugs within ports
   remained in the machine. Oligomycin, FCCP, rotenone/antimycin A, and
   monensin were automatically injected overtop cells at 20, 40, 60, and
   80 min, respectively. OCR and ECAR were both measured. OCR and ECAR
   values were normalized to cell number based on crystal violet staining
   and absorption at 595 nm following solubilization.

Measurement of ATP levels

   Relative ATP levels were assessed using the Incucyte CytoATP Lentivirus
   Reagent Kit (Sartorius, #4772) according to kit instructions. Data were
   acquired using an Incucyte SX5 Live-Cell Analysis System configured
   with an SX5 Metabolism Optical Module (Sartorius, #4820). Briefly, 4T1
   cells (Cas9CTL and sg-MKNK1) were transduced with CytoATP, and CytoATP
   nonbinding control lentivirus and stably expressing populations were
   derived via puromycin selection. Selected cells were then seeded in
   96-well plates at a density of 2000 cells per well in 4T1 medium with a
   glucose concentration of 10 mM (low glucose). The following day, cells
   were treated with oligomycin and placed into the Incucyte SX5 for
   immediate scanning. Images were subsequently acquired every 30 min for
   24 hours in total and analyzed using built-in software. Corrected ATP
   ratios were calculated by normalizing against CytoATP nonbinding
   controls and are expressed relative to Cas9CTL at baseline.

Cell assessment by flow cytometry

   4T1 cells (Cas9CTL and sg-MKNK1) were trypsinized, centrifuged at 300g
   for 5 min, and washed in PBS. For mitochondrial assays, 1 × 10^5 cells
   were stained with 1 μM MitoSOX mitochondrial superoxide indicator
   (Invitrogen, #[294]M36008) or 200 nM MitoTracker CMXRos (Invitrogen,
   #M7512) according to the manufacturer’s directions. For stemness
   assays, 2 × 10^5 cells were stained with fluorescently tagged
   antibodies that recognize CD44 and CD24 for 30 min on ice. Antibody
   information is listed in table S1. After staining, cells were washed in
   PBS, and fluorescence was assessed on an LSRFortessa (BD Biosciences).

ALDEFLUOR assay

   ALDH activity in 4T1 cells (Cas9CTL and sg-MKNK1) was measured
   according to ALDEFLUOR kit instructions (STEMCELL Technologies,
   #01700). Briefly, 1 × 10^6 cells were suspended in the provided buffer
   and incubated with 1.5 μM BODIPY-aminoacetaldehyde for 30 min at 37°C.
   Using this protocol, ALDH activity is proportional to cellular
   fluorescence, as measured using an LSRFortessa (BD Biosciences).
   Background fluorescence was accounted for by treating cells with 15 μM
   diethylaminobenzaldehyde, a selective inhibitor of ALDH (negative
   control). Data were analyzed using FlowJo version 10.7.1.

Analysis of publicly available TCGA patient data

   The TCGA Breast Invasive Carcinoma PanCancer Atlas database was
   accessed through cBioPortal. Analysis was limited to only patients for
   whom MNK1 protein expression z-score, and overall survival data were
   available (n = 102). Full protein expression data and matching clinical
   correlates were downloaded for these patients only. Expression heatmaps
   were generated using Prism 9. ssGSEA was subsequently performed
   according to methodology described in Barbie et al. ([295]78), using
   KEGG Glycolysis_Gluconeogenesis, Oxidative_Phosphorylation,
   Citric_Cycle_TCA_Cycle, Wnt_Signaling, and Hedgehog_Signaling gene
   sets. Analysis was performed on the GenePattern cloud–based public
   server, using the “ssGSEA” module (v10.1.0). Input data (protein
   expression z-scores from MS by the National Cancer Institute’s Clinical
   Proteomic Tumor Analysis Consortium) for relevant patients were
   downloaded from cBioPortal. Gene set database files were downloaded
   from the GSEA MSigDB. Default settings were used for all other analysis
   parameters. Pearson rank-order calculations were performed to correlate
   MNK1 protein expression with individual proteins of interest and ssGSEA
   enrichment scores. Analysis of overall survival data was performed on
   the highest and lowest quartile of patients when ranked by MNK1 protein
   expression.

Histology and IHC

   Following FFPE processing, all samples were sectioned at 4 μm in
   thickness according to staining requirements. Tumor-bearing livers were
   step-sectioned (two steps, 200 μm between steps) for subsequent
   staining with H&E as previously described ([296]17). IHC stainings were
   performed on primary tumor and tumor-bearing liver tissues. Slides were
   deparaffinized, rehydrated, and subjected to heat-induced antigen
   retrieval. Endogenous peroxidase activity was quenched for 15 min in
   10% hydrogen peroxide, and tissues were blocked in 10% donkey serum
   (Jackson ImmunoResearch, #017-000-121). Tissues were then stained with
   antibodies specific to MNK1 (tumors; 1:50 dilution) or phospho-eIF4E
   (tumor-bearing livers; 1:50 dilution). Stainings were visualized using
   ImmPACT DAB (3,3′-diaminobenzidine) (Vector Laboratories, #SK-4105) and
   counterstained with Harris-modified hematoxylin (EMD Millipore,
   #638A-85). All stained slides were imaged using a Zeiss AxioScan.Z1
   slide scanner and analyzed using QuPath version 0.2.3. See table S1 for
   antibody details.

Immunoblotting

   Immunoblotting was performed as previously detailed ([297]15). Briefly,
   protein lysates were isolated from cell pellets using
   radioimmunoprecipitation assay buffer [150 mM tris-HCl (pH 7), 150 mM
   NaCl, 1% NP-40, 1% sodium deoxycholate, and 0.1% SDS]. Protein
   concentrations were measured by Bradford reagent assay (Bio-Rad),
   separated on a 10% SDS-polyacrylamide gel electrophoresis gel, and
   transferred to nitrocellulose membranes. Membranes were blocked in 5%
   nonfat milk and incubated in primary antibody dilutions overnight at
   4°C. The following day, membranes were incubated with secondary
   antibody at room temperature for 1 hour. Membranes were then revealed
   using ECL Prime Western Blotting Detection Reagent (Amersham) and
   Ultra-High Contrast Western Blotting Film. Antibody information is
   listed in table S1. Immunoblot densitometry analysis was performed for
   all proteins, normalizing against a β-actin loading control, and is
   displayed in fig. S9.

Cell proliferation growth curves

   4T1 cells (Cas9CTL and sg-MKNK1) were serum-starved for 16 hours before
   seeding in 96-well plates at a density of 1000 cells per well. At
   24-hour intervals, up to 96 hours after seeding, cells were fixed in 4%
   PFA for 15 min at room temperatures and stored in 0.4% PFA. Fixed cells
   were stained with 0.5% crystal violet at end point for 30 min at room
   temperature. The following day, crystal violet was dissolved in 10%
   acetic acid, and absorbance was measured at 590 nm using a plate
   reader. Cell proliferation over time was calculated relative to a
   standard curve of known cell numbers.

Statistics

   Prism 9 software (GraphPad) was used to perform statistical testing.
   Details of the statistical testing performed for all relevant figure
   panels can be found in in figure legends. P values for the differences
   of means between groups were calculated using unpaired t tests, one-way
   ANOVA, two-way ANOVA, and Pearson rank-order correlations, with
   multiple comparison tests as appropriate. P values less than 0.05 were
   considered statistically significant.

Acknowledgments