Abstract

   Approximately 30% of human cancers carry various RAS mutations,
   including KRAS, NRAS, and HRAS. Among these mutations, KRAS is the most
   prevalent isoform detected in lung cancer. While several small
   molecular inhibitors targeting specifically KRAS^G12C have been
   developed and tested clinically, alternative approaches are still
   necessary due to expected drug resistance. In this study, we present
   evidence that the loss of DDX3X significantly delays tumor progression
   in various KRAS-driven lung cancer models. Inhibition of DDX3X disrupts
   cysteine and glutathione metabolism, thereby inducing ferroptosis in
   lung cancer cells. This effect is primarily mediated by the
   downregulation of Cystathionine-β-synthase (CBS), the rate-limiting
   enzyme in cysteine generation. Mechanistically, DDX3X directly binds to
   the transcription factor JUND, which mediates the transcriptional
   regulation of METTL16, a key N^6-methyladenosine methyltransferase, and
   subsequently regulates m^6A modification and translation of CBS
   transcripts. This cascade induces hypermethylation and high expression
   of CBS, consequently triggering cysteine production and maintaining
   antioxidative homeostasis, which is essential for the survival of
   KRAS-driven lung cancer cells. Finally, we demonstrate that a newly
   developed DDX3X PROTAC degrader J10 efficiently delays lung cancer
   progression with multiple advantages compared to DDX3X small molecular
   inhibitor RK-33 and limited side effects. These findings unveil the
   potential of DDX3X as a valuable target for adjuvant therapies in
   managing KRAS-driven lung cancer.

   Subject terms: Cancer, Drug development

Introduction

   Lung cancer stands as the leading cause of cancer-related fatalities
   worldwide, with non-small cell lung cancer (NSCLC) accounting for
   approximately 85% of all cases [[68]1]. NSCLC frequently arises from
   oncogenic mutations, including EGFR, KRAS, ALK, MET and so on [[69]2].
   For a considerable time, KRAS has been perceived as an elusive target
   for drug development. However, recent advances have led to the creation
   and clinical testing of small molecular inhibitors specifically
   targeting KRAS^G12C [[70]3]. Despite these strides, it is indicated
   that lung cancer cells can swiftly develop resistance to these
   treatments [[71]4]. Consequently, there remains an urgent and pressing
   need to decipher the molecular mechanisms driving KRAS-driven lung
   cancer progression and to explore alternative therapeutic strategies.

   DDX3X is an RNA-binding protein classified within the DEAD-box helicase
   family. It serves diverse functions in RNA homeostasis, encompassing
   transcriptional regulation [[72]5], splicing [[73]6], mRNA stability
   [[74]7], mRNA nuclear export [[75]8], ribosomal biogenesis [[76]9], and
   translation [[77]10]. Accumulating evidence underscores DDX3X
   involvement in various cellular processes, including cell-cycle
   regulation [[78]11], innate immune response [[79]12], apoptosis
   [[80]13], DNA damage [[81]14], embryogenesis [[82]15] and stress
   response [[83]16]. Whether DDX3X is involved in cellular metabolic
   homeostasis remains poorly understood. While genetic alterations in
   DDX3X have been identified in several cancers, its role in cancer
   progression remains a subject of debate and varies depending on the
   cancer context. For instance, DDX3X acts as an oncogenic gene in
   pancreatic cancer, regulating either transcription or translation
   pathways to promote tumor progression [[84]17]. DDX3X enhances tumor
   invasion or cell growth in colorectal cancer [[85]18]. DDX3 regulates
   castration-resistant prostate cancer by mediating posttranscriptional
   regulation of the androgen receptor [[86]19]. The loss of DDX3X
   perturbed lymphomagenesis in a sex-dependent manner [[87]20]. However,
   recent studies have revealed that DDX3 might be a tumor suppressor in
   specific cancer contexts. In advanced colorectal cancer [[88]21] and
   p53-inactivated lung tumors [[89]22], diminished expression of DDX3 was
   observed, correlating with poor prognosis in patients, while in vitro
   experiments, loss of DDX3 promoted soft-agar growth and invasion. These
   observations highlight the important and multiple roles of DDX3X in
   cancer progression.

   In lung cancer, high expression of DDX3X has been associated with more
   aggressive tumor behavior. Inhibition of DDX3X disrupts the
   DDX3X-β-catenin axis and the non-homologous end joining (NHEJ) DNA
   damage repair pathway [[90]23]. A small molecule DDX3X inhibitor,
   RK-33, has demonstrated significant potential in promoting tumor
   regression and enhancing sensitivity to radiotherapy [[91]23]. In our
   study, by utilizing a spontaneous KRAS-driven lung cancer model, human
   lung cancer organoids, and cell lines, we have unveiled a pivotal role
   of DDX3X in lung cancer progression. Specifically, inhibiting DDX3X
   disrupts cysteine and glutathione anabolism, consequently inducing
   ferroptosis in lung cancer cells. This effect is primarily mediated by
   the suppression of CBS translation through JUND/METTL16-mediated m^6A
   modification. Additionally, we have developed novel DDX3X PROTAC
   degraders and compared their anti-tumor efficiency with RK-33,
   showcasing their great potential and advantages in targeting DDX3X for
   lung cancer therapeutics. In conclusion, our findings elucidate the
   critical function of DDX3X in maintaining antioxidant homeostasis and
   provide an alternative strategy to treat KRAS-driven lung cancer.

Results

Disrupting Ddx3x impedes lung cancer progression and markedly extends
survival in mice harboring Kras^G12D-driven lung cancer

   To investigate the correlation between DDX3X expression and lung cancer
   patient prognosis, we conducted a comprehensive analysis of survival
   data from lung cancer cases profiled by TCGA. Notably, elevated DDX3X
   expression was associated with poorer overall survival and
   progression-free intervals in lung adenocarcinoma (LUAD) (Fig. S[92]1A,
   B). To explore the role of DDX3X in lung cancer, we initially bred
   Ddx3x ^fl/ fl mice to lox-stop-lox-Kras^G12D mice to generate
   Kras^G12D;Ddx3x^fl/fl and Kras^G12D;Ddx3x^fl/y offspring. Lung cancer
   was induced in a stepwise process via intratracheal administration of
   Ad5-Spc-Cre virus [[93]24]. Notably, we observed no lung cancer related
   deaths in Kras^G12D;Ddx3x^fl/fl and only a limited number in
   Kras^G12D;Ddx3x^fl/y mice (two out of 12), underscoring a remarkable
   survival advantage compare to their littermates (Fig. S[94]1C).
   Subsequently, by introducing p53^fl/fl allele, we obtained
   Kras^G12D;Ddx3x^fl/fl p53^fl/fl (termed KP;Ddx3x^fl/fl hereafter),
   KP;Ddx3x^fl/+ and KP;Ddx3x^fl/y mice, similarly, our data showed that
   depletion of Ddx3x also prolonged survival of mice bearing lung cancer
   dramatically (Fig. [95]1A), such survival benefit was also evident in
   KP;Ddx3x^fl/+ mice when only a single Ddx3x gene copy was eliminated
   (Fig. S[96]1D). Furthermore, histological evaluation 12 weeks after
   tumor induction demonstrated much less tumor burden and fewer tumor
   foci in KP;Ddx3x^fl/fl and KP;Ddx3x^fl/y mice compared to their DDX3X
   competent littermates (Fig. [97]1B, C). Consistently, micro-CT analysis
   also demonstrated that deletion of DDX3X significantly delayed lung
   tumor progression (Fig. [98]1D and Fig. S[99]1E). Next, we purified
   primary lung tumor cells and conducted tumor spheroid assays, which
   indicated that ablation of DDX3X significantly impaired tumor spheroid
   formation (Fig. [100]1E, F). Similarly, in lung cancer patient-derived
   organoids (PDOs), which faithfully reproduce the morphological and
   genomic features of human lung tumors (Fig. S[101]2), we observed that
   both the size and viability were markedly decreased in DDX3X deficient
   lung tumor organoids (Fig. [102]1G–I). Moreover, we generated
   DDX3X-knock down cells in a KRAS mutant human lung cancer cell A549
   (KRAS^G12S) and found suppression of DDX3X halted the proliferation of
   A549 cells (Fig. [103]1J–M). A similar phenotype was reproduced in PC-9
   cells (KRAS wildtype) and NCI-H23 cells (KRAS^G12C) (Fig. S[104]1F–M).

Fig. 1. Depletion of Ddx3x delays lung cancer progression and markedly
extends survival in mice harboring Kras^G12D-driven lung cancer.

   [105]Fig. 1
   [106]Open in a new tab

   A Kaplan-Meier analysis of the overall survival of mice bearing lung
   cancer: KP;Ddx3x^+/+ (n = 17), KP;Ddx3x^fl/fl (n = 17), KP;Ddx3x^+/y
   (n = 25), and KP;Ddx3x^fl/y (n = 25). (Log-rank (Mantel-Cox) test). B
   Representative hematoxylin and eosin (H&E) stained lung tissue sections
   from 12-week-old mice of KP;Ddx3x^+/+, KP;Ddx3x^fl/fl, KP;Ddx3x^+/y,
   and KP;Ddx3x^fl/y mice; Scale bar: 6 mm. C Quantification of lesion
   numbers in (B) (mean ± SEM, unpaired Student’s t test). D
   Representative transverse micro-CT images of wild-type (WT) and DDX3X
   knockout (KO) mice were captured at uniform endpoint event time points.
   The selected images correspond to the location of the largest tumour
   diameter. Tumors are marked with yellow arrows; Scale bar: 1 mm. E
   Representative bright-field microscopy images of lung cancer organoids
   captured on day 6 of culture, derived from mice with genotypes:
   KP;Ddx3x^+/+, KP;Ddx3x^fl/fl, KP;Ddx3x^+/y, and KP;Ddx3x^fl/y; Scale
   bar: 50 μm. F Quantification of the diameters of organoids from (E)
   (mean ± SEM, One-way ANOVA). G Representative bright-field images of
   control and DDX3X knockdown patient-derived lung cancer organoids.
   Scale bar: 50 μm. H Analysis of DDX3X mRNA levels in patient-derived
   organoids following lentiviral transduction with non-targeting shRNA
   (shNS) or shRNA targeting DDX3X (shDDX3X) (n = 3, mean ± SEM, unpaired
   Student’s t test). I Quantification of the viability of patient-derived
   lung cancer organoids using an ATP release assay (CellTiter-Glo 3D)
   (n = 4, mean ± SEM, unpaired Student’s t test). J Immunoblot analysis
   of DDX3X expression in A549 cells transduced with shNS or shDDX3X.
   β-actin was employed as a loading control. K A 5-day period analysis of
   the cell proliferation in shNS or shDDX3X A549 cells (n = 3,
   mean ± SEM, Two-way ANOVA). L Clonogenic assay of A549 cells transduced
   with shNS or shDDX3X, the colonies were stained and assessed after a
   2-week incubation period. M Quantification of colony formation in A549
   cells transduced with shNS or shDDX3X. (n = 3, mean ± SEM, unpaired
   Student’s t test).

DDX3X is a pivotal regulator of cysteine-glutathione metabolism and
ferroptosis in lung cancer cells

   To study the role of DDX3X in lung cancer progression, we undertook a
   label-free proteomic analysis using DDX3X knockdown (shDDX3X) A549
   cells. Our findings unveiled a significant alteration in 248
   upregulated proteins and 192 downregulated proteins (Fig. [107]2A).
   Pathway enrichment analysis (PEA) highlighted the enrichment of
   proteins associated with glutathione metabolism, ferroptosis, and
   cysteine and methionine metabolism in DDX3X knockdown cells (Fig.
   [108]2B). Concurrently, we executed an untargeted metabolome analysis
   on the DDX3X knockdown A549 cells. Consistent with the proteomic
   findings, key metabolites linked to cysteine and glutathione including
   s-adenosylmethionine (SAM), glutathione (GSH), glutathione oxidized
   (GSSG), and γ-Glutamylcysteine (gamma-Glu-Cys) [[109]25] were markedly
   reduced in shDDX3X cells (Fig. S[110]3A). This was further corroborated
   through KEGG metabolic enrichment pathway analysis via MetaboAnalyst
   [[111]26](Fig. S[112]3B) and targeted liquid chromatography with tandem
   mass spectrometry (LC-MS/MS) (Fig. S[113]3C, D). Integrating the
   proteomic and metabolic datasets, we discovered that enzymes pivotal to
   cysteine biosynthesis, such as cystathionine beta-synthase (CBS)
   [[114]27], and key glutathione biosynthesis genes, including
   glutamate-cysteine ligase catalytic subunit (GCLC) [[115]28] and
   Glutamate-cysteine ligase regulatory subunit (GCLM) [[116]29], were key
   downstream targets downregulated upon DDX3X inhibition (Fig. [117]2C).
   This was further validated through immunoblotting assays that confirmed
   the downregulation of CBS, GCLC, and GCLM in shDDX3X A549 cells (Fig.
   [118]2D), NCI-H23 cells (Fig. S[119]3E), and primary lung tumor cells
   from KP;Ddx3x^fl/y mice (Fig. [120]2E). Accordingly, we observed a
   significant decrease in cysteine levels, which is an antioxidant
   metabolite regulated by cystathionine β-synthase (CBS), in shDDX3X
   cells (Fig. [121]2F and Fig. S[122]3F). Given that compromised cysteine
   and glutathione metabolism can elevate lipid peroxidation [[123]30],
   and with ferroptosis-associated genes also being prominent in our
   proteomic analysis (Fig. [124]2B), we evaluated intracellular Fe^2+ and
   lipid peroxidation levels in shDDX3X cells using Ferene-S and the lipid
   oxidation probe C11 BODIPY 581/591, respectively. Consistently, both
   parameters were elevated in DDX3X-deficient A549, PC-9 and NCI-H23
   cells (Fig. [125]2G–J and Fig. S[126]3G–L). Furthermore, the levels of
   malondialdehyde (MDA) and 4-hydroxynonenal (4-HNE), both markers of
   oxidative stress-induced lipid peroxidation and ferroptosis, were
   significantly elevated in shDDX3X cells (Fig. [127]2K, L, Fig.
   S[128]3M, N), indicating that DDX3X ablation instigates ferroptotic
   cell death in lung cancer cells. Notably, when we treated lung cancer
   cells with erastin, a ferroptosis inducer known to inhibit system
   Xc-mediated cystine import [[129]31], we observed enhanced resistance
   to erastin treatment in shDDX3X cells (Fig. [130]2M). This was
   attributed to diminished efficacy of erastin in elevating lipid
   peroxidation levels in DDX3X-deficient cells compared to their
   wild-type control cells (Fig. [131]2N, O), likely due to the inherent
   saturation of lipid peroxidation in the absence of DDX3X.

Fig. 2. DDX3X controls cysteine-glutathione metabolism and ferroptosis in
lung cancer cells.

   [132]Fig. 2
   [133]Open in a new tab

   A Volcano plot illustrates the fold change (FC) and variance for all
   expressed proteins in shNS or shDDX3X A549 cells. Differentially
   expressed proteins (DEP) are denoted by red and blue points, indicating
   up-regulated (fold change>1.5 and p-Value < 0.05) and down-regulated
   (fold change<0.6 and p-Value < 0.05) changes, respectively. B Scatter
   plot shows the enrichment of top KEGG pathways based on differentially
   expressed proteins (DEP) from (A). Point size and color are
   proportional to the DEP hit count (number shown) and q value,
   respectively. The top ten KEGG pathways are ranked by gene ratio. C A
   pathway map highlights proteins and biochemicals within the ferroptosis
   pathway, specifically highlighting Cysteine-Glutathione signature
   proteins as indicated in (A). D Immunoblot analysis of DDX3X, CBS,
   GCLC, and GCLM in A549 cells. β-actin was employed as a loading
   control. E Immunoblot analysis of DDX3X, CBS, GCLC, and GCLM in the
   indicated cells. β-actin was employed as a loading control. F
   Quantification of cysteine levels in A549 shDDX3X or shNS cells. n = 3,
   mean ± SEM with ordinary One-way ANOVA. G Flow cytometry assay of
   ferrous ion (Fe^2+) fluorescence in shNS or shDDX3X A549 cells. H
   Quantification of mean fluorescence intensity (MFI) in (G) (n = 3,
   mean ± SEM, One-way ANOVA). I Flow cytometry analysis assessed lipid
   peroxidation through C11-BODIPY fluorescence in shNS or shDDX3X A549
   cells. FITC-BODIPY C11 and PE-BODIPY C11 represent oxidized (ox) and
   reduced (re) populations, respectively. J Quantification of the ratio
   of oxidized (ox) to reduction (re) fluorescence intensity in (I)
   (n = 3, mean ± SEM, One-way ANOVA). K, L Quantification of
   malondialdehyde (MDA) (K), and 4-hydroxynonenal (4-HNE) (L) levels in
   A549 shDDX3X or shNS cells. n = 3, mean ± SEM with ordinary One-way
   ANOVA. M Cell viability assay of shNS or shDDX3X A549 cells treated
   with or without 40 μM Erastin for 48 h (n = 3, mean ± SEM, Two-way
   ANOVA). N Flow cytometry analysis of shNS or shDDX3X A549 cells in the
   presence or absence of 40 μM Erastin for 48 h to assess lipid
   peroxidation through C11-BODIPY fluorescence. O The relative ratio of
   oxidized (ox) to reduction (re) fluorescence intensity, normalized to
   the untreated condition in (N) (n = 3, mean ± SEM, NS. p > 0.05,
   Two-way ANOVA).

CBS emerges as the pivotal downstream effector of DDX3X in orchestrating lung
cancer progression

   Given the downregulation of CBS, GCLC, and GCLM in DDX3X-deficient lung
   cancer cells, we then questioned which enzyme primarily mediates the
   effects of DDX3X inhibition. Since existing knowledge underscores the
   predominant role of GCLC over GCLM in modulating cancer progression
   [[134]32, [135]33], we thereby individually silenced CBS and GCLC in
   lung cancer cells. Inhibition of CBS markedly suppressed proliferation
   in A549, PC-9 and NCI-H23 cells (Fig. [136]3A–D and Fig. S[137]4A–D),
   accompanied by a notable rise in intracellular Fe^2+ and lipid
   peroxidation levels (Fig. [138]3E–H and Fig. S[139]4E–H). Notably,
   while knocking down GCLC yielded a similar phenotype, its impact was
   comparatively milder than what was observed in CBS suppression (Fig.
   S[140]4I–T). Conversely, the overexpression of either CBS or GCLC in
   A549 cells led to accelerated cell proliferation and diminished Fe^2+
   and lipid peroxidation levels (Fig. S[141]5A–H), underscoring the vital
   roles of these enzymes in modulating antioxidant homeostasis and lung
   cancer progression.

Fig. 3. CBS is the key downstream effector of DDX3X in orchestrating lung
cancer progression.

   [142]Fig. 3
   [143]Open in a new tab

   A Representative western blot showing CBS-silenced A549 cells. β-actin
   was used as loading control. B Assessment of cell proliferation in CBS
   knockdown A549 cells using a CCK-8 assay for 6 days (n = 3, mean ± SEM,
   One-way ANOVA). C Clonogenic assay in control or CBS knockdown A549
   cells. D Quantification of colony formation in (C). E FACs analysis of
   ferrous ion (Fe^2+) levels in CBS knockdown A549 cells. F
   Quantification of the mean fluorescence intensity (MFI) of ferrous ions
   (Fe^2+) in (E) (n = 3, mean ± SEM, One-way ANOVA). G Assessment of
   lipid peroxidation via C11-BODIPY 581/591 staining in control or CBS
   knockdown A549 cells. H Quantification of C11-BODIPY fluorescence
   intensity in the specified cells in (G), Data normalization was
   performed against measurements in shNS for comparative analysis (n = 3,
   mean ± SEM, One-way ANOVA). I Western blot analysis of DDX3X and CBS in
   A549 cells transduced with lentiviral expressing shNS, shDDX3X, shDDX3X
   plus overexpressed (OE) CBS. β-actin was used as loading control. J
   Western blot analysis of DDX3X and GCLC in A549 cells transduced with
   lentiviral shNS, shDDX3X, shDDX3X plus overexpressed GCLC. β-actin was
   used as loading control. K A 6-day assessment of relative cell
   viability for each indicated cell line. The shNS value was used for
   data normalization (n = 3, mean ± SEM, NS. p > 0.05, Two-way ANOVA). L
   Cell viability assay of control or DDX3X knockdown A549 cells treated
   with or without 0.5 mM cysteine for 24 hours. The untreated shNS-A549
   cells was used for data normalization (n = 3, mean ± SEM, unpaired
   Student’s t test). M, N Cell viability analysis of DDX3X knockdown A549
   cells with or without 0.5 mM cysteine supplementation for 24 h, under
   hypoxia alone (M) and in combination with nutrient deprivation (N). O
   Kaplan-Meier survival curves depicting the overall survival of mice
   with lung cancer (KP;Ddx3x^+/y and KP;Ddx3x^fl/y n = 6), with mice
   subjected to either 1 mM cysteine-supplemented water or regular water
   (NS. p > 0.05, Log-rank (Mantel-Cox) test).

   Next, we successfully achieved overexpression of CBS and GCLC in
   DDX3X-deficient cells (Fig. [144]3I, J). Notably, the overexpression of
   CBS nearly restored the growth impairment seen in DDX3X-knockdown
   cells, while GCLC overexpression had a less pronounced effect on the
   proliferation of DDX3X-deficient cells (Fig. [145]3K and Fig.
   S[146]5I–K). Consistent with this, cysteine supplementation—synthesized
   by CBS—enhanced the viability of DDX3X-deficient cells but had little
   impact on DDX3X-competent cells (Fig. [147]3L). Interestingly, cysteine
   was more effective at restoring the viability of DDX3X-knockdown cells
   under stressed conditions (Fig. [148]3M, N). Moreover, the provision of
   cysteine-enriched drinking water expedited lung tumor progression
   solely in KP;Ddx3x^fl/y mice but not in their KP; Ddx3x^+/y littermates
   (Fig. [149]3O). Thus, these findings underscore CBS-mediated cysteine
   metabolism as a pivotal downstream effector in DDX3X-driven lung cancer
   progression.

DDX3X controls the translation of CBS through METTL16

   To elucidate how DDX3X governs CBS expression, we initially assessed
   the production of nascent CBS RNA in DDX3X knockdown and control cells.
   Quantitative real-time polymerase chain reaction (qRT-PCR) with primers
   spanning intron-exon boundaries revealed that DDX3X had no discernible
   impact on the generation of nascent CBS transcripts (Fig. [150]4A).
   Furthermore, DDX3X did not influence the stability of CBS mRNA (Fig.
   [151]4B). Intriguingly, polysome profiling indicated a significant
   impairment in the translation of CBS mRNA in DDX3X-deficient cells
   (Fig. [152]4C). To unravel the mechanism of DDX3X-mediated CBS
   translation, we conducted RNA sequencing in DDX3X knockdown cells,
   revealing 768 downregulated genes and 1050 upregulated genes (Fig.
   [153]4D). Integrating the RNAseq with the proteomics data, focusing
   specifically on downregulated genes/proteins, we identified 11 genes
   that exhibited alterations at both the transcriptional and protein
   levels (Fig. [154]4E). Among these, METTL16, a methyltransferase
   involved in RNA m^6A modification [[155]34, [156]35] and metabolism
   [[157]36], emerged as significantly downregulated in the absence of
   DDX3X (Fig. [158]4E). Intriguingly, the expression of other crucial
   m^6A regulators, including METTL3, METTL4, FTO, and ALKBH5, remained
   unaffected in DDX3X-deficient A549, NCI-H23 cells and primary lung
   tumor cells derived from KP;Ddx3x^fl/y mice (Fig. [159]4F, G, Fig.
   S[160]6A). Consistent with these findings, liquid chromatography-tandem
   mass spectrometry (LC-MS/MS) was employed to assess the m^6A
   modification level of total mRNA, revealing a marked decrease in global
   m^6A modification in DDX3X-knockdown cells (Fig. [161]4H). To establish
   a direct link between METTL16 and CBS, we knocked down METTL16 in A549
   and NCI-H23 cells, resulting in clear inhibition of CBS expression
   (Fig. [162]4I, J) and cell proliferation (Fig. [163]4K–P). These
   findings collectively suggest that METTL16 plays a pivotal role in
   mediating the regulation of CBS expression by DDX3X.

Fig. 4. DDX3X controls the translation efficacy of CBS through METTL16.

   [164]Fig. 4
   [165]Open in a new tab

   A Quantitative real-time PCR analysis of pre-mRNA levels measured by in
   shNS or shDDX3X A549 cells (n = 3, mean ± SEM, NS. p > 0.05, unpaired
   Student’s t test). B Assessment of CBS mRNA stability in shNS or
   shDDX3X A549 cells (n = 5, mean ± SEM, NS. p > 0.05, unpaired Student’s
   t test). C Polysomal analysis of CBS mRNA in shNS or shDDX3X A549
   cells, mean CBS mRNA abundance in each polysomal fraction was presented
   (n = 3, mean ± SEM, One-way ANOVA). D Volcano plot from RNA-seq data
   showing downregulated mRNAs (fold change <0.6 and p-value <0.05, green
   dots) and upregulated mRNAs (fold change >1.5 and p-value <0.05, orange
   dots) in A549 cells upon DDX3X knockdown. E Venn diagram highlights
   overlaps between downregulated transcripts and proteins in shDDX3X A549
   cells compared with shNS control cells. F, G Representative western
   blot analysis of DDX3X, METTL16, METTL3, METTL14, FTO, and ALKBH5 in
   DDX3X knockdown A549 cells (F) and DDX3X knockout primary lung cancer
   cells (G). GAPDH was used as loading control. H Analysis of the
   m^6A-to-A ratio in mRNAs from shNS or shDDX3X A549 cells, measured by
   LC-MS/MS (n = 6, mean ± SEM, One-way ANOVA). I, J Representative
   analysis of METTL16 and CBS protein level in METTL16-knockdown A549
   cells (I) and NCI-H23 cells (J). K, L A 6-day period proliferation
   assay of control or METTL16 knockdown A549 cells (K) and NCI-H23 cells
   (L) (n = 3, mean ± SEM, Two-way ANOVA). M, N Colony formation assay in
   control or METTL16 knockdown A549 cells (M) and NCI-H23 cells (N), cell
   colonies were stained and assessed after a 2-week incubation period. O,
   P Quantification of colony formation in (M) and (N), respectively.
   (n = 3, mean ± SEM, unpaired Student’s t test).

METTL16-dependent m^6A modification is required for DDX3X-regulated CBS
expression and lung cancer proliferation

   To investigate the correlation between METTL16 expression and lung
   cancer patient prognosis, we conducted a comprehensive analysis of
   survival data from lung cancer cases profiled by TCGA. Notably,
   elevated METTL16 expression was associated with poorer overall survival
   and progression-free intervals in lung adenocarcinoma (LUAD) (Fig.
   S[166]6B, C). Next, we wondered whether METTL16-mediated m^6A
   modification was required for DDX3X-regulated CBS expression. Thus, a
   mutant variant of METTL16 (devoid of m^6A modification activity) (Table
   S[167]5) was established in A549 cells. Notably, only overexpression of
   wildtype METTL16, but not the mutant METTL16 led to elevated CBS
   expression (Fig. [168]5A). As expected [[169]37], only overexpressing
   wildtype METTL16 triggers lung cancer cells proliferation (Fig.
   [170]5B–D). With strict peak-calling criteria, we observed altered m^6A
   peaks in shDDX3X as compared to shNS A549 cells, with a predominant
   clustering toward the boundary between the 3’UTR and CDS (Fig.
   S[171]6D–F). Notably, these modified peaks were enriched for the
   canonical DRACH motif associated with sites of m^6A modification (Fig.
   S[172]6G).

Fig. 5. METTL16-dependent m^6A modification is required for DDX3X regulated
CBS expression and lung cancer proliferation.

   [173]Fig. 5
   [174]Open in a new tab

   A Protein expression of METTL16, and CBS in overexpressed WT or Mut
   METTL16 A549 cells. β-actin was employed as a loading control. B A
   5-day period proliferation assay of overexpressed WT or Mut METTL16
   A549 cells (n = 3, mean ± SEM, NS. p > 0.05, Two-way ANOVA). C
   Representative crystal violet-stained plates depicting the clonogenic
   capacity of A549 cells with wild-type (WT) or mutant (Mut) METTL16
   expression. D Quantification of colony formation in (C). E IGV
   snapshots of MeRIP-seq reads along CBS mRNAs, showcasing differentially
   methylated sites in shDDX3X compared to shNS A549 cells. Red peaks
   indicate methylated sites. Dark peaks indicate input. F, G
   m^6A-RIP-qPCR analysis of CBS 3’ UTR fragments. Quantitative assessment
   of m^6A modification in CBS 3’UTR fragments in DDX3X knockdown (F) or
   METTL16 knockdown (G) A549 cells (n = 3, mean ± SEM, NS. p > 0.05,
   unpaired Student’s t test). H-J Evaluation of luciferase reporter
   activity in DDX3X knockdown (H), METTL16 knockdown (I) and
   overexpressed WT or Mut METTL16 (J) A549 cells with different 3’UTR
   mutants, Renilla luciferase activity served as an internal control
   (n = 3, mean ± SEM, NS. p > 0.05, unpaired Student’s t test). K
   Evaluation of METTL16 and DDX3X expression in A549 cells transduced
   with lentiviral shNS or shDDX3X as well as overexpressed WT or Mut
   METTL16. β-actin was employed as a loading control. L Quantification of
   relative cell viability in A549 cells transduced with shNS or shDDX3X
   as well as overexpressed WT or Mut METTL16 (n = 3, mean ± SEM, NS.
   p > 0.05, Two-way ANOVA). M Colony formation assay of A549 cells
   transduced with shNS or shDDX3X as well as overexpressed WT or Mut
   METTL16, cell colonies were stained and evaluated after a 2-week
   incubation period. N Quantification of colony formation in (M).

   Importantly, hypomethylated differential peaks were observed in the CBS
   3’UTR region affected by DDX3X knockdown (Fig. [175]5E and Fig.
   S[176]6H). Using the sequence-based m^6A modification site predictor
   SRAMP [[177]38], we evaluated the m^6A level of 3 sites (3’UTR-1,
   3’UTR-2, 3’UTR-3) in the 3’UTR domain of CBS mRNA and observed a
   significant decreased m^6A modification in all these sites in DDX3X
   knockdown cells (Fig. [178]5F), similar phenotype was also found in
   METTL16 knockdown cells (Fig. [179]5G). To examine which m^6A
   modification sites were required for CBS expression, we generated 3
   mutants using dual-luciferase reporter vectors containing every segment
   of CBS 3’UTR with different m^6A modification sites’ mutation (Fig.
   S[180]6I). Our data suggested that mutation of 3’UTR-1 had no obvious
   effect to rescue the phenotype caused by DDX3X or METTL16 deficiency,
   whereas mutation of 3’UTR-2 or 3’UTR-3 fully eliminated the outcome
   induced by DDX3X or METTL16 inhibition (Fig. [181]5H, I), suggesting
   these two m^6A modification sites were essential to mediate the METTL16
   regulation on CBS expression. In line with this, overexpression of
   wildtype METLL16 enhanced the expression of CBS with 3’UTR-1 mutation
   but not with 3’UTR-2 or 3’UTR-3 mutation, further confirming the
   importance of these two sites in controlling CBS expression (Fig.
   [182]5J). Finally, we introduced overexpression of either wildtype or
   mutant METTL16 in different DDX3X knockdown cells (Fig. [183]5K), and
   only wildtype METTL16 could restore the proliferation and growth
   disadvantage in DDX3X deficient cells (Fig. [184]5L–N, Fig. S[185]6J,
   K).

DDX3X orchestrates METTL16 transcription through direct interaction with JUND
transcripts

   To elucidate the mechanism by which DDX3X controls METTL16 expression,
   we initiated a study employing qRT-PCR primers designed to span both
   introns and exons of METTL16 [[186]39], facilitating the detection of
   nascent RNA production (Fig. S[187]7A). Notably, a significant
   reduction in nascent METTL16 RNA levels was observed in DDX3X knockdown
   A549 and PC-9 cells (Fig. [188]6A and Fig. S[189]7B). Intriguingly,
   there was no significant difference in the decay of METTL16 mRNA when
   DDX3X was deficient (Fig. [190]6B). Importantly, the overall METTL16
   mRNA levels exhibited a decrease in DDX3X knockdown A549 cells and PC-9
   cells (Fig. [191]6C and Fig. S[192]7C), with no obvious alterations in
   METTL16 translation efficiency in the absence of DDX3X (Fig. [193]6D).
   These findings strongly suggest that DDX3X regulates METTL16 expression
   at the transcriptional level. To delve into the regulatory mechanism of
   DDX3X over METTL16 transcription, we conducted an integrated study,
   combining differentially expressed proteins (DEPs) from our proteomics
   datasets, DDX3X-binding transcripts from the POSTAR3 database, and
   predicted transcription factors for METTL16 using both hTFtarget and
   ENCODE databases. Through this approach, two downregulated
   transcription factors (JUND, CHD1) [[194]40, [195]41] and one
   upregulated transcription factor (MAX) [[196]42] were identified as
   direct binding targets of DDX3X involved in METTL16 transcriptional
   regulation (Fig. [197]6E). Next, RNA immunoprecipitation (RIP) followed
   by qRT-PCR analyses revealed a significant enrichment of the JUND
   transcript with DDX3X protein, confirming the direct interaction
   between JUND mRNA and DDX3X protein endogenously (Fig. [198]6F).
   Notably, knocking down of DDX3X resulted in a reduction in JUND
   expression at both the mRNA and protein levels (Fig. [199]6G, H),
   however, DDX3X inhibition had no effect on nascent JUND transcript
   production but significantly impaired the stability of JUND mRNA (Fig.
   [200]6I,J). Given that JUND is a transcription factor in lung cancer
   progression [[201]43], we further investigated whether JUND acts as a
   nexus between METTL16 and DDX3X. Silencing JUND with siRNA in A549 and
   PC-9 cells clearly reduced METTL16 nascent RNA, mRNA production, and
   protein expression (Fig. [202]6K, L, Fig. S[203]7D, E). Finally, by
   performing co-immunoprecipitation and immunofluorescence assay, we
   excluded the possibility of direct interaction between DDX3X and JUND
   in lung cancer cells (Fig. S[204]7F–I). These observations substantiate
   our hypothesis that JUND plays a crucial role in mediating the
   transcriptional regulation of METTL16 by DDX3X.

Fig. 6. DDX3X regulates METTL16 transcription through direct interaction with
JUND transcripts.

   [205]Fig. 6
   [206]Open in a new tab

   A Quantification of METTL16 pre-mRNA in A549 cells transduced with shNS
   or shDDX3X (n = 3, mean ± SEM, One-way ANOVA). B Degradation assay of
   METTL16 mRNA evaluated by qRT-PCR in shNS or shDDX3X A549 cells treated
   with Actinomycin D (n = 3, mean ± SEM, NS. p > 0.05, Two-way ANOVA). C
   qRT-PCR analysis of METTL16 total mRNA in control or DDX3X knockdown
   A549 cells (n = 3, mean ± SEM, One-way ANOVA). D Mean abundance of
   METTL16 mRNA in each polysome fraction from polysomal analysis
   conducted in shNS and shDDX3X A549 cells (n = 3, mean ± SEM, NS.
   p > 0.05, One-way ANOVA). E Venn diagram showing the distribution of
   transcripts among the CHIP dataset of transcription factors for METTL16
   (hTFtarget or ENCODE database), DEPs (differential proteins upon
   DDX3X-knockdown), and DDX3X-directly bound transcripts (RBPDP
   database). The shared transcripts among all these datasets were
   highlighted. F Enrichment of JUND, CHD1, and MAX mRNA was assessed
   through an anti-DDX3X immunoprecipitation assay followed by qRT-PCR
   with normalization to input mRNA levels (n = 3, mean ± SEM, NS.
   p > 0.05, unpaired Student’s t test). G qRT-PCR assay of relative
   levels of JUND mRNA in shNS or shDDX3X A549 cells (n = 3, mean ± SEM,
   One-way ANOVA). H Representative JUND expression in shNS or shDDX3X
   A549 cells. I qRT-PCR assay of relative JUND pre-mRNA levels in shNS or
   shDDX3X A549 cells (n = 3, mean ± SEM, NS. p > 0.05, One-way ANOVA). J
   Degradation assay of JUND mRNA analyzed by qRT-PCR in shNS or shDDX3X
   A549 cells treated with Actinomycin D (n = 3, mean ± SEM, Two-way
   ANOVA). K Relative pre-mRNA and mRNA levels of METTL16 in control or
   JUND knockdown A549 cells (n = 3, mean ± SEM, unpaired Student’s t
   test). L Representative expression of JUND, METTL16, and CBS in control
   and JUND knockdown A549 cells.

   Finally, we evaluated the prognostic value of DDX3X and its associated
   genes by interrogating DDX3X/CBS, DDX3X/METTL16 and DDX3X/JUND
   respectively using TCGA. Our data indicated that patients with higher
   expression of DDX3X/CBS, DDX3X/METTL16 and DDX3X/JUND exhibited poorer
   overall survival and progression-free interval (Fig. S[207]8A–F),
   suggesting the prognostic value of DDX3X and its associated genes for
   lung cancer patients.

The DDX3X degrader J10 shows enhanced cytotoxic activity relative to RK-33

   In our pursuit of evaluating DDX3X as a therapeutic target in lung
   cancer, we conceived twelve potential heterobifunctional DDX3X
   degraders using Proteolysis Targeting Chimeras (PROTACs) technology
   [[208]44], which leverage the ubiquitin-proteasome system to induce
   DDX3X degradation (Fig. [209]7A). Drawing inspiration from the reported
   binding modes of the known DDX3X inhibitor RK-33 [[210]11, [211]23],
   which demonstrated that the methoxybenzyl group on the imidazolone
   moiety positioned toward the solvent-exposed region could serve as a
   starting point for molecular design. We connected RK-33 to cereblon
   (CRBN) ligand thalidomide, generating PROTAC DDX3X degraders with
   linker arms of varying lengths for the investigation of DDX3X
   protein-degrading activity and antitumor efficacy. The synthesis of the
   diimidazolodiazepine DDX3X PROTAC degraders used in this study is
   outlined (Fig. S[212]9A). Our initial assessment of DDX3X protein
   degradation efficacy revealed that among the compounds tested, J10
   exhibited superior degrading activity compared to others, suggesting a
   preference for a linker with a suitable length at the 4-position of
   thalidomide (Fig. [213]7B). In addition, we showed that J10 efficiently
   degraded DDX3X in all 6 lung cancer cell lines tested in this study
   (Fig. [214]7C). Subsequently, we evaluated the impact of these
   compounds on lung cancer cell viability in A549 cell lines. Notably,
   J8, J10, J11, J12, J18, and J20 demonstrated the ability to suppress
   lung cancer cell growth, with IC[50] values of 0.52, 0.15, 1.14, 0.41,
   4.2, and 1.1 μM, respectively (Fig. [215]7D, E). To ensure these
   effects were not cell-line specific, we investigated the impact of J8
   and J10 in two other lung cancer cell lines with elevated DDX3X
   expression, NCI-H460 and NCI-H1975, obtaining similar results (Fig.
   S[216]9B–D). However, the overall cytotoxic effects of RK-33 and J10
   were comparable in these lung cancer cell lines (Fig. [217]7F, G).
   Given that KRAS^G12D is the most frequent KRAS mutation in both
   pancreatic and colorectal cancers, we further extended our analysis to
   include KRAS^G12D-mutant pancreatic (PANC-1) and colorectal (LS174T)
   cell lines. Consistent with our previous findings, J10 treatment
   effectively decreased DDX3X protein levels (Fig. [218]7H) and showed
   comparable cytotoxicity with RK-33 in pancreatic cancer and colon
   cancer (Fig. [219]7I, J).

Fig. 7. The DDX3X degrader J10 exhibits anti-tumor activity against lung
cancer.

   [220]Fig. 7
   [221]Open in a new tab

   A Chemical structures of PROTAC DDX3X degraders J4, J5, J6, J8, J10,
   J11, J12, J14, J15, J16, J18, J20. B Evaluation of DDX3X protein levels
   in A549 cells treated with different doses of PROTAC DDX3X degraders
   for 48 h (n = 3). C Immunoblot analysis of DDX3X in PC-9, HCC827, A549,
   NCI-H460, NCI-H1975, and NCI-H23 cells after treatment with J10, RK-33
   or DMSO for 24 h. D Calculated IC[50] values from dose response assays
   in (E). E Representative concentration-response curves were determined
   by CCK8 assay after treatment with different PROTAC DDX3X degraders for
   48 h (n = 3, mean ± SEM). F Display of IC[50] values of (G). Data
   represent the mean ± SD from three biological replicates (n = 3). G
   IC[50] analysis of J10 and RK-33 in indicated human lung cancer cell
   lines for 24 h. H Immunoblot analysis of DDX3X protein level after 24 h
   treatment with J10, RK-33, or DMSO (vehicle control) in PANC-1 and
   LS174T cells. β-Actin was used as loading control. I, J Dose-response
   curves and calculated half-maximal inhibitory concentrations (IC[50])
   of J10 and RK-33 in PANC-1 pancreatic ductal adenocarcinoma (I) and
   LS174T colorectal adenocarcinoma cell lines (J) (both harboring
   KRAS^G12D mutations) after 24 h treatment.

The DDX3X degrader J10 exhibits superior tumor-suppressive effects and
enhances radiosensitivity compared to RK-33

   Our next objective was to compare the anti-tumor efficacy of J10 with
   RK-33, a small molecule DDX3X inhibitor that has been shown to have
   anti-tumor effects [[222]23]. We first evaluated the therapeutic
   potential of J10 using an A549 xenograft model, which demonstrated
   significant tumor growth inhibition when administered at 10 mg/kg every
   other day (Fig. S[223]10A–E). Next, we test the antitumor efficacy of
   J10 in spontaneous Kras^G12D; p53^fl/fl lung cancer model, where J10
   treatment significantly extended overall survival and markedly delayed
   disease progression (Fig. S[224]10F, G). Additionally, J10 exhibited
   robust antitumor effects in lung cancer patient-derived organoids
   (PDOs) (Fig. S[225]10H, I).

   We then compared the therapeutic potential of J10 and RK-33 in
   different mouse lung tumor models. In subcutaneous models with A549 and
   PC-9 cells, J10, administered orally at 10 mg/kg, significantly
   outperformed RK-33 at 20 mg/kg in suppressing tumor proliferation,
   growth, and weight (Fig. [226]8A–G). Additionally, in an orthotopic
   lung cancer model, established by injecting KRAS^G12D;p53^-/-
   luciferase-expressing cells into the tail vein, J10 also exhibited a
   superior tumor suppression effect compared to RK-33 under the same
   treatment regimen (Fig. [227]8H–L). We also assessed the potential
   toxicity of RK-33 and J10 in human normal lung epithelial cells
   BEAS-2B. Our results indicated that J10 had only a mild degradative
   effect in BEAS-2B cells (Fig. S[228]11A), and J10 showed significantly
   lower cytotoxicity compared to RK-33 (Fig. S[229]11B), suggesting that
   J10 has a superior safety profile. Remarkably, J10 did not induce
   significant weight loss (Fig. S[230]11C) or hepatotoxicity (Fig.
   S[231]11D). Administering J10 did not cause any systemic toxicity, as
   determined through hematology analyses. Specifically, there were no
   significant changes observed in the counts of platelets, red or white
   blood cells, lymphocytes, monocytes, neutrophils, or eosinophils in
   mice treated with J10 compared to control mice (Fig. S[232]11E–L).
   Intriguingly, J10 showed similar capacity to induce radiosensitization
   as RK-33 in lung cancer cells (Fig. S[233]12A–D). These data
   demonstrate that J10 exhibits superior and more sustained
   tumor-suppressive effects compared to RK-33 in vivo.

Fig. 8. The DDX3X degrader J10 shows superior tumor suppressing effect
compared to RK-33 in vivo.

   [234]Fig. 8
   [235]Open in a new tab

   A Schema for the pharmacodynamic (PD) study in A549 and PC-9
   xenografts. BALB/c nude mice bearing subcutaneous xenografts of A549
   (5 × 10^6 cells) or PC-9 (2.5 × 10^6 cells) were treated with 10 mg/kg
   J10, 20 mg/kg RK-33 or vehicle via oral gavage (p.o.) on day 1, 3, and
   5 each week for 2 weeks and day 1 for the third week. Mice were
   continuously monitored for 10 days after the last treatment. n = 8 mice
   per group were used. B-D The tumor growth curves of A549 xenograft
   tumors (B), representative images of A549 xenograft tumors (C) and
   tumor weight analysis of A549 xenograft tumors (D) treated with
   10 mg/kg J10, 20 mg/kg RK-33 or vehicle. Data was presented as
   mean ± SEM. The tumor growth curves were assessed using two-way ANOVA,
   tumor weight was assessed using one-way ANOVA, n = 8 mice per group.
   E-G The tumor growth curves of PC-9 xenograft tumors (E),
   representative images of PC-9 xenograft tumors (F) and tumor weight
   analysis of PC-9 xenograft tumors (G) treated with 10 mg/kg J10,
   20 mg/kg RK-33 or vehicle. Data was presented as mean ± SEM. The tumor
   growth curves were assessed using two-way ANOVA, tumor weight was
   assessed using one-way ANOVA, n = 8 mice per group. H Schema for the in
   vivo tumor suppressing analysis using KP tumor cells (KRAS^G12D;p53^-/-
   derived from mouse lung adenocarcinoma) in C57BL/6 mice. 5×10^5 KP-luc
   (luciferase) cells was intravenously (through the tail vein) injected
   into C57BL/6 mice. One week after injection, mice were treated with J10
   (10 mg/kg), RK-33 (20 mg/kg), or vehicle via oral gavage (p.o.) three
   times per week for two weeks. After 14 days of treatment, three mice
   from each group (vehicle, J10, and RK-33) were harvested for analysis.
   I Quantification of flux (photons/sec) with an exposure time of 60 s in
   (J). n = 3, mean ± SEM with ordinary One-way ANOVA. J Representative
   bioluminescence images of mice obtained 14 days after treatment with
   vehicle, J10, or RK-33. K Representative images of H&E-stained lung
   tissue sections from mice treated with vehicle, J10, or RK-33 for 14
   days. Scale bar: 2 mm. L Quantification of the lesion area in sections
   shown in (K). n = 3, mean ± SEM with ordinary One-way ANOVA.

Discussion

   Despite the growing recognition of the critical role of RNA metabolism
   in various pathological processes [[236]45], many members of the DDX
   (DEAD-box RNA helicase) family lack established gene-disease
   associations, particularly in cancer progression [[237]46, [238]47].
   However, DDX3X has garnered significant attention in recent years and
   is emerging as a potential therapeutic target for cancer therapy due to
   its frequent mutations and altered expression in a range of
   malignancies [[239]48, [240]49], including prostate cancer [[241]19],
   pancreatic cancer [[242]50], colorectal cancer [[243]13], head and neck
   squamous cell carcinoma (HNSCC) [[244]51], oral squamous cell carcinoma
   [[245]52], breast cancer [[246]53], hepatocellular carcinoma [[247]54],
   lymphomagenesis [[248]20], and lung cancer [[249]22, [250]23]. While
   previous studies have highlighted DDX3X multifaceted roles in lung
   cancer progression and its potential as a therapeutic target, a
   comprehensive understanding of its involvement in lung cancer
   necessitates further in-depth investigations.

   Given that KRAS mutations are prevalent in almost 30% of non-small cell
   lung cancer (NSCLC) patients [[251]2], we chose a Kras^G12D-driven
   spontaneous lung cancer murine model to elucidate the detailed
   mechanism by which DDX3X regulates lung cancer progression, which might
   be also extended to KRAS wildtype lung cancers. Specifically, our study
   focuses on how DDX3X regulates lung cancer metabolism. Our findings
   reveal that DDX3X serves as a key regulator of cysteine and glutathione
   metabolism, thereby controlling ferroptosis in lung cancer cells. This
   discovery establishes a direct link between DDX3X, metabolism, and cell
   death in lung cancer.

   In our subsequent investigations, we sought to elucidate how DDX3X
   regulates CBS expression. Given DDX3X role as a crucial RNA-binding
   protein governing various aspects of gene expression [[252]8, [253]10,
   [254]55], our study in KRAS-driven lung cancer cells revealed that
   DDX3X influences CBS translation without affecting transcription or
   mRNA stability. Upon examining genes downregulated by DDX3X inhibition,
   our integrated analysis pinpointed a methyltransferase, METTL16, known
   for its involvement in RNA m^6A modification [[255]34–[256]36].
   Subsequent analyses confirmed a global downregulation of m^6A in DDX3X
   knockdown cells. Notably, among key m^6A regulators, only METTL16
   expression was suppressed in DDX3X knockdown cells. Furthermore, we
   demonstrated that METTL16-dependent m^6A modification is essential for
   DDX3X-regulated CBS expression and lung cancer proliferation. This
   represents one of the first pieces of evidence [[257]52] showing that
   DDX3X can directly regulate m^6A modification via METTL16. Lastly, we
   uncovered that DDX3X controls METTL16 transcription through a direct
   interaction with JUND transcripts, thus delineating an intact pathway
   elucidating how DDX3X regulates lung cancer progression. These findings
   add a new perspective on metabolic control to supplement the
   multifaceted roles of DDX3X in regulating cancer progression. However,
   we acknowledge the possibility of additional functions of DDX3X in
   tumorigenesis, which requires further investigations.

   While previous studies have highlighted the significance of DDX3X in
   lung cancer progression [[258]22], and small molecule inhibitors such
   as RK-33 have been developed for treating lung cancer [[259]23], no
   clinical trials involving DDX3X inhibitors have been initiated thus
   far. One potential reason for this could be the concern that inhibiting
   DDX3X RNA helicase activity might lead to severe side effects. In this
   study, we have developed a novel DDX3X degrader, J10, using the PROTAC
   strategy. PROTAC degraders offer several advantages over traditional
   small molecule inhibitors, including its intrinsic event-driven
   pharmacology and nonlinear pharmacodynamics, improved safety profiles,
   and reduced risk of drug resistance. Indeed, using different lung
   cancer models, we demonstrated that the DDX3X degrader J10 efficiently
   induces DDX3X degradation and exhibits significantly more anti-tumor
   effects in all in vivo models with less toxicity compared to RK-33.
   Therefore, we propose that inducing DDX3X degradation may be a safer
   approach than inhibiting DDX3X activity, as it might allow for DDX3X to
   perform its essential functions within a controlled time window. This
   strategy holds greater potential for future clinical translation.

   Our study has several limitations. DDX3X plays a crucial role in
   maintaining cell survival and growth, thus generating a DDX3X knockout
   cell line is challenging, as complete elimination of endogenous DDX3X
   is unattainable. Furthermore, upon inhibition of DDX3X, we observed an
   upregulation of DDX3Y expression, a homologous gene located on the Y
   chromosome (Fig. S[260]12E–H). This phenomenon could potentially
   compromise the efficacy of DDX3X inhibition. However, due to the lack
   of a specific antibody for DDX3Y, we were unable to examine the
   distinctions between DDX3X and DDX3Y in this study. While our
   investigation highlights one mechanism by which DDX3X regulates lung
   cancer progression, it does not preclude the existence of other
   potential roles of DDX3X in tumorigenesis. Finally, the applicability
   of our findings to other tumor types necessitates further
   investigation.

Materials and methods

Cell lines

   The A549, PC-9, NCI-H460, NCI-H1975, HCC827, NCI-H23, PANC-1, LS174T,
   HEK293T, and BEAS-2B cells, were purchased from the China Center for
   Type Culture Collection (CCTCC). A549 and BEAS-2B cells were cultured
   in Ham’s F-12K Medium, while PC-9, NCI-H460, NCI-H1975, HCC827and
   NCI-H23 tumor cells were maintained in RPMI 1640 Medium. HEK293T,
   PANC-1, LS174T cells were cultured in high glucose Dulbecco’s modified
   Eagle’s medium (DMEM). The medium was supplemented with 10%
   heat-inactivated fetal bovine serum (FBS) and 1%
   Penicillin-Streptomycin (P/S). All cells were confirmed to be free of
   mycoplasma contamination. Additionally, all cell cultures were
   maintained at 37°C in a humidified incubator under a 5% CO[2]
   atmosphere. All reagent information is detailed in Table S[261]1.

Animal studies

   C57BL/6 and BALB/c nude mice aged 4–6 weeks were purchased from
   Guangzhou Southern Medical University Laboratory Animal Sci.&Tech.Co.
   The Kras^LSL-G12D;p53^flox/flox, and Ddx3x^flox/flox transgenic mice
   were gifts from the Institute of Molecular Biotechnology of the
   Austrian Academy of Science (IMBA). Mice were maintained under
   specific-pathogen free (SPF) conditions in individually ventilated
   cages at 21–22 °C and 39–50% humidity, with a 12-hour light–dark cycle.
   All mouse experiments were performed in compliance with the guidelines
   of the Institutional Animal Care and Use Committee of Southern Medical
   University Nanfang Hospital and were approved under protocol number
   IACUC-LAC-20220926-001.

Subcutaneous implantation of lung cancer xenografts

   A total of 5 × 10^6 A549 or 2.5 × 10^6 PC-9 cells were subcutaneously
   injected into the right flanks of 6-week-old male nude mice. One week
   post-injection, the mice were randomly divided into three groups, with
   8 mice per group (24 mice in total): vehicle (group 1), 10 mg/kg J10
   (group 2), and 20 mg/kg RK-33 (group 3, Ambeed, A179326). J10 and RK-33
   was administered via oral gavage on days 1, 3, and 5 per week for two
   weeks, followed by a one-week observation period without treatment.
   Tumor volume was measured regularly using vernier calipers and
   calculated using the formula [V = ½ (Length × Width²)]. On day 25, mice
   were anesthetized with 1% pentobarbital, and peripheral blood, tumor
   tissues, and organs were collected for subsequent analysis.

Orthotopic model of lung cancer

   Stably transfected KP tumor cells expressing luciferase (KP-Luc) were
   injected into the tail vein of 6-week-old C57BL/6 mice, with each mouse
   receiving a 100 μL PBS suspension containing 5 × 10^5 proliferating
   cells. Seven days post-injection, the mice were randomly assigned to
   one of three groups: a control group and two treatment groups. The
   control group received vehicle treatment, while the treatment groups
   were administered either 10 mg/kg J10 or 20 mg/kg RK-33 via oral gavage
   on days 1, 3, and 5 per week, continuing until day 21. Body weight was
   monitored regularly throughout the study. On day 21, tumor progression
   was evaluated using bioluminescence imaging with the IVIS system
   (Caliper Life Sciences) following intraperitoneal injection of
   D-Luciferin potassium salt (3 mg/kg per mouse, administered 5 minutes
   prior to imaging, Beyotime, ST198) and anesthesia with 1% pentobarbital
   sodium (80 mg/kg per mouse). At the conclusion of the experiment,
   peripheral blood and tissues from the lungs, heart, liver, spleen, and
   kidneys were harvested. Tumor and organ samples were fixed in formalin
   for histopathological analysis, while a portion of the tumor tissue was
   cryopreserved for molecular studies. Cellular immunoassays were
   performed on peripheral blood, and serum samples were analyzed for
   hepatotoxicity.

Intranasal adenovirus administration

   The Cre-expressing adenovirus (Ad5-mSPC-Cre) utilized in this study was
   obtained from Genechem Biotech Co., China. Virus administration was
   carried out via intranasal delivery using previously reported protocols
   [[262]56]. Briefly, 6- to 8-week-old mice were anesthetized with 1.5%
   sodium pentobarbital and positioned on a heating pad. The 62.5 μL
   adenovirus mixture (480 μL Minimum essential medium (MEM), 40 μL
   adenovirus (3 × 10^10 TU/ml), and 5.2 μL of 1 M CaCl₂), was then
   administered intranasally. All viral injections were performed in a
   biosafety level 2+ facility, strictly following the guidelines set by
   the Biosafety Committee of Southern Medical University to ensure safety
   and compliance with biosafety standards. The primers used for
   genotyping the mice are listed in Table S[263]4.

Isolation and culture of primary murine lung tumor cells

   Primary murine lung cancer cells were isolated and purified using
   established protocols. Mice, 10-12 weeks after adenovirus infection,
   were anesthetized, and their thoracic cavities were exposed. Two
   milliliters of digestive solution [4 mL of 2 μg/mL Dispase (Corning,
   354235) and 40 μL of 0.05 mg/mL DNase I (Sigma, DN25)] was then quickly
   instilled through the trachea into the lungs, after which the trachea
   was tied off. The lungs (excluding the heart) were carefully removed
   and incubated in 5 mL of digestive solution in a 15 mL tube for
   60–90 min in a 37 °C water bath. The tube was gently agitated every
   5–8 min during the incubation to facilitate digestion. Following
   digestion, the lung tissue was minced using a scalpel blade, and the
   resulting tissue/cell suspension was filtered twice through a 70 μm
   strainer while grinding. The filtered suspension was then centrifuged
   for 4 min at 1000 rcf. After discarding the supernatant, the cell
   pellet was resuspended in 4 mL of erythrocyte lysis buffer and allowed
   to stand for 3 min to lyse erythrocytes. After this, 15 mL of DMEM was
   added, and the cell suspension was filtered again through a 70 μm
   strainer. The suspension was centrifuged once more, and the cell pellet
   was resuspended in 10 mL of DMEM. To eliminate immune cells, the cell
   suspension was incubated in dishes coated with 50 μL of anti-mouse
   CD16/32 Antibody (Biolegend, 101302, 0.5 mg/mL) for one hour at 37 °C.
   Non-adherent cells were collected, resuspended in DMEM supplemented
   with 10% FBS, and transferred to 10 cm culture dishes. These dishes
   were incubated for one hour at 37 °C to further deplete fibroblasts.
   Finally, the non-adherent cells were centrifuged again, and the cell
   pellet was resuspended in 10 mL of culture medium (Ham’s F-12 medium
   containing 1x GlutaMAX, 15 mM HEPES, 0.8 mM CaCl₂, 0.25% BSA,
   Insulin-transferrin-selenium (ITS, Sigma, I3146), 2% FBS, and 1% P/S).
   The cells were then plated onto 10-cm dishes coated with collagen I
   (Corning, 354236, 200 μg per dish) for further culture.

Patient lung cancer organoids generation and culture

   Patient-derived fresh tissue samples were collected with written
   informed consent, under the approval of the Institutional Review Board
   at Nanfang Hospital of Southern Medical University. Organoid lines were
   developed following established protocols with slight modifications.
   Fresh tissue samples were washed three times with PBS, placed in
   sterile 10 cm petri dishes, and mechanically dissected into small
   pieces before enzymatic digestion. Digestion was performed using 2–3 mL
   of collagenase media [prepared with 4 mL Dispase (Corning, 354235),
   40 μL of 0.05 mg/mL DNase I (Sigma, DN25), 4 μL anti-mouse CD16/32
   antibody (Biolegend, 101302), 200 μL collagenase I (Gibco, 17100017),
   and 10 μM Y-27632 (APExBIO, A3008)]. The mixture was incubated at 37 °C
   on a shaker at 200 rpm until the solution turned cloudy (approximately
   10 minutes). The resulting tissue/cell suspension was filtered through
   a 70 μm strainer while grinding. To eliminate immune cells, the
   filtered suspension was incubated in dishes coated with 50 μL of
   anti-mouse CD16/32 antibody (Biolegend, 101302) for 1 hour at 37 °C.
   The suspension was centrifuged at 1000 rcf for 3 min, and the cell
   pellet was resuspended in tissue-specific culture media, including
   Advanced DMEM/F-12 (Gibco, 12634028) with glutamax (Invitrogen,
   35050061), B27 (Gibco, 17504044), 1.25 mM N-acetylcysteine (Macklin,
   N800425), 50 ng/mL Recombinant Human EGF (R&D, 236-EG-01M), 100 ng/mL
   Recombinant Human FGF-10 (Peprotech, 100-26-5), 100 ng/mL Recombinant
   Human FGF-4 (Peprotech, 100-31-25), 100 ng/mL Recombinant Human Noggin
   (R&D, 6057-NG), 10 μM SB202190 (R&D, 1264), 250 nM CHIR99021 (Sigma,
   SML1046), anti-mouse CD16/32, 10 μM Y-27632 (APExBIO, A3008), and
   500 nM A 83-01(MCE, HY-10432). Up to 50 μL drops of the Matrigel/cell
   suspension were distributed into a 24-well cell suspension culture
   plate (Gibco) and allowed to polymerize for 20 min at 37 °C.
   Tumor-specific primary culture media was then added to each well, with
   media changed every 3 days. All human lung cancer patient’ samples were
   proceeded with approval of Medical Ethics committee of Nanfang hospital
   of Southern Medical University with approve number NFEC-202108-K7.

Culture of murine lung cancer organoids

   5 × 10³ lung tumor cells were mixed with 5 μL of Matrigel (Corning,
   354234) and kept on ice until dispensed as rounded drops onto 12-well
   plates. The plate was incubated at 37 °C for 5–20 min to allow the
   Matrigel to solidify. Once solidified, 1 mL of primary culture medium
   was added, and incubation was continued. Images were collected and
   analyzed after 6 days of incubation.

Hematoxylin and eosin (H&E) and immunohistochemistry (IHC) staining

   H&E staining and IHC staining were performed as previously described.
   In brief, mouse organs or tumor tissues were fixed with formalin or 4%
   paraformaldehyde (PFA) respectively overnight, and dehydrated in
   ethanol, embedded in paraffin, sectioned (4 μm) followed by staining
   with hematoxylin and eosin. For IHC staining, slides were
   de-paraffinized in xylene and ethanol, and rehydrated in water. Antigen
   retrieval was performed by high-pressure heating of slides in sodium
   citrate buffer (pH 6.0) or 1 mM EDTA buffer for 5 minutes. Slides were
   quenched in hydrogen peroxide (3%) to block endogenous peroxidase
   activity. Slides were incubated overnight at 4 °C with primary
   antibody, rinsed with PBS, and subsequently incubated with
   HRP-conjugated secondary antibody at 37 °C for 30 min. Following this,
   slides were stained using a diaminobenzidine (DAB) kit (ZSGB, ZLI-9018)
   according to the manufacturer’s instructions. Following this, the lung
   sections were scanned using a Mirax slide scanner. Visual analysis was
   conducted in a double-blinded manner to ensure unbiased assessment.

Synthetic PROTAC DDX3X degraders

   In the synthesis process, Compound A undergoes a nucleophilic reaction
   with tert-butyl bromoacetate in the presence of cesium carbonate to
   yield Compound B. Compounds C and D undergo a nucleophilic reaction in
   the presence of organic bases to produce Compound E. Subsequently,
   Compounds B and D undergo tert-butyl and tert-butyloxycarbonyl removal,
   respectively, in the presence of trifluoroacetic acid. The resulting
   compounds, F and G, are then evaporated to dryness under reduced
   pressure. Without purification, Compounds F and G directly undergo a
   condensation reaction in the presence of a condensing agent and organic
   base to obtain the diimidazolodiazepine compound described in formula
   (refer to Fig. [264]7A) or a pharmaceutically acceptable salt thereof
   (Fig. S[265]9A).

Lung tumor histology

   Histological analysis was carried out on all lung tumors in accordance
   with established protocols [[266]57]. In summary, the lungs were
   sectioned into 4 μm thick slices from at least three different planes
   and stained with hematoxylin and eosin. Following this, the lung
   sections were scanned using a Mirax slide scanner, and the lung/tumor
   areas were automatically scored using an algorithm programmed and
   executed with the Definiens software suite. Visual analysis was
   conducted in a double-blinded manner to ensure unbiased assessment.

siRNAs and shRNA plasmids

   To generate constitutive knockdown cell lines, plasmid DNA was purified
   from pLKO.1 backbone vectors expressing shRNAs targeting DDX3X, CBS,
   GCLC, METTL16 and a scrambled control. SiRNA was synthesized
   specifically for transient transfection experiments. The target
   sequences for shRNA and siRNA are detailed in Table S[267]2.

Lentiviral production and transduction

   Lentivirus production adhered to established procedures. Briefly,
   HEK293T cells, cultured to approximately 90% confluence on the day of
   transfection, were subjected to transfection using Lipofectamine 3000.
   This involved the introduction of packaging plasmids (psPAX2 and
   pMD2.G) and target plasmids. For every 2 μg of the target plasmid,
   18 μL of Lipofectamine 3000 was employed. The cells were incubated with
   the plasmid/liposome transfection mixture for 8–12 h. Subsequently, the
   medium was replaced with OPTI-MEM supplemented with 5% FBS and 1%
   sodium pyruvate. Lentivirus was harvested 72 h post-transfection. For
   lentiviral transduction, cells were exposed to lentiviral particles for
   8–12 h in the presence of polybrene (10 μg/ml) (Santa Cruz
   Biotechnology).

In vitro hypoxia and nutrient deprivation model

   Cells were cultured to approximately 80% confluence, then the medium
   was replaced with EBSS containing 200 μM cobalt chloride for the
   indicated times [[268]55]. For the hypoxia-only model, cells were
   similarly cultured to 80% confluence and replaced with complete medium
   containing 200 μM cobalt chloride for the indicated times.

Cell proliferation assay

   Cells were seeded in 96-well plates at a density of 3000–4000 cells per
   well and allowed to adhere overnight. After incubation, cells were
   treated with various concentrations of DDX3X degraders for 48 h to
   assess their inhibitory effects. To compare the efficacy of J10 and
   RK-33, cells were exposed to different concentrations of each compound
   for 24 h. The medium was then replaced with 10 μL of CCK-8 solution and
   100 μL of culture medium per well. The plates were agitated for 2 h at
   37 °C before measuring luminescence. The IC[50] values were determined
   using GraphPad Prism.

   For the proliferation assay, cells were seeded in 96-well plates at a
   density of 1000 cells per well and allowed to adhere overnight.
   Following incubation, 10 μL of CCK-8 solution and 100 μL of culture
   medium were added to each well. The plates were incubated for 4 h at
   37 °C before absorbance was measured at 450 nm to obtain the Day 0
   reading. Subsequent readings were taken on different days to monitor
   cell viability, with each measurement referenced to the initial Day 0
   value. All experiments were conducted independently in triplicate.

RIP-qRT-PCR

   RNA immunoprecipitation (RIP) was performed using the Magna RIP Kit
   (Millipore) following the manufacturer’s instructions. Initially, cells
   were cultured in 15 cm dishes, washed with ice-cold PBS, and then
   collected. Subsequently, they were lysed in RIP lysis buffer with added
   protease and RNase inhibitors. DDX3X antibodies and control mouse IgG
   (Millipore) were coupled to protein A/G magnetic beads through a 30 min
   incubation at room temperature. This was followed by three washes and
   subsequent incubation with cell lysate at 4 °C overnight. After six
   additional washes, the beads were resuspended, and the proteins were
   digested using proteinase K at 55 °C for 30 min. The RNA, both input
   and immunoprecipitated, was extracted and subsequently subjected to
   either next-generation sequencing (NGS) or qRT-PCR.

Polysome profiling

   Linear sucrose gradients ranging from 5% to 50% were freshly prepared
   in an ultracentrifuge tube from Beckman, using an automated gradient
   maker (BioComp). Prior to this, cells were treated with 100 μg/ml
   Cycloheximide (CHX) (MCE) at 37 °C for 10 min, washed twice with
   ice-cold PBS containing 100 μg/ml CHX, and then collected.
   Subsequently, the cells were lysed in a hypertonic buffer comprising
   20 mM Tris-HCl (pH 7.4), 300 mM NaCl, 10 mM MgCl[2], 0.5% sodium
   deoxycholate, 1% Triton X-100, 1 mM DTT, 100 µg/ml CHX, and 300 U/ml of
   RNase inhibitor. After lysis, cell debris was removed by centrifugation
   at 16,000 × g for 10 min at 4 °C. Subsequently, 500 µL of supernatant
   was loaded into the sucrose gradients, followed by centrifugation at
   4 °C and 36,000 rpm for 2 h using the SW41Ti rotor from Beckman.
   Samples were then fractionated and analyzed using a Piston Gradient
   Fractionator from BioComp and a fraction collector from Gilson.
   Briefly, a density gradient fractionation system was used to generate a
   polysome profile, distinguishing fractions that correspond to 40S
   (fractions 4 to 6), 60S (fractions 6 to 8), 80S monosomes (fractions 8
   to 10), low-molecular-weight polysomes (fractions 10 to 13, LP), and
   high-molecular-weight polysomes (fractions 13 to 15, HP). For
   normalization purposes, 5 ng of polyadenylated synthetic luciferase
   mRNA (Promega) was added to each fraction. RNA was extracted from each
   fraction and subjected to qRT-PCR analysis.

Label-free proteome

   Begin by resuspending cell pellets in SDT buffer, consisting of 4% SDS
   and 100 mM Tris-HCl at pH 7.6. The subsequent steps involve protein
   digestion using trypsin through the Filter-Aided Sample Preparation
   (FASP) procedure. To each sample, add 40 mM DTT and shake at 37 °C for
   1.5 h. Introduce 20 mM IAA, incubate in darkness for 30 min, and mix
   each sample with 200 μL of 8 M urea before transferring to filters.
   Centrifuge the filters, wash with 8 M urea buffer three times, followed
   by two washes with 25 mM NH[4]HCO[3] buffer. Proceed with trypsin
   digestion by mixing units with trypsin at a 1:100 ratio. Incubate at
   37 °C for 12–18 h, and after the incubation period, spin down the
   condensate from the filters at 1000 x g for 10 s to collect the
   resulting peptides. Perform peptide desalting using C18 Cartridges,
   concentrate peptides through vacuum centrifugation, and reconstitute in
   40 µl of 0.1% formic acid. Estimate peptide content by measuring UV
   light spectral density at 280 nm.

   For LC-MS/MS analysis, use the timsTOF Pro mass spectrometry system
   coupled with Nanoelute. Load peptides onto a C18-reversed phase
   analytical column with a linear gradient of 99.9% acetonitrile and 0.1%
   formic acid. Operate the mass spectrometer in positive ion mode with an
   applied electrospray voltage of 1.5 kV. Employ Parallel Accumulation
   Serial Fragmentation (PASEF) mode for data collection, with specific
   parameters, such as ion mobility coefficient (1/K0) set between 0.6 to
   1.6 Vs cm2, 1 MS scan, and 10 MS/MS PASEF scans. Combine MS raw data
   for each sample and analyze using MaxQuant (v1.6.14) software. Utilize
   specific parameters, including Max Missed Cleavages (2), Fixed
   modification (Carbamidomethyl-C), Variable modification (Oxidation-M),
   MS/MS Tolerance (20 ppm), Protein FDR ( ≤ 0.01), Peptide FDR ( ≤ 0.01).
   Select razor and unique peptides for protein quantification, set a time
   window (match between runs) of 2 minutes, use LFQ for protein
   quantification, and ensure a Minimum ratio count of 1.

MeRIP- qRT-PCR

   For the fragmentation of 10 μg of sample RNA, 1 μL of the 10X
   Fragmentation Buffer (Invitrogen, AM8740) was added to the 10 μL RNA
   sample. The mixture was thoroughly combined, briefly spun, and
   incubated at 70 °C for 3 min in a heating block. Subsequently, 1 μL of
   the Stop Solution was added, and the mixture was placed on ice until
   further use. Five percent of fragmented RNA was preserved as input,
   while the remaining RNA underwent m6A-immunoprecipitation (m^6A-IP)
   using the EpiMark N6-Methyladenosine Enrichment Kit.For m^6A-IP, 1 μg
   of m6A-specific antibody or 1 μg of IgG (Pierce) was added to 40 μL of
   Protein G Magnetic Beads and incubated with orbital rotation for 1 h
   with gentle rotation at RT. The fragmented RNA was then divided into
   two conditions: m^6A-RIP, incubated with 1 μg of m^6A-specific
   antibody, and IgG control, incubated with 1 μg of IgG, rotating head
   over tail for 20 h with gentle rotation at 4 °C. The bead mixture was
   eluted with 75 μL of Monarch RNA Cleanup Binding Buffer (0.5 mg/ml
   N6-Methyladenosine) with gentle rotation at RT for 1 h. Eluted RNA was
   purified using the Monarch RNA Cleanup Kit. Subsequently, input and IP
   RNA were utilized for qRT-PCR, as described above. The m^6A enrichment
   is determined by calculating the IP/input ratio for the gene of
   interest, normalizing for expression levels [[269]58]. Specific primers
   for CBS 3’UTR can be found in Table S[270]3.

Western blot

   Western blotting was conducted following a previously established
   protocol [[271]59]. Total proteins were extracted from cultured cells
   using a Total Protein Extraction Kit according to the manufacturer’s
   instructions. Protein concentrations were determined using a BCA
   Protein Assay Kit. Lysates were then mixed with the appropriate volume
   of 5× loading buffer, boiled, and loaded into SDS-PAGE gels for
   separation. Subsequently, proteins were transferred to polyvinylidene
   difluoride (PVDF) membranes using a Mini Trans-Blot™ system (Bio-Rad).
   The membranes were blocked with 5% skim milk in TBST, followed by
   overnight incubation with primary antibodies at 4 °C, and subsequent
   incubation with HRP-conjugated secondary antibodies for one hour at
   room temperature. Blots were visualized using Immobilon™ Western
   Chemiluminescent HRP Substrate and the ChemiDoc™ XRS + imaging system
   (Bio-Rad). Image analysis and quantification were performed using
   Bio-Rad Image Lab software. The information of antibodies is detailed
   in Table S[272]1.

RNA extraction and qRT–PCR analysis

   Total RNA was extracted using RNAiso plus (Takara) following the
   manufacturer’s protocol and subsequently reverse transcribed using the
   cDNA Reverse Transcription Kit. qRT–PCR reactions were performed with
   the SYBR Green PCR Kit and a ROCHE LightCycler 96 instrument.
   Expression data were normalized to β-actin mRNA levels and presented in
   arbitrary units calculated as 2^^Ct(β−actin−gene of interest). Primers
   used for qRT-PCR can be found in Table S[273]3.

Colony forming assay

   The cells were seeded in a 6-well plate at a density of 500 cells per
   well in 2 mL of media. After a 2-week incubation period, the assays
   were terminated by fixing the cells in a methanol solution for 10 min,
   followed by staining with a 0.5% crystal violet solution for 30 min.
   The colony numbers were manually counted by a blinded observer.

Flow cytometry-based assay

   2 × 10^5 cells were seeded overnight, collected, and labeled with 1 µM
   FerroOrange dye for 15–30 min. After centrifugation, cells were
   resuspended in PBS. Flow cytometry data (10,000 events) were acquired
   using a BD LSRFortessa™ Cell Analyzer and analyzed with FlowJo
   software.

   2 million lung cancer cells were seeded overnight, exposed to Erastin,
   then stained with 1:1000 C11-BODIPY581/591. After digestion and
   collection, cells were resuspended in PBS. Fluorescent intensity was
   measured using BD LSRFortessa™, and lipid peroxidation was evaluated by
   analyzing FITC and PE channel ratios with FlowJo software.

LC-MS/MS-based m^6A level assay

   Total RNA was isolated using RNAiso plus (Takara) following the
   manufacturer’s protocol. mRNA was subsequently extracted using a
   Dynabeads™ mRNA Purification Kit (Invitrogen) as previously described.
   For m^6A quantification, 200 ng polyadenylated mRNAs were digested with
   1.2 U of nuclease P1 (Sigma) dissolved in 25 µl of NH[4]OAc buffer
   (20 mM, pH = 5.3) at 42 °C for 2 h. Then, 3 µl of NH[4]HCO[3] (1 M) and
   1 U alkaline phosphatase (Sigma) were added, followed by an additional
   two hours of incubation at 37 °C. The samples were diluted with formic
   acid to 50 µl, filtered with a 0.22 µm syringe filter (Millipore), and
   5 µl of the solution was injected into the LC-MS/MS. Nucleosides were
   separated by reversed-phase ultra-performance liquid chromatography on
   a C18 column, coupled to mass spectrometry detection with an Orbitrap
   Fusion™ Tribrid™ LC mass spectrometer (Thermo) in positive electrospray
   ionization mode. Nucleoside quantification was based on retention time
   and nucleoside-to-base ion transitions: 268-136 for A, and 282-150 for
   m^6A. Using the standard curve generated from reference standards
   running in the same batch, the concentrations of A and m^6A in the
   samples were calculated. The modification level of m^6A was determined
   as the percentage of m^6A out of the total amount of A, to normalize
   the amount of mRNA injected from different samples.

LC-MS/MS-based GSH and GSSG detection

   In the intricate LC-MS/MS-based sample preparation process, the initial
   steps involved the preparation of Solution 1 (comprising 100% LC-MS
   methanol) and Solution 2 (containing 25 mM ammonium acetate and 2.5 mM
   Na-Ascorbate in LC-MS water). Subsequently, Ellman’s reagent (20 mM)
   was meticulously added to Solution 2 in anticipation of subsequent
   experiments. During the sample collection phase, A549 cells were
   initially gathered at 4 °C and subjected to a brief centrifugation.
   Following this, 240 µL of Solution 1 and 60 µL of Solution 2
   (containing Ellman’s Reagent) were judiciously added. After a brief
   vortexing and sonication, the samples underwent centrifugation, and the
   supernatant, rich in metabolites, was carefully transferred to ice and
   dried using a nitrogen desiccator. LC-MS analyses were promptly
   conducted on the same day as the extraction [[274]60].

ELISA assay

   4-HNE and cysteine levels were measured using ELISA kits: the 4-HNE
   ELISA Kit (MEIMIAN, MM-13089H1) and the cysteine ELISA Kit
   (MyBioSource, MBS269913). Cell samples were lysed in RIPA buffer with
   protease and phosphatase inhibitors, then centrifuged at 12,000 rpm for
   10 min to collect the supernatant. Levels of 4-HNE and cysteine in the
   lysates were measured following the manufacturer’s protocols and
   normalized to protein concentration using the BCA assay (KEYGEN,
   KGB2101).

MDA measurement

   Approximately 1 mL of pre-cooled extraction solution was added to
   prepare the cell lysate, followed by 30 cycles of sonication. The
   lysate was then centrifuged at 8000 g at 4 °C for 10 min, and the
   supernatant was collected. Test reagents were added according to the
   manufacturer’s instructions and incubated in a 100 °C water bath for
   60 min. The mixture was then centrifuged at 10,000 g at room
   temperature for 10 min, and supernatant was used to detect absorbance
   at 532 and 600 nm using a microplate reader (BioTek).

mRNA decay assay

   The lung cancer cells were treated with actinomycin D at a final
   concentration of 5 μg/mL for the specified time intervals, after which
   total RNA was extracted. qRT-PCR was performed as described above. The
   mRNA half-life (t[1/2]) was computed using the formula ln 2/-slope,
   with GAPDH serving as the normalization reference.

Dual luciferase assay

   Fragments of either CBS-3′UTR-WT or CBS-3′UTR-Mut (where m^6A was
   changed to T) were cloned into pmirGLO vectors to generate CBS-3′UTR-WT
   and CBS-3′UTR-Mut plasmids (Table S[275]5). Luciferase activity was
   assessed using the Dual-Luciferase Reporter Assay System (Promega) 72 h
   post-transfection with CBS-3′UTR-WT or CBS-3′UTR-Mut. The relative
   Fluc/Rluc activity was determined by normalizing firefly luciferase
   activity to renilla luciferase activity.

RNA-sequence

   Paired-end reads were sequenced using Illumina NovaSeq 6000 with 150 bp
   read length. Quality control was firstly performed via Q30
   calculations. After adapter trimming with Cutadapt(v 1.14)and
   low-quality read elimination with Trimmomatic(v 0.36), FastQC(v 0.11.5)
   software was used to do some quality control checks. Sequences were
   filtered to remove mitochondrial, ribosomal RNA and tRNA sequences with
   Bowtie2(v 2.2.9). Clean reads were aligned to Homo Sapiens genome
   (GENCODE Release 36, GRCh38.p13) by STAR(v 2.4.0). Reads were counted
   using featureCounts(v 1.6.3). DESeq2(v 1.42.0) was used for
   differential gene expression analysis. (p-value < 0.05 and |Fold
   change | > 1.5).

Metabolome analysis

   Beginning with the careful removal of A549 cell culture medium and a
   subsequent PBS wash, the extraction process employed 800 μL of cold
   methanol/acetonitrile/water (2:2:1,v/v/v). The resulting mixture
   underwent centrifugation at 14000 g for 5 min, and the supernatant was
   judiciously collected and subjected to drying in a vacuum centrifuge.
   For LC-MS analysis, the dried samples were re-dissolved in 100 μL
   acetonitrile/water (1:1, v/v) solvent. Maintaining analytical stability
   and repeatability, quality control (QC) samples were meticulously
   prepared by pooling 10 μL from each sample, systematically inserted for
   analysis every 5 samples.

   The LC-MS analysis employed state-of-the-art equipment, including an
   UHPLC (1290 Infinity LC, Agilent Technologies) coupled with a
   quadrupole time-of-flight (AB Sciex TripleTOF 6600) at Shanghai Applied
   Protein Technology Co., Ltd.Utilizing both HILIC and RPLC separation
   techniques, chromatographic separation ensured robust performance in
   both ESI positive and negative modes. The ESI source conditions were
   set as follows: Ion Source Gas1 (Gas1) as 60, Ion Source Gas2 (Gas2) as
   60, curtain gas (CUR) as 30, source temperature: 600°C, IonSpray
   Voltage Floating (ISVF) ± 5500 V. In MS only acquisition, the
   instrument was set to acquire over the m/z range 60-1000 Da, and the
   accumulation time for TOF MS scan was set at 0.20 s/spectra. In auto
   MS/MS acquisition, the instrument was set to acquire over the m/z range
   25-1000 Da, and the accumulation time for product ion scan was set at
   0.05 s/spectra. The product ion scan is acquired using information
   dependent acquisition (IDA) with high sensitivity mode selected. The
   parameters were set as follows: the collision energy (CE) was fixed at
   35 V with ± 15 eV; declustering potential (DP), 60 V (+) and −60 V (−);
   exclude isotopes within 4 Da, candidate ions to monitor per cycle: 10.

   The raw MS data underwent thorough processing using ProteoWizard MS
   Convert before importing into freely available XCMS software. For peak
   picking, the following parameters were used: centWave m/z = 10 ppm,
   peakwidth = c (10, 60), prefilter = c (10, 100). For peak grouping, bw
   = 5, mzwid = 0.025, minfrac = 0.5 were used. CAMERA (Collection of
   Algorithms of MEtabolite pRofile Annotation) was sued for annotation of
   isotopes and adducts. In the extracted ion features, only the variables
   having more than 50% of the nonzero measurement values in at least one
   group were kept. Compound identification of metabolites was performed
   by comparing of accuracy m/z value (<10 ppm), and MS/MS spectra with an
   in-house database established with available authentic standards.The
   processed data were then subjected to multivariate data analysis using
   the R package (ropls), including Principal Component Analysis (PCA) and
   Orthogonal Partial Least-Squares Discriminant Analysis (OPLS-DA). To
   ensure the robustness of the model, a 7-fold cross-validation and
   response permutation testing were conducted. Statistical significance
   was assessed through Student’s t-test, with a stringent threshold of p
   value < 0.05 indicating significant changes.

MeRIP-sequence

   MeRIP-seq assay was performed by Cloudseq Biotech Inc. In brief, m^6A
   RNA immunoprecipitation (IP) was carried out utilizing a GenSeq™ m^6A
   RNA IP Kit (GenSeq). Both the input and m^6A IP samples were prepared
   for NGS. The library was constructed using a NEBNext® Ultra II
   Directional RNA Library Prep Kit (NEB). The library’s quality was
   assessed using a BioAnalyzer 2100 system (Agilent).

   For MeRIP-seq analysis: Paired-end reads were sequenced on the Illumina
   NovaSeq 6000 platform with a 150 bp read length. Quality control was
   initiated by calculating Q30 scores. Subsequently, adapter trimming was
   performed using Cutadapt (v1.14), and low-quality reads were removed
   with Trimmomatic (v0.36). Quality checks were conducted using FastQC
   (v0.11.5) software. To eliminate mitochondrial, ribosomal RNA, and tRNA
   sequences, Bowtie2 (v2.2.9) was employed. Clean reads were then aligned
   to Homo sapiens transcripts (GENCODE Release 36, GRCh38.p13) using
   Bowtie2 (v2.2.9) with the parameters (–local -x). Methylated sites on
   peaks were identified using MACS2 (v2.2.7.1) peak-calling software,
   using corresponding input samples as controls. Differentially
   methylated sites were determined with the BEDTools map function in the
   R language. For peak annotation, the R package ChIPseeker (v1.38.0)
   [[276]61] was utilized. The distribution of m^6A peaks on mRNA features
   was visualized using the R package Guitar (v2.18.0). Enriched motifs
   within m^6A peaks were identified with HOMER [[277]62].

Whole exome sequencing of PDOs and matched primary lung tumor tissues

   Genomic DNA was extracted from patient-derived tumor organoids (PDOs)
   and matched primary lung tissues for whole-exome sequencing. Libraries
   were prepared using the Agilent SureSelect Human All Exon V6 kit
   (64 Mb) and sequenced on the Illumina NovaSeq X platform with 150 bp
   paired-end reads, achieving an average coverage depth of over 100×. Raw
   sequencing reads were subjected to quality control using fastp (v1.0.1)
   [[278]63], which automatically trimmed adapter sequences and filtered
   out low-quality bases at read ends. Quality reports were generated as
   part of this process. Clean reads were then aligned to the human
   reference genome (hg19) using BWA (v0.7.19). The resulting BAM files
   were sorted and indexed with Samtools, and duplicate reads were removed
   [[279]64]. Base quality score recalibration (BQSR) was performed using
   GATK (v4.6.2.0). High-quality BAM files were processed with GATK
   HaplotypeCaller for variant calling, with single nucleotide
   polymorphisms (SNPs) and insertions/deletions (INDELs) filtered based
   on GATK best practice recommendations. Somatic mutations were
   identified using GATK Mutect2, and high-confidence calls were obtained
   through population database filtering and stringent post-processing
   criteria. All variants were functionally annotated with ANNOVAR
   (2025-06), incorporating multiple public databases, including RefSeq,
   ClinVar, COSMIC, the 1000 Genomes Project, and gnomAD. Copy number
   variation (CNV) analysis was performed in tumor-only mode using CNVkit
   (v0.9.12) with a pre-built flat reference. Statistical analyses and
   data visualization were conducted in Python 3.10 using packages such as
   Pandas, Matplotlib, and Seaborn.

Survival analysis and risk score model construction

   To assess the prognostic significance of candidate genes in lung
   adenocarcinoma (LUAD) patients, clinical survival data—including
   overall survival (OS) and progression-free interval (PFI)—were
   retrieved from the TCGA-LUAD cohort. Kaplan–Meier survival curves were
   generated to compare outcomes between high and low gene expression
   groups, with statistical significance evaluated using the log-rank
   test. To capture the combined prognostic impact of multiple genes, a
   multigene risk score model was constructed based on the Cox
   proportional hazards regression. Gene expression levels served as
   independent variables in the multivariate Cox model, and a risk score
   for each patient was calculated using the following formula:
   [MATH: <mrow><msub><mrow><mi mathvariant="normal">Risk
   Score</mi></mrow><mrow><mi>i</mi></mrow></msub><mo>=</mo><munderover><m
   o>∑</mo><mrow><mi>j</mi><mo>=</mo><mn>1</mn></mrow><mrow><mi>n</mi></mr
   ow></munderover><mrow><msub><mrow><mi>β</mi></mrow><mrow><mi>j</mi></mr
   ow></msub></mrow><mo>×</mo><msub><mrow><mi
   mathvariant="normal">Exp</mi></mrow><mrow><mi
   mathvariant="italic">ji</mi></mrow></msub></mrow> :MATH]

   Where βj is the regression coefficient for gene j, and Expji represents
   the expression level of gene j in patient i. Patients were then
   stratified into high-risk and low-risk groups according to the median
   risk score. Kaplan–Meier analysis and the log-rank test were performed
   to evaluate survival differences between the two groups. All analyses
   and visualizations were carried out using the survival and survminer
   packages in R.

Statistical analysis

   To determine statistical significance, comparisons between two groups
   were performed using an unpaired Student’s t-test, comparisons between
   multiple groups were performed using a one-way ANOVA, comparisons
   between two or multiple groups with additional treatment conditions or
   repeated measurements at different time periods were performed using a
   two-way ANOVA, and detailed statistical analyses of high-throughput
   sequencing are described in the “Materials and Methods” section. For
   additional information on statistical analyses, please refer to the
   figure legends.

Supplementary information

   [280]Figure S1^ (5.4MB, tif)
   [281]Figure S2^ (29.6MB, tif)
   [282]Figure S3^ (3MB, tif)
   [283]Figure S4^ (3.9MB, tif)
   [284]Figure S5^ (2.1MB, tif)
   [285]Figure S6^ (2.6MB, tif)
   [286]Figure S7^ (7.1MB, tif)
   [287]Figure S8^ (26.4MB, tif)
   [288]Figure S9^ (2.2MB, tif)
   [289]Figure S10^ (8.1MB, tif)
   [290]Figure S11^ (3.3MB, tif)
   [291]Figure S12^ (10.3MB, tif)
   [292]Supplementary information^ (42.2KB, docx)
   [293]original data-ddx3x^ (93MB, xlsx)

Acknowledgements