The Raji, DG-75 and Ramos Burkitt lymphoma (BL) cell lines were authenticated using ATCC’s STR Profiling Authentication Services. The CA46, Daudi, GA-10, EB1 and ST486 cell lines were obtained from ATCC, while the BL41 cell line was obtained from DMSZ. The Namalwa cell line was a generous gift from Dr. Javier Di Noia at IRCM (Montreal, QC, Canada). Daudi-Fluc-Puro and Raji-Fluc-Puro cell lines were obtained from Imanis Life Sciences. Data concerning mutation in the gene DDX3X carried by the different cell lines was extracted from the Broad depmap portal (https://depmap.org/portal).

The RK-33 inhibitor was purchased from Focus Biomolecules and prepared as a 50 mM stock solution in dimethyl sulfoxide (DMSO) for both in vitro and in vivo experiments. MLN4924 (pevonedistat) was obtained from AdooQ BioScience and prepared as a 50 mM stock solution in DMSO for in vivo experiments. The C1 compound was custom synthesized by ChemSpace (Latvia) and prepared as a 50 mM stock solution in DMSO.

DL-buthionine-sulfoximine (BSO) and inosine were purchased from Millipore Sigma and prepared as 100 mM stock solutions in water. TAK-243 and omaveloxolone were obtained from Selleck Chemicals, while CPUY192018 was obtained from MedChemExpress Co., Ltd.; all were prepared as 100 mM stock solutions in DMSO.

For Western blot, The DDX3Y-specific antibody was purified from rabbits immunized as published (Gong et al., 2021) by Biomatik. The DDX3X antibody A300-474A was from Bethyl laboratory.

NSG (NOD.Cg-Prkdc^scid Il2^rgtm1Wjl /SzJ, Strain #:005557, RRID:IMSR_JAX:005557) mice were obtained from Jackson Laboratory and bred in the IRCM SPF+ (free of pathogens and opportunistic germs) animal facility. Mice were maintained on a 12-h light/dark cycle with ad libitum access to food and water.

All animal experiments were reviewed and approved by the Animal Care Committee of IRCM (Protocol #2025–1282) in accordance with the guidelines of the Canadian Council on Animal Care.

Cell viability for all cell lines was determined using an XTT assay from Biotium following the manufacturer’s instructions. Briefly, 5 × 10³ to 1 × 10⁴ cells were seeded in 96-well plates with 100 μL of RPMI supplemented with 10% FBS, containing serial dilutions of the compounds from 100 μM down to 1% DMSO (vehicle control). Cells were cultured at 37°C for 4 days. 50 μL of XTT labeling reagent was then added to each well, and plates were incubated for 1 h at 37°C. Absorbance was measured at 490 nm (signal) and 650 nm (background) using a SpectraMax 190 Microplate Reader (Molecular Devices) with SoftMax Pro 5.2 software. Cell viability was calculated by subtracting 650 nm background values from 490 nm absorbance values and normalizing to the DMSO-treated control. Nonlinear regression curves and the 50% cytotoxic concentration (CC₅₀) were determined using GraphPad Prism version 10.4.2.

The synthetic lethality and/or rescue potential of combinational treatments was also assessed using the XTT assay. The CC₅₀ values for additional drugs were determined as described above and were then used at concentrations below their CC₅₀: DL-buthionine-sulfoximine (10 μM); inosine (500 μM); TAK-243 (4 nM); omaveloxolone (50 nM); CPUY192018 (5 μM). These compounds were combined with serial dilutions of RK-33 (ranging from 150 μM to 1% DMSO vehicle control) to evaluate combinatorial effects.

To assess apoptosis induced by DDX3X inhibitors, Raji, CA46, and Daudi cells were treated for 24 h with RK-33 at concentrations ranging from 0 to 50 μM and C1 at 0–5 μM. Cells were then stained with FITC-labeled annexin V (BioLegend) and propidium iodide (PI), acquired using a Sony SA3800 Spectral Cell Analyzer, and analyzed with FlowJo (version 10.9.0).

To establish a luciferase-expressing CA46 cell line, parental CA46 cells were transfected with the pF CAG luc hygro plasmid (a gift from Brett Stringer; Addgene plasmid #67502; http://n2t.net/addgene:67502; RRID: Addgene_67502). This plasmid contains the firefly luciferase gene under the control of the chicken beta-actin promoter and hygromycin resistance gene (Stringer et al., 2019). Selection was carried out using hygromycin B (Millipore Sigma) in a methylcellulose-based semi-solid medium (ClonaCell-TCS, Stemcell Technologies). The resulting CA46-Fluc-Hygro cell line was compared to the commercially available Raji-Fluc-Puro and Daudi-Fluc-Puro cell lines, exhibiting a bioluminescence level approximately 1/30 of the latter two (data not shown).

To determine the optimal cell number for the lymphoma dissemination monitoring, five-week-old NSG mice were injected intravenously (i.v.) with increasing doses (1 × 10⁵, 2 × 10⁵, 3 × 10⁵, 4 × 10⁵, and 5 × 10⁵ cells in 100 µL PBS) of Fluc-expressing BL cell lines. Dissemination was evaluated using bioluminescence imaging with the IVIS Imaging System 200 (Xenogen). Based on these results, the following cell numbers were selected: Raji-Fluc-Puro: 1 × 10⁵; Daudi-Fluc-Puro: 2.5 × 10⁵; CA46-Fluc-Hygro: 5 × 10⁵.

For each cell line, 20 five-week-old NSG mice were injected i.v. with the optimized number of cells in 100 µL PBS and mice were randomized into control (vehicle-treated) and treatment (RK-33 or Pevonedistat) groups. For bioluminescence imaging, mice were injected intraperitoneally (i.p.) with 150 mg/kg D-luciferin (15 mg/mL in PBS; 250 µL (∼3.75 mg) per 25 g mouse) and anesthetized using 2% isoflurane. Imaging was performed with the Xenogen IVIS Imaging System 200 using exposure times of 1–30 s. A maximum of five mice were imaged simultaneously, with continuous isoflurane exposure to maintain anesthesia. Bioluminescence was recorded every other day, and mice reaching a predefined dissemination threshold, set at an average radiance of 1 × 10⁶ photons/s/cm²/sr for both Raji and Daudi xenografts, and 3 × 10⁵ photons/s/cm²/sr for the CA46 xenograft, were assigned to treatment. Mice in the RK-33 treatment group received 50 mg/kg RK-33 (5 mg/mL in 20% DMSO/corn oil, 250 µL per 25 g mouse, corresponding to 1.25 mg), administered i.p. every other day for 2 weeks. For the Pevonedistat treatment group, mice received subcutaneous injections of 90 mg/kg Pevonedistat (14.7 mg/mL in 5% DMSO/30% PEG 300/5% Tween 80, 150 µL per 25 g mouse, corresponding to 2.2 mg) every other day for 2 weeks. Control mice received vehicle injections consisting either of 10 µL of 20% DMSO in corn oil per gram of body weight or 150 µL of 5% DMSO/30% PEG 300/5% Tween 80, depending on the treatment group at the same frequency. Lymphoma progression was monitored every other day for 3 weeks after treatment initiation or until mice reached endpoint criteria. Results were reported as the radiance fold change over the first day of treatment. BL progression curves were compared by a CGGC permutation test using 10,000 permutations (Elso et al., 2004).

Due to differences in bioluminescence intensity among cell lines, sample imaging was standardized for visual comparison. Imaging of mice injected with Raji-Fluc-Puro or Daudi-Fluc-Puro cells was conducted with an exposure time of one second, while imaging of mice injected with CA46-Fluc-Hygro cells required an exposure time of 30 seconds due to the lower signal intensity.

To target the human DDX3X and DDX3Y genes, Small interfering RNAs (siRNAs) were obtained from Origene. The siRNAs used were the Trilencer-27 siRNA oligo duplex kits SR319906BL for DDX3X, SR305689BL for DDX3Y, and the universal scrambled negative control siRNA duplex (SR30004). For the RNA-seq experiments, 2.5 × 10⁵ Raji and CA46 cells were electroporated with 10 pmol of either the scrambled control, DDX3X, DDX3Y, or a combination of DDX3X and DDX3Y siRNAs using the Neon Transfection System (Thermo Fisher Scientific) and the 10 μL Neon Transfection kit with Buffer T. The parameters were as follow: for Raji cells, a pulse voltage (PV) of 1,100 V, pulse width (PW) of 30 ms, and pulse number (PN) of 2; for CA46 cells, a PV of 1,150 V, PW of 20 ms, and PN of 2. For the mass spectrometry experiments, 2.5 × 10⁶ cells were electroporated with 100 pmol of siRNA using the 100 μL Neon kit with the same electroporation parameters. After the electroporation, cells were seeded into 24-well culture plates and incubated for 24 h in RPMI medium supplemented with 10% fetal bovine serum (FBS), then harvested for RNA or protein extraction. All these conditions were done in duplicate.

This research involves human participants who gave informed consent to participate in the study under the approbation of the research ethics committees of Héma-Québec (Project # 2021–016) and received approval from the Research ethics committee (CÉR) of the IRCM (protocol # 2022–1138). Three healthy male donors provided primary human B cells which were derived from leukapheresis products. Héma-Québec supplied patient samples that were stored in Leukoreduction System (LRS) Chambers (Excellos). The leukocyte/red blood cell mixture collected from the chambers was separated by density gradient centrifugation using Lympholyte®-H (Cedarlane) to isolate peripheral blood mononuclear cells (PBMCs). Untouched B cells were then purified from the PBMC fraction using the MojoSort™ Human Pan B Cell Isolation Kit (BioLegend). Isolated B cells were resuspended in RPMI medium supplemented with 20% charcoal-stripped FBS, 10 mM HEPES and 55 nM β-Mercapthoethanol and a total of 1 × 10⁶ purified B cells were seeded into 12-well plates pre-coated with an irradiated feeder cell layer (YK6-CD40L-IL21), which were provided as a generous gift by Dr. Daniel Hodson (Wellcome-MRC Cambridge Stem Cell Institute, Cambridge, UK), and express CD40 ligand and secrete interleukin-21 to support B cell activation and proliferation (Caeser et al., 2019). Cells were cultured under these conditions for a 48 h activation period, and then treated with either 5 µM RK-33 or 0.4% DMSO (vehicle control) for 24 h. After treatment, B cells were carefully collected to prevent feeder cell contamination. Total RNA was then extracted for RNA sequencing.

To assess the impact of DDX3X inhibition on the global transcriptome, total RNA was extracted from 2.5 × 10⁵ Raji and CA46 cells after 24-h treatments with either RK-33 (50 μM for Raji and 10 μM for CA46) or siRNAs targeting DDX3X, DDX3Y, or both DDX3X/DDX3Y and their respective controls (DMSO or Scramble siRNA). Similarly, total RNA was extracted from primary human B cells treated with RK-33 or DMSO for 24 h. In all cases, RNA was extracted using the RNeasy Mini Kit (Qiagen). RNA integrity and quality were verified using the RNA 6000 Pico Kit on an Agilent Bioanalyzer system.

RNA-seq libraries were prepared using the Illumina TruSeq Stranded mRNA Library Prep Kit, following the manufacturer’s protocol. Sequencing was conducted on an Illumina HiSeq 2000 system using the TruSeq PE Cluster Kit v3-cBot-HS. Quality control of the raw sequencing reads was performed with FASTQC v0.12.1. Reads were aligned to the human reference genome GRCh38 using STAR v2.7.11b, with an average of 87% of reads uniquely mapped. Raw read counts were obtained using FeatureCounts v2.0.6, based on the GRCh38 (release 110) genome annotation. Differential gene expression analysis was performed using the DESeq2 package in R. KEGG pathway enrichment analysis was performed using the GSEA (Gene Set Enrichment Analysis) tool with 1,000 gene set permutations (Joshi et al., 2013) or using ShinyGO v0.82, based on significantly deregulated genes (adjusted P < 0.05) (Ge et al., 2020). Expression heatmaps were created using the Heatmapper online tool, applying hierarchical clustering to both rows and columns using the Average Linkage method and Kendall’s Tau distance metric (Babicki et al., 2016). Principal component analysis (PCA) was done using the Statistics Kingdom platform (https://www.statskingdom.com). Additional functional enrichment analysis, including gene ontology and pathway analysis, was performed using the gprofiler2 package in R. All RNA-seq data are available through GEO under accession number GSE294862. A summary of RNA-seq data are available in Supplementary Tables S1–S3.

To evaluate the impact of DDX3X inhibition on the chemogenomic interactions from nonglobal proteome, total protein was extracted from 2.5 × 10⁶ Raji and CA46 cells following 24-h treatments with RK-33, DMSO, or siRNAs targeting DDX3X, DDX3Y, or both. Cells were lysed in RIPA buffer supplemented with EDTA-free protease inhibitors (Roche) and disrupted by brief sonication (3 s) using a Branson 250 Digital Sonifier. Lysates were centrifuged, and supernatants were collected for downstream analysis.

Protein concentration was determined using the BCA Protein Assay Kit (Thermo Scientific). For each condition, approximately 100 µg of total protein was digested, isobarically labeled with TMT 6-plex reagents (Thermo Fisher Scientific) and fractionated into five parts. Mass spectrometry analysis was done on an Orbitrap Fusion mass spectrometer (Thermo Scientific). To control for labeling bias, samples were organized into four independent TMT 6-plex sets. Two sets (TMT-1 and TMT-2) each contained one replicate of CA46 cells (for a total of two biological replicates for each condition) treated with DMSO, RK-33, scramble siRNA, DDX3X siRNA, DDX3Y siRNA, and combined DDX3X/DDX3Y siRNAs. The remaining two sets (TMT-3 and TMT-4) included one replicate each of Raji cells similarly treated. Peptide identification and quantification were performed using Proteome Discoverer v2.1 (Thermo Scientific) against the UniProt human protein database (downloaded October 2024). Within each TMT experiment, relative protein abundances were normalized to total protein abundance per sample and corrected for TMT batch effects by first averaging the expression values for each gene across all samples within each TMT set. For each gene, a normalization ratio was then calculated between paired TMT sets (e.g., TMT-1/TMT-2 and TMT-3/TMT-4). This gene-specific ratio was used to scale the corresponding values in one TMT set (e.g., TMT-2), thereby minimizing inter-set variation and effectively correcting for batch effects. Proteins exhibiting a fold change ≥1.4 relative to their corresponding controls (DMSO for RK-33 treatment, scramble siRNA for siRNA-treated samples) were considered differentially expressed and selected for pathway enrichment analysis. Enrichment analysis of biological pathways was performed using GSEA and ShinyGO v0.82, following the same procedures applied to the transcriptomic dataset. Both upregulated and downregulated proteins were analyzed based on the 1.4-fold change threshold to identify significantly altered functional categories and pathways. Heatmaps and PCA were performed as described for the transcriptome profiling, using the same tools and parameters. Proteomic data are available in Supplementary Table S4.

To identify genetic alterations that either sensitize or confer resistance to DDX3X inhibition, a genome-wide pooled CRISPR/Cas9 knockout screen in the human pre-B acute lymphoblastic leukemia cell line NALM-6 was performed by the ChemoGenix platform at the Université de Montréal (Montreal, QC, Canada), as described previously (Bertomeu et al., 2018). Briefly, a NALM-6 clone stably expressing a doxycycline-inducible Cas9 made from pCW-Cas9 (Addgene #50661) was transduced with the EKO library, a genome-wide sgRNA targeting protein-coding genes (278,754 different sgRNAs). sgRNA-mediated gene disruption was then initiated by adding doxycycline (2 μg/mL) for a period of 7 days to allow for robust gene knockout across the population. Post-induction, the pooled library was then split into different T-75 flasks (28 × 10⁶ cells per flask; a representation of 100 cells/sgRNA) in 70 mL at 4 × 10⁵ cells/mL. One set was treated with 7 µM RK-33 and the other with DMSO as vehicle control. Cells were cultured under these conditions for 8 days, with monitoring of growth every 2 days, diluting back to 4 × 10⁵ cells/mL and adding more compound to maintain the same final concentration whenever cells reached 8 × 10⁵ cells/mL. Viable cells were harvested at the end of the treatment window.

Genomic DNA was isolated from cell pellets using the Gentra Puregene Cell Kit (Qiagen), following the manufacturer’s guidelines. The sgRNA cassette integrated into each genome was PCR-amplified as described (Bertomeu et al., 2018) and prepared for next-generation sequencing (NGS). Sequencing was performed using the Illumina NextSeq 2000 platform to determine sgRNA abundance in each treatment condition. Reads were aligned using Bowtie 2.4.4 in forward-strand only mode (using the norc parameter) with otherwise default parameters and total read counts per gRNA were tabulated. Context-dependent chemogenomic interaction scores were calculated using a modified version of the RANKS algorithm (Bertomeu et al., 2018) that uses guides targeting similarly essential genes as controls to distinguish condition-specific chemogenomic interactions from non-specific fitness/essentiality phenotypes. The chemogenomic screen data are available in Supplementary Table S5.

Intracellular total reactive oxygen species (tROS) levels were measured using the Total ROS Assay Kit 520 nm from Thermo Fisher Scientific. Raji cells were treated with 12.5 μM RK-33 or DMSO, and CA46 cells were treated with 10 μM RK-33 or DMSO for 24 h prior to analysis. Cells were analyzed on a BD LSRFortessa™ Cell Analyzer (BD Biosciences) using the 488 nm excitation laser and FITC emission filter (530/30 nm). The extent of ROS production was quantified by comparing the mean fluorescence intensity (MFI) between RK-33-treated and vehicle-treated control cells.

The sensitivity of Burkitt lymphoma (BL) cell lines to DDX3X inhibition was evaluated by treating BL cell lines with two different DDX3X inhibitors: RK-33 (Bol et al., 2015a; Xie et al., 2016; Yang et al., 2020) and the recently developed, more potent inhibitor C1 (Abate et al., 2015; Nakao et al., 2020). Nine BL cell lines were analyzed, six derived from male patients and three from female patients. Of the six male lines, two carried a wild-type (wt) DDX3X allele (CA46, BL41), while four harbored a mutant DDX3X allele (Raji, DG75, Daudi, GA10) (Supplementary Figure S1A). Most of these cell lines have upregulated DDX3Y possibly as a compensatory mechanism to counteract the loss of DDX3X function (Supplementary Figure S1B). Additionally, of the three female cell lines, one carried a wild-type DDX3X allele (Namalwa) and two carried a single mutant allele (EB1, ST486) (Supplementary Figure S1).

All BL cell lines exhibited varying degrees of sensitivity to DDX3X inhibition, including those that lack functional DDX3X and rely on DDX3Y for survival (Figures 1A,B; Supplementary Figure S2). Among them, Daudi, in which DDX3X was undetectable but DDX3Y was present at high levels (Supplementary Figure S1B), was highly sensitive to inhibition (Figure 1A). In contrast, Raji, which expresses a mutated DDX3X protein lacking helicase activity and relies on DDX3Y expression for survival (Wang et al., 2015) (Supplementary Figure S1B), was the most resistant (Supplementary Figure S1A). The CA46 cell line, which expresses wild-type DDX3X but no DDX3Y due to loss of the Y chromosome, displayed intermediate sensitivity to the inhibitors (Figure 1A; Supplementary Figure S1B).

Notably, the effects of DDX3X inhibition in Daudi, Raji, and CA46 BL cell lines were rapid, with detectable apoptosis occurring within 24 h at concentrations near the CC₅₀ (Figures 1C,D). In these experiments, C1 demonstrated an order of magnitude higher cytotoxicity than RK-33, highlighting that more potent DDX3X inhibitors can effectively kill BL cell lines in vitro.

DDX3X inhibition effectiveness in vivo was evaluated through a dissemination model which emulates human Burkitt lymphoma (BL) (Figure 2A) (Matis et al., 2009). Immunodeficient NSG mice received intravenous injections with luciferase-expressing CA46 (Fluc-CA46-Hygro), Daudi (Fluc-Daudi-Puro), or Raji (Fluc-Raji-Puro) BL cells and were then monitored for tumor progression using an IVIS 200 Xenogen device to detect bioluminescence. Mice received RK-33 treatment after their tumor burden reached defined threshold values of 6 × 10⁶ photons/s/cm²/sr for Fluc-Raji-Puro and Fluc-Daudi-Puro cells and 3 × 10⁵ photons/s/cm²/sr for Fluc-CA46-Hygro cells before being monitored for 2 weeks (Figures 2A–D). We chose RK-33 instead of C1 because RK-33 had previously been used in cancer xenograft models (Wilky et al., 2016; Tantravedi et al., 2019) and C1 displayed poor bioavailability (data not shown). For comparison, an independent cohort of mice received Pevonedistat as a reference drug because it represents a potent new B-cell lymphoma treatment undergoing clinical trials that targets Neddylation, a protein modification similar to ubiquitination (Czuczman et al., 2016; Paiva et al., 2017; Torka et al., 2020) (Figures 2B–D).

Mice injected with the mildly aggressive wild-type DDX3X-expressing Fluc-CA46-Hygro cells demonstrated substantial responsiveness to Pevonedistat treatment and a moderate sensitivity to RK-33 compared to the vehicle-treated group (Figure 2B) even at near-maximal tolerated dose. Mice injected with the highly aggressive Fluc-Raji-Puro cells, which express mutated DDX3X and functional DDX3Y showed only a modest response to both Pevonedistat and RK-33 treatments. As seen with the CA46 model, Pevonedistat demonstrated superior effectiveness in impeding tumor progression than RK-33 (Figure 2C). However, the overall reduction in tumor burden for both treatments was marginal, resulting in only 2–3 days of extended survival time. This suggested that the mutant DDX3X in Raji cells may still contribute to tumorigenicity while remaining unaffected by RK-33 since its catalytic activity is already inactive, or that DDX3Y retains some residual activity in the presence of the inhibitor.

Interestingly, the RK-33 treatment of mice injected with Fluc-Daudi-Puro cells, which exhibit an intermediate level of aggressiveness compared to the CA46 and Raji cell lines, resulted in tumor suppression similar to Pevonedistat (Figure 2D). Since the survival of Daudi cells depends on DDX3Y overexpression because DDX3X expression is undetectable, these results suggest RK-33 is also effective in blocking DDX3Y to slow down tumor progression and that inhibition of both DDX3 paralogues can contribute to tumor suppression. The results also show that RK-33 slowed down tumor progression across all three BL models, but that Pevonedistat proved more effective which suggests the cell lines developed alternative survival mechanisms against DDX3X inhibition in vivo.

To identify changes in cellular pathways caused by DDX3X inhibition, we treated Raji cells and CA46 cells with the DDX3X inhibitor RK-33. The Raji cell line, which carries a catalytically inactive DDX3X and relies on wild-type DDX3Y was chosen because it represents a typical case of Burkitt lymphoma, in which a mutant DDX3X is present alongside DDX3Y, both potentially contributing to the malignancy. The CA46 cell line was selected because it expresses only wild-type DDX3X and lacks DDX3Y, therefore providing a simplified model and a control. These cell lines represent a range of situations found in human BL. Cells were treated for 24 h and then total RNA and protein were extracted for next-generation RNA sequencing (RNA-seq, Figures 3A,B) and Tandem Mass Tag mass spectrometry (TMT-MS, Figures 3C,D) respectively. Because the TMT-MS data produced a strong batch effect between the different TMT sets, as highlighted by the strong clustering between TMTs (Supplementary Figure S3A), data were normalized to account for this effect, allowing to assess the treatment-specific effects by minimizing the influence of TMT-specific variation (Figure 3C).

The results revealed cell line-specific responses to RK-33 treatment at both the transcriptional and proteomic levels. The CA46 cells showed a very limited number of differentially expressed transcripts after RK-33 treatment, indicating that the transcriptional changes were relatively small (Figure 3E). Similar to the transcriptomic data, proteomic profiling revealed very little disruption with only a small number of proteins having altered abundance (Figure 3E). In contrast, Raji cells, which were more resistant to RK-33 at the viability level, had a much broader response RK-33 treatment affected thousands of genes as shown by RNA-seq, and mass spectrometry also revealed widespread proteomic changes (Figure 3F).

To obtain further understanding of the biological pathways affected by RK-33, we used Gene Set Enrichment Analysis (GSEA) and ShinyGO for pathway enrichment analysis. These analyses identified different patterns of pathway regulation between the 2 cell lines. CA46 cells showed minimal pathway enrichment with only a few categories showing some marginal dysregulation (Figures 3B,D,E). On the other hand, Raji cells had robust enrichment in a wide range of biological pathways at both the mRNA and protein levels (Figures 3B,D,F).

Although there was little overlap of the differentially expressed genes or proteins between the 2 cell lines and between transcriptomic and proteomic datasets within the same cell line, several biological functions were repeatedly identified between the different conditions. Of those, protein metabolism (protein export, ribosome function, unfolded protein response), oxidative stress (oxidative phosphorylation, mTORC1 signaling, ROS related pathways), RNA processing and degradation, and cell cycle regulation (G2/M checkpoint, mitotic spindle) were some of the pathways commonly affected (Figures 3E,F). This indicated that RK-33-induced DDX3X inhibition disrupts general cellular homeostasis rather than disrupting specific molecular pathways.

These results confirmed that RK-33 likely acts, at least in part, through the induction of cellular stress responses that may eventually result in apoptosis. Furthermore, the data suggested that transcriptional dysregulation precedes proteomic alterations, suggesting that transcriptional reprogramming is among the earliest detectable cellular responses to DDX3X inhibition. This is consistent with the observation that despite the high transcriptional effects observed in Raji cells, the proteomic variability was important, even between biological replicates, which might suggest a stochastic and dynamic nature of the downstream responses to cellular stress (Figure 3C).

Taken together these results indicated that RK-33 has biological effects that are highly context dependent. The observed differential sensitivity and variability in pathway responses among BL cell lines suggested a complex, multifaceted role of DDX3X in maintaining cellular function and the importance of cellular context in determining therapeutic efficacy.

To determine whether any of the effects described above could be attributed to non-specific effects of RK-33, we conducted siRNA to knockdown DDX3X, DDX3Y, or both genes at the same time. This strategy not only allowed us to compare the results of specific gene silencing to those observed with pharmacological inhibition, but also to explore the contribution of wild-type DDX3X (in CA46) and the catalytically inactive DDX3X variant (in Raji). This was important because, some studies had shown that DDX3X may exhibit some cellular functions that do not depend on its RNA-helicase activity (Soulat et al., 2008; Geissler et al., 2012; He et al., 2024; Owens et al., 2024). Therefore, we conducted knockdown experiments to help distinguish between phenotypes that arise primarily from helicase inhibition and those caused by the loss of the protein.

Since CA46 cells lack a Y chromosome, we only knocked down DDX3X in this line. However, Raji cells carry both a catalytically inactive DDX3X and the wild-type DDX3Y, thus enabling us to determine the functional significance of each variant by depleting them separately or in combination. The cells were transfected with siRNAs for 24 h before RNA-seq and TMT-based mass spectrometry analyses (Figure 4; Supplementary Figure S4). Importantly, both RNA-seq and mass spectrometry confirmed the successful knockdown of target genes on the RNA and protein levels but showed that siDDX3X was more efficient than siDDX3Y (Supplementary Figure S4). RNA-seq data showed that siRNA-mediated knockdown resulted in gene expression changes that were generally less pronounced than those seen with RK-33 treatment (Figures 4A,C). In both CA46 (Figure 4B) and Raji (Figure 4C) cells, only a small number of hallmark pathways were altered, and the changes in proteomics were even smaller (Figures 4F,G). Nonetheless, in Raji cells, PCA analysis of the transcription profiles showed that cells treated with siDDX3X alone or with both siDDX3X/DDX3Y were more similar to each other than to the scramble control or siDDX3Y alone (Figure 4D). Suggesting that the catalytically inactive DDX3X in Raji makes a substantial contribution to the transcriptome, pointing to a non-helicase-dependent functional role in these cells.

Although transcriptional changes were not massive, there was little overlap in differentially expressed genes between the different knockdowns in Raji cells (Figure 4E). However, using ShinyGO for gene set enrichment analysis, we observed a consistent enrichment in stress-related pathways, especially those related to the generation of reactive oxygen species and oxidative stress such as the MTORC1 Signaling hallmark (Figure 4E) which was also enriched in RK-33-treated cells along the oxydative phosphorylation hallmark. Moreover, ShinyGO analysis of the genes overlapping between the three knockdowns and RK-33-treated cells identified the same oxydative stress-related hallmarks with the reactive oxygen species pathway being the most enriched (Supplementary Figure S5). Broader pathway enrichment analyses across datasets also identified stress as a common theme, especially with respect to oxidative phosphorylation, protein metabolism, and cell cycle regulation (Supplementary Figure S4). These observations suggest that the knockdown of DDX3X or DDX3Y by siRNA causes a disruption of cellular homeostasis and activates a general stress response.

As observed in the proteomic data of RK-33-treated cells, the TMT-MS data produced a strong batch effect between the different TMT sets of siRNA-treated cells (Supplementary Figure S3B,C), and data were thus normalized to account for this effect (Figures 4F,G). At the proteomic level, siRNA knockdown had an even milder effect. Only a handful of proteins were differentially expressed across multiple knockdowns (Figures 4F–I). Importantly, PCA confirmed the heatmap clustering indicating that the main source of variance between proteomic datasets was not due to biological effects of knockdowns but rather due to technical aspects such as separating biological replicates into different TMT batches (Figure 4H). In both heatmap clustering of normalized data and PCA after removing the TMT batch effect by analyzing the components PC2 and PC3, the DDX3Y knockdown alone appeared to cluster closer to the dual knockdown than to the DDX3X knockdown alone, which shows a larger variation between two replicates than the other sets (Figures 4G,H). ShinyGO analysis of dysregulated peptides that are shared by different knockdowns did not identify any specific pathway enrichment, suggesting that the few observed alterations in protein levels may be random rather than specific biological responses (Figure 4I).

Despite proteomic changes that were very small and limited replicate clustering, the most marked effects observed when using broader gene set enrichment analyses were seen in Raji cells following the knockdown of the catalytically inactive DDX3X (Supplementary Figure S4B). This implies that despite the fact that the mutant DDX3X is inactive enzymatically, it still has a regulatory role in this particular case, possibly even more so than the wild-type DDX3Y. However, it is also possible that the greater impact of DDX3X knockdown may be due to the more efficient silencing of DDX3X than of DDX3Y.

In summary, depleting DDX3X and DDX3Y using siRNA recapitulated partially the transcriptional changes caused by inhibition of DDX3 by RK-33 with the induction of stress-related gene expression signatures. These data further support a model in which a catalytically inactive mutant forms of DDX3X may have non-canonical roles in Burkitt lymphoma.

To identify modulators of sensitivity or resistance to DDX3X inhibition by RK-33, a genome-wide CRISPR/Cas9 chemogenomic screen was conducted in NALM-6 cells, a human pre-B acute lymphoblastic leukemia cell line for which the screening protocol had been previously established (Bertomeu et al., 2018). This screen revealed a range of genes whose knockout either rescued cells from the effects of RK-33 or rendered them more sensitive to it (Figure 5A). Notably, DDX3X itself emerged as a “rescuer” hit, suggesting that complete loss of the gene can mitigate the cytotoxic impact of its pharmacological inhibition. This finding further supported our hypothesis that catalytically inactive DDX3X may retain abnormal functions within the cell, the loss of which reduces the detrimental effects induced by RK-33. Among the most significant rescue hits were several genes involved in the glutathione metabolic pathway, including GCLM, GSTO1, GCLC, GSS, and GSR (Figures 5A,B) and an ontology analysis of the top hits from the rescue group confirmed the enrichment of glutathione-related pathways (Figure 5B), which are central to the regulation of intracellular reactive oxygen species (ROS).

To further explore whether glutathione metabolism is altered upon RK-33 treatment, we re-examined RNA sequencing data from RK-33-treated Raji and CA46 BL cell lines. Gene set enrichment analysis using the KEGG glutathione metabolism pathway, revealed that this gene set was negatively correlated with the RK-33 treatment compared to DMSO controls (Figure 5C). This effect was particularly pronounced in Raji cells, which showed a greater sensitivity to RK-33 than the CA46 cells. These observations suggested that RK-33 may increase oxidative stress in BL cells through disruption of glutathione metabolism.

To investigate whether RK-33 directly induces ROS accumulation, we treated Raji and CA46 cells with RK-33 for 24 h at concentrations near their respective CC₅₀ values and quantified intracellular ROS levels using flow cytometry. Both cell lines displayed a significant increase in total ROS following treatment, confirming that RK-33 induces oxidative stress in BL cells (Figure 5D). This increase in ROS is consistent with the involvement of redox imbalances in the cytotoxic mechanism of RK-33.

Given these findings, we next asked whether small molecule inhibitors targeting top screen hits could modulate the cellular response to RK-33. From the sensitizer group, we selected KEAP1 and UBA6, two genes for which pharmacological inhibitors are available. KEAP1, a central component of the Keap1-Nrf2-ARE pathway that regulates oxidative stress responses (Tu et al., 2019), was inhibited using CPUY192018 (5 µM)(Lu et al., 2019) and Omaveloxolone (50 nM)(Reisman et al., 2019). Despite its strong representation among synthetic lethal hits, co-treatment with either inhibitor and RK-33 did not result in a measurable synergy (Figure 5F; Supplementary Figure S6B,C). Similarly, inhibition of UBA6, an E1 ubiquitin-activating enzyme, using inosine (500 µM) (Zhang et al., 2022) and TAK-243 (4 nM) (Barghout et al., 2019), failed to produce any notable enhancement of RK-33 cytotoxicity in either cell line (Figure 5F; Supplementary Figure S6A,D).

We then focused on genes from the rescue group, particularly those involved in glutathione metabolism. Since several glutathione-related enzymes were identified as the most frequent rescuer hits in the screen, we tested whether inhibition of glutathione synthesis would affect RK-33 sensitivity. Cells were treated with Buthionine Sulfoximine (BSO, 7.5 µM), an inhibitor of glutamate-cysteine ligase (GCL), the first enzyme in glutathione biosynthesis (Griffith and Meister, 1979). Interestingly, whereas the chemogenomic screen in NALM-6 cells suggested that loss of glutathione pathway components would rescue RK-33 toxicity, treatment of Raji and CA46 cells with BSO resulted in a marked synthetic lethality when combined with RK-33 (Figures 5E,F). This discrepancy may be due to lineage-specific differences in redox buffering capacity and suggests that BL cells are dependent on the glutathione synthesis pathway to counterbalance RK-33-induced oxidative stress, whereas the NALM-6 cells might be relying more on the Keap1-Nrf2-ARE pathway to achieve the same results.

To assess if DDX3X inhibition would affect normal B cells differently than BL cell lines, we used primary human B cells from peripheral blood of healthy donors who had undergone leukapheresis (Figure 6A). The RK-33 treatment produced widespread transcriptional alterations in primary B cells which showed similarities to Raji cell changes, but at reduced intensity (Figures 6B,C).

Hallmark analysis using ShinyGO showed that RK-33 treatment caused significant enrichment of stress-related hallmarks including oxidative stress responses together with cell cycle regulation and general protein metabolism, which showed substantial overlap with Raji cell results (Figure 6D). In addition, the KEGG pathways identified by GSEA and ShinyGO further supported an effect of general cellular stress programs (Figures 6E,F) with affected pathways appearing to be deregulated in a stochastic manner with no clear indication of selective targeting. This strongly suggested that the transcriptional changes may result from a general cellular destabilization rather than specific pathway inhibition.

Interestingly, although ROS-related pathways showed modest enrichment following RK-33 treatment, the glutathione metabolism pathway exhibited an opposite and much milder trend in primary B cells (Figure 6G) compared to BL cell lines. This suggested that the mechanisms leading to ROS accumulation in normal B cells upon RK-33 treatment might differ from those in malignant B cells despite oxidative stress being a common outcome of DDX3X inhibition in normal and malignant cells.