Pharmacologic Inhibition of EIF4A Blocks NRF2 Synthesis to Prevent Osteosarcoma Metastasis Open Access

Osteosarcoma (OS) is characterized by high genomic complexity and clinical heterogeneity (1). Although aggressive surgical resection of the primary tumor and multiagent MAP (methotrexate, adriamycin, and cisplatin) chemotherapy has improved relapse-free survival in up to 65% for patients with localized OS, treatment options for patients with metastatic disease are extremely limited (1). Indeed, many patients exhibit resistance to therapy and develop metastasis, particularly to the lungs. Although a subset of patients with resectable lung metastases can benefit from thoracotomy and metastasectomy (2), the overall survival for metastatic OS remains at ∼20% and has not improved for four decades (3). Defining new therapies that prevent metastasis or targeting tumor cells at the metastatic site is therefore a critical unmet need in OS. However, we lack a detailed mechanistic understanding of what connects intratumor heterogeneity, metastatic capacity, and therapy resistance, as well as the role of specific oncogenic alterations in these processes.

Survival of tumor cells during the metastatic cascade, such as following intensive chemotherapy, is likely due to both preexisting genetic alterations that increase fitness and epigenetic plasticity encoded in a subset of cells within the primary tumor. However, although copy number gains, deletions, and structural rearrangement of specific oncogenic drivers are common and may define some features that facilitate OS metastasis, no studies have clearly defined the role of such drivers in OS metastasis (2). For example, MYC amplification and/or increased expression may portend a worse prognosis (3), yet the specific role of MYC in driving metastatic cell survival or chemotherapy resistance is not well understood. It has been known for decades that tumor metastasis is a highly inefficient process, with only a very small percentage of primary tumor cells retaining metastatic competency (4). A prevailing view is that disseminated tumor cells must successfully adapt to diverse microenvironmental stresses to survive, including local oxygen and nutrient deprivation, anoikis and shear stresses in the bloodstream, immune attack (5), nitrosative (6) and oxidative stress (7), and endoplasmic reticulum stress (8), each of which can potentially cull pre-metastatic cells. For example, cancer cells can experience high oxidative stress from extrinsic sources such as microvascular endothelial cells to diminish metastatic capacity (6), and high levels of cellular reactive oxygen species (ROS) block melanoma metastasis (7).

Although the majority of tumor cells that reach the lungs, the major site of OS metastasis, do not survive (9–11), a small subpopulation seems to adapt, survive, and proliferate in the lung microenvironment, as demonstrated in OS lung seeding studies (12). What exactly characterizes such persisting cells at the molecular level remains largely unknown. Emerging literature suggests that reprogramming of mRNA translation plays a major role in adaptation to adverse stress of the tumor microenvironment and is an important contributor to metastatic fitness (13–16). For example, we and others have shown that the eEF2K translation elongation factor, which inhibits translation elongation and energy-demanding protein synthesis, protects tumor cells from nutrient deprivation and is associated with aggressive features (17, 18). Translation regulation is broadly divided into three phases, namely translation initiation, elongation, and termination (13), and small-molecule inhibitors targeting these processes are now in development, including various tool compounds and clinical drugs (15). In OS, highly metastatic OS cells can more efficiently initiate the translation of mRNAs with complex 5ʹ untranslated regions (5ʹ-UTR) during metastatic colonization of the lungs compared with low metastatic cells (19). Because the features of 5ʹ-UTR strongly influence translation initiation (20), we wondered whether members of the eukaryotic initiation factor 4F (EIF4F) complex, namely EIF4G, EIF4A, and EIF4E, might also influence the translation of metastasis-associated mRNA transcripts, given that deregulation of these factors is implicated in many different cancer types (15). Phosphorylation of EIF4E through TGFβ-induced MNK1 activity correlates with enhanced translation of metastasis-associated mRNAs encoding MMP9 and SNAIL (21). Increased expression of EIF4G1 promotes tumor cell invasion and tumor embolization in inflammatory breast cancer (22), and EIF4A1 overexpression is associated with increased invasiveness and metastatic capacity in pancreatic cancer cells (23). Although previous RNA sequencing (RNA-seq) data show that various eukaryotic initiation factors are upregulated in high versus low metastatic OS cells (24), a functional link between translational control and metastatic propensity in OS has yet to be established.

In the present study, we explored whether deregulated mRNA translation drives rapid adaptive translational reprogramming to support OS cell dissemination during the metastatic cascade. Our initial small-molecule screen of a series of translation inhibitor molecules targeting EIF4A, EIF4E, EIF4G, and EEF2K identified the specific EIF4A1/2 inhibitor, CR-1-31B, as having potent cytotoxic and antimetastatic activities in metastatic OS models. CR-1-31B and its natural product base, silvestrol, are members of the rocaglate class of small-molecule compounds, which contain a common cyclopenta[b]benzofuran core that clamps the EIF4A helicase onto polypurine (A/G)–rich sequences (25, 26). A clinical-grade rocaglate, eFT226 (zotatifin; eFFECTOR Therapeutics, Inc.), a potent EIF4A1 inhibitor, is currently in phase 1 to 2 clinical trials for adult malignancies (27, 28). We found that CR-1-31B enhanced apoptosis in OS cells subjected to oxidative stress in vitro and blocked primary tumor growth and lung metastasis in OS mouse models. CR-1-31B reversed the synthesis of oxidative stress–induced cytoprotective factors under oxidative stress, including NRF2, the master regulator of the antioxidant response (29), which has been recently implicated in the survival of drug-tolerant persister cells in human tumors (30). Mechanistically, CR-1-31B strongly inhibited translation of NRF2-encoding NFE2L2 transcripts and the expression of downstream NRF2 transcriptional targets. Effects of CR-1-31B were phenocopied by genetic inactivation of NRF2 and by a second rocaglate, namely the clinical-grade molecule, zotatifin (eFT226), currently in clinical trials for other solid tumors. Together, our studies identify small-molecule EIF4A1 inhibitors as promising antimetastatic agents that potently block OS tumor growth and progression of metastatic OS.

The eGFP-expressing OS cell lines MG63.3 (RRID: CVCL_WL01), MG63 (RRID: CVCL_0426), MNNG, and HOS were obtained from Dr. Rosandra Kaplan (Pediatric Oncology Branch, NIH, Bethesda, Maryland, USA); patient-derived xenograft (PDX) cell line models OS742 (RRID: CVCL_C8FY), OS742-ZsGreen/Luc, and OS384 (RRID: CVCL_C8FU) were kindly provided by coauthor Dr. Alejandro Sweet-Cordero (University of California San Francisco, San Francisco, California, USA); canine OS cell lines OSCA29 (RRID: CVCL_L384) and OSCA78 (RRID: CVCL_L404) were obtained from Dr. Jaime Modiano (University of Minnesota, St. Paul, Minnesota, USA); and murine OS F420 cells were from Dr. Jason Yustein (Emory University, Atlanta, Georgia, USA). The OS PDX models and cell line models from A. Sweet-Cordero. were acquired with patient informed written consent or parental informed written consent (if patients were younger than the age of consent), according to the Belmont Report guidelines and local research ethics board, as previously described by Sayles and colleagues (31). Human OS tissue microarrays (TMA) from the Children’s Oncology Group and OS PDX were used in accordance with recognized ethical guidelines and approved by the local University of British Columbia Human Research Ethics Board under local protocols #H20-00908 and #H22-03018. All OS cell lines and PDX cell lines were tested for Mycoplasma using a PCR method previously described (32). The OS cells were grown in Dulbecco’s Modified Essential Media supplemented with 10% FBS and 1× penicillin/streptomycin/L-glutamine (Thermo Fisher Scientific, Cat# 10378016). The PDX cell line models were grown in Dulbecco’s Modified Essential Media supplemented with 10% bovine growth serum (Fisher Scientific, Cat# SH3054103) and 1× penicillin/streptomycin/L-glutamine. Ex vivo lung slices were grown in B-media (33). Several of the small-molecule inhibitors (SMIs) used in this study were obtained from MedChemExpress: SBI-756 (Cat# HY-19560), eFT508 (Cat# HY-100022), and CR-1-3-1B (Cat# HY-136453). Silvestrol was obtained from the Developmental Therapeutics Program (https://dtp.cancer.gov/; RRID: SCR_003057) under the Division of Cancer Treatment and Diagnosis, NCI (NIH). The eEF2K inhibitor, A-484954 (Cat# 324516), was obtained from MilliporeSigma. Fluorescent indicator dyes such as SYTOX Orange and CellRox Orange/Deep Red were purchased from Thermo Fisher Scientific. Tert-butylhydroquinone (tBHQ) and hydrogen peroxide (H₂O₂) were purchased from Sigma-Aldrich.

Several OS- and bone-related datasets were accessed on the R2 Platform (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi), and log₂-transformed expression levels for patients with OS (Buddingh cohort), OS PDX samples (Dai cohort), normal bone marrow cells (Ambros cohort), along with RNA-seq data were made available from A. Sweet-Cordero’s group, including patients with OS, PDX models, and cell lines downloaded to construct heatmaps. For each dataset, outliers' genes that had zero counts were removed from each dataset prior to Z-scoring. The six Z-scored datasets were merged to construct a heatmap of the gene set listed above that was common in all datasets. The transcript levels of EIF4A1 in various adult and pediatric cancer cell lines were obtained from online Cancer Dependency Map Portal (RRID: SCR_017655). The survival data from patients with high or low abundance of EIF4A1 were obtained from the Gene Expression Profiling Interactive Analysis (RRID: SCR_018294) and were compared with each other via the log-rank test.

The use of patient TMAs was approved by the local University of British Columbia Research Ethics Board, under the human ethics protocol #H20-00908. Patients' written informed consent was obtained for the use of archived material for research purposes prior to the creation of the TMAs at their respective institutions. A TMA containing normal pediatric tissue (tissue from children diagnosed with non–cancer-related pathologies) was obtained from the Research Institute at Nationwide Children’s Hospital (Columbus). For the normal tissue TMA construction, formalin-fixed, paraffin-embedded control tissue blocks from 48 patients were selected for use as donor blocks. A total of 138 cores were used, and the tissue types represented are as follows: aorta, adipose, bladder, breast, cartilage, cervix, colon, connective tissue, diaphragm, epididymis, heart, kidneys, liver, lungs, mesentery, ovary, pancreas, small intestine, skeletal muscle, smooth muscle, spleen, testes, thymus, thyroid, tonsil, trachea, and skin. For the pediatric tumor TMA (Children’s Oncology Group), 21 OS cases were included. Hematoxylin and eosin (H&E)–stained slides from these tissue blocks were annotated for desired areas, and the slides were then used as guides for core selection. TMAs were constructed on a Beecher manual tissue arrayer I. Processing of slides of formalin-fixed, paraffin-embedded TMA sections were as previously described (34, 35). Briefly, TMA sections were probed with EIF4A1 antibody (Abcam #ab31217) at 1:2,000 for 1 hour. Slides were then incubated with secondary antibodies (The Jackson Laboratory) with 1:500 dilution. Intensity scoring was performed on a common four-point scale: 0, no staining; 1, low but detectable staining; 2, clearly positive staining; and 3, strong expression. Expression was quantified as an H-score: the product of staining intensity × percentage of stained cells.

All animal experiments described herein have been approved by the University of British Columbia Animal Care Committee (protocol# A21-0209). Paraosseous primary leg tumors were induced in 6- to 8-week-old female NOD/SCID mice (RRID: IMSR_ARC:NODSCID, The Jackson Laboratory) via intramuscular injection of 1 × 10⁶ MG63.3 or MNNG cells (in 25 μL volume) into the right gastrocnemius muscle. Mice were randomized to either (i) vehicle group: 5.2% polyethylene glycol—400/5.2% Tween-80 in 0.9% sterile saline or (ii) CR-1-31B treatment group: at a dose of 0.2 mg/kg, administered via i.p. injection, 1× injection per day for 15 consecutive days. Vehicle and drug treatments started when their primary tumors reached a diameter of 7 to 9 mm. Experimental endpoints occurred when primary tumors reached 17 mm in diameter or 60 days had elapsed. Humane endpoints are defined by a scoring system of several clinical parameters including the presence of tumor ulcerations, body weight, hydration, breathing rate, appearance, and activity level.

To assess the effects of intervention on the development of lung metastases, 3 to 5 × 10⁵ MG63.3 or MNNG cells (in 100 μL sterile saline) were injected into the tail vein. For CR-1-31B studies, mice were randomized to vehicle or CR-1-31B treatment groups as previously described. For dox-shRNA knockdown (KD) studies, mice were fed either with regular chow or with dox chow (625 mg/kg; Envigo, Cat# TD.08541) 1 week prior to injecting MG63.3 cells with dox-inducible control nontargeting shRNAs, and shRNA#1 and shRNA#2 against NRF2. Experimental endpoint was based on a predetermined time point when animal subjects were euthanized at 3 to 4 weeks, after which the lungs were harvested for histological analysis.

The ex vivo pulmonary metastasis assay (PuMA) was used to study the effects of intervention on lung colonization of murine lung tissue; methods are previously described (8, 33, 36). Briefly, 6- to 8-week-old NOD/SCID mice were injected with 3 × 10⁵ to 1 × 10⁶ (in 100 μL nonpyrogenic sterile saline) MG63.3 or MNNG cells via tail vein intravenous injection. The lungs were processed as previously described (5, 6). Eight to ten lung slices were used in each experimental condition. For CR-1-31B or eFT226 drug studies, media were refreshed daily. Image acquisition and analysis of PuMA micrographs were done as previously described (6). Then, 8 to 10 lung slices were quantified for each condition. Each PuMA experiment was repeated at least two times; representative micrograph images and results are shown in the figures.

The percent lung tumor burden for 8 to 10 lung slices per condition or lung tissue sections per in vivo experimental group was calculated and compared by statistical tests.

To assess changes in global translation rates in response to CR-1-31B treatment, MG63.3, MNNG, and OS742 cells were incubated for 24 hours with DMSO, 2, 4, 8 nmol/L CR-1-31B, followed by a 1-hour treatment with 10 μmol/L cycloheximide (CHX; Sigma-Aldrich, Cat# 01810). For the last 15 minutes of treatment, cells were treated with 0.3 μg/mL puromycin for 15 minutes. For Western blotting, cells were harvested and 20 μg of whole cell lysates was electrophoresed in a 12% polyacrylamide gel. Proteins were transferred to a nitrocellulose membrane and immunoblotted with anti-puromycin antibodies (1:1,000; Kerafast, Cat# EQ0001, RRID: AB_2620162). For semiquantitative immunofluorescence staining of puromycilated proteins in OS cells treated with CR-1-31B, ∼2 × 10⁴ OS cells were seeded in an 8-well chamber slide (EMD Millipore, Cat# PEZGS0816). OS cells were exposed to the same previous experimental conditions and fixed in 3.7% paraformaldehyde in PBS for 12 minutes and permeabilized with PBS with 0.05% Triton X-100. Slides were incubated with anti-puromycin antibodies directly conjugated to Alexa Fluor 647 (1:100; Millipore, Cat# MABE343-AF647, RRID: AB_2736876) overnight at 4°C. Coverslips were placed on slides using VECTASHIELD with DAPI (VectorLabs, Cat# H-1200). For image acquisition, images were acquired using a Zeiss Airyscan 800 laser scanning microscope and Plan Apochromat 20×/0.8 M27 objective lens. For semiquantitative analysis of puromycilated proteins (Alexa 647 fluorescence) using ImageJ (RRID: SCR_003070), see the “Cell Rox Staining and Live-Cell Imaging” section in the Supplementary Data.

Adherent cell cultures for polysome analysis were pretreated for 10 minutes with CHX (100 µg/mL final) at 37°C. Media containing CHX were removed, and cells were trypsin harvested prior to snap-freezing and stored at −80°C until additional processing. Sucrose gradients (10%–34%–45%) were prepared in ultra-clear ultracentrifuge tubes (4 mL, Cat: 344062, Beckman) using a stepwise approach. Then, 10%, 34%, and 45% (w/v) solutions were prepared in polysome base buffer [20 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), pH 7.3; 150 mmol/L KCl; 5 mmol/L MgCl₂]. For cell lysis, frozen cell pellets were reconstituted in an initial polysome lysis buffer [20 mmol/L HEPES, pH 7.3; 150 mmol/L KCl; 5 mmol/L MgCl₂; 0.5 cOmplete protease inhibitor cocktail, EDTA free (Sigma-Aldrich, Cat# 11836170001); 0.5% (v/v) NP-40; 0.5% (w/v) deoxycholate; 10 U TURBO DNase (Thermo Fisher Scientific, Cat# AM2239); and 100 μg/mL CHX], and incubated on ice for 5 minutes. For total RNA analysis, a volume equivalent to 10% (v/v) of this initial lysate was diluted in a separate tube with three parts of TRIzol (Zymo Research, Cat# R2063) per one volume of sample. For total proteome analysis, a volume equivalent to 30% (v/v) of this initial lysate was transferred to a separate tube. Total RNA and proteome samples were further processed as described below. The remaining material was centrifuged for 5 minutes at 5,000 g at 6°C. The supernatant was recovered and loaded onto a gradient prepared 24 hours before and centrifuged for 2 hours at 34,000 rpm in the SW 60 Ti rotor at 6°C. Centrifuged polysome samples were fractionated using the Biocomp fractionation station equipped with a Triax UV cell (absorbance at 254 nm) using a speed setting of 0.2 and collection of 48 total fractions. With this configuration, we ended up with a total of eight fractions covering the region beyond the 80S peak up to the signal drop after the heaviest polysomes. The first four fractions, up to approximately the end of the n = 3 peak, were retained for pooling as “light” and the next four as “heavy.”

Total RNA was extracted using the Direct-zol RNA MicroPrep Kit (Zymo Research, Cat# R2063) with the optional on-column DNaseI digest, according to the manufacturer’s instructions. For total RNA, the TRIzol diluted samples were processed directly using the Direct-zol kit. Extracted RNA was prepared for RNA-seq using the NEBNext Ultra II Directional RNA Library Preparation Kit with the NEBNext rRNA Depletion kit (New England Biolabs, Cat# E6310L) according to the manufacturer’s instructions. Prepared samples were dual-barcode multiplexed during PCR for sequencing on NovaSeq hardware to a minimum depth of 50 million paired reads per sample. The resulting FASTQ files were processed using BBDuk (ktrim = r k = 23 mink = 11 hdist = 1 tpe tbo) to remove any adapter sequences. For quantification, BBDuk-processed files were processed using Salmon (version 1.5.2; ref. 37) using selective alignment (–validateMappings) with a decoy-aware transcriptome based on the full genome (GRCh38, Gencode release 38) and GC bias correction (–gcBias). The quantification data were further parsed in R using the tximport (RRID: SCR_016752) and DESeq2 (RRID: SCR_000154) packages to facilitate comparisons between sample sets. The gene set enrichment analyses (GSEA) were obtained in R using the clusterprofiler (RRID: SCR_016884) package. Cutoff P values were set at 1 to include all pathways. GSEA plots were created in R using the gseaplot2 function of the enrichplot package. An adjusted P value of < 0.05 was considered significant. Bubble charts of pathways relevant to oxidative stress, oxidoreductase activity, NRF2/antioxidant response elements (ARE) signaling, endoplasmic reticulum stress, and hypoxia and their associated changes in normalized enhancement score, gene set member number, and adjusted P values were constructed using Visomics online tool (http://bioinfo.ihb.ac.cn/visomics).

The total proteome material from the previous protocol was first digested with a mixture of 1 unit of Benzonase Nuclease and 1 µg of RNaseA for 10 minutes at 21°C. After cell disruption was completed through addition of a 4× lysis buffer [400 mmol/L HEPES, pH 7.3; 8% (v/v) SDS; 40 mmol/L dithiothreitol] to a final 1× concentration, the mixtures were incubated at 95°C for 5 minutes. After heating, the samples were left at 21°C for 10 minutes prior to the addition of chloroacetamide to a final concentration of 40 mmol/L and a further incubation at 21°C for 30 minutes. At this stage, protein samples were processed for data-independent acquisition mass spectrometry (DIA-MS). For DIA-MS analysis, protein lysates (∼20 μg) were initially processed using a modified version of the SP3 protocol (38, 39). DIA-MS raw data were demultiplexed using msConvert, a command line utility program (ProteoWizard, peakPicking = “vendor”, demultiplex = overlap only, 10 ppm, SIM as spectra), and processed using DIA-NN (version 1.8.1, RRID: SCR_022865). The resulting data were analyzed in R using the “iq” package to generate estimates of protein abundance and “DEqMS” to perform differential abundance calculations between conditions. Proteins that were significantly (P < 0.05) upregulated at least twofold were used to construct volcano plots for single-agent CR-1-31B treatment. For proteomic studies examining changes in response to CR-1-31B and tBHQ treatment, a heatmap was constructed with a subset of proteins that met the following criteria: (i) upregulated by 75 μmol/L tBHQ treatment alone and (ii) not increased by 4 nmol/L CR-1-31B treatment alone and not increased by combination conditions.

The day before the start point for proliferation/drug studies, 1 × 10³ MG63.3 cells, 1 × 10³ MNNG cells, and 4 × 10³ OS742 cells in 100 μL were seeded in Nunc Edge 96-well plates (Thermo Fisher Scientific, Cat# 167425). The next day 2× drug concentration in 100 μL was added to each well, and the plate placed in the Incucyte S3 (Sartorius) for 96 hours, where images and percentage confluency measurements were acquired every 4 hours. Percent confluency values were normalized to the percent confluency at the starting point (t = 0) to obtain fold change in percent confluency over time.

To generate reporter OS cell lines (MG63.3, MNNG) for the antioxidant response, a reporter system based on pGL24.6 vector (Promega, Cat# E8441) was constructed whereby a reporter protein (mCherry) was driven by two tandem antioxidant response elements (GTGACAAAGCA; ref. 40), which were spliced upstream of 5ʹ of a minimal reporter driving the expression of mCherry with a destabilizing 3ʹ-proline/glutamic acid/serine/threonine (PEST) sequence. OS cells were transfected with 6 μg plasmid and transfected with Lipofectamine 2000 (Thermo Fisher Scientific, Cat# 11668019) according to the manufacturer’s instructions. A stable cell line was established by selection with 300 μg/mL hygromycin B (Thermo Fisher Scientific, Cat# 10687010) for 2 weeks, pooling of resistant colonies, followed by fluorescent activated cell sorting for positive mCherry cells following tBHQ treatment. For semiquantitative analysis of ARE-mCherry fluorescence using ImageJ (RRID: SCR_003070), see the previously mentioned “Cell Rox Staining and Live-Cell Imaging” section in the Supplementary Data.

Total RNA from cell pellets were isolated and purified using the RNeasy Plus Universal Kit (QIAGEN) according to the manufacturer’s instructions. RNA yield and purity were assessed on a NanoDrop 1000 using the 260/280 method. To make cDNA using the High-Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific, Cat# 4368813), 1 µg of RNA was used as the template for reverse transcription. Then, 50 ng of the cDNA template was used in a RT-PCR using the Fast SYBR Green Master Mix (Thermo Fisher Scientific, Cat# 4385612) using the following primers (IDT): NFE2L2: forward: 5ʹ-CAG​ATG​CCA​CAG​TCA​ACA​CA-3ʹ, reverse: 5ʹ-GGG​CTC​AGC​TAT​GAA​AGC​A-3ʹ; MYC: forward: 5ʹ-TCG​GAT​TCT​CTG​CTC​TCC​TC-3ʹ, reverse: 5ʹ-TCG​GTT​GTT​GCT​GAT​CTG​TC-3ʹ; GAPDH: forward: 5ʹ-ACC​CAC​TCC​TCC​ACC​TTT​GA-3ʹ, reverse: 5ʹ-CTG​TTG​CTG​TAG​CCA​AAT​TCG​T-3ʹ. NFE2L2 transcript levels were normalized to GAPDH, and changes were expressed as fold changes.

OS cells were treated with 100 μmol/L CHX at 120, 60, 30, and 0 minutes after these were treated with 75 μmol/L tBHQ to induce NRF2 expression. They were harvested after CHX treatment time points, and whole cell lysates were isolated using RIPA buffer. Then, 20 µg of cell lysates was electrophoresed on a 12% polyacrylamide gel, after which proteins were transferred to nitrocellulose membranes. The membranes were immunoblotted for NRF2 and β-actin.

To study newly synthesized proteins during oxidative stress, 1 × 10⁶ MG63.3 cells were incubated with methionine-free DMEM for 1 hour prior to refreshing with the same media but supplemented with 300 μmol/L azidohomoalanine (AHA) and co-treated with 75 μmol/L tBHQ for 2 hours. For combination conditions (75 μmol/L tBHQ + 2 nmol/L or 4 nmol/L CR-1-31B), OS cells were pretreated with 2 and 4 nmol/L CR-1-31B for 24 hours prior to labeling with AHA and 2 hours of treatment with 75 μmol/L tBHQ. Twenty-four hours of incubation with 10 μmol/L CHX was a positive control for translation inhibition. After treatment conditions, the cells were harvested; whole cell lysates were processed by the Click-iT Protein Buffer Kit (Thermo Fisher Scientific, Cat# C10276) according to the manufacturer’s instructions; and AHA-modified proteins were detected by chemoligation to biotin alkynes (Thermo Fisher Scientific, Cat# B10185). Protein samples were enriched for biotinylated proteins by incubation with 50 μL of M-280 streptavidin-coated magnetic Dynabeads (Thermo Fisher Scientific, Cat# 11205D) according to the manufacturer's instructions. Biotinylated protein–rich samples were denatured at 95°C for 5 minutes in protein loading buffer and were probed for NRF2 and β-actin via the Western blotting procedure described below.

Protein whole cell lysates were prepared using RIPA cell lysis and extraction buffer. Protein concentration for each sample was determined by the bicinchoninic acid assay (Thermo Fisher Scientific, Cat# 23227) per the manufacturer’s instructions. About 20 µg of protein was loaded in 12% polyacrylamide mini-gel (Thermo Fisher Scientific, Cat# NP0321BOX) and electrophoresed for 120 minutes at 120 V and 40 mA. Proteins were then transferred to a nitrocellulose membrane overnight at constant 40 V. Prior to the addition of primary antibodies, membranes were blocked in Tris-buffered saline with 0.1% Tween-20 and 5% milk powder for 1 hour. Antibodies used in this study include anti-NRF2 (1:1,000; Abcam, Cat# ab62352, RRID: AB_944418), anti-NQO1 (1:1,000; Abcam, Cat# ab34173, RRID: AB_2251526), anti-mCherry (1:1,000; Thermo Fisher Scientific, Cat# M11217, RRID: AB_2536611), anti-MYC (1:1,000; Cell Signaling Technology, Cat# 9402, RRID: AB_2151827), anti-cyclin D1 (1:1,000; Cell Signaling Technology, Cat# 2922, RRID: AB_2228523), anti-biotin (1:1,000; Abcam, Cat# ab53494, RRID: AB_867860), anti-total PARP (Cell Signaling Technology, Cat# 9542, RRID: AB_2160739), anti-β-tubulin (1:1,000; Cell Signaling Technology, Cat# 15115, RRID: AB_2798712), anti-vinculin (Cell Signaling Technology, Cat# 13901, RRID: AB_2728768), and mouse anti-β-actin (1:1,000; Sigma-Aldrich, Cat# A1978), which were incubated overnight at 4°C. Membranes were washed with Tris-buffered saline with 0.1% Tween-20 and incubated with either goat anti-rabbit (Cat# 111-035-003; RRID: AB_2313567) or goat anti-mouse IgG (Cat# 115-035-003; RRID: AB_621842) conjugated to horseradish peroxidase. Protein bands were developed using the SuperSignal West Pico Chemiluminescent Substrate kit (Thermo Fisher Scientific, Cat# 34077) per the manufacturer’s instructions. Protein bands were imaged and digitized using an ImageQuant LAS 4000 imaging system and ImageQuant TL software (GE Healthcare; RRID: SCR_018374). Densitometry of protein bands was performed using ImageJ software (RRID: SCR_003070).

The Graph Prism (RRID: SCR_002798) software package was used for statistical analyses. Numerical datasets were tested for normality using the D’Agostino–Pearson test. When comparing two groups of normal data, an unpaired t test was used. In the case of non-normal data, the Mann–Whitney U test was used. When comparing three groups or more of normal data, ANOVA and post hoc Tukey multiple comparisons were used to compare the means of each group. When comparing three groups or more of non-normal data, the Kruskal–Wallis test and Dunn multiple comparisons test were used to compare the medians of the groups. For animal survival studies, Kaplan–Meier curves were constructed and experimental groups compared by the log-rank test. Significance values are indicated on the graphs.

The proteomics data generated in the current study have been deposited to the ProteomeXchange consortium via the PRIDE (RRID: SCR_012052) partner repository with the following dataset identifier: PXD047420. The datasets include (i) single-agent treatment of MG63.3 cells with DMSO, 2, 4, 8 nmol/L CR-1-31B; (ii) MG63.3 cells treated with DMSO, 75 μmol/L tBHQ for 1 hour, 4 nmol/L CR-1-31B treated for 24 hours, and combination of 75 μmol/L tBHQ for 1 hour, and 4 nmol/L CR-1-31B treated for 24 hours. The repository contains raw data, search results, detailed MS acquisition method parameters, and tandem mass tag (TMT) labeling schemes for all proteomic analyses performed in this work. Values for all data points in the graphs are reported in the Supporting Data Values file. All other relevant data are available from the corresponding author on reasonable request.

See the Supplementary Data file for an expanded methodology section.

All schematic diagrams were made using the online tool BioRender (RRID:SCR_018361).

Recent studies point to a role for altered mRNA translation in tumor cell plasticity and stress adaptation, leading to increased capacity for aggressive behavior, including metastasis (21, 41, 42). To assess deregulated translational control as a potential driver in metastatic OS, we first analyzed publicly available RNA-seq datasets for expression of known mRNA translation factors, including EIF2A, EIF3, EIF4A1, EIF4A2, EIF4A3, EIF4B, EIF4E, EIF4E2, EIF4E3, EIF4EBP1, EIF4G1, EIF4G2, EIF4G3, EEF2, and EEF2K. Comparison of the OS Buddingh patient cohort (GEO ID: GSE21257) and the Dai OS PDX cohort (GEO ID: GSE124768) to normal bone marrow samples of patients (Ambros cohort, GEO: GSE94035) revealed cross-cohort elevation of EIF4A1, EEF2, and EIF4G1/2 mRNAs in patients with OS and PDX specimens compared with normal bone marrow samples (Fig. 1A). Of the translation factors analyzed, the most consistently upregulated transcript was EIF4A1, and OS was among the highest EIF4A1 expressors in the Dependency Map database (Fig. 1B). Elevated EIF4A1 expression was strongly predictive of poor patient survival in patients with sarcoma (Gene Expression Profiling Interactive Analysis 2/The Cancer Genome Atlas database; Fig. 1C). We next assessed EIF4A1 protein expression by IHC in a tissue microarray containing OS tumors from 21 patients (obtained from the Children’s Oncology Group) and normal tissue controls (Fig. 1D). Tumor cell positivity for EIF4A1 (expressed as H-scores) was significantly higher in OS tumor tissue versus normal control tissue (Fig. 1E). Moreover, EIF4A1 protein expression was highly predictive of overall survival, whereby high-expressing tumors had significantly lower survival rates compared with medium-to-low expressors (Fig. 1F). Together, these data demonstrate that EIF4A1 expression is elevated in OS, and particularly in aggressive forms of OS.

Several promising agents are currently in development to target different phases of mRNA translation in human cancers (15). We therefore evaluated a panel of known experimental and clinical-grade small-molecule inhibitors targeting EIF4A, EIF4E, EIF4G, and EEF2K, for in vitro IC₅₀ values in OS (Supplementary Table S1). We included the EEF2K inhibitor A484954 even though EEF2K transcript levels were low in OS (Fig. 1A), given that EEF2K protects tumor cells from nutrient deprivation (17), particularly in MYC-driven tumors (18). From these screens, the rocaglate-based compounds, silvestrol and CR-1-31B, which act by clamping the EIF4A1/2 helicase onto polypurine (A/G)–rich sequences (25, 26), each demonstrated potent in vitro cytotoxicity in the low nanomoles per liter range (4–8 nmol/L) in sulforhodamine B (SRB) assays of MG63.3 cells compared with much higher IC₅₀ values for the MNK1/2-EIF4E inhibitor eFT508, the EIF4G1/2 inhibitor SBI-756, and the EEF2K inhibitor A-484956 (Supplementary Table S1). Although silvestrol had lower IC₅₀ values than CR-1-31B, silvestrol has suboptimal drug-like properties and poor solubility and induced pulmonary toxicity in canine studies (43). Because CR-1-31B is well tolerated in vivo (44, 45) and demonstrates low nanomoles-per-liter potency in MG63.3 cells, we focused on this compound in further studies. Single-agent activity of CR-1-31B reduced cell viability and cell proliferation in a dose-dependent manner from 2 to 8 nmol/L across three independent human OS models, including MG63.3 and MNNG cells, and the OS742 OS PDX model (Supplementary Fig. S1A and S1B; refs. 31, 46). Low nanomoles-per-liter levels of antiproliferative activity of CR-1-31B were confirmed in an expanded panel of established human, mouse, canine, and PDX-derived OS cell line models, revealing potent activity across each species (Supplementary Table S2). Finally, CR-1-31B significantly reduced cell migration at 8 nmol/L and invasion through Matrigel at 4 to 8 nmol/L in human MG63.3 OS cells at 16 hours of treatment (Supplementary Fig. S1C and S1D).

To directly assess the in vivo efficacy of CR-1-31B, we used a paraosseous orthotopic OS model based on xenotransplantation of MG63.3 cells into the hind gastrocnemius muscle of NOD/SCID mice to form primary tumors adjacent to the femur, a well-characterized model for in vivo OS studies (47, 48). Once tumor sizes reached a diameter of 7 to 9 mm, the mice were dosed intraperitoneally with CR-1-31B at 0.2 mg/kg, a dose used in previous studies (44, 45), with one cycle of 15 days at 1× per day. CR-1-31B significantly inhibited primary tumor growth compared with vehicle alone, as was clearly evident by ∼day 35 to 40 after treatment (Fig. 2A). CR-1-31B was well tolerated in mice; no overt toxicities nor significant changes in body weights were observed after drug treatment (Fig. 2B). CR-1-31B treatment markedly increased survival by Kaplan–Meier analysis, with over half of the drug-treated group surviving to the 60-day endpoint, whereas there were no surviving mice in the control group (Fig. 2C). Notably, tumors regrew in the CR-1-31B group after treatment cessation, which was evident until the 60-day experimental endpoint. As no spontaneous lung metastases were observed in the MG63.3 cohort (data not shown), we next performed an experimental metastasis assay, whereby 3 × 10⁵ MG63.3 cells were injected into the tail vein of NOD/SCID mice. CR-1-31B treatment at 0.2 mg/kg starting at day 21 for 15 additional days strongly inhibited lung metastases, as shown by representative IHC staining for eGFP in mouse lung tumors from control versus CR-1-31B–treated groups in Fig. 2D, with quantification in Fig. 2E. Similar antitumor activity was observed in a second human OS model using MNNG cells. CR-1-31B significantly delayed tumor growth (although not as effectively as for MG63.3 cells; Supplementary Fig. S2A), with no toxicity or weight loss observed (Supplementary Fig. S2B), and CR-1-31B significantly increased survival times compared with controls (Supplementary Fig. S2C). In contrast to MG63.3, orthotopic MNNG xenografts formed spontaneous lung metastases, and the CR-1-31B–treated group demonstrated a highly significant reduction in lung metastasis by eGFP IHC in Fig. 2F, with quantification in Fig. 2G. Together, these data provide compelling evidence that CR-1-31B blocks OS tumor growth and progression in vivo.

To further explore the antimetastatic activity of CR-1-31B, we utilized an alternative method, namely the ex vivo PuMA model, which directly interrogates the capacity of tail vein–injected GFP-labeled tumor cells to survive within and colonize the lungs (33, 36) in a relevant three-dimensional (3D) microenvironment. Following injection, mouse lung sections can be maintained ex vivo for up to 28 days, as described previously (33, 36). We observed that by 14 days after injection of GFP-labeled tumor cells, outgrowth of MNNG cells in ex vivo lung tissues was dramatically reduced by CR-1-31B compared with vehicle alone (Figs. 2H-I), even at 2 nmol/L CR-1-31B, which only weakly reduced MG63.3 cell proliferation in vitro (Supplementary Fig. S1B). This suggests that the levels of CR-1-31B that are minimally lethal in vitro are not permissive for metastatic colonization of lung tissue in this model. PuMA GFP fluorescence imaging data were validated by H&E and IHC staining for eGFP in MNNG PuMA tissue sections (Supplementary Fig. S2D and S2E). Similar data were observed using MG63.3 cells (Supplementary Fig. S2F–S2I), whereby CR-1-31B almost completely blocked the ability of metastatic OS cells to colonize and grow in the lungs. Taken together, these data demonstrate the potent activity of CR-1-31B in reducing primary OS growth and particularly lung colonization and metastatic capacity.

We next wished to determine the mechanism by which CR-1-31B exerts its effects in OS. Given that rocaglates are translation inhibitors, we first confirmed that CR-1-31B blocks translation in OS cells using SUrface SEnsing of Translation assays, which assess rates of mRNA translation by monitoring puromycin incorporation into newly synthesized proteins (49). CR-1-31B treatment led to dose-dependent in vitro reductions in MG63.3 mRNA translation, as shown by Western blotting (Fig. 3A) and fluorescence immunostaining for puromycin (Fig. 3B). Similar findings were observed in MNNG and OS742 cells (Supplementary Fig. S3A–S3F). Notably, a global block in translation was observed only at 8 nmol/L CR-1-31B or above for MG63.3 and MNNG cells and at 16 nmol/L or above for OS742 cells (Supplementary Fig. S3D–S3F). Therefore, CR-1-31B blocks proliferation and viability of OS cells in vitro at concentrations below those that inhibit global translation, pointing to selective effects on translation, consistent with the minimal in vivo toxicity observed. Moreover, using sucrose gradient fractionation and polysome profiling to catalog actively translating heavy transcripts (i.e., mRNAs bound by four or more ribosomes) versus light transcripts (i.e., binding ≤3 ribosomes) in MG63.3 cells, 4 nmol/L CR-1-31B increased the translationally inactive 80S peak compared with vehicle (Fig. 3C). Notably, CR-1-31B reduced protein levels of both cyclin D1 and c-MYC, known rocaglate targets (50, 51), in MG63.3, MNNG, and OS742 cells (Supplementary Fig. S4A and S4B), albeit at variably effective concentrations, ranging from 2 to 20 nmol/L for cyclin D1 and higher amounts for c-MYC, confirming that CR-1-31B inhibits translation of rocaglate-sensitive transcripts in metastatic OS cells. These studies support the notion that CR-1-31B selectively inhibits translation within specific pathways in OS to block growth and metastatic progression, rather than by affecting global translation.

We next used an unbiased multiomics approach to further uncover how rocaglates block OS tumor growth and progression. MG63.3 cells were treated with 4 nmol/L CR-1-31B or vehicle for 24 hours and total mRNA and proteins were isolated for RNA-seq and MS-based proteomics, respectively. Differential transcript levels detected by RNA-seq under these conditions are shown as a volcano plot in Fig. 3D. GSEA of transcript fold changes revealed several significantly different pathways between the two groups, as summarized in a bubble chart (Fig. 3E). Of the top pathways most significantly altered by CR-1-31B were ones linked to redox biology [e.g., oxidoreductase activity (OR)], oxidative stress (e.g., hypoxia, oxidative stress–induced senescence), and the NRF2 pathway (Biocarta: ARE-NRF2 pathway, WikiPathways NRF2-ARE regulation); oxidative stress–related GSEA plots are provided in Supplementary Fig. S5A. Identically treated cells were also subjected to TMT proteomics to catalog proteins specifically reduced by CR-1-31B, revealing significant global proteomic changes as shown as a volcano plot in Fig. 3F; full proteomic datasets are available via ProteomeXchange with identifier PXD047420 (http://proteomecentral.proteomexchange.org). Of these, GSEA demonstrated several protein categories overlapping with the above RNA-seq analysis, namely those linked to oxidoreductase activity (e.g., intramolecular OR, OR acting on NADPH, quinone or similar compound, and disulfide OR), as shown in the bubble chart of Fig. 3G; oxidative stress–related GSEA plots are provided in Supplementary Fig. S5B. Together, these results demonstrate the potent biological activity of CR-1-31B in metastatic OS cells and point to potential alterations in oxidative stress signaling by this agent.

Given the above RNA-seq and proteomic analyses, as well as literature that oxidative stress inhibits metastasis (6, 7, 35, 52), we further explored whether CR-1-31B modifies the oxidative stress response in OS cells. Treatment of OS cell line monolayers in vitro with the chemical oxidative stress inducer, tert-butyl hydroquinone (tBHQ), in combination with CR-1-31B synergistically augmented cell death over a range of 1 to 2 nmol/L CR-1-31B combined with 75 to 110 μmol/L tBHQ (Supplementary Fig. S6A) and reduced cell proliferation (Fig. 4A) across three different OS cell line models. Next, these studies were repeated in low-attachment 3D multicellular tumor spheroids, known to exhibit high ROS levels (53), as described (35, 54). Using SYTOX Orange staining to measure cell death, CR-1-31B co-treatment with tBHQ significantly increased cell death of OS spheroids compared with vehicle or single agents alone (Fig. 4B), and CR-1-31B and tBHQ co-treatment also increased PARP cleavage (Supplementary Fig. S6B). Moreover, CR-1-31B co-treatment with a second oxidative stress inducer, H₂O₂, also significantly enhanced levels of ROS compared with single agents alone, as measured using Cell Rox assays (Supplementary Fig. S6C). These data indicate that CR-1-31B sensitizes OS cells to redox stress and that CR-1-31B itself increases ROS levels in OS cells exposed to oxidative stress–inducing agents. To directly assess the effects of CR-1-31B on the oxidative stress response, we constructed a genetic fluorescence-based mCherry plasmid reporter to monitor transcriptional activity of NRF2, a key mediator of the antioxidant response (55). In the absence of oxidative stress, NRF2 is bound by the KEAP1 adaptor and undergoes rapid proteasomal degradation, whereas under oxidative stress, KEAP1/NRF2 interactions are blocked and NRF2 becomes stabilized. Based on a previously described reporter (40), two tandem regulatory NRF2-binding AREs were placed 5ʹ-upstream of a minimal promoter-driven destabilized mCherry, permitting fluorescence detection of NRF2 transcriptional activity (see schematic in Fig. 4C). Stable MG63.3 clones expressing ARE-mCherry (designated MG63.3 eGFP/ARE-mCherry) were isolated and subjected to tBHQ treatment, which led to activation of ARE-mCherry in MG63.3 cells, as shown by fluorescence of ARE-driven mCherry expression (Fig. 4D), as well as upregulation of NRF2, its known target gene, NQO1, and mCherry itself (Fig. 4E). Live-cell imaging showed that while neither vehicle nor CR-1-31B alone had any effect, tBHQ treatment for 24 hours significantly enhanced mCherry fluorescence. However, fluorescence was completely blocked when MG63.3 cells were co-treated with tBHQ and CR-1-31B in either two-dimensional monolayers (Fig. 4F) or 3D spheroids (Supplementary Fig. S7A). This was confirmed by Western blotting, whereby CR-1-31B co-treatment blocked mCherry and NQO1 protein expression in tBHQ-treated cells, which in addition showed that expression of NRF2 itself was downmodulated by CR-1-31B/tBHQ co-treatment (Fig. 4G). Identical results were observed in a second OS cell line, namely ARE-mCherry–expressing MNNG cells (Supplementary Fig. S7B–S7E). Together, these data indicate that CR-1-31B inhibits NRF2-mediated antioxidant response and induction of ARE-regulated genes under oxidative stress in OS cells.

To determine how CR-1-31B blunts the NRF2 antioxidant response in an unbiased manner, we used TMT proteomics to catalog proteins in MG63.3 cells that are specifically enhanced by oxidative stress, but in which co-treatment with CR-1-31B inhibits this induction. We therefore specifically focused on proteins (i) that were upregulated at least twofold by 75 μmol/L tBHQ treatment; (ii) that were not induced by 4 nmol/L CR-1-31B alone; and (iii) in which induction by tBHQ was blocked or reduced by co-treatment for 2 hours. A heatmap of the major proteomic changes under these conditions compared with vehicle alone (DMSO) is shown in Fig. 5A. The top 10 proteins fulfilling these criteria included inhibitor of growth family member 5 (ING5), NFE2-like bZIP transcription factor 2 (NRF2), potassium voltage–gated channel subfamily H member 8 (KCNH8), PZP alpha-macroglobulin-like protein (PZP), thymosin beta-10 (TMSB10), alpha-2-macroglobulin (A2M), phosphoenolpyruvate carboxykinase-1 (PCK1), solute carrier family 51 subunit beta (SLC51B), HGF activator (HGFAC), and early growth response 1 (EGR1). Of these, the most compelling protein for further analysis was NRF2, given its known role in the antioxidant response (29). NRF2 expression was induced by tBHQ but not by CR-1-31B itself at 2 to 4 nmol/L concentrations (Fig. 5B), but tBHQ-mediated NRF2 induction was blocked by CR-1-31B co-treatment in a dose-dependent manner. This effect was nontranscriptional, as CR-1-31B failed to reduce NFE2L2 mRNA levels, and in fact co-treatment with tBHQ actually increased NFE2L2 levels (Fig. 5C). Moreover, CR-1-31B at 2, 4, or 8 nmol/L had no effect on the stability of NRF2 protein, known to have a half-life of ∼30 minutes in other cell types (56), as shown using CHX pulse chase assays over a period of at least 2 hours (Fig. 5D); Western blots for NRF2 stability are shown in Supplementary Fig. S8. To test if CR-1-31B instead translationally controls NRF2 expression, we performed bioorthogonal noncanonical amino acid tagging to detect newly synthesized proteins (Fig. 5E). This technique utilizes L-AHA combined with “Click-It” chemistry and biotin alkenes to biotinylate and identify acutely synthesized proteins in MG63.3 cells treated with tBHQ +/− CR-1-31B (Fig. 5F). This clearly demonstrated that while NRF2 is acutely synthesized over a 2-hour period in response to tBHQ (but not vehicle or CR-1-31B alone), NRF2 induction was blocked when cells were co-treated with tBHQ and 2 to 4 nmol/L CR-1-31B, especially at 4 nmol/L (Fig. 5G). These data strongly suggest that under conditions whereby global mRNA translation remains active, NFE2L2 translation is selectively inhibited under tBHQ/CR-1-31B co-treatment compared with tBHQ alone. Accordingly, NFE2L2 distribution to early (i.e., translationally inactive) versus late polysomal fractions was enhanced in OS cells treated with CR-1-31B (Supplementary Fig. S9A). The known rocaglate-sensitive transcript, c-MYC, was similarly affected, with transcript partitioning to light polysomal fractions upon CR-1-31B single or tBHQ co-treatment (Supplementary Fig. S9B). These findings support the above ARE-mCherry studies, showing that NRF2 upregulation and subsequent induction of its downstream targets, NQO1 and ARE-driven mCherry, are reduced by CR-1-31B treatment (Fig. 5G; quantified in Fig. 5H). Notably, ectopic NRF2 expression driven by a ubiquitous cytomegalovirus (CMV) promoter in MG63.3 cells was completely unaffected by tBHQ/CR-1-31B co-treatment (Supplementary Fig. S9C), in contrast to the effects on wild-type NFE2L2 observed in Figs. 4G, 5B and G, and Supplementary Fig. S7E. Furthermore, IHC staining of MG63.3 primary tumor (Supplementary Fig. S10A) and lung metastasis sections (Supplementary Fig. S10B) demonstrated markedly reduced NRF2 immunostaining in the CR-1-31B–treated cohort compared with controls. Together, these data show that CR-1-31B potently inhibits acute NRF2 synthesis in OS cells under oxidative stress.

To verify the role of NRF2 in reducing oxidative stress and enhancing lung colonization of OS cells, we genetically inactivated NFE2L2 in MG63.3 cells using two independent lentiviral-based doxycycline (dox)–inducible NFE2L2-targeting constructs, shNRF2#1 or shNRF2#2, to mediate NRF2 KD. In the absence of dox, neither a nontargeting control shRNA (shNTC) nor the two targeting shRNAs blocked NRF2 upregulation under tBHQ treatment, as expected. However, in the presence of dox, both targeting shRNAs inhibited NRF2 induction, whereas the shNTC control had no effect (Supplementary Fig. S11A). Dox significantly reduced proliferation of shNRF2-expressing MG63.3 cells under tBHQ treatment in two-dimensional cultures compared with the shNTC control, whereas no differences were observed in the absence of dox (Supplementary Fig. S11B). Moreover, compared with controls, NRF2 KD significantly increased cell death by SYTOX staining in MG63.3 3D tumor spheroids expressing shNRF2#1 or shNRF2#2 (Supplementary Fig. S11C and S11D). Next, using PuMA, growth of MG63.3 lung lesions was similar between shNTC controls and cells with NFE2L2-targeting constructs in the absence of dox. However, dox treatment significantly reduced lung lesions of shNRF2#1- or shNRF2#2-expressing cells compared with shNTC controls, and indeed lung colonization was almost completely blocked by NRF2 KD (Supplementary Fig. S11E and S11F). Finally, using the above in vivo tail vein injection model, mice injected with shNTC-expressing MG63.3 cells showed no significant differences in metastatic burden when fed regular versus dox chow. By contrast, mice injected with shNRF2#1- or shNRF2#2-expressing cells had significantly reduced metastatic burdens when fed on a dox chow diet versus their regular chow fed counterparts (Supplementary Fig. S11G and S11H). Together, these data confirm that loss of NRF2 is sufficient to block OS growth and metastatic capacity, phenocopying the inhibitory effects of CR-1-31B on metastatic progression in OS.

We next assessed if a second rocaglate-based EIF4A1 inhibitor, the clinical-grade compound eFT226 (zotatifin; kindly provided by eFFECTOR Therapeutics), has similar activity as CR-1-31B. eFT226 is currently in phase 1 to 2 clinical trials for adult malignancies (57) and therefore likely closer to clinical application in OS. We first used in vitro functional assays to show that similar to CR-1-31B, combining tBHQ with eFT226 at low nanomolar levels strongly sensitized MG63.3, MNNG, and OS742 cells to tBHQ-induced oxidative stress, which was not observed with eFT226 alone (Fig. 6A). Notably, inhibitory effects on OS742 cell proliferation were observed at eFT226 concentrations as low as 0.1 nmol/L when combined with tBHQ (Supplementary Fig. S12). Treatment with eFT226 markedly downregulated NRF2 protein induction in response to tBHQ when MG63.3 cells were co-treated with tBHQ and 2 to 8 nmol/L eFT226 (Fig. 6B). Using the above ARE-mCherry reporter, nanomolar eFT226 concentrations dramatically reduced ARE-mCherry fluorescence in MG63.3 (Fig. 6C) treated with tBHQ. Finally, eFT226 robustly prevented growth of MG63.3 cells in lung tissue in the PUMA model. Although lung tissue seeding on day 0 of tumor cell injection did not significantly vary between DMSO and eFT226-treated PuMA slices, by day 14 after injection, 4 to 6 nmol/L eFT226 treatment of PuMA slices strongly inhibited lung metastatic growth compared with vehicle controls (Fig. 6D). Similar results were observed in MNNG cells, where eFT226 sensitized MNNG cells to tBHQ, downmodulated NRF2 protein under oxidative stress, downmodulated the ARE-mCherry reporter under oxidative stress, and inhibited growth in the PuMA model (Supplementary Fig. S13A–S13D). Collectively, our data support the working model of Fig. 6E, whereby EIF4A1 inhibition reduces lung metastatic burden; in the absence of CR-1-31B/eFT226, OS cells that have disseminated to the lungs mount an NRF2-mediated adaptive response against oxidative stress, permitting metastatic OS cells to survive and proliferate to form lung metastases. However, in the presence of CR-1-31B/eFT226, NFE2L2 translation and NRF2 synthesis are blocked or attenuated, resulting in a diminished antioxidant response. Drug-treated OS cells thus fail to survive in the lung microenvironment and undergo cell death, thereby inhibiting or reducing lung metastases.

There is a profound unmet clinical need to identify effective antimetastatic therapies for high-risk childhood bone cancers such as OS. Here, we demonstrate that small-molecule inhibitors of mRNA translation initiation severely limit metastatic competence in OS. The EIF4F complex member, EIF4A1, is markedly overexpressed in OS across diverse public OS RNA-seq datasets, which was confirmed by IHC in clinical specimens. Accordingly, we uncover that CR-1-31B, an EIF4A1/2 inhibitor of the rocaglate class of compounds, has potent in vitro antiproliferative activity in OS cells across three different mammalian species, namely mouse, canine, and human cells, at low nanomolar concentrations. Furthermore, CR-1-31B blocks both primary OS growth and metastatic capacity in vivo. CR-1-31B unexpectedly increased sensitivity to oxidative stress, known to diminish metastatic capacity (6, 7). Notably, the ability of CR-1-31B to decrease metastatic capacity and adapt to oxidative stress was observed at concentrations below those at which global translation was inhibited, suggesting that transcripts involved in the antioxidant stress response and the metastatic phenotype may be selectively inhibited by CR-1-31B. Indeed, we identified NFE2L2, which encodes NRF2, a master regulator of the antioxidant response (29), as being translationally repressed by CR-1-31B in vitro and in vivo and that genetic inactivation of NRF2 strongly inhibited OS colonization of the lungs, the main site of OS metastatic dissemination. Moreover, we showed that a clinical-grade rocaglate, eFT226 (zotatifin), shows identical capacity to sensitize OS cells to oxidative stress and also dramatically inhibits the metastatic phenotype of aggressive OS cells.

NRF2 is a highly attractive molecular target in metastatic OS as a key transcriptional mediator of the antioxidant response, and its upregulation has already been linked to poor outcome in OS patients (58). NRF2 is critical for mounting an effective antioxidant response through ARE-driven downstream effector proteins such as NQO-1, HO-1, GCL, and other NRF2 transcriptional targets, which are collectively responsible for neutralizing excess ROS and restoring cellular redox balance (29). This is likely critical for OS lung metastasis, since high oxidative stress attributable to either intrinsic sources of high ROS (i.e., mitochondrial respiration; ref. 7) or exerted by the host cells (microvascular endothelium) has been reported in the lung microenvironment (6). Moreover, upregulation of NRF2 and antioxidant gene programs are associated with drug-tolerant persister cell growth across multiple cancer types (30). NRF2 also modulates several other cellular processes associated with the metastatic phenotype, including enhanced cell migration (59), and metabolic reprogramming and mitochondrial functions (60, 61). Presumably, the polypurine-rich region within the NFE2L2 5ʹ-UTR (62) makes this transcript susceptible to CR-1-31B/eFT226-mediated clamping of EIF4A1 onto rocaglate-sensitive mRNAs. Our findings suggest that modulating the NRF2 response may be of particular importance in the setting of metastatic OS. However, whether rocaglate-based inhibition of other pathways also plays a role remains to be determined, as EIF4A inhibition is likely to target other relevant metastasis-associated pathways in OS cells. In support of the latter, our NRF2 shRNA KD studies showed that although lung tumor burden was reduced by this intervention, it was not completely eliminated.

Previous work from our group has shown that in Ewing sarcoma, another aggressive childhood bone cancer, enhanced translation of multiple cytoprotective mRNAs by the YB-1 RNA binding protein supports the metastatic competence of Ewing sarcoma cells (41, 61, 63). Collectively, these data and those of the present study strongly suggest that targeting the translation of pro-metastatic mRNAs such as NFE2L2 is a promising therapeutic strategy in metastatic OS and other high-risk childhood sarcomas. Indeed, Robichaud and colleagues (21) showed that EIF4E phosphorylation is required for translational activation of SNAIL and MMP3 mRNAs, which is required for an invasive and metastatic phenotype in a PyMT breast cancer model. Truitt and colleagues (64) reported that EIF4E-mediated translational control of ROS levels facilitates transformation and cancer cell survival in vivo. Our group showed that YB-1 also enhances stress-selective translation of cytoprotective proteins such as HIF1ɑ and G3BP1, both of which contribute to sarcoma metastases (41, 61). Moreover, although our studies support the notion that CR-1-31B–mediated inhibition of NRF2 expression is necessary and sufficient to inhibit OS lung colonization, it is likely that other CR-1-31B–downregulated proteins also contribute to inhibition of OS metastasis and will be pursued in future studies.

Compared with inhibitors of EIF4G1 (SBI-756), EIF4E1 (Tomivosertib), and EEF2K (A484954) tested in this study, those targeting EIF4A1 (silvestrol, CR-1-31B, and eFT226) were the most potent in metastatic OS cells, suggesting that EIF4A1 activity may be crucial for the translation of OS oncogenic signaling pathways. For example, CR-1-31B has been shown to have antitumor activity in several preclinical cancer models (26, 44, 45, 65). Cao and colleagues (66) found that CR-1-31B translationally downregulates c-FLIP, resulting in reduced TRAIL resistance in gallbladder cancer cells and reduced TRAIL-treated tumors in vivo, whereas Fooks and colleagues (67) demonstrated that single-agent treatment with CR-1-1B increases oxidative stress in acute myeloid leukemia (AML) cells. A related EIF4A1/2 natural product–based inhibitor, Rocaglamide, had antitumor activity in malignant peripheral nerve sheath tumor and other sarcomas, including OS (43). However, effects on oxidative stress pathways and metastatic capacity in OS were not studied. Our data add to these studies by showing that another rocaglate, CR-1-31B, has a higher potency and shows antitumor activity at doses 15-fold lower than Rocaglamide and extends the effects of rocaglates to include antimetastatic activity for CR-1-31B and eFT226 using the PuMA model, as well as for CR-1-31B in vivo. The clinical-grade compound, eFT226, showed preclinical antitumor activity against hepatocellular carcinoma (68), receptor tyrosine kinase (RTK)–driven solid tumors (69), B-cell lymphoma (70), and TNBC (27). More recently, eFT226 was found to inhibit primary tumor growth and enhance interferon response and T-cell–dependent tumor targeting in TNBC (28). Although the chemical structures of CR-1-31B and eFT226 share the same cyclopenta[b]benzofuran core, each has unique chemical properties which suggest that their translation inhibitory mechanisms might differ slightly (71). Having said that, we found that both rocaglates block NRF2 synthesis and lung colonization in our OS models. To our knowledge, this work is the first to report that CR-1-31B or eFT226 monotherapy can reduce OS lung metastases in the PuMA model, and for CR-1-31B, in both spontaneous and tail vein injection metastatic OS mouse models.

In conclusion, our study provides compelling preclinical evidence to support the therapeutic application of CR-1-31B and eFT226 to inhibit lung colonization in metastatic OS, potentially mediated through the translational inhibition of NRF2. Future studies will be necessary to assess whether rocaglate-based modulation of NRF2 can also enhance responses to standard-of-care chemotherapy agents in OS, such as methotrexate, doxorubicin, and cisplatin (i.e., MAP therapy). Notably, eFT226 is under development and testing in clinical trials for adult malignancies (#NCT04092673; NCT05101564), providing an exciting opportunity for testing in OS as a clinical intervention.

M.M. Lizardo reports nonfinancial support from eFFECTOR Therapeutics during the conduct of the study. E. Sweet-Cordero reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

M.M. Lizardo: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. C. Hughes: Conceptualization, data curation, formal analysis, methodology. Y.Z. Huang: Data curation, validation, investigation, methodology. T. Shyp: Data curation, investigation, methodology. A. Delaidelli: Data curation, validation, methodology. H.-F. Zhang: Data curation, methodology. S.S. Shaool: Data curation, methodology. A.F. Renner: Data curation, formal analysis. F. Burwag: Data curation. L.C. Sayles: Resources. A.G. Lee: Data curation, software, formal analysis. E. Sweet-Cordero: Resources. P.H. Sorensen: Conceptualization, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Melanie Rouleau for critical reading of the article and Sylvia Cheng of BC Cancer Research Histology and Imaging Core for technical assistance. This work was supported in part by Alex’s Lemonade Stand Foundation grant #21-23176 (to P.H. Sorensen and A. Sweet-Cordero), by CIHR Foundation grant #FDN-143280 (to P.H. Sorensen), and by the British Columbia Cancer Foundation through generous donations from Team Finn and other riders in the Ride to Conquer Cancer (to PHB). M.M. Lizardo was funded in part by the British Columbia Cancer Foundation and Alex’s Lemonade Stand Foundation (#21-23176). We would also like to thank Stephen Worland and others from eFFECTOR Therapeutics, Inc., for kindly providing eFT226 (zotatifin) and for scientific discussions and critical feedback on the article.