CELF1 is a non-canonical eIF4E binding protein that promotes translation of epithelial-mesenchymal transition effector mRNAs
Arindam Chaudhury, Natee Kongchan, Shebna A Massey, Rajesh Sharma, Rituraj Pal, Na Zhao, Phoebe Tsoi, Yingmin Zhu, Emuejevoke Olokpa, Sufeng Mao, Sonia del Rincon, Lucas C Reineke, Richard E Lloyd, Marco Sardiello, Jeffrey M Rosen, Choel Kim, Josephine C Ferreon, Joel R Neilson
TL;DR
The study shows how CELF1 promotes translation of specific mRNAs during cell transformation, using a non-standard pathway involving eIF4E and PABPC1.
Contribution
The paper identifies a novel non-canonical translation mechanism involving CELF1, eIF4E, and PABPC1 that regulates epithelial-mesenchymal transition.
Findings
CELF1 binds eIF4E and PABPC1 to promote translation of GRE-containing mRNAs independently of eIF4G1.
Disrupting the CELF1/eIF4E interaction inhibits epithelial-mesenchymal transition and metastasis in vivo.
Abstract
Mounting evidence is revealing an increasing complexity of gene regulation at the level of messenger RNA (mRNA) translation. Within mammalian cells, canonical cap-dependent mRNA translation depends on the eIF4F complex, consisting of the m7G mRNA cap-binding protein eukaryotic initiation factor 4E (eIF4E), the helicase eIF4A (eIF4A), and the eIF4G (eIF4G1) scaffolding protein. eIF4G1 additionally binds poly(A) binding protein (PABPC1) to facilitate mRNA circularization and nucleates pre-translation initiation complex assembly to initiate ribosomal scanning. In breast epithelial cells, the CELF1 RNA-binding protein specifically promotes the translation of select epithelial-to-mesenchymal transition (EMT) effector mRNAs by binding GU-rich elements (GREs) within their 3′ untranslated regions (UTRs). Here we show that CELF1 directly binds to both eIF4E and PABPC1 to promote…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7- —Samuel Waxman Cancer Research Foundation10.13039/100001384
- —Adrienne Helis Malvin Medical Research Foundation10.13039/100006387
- —National Cancer Institute10.13039/100000054
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsPI3K/AKT/mTOR signaling in cancer · RNA Research and Splicing · Genetic Syndromes and Imprinting
Introduction
Regulatory control at the level of messenger RNA (mRNA) translation contributes significantly to gene expression and function [1–4]. Recent findings indicate that coordinated changes in post-transcriptional regulatory networks can alter cellular phenotype and behavior [2, 5]. A core cellular process underlying development, tumor metastasis, and tumor radiation and chemoresistance is epithelial-to-mesenchymal transition (EMT) [6, 7]. Since metastasis and therapy resistance to insult associated with cellular de-differentiation are the foremost causes of cancer lethality [8, 9], it is critical to better understand the mechanisms by which EMT, and thus these two characteristics, are promoted.
It has been well-documented that both oncogenes and signaling pathways regulating expression of these oncogenes converge on the regulation of mRNA translation [2, 10]. Indeed, within the process of cellular transformation, mechanisms at each stage of translation and the ribosomal machinery may be co-opted to foster translation that perpetuates oncogenic programs [11–13]. We previously defined a translational regulatory circuit driving EMT in vitro and cancer progression in vivo in breast epithelial cells [14]. Via polysomal profiling, we identified the RNA-binding protein CELF1 as the key regulator of this circuit. CELF1, which is both necessary and sufficient for EMT in vitro and experimental metastasis in vivo, promotes translation of select EMT effector mRNAs by binding to GU-rich elements (GREs) within the 3′ untranslated regions (UTRs) of these mRNAs. Molecular genetic epistasis analysis confirmed that CELF1’s role in EMT is mediated by its translational targets. However, our previous work did not establish a mechanism by which CELF1 drives translation within this context.
Canonical eukaryotic translation initiation is a well-defined and ordered process that guides the assembly of ribosomes onto m^7^G-capped mRNAs [15]. The m^7^G cap structure on the mRNA is first bound by the eIF4E protein. eIF4E recruits the other two members of the tripartite eIF4F complex—the scaffold protein eIF4G1 and the RNA helicase eIF4A—to the cap structure [15]. eIF4G1 bridges eIF4E and eIF4A to poly(A) binding protein (PABPC1). In addition, eIF4G1 binds the multi-subunit eIF3 factor, which in turn coordinates a pre-assembled 43S pre-initiation complex (PIC) comprised of eIF1, eIF1a, eIF5, and the (small) 40S ribosomal subunit bound by the eIF2-GTP-Met-tRNAi ternary complex, to form the 48S PIC [16]. eIF4G1’s role is so central to cap-dependent translational initiation that cleavage of this protein in virally infected cells is sufficient to broadly disrupt cap-dependent translation, giving the virus complete control of the cell [17, 18].
In this work, we demonstrate that CELF1 promotes translation of its GRE-containing mRNA targets in mesenchymal cells via a cap-dependent mechanism. We further show that CELF1 associates with both eIF4E and PABPC1 in cellular extracts, and within intact cells. We find that the translation of CELF1’s targets is both eIF4G1-independent and dependent upon the GRE within the 3′ UTR of these targets. The interaction of CELF1 with eIF4E depends on phosphorylation of eIF4E. We identify both the conserved peptide sequence within CELF1 mediating this interaction and the residues within the eIF4E protein that mediate binding of a CELF1-derived peptide. We demonstrate that CELF1 effectively competes with eIF4G1 from eIF4E and directly binds PABPC1. Finally, we show that CELF1’s interaction with eIF4E is required for both EMT in vitro and experimental metastasis in vivo.
Materials and methods
Cell culture and treatment
The MCF-10A cell line was obtained from the ATCC (Manassas, VA) and cultured as described previously [14]. MCF-10A stable cells expressing HA-EIF4E or HA-EIF4E^S209A^ were kind gifts from Dr Wilson H. Miller, Jr (Lady Davis Institute for Medical Research, Segal Cancer Centre, Jewish General Hospital, and McGill University, Montreal, Quebec, Canada) [19]. MCF-10A cells were treated for 3 days with 5 ng/ml of TGF-β1 (R&D Systems, Minneapolis, MN).
Cell lysis, immunoblot, and immunoprecipitation
Cell lysis and immunoblotting were performed as described previously [14, 20]. Supplementary Table S1 provides the list and associated information of antibodies used in the current study. All blots were also probed with GAPDH to confirm equal loading. For immunoprecipitation, cells were lysed with NP-40 lysis buffer (150 mM NaCl, 1% NP-40, and 50 mM Tris-Cl (pH 8.0) supplemented with protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, Waltham, MA). Lysates (800 µg) were incubated at 4°C with 10 µg of indicated antibodies (Supplementary Table S1) or IgG (mouse or rabbit based on the antibody used for the immunoprecipitation) for 4 h followed by addition of 20 μl of protein A/G agarose beads (Thermo Fisher Scientific) for 2 h. Immune complexes were washed 3 × 5 min in lysis buffer at 4°C with constant rotation. Beads were quickly centrifuged at maximum speed and immunoprecipitates were subjected to immunoblot analysis. Where indicated immunoprecipitates were treated with RNase A (100 µg per mg of lysate; Sigma–Aldrich, St. Louis, MO), RNase I (1000 units per mg of lysate; Thermo Fisher Scientific AM2295) [21], and/or recombinant coxsackievirus B3 2A protease (4 ng/µl) for 30 min at 37°C [22].
Proximity ligation assays
Cells were seeded on poly-D-lysine-coated 12 mm circle glass coverslips in 24-well plates. After fixation with 4% paraformaldehyde in PBS for 30 mins, cells were permeabilized with 0.25% Triton X-100 in PBS for 10 min. Proximity ligation assays were performed using the Duolink^®^ In Situ Red Starter Kit Mouse/Rabbit (Sigma–Aldrich) as per the manufacturer’s instructions. Briefly, coverslips were placed in a humidified chamber, blocked with Duolink^®^ blocking solution for 60 min at 37°C, and then incubated with 1:1000 dilutions of primary antibodies for 1 h at room temperature. Coverslips were washed twice for 5 min, and PLA probe was added for 1 h at 37°C. Coverslips were again washed twice for 5 min, and ligase was incubated for 30 min, again at 37°C. Coverslips were again washed prior to addition of polymerase and incubation at 37°C for 100 min, then washed again and mounted in Duolink^®^ In Situ Mounting Media with DAPI. Images were acquired on an ECHO Revolve FL or Nikon A1 confocal fluorescent microscope using the 40× objective lens and processed using ImageJ and Photoshop software.
Polysome profiling, RNA immunoprecipitation, and quantitative real-time PCR
Polysome profiling from MCF-10A cells stably expressing either wild-type (WT) or S209A mutant eIF4E and transiently transfected with a CELF1 expression plasmid [19] was performed as described previously [14] by fractionating 30 OD units of cytoplasmic extract over 10%–50% sucrose gradients. We used TRIzol LS reagent (Thermo Fisher Scientific) to extract RNA from equal volumes of the various polysome fraction and total lysate aliquots as described before [14]. RNA was quantitated on a Tecan M200 plate reader via the A260/280 method. One microgram of RNA isolated from equal volume of polysome fractions or input total lysate was primed with random hexamers and reverse transcribed using SuperScript III Reverse Transcriptase (Thermo Fisher Scientific) as per the manufacturer’s instructions. Control reactions omitted reverse transcriptase. Equivalent volumes of complementary DNA (cDNA) were subsequently used to template triplicate 10 μl quantitative polymerase chain reaction (qPCR) reactions using KAPA SYBR FAST Universal 2× qPCR Master Mix (KAPA BIOSYSTEMS, Wilmington, MA) and 0.1 μM of the primers indicated in Supplementary Table S2, using the manufacturer’s recommended cycling parameters on a VIIA7 cycler (Applied Biosystems) and using the melting curve to assess reaction specificity. Reaction data were analyzed on QuantStudio Software (Applied Biosystems). Comparative expression data were calculated via the -^ΔΔ^C_t_ method, using β-actin as a reference. RIP and data analyses were performed as described previously using primers described in Supplementary Table S2 [14].
m7GTP chromatography
Cytoplasmic extracts [14] were subjected to m^7^GTP chromatography using immobilized γ-aminophenyl-m^7^GTP (C10-spacer) beads (Jena Bioscience, Germany). Beads were equilibrated with buffer A (100 or 200 mM KCl, 50 mM Tris–HCl [pH 7.5], 5–10 mM MgCl_2_, and 0.5% Triton X-100) plus BSA (0.1 mg/ml) at 4°C for 30 min. The resin was washed and incubated with 500 µg of protein extract, and either left untreated or treated with 4 ng/µl of recombinant coxsackievirus B3 2A protease for 60 min at 4°C. GTP (100 mM) was added to reduce nonspecific binding. The beads were washed with 0.4 ml of buffer A and then incubated for 30 min with 200 mM of m^7^GTP or GTP.
Plasmid constructs
The pEGFP-N1-CELF1 putative eIF4E binding mutants were generated via site-directed mutagenesis using the QuickChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies, La Jolla, CA) and primers listed in Supplementary Table S2. For recombinant protein expression, the WT and Δ365–371 mutant CELF1 were cloned in the pET His10 TEV LIC cloning vector (2B-T-10) (a gift from Scott Gradia, Plasmid #78173, Addgene, Cambridge, MA) using NEBuilder HiFi DNA Assembly Master Mix (NEB) and primers listed in Supplementary Table S2. Other CELF1 and Renilla luciferase expression plasmids utilized herein have been described [14]. The GST-EIF4E (WT and W73A mutant) cloned into the pGEX-5X-1 vector were kind gifts from Dr Yan-Hwa Wu Lee [23] and were used to generate the phosphomimic WT and W73A mutant GST-EIF4E by site-directed mutagenesis using primers listed in Supplementary Table S2. The WT and W73A EIF4E coding sequence entry constructs for overexpression were generated from FLAG-EIF4E, and FLAG-EIF4E (W73A) plasmids gifted by Dr Katherine L. Borden [24]. Expression constructs for WT and W73A mutant EIF4E were generated by Gateway Cloning (Thermo Fisher Scientific) into the pLenti6.3 vector. shRNA plasmid constructs targeting the 3′ UTR of EIF4E were cloned into the pGIPZ backbone using oligonucleotides listed in Supplementary Table S2.
Transfection and transduction
Transient transfection was performed using Lipofectamine LTX (Life Technologies, Carlsbad, CA), per the manufacturer’s instructions. The pGIPZ lentiviral particles were generated by transfection of HEK-293T cells using Lipofectamine 2000 (Thermo Fisher Scientific), per the manufacturer’s instructions. The pLenti6.3 lentiviral particles were generated by transfection of HEK-293T cells as described previously [25]. For transduction, early passage cells were seeded at 500 000 cells per 10-cm^2^ dish one day prior to infection, and transduction was performed as described previously [26]. To generate the EIF4E (W73A) overexpressing stable cells, the MCF-10A cells were first transduced with the pLenti6.3-EIF4E (W73A) and selected with Blasticidin (5 µg/ml; R&D, Minneapolis, MN). Selected cells were then transduced with pGIPZ shRNA targeting the 3′ UTR of EIF4E and selected with puromycin (2 µg/ml; R&D, Minneapolis, MN).
Surface sensing of translation (SuNSET)
The SuNSET technique was performed as described previously [27]. Briefly, indicated cells were labeled for 10 min using puromycin (10 µg/ml). Cells were harvested, lysed as described earlier, and resolved using sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE). Blots were probed with anti-puromycin antibody (Kerafast, Boston, MA) and subsequently stained with Coomassie Blue R250 (Sigma–Aldrich).
Recombinant protein expression and purification
The pGEX-5X-1 WT/S209D and W73A/S209D EIF4E constructs were transformed into competent BL21(DE3) Escherichia coli (Thermo Fisher). One colony was picked the following day, inoculated in 10 ml of Luria-Bertani (LB) broth, and grown overnight (200 rpm at 37°C). The inoculum was added into 1000 ml LB (1:100) and grown at 200 rpm at 37°C for 2 to 3 h until OD_600_ was between 0.6 and 0.8, at which point it was induced with 0.4 mM isopropyl β-d-1-thiogalactopyranoside (IPTG, Thermo Fisher Scientific) and grown overnight at 200 rpm at 20°C. Following growth, cells were harvested and pelleted (5000 × g, 10 min). The cell pellet was resuspended in 50 ml lysis buffer [50 mM PBS, pH 7.6, 500 nM NaCl, 1 nM EDTA, and 1 mM PMSF] and disrupted by passing through a microfluidizer. Lysate was clarified by centrifugation at 50 000 g for 30 minutes at 4°C. Clarified lysate was passed through a 0.22 µM syringe filter and loaded onto a 1 ml GE GSTrap 4B column (GE AKTA) at 0.5 ml/minute. The column was washed with 20 column volumes of lysis buffer before elution using freshly prepared lysis buffer containing 20 mM glutathione at 0.5 ml/minute. The expression plasmids for the 6xHis-tagged CELF1 proteins were transformed, cultured, and induced similarly as above. The His SpinTrap Kit (GE Healthcare) was used according to manufacturer’s recommendations to purify the recombinant CELF1 proteins. Finally, buffer exchange was performed using 5 ml desalting spin columns (MilliporeSigma, Burlington, MA). Desalted and purified protein samples were aliquoted, glycerol was added at 10% (v/v), snap frozen, and stored at -80°C until further use. The C-MYC/DDK-tagged purified recombinant human eIF4G1, eIF4A1, and PABPC1 proteins were purchased from OriGene (Rockville, MD).
In vitro binding assays
Recombinant, bait GST-tagged eIF4E or 6xHis-tagged CELF1 proteins (1 μM) were initially bound to glutathione-agarose beads (Thermo Fisher Scientific) or Ni^2+^ beads (GE Healthcare), respectively. Post-binding, beads were washed three times with wash buffer containing 50 mM Tris-HCl (pH 7.4) and 20% glycerol before addition of prey proteins (0.1–10 μM). Mixtures were incubated at room temperature with rocking for 2 h. Beads were washed three times with wash buffer containing 500 mM NaCl. The protein bound to the beads was resolved by SDS-PAGE and analyzed by both Coomassie staining and immunoblot analysis.
Mammalian cell-free in vitro translation
WT or GRE deletion mutant CRLF1 and SNAI1 3′ UTRs were fused downstream of the Renilla luciferase coding sequence in the broadly used pRL-TK CXCR4 6x reporter plasmid ([28], Plasmid #11 308, Addgene, Cambridge, MA) as described before [14]. T7 forward and respective 3′ UTR-specific reverse primers, listed in Supplementary Table S2, were used to generate PCR templates for in vitro transcription from pRL-TK CXCR4 6x, pRL-TK CRLF1, pRL-TK CRLF1 ΔGRE, pRL-TK SNAI1, and pRL-TK SNAI1 ΔGRE plasmids. Capped and polyadenylated mRNA templates were generated using the mMESSAGE mMACHINE kit and Poly(A) Tailing kit, respectively (Thermo Fisher Scientific).
MCF-10A cells were transiently transfected with shRNA targeting either GLB1 or CELF1 as described previously [14] and subsequently treated with TGF-beta for 72 h. Cell-free extracts from these cells were prepared and in vitro translation was performed as described previously [29]. Briefly, cells were lysed in freshly prepared ice-cold hypotonic lysis buffer [10 mM HEPES, pH 7.6, 10 mM potassium acetate, 0.5 mM magnesium acetate, 5 mM DTT, EDTA-free protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific)]. Where indicated extracts were treated with 4 ng/µl recombinant coxsackievirus B3 2A protease as described earlier. For immunodepletion of eIF4G1, extracts were processed as described above for immunoprecipitation with an anti-eIF4G1 antibody or beads alone (Supplementary Table S2), and the process was repeated twice.
In vitro translation reactions were set up in a total reaction volume of 10 µl per sample [containing 4 µl of cell-free extract, 1 µl of translation buffer (16 mM HEPES, pH 7.6, 20 mM creatine phosphate, 0.1 µg/µl creatine kinase, 0.1 mM spermidine and 100 µM amino acids (Sigma–Aldrich)], 0.4 µl of 1M potassium acetate, 0.2 µl of 100 mM magnesium acetate, 20 U RNasin (Promega), and 0.04 pmol of mRNA template. Reactions were incubated for 30 min at 37°C. A reaction with no added mRNA template was used as a background control. Translation read-out was performed using the dual-luciferase reporter assay system (Promega) as per the manufacturer’s protocol on a Tecan M200 multimode reader running Tecan Magellan software (Tecan). Data is presented as mean ± standard deviation (SD) of luciferase light units.
Nuclear magnetic resonance
Recombinant plasmids encoding eIF4E^S209D^ were transformed into E. coli BL21 Star (DE3) chemically competent cells (Invitrogen). Transformed cells were grown at 37°C in Terrific Broth with 100 μg/ml carbenicillin for 6 h. One hundred microliters of bacterial cells were added to 50 ml of 1× M9 salts (GenDEPOT) containing 1 g/l [^15^N]-NH_4_SO_4_, 4 g/l D-glucose, 1 mM MgCl_2_, 0.1 mM CaCl_2_, 1× BME Vitamins (Millipore Sigma), and 100 μg/ml carbenicillin and grown at 37°C overnight. Overnight cultures were added to 950 ml of the same medium described earlier and incubated at 37°C to an OD_600_ of 1.2–1.4, at which point protein expression was induced with 1 mM IPTG. Cells were further incubated at 18°C overnight and harvested by centrifugation at 7900 × g.
NMR titration experiments were performed on a Bruker Avance III HD 800 MHz spectrometer equipped with a cryogenic triple-resonance probe. Uniformly ^15^N-labeled eIF4E S209D protein was prepared at a concentration of 100 µM in 50 mM phosphate buffer (pH 6.9), containing 100 mM NaCl, and 10% D_2_O for lock purposes. A saturating concentration of 7-methylguanosine (m^7^G, 150 µM) was pre-incubated with the protein to ensure complete binding saturation prior to peptide titration. The titration was conducted with a synthetic CELF1-derived long peptide (>95% purity, New England Peptide) added sequentially to achieve final protein-to-peptide molar ratios of 1:0.2, 1:0.5, 1:1, 1:3, and 1:7. At each titration point, sensitivity-enhanced ^1^H-^15^N 2D HSQC spectra were recorded at 25°C, employing standard Bruker pulse sequences with 2048 (^1^H) × 256 (^15^N) data points. Spectral widths of 16 ppm (^1^H) and 36 ppm (^15^N) were used with carrier frequencies set at water resonance (∼4.7 ppm) and central amide region (∼119 ppm), respectively. All NMR spectra were processed and analyzed using NMRPipe [30] and NMRFAM-Sparky [31]. Backbone amide assignments were based on previously reported BMRB entries 5712 and 7115 [32].
In vitro migration and invasion assay
In vitro migration and invasion assays were performed using Culturex 96 Well Cell Migration and Invasion Assay kits (Trevigen, Gaithersburg, MD) as described before [14]. Data obtained were used to analyze percent migration and invasion and were expressed as percent mean ± SD.
Animal studies
All mouse procedures were approved by the Institutional Animal Care and Use Committees of the Baylor College of Medicine and were performed as previously described [14]. Six-week-old T-cell-deficient female homozygous nude mice (NCr-Foxn1^nu^) (Taconic, Hudson, NY, USA) were used for the experimental metastasis experiments (n = 4 in parental group, n = 4 in WT CELF1 group, and n = 6 in CELF1^ΔY^ mutant group). To assess the experimental metastatic potential of cells, 10^6^ indicated MCF-10AT1 cell variants labelled with GFP-Firefly luciferase were injected into animals via the tail vein. Mice were assessed weekly for metastasis using in vivo bioluminescence imaging using an IVIS Imaging System (IVIS imaging system 200, Xenogen Corporation, PerkinElmer, Waltham, MA, USA). Mice were euthanized on day 15 post tail-vein injection at which time the lungs were surgically removed and fixed using 10% neutral buffered formalin. Lungs were further subjected to hematoxylin and eosin staining and immunohistochemistry using anti-CUGBP1 antibody [3B1] (ab9549; Abcam, Cambridge, MA, USA) (1:500). Images were obtained using Axio Zoom.V16 microscope (stereo).
Statistical analysis
Laboratory data are presented as mean ± standard deviation about the mean (SD) unless otherwise stated. When two groups were compared, the Student’s t-test (two-sided) was used unless otherwise indicated, and a P-value < 0.05 was considered significant.
Results
CELF1 interacts directly with eIF4E and PABPC1 at the m7G cap, independent of intact eIF4G1
Our previous work suggested that CELF1’s control of the translation of EMT effector mRNAs containing GU-rich elements might be at the level of 5′ m^7^G cap-dependent translation initiation [14]. We first confirmed and extended these findings by examining additional CELF1 targets. We fused the 3′ UTRs of a subset of GRE-containing mRNAs (JUNB, CRLF1, SNAI1, and SSBP2) downstream of the Renilla luciferase coding sequence in the pRL-TK-CXCR4-6x [28] reporter plasmid (Fig. 1a). Transfection of the GRE-containing 3′ UTR reporters into MCF-10A cells, either untreated or treated with TGF-β to induce EMT, revealed a significant increase in reporter activity specific to treated cells (Fig. 1a), independent of any corresponding differences in relative mRNA expression. Indeed, we consistently noticed a 50% reduction of reporter mRNA in these experiments (Fig. 1b). To test whether this phenomenon was conserved in the absence of cap-dependent translation initiation, we built a battery of bicistronic constructs in which the same thymidine kinase promoter drove expression of the firefly luciferase coding sequence, followed by the encephalomyocarditis virus (EMCV) internal ribosome entry site (IRES), the Renilla luciferase open reading frame, and individual GRE-containing 3′ UTRs (pFR-EMCV) (Fig. 1c). In contrast to the results obtained with monocistronic Renilla luciferase reporters, no differences in relative reporter expression driven by the IRES were observed in a comparison of the epithelial and mesenchymal states (Fig. 1c and d). Interestingly, the previously observed modest reduction of reporter mRNA levels was blunted in these experiments, raising the possibility that translation and modest reductions of mRNA levels might somehow be related. Nonetheless, these results confirmed and extended our previous observation that translational control of mRNAs harboring GREs within their 3′ UTRs during EMT is mediated by 5′ end-dependent translational initiation [14].
*CELF interacts with eIF4E at the m7G cap, independent of intact eIF4G1. (a) Reporter assay quantifying the relative Renilla luciferase expression from the indicated 3′ UTR luciferase reporters in untreated and TGF-β-treated MCF-10A cells. Data were normalized to Firefly luciferase expression and are presented as fold change of this normalized signal relative to CXCR4 reporter in untreated MCF-10A cells. (b) Quantitative reverse transcription polymerase chain reaction (qRT-PCR) of indicated Renilla luciferase reporters (pRL-TK) in untreated and TGF-β-treated MCF-10A cells. Data were normalized to endogenous ACTB expression. (c) As in (a), with the indicated 3′ UTR luciferase reporters driven from an EMCV internal ribosomal entry site (IRES) in untreated and TGF-β-treated MCF-10A cells. (d) As in (b), for reporter assays in panel (c). (e) Right six lanes—immunoblots of indicated immunoprecipitates from whole-cell lysates derived from MCF-10A cells treated with TGF-β for 72 h. One half of each total immunoprecipitate was digested with RNase A prior to immunoblotting with the indicated antibodies. Right six lanes—as in the left six lanes, but lysates were digested with coxsackievirus 2A protease to cleave eIF4G1 before immunoprecipitation. CT = C-terminal; FL = full length; NT = N-terminal. (f) m7GTP cap analog binding assays utilizing cytosolic extracts derived from MCF-10A cells treated with TGF-β for 72 h. As above, one half of each extract was digested with coxsackievirus 2A protease to cleave eIF4G1 before the assay. (g) Proximity ligation assays using the indicated pairs of antibodies on MCF-10A cells treated with TGF-β for 72 h. In all panels, results are representative of at least three independent experiments and error bars depict mean ± standard deviation (SD) of aggregate replicates performed in triplicate. NS: not significant; P-value < 0.05 (Student’s t-test).
We next assessed whether CELF1 physically interacts with established components of the canonical translation initiation complex. EMT was induced in MCF-10A cells via either TGF-β treatment or GFP-CELF1 overexpression and monitored by loss of expression of the epithelial cell marker E-cadherin (CDH1) concomitant with induction of expression of the mesenchymal cell markers Fibronectin (FN1) and Vimentin (VIM) (Supplementary Figure S1a). As previously reported, we observed an increased level of phosphorylation of eIF4E at serine 209, which has been previously shown as required during TGF-β-induced EMT [19]. However, no consistent change in steady-state protein expression of any of the tested translation initiation factors was observed between the epithelial and mesenchymal states (Supplementary Fig. S1a). Co-immunoprecipitation experiments using extracts from TGF-β-induced cells revealed an association among CELF1, eIF4E, eIF4G1, eIF4B, eIF3H, eIF3C, and poly(A)-binding protein (PABPC1) (Fig. 1e and Supplementary Fig. S1b). CELF1 did not physically associate with eIF4A, eIF4G2, or eIF4G3 in any of the conditions assayed (Fig. 1e and Supplementary Fig. S1b). We next asked whether these interactions might be RNA-dependent. CELF1 retained its physical association with eIF4E, eIF4G1, and PABPC1 following RNase A digestion, whereas this digestion eliminated the association of CELF1 with eIF4B, eIF3H, and eIF3C, consistent with the notion that these latter interactions are RNA-dependent (Fig. 1e). Indeed, the physical association with PABPC1 and EIF4E was maintained following double digestion with both RNase A and RNase 1, reducing the possibility that the observed interaction was due to incomplete digestion of polyadenylate sequences (Supplementary Fig. S1c and d).
Given the above-established physical association of eIF4E, eIF4G1, and PABPC1, our results thus far were consistent with a model in which CELF1 directly binds one or more of these three proteins. To further investigate this notion and begin to attempt to differentiate among several inherent possibilities, we again immunoprecipitated cellular extracts derived from TGF-β-treated MCF-10A cells that were either untreated or digested with purified coxsackievirus 2A protease. Coxsackievirus 2A protease impairs m^7^G cap-dependent translation during the coxsackievirus replication cycle by cleaving the N-terminal fragment of eIF4G1, which binds eIF4E and PABPC1, from eIF4G1’s C-terminal fragment, which binds eIF4A and eIF3 [33–35]. Following digestion with 2A protease, CELF1’s interaction with the C-terminal, but not N-terminal, fragment of eIF4G1 was maintained. Surprisingly, the association between CELF1 and both eIF4E and PABPC1 was preserved within this context (Fig. 1e), demonstrating that this tripartite interaction was independent of eIF4G1’s N-terminus. eIF3C, eIF3H, and eIF4B retained their association with CELF1 upon 2A protease digestion, unless also digested with RNase A (Fig. 1e), indicating that these factors are a stable part of a remaining RNA-dependent complex that is not dependent on intact eIF4G1. We concluded that in the context of a whole cellular lysate, CELF1 likely binds eIF4E and PABPC1 directly, and that this complex may be tethered to eIF4B and eIF3 in an RNA-dependent fashion. The results also suggested that CELF1 continued to interact with eIF4G1’s C-terminus in a manner independent of intact RNA.
Focusing on the notion that the abovementioned experiments represented interactions in the context of a whole cellular lysate, we next sought to determine whether these interactions exist specifically at the mRNA m^7^G cap. To do this, we employed direct capture of eIF4E and its binding partners from extracts from TGF-β-induced cells via cap-analog affinity resin, precipitation, and immunoblot analysis. Following binding to cap analog, competitive elution with m^7^GTP, but not GTP, revealed co-immunoprecipitation between eIF4E and CELF1, eIF4G1, eIF4A, eIF4B, eIF3H, eIF3C, and PABPC1, suggesting that these associations were present at the m^7^G cap (Fig. 1f). We next asked whether these interactions might be preserved in the absence of intact eIF4G1, within the context of a preliminary digestion of these extracts with 2A protease. Under these conditions, the N-terminal but not C-terminal fragment of eIF4G1 co-eluted with eIF4E, which is consistent with several previous reports [34, 35] (Fig. 1f). Digestion with 2A protease also disrupted co-elution of eIF4A, but did not appreciably inhibit co-elution of eIF4B, eIF3C, eIF3H, and PABPC1, or CELF1 (Fig. 1f). Crucially, however, while the N-terminal fragment of eIF4G1 was observed in the context of cap-analogue binding, the interaction between eIF4G1’s C-terminus and CELF1 that exists in whole cell lysate (Fig. 1e) was not recapitulated in these assays (Fig. 1f). Since eIF3 binds a sequence within the C-terminal fragment of eIF4G1, the presence of eIF3C and eIF3H at the cap in the 2A-digested reactions raised the possibility that the interaction of CELF1, eIF4E, PABPC1, and eIF3 at the m^7^G cap of mRNA might be independent of a direct association between CELF1 and eIF4G1, and we explored this possibility further.
We next employed proximity ligation assays (PLAs) to assess interactions among CELF1, eIF4E, PABPC1, and eIF4G1 in intact cells. While the combination of the antibody recognizing CELF1 with antibodies recognizing eIF4E or PABPC1 resulted in robust PLA signal, the signal upon combination of the CELF1 antibody and an antibody recognizing eIF4G1 could not be distinguished from background signal associated with the latter antibody (Fig. 1g and Supplementary Fig. S2). As expected, robust signal was observed when the antibody recognizing eIF4G1 was combined with antibodies recognizing eIF4E or PABPC1. We thus concluded that CELF1 interacts with eIF4E and PABPC1, but not eIF4G1, in intact, TGF-β-treated MCF-10A cells, and turned our attention to how manipulation of CELF1 and eIF4G1 might impact translation of CELF1’s regulatory targets.
CELF1 reduces the necessity for eIF4G1 in translation of GRE-containing EMT effector mRNAs
To assess the interactions of CELF1 and eIF4G1 with CELF1’s regulatory targets, we performed UV crosslinking/immunoprecipitation/qRT-PCR assays from extracts derived from TGF-β-treated MCF-10A cells (Fig. 2a). As expected, immunoprecipitation with an anti-eIF4E antibody resulted in significant enrichment for each of the mRNAs assayed. In contrast, immunoprecipitation with an anti-CELF1 antibody from cellular extracts derived from these cells resulted in significant enrichment of GRE-containing (EGR3, FOSB, JUNB, and SNAI1) but not control (GAPDH and ACTB) mRNAs. Furthermore, immunoprecipitation with anti-eIF4G1 revealed an essentially opposite pattern of enrichment—while each of the control mRNAs was enriched in this case, the immunoprecipitation only significantly enriched one GRE-containing mRNA (SNAI1), and this enrichment was comparatively modest. To further refine these observations, we repeated these experiments, employing tandem co-immunoprecipitation. Anti-eIF4E immunoprecipitates were themselves immunoprecipitated a second time, using either anti-CELF1 or anti-eIF4G1 antibodies (Fig. 2b). Fully consistent with our single-IP crosslinking assays, a secondary immunoprecipitate with anti-CELF1 enriched our GRE-containing (EGR3, FOSB, JUNB, and SNAI1) but not control (GAPDH and ACTB) mRNAs, while a secondary immunoprecipitate with anti-eIF4G1 displayed the opposite pattern, enriching our control but not GRE-containing transcripts. Importantly, deletion of the GRE within the 3′ UTR of reporters corresponding to CELF1’s regulatory targets abolished CELF1-dependent enrichments in tandem immunoprecipitations, with these reporters instead co-immunoprecipitating with eIF4G1 (Supplementary Fig. S3a). While these data must be interpreted with some degree of caution given poor efficiency and sequence bias associated with UV254 crosslinking, they are fundamentally consistent with a model in which within mesenchymal breast cancer cells, CELF1 specifically associates with these GRE-containing transcripts but not transcripts lacking a GRE; whereas eIF4G1 specifically associates with transcripts lacking a GRE but not those containing one. This poses a conundrum given the fundamental notion that eIF4G1 is required for mRNA translation initiation.
*CELF1 stimulates translation of GRE-containing EMT effector mRNAs in the context of reduced eIF4G1 function. (a) RNA crosslinking-immunoprecipitation/qRT-PCR of GRE-containing mRNAs (EGR3, FOSB, JUNB, SNAI1) from TGF-β-treated MCF-10A cells using anti-CELF1, anti-eIF4E, and anti-eIF4G1 antibodies or mouse and rabbit IgG. ACTB and GAPDH are non-GRE-containing negative control mRNAs. (b) RNA crosslinking-immunoprecipitation/qRT-PCR of GRE-containing mRNAs (EGR3, FOSB, JUNB, SNAI1) from TGF-β-treated MCF-10A cells using tandem anti-eIF4E/anti-CELF1 immunoprecipitation, tandem anti-eIF4E/anti-eIF4G1 immunoprecipitation, or tandem immunoprecipitation with mouse and rabbit IgGs. ACTB and GAPDH are non-GRE-containing negative controls. (c, d) Efficiency of in vitro translation of indicated capped and polyadenylated Renilla luciferase reporter mRNAs in mock or 2A protease-digested cell-free extract. (e, f) Efficiency of in vitro translation of reporter mRNAs as described in panels (c) and (d), but with mock-depleted (Beads) cell-free extract, eIF4G1-immunodepleted (ID) cell-free extract, or eIF4G1-immunodepleted cell-free extract reconstituted by addition of 20 nM enriched eIF4G1 and/or an equivalent concentration of recombinant CELF1. In panels (c–f), all extracts were derived from TGF-β-treated MCF-10A cells transiently transfected with shRNAs targeting either GLB1 (c, e) or CELF1 (d, f). CXCR4 = control, WT = wild-type 3′ UTR, ΔGRE =3′ UTR with deletion of GRE. In all panels, results are representative of at least three independent experiments. Error bars depict mean ± standard deviation (SD) of aggregate replicates performed in triplicate. NS: not significant; P-value < 0.05 (a, b, e, f: ANOVA with Dunnet’s post-hoc test; c, d: Student’s t-test).
We thus examined the translation of CELF1 regulatory targets more closely, performing in vitro luciferase assays with translationally competent extracts derived from TGF-β-treated MCF-10A cells transiently transfected with shRNAs targeting β-galactosidase (GLB1; negative control) or CELF1. Importantly, translation of control constructs in our extracts was characterized by synergy in the presence of both an m^7^G cap and poly(A) tail, indicating that the extracts effectively recapitulated basal mRNA translation (Supplementary Fig. S3b). As expected, robust translation of a capped and polyadenylated control Renilla luciferase reporter mRNA (CXCR4) within these extracts was significantly attenuated following cleavage of eIF4G1 by 2A protease (*P *< 0.01; Fig. 2c and d). In contrast, while cleavage of eIF4G1 had no significant effect on the translation of reporter mRNAs corresponding to established CELF1 target 3′ UTRs (CRLF1 and SNAI1), deletion of the GRE rendered the reporters sensitive to 2A cleavage (*P *< 0.01; Fig. 2c). Interestingly, performing this assay in extracts from cells in which CELF1 had been knocked down by shRNAs revealed a decrease in translation of the CRLF1 and SNAI1 reporters irrespective of 2A protease digestion (again, contingent upon the presence of the GRE within these reporters) that was restored by addition of recombinant CELF1 protein (*P *< 0.01; Fig. 2d). This strongly suggests that the GRE within these reporters is repressive unless bound by CELF1, and that binding of CELF1 to these elements promotes translation of GRE-containing mRNAs with a reduced requirement for intact eIF4G1. These results do not, however, rule out that the N- and C-terminal cleavage products of eIF4G1 may retain some function in promoting the translation of the GRE-containing reporters.
To directly address this caveat, we repeated the in vitro translation assay using extracts from which eIF4G1 had been immunodepleted. As previously observed in the context of 2A protease digestion, immunodepletion of eIF4G1 in extracts derived from MCF-10A cells treated with a control shRNA (Supplementary Fig. S3c–e) ablated the translation of a control reporter, whereas reporters fused to GRE-containing CELF1 target UTRs were unaffected (*P *> 0.01, Fig. 2e and Supplementary Fig. S3f). Like the control reporter, translation of mutant CELF1 target UTRs lacking a GRE was significantly attenuated upon eIF4G1 depletion (P < 0.01; Fig. 2e and f, Supplementary Fig. S3f and g). As expected, addition of an excess of recombinant eIF4G1 to the depleted extracts (Supplementary Fig. S3d) restored translation of the control and ΔGRE reporters to levels matching what was observed in the mock-depleted extracts (Fig. 2e and f, Supplementary Fig. S3f and g). Again, CELF1 knockdown diminished translation of the WT SNAI1 and CRLF1 reporters (*P *< 0.01 in each case), even in extracts retaining eIF4G1 function, and this decrease in translation was rescued by addition of recombinant CELF1 (Fig. 2f; Supplementary Fig. S3e and g). We concluded that in TGF-β-treated MCF-10A cells, the GRE element within CELF1 target mRNAs inhibits the translation of these mRNAs unless this element is bound by CELF1, which in turn robustly reduces the dependency of translation of the GRE-containing mRNAs on eIF4G1.
Phosphorylation of eIF4E is required for CELF1-driven EMT in MCF-10A cells
It has been previously established that phosphorylation of eIF4E at serine 209 is required for TGF-β-induced EMT in MCF-10A cells [19]. Since CELF1 both promotes the translation of EMT effector mRNAs at the level of translational initiation (Fig. 1a–d) and interacts with eIF4E at the m^7^G cap of GRE-containing mRNAs (Fig. 1f), we asked whether disruption of eIF4E phosphorylation at serine 209 would be sufficient to block CELF1-driven EMT in MCF-10A cells. We employed a well-established system in which MCF-10A cells are stably transduced with either WT or S209A phospho-null mutant murine Eif4e, concomitant with shRNA-mediated knockdown of endogenous human EIF4E [19]. While cells expressing a WT Eif4e transgene underwent EMT normally in response to CELF1 overexpression, CELF1-driven EMT was blocked by expression of the S209A mutant Eif4e (Fig. 3a). Reciprocal co-immunoprecipitation experiments again demonstrated an association between CELF1 and eIF4E, but that association with the S209A phosphomutant was essentially ablated, suggesting that the interaction was specific to eIF4E phosphorylated in this position (Fig. 3b). Polysomal fractionation and enrichment analysis (Fig. 3c) again revealed that CELF1’s GRE-containing mRNA targets [14] were enriched within MCF10A cells expressing murine Eif4e and treated with TGF-β (Fig. 3d), indicating conservation of function between the murine and human proteins. In contrast, these mRNAs were robustly depleted within polysomal fractions derived from TGF-β-treated cells expressing the Eif4e S209A mutant (Fig. 3d). Consistent with these results, immunoblot analysis of SSBP2, SNAI1, and FOSB expression revealed a robust increase in relative expression upon rescue of EIF4E knockdown by murine Eif4e, but not upon rescue of EIF4E knockdown with the murine Eif4e S209A mutant (Fig. 3e). These results are consistent with the notion that eIF4E phosphorylation is required for CELF1’s functions as both a promoter of mRNA translation and a driver of EMT.
*Phosphorylation of eIF4E is required for CELF1-driven EMT in MCF-10A cells. (a) Immunoblots of lysates derived from MCF-10A cells stably expressing either HA-tagged WT or S209A mutant murine EIF4e and shRNA targeting human EIF4E or control shRNA, and either mock transfected or transiently transfected with a CELF1 overexpression construct for 72 h. GAPDH = loading control. (b) Immunoblot of indicated immunoprecipitates from lysates derived from TGF-β-treated MCF-10A cells, stably expressing either HA-tagged WT or S209A mutant murine Eif4e and shRNA targeting human EIF4E or control shRNA. IgG: negative immunoprecipitation control. (c) Polysomal profiles from MCF-10A cells in which endogenous EIF4E expression had been knocked down via shRNA and then rescued via stable transduction of either WT or S209A mutant Eif4e. (d) qRT-PCR validation of polyribosomal enrichment and depletion of indicated mRNAs via total and polysomal mRNA from MCF-10A cells stably expressing WT or S209A mutant Eif4e, treated with TGF-beta for 72 h. (e) MCF-10A cells in which endogenous EIF4E expression had been knocked down via shRNA and then rescued via stable transduction of either WT or an S209A mutant Eif4e were transiently transfected with CELF1 expression construct. After 72 h, extracts were assessed via immunoblot for relative protein expression of CELF1-regulated EMT effectors. In all panels, results are representative of at least three independent experiments. Error bars in panel (d) depict mean ± standard deviation (SD). NS: not significant; P-value < 0.05 (Student’s t-test).
CELF1 directly binds the eIF4E S209D phosphomimetic via a conserved motif
We next sought to further explore direct interactions CELF1 might make within the translation pre-initiation complex. The YXXXXLΦ motif (where X is any amino acid and Φ is a hydrophobic amino acid residue) is common among eIF4G proteins and eIF4E binding proteins (4E-BPs), which utilize the motif to compete for binding to the dorsal surface of eIF4E [36]. We identified three putative eIF4E binding motifs in the CELF1 peptide sequence (Fig. 4a). We transiently transfected MCF-10A cells with GFP-tagged WT CELF1 or with individual mutants of CELF1 in which these putative eIF4E-binding motifs were deleted, either individually or in combination. Immunoprecipitation of WT CELF1 or these CELF1 mutants revealed that deletion of CELF1 amino acids 365–371 (YAAAALP) abrogated precipitation of eIF4E (Fig. 4b). Strikingly, although the YAAALP peptide sequence resides within an intrinsically disordered region, comparison of peptide sequences among human CELF1, murine Celf1, zebrafish Celf1, chicken CELF1, Xenopus celf1 (formerly Eden-bp), and human CELF2 revealed that the sequence was perfectly conserved within these homologs, potentially indicating a broadly conserved function (Supplementary Fig. S4a).
CELF1 directly binds eIF4E via interactions via the canonical dorsal cleft region and the lateral hydrophobic patch. (a) Schematic of CELF1 domain structure and candidate eIF4E binding motifs. RRM, RNA-recognition motif. (b) Immunoblots of immunoprecipitations from lysates of MCF-10A cells transfected with WT or indicated mutant GFP-CELF1 plasmids for 72 h. (c) Immunoblots of binding assays using affinity-purified phosphomimic eIF4E (GST-EIF4ES209D), affinity-purified WT CELF1 (6xHis-CELF1), or affinity-purified mutant CELF1 (6x-His-CELF1Δ365–71). (d) Immunoblots of immunoprecipitations derived from lysates of MCF-10A cells stably expressing an shRNA targeting the 3′ UTR of EIF4E and co-expressing either WT or EIF4EW73A mutant, treated with TGF-β or transiently transfected with a GFP-CELF1 plasmid for 72 h. IgG: negative control. (e) Immunoblots of binding assays in which affinity-purified 6xHis-CELF1 was mixed with affinity-purified phosphomimic (GST-EIF4ES209D) or mutant (GST-EIF4ES209D/W73A) eIF4E. (f) HSQC-NMR spectra of 15N-labeled eIF4ES209D, alone or mixed with a seven-fold excess of CELF1 YAAAALP-containing peptide (sequence shown). Representative shifts are magnified. (g) Dorsal surface view of the crystal structure of eIF4E complexed with m7GTP (PDB 1IPC - [37]), depicting chemical shifts observed in HSQC-NMR. The canonical eIF4E binding cleft is colored in red, and chemical shifts induced by the CELF1 peptide are indicated in blue. Shifts overlapping the canonical binding cleft are indicated in purple. (h) As in (g), rotating the eIF4E structure ninety degrees along a roughly fifteen-degree bearing for depiction of the lateral surface mediating non-canonical binding. Coloring and annotations are as in (g). (i) Purified, untagged CELF1 was mixed with purified eIF4E, eIF4ES209D, eIF4ES209D/W73A (disrupts canonical dorsal binding), or eIF4ES209D/I63A/I79A (disrupts non-canonical lateral binding) and then immunoprecipitated with IgG (negative control) or anti-CELF1 antibody and immunoblotted with the indicated antibodies. Results in (a–e, i) are representative of at least three individual experiments.
To determine whether the molecular interaction between CELF1 and eIF4E was direct, we generated affinity-tagged purified bacterial proteins for pulldown assays: a GST-tagged phosphomimetic eIF4E (eIF4E^S209D^), a 6xHis-tagged WT CELF1, and a 6xHis-tagged CELF1 mutant in which amino acids 365–371 were deleted (CELF1^Δ365–71^, Supplementary Fig. S4b). Consistent with our data from the previous experiment, WT CELF1, but not CELF1^Δ365–71^, co-precipitated with GST-tagged eIF4E^S209D^, indicating a direct interaction between CELF1 and eIF4E mediated by the YAAAALP sequence (Fig. 4c). Importantly, CELF1’s interaction with eIF4E was independent of both CELF1’s RNA binding activity and previously described post-translational modifications (PTMs) even though mutations abrogating these features within CELF1 [14, 38] disrupted CELF1 overexpression-driven EMT (Supplementary Fig. S4c–f). Importantly, these results indicate that the interaction between CELF1 and eIF4E is predicated neither by CELF1 phosphorylation nor by CELF1’s RNA binding activity—the latter finding is consistent with the data presented in Fig. 1b and Supplementary Fig. S1b and c. However, the findings imply that the CELF1/eIF4E interaction is not sufficient to drive EMT in and of itself, and that this process is dependent upon CELF1’s interaction with its RNA targets.
Given the necessity of the 4E-BP-like consensus sequence for CELF1’s binding to eIF4E, we next asked whether this interaction could be abrogated by substitution of tryptophan 73 with alanine (W73)—a mutation used to disrupt binding by eIF4G1 and other 4E-BPs [39]. We transduced MCF-10A cells with either WT or W73A mutant EIF4E (EIF4E^W73A^), followed by knockdown of the endogenous EIF4E using an shRNA targeting the 3′ UTR. As assessed by the surface sensing of translation using puromycin labeling (SUnSET) assay [27], stable overexpression of the W73A mutant EIF4E did not appear to markedly inhibit global translation in this context (Supplementary Fig. S4g), similar to previous descriptions [40–42]. Immunoprecipitation of eIF4E and eIF4E^W73A^ from MCF-10A cell extracts derived from cells either treated with TGF-β or overexpressing a GFP-CELF1 fusion protein revealed that CELF1’s interaction with eIF4E was dependent upon the W73 residue (Fig. 4d). These results were recapitulated in vitro using affinity-tagged purified proteins, indicating that the W73A mutation ablated the direct interaction between CELF1 and eIF4E in this latter context (Fig. 4e). Taken together, these data indicate that CELF1 directly interacts with eIF4E via the former’s YAAAALP motif, and that this interaction is disrupted by substitution of tryptophan 73 to alanine within the eIF4E protein—this latter observation strongly suggesting that CELF1 protein binds eIF4E at the same site as both eIF4G1 and the uniformly inhibitory 4E-BPs (Fig. 4d).
To directly assess the interaction between eIF4E^S209D^ and a synthetic CELF1 peptide, we employed ^1^H-^15^N heteronuclear single quantum coherence nuclear magnetic resonance spectroscopy (HSQC-NMR), which is sensitive to changes in the chemical environment of backbone amide groups and ideal for detecting protein-peptide interactions at atomic resolution. Uniformly ^15^N-labeled eIF4E^S209D^ (Supplementary Fig. S5a) was pre-bound with a saturating concentration (150 µM) of 7-methylguanosine and then titrated with increasing concentrations of a 62-mer peptide derived from CELF1 amino acids 354–416 (Fig. 4f), designed to correspond closely to an eIF4G1 peptide used in previous structural studies [43]. We observed over a dozen significant chemical shift perturbations (CSPs), correlating with increasing peptide concentrations, flanking and within both the previously defined dorsal (canonical binding) and lateral (non-canonical binding) surfaces on the labeled eIF4E^S209D^ protein. Interestingly, while these CSPs traced a clear path from the dorsal to lateral faces of eIF4E^S209D^ (Fig. 4g and h), CSPs on the dorsal face were distinct from contacts previously defined for eIF4G1 binding (Fig. 4g). In contrast, CSPs on the lateral face were observed at tryptophan 46 and isoleucine 79, consistent with previously defined contacts in this hydrophobic region [43], but these perturbations extended to the flanking hydrophobic residues leucine 62 and phenylalanine 47 (Fig. 4h).
To functionally assess the lateral contacts, we again turned to co-immunoprecipitation experiments, this time employing untagged bacterially expressed CELF1, eIF4E^S209D^, eIF4E^S209D/W73A^, and eIF4E^S209D/I63A/I79A^ proteins (Supplementary Fig. S5b and c), with mutations in the latter mutant mirroring previous mutations utilized in Drosophila eIF4E to disrupt lateral binding [43, 44]. As predicted, both the tryptophan and tandem isoleucine mutations disrupted eIF4E’s association with CELF1 (Fig. 4i). These data strongly suggest that CELF1 and eIF4G1 bind to eIF4E in a mutually exclusive fashion.
CELF1 displaces eIF4G1 from eIF4ES209D while directly binding PABPC1
Given that canonical 4E-BPs block translation by competing with eIF4G1 for binding to eIF4E, it was counterintuitive that CELF1 would use a similar mechanism to promote translation. To build further evidence for a competitive mode of binding, we investigated associations among the binding of eIF4E, CELF1, and eIF4G1, again using affinity-purified proteins. Consistent with established literature and our own data described above, in vitro binding assays confirmed that eIF4G1 binds WT eIF4E, but neither the eIF4E W73A mutant (Fig. 5a) nor CELF1 (Fig. 5b). Co-incubation of both eIF4G1 and CELF1 with eIF4E^S209D^ revealed that CELF1 effectively and stoichiometrically competed with eIF4G1 for the latter protein (Fig. 5c). This competition was dependent upon CELF1’s direct interaction with eIF4E^S209D^, given that coincubation of CELF1 protein with deletion of the conserved YAAAALP motif was unable to displace eIF4G1 in the same experiment. Conversely, eIF4G1 was unable to displace WT CELF1 protein bound to eIF4E^S209D^, even at ten-fold molar excess (Fig. 5d). These results strongly suggest that while both CELF1 and eIF4G1 are able to bind to phosphorylated eIF4E, CELF1 displaces eIF4G1 in this context.
CELF1 displaces eIF4G1 from eIF4E and directly binds PABPC1. (a) Purified GST-tagged eIF4E or eIF4ES209D was mixed with purified DDK-eIF4G1. Complexes were pulled down with purified glutathione beads and immunoblotted with indicated antibodies. (b) Purified DDK-tagged eIF4G1 was mixed with purified GST-tagged eIF4ES209D (positive control) or purified 6x-His-tagged CELF1 and precipitated on glutathione or Ni2+ resin and immunoblotted with the indicated antibodies. (c) Immunoblots of binding assays in which purified GST-eIF4ES209D was mixed with purified DDK-eIF4G1 in combination with increasing amounts of either purified WT (6xHis-CELF1) or mutant (6x-His-CELF1Δ365–71) CELF1. Complexes were captured on glutathione resin. (d) Immunoblots of binding assays in which purified GST-eIF4ES209D was mixed with purified wild type (6xHis-CELF1) and increasing amounts of purified DDK-eIF4G1. Complexes were captured on Ni2+ resin. (e) Immunoblots of competitive binding assays in which purified DDK-tagged PABPC1 was incubated with the indicated ratios of DDK-tagged eIF4G1 and 6xHis-tagged CELF1. Complexes were captured on Ni2+ resin. (f) Purified, untagged CELF1 was mixed with purified, untagged eIF4ES209D, purified, untagged PABPC1, or both. Complexes were immunoprecipitated with IgG (negative control) or anti-CELF1 antibody and blotted with the indicated antibodies. (g) Immunoblots of binding assays in which the indicated concentrations of purified DDK-tagged eIF4A were mixed with increasing concentrations of purified DDK-eIF4G1 or purified 6xHis-tagged CELF1. Complexes were captured on Ni2+ resin or via immunoprecipitation with anti-eIF4G1 antibody, as indicated. All results are representative of at least three independent experiments.
We next asked whether CELF1 was able to bind PABPC1 and thus, in theory, to circularize mRNA during the formation of a translation pre-initiation complex, again using affinity-purified proteins. Consistent with our observations within cellular extracts and proximity ligation assays, in vitro binding experiments confirmed that CELF1 binds recombinant purified PABPC1 directly (Fig. 5e and Supplementary Fig. S5d). As expected, PABPC1 also directly bound eIF4G1. Interestingly, however, competitive binding assays revealed that eIF4G1 did not displace CELF1 from PABPC1 but was instead able to bind the latter two complexed proteins (Fig. 5e), suggesting eIF4G1 and CELF1 bind to PABPC1 on distinct sites. Nonetheless, direct binding between CELF1 and PABPC1 suggests that CELF1 may have the ability to circularize mRNA. Additional binding and immunoprecipitation assays revealed that CELF1 directly binds eIF4E^S209D^ and PABPC1 simultaneously (Fig. 5f), consistent with CELF1 bridging the latter two molecules.
Consistent with our immunoprecipitation and affinity binding experiments from cellular extracts (Fig. 1e and f), purified recombinant CELF1 did not directly bind purified recombinant eIF4A (Fig. 5g). While a full mechanistic and functional demonstration of eIF4A-independence in this context is beyond the scope of the current study, our results suggest translation of CELF1’s targets in mesenchymal MCF-10A cells may be independent of this helicase. Whether or not this is the case, our results are consistent with a model in which CELF1 displaces eIF4G1 from eIF4E^S209D^ and binds PABC1 to promote translation of its targets.
Interaction of CELF1 and eIF4E is required for CELF1-driven EMT in vitro and experimental metastasis in vivo
Given that we have previously demonstrated that overexpression of CELF1 is sufficient to induce EMT in several breast epithelial cell lines [14], we next set out to determine whether disruption of CELF1’s interaction with eIF4E would abrogate this induction. Individual overexpression of our battery of CELF1 YXXXXLΦ mutants in MCF-10A cells revealed that the CELF1^Δ365–71^ mutant specifically failed to induce EMT despite comparable levels of expression among the mutant and WT CELF1 proteins (Fig. 6a). Conversely, overexpression of the eIF4E^W73A^ mutant in MCF-10A cells disrupted EMT, whether induced by TGF-β or CELF1 overexpression (Fig. 6b). We next examined two additional breast epithelial cell lines—the MCF-10AT1 line (a non-malignant isogenic derivative of the parental non-transformed MCF-10A line) and the MDA-MB-468 line. As we have previously demonstrated [14], overexpression of WT GFP-CELF1 induced loss of epithelial cell marker CDH1 and an increase in the mesenchymal cell markers FN1 and VIM in both lines (Fig. 6c). However, these changes were not observed upon overexpression of the CELF1^Δ365–71^ mutant, consistent with the notion that the role of the CELF1/eIF4E interaction in EMT is common among breast epithelial cell lines. Consistent with these results, while overexpression of WT CELF1 drove increases in both cellular migration and invasion in the MCF-10AT1 and MDA-MB-468 lines, overexpression of the CELF1^Δ365–71^ failed to drive these processes (Fig. 6d–g).
*Interaction of CELF1 and eIF4E is required for CELF1-driven EMT and experimental metastasis. (a) Immunoblot analysis of indicated EMT markers in lysates derived from MCF-10A cells transfected with WT or indicated mutant GFP-CELF1 plasmids for 72 h. (b) Immunoblot analysis of indicated EMT markers in lysates derived from MCF-10A cells expressing either WT or W73A mutant human EIF4E and shRNA targeting the 3′ UTR of human EIF4E and induced to undergo EMT via stable expression of GFP-CELF1 or TGF-β treatment for 72 h. (c) Immunoblot analysis of indicated EMT markers and GFP-CELF1 in lysates derived from parental MCF-10AT1 cells (left column) and MDA-MB-468 (right column), or each cell line stably transduced with either WT or Δ365–71 mutant GFP-CELF1. GAPDH = loading control in panels (a), (b), and (c); black line in panels (a) and (b) denotes lysates derived from the same experiment, but gels processed in parallel. All results (a–c) are representative of at least three independent experiments. Quantification of relative in vitro cellular migration (d, f) and invasion (e, g) in transwell assays in parental MCF-10AT1 and MDA-MB-468 cells, respectively, or stably transduced with either WT or Δ365–371 mutant GFP-CELF1. Data represents mean ± SD of at least three independent experiments, each performed in triplicate. P-value < 0.05 (ANOVA with Dunnet’s post-hoc test). (h, i) Parental MCF-10AT1 cells, or cells stably overexpressing either WT or Δ365–71 mutant GFP-CELF1, were injected into the tail vein of athymic nude mice. The incidence and progression of metastasis were measured by luciferin injection and bioluminescence imaging of Firefly luciferase (h), and ex vivo excised lungs on day 15 (i). (j) Representative hematoxylin and eosin (H&E) (top) and immunohistochemical (IHC) (bottom) staining, respectively, of the lungs from mice shown in panel (h). Scale bar, 200 µm (top); 50 µm (bottom). Black arrows (bottom) indicate micrometastases. Dotted lines indicate area shown in corresponding H&E staining of serial sections shown in panel (j). For (h–j), representative images are from n = 4 for parental, n = 4 for WT GFP-CELF1, and n = 6 for mutant GFP-CELF1Δ365–71 experimental groups.
To determine whether the importance of the CELF1/eIF4E interaction was limited to EMT in vitro, we next assessed the importance of this interaction in a model of experimental metastasis. Overexpression of CELF1 induces experimental lung metastasis from the normally tumorigenic but non-metastatic MCF-10AT1 cell line [14, 45]. Consistent with our previous work, ectopic expression of WT CELF1 within MCF-10AT1 cells injected into the tail vein of nude mice conferred these cells with the ability to rapidly induce lung colonization. In contrast, when mice were injected with MCF-10AT1 cells overexpressing the CELF1^Δ365–71^ mutant, lung colonization was severely attenuated (Fig. 6h–j). In aggregate, our results strongly support a model in which, within the context of breast epithelial cells undergoing EMT and metastatic dissemination, CELF1 is recruited to GRE-containing EMT effector mRNAs, acting in cis on these mRNAs to promote non-canonical translation of these mRNAs via a direct interaction with eIF4E and PABPC1 (Fig. 7).
A working model for CELF1-dependent cap-dependent translational initiation during TGF-β-mediated EMT of breast epithelial cells. During EMT of breast epithelial cells, CELF1 is stabilized and recruited to GREs within the 3′ UTRs of EMT effector mRNAs. In cis on the RNA, CELF1 then bridges eIF4E and PABPC1 to promote translation of these mRNAs. Additional mechanisms conferring specificity on this process, as well as the mechanism by which CELF1 recruits the full 43S pre-initiation complex, remain to be defined. Components are not drawn to scale.
Discussion
In this work, we demonstrate that CELF1 promotes translation of its GRE-containing mRNA targets during EMT via a cap-dependent mechanism. We go on to show that CELF1 associates with both eIF4E and PABPC1 in both extracts and intact cells, and that the translation of CELF1’s targets is both eIF4G1-independent and dependent upon the GRE within the 3′ UTR of these targets. We show that CELF1’s interaction with eIF4E is dependent upon phosphorylation of the latter protein and identify both the conserved peptide sequence within CELF1 mediating the interaction and contacts upon the eIF4E protein. We demonstrate that CELF1 effectively competes with eIF4G1 from eIF4E and can also directly bind PABPC1. Finally, we show that CELF1’s interaction with eIF4E is required for both EMT in vitro and experimental metastasis in vivo. Thus, our results identify a cap-dependent non-canonical translation initiation mechanism that facilitates EMT and metastatic progression by selectively promoting translation of GRE-containing EMT effector mRNAs.
Although the canonical eIF4F complex has long been considered core to cap-dependent translation initiation [46], evidence that cap-dependent translation initiation independent of eIF4F may exist as a cellular mechanism for adaptive translation continues to emerge [47]. This is especially true in the context of physiological or pathological conditions where canonical cap-dependent translation is broadly inhibited [48–52]. In contrast to this, CELF1 was identified in a polysome profiling approach by virtue of association with the GREs common among the 3′ UTRs of only a handful of mRNAs whose translation increases in the mesenchymal or de-differentiated state [14]. Given that roughly 5% of human genes contain GREs within their 3′ UTRs [53], it is clear that as-yet unidentified additional factors are likely to confer specificity to CELF1-mediated translational activation. In addition, the apparent block in translation of CELF1’s targets, again GRE-mediated, when CELF1 is knocked down (Fig. 2d and f) implies that the regulation of these targets is highly regulated, and that CELF1 must evict another regulator from these elements to promote translation. This will be an exciting area for future study.
The model we propose (Fig. 7) is unique, and to our knowledge the first example of an RNA-recognition motif (RRM)-containing RNA-binding protein directly bridging a cis 3′ UTR regulatory element and eIF4E to promote eIF4G-independent translation of target mRNAs. To our knowledge, only two additional stimulatory 4E-BPs have been described—Mextli in Drosophila and threonyl-tRNA synthetase (TRS) in vertebrates [54, 55]. While Mextli contains a type I K homology RNA-binding domain, TRS contains a TGS domain. Alignment of the CELF1 protein sequence to either of these proteins (or eIF4G1) reveals no significant homology. In contrast, Mextli contains clear homology to eIF4G1 and, by binding both eIF4E and eIF3, functions as an eIF4G1 surrogate. Interestingly, Mextli appears to promote translation in the absence of eIF4A, instead associating with the Drosophila homolog of the DEAD-box helicase DHX35 [54]. While we do not here definitively establish that CELF1-directed translation occurs independently of eIF4A function, it is certainly a possibility that another helicase fills eIF4A’s role in this context.
While Mextli-directed translation is clearly important for germline stem cell maintenance, the specificity of Mextli-directed translation as related to the transcriptome, or whether it directly binds its mRNA targets, has not to our knowledge been established. TRS directly interacts with mRNAs involved in vertebrate development and has been shown to be critical for vasculogenesis. TRS-directed translation of the VEGF mRNA is mediated via a direct interaction with an anticodon-like loop in the 5′ UTR of this transcript. However, while TRS interacts with eIF4A and PABPC1, this translation is directed via the eIF4E homolog eIF4E2 (4EHP) rather than eIF4E itself. [55]. Interestingly and similarly to TRS, CELF1 has been previously implicated in the translation of mRNAs encoding the C/EBPβ transcription factor [56] and CDKN1A (p21) cyclin-dependent kinase inhibitor protein [57] via binding the 5′ region of these transcripts. It will be of particular interest to determine whether CELF1’s regulation of these genes in fibroblasts and liver cells, respectively, occurs via similar mechanisms.
Although our data clearly indicate that CELF1 has the potential to circularize mRNAs, the mechanism by which it recruits the 40S ribosomal subunit to initiate scanning to the start codon remains somewhat unclear. Our data indicate that CELF1 associates with eIF3C and eIF3H, both within cellular extracts and in m^7^G cap analogue binding assays. On one hand, the cap binding assays strongly support the notion that this interaction is independent of eIF4G1, since the C-terminus of eIF4G1, which binds eIF3, is not retained on the m^7^G beads. On the other hand, our immunoprecipitation experiments from cellular extract demonstrate that the interaction between CELF1 and eIF3 is RNA-dependent. Interestingly, work by the Cate group has shown that eIF3 promotes translation by directly binding the 5′ UTR of the JUNB homolog JUN [51], and that there is a direct association of eIF3 with the 3′ UTRs of highly translated mRNAs in human pluripotent stem cell-derived neural progenitor cells [58]. Whether eIF3 is similarly directly associated with CELF1 mRNA targets in de-differentiated breast epithelial cells, or whether this complex is brought in via another RNA-binding protein, is an exciting area for further study. Either of these mechanisms would likely provide some additional level of understanding regarding CELF1’s target specificity in this regulatory context.
In the scanning model of translation initiation, the decoding site and latch of the 40S subunit must open to allow the recruitment and migration of mRNA, and this unwinding process is catalyzed by the eIF4A helicase [15]. Surprisingly, our results support the conclusion that CELF1’s role in promoting translation of GRE-containing EMT effector mRNAs is independent of an interaction with eIF4A. Much work has led to the now generally accepted notion that the 5′ UTRs of mRNAs encoding oncogenic factors are characterized by increased length and/or more complex structure, and thus that translation of these mRNAs is more dependent upon eIF4A helicase activity than translation of mRNAs whose 5′ UTRs are less structured [59–61]. Importantly, most of the studies establishing this notion have focused specifically on mRNAs encoding regulators of the cell cycle rather than EMT or de-differentiation [50–52, 62]. Indeed, a compelling recent study robustly demonstrated the efficacy of zotatifin, a well-tolerated rocaglate-derived eIF4A inhibitor, in several mouse models of triple-negative breast cancer [63]. Beyond inducing a reduction in tumor cell proliferation, zotatifin treatment also sensitized the tumors to immune checkpoint blockade and treatment with existing first-line chemotherapeutic agents. Similar results were shortly thereafter shown for the rocaglate inhibitor MG-002 [64]. Taken together with the observations described within several recent published studies highlighting non-canonical modes of translation initiation in transformed cells [51, 52, 62], these and similar findings engender speculation that different facets of tumorigenesis or tumor progression may incorporate discrete modes of canonical or non-canonical translation initiation specific to these facets. Our demonstration that CELF1 interacts with eIF4B may hint that eIF4A is recruited to CELF1-containing translation pre-initiation complexes in a non-canonical fashion; alternatively, it is possible that another RNA helicase may function within this context. While this is a fascinating topic for further study, we speculate that inhibition of a second helicase specific to this process might serve as a robust form of differentiation therapy in advanced breast cancer, rendering late-stage tumors and metastases vulnerable to immune checkpoint blockade or existing first-line chemotherapeutic treatments.
While the components, mechanisms, and underlying signaling pathways dictating canonical translation pre-initiation complex formation in mRNA translation are well-established, the contribution of non-canonical complexes to translation is only beginning to be revealed. Already, however, it is clear that regulation of translation is both context- and cell-type dependent in terms of target specificity and activity [51, 52, 62, 65]. In theory, encoding more than one mechanism of cap-dependent translation initiation allows cells additional control over protein synthesis, perhaps where additional specificity in this control might provide the cell an adaptive advantage in response to a specific physiological stimulus or genetic insult. It is thus tempting to speculate that additional specialized translational complexes remain to be discovered, both in normal cellular function and within the diseased or transformed state. From the latter perspective, a more robust mechanistic understanding of analogous specialized mechanisms of translation may reveal novel vulnerabilities that may be exploited in the clinic, especially given the initial promise within clinical trials of drugs targeting canonical translation initiation as a bulk process [63, 66, 67].
Supplementary Material
gkag123_Supplemental_File
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Schwanhausser B, Busse D, Li N et al. Global quantification of mammalian gene expression control. Nature. 2011;473:337–42. 10.1038/nature 10098.21593866 · doi ↗ · pubmed ↗
- 2Robichaud N, Sonenberg N, Ruggero D et al. Translational control in cancer. Cold Spring Harb Perspect Biol. 2018;11:a 032896. 10.1101/cshperspect.a 032896.PMC 660146529959193 · doi ↗ · pubmed ↗
- 3Vogel C, Marcotte EM. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. Nat Rev Genet. 2012;13:227–32. 10.1038/nrg 3185.22411467 PMC 3654667 · doi ↗ · pubmed ↗
- 4Zhao J, Qin B, Nikolay R et al. Translatomics: the global view of translation. Int J Mol Sci. 2019;20:212. 10.3390/ijms 20010212.30626072 PMC 6337585 · doi ↗ · pubmed ↗
- 5Gobet C, Naef F. Ribosome profiling and dynamic regulation of translation in mammals. Curr Opin Genet Dev. 2017;43:120–7. 10.1016/j.gde.2017.03.005.28363112 · doi ↗ · pubmed ↗
- 6Pastushenko I, Blanpain C. EMT transition states during tumor progression and metastasis. Trends Cell Biol. 2019;29:212–26. 10.1016/j.tcb.2018.12.001.30594349 · doi ↗ · pubmed ↗
- 7Dongre A, Weinberg RA. New insights into the mechanisms of epithelial-mesenchymal transition and implications for cancer. Nat Rev Mol Cell Biol. 2019;20:69–84. 10.1038/s 41580-018-0080-4.30459476 · doi ↗ · pubmed ↗
- 8Sporn MB . The war on cancer. Lancet. 1996;347:1377–81. 10.1016/S 0140-6736(96)91015-6.8637346 · doi ↗ · pubmed ↗