Targeting eIF4A-Dependent Translation of KRAS Signaling Molecules

The translation of mRNAs into protein is tightly controlled at the level of the multisubunit eIF4F initiation complex (1). The eIF4F complex assembles on the 5′cap structure and scans the 5′UTR (untranslated region) for a translation start site. EIF4A (DDX2) is the RNA helicase component of the eIF4F translation initiation complex and is especially required to initiate translation of mRNAs with long 5′UTRs that contain highly structured RNA sequences such as multiple G-quadruplex (GQ) elements (2–4). To some extent this insight enables the predictive identification of eIF4A-dependent mRNAs. Notably, the NRAS and KRAS genes have predicted GQ-forming sequences in their 5′UTRs, although the therapeutic impact of this prediction is not known (5, 6). Importantly, the natural compound silvestrol binds eIF4A with nanomolar affinity and disables its RNA unwinding activity (7–10). Synthetic analogues of silvestrol (e.g., CR-1–31B) have shown promise in models of leukemia and lymphoma (2, 11, 12). In principle, cancers driven by a GQ-controlled oncogene such as KRAS should be susceptible to eIF4A blockade.

Genomic studies have catalogued the genetic drivers of pancreatic adenocarcinoma (PDAC) and show nearly ubiquitous activation of KRAS and loss of tumor-suppressor genes p53, p16/INK4A, and SMAD4 (13–17). Activation of mRNA translation is an important biological consequence of KRAS activation. Accordingly, PDACs show increased levels of mRNA translation and mediated through activation of MAPK, PI3K–AKT–mTOR signals, NRF2, and MYC (18–20). mTORC1/S6K1 signaling activates translation initiation through phosphorylating eIF4B, cofactor for eIF4A helicase and enhance translation of MYC in pancreatic cancer (21, 22). There have been significant advances in therapeutic development against PDAC (23). These include, inhibitors of G12C KRAS mutation (24–26), and strategies to cotarget KRAS downstream signaling molecules like (e.g., MAPK, PI3K, mTOR, EGFR, c-RAF; refs. 27–30). Although we have pharmacological inhibitors of the G12C KRAS mutation, this specific mutation is present in only 3% of PDACs (24, 25). A recent study reports effect of eIF4A inhibition on mouse PDAC (31). In this study, we explore exactly how eIF4A inhibition affects human PDACs and we measure effects on global and specific mRNA translation programs.

Pancreatic cancer cell lines were cultures as per specified by the ATCC. PANC1 and MiaPaca2 cells were purchased from the ATCC. PANC1 cells were cultured in DMEM supplemented with 10% FBS and penicillin–streptomycin (100 U/mL; 100 μg/mL; all Life Technologies). MiaPaca-2 cells were cultured in DMEM supplemented with 10% FBS, 2.5% horse serum penicillin–streptomycin (100 U/mL; 100 μg/mL; all Life Technologies). All the cell lines used were regularly tested for Mycoplasma contamination using PCR. The IC₅₀ data on the panel of cancer cell lines were done in collaboration with Tri-Institutional Drug Initiative program at Memorial Sloan Kettering Cancer Center (MSKCC, New York, NY).

Silvestrol was purchased from ChemScene (CS-0543). [±]CR-1–31B and [−]CR-1–31B were synthesized in house at organic chemistry core at MSKCC and Tri-Institutional Drug Development Initiative at MSKCC, respectively. Each was suspended in DMSO for in vitro experiments and 5.2% Tween 80 5.2% PEG 400 for in vivo experiments. Cycloheximide (C7698) was purchased from Sigma. pLEX-HA-birA*-K-Ras(G12D)-IRES-Puro was a gift from Paul Khavari (Addgene, plasmid #120562).

Human pancreatic cancer PANC1 cells were treated with DMSO or [±]CR-1–31B (25 nmol/L; 45 minutes) followed by cycloheximide treatment for 10 minutes. Total RNA and ribosome-protected fragments were isolated following published protocol (32). Deep sequencing libraries were generated from these fragments and sequenced on the HiSeq 2000 platform. Genome annotation was from the human genome sequence GRCh37 downloaded from Ensembl public database: http://www.ensembl.org.

First, ribosome footprint (RF) reads were filtered on the basis of the quality score, which kept reads that have a minimum quality score of 25 for at least 75% of the nucleotides. Second, the linker sequence (5′-CTGTAGGCACCATCAAT-3′) was trimmed from the 3′ end of the reads. Next, we filtered out the reads shorter than 15nt after the linker-trimming step. All these aforementioned steps were done by using FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/index.html). To remove ribosomal RNA, the footprint reads were then aligned to the ribosome RNA sequences of GRCh37 downloaded from UCSC Table Browser (https://genome.ucsc.edu/cgi-bin/hgTables). After removing the reads aligned to the ribosome RNAs, RF reads were mapped to the human genome sequence GRCh37 downloaded from Ensembl public database: http://www.ensembl.org using HISAT2 with default parameters. We only used the uniquely aligned reads for further analysis.

Total mRNA-sequencing reads were aligned to the GRCh37 reference using HISAT2. Similarly, as RF reads alignment, we performed the splice alignment for the paired-end mRNA-seq datasets with the default parameters. We only kept the uniquely aligned reads for the downstream analysis. The alignment quantification for both RF and mRNA sequencing was done using featureCounts with the annotations of the protein coding genes of GRCh37 as input. Only reads aligned to the exonic regions of the protein coding genes were used for the downstream analysis.

We used Ribo-diff to analyze the translation efficiency (TE) based on the ribosome footprinting and mRNA sequencing data. Genes with at least 10 normalized read counts as the sum of RF and RNA sequencing data were used as input, which resulted in 10,861 genes in total. Genes with significantly changed TE were defined by the q value cutoff equal to 0.001.

The longest transcript was selected to represent each corresponding gene. The 5′UTR sequences of the transcripts were collected for predicting motifs. Both the significant genes with increased or decreased TE and the corresponding background gene sets were used to predict motifs by DREME (33). The occurrences of the significant motifs (E < 0.05 and P < 1 × 10^–8 from DREME) were called using FIMO (33) with default parameters for strand-specific prediction of all the 5′UTR sequences.

We used RNAfold (version 2.1.6) to predict the GQ formation in the RNA secondary structure with the –g option. The number of different types of GQ (GG×4, GGG×4, and GGGG×4) was then calculated by counting the number of consecutive “G”s using customized python scripts.

Full-length 5′UTR of KRAS transcript variant b was cloned in MSCV-CMV-dsRed-IRES d1eGFP using Age I and Dra III restriction enzyme sites. This strategy replaced the IRES with the full-length 5′UTR of KRAS. For mutant version, we changed the GG pattern in the full-length 5′UTR of KRAS that disrupted the RNA GQ structure. The clones were validated by PCR sequencing using CMV forward primers and the presence of full-length 5′UTR of KRAS (wild-type and mutant, mentioned below) was confirmed downstream of CMV promoter. Reporter plasmids were used for transient transfection in 293T cells. Destabilized d1eGFP has a half-life of 1 hour and the translation activity driven by respective 5′UTRs was measured by d1eGFP intensity using flow cytometry. The 5′UTR sequence for full-length wild-type and mutant version with restriction enzyme sites is provided below.

CCGGTGGCCGCACCACCACACCCAGCAGCGCGCGCCGCAGTACCCGCACCGAGCCTGCCACCGCCGCGCCCCGTGCTCCCGGCCCCCGCCGTTGCACACTGACAGCGAGCGCACCGCAACCGCTGGAACCACCACCACACCCAGAACCTCAGCACCTCCCAACTGCGGGCGCGCGGCCTGCTGAGAATGCACGTT

GTGCATTCTCAGCAGGCCGCGCGCCCGCAGTTGGGAGGTGCTGAGGTTCTGGGTGTGGTGGTGGTTCCAGCGGTTGCGGTGCGCTCGCTGTCAGTGTGCAACGGCGGGGGCCGGGAGCACGGGGCGCGGCGGTGGCAGGCTCGGTGCGGGTACTGCGGCGCGCGCTGCTGGGTGTGGTGGTGCGGCCA

CCGGTGGCCGCGGCGGCGGAGGCAGCAGCGGCGGCGGCAGTGGCGGCGGCGAAGGTGGCGGCGGCTCGGCCAGTACTCCCGGCCCCCGCCATTTCGGACTGGGAGCGAGCGCGGCGCAGGCACTGAAGGCGGCGGCGGGGCCAGAGGCTCAGCGGCTCCCAGGTGCGGGAGAGAGGC CTGCTGAAACACGTT

GTGTTTCAGCAGGCCTCTCTCCCGCACCTGGGAGCCGCTGAGCCTCTGGCCCCGCCGCCGCCTTCAGTGCCTGCGCCGCGCTCGCTCCCAGTCCGAAATGGCGGGGGCCGGGAGTACTGGCCGAGCCGCCGCCACCTTCGCCGCCGCCACTGCCGCCGCCGCTGCTGCCTCCGCCGCCGCGGCCA

Restriction enzyme sites are shown as underlined.

CMV Forward Primer-CGCAAATGGGCGGTAGGCGTG

To delete the RNA GQs from KRAS 5′UTR, we designed sgRNAs targeting the genomic regions on KRAS. Each sgRNA was cloned into the lenti-CRISPRv2 (Addgene, cat no. 52961) using BsmB I restriction enzyme site. PANC1 cells were cotransfected with sgRNA containing plasmids and then selected with puromycin (2 μg/mL) for 24 hours. Selected colonies were expanded for 2 weeks. The deletion was confirmed using PCR primers spanning the KRAS 5′UTR region in genomic DNA. The sequence of sgRNA and PCR primers is provided below.

KRAS-5′UTR-5′sg1-F: CACCGCTAGGCGGCGGCCGCGGCGG

KRAS-5′UTR-5sg1′-R: AAACCCGCCGCGGCCGCCGCCTAGC

KRAS-5′UTR-3′sg1-F: CACCGCCAGAGGCTCAGCGGCTCCC

KRAS-5′UTR-3′sg1-R: AAACGGGAGCCGCTGAGCCTCTGGC

KRAS-5′UTR-3′sg2-F: CACCGCTCAGCGGCTCCCAGGTGC

KRAS-5′UTR-3′sg2-R: AAACGCACCTGGGAGCCGCTGAGC

KRAS 5′UTR F2: GACCGCCTCCAGCCTCA

KRAS 5′UTR R2: AAGAAGAATCGAGCGCGGAA

Lysates were made using TNN lysis buffer (50 mmol/L Tris-Cl, 250 mmol/L NaCl, 5 mmol/L EDTA, 0.5% NP-40 supplemented with protease inhibitor). 60 μg of protein was loaded onto SDS-PAGE gels then transferred onto Immobilon-FL Transfer Membranes (Millipore IPFL00010). The antibodies used were KRAS2B (Proteintech), MYC (Cell Signaling Technology), HRAS (Abcam), ERK (BD Pharmingen), p-ERK (p42/44), MET, YAP1, XPO1, DDX6, PARP1, RALBP1, PI3KCA, RAC1/2, β-tubulin, GAPDH (Cell Signaling Technology), and β-actin (Sigma A5316). Quantification of Western blot images was done by using ImageJ software.

Human pancreatic cancer cell lines PANC1, were used for annexin V staining using kit (Invitrogen) and following the manufacturer's instruction. Annexin V staining was detected by FACS analysis.

We performed polysome profiling to evaluate the effect of [±]CR-1–31B on global translation and translation of KRAS4B transcript. Briefly, we used polysome lysate from PANC1 cells treated with DMSO or [±]CR-1–31B (50 nmol/L) for 1 and 4 hours in duplicates and performed polysome fractionation using a published protocol from Panda and colleagues (34). The relative distribution of the percentage of mRNA of KRAS and B2M over the sucrose gradient was studied by RT-qPCR analysis of the RNA in each of the 12 gradient fractions.

To evaluate the eIF4A activity, we have used a fluorescent duplex unwinding assay as reported in refs. 35 and 36. Briefly, we designed oligo containing 12-mer CGG4 (GQ) motif and (AG)7 repeats labeled with Cy3 on the positive strand and with BHQ (Black Hole Quencher) on the negative strand. We purified human eIF4A1, eIF4H isoform 2, and eIF4GΔ using previously reported study (37). Purified proteins eIF4A and eIF4GΔ proteins are stored at −80°C in storage buffer (20 mmol/L Hepes, pH 7.5, 200 mmol/L KCl, 1 mmol/L DTT) supplemented with 10% glycerol. To maintain stability, purified eIF4H is stored at −80°C in storage buffer supplemented with 20% glycerol as reported in refs. 35 and 36. We performed the duplex unwinding exactly as reported in Ozes and colleagues (35) using eIF4A1 (10 μmol/L), eIF4H isoform 2 (10 μmol/L), eIF4GΔ (5 μmol/L) with 12 mer (CGG) and (AG)7 repeat oligos (50 nmol/L) in the presence of eIF4A inhibitor [±]CR-1–31B at (50 and 100 μmol/L).

Oligo sequences are provided below.

1. 12 mer (CGG)-Cy3 REV- 5′-Cy3-CGGCGGCGGCGG-3′

2. 12 mer (CGG)-BHQ FOR-5′-GCCGCCGCCGCC-BHQ-3′

3. 12 mer (CGG)-Competitor DNA-5′-GCCGCCGCCGCC-3′

4. (AG)7 repeat-Cy3 REV-5′-Cy3-AGAGAGAGAGAG-3′

5. (AG)7 repeat-BHQ FOR-5′-CUCUCUCUCUCU-BHQ-3′

6. (AG)7 repeat-Competitor DNA-5′-CTCTCTCTCTCT-3′

Human pancreatic cancer MiaPaca-2 cells expressing stable GFP-Luciferase reporter were injected in subcutaneous flank in J:Nu mice (5 million cells per flank). IVIS imaging was performed weekly to monitor the tumor growth. When tumors were between 80 and 100 mm³, Silvestrol (0.5 m/kg) or [±]CR-1–31B (0.5 mg/kg) or [−]CR-1–31B (0.5 mg/kg) was injected in mice intraperitoneally twice a week until the control mice developed fully grown tumors. P values were calculated using two-way repeated measures ANOVA. PDX tumors were established by the antitumor facility at MSK under an approved Institutional Review Board protocol and were transplanted subcutaneously in nude mice. Once tumors reached 80–100 mm³, the mice were randomized into treatment groups as above. J:nu mice were purchased from The Jackson Laboratory. All animal experiments were performed in accordance with regulations from Memorial Sloan-Kettering Cancer Center's Institutional Animal Care and Use Committee.

The PDA cell line KPC4662 used for orthotopic implantation was obtained from the RH Vonderhide group and previously described (38) and stably transfected with GFP-Luciferase reporter. C57BL/6 mice were obtained from The Charles River Laboratories. For orthotopic implantation of PDEC, we followed previously described procedure with some modifications (39). In brief, mice were anesthetized using isoflurane and then the pancreas was exposed through an abdominal incision (laparotomy). PDAC (2 ×10⁵ cells/mouse) was suspended in Matrigel (Corning) diluted 1:1 with cold PBS (total volume of 25 μL) and injected into the tail region of the pancreas using a Hamilton Microliter Syringe. A successful injection was verified by the appearance of a fluid bubble without intraperitoneal leakage. The abdominal wall was closed with absorbable Vicryl RAPIDE sutures (Ethicon) and the skin was closed with wound clips (Roboz). Mice with luciferase imaging diagnosed tumors of volume 50 to 150 mm³ were enrolled and block randomized into treatment groups. Tumors were visualized and reconstructed for quantifying tumor volume using the integrated Vevo 2100 Workstation software package. The Institutional Animal Care and Use Committee at MSKCC approved all animal care and procedures.

For the in vivo application in KPC mouse model of PDAC (PdxCre; KrasG12D, TP53fl/fl), [±]CR-1–31B was diluted in 5.2% PEG400 (Sigma-Aldrich), 5.2% Tween 80 (Sigma-Aldrich), 2% DMSO (Sigma-Aldrich), and administered intraperitoneally at 0.5 mg/kg, twice a week, from d28 to d83. For the in vivo application in PDA cell line KPC4662-derived orthotopic tumors in pancreas of C57/Bl mice, [±]CR31B was diluted in 5.2% PEG400 (Sigma-Aldrich), 5.2% Tween 80 (Sigma-Aldrich), 2% DMSO (Sigma-Aldrich). Vehicle or [±]CR-1–31B was administered at 0.5 mg/kg dose via intraperitoneal injection on Monday–Wednesday–Friday (3 days a week). For the in vivo application in MiaPaca-2 cell line–derived xenografts nude mice, silvestrol was diluted in 5.2% PEG400 (Sigma-Aldrich), 5.2% Tween 80 (Sigma-Aldrich), 2% DMSO (Sigma-Aldrich). Vehicle or silvestrol was administered at 0.5 mg/kg dose via intraperitoneal injection on Monday–Wednesday–Friday (3 days a week). For the in vivo application in human PDAC-derived tumors in nude mice, [−]CR-1–31B was diluted in 10% Captisol (Sigma-Aldrich) in sterile water. Vehicle or [−]CR-1–31B was administered at 0.5 mg/kg dose via intravenous injection on Monday–Thursday (e.g., twice a week). Toxicity in all the in vivo treatment experiments was monitored by weight loss and daily clinical observation until the end of the experiment. 24 hours after the last test article administration, 4–5 mice in each group were sacrificed and clinical chemistry, hematology and tissue-specific histopathology were done at autopsy.

The study was conducted under Memorial Sloan-Kettering Cancer Center Institutional Review Board approval (MSKCC IRB 15–149 or 06–107) and all patients provided written informed consent before tissue acquisition. Tissue resections and biopsies from patients with pancreatic cancer were processed according to protocols previously described by Boj and colleagues (40, 41), slightly modified to ensure maximum viable cell recovery and organoid formation efficiency. Briefly, resected tissue or biopsy were minced in less than 1-mm³ pieces and digested in organoid medium containing collagenase II (2.5 mg/mL) and Rho kinase inhibitor Y-27632 10.5 μmol/L (Selleck Chemicals). Tissue digestion was performed up to 4 hours at 37°C. Red blood cells were removed by specific lysis with ACK buffer (Gibco). After 3 successive wash in PBS and organoid medium, dissociated cells were seeded in growth factor reduced Matrigel (BD Biosciences) and cultured in WNT-driven expansion medium containing: DMEM-F12 Advanced (Gibco), Hepes 10 mmol/L (Gibco), antibiotics 500μg/mL (Gibco), Glutamax 2 mmol/L (Gibco), A83–01 0.5 μmol/L (Tocris), human EGF 50 ng/mL (Peprotech), human FGF10 100 ng/mL (Peprotech), human Noggin 100 ng/mL (Peprotech), human Gastrin I 10 nmol/L (Sigma), N-acetylcysteine 1.25 mmol/L (Sigma), Nicotinamide 10 nmol/L (Sigma), B-27 supplement 1X (Gibco), Wnt3A conditioned media 50% (v/v, from Hans Clevers, Hubrecht Institute, the Netherlands), R-spondin1 conditioned media 10% (v/v, from Calvin Kuo, Stanford University, Stanford, CA). Organoid lines were considered established when sustained epithelial proliferation was maintained over 5 passages (about 3- to 5-week culture).

Evaluation of PDAC organoids sensitivity to [±]CR-1–31B was performed in 384-well plate format. Briefly 2,000 single organoid cells were plated into Matrigel-coated well and precultured for 4 days. Organoids were then exposed to the drug over a 7-concentration range (from μmol/L to sub-nmol/L) or vehicle control for 6 days. Cell viability was assessed at day 10 using the Cell Titer-Glo 3D assay (Promega).

Cells were seeded in 6-well plates (20 × 10³ cells per well) and allowed to adhere overnight in regular growth media. DMSO or [±]CR-1–31B (10 nmol/L) was added and refreshed every 3 days until the end of the experiment (14 days). For each independent experiment, the DMSO or [±]CR-1–31B–treated cells were fixed in 4% formaldehyde for 15 minutes at room temperature and subsequently stained with 0.1% crystal violet and digitalized on an image scanner. All experiments were performed at least three times in triplicates and representative results are shown.

Tissue specimens were fixed in 4% buffered paraformaldehyde, dehydrated, and embedded in paraffin wax. Formalin-fixed paraffin-embedded sections of 3 μm were stained with hematoxylin and eosin, TUNEL, and Ki-67. Immunohistochemical analysis was done on five samples in each vehicle and drug-treated group by counting number of positive tumor cells at 10 high-power field (×400) and for Ki-67 staining one hot spot area at ×400 was selected for each case.

Total RNA was extracted using AllPrep DNA/RNA/Protein Mini Kit (Qiagen 80004). cDNA was made using SuperScript III First-Strand (Invitrogen 18080–400). Analysis was performed by ΔΔC_t. Applied Biosystems TaqMan GeneExpression Assays: Human KRAS Hs00364284_g1, Myc Hs00153408_m1, MET Hs01565584_m1, YAP1 Hs00902712_g1, XPO1 Hs00185645_m1, DDX6 Hs00898915_m1, and B2M Hs00187842_m1.

All the results were analyzed with two-tailed t tests unless specified. The significance of motif enrichments was from DREME program based on the Fisher exact test. The hypergeometric test was performed to test for the significance in the enrichment of the gene overlap in the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway.

Additional Methods, Supplementary display items and source data are available in the online version of the article; references unique to these sections appear only in the online article.

The ribosome footprinting and total mRNA sequencing raw and processed data were deposited in the NCBI Gene Expression Omnibus database GSE120159 accession number available at following link. (To review GEO accession GSE120159: Go to https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE120159 Enter token yhkpamwsnpinfkt into the box).

The KRAS protein is encoded by two variant transcripts (a and b) and two pseudogenes that do not encode a protein (Supplementary Fig. S1A). 5′UTR sequence of human and mouse KRAS transcripts is identical as shown by ClustalX alignment (Supplementary Fig. S1A and S1B). The KRAS4B 5′UTR is 192 bp long and computational prediction using M-Fold indicates the presence of three highly structured and potentially GQ-forming regions in the 5′UTR of both KRAS transcripts that are reminiscent of eIF4A-dependent mRNAs (Fig. 1A and B; refs. 2, 3, 5, 6). The transcription start site (TSS) can vary within different cell types and tissue (32). We have compared the TSS in various PDAC cell lines (RNA-seq data reported in GSE dataset GSE160434) and observed that KRAS variant b (KRAS4B) is the most expressed transcript across the nine PDAC cell lines, suggesting that the TSS for KRAS is very much stable in these PDAC cell lines. We built a KRAS4B 5′UTR translation reporter assay to test the eIF4A requirement using the eIF4A inhibitor [±]CR-1–31B (11). Briefly, destabilized eGFP (d1eGFP, T1/2 = 1 hour) is expressed from a CMV promoter and its translation controlled by the KRAS4B 5′UTR or a mutant 5′UTR where predicted GQ structures are disrupted without changing 5′UTR length or GC content (Fig. 1C). [±]CR-1–31B is a synthetic analogue of silvestrol synthesized by the J. Porco's group at University of Boston and shows nanomolar eIF4A inhibition (2, 11, 12, 31, 42–44). We measured the basal expression of d1eGFP driven by KRAS4B 5′UTR or a mutant 5′UTR in 293T cells (Supplementary Fig. S1C). Next, we confirmed the half-life of d1eGFP by adding a pan translation inhibitor cycloheximide. Both KRAS4B 5′UTR or a mutant 5′UTR-driven d1eGFP expression was reduced to 50% following 1 hour treatment with cycloheximide (Supplementary Fig. S1D and S1E). In 293T cells, transiently expressing the KRAS4B 5′UTR reporter constructs treatment with [±]CR-1–31B (50 nmol/L) reduces translation of the d1eGFP reporter controlled by the wild-type KRAS 5′UTR by 70% and has little effect on the mutant 5′UTR (P < 0.001; Fig. 1D; Supplementary Fig. S1F). Consistently, treatment of KRAS mutant PDAC cell lines (PANC1, in MiaPaca-2) with different doses of [±]CR-1–31B (1–50 nmol/L, 72 hours) reduces the KRAS protein levels consistent with reported KRAS protein half-life (Fig. 1E–H; ref. 45). EIF4A inhibition resulted in slight reduction of phospho-ERK levels at 72 hours of [±]CR-1–31B (1–50 nmol/L) treatment (Fig. 1E and F). KRAS mRNA expression was not affected by [±]CR-1–31B (Supplementary Fig. S1G). Importantly, eIF4A dependence of KRAS translation depends on an intact, endogenous KRAS 5′UTR. For example, using two pairs of CRISPR guideRNAs (sgRNAs) we induce KRAS 5′UTR deletions sg1–sg2 and sg1–sg3, respectively, that abrogate [±]CR-1–31B sensitivity and eIF4A-dependent KRAS translation in PANC1 cells (Fig. 1I and J; Supplementary Fig. S1H and S1I). Deletion of KRAS 5′UTR did not affect the protein expression of KRAS, phospho-ERK levels and mRNA of KRAS in the PANC1 cells (Supplementary Fig. S1J and S1K). Hence, GQ elements in the KRAS 5′UTR confer dependence on eIF4A in translation reporter assays and in human PDAC cells.

Nanomolar concentrations (2–10 nmol/L) of [±]CR-1–31B inhibit cell growth in KRAS mutant PDAC and other cancer cell lines indicated by IC₅₀ analysis, PARP1 cleavage, and block colony formation in human PDAC cell lines (PANC1, MiaPaca-2; Fig. 2A and B; Supplementary Fig. S2A and S2B). Primary fibroblast lines showed at least approximately 1,000-fold higher IC₅₀ for [±]CR-1–31B (Supplementary Fig. S2A). Importantly, CRISPR–cas9-engineered KRAS 5′UTR deletions (sg1–sg2 and sg1–sg3) are largely (>80% colonies at 4 and 6 nmol/L [±]CR-1–31B) able to rescue colony formation by PANC1 cells under continued [±]CR-1–31B exposure, indicating that KRAS is a significant target of the drug's action (Fig. 2C). Deletion of KRAS 5′UTR in PANC1 cells increased the IC₅₀ for [±]CR-1–31B to approximately 10 nmol/L compared with approximately 6 nmol/L in Cas9-expressing PANC1 control cells (Supplementary Fig. S2C and S2D). Next, overexpression of HA-BirA*-K-RAS4B (G12D) in PANC1 cells rescued the cell death induced by [±]CR-1–31B. We observed only 30% inhibition in cell proliferation at the highest dose of [±]CR-1–31B (10 μmol/L) in PANC1 cells with ectopic expression of KRAS4B (G12D; Supplementary Fig. S2E). Although the [±]CR-1–31B treatment inhibited the endogenous KRAS4B protein, it did not affect the ectopic expression of HA-BirA*-KRAS4B(G12D) as determined by the Western blot analysis (Supplementary Fig. S2F). [±]CR-1–31B treatment also equally inhibited the MYC protein levels similar to the wild-type PANC1 cells (Supplementary Fig. S2F). We observed higher levels of total ERK in PANC1 cells with ectopic expression of KRAS4B (G12D) compared with wild-type PANC1 cells but the p-ERK levels were equally affected by the [±]CR-1–31B treatment (Supplementary Fig. S2F).

Organoid culture more closely models the three-dimensional tissue organization of tumors that may also affect drug action (40). Here, [±]CR-1–31B effectively blocked organoid formation from murine KRAS/p53 PDACs (Fig. 2D) and similarly inhibited the growth of primary human PDAC organoids ([±]CR-1–31B IC₅₀s: 0.4–22 nmol/L; 72 hours; Fig. 2E and F). Hence, the silvestrol analogue [±]CR-1–31B kills PDAC cells and the effects are reduced upon disruption of the KRAS 5′UTR.

We performed ribosome footprinting (ribosome profiling) and deep sequencing (32) in the presence and absence of the eIF4A inhibitor [±]CR-1–31B to identify eIF4A-dependent translation. Briefly, we normalized ribosome-protected RNA fragments (RF reads) to the total RNA abundance to isolate changes in TE. We performed ribosome footprinting on three control (DMSO) and three [±]CR-1–31B (25 nmol/L)-treated PANC1 samples (Fig. 3A). We chose an early time point (45 min) to minimize secondary and knockon effects. Read mapping to ribosomal RNAs, noncoding RNAs, library linkers, and incomplete alignments were removed from the analysis (Supplementary Fig. S3A–S3F). Most of the remaining reads range from 25 to 35 nucleotides in length and map to protein coding genes (Supplementary Fig. S3A–S3F). The total number of RF reads mapped to exons was 4.1 million in control and 3.5 million in [±]CR-1–31B–treated samples. This corresponds to 19,821 protein coding genes. Quality control analysis of replicates showed significant correlations among the replicates with a Pearson coefficient >0.97 (Supplementary Fig. S3G and S3H). We used the RiboDiff statistical framework to isolate the effect on mRNA translation (46). With a very stringent statistical cutoff at q < 0.001 (FDR <1%), we identified 614 mRNAs whose translation was significantly repressed (TE down: n = 614; q < 0.001), and we also detected a set of mRNAs showing a relative increase in ribosome occupancy (TE up: n = 456; q < 0.001; Fig. 3B; Supplementary Table S1A and S1B; Complete dataset at GEO# GSE120159). A full list of genes differentially affected on TE by eIF4A inhibitor [±]CR-1–31B in PANC1 cells is provided in Supplementary Table S2. Importantly, we notice that eIF4A-dependent (TE down) mRNAs included key oncogenic drivers of PDAC, such as MYC, MET, YAP, TGFβ1/2, PI3K, and other proteins involved in RAS signaling (Fig. 3B). KRAS TE was reduced by 35% at q = 0.5 whereas MYC TE decreased by 40% at q < 0.001 (Supplementary Table S2).

These findings differ from results reported in a recent polysome fractionation study on PDAC cells treated with the same inhibitor (31). For example, we do not see the reported changes in the translation of redox and central carbon metabolism genes (31). A review of the published dataset (Supplementary Table S2 in ref. 31) shows that the polysome fractionation experiment identified exactly two genes (Lag3 and Tmcc2!) whose translation was significantly (q < 0.05) reduced upon eIF4A inhibition. Other genes implicated in the study as eIF4A targets are shown with significance values as high as q = 0.25 and in some instances q = 0.45 are chosen on the basis of the manual analysis. Most likely this reflects poor inter-sample reproducibility of manual or instrumental separation of heavy and light polysome fractions using this older methodology.

We confirmed the effects on protein levels for KRAS and several targets across identified with as highly significant cell lines (PANC1, MiaPaca-2), organoids, in PDACs arising in vivo (PdxCre; KrasG12D, TP53fl/fl; Fig. 3C–F; ref. 47). [±]CR-1–31B did not affect the mRNA levels of these targets in PANC1 cells (Supplementary Fig. S3I). We also validated the effect of [±]CR-1–31B on global and KRAS translation by polysome profiling. We observed a significant increase of total mRNA in the light molecular weight (LMW; P < 0.05) and corresponding reduction in the heavy molecular weight (HMW) polysome fraction following [±]CR-1–31B treatment (50 nmol/L for 1 and 4 hours) in PANC1 cells (Supplementary Fig. S3J). KRAS mRNA was highly and significantly reduced in the HMW polysome fraction (P < 0.05) whereas B2M mRNA showed slight reduction in the HMW polysome fraction following [±]CR-1–31B treatment, indicating that [±]CR-1–31B selectively target KRAS translation (Supplementary Fig. S3K and S3L). Next, we evaluated the effect of [±]CR-1–31B on the kinetics of KRAS protein degradation. Using cycloheximide treatment (1 ng/mL), we observed approximately 40% degradation of KRAS protein between 4 and 24 hours in PANC1 cells whereas [±]CR-1–31B treatment (50 nmol/L) reduced KRAS protein expression to similar extent (40%–50%) in 4–24 hours as observed and quantified by Western blotting analysis (Supplementary Fig. S3M and S3N). Both cycloheximide (1 ng/mL) and [±]CR-1–31B (50 nmol/L) combined showed no significant difference on KRAS protein degradation suggesting that [±]CR-1–31B do not affect the half-life of KRAS protein in PANC1 cells (Supplementary Fig. S3M and S3N). An unbiased gene ontology analysis of eIF4A-dependent genes (TE down) further supported the enrichment (P <1.18E−09) of RAS/MAPK pathway genes, including KRAS, RALA, RALBP1, MEK1/2, RAC2, and MYC (Fig. 3G and H). Together, these findings reveal coordinate translational downmodulation of key KRAS–MAPK–MYC signaling proteins following eIF4A inhibition.

Next, we explored these significant and confirmed eIF4A targets for common molecular features. We compared the [±]CR-1–31B sensitive (TE down) group of mRNAs with annotated 5′UTRs (n = 591), the [±]CR-1–31B independent (TE up; n = 431) to each other and a background list of 623 equally expressed and annotated mRNAs that showed no significant change in their translation (Supplementary Table S3A–S3C). We noticed that [±]CR-1–31B–sensitive mRNAs had significantly longer 5′UTRs (TE down vs. Bkg P = 4.0e−13; TE down vs. TE up P = 3.2e−24; Supplementary Fig. S4A). Next, we applied the MEME (Motif-Based sequence analysis tool; ref. 48) search tool to investigate sequence elements that were either over- or underrepresented in any of the three groups. Comparing the TE down group with TE up and background lists, we identify one significantly overrepresented 12-mer sequence (CGGCGGCGGCGG) and two 9-mer motifs (CGGCGGCGG and CCGCCGCCG; P = 3.8e−10; P = 1.6e−14; and P = 1.1e−8, respectively; Fig. 4A; Supplementary Table S4). We observed no significant enrichment or depletion of sequences/motifs in the coding or 3′UTR sequences. We also performed a separate search for differentially represented structural elements using RNAfold program. This search identified an enrichment of predicted RNA GQ structures (sequence: GGX4) in the 5′UTRs of eIF4A-dependent (TE down) genes compared with the background genes (P = 8.4e−7) and TE up genes (P = 3.8e−9); other potentially GQ forming sequences (GGGX4 and GGGGX4), TOP motif, IRES, uORFs, pyrimidine-rich translation element, and putative eIF4A-binding sites [GAAG (AG)₃ repeats] were too infrequent to detect changes in their representation between groups (Supplementary Fig. S4B and S4C). Next, the 12-mer sequence (CGGCGGCGGCGG) and the 9-mer motif (CGGCGGCGG) significantly overlapped with GQ-forming sequences whereas the 9-mer motif (CCGCCGCCG) did not coincide with the GQ sequences (Fig. 4B). 36% of the TE down transcripts contains GQ sequences their 5′UTR in whereas only 19% and 23% of the TE Up and background genes harbor GQ sequences, respectively (Fig. 4C). Consistently, we found that 5′UTR GQ sequence elements were highly significantly (P < 0.0001) associated with changes in TE across the transcriptome upon eIF4A inhibition (Fig. 4D). A swimmer plot illustrates how the number of GQ elements per 5′UTR corresponds to the degree of translational repression [±]CR-1–31B treatment, indicating a dosage effect (Fig. 4E). We also mapped the relative position and number of GQ sequence elements in the 5′UTR of key oncogenes using QGRS mapper and a stringent cutoff (QGRS score >20); for example, the algorithm identifies three GQ sequence elements in the KRAS 5′UTR, four in MYC, five in RALBP1, four in PI3K, four in HRAS, and seven in YAP1 (Fig. 4F). Again, our findings are in stark contrast with Chan and colleagues (31) that report effects on short and unstructured 5′UTRs, instead we find that long and highly structured 5′UTRs are highly enriched among eIF4A-dependent mRNAs.

We directly tested unwinding by the eIF4A helicase in the presence and absence of the inhibitor ([±]CR-1–31B) in a sequence-specific manner. Briefly, we used an in vitro duplex unwinding assay using the purified initiation factors (eIF4A, eIF4H, eIF4G and ATP) to measure resolution of the 12-mer CGG4 (GQ) motif, and we used a ubiquitous (AG)₇ that has been proposed as preferred eIF4A-binding sequence (35), as a control. The assay uses duplex probes labeled with Cy3 on the positive strand and with BHQ on the negative strand such that unwinding results in an increased Cy3 signal (Fig. 4G). ATP addition results in activation of 12-mer CGG4 (GQ) motif unwinding that is blocked by the eIF4A inhibitor [±]CR-1–31B (p_{vehicle vs. [±]CR-1–31B} = 0.0001; Fig. 4H). On the other hand, [±]CR-1–31B led to increased unwinding of the (AG)₇ sequence (p_{vehicle vs. [±]CR-1–31B} < 0.001), potentially consistent with reports of rocaglamide increasing eIF4A's affinity ubiquitous and AG-rich sequences (Fig. 4I; refs. 7, 49). Hence, [±]CR-1–31B inhibits the eIF4A helicase activity in a sequence-specific manner on 12-mer CGG4 (GQ) sequence motif that are also highlighted in our ribosome profiling data.

Chan and colleagues report activity of an eIF4A inhibitor against mouse PDAC (31) and we are pleased to confirm and expand the therapeutic benefit in the same KPC model and also in human PDAC tumors. Briefly, in murine KPC tumors we observe an increase in median overall survival from 49 days (control) to 69 days upon treatment with [±]CR-1–31B (0.5 mg/kg, i.p., twice a week, d28–d83 n = 7/5, P < 0.02 log-rank Mantel-Cox test; Fig. 5A). In orthotopically engrafted KPC tumors engineered to express luciferase, responses were more varied and [±]CR-1–31B (0.5 mg/kg, i.p., twice a week, d10–d42) induced responses in 3 out of 5 animals, whereas control animals uniformly succumbed to PDAC (n = 5, p_{vehicle vs. CR-1–31B} < 0.01; Fig. 5B; Supplementary Fig. S5A and S5B). Histology showed a cytostatic effect with loss of proliferation (Ki-67 positive; Fig. 5C; Supplementary Fig. S5C), animal weights, blood, and platelet counts were unchanged, indicating tolerability in mice (Fig. 5D; Supplementary Fig. S5D–S5F). Expression of KRAS, MYC, PI3K, and YAP1 was significantly reduced in orthotopically engrafted KPC tumors treated with [±]CR-1–31B (0.5 mg/kg, i.p., twice a week, d10–d42) compared with vehicle-treated tumors as observed by Western blotting and quantification (P < 0.05; Fig. 5E and F).

We further find that the eIF4A inhibitor has single-agent activity against human PDACs. Specifically, we purified the active [−] enantiomer of CR-1–31B. We treated a primary, patient-derived PDX model of PDAC transplanted into the flanks of five nude mice once tumors had reached a volume of approximately 80 mm³ (dosing schedule: [−]CR-1–31B, 0.5 mg/kg, i.v., twice a week from d22–d54). We observed a significant delay in human PDX growth during the treatment period, although tumors resumed growth when treatment was stopped, potentially indicating the need for combination therapies (p_{vehicle vs. [−]CR-1–31B} < 0.03, n = 5; Fig. 5G). When tumors in either group reached a size of approximately 2,000 mm³ animals were euthanized and we considered this as an endpoint for a survival analysis. Notably, [−]CR-1–31B treatment of tumor-bearing animals significantly increased the median overall survival from 49 days to 62 days (0.5 mg/kg i.v., twice a week, n = 10 in vehicle group, n = 13 in treated group, Kaplan–Meier analysis: P < 0.0001 Log-rank Mantel-Cox test; Fig. 5H). Histology showed a striking reduction of proliferating (Ki-67 positive) adenocarcinoma cells (indicated by arrows) surrounded by tumor stroma (Fig. 5I). We confirmed significant loss of KRAS, MYC, and phospho-ERK (P < 0.05) whereas total ERK remained unchanged as shown by immunoblots on lysates of in vivo treated tumors collected from controls or 24 hours after the last [−]CR-1–31B dosing (Fig. 5J). ImageJ quantification is shown where the KRAS, MYC protein expression is normalized to β-actin (Fig. 5K). p-ERK is normalized to total ERK (Fig. 5K). As with the racemic mixture, we found [−]CR-1–31B was well tolerated without evidence of drug-related mortality or weight loss (Fig. 5L).

To enhance reproducibility of these important results, we also tested the natural compound silvestrol and the analogue [−]CR-1–31B in more widely available PDAC xenograft models. For example, silvestrol (0.5 mg/kg, i.p., twice a week, from d10–d63) caused tumor growth arrest in MiaPaca-2 xenografts in nude mice by luciferase imaging, and ex vivo tumor weights and volumes (n = 5, P < 0.05; Supplementary Fig. S5G–S5I). TUNEL stains (24 hours after last dosing) indicated tumor cell apoptosis in silvestrol-treated tumors (Supplementary Fig. S5J and K). Silvestrol was well tolerated without weight loss or significant changes in blood counts (24 hours after last dosing; Supplementary Fig. S5L–S5O). [−]CR-1–31B showed similar tumor growth delay in this model (Supplementary Fig. S5P), and endpoint analysis (tumor >2,000 mm³) indicates a significant survival advantage in [−]CR-1–31B-treated MiaPaca-2 bearing animals (0.5 mg/kg, i.p., twice a week, n = 5/8, P < 0.0012; Supplementary Fig. S5Q).

Our findings provide new insight into the therapeutic utility and the mechanism of eIF4A inhibitor treatment. Using ribosome footprinting on human PDAC cells, we find that genes with long- and highly structured 5′UTRs depend on eIF4A for their translation. These include important KRAS signaling molecules such as RALA, RALGDS, RAF, RAC, PI3K, RALBP1, and MYC, whereas KRAS itself falls just outside our stringent statistical criteria and is still affected in protein studies. Consistent with our previous findings in T-cell leukemia, we notice that many eIF4A-dependent mRNAs harbor sequence elements that are predicted to fold into secondary, noncovalent GQ structures (2). Inserting GQ sequences confers eIF4A dependence in translation reporter and helicase assays. Conversely, removing them from the endogenous KRAS gene reduces the eIF4A requirement. However, it is not clear to what extent these sequences exist in a stably folded state in cells. For example, one RNA sequencing-based method suggested that GQ elements are largely unstructured in cells (50), another method (rG4-seq) indicates that they are structured, and a third, antibody-based method, also detects abundant GQ structures in cells (51, 52). Our study does not directly address this question, although consistent and unbiased identification of GQ sequences in eIF4A-dependent RNAs and functional evidence consistently support a role for GQ elements in eIF4A dependence. By contrast, ubiquitous AG-rich sequences that act as accessible sites for eIF4A binding (49) do not confer eIF4A dependence and may act to sequester eIF4A.

Ribosome profiling has emerged as the experimental standard to precisely map and measure ribosome occupancy across the transcriptome and provides a surrogate measure for mRNA translation into protein (53). The previous polysome fractionation method relied on manual separation of heavy and light polysome fractions from sucrose gradients and suffered low reproducibility. Highlighting these differences, our findings differ strongly from a recent polysome study on PDAC cells treated with the same eIF4A inhibitor–treated PDAC cells (31). The other study by Chan and colleagues ruled out MYC and KRAS, and instead implicated a number of redox enzymes and central carbon metabolism genes, however, using conventional statistical cutoffs (q < 0.05), none of named genes showed a significant difference in the Chan and colleagues study as well as in our study (31). Relying on these data, the study reports preferential eIF4A effects on genes with short and unstructured 5′UTRs that contradicts much prior work on this helicase (54, 55). On the other hand, we happily agree that eIF4A inhibition has promising single-agent activity against PDAC in vitro and in vivo (31). A prior study on a different rocaglamide (Infinity's Compound 76) showed toxicity and low efficacy and these effects may be specific to that inhibitor (56). Our findings show in vivo efficacy at tolerable dose levels for the CR-1–31B compound in mouse and human models of PDAC and indicate a therapeutic window in treating this deadly cancer with this new and exciting class of drugs.

P.B. Romesser reports grants and personal fees from EMD Serono, as well as personal fees and non-financial support from Elekta outside the submitted work. Y. Fukase reports a patent for WO2019161345A1 issued. G. Yang reports grants from NCI during the conduct of the study, as well as reports intellectual property rights in Memorial Sloan Kettering Cancer Center and Angiogenex. O. Ouerfelli reports intellectual property rights in Jazz Pharmaceutical, MSKCC, Johnson & Johnson, and Y-mAbs Therapeutics, as well as reports ownership, intellectual property rights, and provision of services in Angiogenex. S.D. Leach reports membership of the scientific advisory board of NYBO Therapeutics and is a co-founder of Episteme Prognostics. Z. Ouyang reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

K. Singh: Conceptualization, resources, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. J. Lin: Data curation, software, investigation, methodology. N. Lecomte: Investigation, methodology. P. Mohan: Validation, methodology. A. Gokce: Formal analysis. V.R. Sanghvi: Methodology. M. Jiang: Validation, methodology. O. Grbovic-Huezo: Investigation, methodology. A. Burčul: Data curation, software, formal analysis. S.G. Stark: Data curation, software, formal analysis. P.B. Romesser: Methodology. Q. Chang: Methodology. J.P. Melchor: Writing–review and editing. R.K. Beyer: Investigation, methodology. M. Duggan: Investigation. Y. Fukase: Methodology. G. Yang: Methodology. O. Ouerfelli: Methodology. A. Viale: Methodology. E. de Stanchina: Investigation, methodology. A.W. Stamford: Investigation, methodology. P.T. Meinke: Investigation, methodology. G. Rätsch: Resources, data curation, supervision. S.D. Leach: Resources, supervision, methodology. Z. Ouyang: Data curation, software, formal analysis, supervision. H.-G. Wendel: Conceptualization, resources, supervision, writing–original draft.

H.G. Wendel is supported by NIH/NCI grants R01CA183876-03, R01 CA207217-04, R01CA190384-05, P50 CA217694-02, P50 CA192937-04. H.G. Wendel is also supported by Lymphoma Research Foundation; William H. Goodwin and Alice Goodwin and the Commonwealth Foundation for Cancer Research; the Center for Experimental Therapeutics at MSKCC; the Starr Cancer Consortium; the Geoffrey Beene Cancer Research Center; David Rubenstein Center for Pancreatic Cancer; Druckenmiller Center for Lung Cancer Research; a Leukemia and Lymphoma Society (LLS) SPORE grant; and the MSKCC Core Grant (P30 CA008748). H.G. Wendel is a scholar of the Leukemia Lymphoma Society. K. Singh is supported by the Pancreatic Cancer Action Network. G. Rätsch is supported by MSK Core funding. S.D. Leach is supported by NIH grant R01CA204228. E. de Stanchina is supported by NIH U54 OD020355-01. P.B. Romesser is supported by a K12 CA184746, G. Rätsch is supported by the core funding of ETH Zürich. Z. Ouyang is supported by NIH R35 GM124998. We acknowledge the use of the Integrated Genomics Operation Core (funded by CCSG, P30 CA08748, Cycle for Survival, Marie-Josée and Henry R. Kravis Center for Molecular Oncology, the Mouse Pharmacology and Organic Synthesis cores funded by the NCI CCSG, P30 CA08748. We thank G. Sukenick, J. McCauley S. Kargman (Tri-I-TDI), T. Tammela (MSK) for assistance with various part of the study. We thank Dr. Christopher S. Fraser for sharing the human eIF4AI (406 amino acids), eIF4H isoform 2 (228 amino acids), and eIF4GΔ (amino acids 682–1166 from eIF4GI) expression plasmids.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.