The exon-junction complex helicase eIF4A3 controls cell fate via coordinated regulation of ribosome biogenesis and translational output

The emerging role of NMD in cancer biology (7) led to its exploitation in the production of new candidate drugs for cancer treatment. Being part of the EJC complex and therefore important for NMD (6), eIF4A3 has already been chemically targeted with promising results for cancer therapy (13). A mutational analysis of eIF4A3 in various cancer types shows great heterogeneity, with amino acid residues known to affect NMD being less prone to become mutational hotspots than those residues critical for eIF4A3’s helicase activity or interaction with other members of the EJC complex (fig. S1A) pinpointing a potential non-NMD therapeutic target window. Analysis of publicly available cancer patient data shows that in most cases studied (16), eIF4A3 expression is higher in cancer compared to the normal tissue counterparts (Fig. 1A), a finding we could further support in a panel of cancer versus normal/nontransformed (diploid) human cell types, both at the RNA (fig. S1, B and C) and the protein level (fig. S1D). In addition, lower eIF4A3 mRNA levels correlate with better prognosis in most of The Cancer Genome Atlas (TCGA)–registered cancer types represented in LinkedOmics (Fig. 1B, fig. S1E, and table S1A) (17). These findings suggest that elevated eIF4A3 levels may be selected for during tumor development and benefit cancer cell survival and growth. To find previously unknown eIF4A3-related cancer vulnerabilities, we performed a pathway enrichment analysis among the top 100 genes whose dependency scores were positively correlated with that of eIF4A3 among 793 different cancer cell lines (DepMap, CRISPR, and Avana 20Q3) (Fig. 1C and table S1, B and C) (18). Apart from mRNA splicing, NMD, and p53 signaling, we noticed a correlation with genes implicated in rRNA processing [e.g., uL5 (RPL11), RLPL17, and SNU13]. Using a curated RiBi gene signature (table S1D) and RNA sequencing (RNA-seq) data from the DepMap consortium, we then categorized all cell lines into low and high RiBi (fig. S1F) and found that eIF4A3 expression is higher in cell lines with high RiBi (Fig. 1D), further supporting a role of the helicase in this process. Moreover, the effect of eIF4A3 knockdown on cell survival (DepMap and DEMETER2) was inversely correlated with the cytotoxicity of RiBi stress–related chemical compounds [e.g., oxaliplatin (19) or the MDM2-p53 inhibitors CGM097 and idasanutlin], strongly implicating eIF4A3 in their mechanism of action (Fig. 1E and table S1E). In conclusion, eIF4A3 expression might serve as a predictive marker in high-RiBi cancers and its underlying role in RiBi points toward a previously unidentified cancer vulnerability with a therapeutic potential.

EIF4A3 has been shown to reside, as part of the EJC, in perispeckles, nuclear subdomains formed around splicing centers (20). Using SC35 (serine and arginine rich splicing factor 2) as a surrogate marker for nuclear speckles, we confirmed the presence of eIF4A3 in these domains by immunofluorescence (IF), using a U2OS cell line ectopically expressing green fluorescent protein (GFP)–tagged eIF4A3 (fig. S2A). The perispeckle staining pattern did not change upon ribosomal stress induced by low concentration of actinomycin D (5 nM; henceforth ActD^L) (21). Besides perispeckles, however, we also noticed variable nucleolar localization of eIF4A3 (Fig. 2A), consistent with an in-depth analysis of the nucleolar proteome, which indicated that the helicase resides also in nucleoli (22). In our cell culture experiments, treatment with ActD^L or BMH-21 (two inhibitors of Pol I) (23) reduced the amount of eIF4A3 colocalizing with the nucleolar marker fibrillarin (FBL) or UBF (upstream binding transcription factor), supporting the notion that active Pol I transcription is required for nucleolar localization of eIF4A3 (Fig. 2, A and B, and fig. S2B). Staining of U2OS cells for endogenous Y14 showed that other components of the EJC complex were also detected in the nucleolus (fig. S2C) in agreement with a previous report (24). Upon ActD^L treatment, however, Y14 did not show the same spatial pattern and relocalization kinetics as eIF4A3, an observation that suggested potentially separate, nucleolar roles for these proteins.

Aberrant rRNA processing is commonly associated with altered nucleolar structure (25). We therefore assessed the nucleolar integrity upon eIF4A3 knockdown by IF, using FBL and UBF proteins as nucleolar markers (fig. S2D) and treatment with ActD^L as a positive control for nucleolar stress. In U2OS (Fig. 2C) and A549 cells (fig. S2E), eIF4A3 knockdown resulted in reshaping of the nucleolus into a necklace-like structure indicative of aberrantly altered rRNA processing (26), whereas ActD^L administration caused formation of nucleolar caps due to Pol I transcriptional inhibition (27). We further analyzed the structure of the outer, granular component (GC), region of the nucleolus where rRNA processing takes place (fig. S2D) (28). Staining for the GC-resident protein nucleophosmin 1 (NPM1) (Fig. 2D) showed a malformed nucleolar exterior in small interfering RNA (siRNA) targeting eIF4A3 (sieIF4A3)–treated U2OS cells, again in contrast to the diffused NPM1 pattern following treatment with ActD^L. Electron microscopy (EM) analysis of cells under the same experimental conditions showed enlarged and deformed nucleoli both in sieIF4A3- and eIF4A3 inhibitor (eiF4A3i)–treated samples (Fig. 2E and fig. S2F, respectively) (29), whereas ActD^L induced highly dense and compacted nucleoli reflecting Pol I inhibition (30). Overall, these findings support a potential protective role of eIF4A3 in the maintenance of nucleolar function and point toward nucleolar stress as a contributing plausible cause of sieIF4A3-mediated induction of p53.

To assess relevance of the above model experiments and potential nucleolar role of eIF4A3 for clinical settings, we next performed a large-scale immunohistochemical analysis of the eIF4A3 protein abundance and subcellular localization, with special emphasis on the nucleolus. Our panel of human tissue sections from archival paraffin blocks included normal tissues and cohorts of tumor specimens from the brain (glioblastoma), colon (adenomas and carcinomas), urinary bladder cancer (early preinvasive and invasive stages), and uterine cervix (dysplasias and squamous cell carcinomas). Representative examples of the eIF4A3 staining patterns are shown in Fig. 2 (F and G) and fig. S2G, and the overall graphical summary from all cancer types is presented in fig. S2H. Notably, consistent with our experimental results so far, eIF4A3 was detectable in nucleoli of both normal cells and tumor cells, in the vast majority of cases with a similar staining intensity in both the nucleoplasm and the nucleolus. This subcellular localization of eIF4A3 was seen also in neurons of normal brain tissue (Fig. 2F), a finding that could indicate a default nucleolar role of eIF4A3 even in nonproliferating yet metabolically highly active normal cells. The staining intensity and patterns of subnuclear localization of eIF4A3 varied widely among individual cells or cell clusters in a given tissue section (Fig. 2, F and G, and fig. S2G) including subsets of cells either with nucleoli being either selectively negative or in contrast nucleoli showing a much stronger signal than the nucleoplasm. We detected the latter pattern of preferential nucleolar localization of eIF4A3 with a higher frequency especially among the high-grade cervical cancer specimens compared to their low-grade counterparts or normal tissue (fig. S2H).

Together, these results were consistent with a potential function of eIF4A3 in the nucleolus and an apparently dynamic localization pattern with even selectively nucleolar localization in subsets of tumor cells. These findings validated the potential pathophysiological relevance of nucleolar eIF4A3 in human cells and tissues, prompting us to deepen our mechanistic insights into such role using model systems and a variety of experimental approaches, as described below.

Ablation of p53 was previously shown to rescue the neurological defects caused by eIF4A3 haploinsufficiency in mice (8). However, the functional relationship of eIF4A3 and p53 remained unexplored. To address this issue, we engineered human U2OS cells to express a double doxycycline (DOX)–inducible system carrying short hairpin–mediated RNAs (shRNAs) against endogenous eIF4A3 (DOX-sheIF4A3-U2OS) together with a FLAG-tagged eIF4A3 cassette for ectopic reexpression of the helicase (DOX-FLAG.WT.eIF4A3-sheIF4A3-U2OS, Fig. 3A). Using two different shRNAs, targeting either eIF4A3 3′ untranslated region (3′UTR) (#1) or its cds (coding sequence) (#2) (Fig. 3A and table S4B), we knocked down eIF4A3 and measured p53 protein levels by immunoblotting. A 80 to 90% reduction in eIF4A3 levels was followed by induction of p53 (Fig. 3B, lanes 1, 3, and 5), whereas ectopic reexpression of eIF4A3 almost completely reversed the induction of p53, indicating a causal link between eIF4A3 ablation and p53 accumulation (lanes 2, 4, and 6). p53 accumulation was not a result of increased TP53 mRNA translation but rather a consequence of enhanced protein stability (fig. S3, A to C), followed by p53 activation as evidenced by the rise of total protein levels of its downstream target, p21 (fig. S3D).

To uncover abnormalities that may explain the p53 induction, we explored the effect of eIF4A3 knockdown (sieIF4A3) on the transcriptome of U2OS cells (fig. S3E). Differential expression (DE) analysis revealed 4493 transcripts (table S2A, P_adj ≤ 0.05, log₂ fold change ≥ |1|), and pathway enrichment analysis indicated an overall suppression of genes regulating the cell cycle, extracellular matrix organization, splicing, and DNA damage response, while “translation,” “RNA metabolism,” and “p53 signaling” terms were enriched among the up-regulated genes (Fig. 3C and table S2B). While these gene expression results are largely consistent with previous reports (8), little is known about any potential role of eIF4A3 in RiBi. A closer inspection of the RNA metabolism gene signature uncovered a positive trend for genes associated with rRNA processing [e.g., small subunit (SSU) components DHX37, UTP3, EXOSC4, and RRP9], while genes regulating RNA nuclear transport were found mostly down-regulated (Fig. 3, D and E, and table S2C). EIF4A3’s involvement in RiBi was further supported by the fact that Reactome pathway enrichment analysis among the 100 most variable genes in human HCT116 colon carcinoma cells treated or not with an inhibitor of eIF4A3 (31) further attested to a plausible role for eIF4A3 in RiBi (fig. S3F and table S2D).

To identify unique molecular signatures related to RiBi, we compared our RNA-seq results obtained from eIF4A3-depleted cells with those produced upon ActD^L known to block Pol I activity. Figure 3F shows that the two experimental conditions share 440 up-regulated genes (quadrant I) whose pathway enrichment analysis points toward p53 signaling (Fig. 3G and table S2, E and F). This and the fact that eIF4A3 knockdown and ActD^L both induce p53 and p21 in a similar fashion (fig. S3G) further implicated eIF4A3 in RiBi. Although eIF4A3 knockdown affects splicing (32) and leads to production of aberrant RNA species such as NMD candidates (fig. S3, H and I, and table S7B), chemical inhibition of neither splicing [by pladienolide B (pladB)] nor nuclear export [by leptomycin B (LMB)], two manipulations known to activate p53 (33, 34), altered the observed sieIF4A3-mediated p53 induction, suggesting that these inhibited processes are unlikely the triggers of p53 activation in cells depleted of eIF4A3 (fig. S3J). Collectively, our data point toward perturbations in RiBi as a plausible mechanism for p53 induction.

Most genes affected by small interfering RNA (siRNA)–mediated eIF4A3 depletion suggest an impact of eIF4A3 on the early steps of rRNA processing (Fig. 3). This and the fact that eIF4A3 localizes in the nucleolus (Fig. 2) prompted us next to investigate the potential nucleolar function of the helicase. To validate biochemically our RNA-seq findings, we used primers covering precursor and mature rRNA regions (30) and detected a decrease in the levels of early rRNA species both in sieIF4A3-treated and in eIF4A3i-treated samples (Fig. 4A and fig. S4A, respectively). To exclude that changes in the levels of various rRNA species follow deregulated/aberrantly enhanced Pol I transcription, we used a luciferase-based assay using a Pol I–targeted promoter (35) and found that eIF4A3 knockdown did not significantly change expression levels from this rDNA reporter promoter, in contrast to treatment with ActD^L (Fig. 4B). This result was further validated by high-content imaging following a short pulse of EU (5-ethynyl uridine) (Fig. 4C) or an antibody against 5.8S rRNA (Fig. 4D). rRNA expression changes in the helicase-depleted cells could not be reversed by the addition of ActD^L, suggesting an altered posttranscriptional “flow” of rRNA processing rather than inhibition of Pol I transcription per se.

As the early rRNA processing events are executed by the SSU processome (36), we then asked whether eIF4A3 could be part of this complex. To address this notion, we performed RNA immunoprecipitation (RIP) between eIF4A3 and the main SSU small nucleolar RNA (snoRNA), U3. Figure 4E shows that eIF4A3 interacts specifically with U3 snoRNA in fixed U2OS cells in agreement with earlier findings in HeLa cells (37), and their interaction is extended onto the rRNA as seen from RIP assays between eIF4A3 and 18S rRNA (Fig. 4F). DE analysis following cross-linking immunoprecipitation sequencing (CLIP-seq) among eIF4A3 and CASC3 (CASC3 Exon Junction Complex Subunit) RNA interactors (Fig. 4G and table S3A) revealed also U3 snoRNA (SNORD3A) as a specific eIF4A3 interactor, supporting our findings and pinpointing that this interaction is irrelevant to eIF4A3’s role in NMD (where CASC3 participates). Moreover, we found that BMS1 (BMS1 Ribosome Biogenesis Factor), which regulates early rRNA processing via SSU (36), is the most positively correlated RiBi gene to eIF4A3 (Fig. 4H and table S3B), further supporting a role for eIF4A3 in rRNA metabolism.

Given the role of some DEAD helicases in R loop clearance (38) and the fact that nucleoli are susceptible to R loop formation (39), we then reasoned that eIF4A3’s role in SSU may relate to the clearance of excessive R loops that could potentially block rRNA processing and lead to genotoxic transcription-replication collision events. To address this, we used an engineered U2OS cell line expressing a DOX-inducible ribonuclease H1 (RNase H1) cassette (40) and measured R loop abundance in nucleoli with IF, using the widely used antibody S9.6, followed by high-content microscopy. We found that whereas sieIF4A3 enhanced the accumulation of R loops, ectopic expression of RNase H1 reversed this effect, thereby directly implicating eIF4A3 in nucleolar R loop clearance (Fig. 4I). Our findings were further validated using a cell line constitutively expressing catalytically inactive GFP-tagged RNase H1 (fig. S4B) (40), to overcome any technical and/or interpretation difficulties linked to the use of the S9.6 antibody (41). sieIF4A3-triggered R loop accumulation in the nucleolus was related to active Pol I, since treatment of U2OS with ActD^L known to selectively block Pol I activity reduced the overall S9.6 signal (fig. S4C). Excessive R loops can cause DNA damage (42), and since our RNA-seq analysis has uncovered relevant gene signatures, we decided to study this notion further. eIF4A3 knockdown led to induced levels of the DNA damage marker γH2AX, shown both by immunoblotting (fig. S4D) and high-content microscopy (fig. S4E). Focusing on the nucleoli, we found that ActD^L treatment partially prevented the sieIF4A3-evoked H2AX phosphorylation, a finding consistent with eIF4A3’s involvement in preventing nucleolar R loop accumulation that can lead to DNA damage (fig. S4F). In conclusion, we show here that eIF4A3 is part of the SSU and clears excessive R loops to ensure proper rRNA processing following Pol I transcription (fig. S4G).

Perturbations in RiBi cause p53 activation via the RPL5/RPL11–5S rRNA–MDM2 complex (IRBC) (2). To assess potential activation of this checkpoint, we measured p53 and p21 levels in U2OS cells depleted of eIF4A3, with or without concomitant depletion of components of the IRBC, and treated or not by ActD^L. Figure 5A shows that sieIF4A3 evoked p53 induction that was partially prevented by concomitant depletion of RPL5 (uL18), RPL11 (uL5), or 5S rRNA (lanes 2 and 6 to 8), supporting the notion that depletion of eIF4A3 caused ribosomal biogenesis stress followed by p53 activation. Similar preventive effect was also observed at the additively enhanced level of p53 in cells depleted of eIF4A3 and treated with ActD^L (lanes 10 and 14 to 16), indicating that the helicase may be involved in mitigating nucleolar stress caused by exogenous factors such as drugs. Analogous data were reproduced in human A549 lung carcinoma cells, documenting that the observed effects were not cell type–restricted (fig. S5A).

To clarify which traits of eIF4A3 mediate the suppressive effect on p53 induction, we expressed eIF4A3 mutated in diverse domains, with selectively disabled functional aspects of the protein, in the DOX-sheIF4A3-U2OS cells. DOX treatment of these cells induced expression of the ectopic FLAG-tagged eIF4A3 versions carrying individual mutations in three sites essential for its major activities of RNA binding, interaction with the EJC complex and ATP binding, respectively (DOX-FLAG.mut.eIF4A3-sheIF4A3-U2OS, Fig. 5B, fig. S5B, and table S4A) (43). As can be seen in Fig. 5C, ectopic expression of either the EJC-binding incompetent (D401KE402R, henceforth referred to as eIF4A3^EJCm) or the RNA binding incompetent (T115R;E117VL118A, henceforth referred to as eIF4A3^RNAm) mutants failed to rescue the effect of knocking down the endogenous wild-type (WT) eIF4A3 on p53 induction, whereas the ATP-binding incompetent mutant (A188Q, henceforth referred to as eIF4A3^ATPm) efficiently rescued the impact of lacking the endogenous eIF4A3. The mutant eIF4A3 proteins defective in EJC binding and RNA binding, respectively, were also impaired in terms of their nucleolar residence (fig. S5, C and D), and they caused perturbed rRNA processing (Fig. 5D). The capacity of these two mutants to cause p53 induction was partially reversed in cells treated with siRNA against RPL11 (Fig. 5E), implicating eIF4A3’s EJC and RNA binding capacity in eIF4A3’s nucleolar activity and prevention of the IRBC-mediated p53 up-regulation. In conclusion, eIF4A3 knockdown causes activation of p53 via IRBC, and the ability to reverse this impact by reexpression of eIF4A3 requires the functional EJC and RNA binding domains of the protein.

To determine whether sieIF4A3 affects translation downstream of perturbed RiBi, we used initially a puromycin pulse-chase assay in U2OS cells treated with ActD^L or siRNA against eIF4A3. Figure S6A shows that whereas exposure to ActD^L substantially reduced global translation, eIF4A3 knockdown showed a less pronounced effect. Quantification of the translation rate using O-propargyl-puromycin (OPP; Fig. 6A) and HPG (‘L-homopropargylglycine) (fig. S6B) click chemistry followed by high-content microscopy indicated a statistically significant but less severe effect of sieIF4A3 on translation compared to that of the protein synthesis inhibitor cycloheximide (CHX) (20 to 25% versus 50%). Furthermore, polysome profiling of U2OS and HeLa cells revealed a higher 80S monosome peak in cells depleted of eIF4A3 (Fig. 6B and fig. S6C, respectively) reminiscent of translation arrest (44).

To see the extent of any residual ongoing translation in sieIF4A3-treated cells, we performed liquid chromatography–mass spectrometry (LC-MS) in U2OS treated with sieIF4A3 or control siRNA and run a proteomics DE analysis. We found around 1700 differentially expressed proteins [false discovery rate (FDR) < 0.05] with gene ontology (GO) analysis pointing to cell cycle modulators (table S5, A and B). To exclude the effect of posttranslational events on protein expression, we then integrated these results with our transcriptomic analysis (Fig. 3) and run a pathway enrichment analysis on genes showing monotonic deregulation (Fig. 6C). Among the up-regulated genes (e.g., MDM2 and CDKN1A), p53 signaling was the most prominent GO term, while, in the down-regulated ones, “cell cycle” and “DNA damage” were among the most enriched terms (table S5C). No “rRNA processing”–related terms were detected, indicating translational buffering events of relevant DE genes at the RNA level (Fig. 3) (45) and further supporting our findings regarding a direct role of eIF4A3 in rRNA metabolism via RNA-protein interactions.

Next, we performed RNA-seq analysis on the polysome profile fractions to understand the net effect of translation on the proteome landscape of sieIF4A3-treated cells (Fig. 6D). Given that 80S monosomes may be translationally active (46), we assessed both the translatome of monosomes (Fig. 6B, magenta box) and polysomes (>3 ribosomes; Fig. 6B, cyan box). Following DE analysis, we separated the transcripts into monosome- and polysome-associated, respectively, and found that the majority are bound to 80S monosomes (circa 10 times more compared to polysomes and 5 times more compared to common regulated transcripts; Fig. 6E). To see whether the 80S-bound transcripts correlate with their protein products, we compared the 80S translatome to our proteomics data (Fig. 6F and table S5D). Unexpectedly, most of the correlated transcripts showed monotonic up- or down-regulation (Fig. 6F, quadrants I and III) supporting active translation in monosomes. Among the up-regulated genes, we detected diverse ubiquitin ligases with seminal roles in cell physiology such as MDM2 and RNF135 (ring finger protein 135), while down-regulated genes were enriched in cell cycle modulators and particularly mitotic factors (Fig. 6F and table S5, E and F). Together, we conclude that eIF4A3 depletion hinders partially protein synthesis while allowing some residual translation to take place via both polysomes and monosomes, with cell cycle modulators representing the functional category most prominently affected by this altered translation.

Given the profound effect of sieIF4A3 on genes regulating cell cycle, we next investigated this aspect in greater detail. Using 5-ethynyl-2′-deoxyuridine (EdU) incorporation followed by high-content microscopy, we first examined the cell cycle progression of nonsynchronized U2OS cells depleted of eIF4A3. Figure 6 (G and H) shows that eIF4A3 knockdown reduced the proportion of S phase cells, similarly to treatment with ActD^L, leading to accumulation of U2OS cells in G₁ and G₂ phases. The concomitant G₁ and G₂ arrest suggests activation of multiple cell cycle checkpoints, an assumption that was validated by the inability of the eIF4A3i-treated cells, synchronized either in G₁ (serum starvation) or in G₂ (nocodazole block), to progress into the subsequent cell cycle phases upon release of the inhibitory treatment (fig. S6, D and E). Using FUCCI (fluorescent ubiquitination-based cell cycle indicator) reporter expressing U2OS cells, we showed that the effect of sieIF4A3 on the G₁-S transition is the dominant one (fig. S6F). G₁ arrest can follow aberrant mitosis (47), and mitotic dysfunction in the absence of eIF4A3 was evident from our omics results. Accordingly, the inability of sieIF4A3-treated cells to undergo proper mitosis was validated using an engineered U2OS cell line stably expressing GFP-tagged histone H2B (fig. S6G) and was further substantiated by the elevated number of micronuclei in cells depleted of eIF4A3 (fig. S6H). The observed G₁ arrest of eIF4A3-depleted cells could reflect p53-mediated G₁ checkpoint activation including p21 up-regulation (Fig. 3) and the ensuing inhibition of the G₁-S transition promoting cyclin E/cdk2 kinase complex, analogous to the effect of DNA damage on cells (48). In support of this notion, we could rescue the cell cycle inhibitory effect of sieIF4A3 (fig. S6I), using a U2OS cell line carrying a dominant negative C-terminal p53 fragment capable of binding to and inhibiting the endogenous WT p53 (ddp53-U2OS). G₂ arrest was also rescued, suggesting that p53 is instrumental in both checkpoints. In summary, cells depleted of eIF4A3 show a p53-mediated cell cycle arrest.

Our omics analyses revealed an apoptotic gene signature upon sieIF4A3 treatment with genes including FAS, BBC3, and BAX found up-regulated both in monosome and polysome fractions (Fig. 7A and table S6A). This observation prompted us to further study the role of eIF4A3 in cell survival downstream of perturbed cell cycle. EIF4A3 depletion reduced cell survival in a time-dependent fashion (Fig. 7B) while promoting apoptosis evidenced by up-regulation of apoptotic genes, caspase activation, poly(ADP-ribose) polymerase (PARP) cleavage, and diminished colony formation capacity (Fig. 7, C to E, and fig. S7, A to D). Using TP53 knockout HCT116 colon carcinoma cells and the ddp53-U2OS cells, it was possible to partially reverse the effect of sieIF4A3 on cell survival by absence of or interference with p53 (Fig. 7, D and E), indicating that the proapoptotic pathways triggered by eIF4A3 depletion encompass both p53-dependent and p53-independent mechanisms. This concept of dual proapoptotic pathways triggered in the absence of eIF4A3 was further corroborated by the fact that using separately either catalytically inactive p53 (Fig. 7F, lanes 2 and 8) or knockdown of the apoptotic genes BAX and BBC3 (PUMA) (lanes 2, 5, and 6) partially rescued the effect of sieIF4A3 on PARP cleavage, while a combination of these manipulations showed no additive effect (lanes 5, 6, 11, and 12). Moreover, sieIF4A3-mediated FAS up-regulation was apparent also in ddp53-U2OS cells, further supporting the notion that some p53-independent proapoptotic program also becomes active in cells depleted of eIF4A3 (Fig. 7G). Using our total and polysome profiling RNA-seq data, we then performed the deltaTE analysis (45) that identifies transcripts regulated at different levels: transcriptional (buffered and offset), translational (exclusive), or both (forward and intensified) (fig. S7E and table S6B). When we plotted the results of this analysis against our proteomics data, we found that apoptotic genes (e.g., BBC3, FAS, and BAX) were monotonically up-regulated both in monosomes and polysomes (Fig. 7, H and I), a finding that could be validated biochemically (fig. S7, F and G) and indicated bimodal translation of these genes. We further noticed that the regulation mode (intensified versus forward) was different among the monosome and the polysome fractions (Fig. 7, H and I). For example, BAX shows intensified expression (regulated both at the transcriptional level and the translational level) only in the monosome fraction, whereas BBC3 follows the same regulation mode both in monosomes and polysomes (Fig. 7, H and I). This concept of intensified expression was then biochemically validated in the case of BBC3. Whereas p53 is essential for sieIF4A3-mediated BBC3 induction, it is redundant for PUMA protein accumulation under the same experimental conditions (fig. S7, H and I). This suggests that PUMA expression is regulated both at the transcriptional and the translational level, overcoming the “p53 bottleneck.” These findings support a multimodal regulation of cell death genes, consistent with concomitant p53-dependent and p53-independent aspects of cell death induction upon depletion of eIF4A3.

On the basis of our findings, we hypothesized that the net transcriptional outcome of p53 activation affecting cell fate could be modulated at the protein translation level (Figs. 6 and 7). Given the effect of sieIF4A3 on posttranscriptional mRNA regulation (6), we then asked whether sieIF4A3 could alter the repertoire of the translated p53 targets. For this purpose, we examined MDM2, a p53 target that was present throughout our omics analyses and was found to be associated with 80S monosomes (Fig. 8A). Under unperturbed growth conditions, MDM2 controls p53 levels via ubiquitination-mediated degradation (49). Degradation or competitive binding of MDM2 to other proteins such as the 5S RNP complex upon RiBi stress (2) causes p53 stabilization. We initially noticed that eIF4A3 ablation (or chemical inhibition) caused production of unique MDM2 protein isoforms not evident following administration of other p53 inducers (Fig. 8, B and C). These isoforms were not byproducts of degradation known to induce p53 (fig. S8, A and B) (50) and could result from translation of alternative MDM2 mRNA isoforms. Using DEXSeq, a differential exon usage–based algorithm, to analyze our RNA-seq data (51), we observed these alternative isoforms that could be also validated by reverse transcription polymerase chain reaction (RT-PCR) (Fig. 8D, fig. S8C, and table S7A). These results are reminiscent of published data regarding the effect of Y14 (another EJC component) on MDM2 splicing (32), likely indicating a universal role for EJC in MDM2 regulation. Integrative analysis of our proteomics data and our 80S monosome RNA-seq data following a transcript-based DE analysis (Salmon; table S7B) revealed that (i) different MDM2 isoforms show different transcription-translation kinetics and (ii) there is at least one NMD MDM2 isoform that is loaded onto the 80S monosomes and could potentially give a protein product (Fig. 8E).

Given that MDM2 mRNA up-regulation following eIF4A3 ablation is p53-dependent (fig. S8D), we hypothesized that production of aberrant mRNA/protein isoforms follows the initial p53 activation that could reflect RiBi stress. In support of this hypothesis, concomitant knockdown of uL5 (RPL11), a component of the 5S RNP complex responsible for the IRBC-mediated p53 up-regulation, prevented the sieIF4A3-mediated induction of all MDM2 isoforms (fig. S8E). In a similar fashion, ectopic expression of the eIF4A3 mutants incompetent in binding either EJC or RNA (eIF4A3^EJCm and eIF4A3^RNAm, respectively) did not reverse the phenotype of aberrant MDM2 isoform pattern seen upon depletion of endogenous WT eIF4A3. In contrast, replacing endogenous eIF4A3 by the mutant that does not bind ATP (eIF4A3^ATPm) could rescue the MDM2 isoform pattern (fig. S8F). As with eIF4A3 knockdown, the production of these alternative MDM2 species was partially reversed by concomitant knockdown of uL5. These isoforms could affect MDM2’s functional interaction with p53, a possibility consistent with the finding that sieIF4A3 treatment markedly enhanced p53 stability (fig. S3A) and augmented the effect of MDM2 chemical inhibition on p53 total levels (fig. S8G). In conclusion, eIF4A3 knockdown leads to both the production of aberrant MDM2 mRNA species (probably via alternative splicing) and their differential translation. It is likely that at least some of these MDM2 species are incompetent to bind p53 (or antagonize normal MDM2 transcripts), rendering the MDM2-p53 feedback loop dysfunctional (52) while, at the same time, sensitizing the cells to stressful stimuli that can induce p53, such as ActD^L. Targeting eIF4A3 in cancer can thus be beneficial in two ways: (i) through activation of p53 via IRBC and (ii) by sustaining p53 in an active state via attenuation of the MDM2-p53 feedback loop.

U2OS was purchased from the American Type Culture Collection. A549, SaOS-2, HeLa, MCF7-1, RPE1, human embryonic kidney (HEK) 293T, and GFP.RNase H1.U2OS were provided by T. Helleday [Karolinska Institutet (KI)]. IMR90 and CCD841 were provided by M. Bienko (KI); Cl2.6 by M. Lindström (KI); and BJ, HT29, ddp53-U2OS, HCT116, and HCT116 TP53^−/− by J. Bartek (Danish Cancer Society). All cell lines were maintained in Dulbecco’s modified Eagle’s medium GlutaMAX (Gibco) supplemented with 10% fetal bovine serum (Sigma-Aldrich). Cell lines were tested for mycoplasma on a monthly basis (Lonza). shRNAs against eIF4A3 (table S4B, uppercase is the backbone and lowercase is the target sequence) were cloned into a U6-Tet-shRNA:Ef1a-TetR-Puro-iRFP670 vector using Age I/Eco RI, as previously described in (54). Viral particles were produced by transfecting the lentiviral and corresponding packaging plasmids (encoding polymerase and envelope proteins) into HEK293T cells. Supernatants were collected two times with 24-hour intervals, filtered, and used to infect U2OS. The cells (DOX-sheIF4A3-U2OS) were selected in the presence of puromycin (1 μg/ml) for 72 hours or until all nontransduced control cells were killed. For the production of DOX-FLAG-eIF4A3(WT/mutant)–sheIF4A3-U2OS cells, EIF4A3 was PCR-amplified from pCMV6-eIF4A3 (Origene, vector: PS100001) using the primer set provided in table S4B and cloned into a N-FLAG–tagged pENTR4 plasmid, which was produced by inserting an oligo-annealed V5-sequence into a pENTR4 vector (55) using Sal I/Not I. EIF4A3 was then subcloned into an DOX-inducible destination vector with blasticidin resistance. The destination vector was produced by inserting an Ef1a-promoter-tta3G-P2A-blasticidin cassette (GenScript) into a Tet-on-inducible plasmid (Addgene #27567) using Pml I/Kpn I. For the production of eIF4A3 mutants, point mutations were inserted into pENTR-eIF4A3 using the GeneArt site-directed mutagenesis kit (Thermo Fisher Scientific, A13282), and the primers are shown in table S4A. Viral particle production and cell transduction were conducted in the same fashion as with shRNA carrying plasmids, and stably transduced cells were selected in the presence of blasticidin (20 μg/ml) for 72 hours or until all nontransduced control cells were killed. GFP-eIF4A3-U2OS cells were produced by transduction of U2OS with pLenti-C-GFP-eIF4A3 (Origene, vector: PS100065).

Cells were treated with 5 nM ActDL (Sigma-Aldrich, A1410) or 2 μM BMH-21 (Sigma-Aldrich, SML1183) for 24 hours. CHX (Sigma-Aldrich, C4859) was used at a working concentration of 100 μg/ml for 30 min or the indicated time points. For DOX induction (Sigma-Aldrich, D9891), cells were treated with DOX (1 μg/ml; Sigma-Aldrich, D9881) for 48 to 72 hours, unless otherwise indicated. Pan-caspase inhibitor Z-VAD-FMK (R&D Systems, FMK001) and caspase 2 inhibitor Z-VDVAD-FMK (R&D Systems, FMK003) were used at a final concentration of 20 μM for 1 hour or the indicated time points. MG132 (Sigma-Aldrich, M7449), Nutlin-3 (Selleckchem, S1061), and MLN4924 (Selleckchem, S7109) were added to cells in a final concentration of 10, 5, and 1 μM, respectively, for 24 hours. Camptothecin (Sigma-Aldrich, C9911) was used at a working concentration of 5 μM and 5-fluorouracil (Sigma-Aldrich, F6627) at 10 μg/ml. Neocarzinostatin was used at 0.5 μg/ml for 24 hours, LMB was used at 20 ng/ml for 6 hours, and pladB was used at 1 μM for 6 hours. For chemical inhibition of eIF4A3, 52a, or 52b, 53a from (31) was used at 5 μM for 24 hours. For cell synchronization at G₂, cells were treated with nocodazole (Sigma-Aldrich, M1404) at 400 nM for 16 hours.

The antibodies used are as follows: mouse monoclonal SC35 (Abcam, ab11826), mouse monoclonal eIF4A3 (Santa Cruz Biotechnology, sc-365549), rabbit monoclonal α-tubulin (Abcam, ab176560), mouse monoclonal α-phospho-H2A.X (Ser¹³⁹) (Sigma-Aldrich, 05-636), mouse monoclonal UBF (Santa Cruz Biotechnology, sc-13125), rabbit polyclonal FBL (Abcam, ab5821), rabbit polyclonal RBM8A (Y14) (Thermo Fisher Scientific, PA5-53790), mouse monoclonal nucleophosmin (Abcam, ab10530), rabbit polyclonal nucleolin (Abcam, ab22758), mouse monoclonal puromycin (Kerafast, EQ0001), rabbit polyclonal RPL5 (uL18) (Abcam, ab86863), mouse monoclonal RPL11 (uL5) (1:200; Life Technologies, 373000), mouse monoclonal p53 (1:1000; Abcam, ab1101), mouse monoclonal FLAG (1:1000; Sigma-Aldrich, F3165), rabbit monoclonal p21 (Cell Signaling Technology, 2947), rabbit polyclonal cyclin A1 (Abcam, ab53699), mouse monoclonal phospho–serine 10 H3 (Abcam, ab14955), rabbit polyclonal FAS (Abcam, ab82419), rabbit monoclonal PARP (Abcam, ab32138), mouse monoclonal MDM2 (Santa Cruz Biotechnology, sc965), mouse monoclonal MDM2^m (mixture of 2A9, 4B2, and 4B11 clones, provided by M. Oren), mouse monoclonal MDM2 (SMP14, sc-965), and mouse S9.6 (provided by O. F. Capetillo). Unless otherwise stated, all antibodies were used at a working dilution of 1:500 for Western blotting and 1:400 for IF. Secondary antibodies used are as follows: mouse Alexa Fluor 647 (Thermo Fisher Scientific, A-21235), rabbit Alexa Fluor 647 (Thermo Fisher Scientific, A-21244), rabbit Alexa Fluor 488 (Thermo Fisher Scientific, A-11008), mouse Alexa Fluor 488 (Thermo Fisher Scientific, A-11029), mouse horseradish peroxidase (HRP; Sigma-Aldrich, A9044), and rabbit HRP (Sigma-Aldrich, A6154). Secondary antibodies for IF were used in dilution 1:500 and for immunoblotting in dilution 1:10,000.

Commercially available SMARTpool ON-TARGET oligonucleotides targeting human eIF4A3 (L-020762), uL5 (RPL11) (L-013703), CASP8 (L-003466), CASP9 (L-003309), TP53 (L-003329), BAX (L-003308), BBC3 (L-004380), and nontargeting siRNA (D-001810) were purchased by Horizon Discovery (Dharmacon). siRNAs against 5S rRNA and RPL5 (uL18) were custom made Horizon Discovery (Dharmacon) according to (56) and (30), respectively. Cells were transfected with 20 nM siRNA using Lipofectamine RNAiMAX reagent (Thermo Fisher Scientific, 13778150) according to the manufacturer’s instructions. After 48 hours or the times indicated, cells were harvested and lysed for RNA or protein extraction.

Cell proliferation was assessed using the resazurin assay. Cells were seeded at a density of 2000 cells per well in 96-well plates; the next day, they were transfected with siRNAs for 48 to 72 hours, while chemical addition took place within the last 24 hours before measuring optical density (OD). Resazurin working solution (Sigma-Aldrich, R7017) was added to supernatants to a final concentration of 50 μg/ml. The plates were further incubated for 4 hours at 37°C, and the emitted fluorescence was measured with a microplate reader using the 560-nm excitation/590-nm emission filter set (Tecan Infinite M1000 Pro). All growth assays were performed a minimum of three times using independent platings of cells to ensure reproducibility. Colony formation assays were performed according to Franken et al. (57).

Following chemical administration or transfection, subconfluent cells were lysed in radioimmunoprecipitation assay (RIPA) buffer supplemented with a cocktail of protease and phosphatase inhibitors (Thermo Fisher Scientific, 78444) and sonicated for five cycles of 30-s on and 15-s off, in a Bioruptor (Diagenode). Following lysate clearance with centrifugation for 10 min at 13,000 rpm and 4°C, protein quantification was performed with the DC Protein Assay Kit II (Bio-Rad, 5000112). Cell lysate (20 μg) was boiled in Laemmli sample buffer for 5 min at 95°C, loaded onto SDS–polyacrylamide gel electrophoresis gels and transferred onto nitrocellulose or polyvinylidene difluoride membranes. Chemiluminescence signal was detected using SuperSignal West Dura (Thermo Fisher Scientific, 34076). Images were acquired with an Amersham Imager 600 scanner. Cell fractionation was performed as previously described (58). Data regarding eIF4A3 subcellular localization were acquired from PepTracker (59). Trichloroacetic acid precipitation was used for protein extraction following polysome profiling.

Total RNA extraction was performed with the PureLink RNA Mini Kit (Thermo Fisher Scientific, 12183025), and RNA extraction from polysome profiling fractions was achieved with TRIzol, followed by clarification with the Norgen Biotek Corp RNA Clean Up/Concentration kit (Norgen Biotek, 298-23600). Quantitative RT-PCR was conducted with the TaqMan RNA-to-CT 1-Step Kit (Thermo Fisher Scientific, 4392938) in a StepOnePlus Real-Time PCR System (Applied Biosystems). TaqMan probes used in this study are listed below. For identification of MDM2 mRNA species, RNA was reverse-transcribed using SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific, 18090050) using random hexamer primers to generate cDNA. The cDNA was treated with 5 U of RNase H (New England Biolabs, M027) for 30 min at 37°C and then analyzed by PCR using the external primer set from Sigalas et al. (60) and a Veriti 96-Well Thermal Cycler (Thermo Fisher Scientific) [30 cycles; T_m (melting temperature): 55°C]. Primers and PCR conditions for rRNA species and apoptotic genes were previously described (30). TaqMan probes (Thermo Fisher Scientific): EIF4A3: 4331182 (Hs01556773_m1); and MDM2: 4400291 (Hs02970282_cn), FAS (Hs00236330_m1), BBC3 (Hs00248075_m1), and TP53 (Hs01034249_m1).

RNA-seq was performed by National Genomics Infrastructure (NGI) at Science for Life Laboratory Stockholm, Sweden. Briefly, following RNA quality assessment by TapeStation electrophoresis (Agilent), libraries were prepared with Illumina RiboZero TruSeq Stranded mRNA, and samples were sequenced in a HiSeq 2500 system (Illumina). Data preprocessing was performed in NGI through nf-core/rnaseq (61). DE analysis was performed using DESeq2 (v1.24.0) (62). Transcript level quantification was performed with Salmon (63), followed by DESeq2-mediated DE. Exon-based differential analysis was achieved with DEXSeq (51) and splicing analysis with IsoformSwitchAnalyzeR (64). For downstream GO analysis, DESeq2-produced data were filtered using a log₂ fold change > |1| and P < 0.05 threshold and processed via ClueGO (v2.5.4, ontologies update 09 April 2018) (65) in Cytoscape (66) or g:Profiler (67). Z score in GO analysis gives the likelihood that a specific GO term is increased or decreased and is described in (68). Transcription factor enrichment analysis was executed with iRegulon (v1.3) (69) in Cytoscape. CLIP-seq data from (70) were analyzed with DESeq2. RNA-seq data following eIF4A3 chemical inhibition were incorporated from (31). All analyses were performed in RStudio (v1.1.453).

Samples were lysed by 4% SDS lysis buffer and prepared for MS analysis using a modified version of the SP3 protein clean-up and digestion protocol (71). Peptides were labeled with TMT10plex reagent according to the manufacturer’s protocol (Thermo Fisher Scientific) and separated by immobilized pH gradient–isoelectric focusing (IPG-IEF) on 3 to 10 strips as described previously (72). Extracted peptide fractions from the IPG-IEF were separated using an online 3000 RSLCnano system coupled to a Thermo Scientific Q Exactive-HF. MSGF+ and Percolator in the Galaxy platform was used to match MS spectra to the Ensembl_92 Homo sapiens protein database (73). DE analysis was performed with DEP (74) in RStudio (v1.1.453). DE proteins passing the threshold of α = 0.85 and lfc (log fold change) = 0.1 were used subsequently for STRING (v. 11) (75) and GO analyses.

RIP was performed according to previously published data (76). Briefly, cells were transfected with 20 nM control siRNA or siRNA against eIF4A3 for 48 hours. The cells were then fixed in 1% formaldehyde and lysed. Following sonication and DNA digestion, equal amounts of lysate were immunoprecipitated with either 3 μg of mouse α-eIF4A3 (Santa Cruz Biotechnology, sc-365549) or mouse immunoglobulin G (IgG) using G-agarose magnetic beads. The next day, the beads were washed thoroughly, and following elution of RNA-protein complexes, RNA was isolated with phenol/chloroform and used as template in qPCR for U3 snoRNA (table S4C) or 18S rRNA (30).

Cells were seeded on 6-well plates containing coverslips (Thermo Scientific Nunc Thermanox Coverslips, 12-565-88) or 96-well plates 1 day before treatment. Following treatment, cells were washed three times with phosphate-buffered saline (PBS), fixed in 4% formaldehyde (Sigma-Aldrich, F8775) for 10 min, and permeabilized for 10 min using PBS with 0.5% Triton X-100 (Sigma-Aldrich, X100). They were subsequently washed three times with PBS and incubated for 1 hour at room temperature in blocking solution [3% bovine serum albumin (BSA) in PBS] (Sigma-Aldrich, A7906). Fixed cells were incubated at room temperature with primary antibodies for 1 hour at dilution 1:400 (unless otherwise stated), washed three times in PBS, and incubated with secondary antibodies for 1 hour at room temperature (dilution, 1:500). Last, cells were stained with 10 μM Hoechst (Thermo Fisher Scientific, 62249) for 30 min at room temperature, and the cover slips were mounted onto slides. FBL was used as a marker for nucleolar segregation. Images were acquired using a Nikon Eclipse Ti2 inverted epifluorescence microscope or an inverted Zeiss LSM 780 confocal microscope. EdU/EU staining was performed with the Click-iT EdU Cell Proliferation Kit for Imaging and Alexa Fluor 647 dye (Thermo fisher Scientific, C10340) or Click-iT RNA Alexa Fluor 488 Imaging Kit (Thermo fisher Scientific, C10329), respectively, OPP staining with the Click-iT Plus OPP Alexa Fluor 488 Protein Biogenesis Assay Kit (Thermo Fisher Scientific, C10456) and HPG staining with the Click-iT HPG Alexa Fluor 488 Protein Synthesis Assay Kit (Thermo Fisher Scientific, C10428) according to the manufacturer’s instructions. Images were acquired using an IN Cell Analyzer 2200 (GE Healthcare) and analyzed using Cell Profiler, Fiji (Image J), and RStudio (v1.1.453).

For puromycin pulse-chase assays, cells were incubated with 1 μM puromycin (Sigma-Aldrich, P8833) for 60 min and lysed in RIPA containing a cocktail of protease and phosphatase inhibitors (Thermo Fisher Scientific, 78444). Puromycin incorporation was visualized by Western blotting with an antibody against puromycin. For polysome profiling, cells in 70 to 80% confluent 15-cm dishes were washed twice in PBS containing CHX (100 μg/ml) and lysed after treatment in hypotonic buffer supplemented with CHX (Sigma-Aldrich, C7968) (44). Lysates were cleared by centrifugation, and equal amounts (adjusted by 260-nm OD measurement) were loaded onto 5 to 50% sucrose gradients produced in sucrose gradient buffer with a gradient forming unit (BioComp). Samples were centrifuged at 36,000 rpm for 3 hours using a SW41Ti rotor in an Optima XE-90 ultracentrifuge (Beckman Coulter). The samples were then analyzed in a piston gradient fractionator (BioComp) using the company’s software. Twenty-eight fractions were collected at a pace of 2.86 mm per fraction (absorbance at 540 nm) and were further used for protein extraction (see section ‘Preparation of cell extracts and Western blotting’).

RNA was isolated from 80S monosomes and polysomes (>3 ribosomes) fractions by using phenol/chloroform extraction method and purified further with Agencourt AMPure RNAClean XP beads (Beckman Coulter, A62987). RNA-seq libraries were prepared using a TruSeq Stranded Total RNA library preparation kit (Illumina, 20020596) and TruSeq RNA Single Indexes (Illumina, 20020492) according to the manufacturer’s protocol, starting with 100 ng of RNA. SuperScript IV Reverse Transcriptase (Thermo Fisher Scientific, 1809001) was used for cDNA synthesis, and DNA was purified with Agencourt AMPure XP beads (Beckman Coulter, A63881). Quality assessment and concentration estimation of the purified RNA and DNA were performed using the Qubit 3 (Life Technologies) and BioAnalyzer 2100 (Agilent). Each library was diluted to 4 nM and combined equimolar into a single pool before sequencing on the Illumina NextSeq500 platform (Illumina) using the NextSeq 500/550 High Output Kit v2.5 (150 cycles) (Illumina, 20024907). RNA-seq data were processed using nf-core/rnaseq (61) pipeline using GRCh37 genome build. Following initial multidimensional scaling analysis, one sample (from polysome fraction) was omitted in downstream DE analyses due to poor alignment efficiency. DE analysis was performed using DESeq2 (v1.24.0) (62) and deltaTE (45).

Following treatment, live FUCCI-U2OS cells were washed twice and resuspended in PBS containing 10% BSA. Samples of 200 to 250 μl (5000 events) in 96-well round-bottom plates were analyzed with the Guava easyCyte 5HT HPL Benchtop Flow Cytometer (Millipore). Cell cycle stages were determined using the InCyte 3.3 software with the gain controls set to 7.66 (forward scatter), 1.0 (side scatter), 3.83 (red/green fluorescence), 3.36 (yellow fluorescence), and 7.8% yellow-green compensation.

Cells were seeded in 96-well plates at low density (2000 cells per well), treated as indicated, and stained for caspase 3/7 with CellEvent Caspase-3/7 Green Detection Reagent (Thermo Fisher Scientific, C10423) according to the manufacturer’s guides. Cells were counterstained with 10 μM Hoechst (Thermo Fisher scientific, 62249) for 30 min, and images were acquired with IN Cell Analyzer 2200 (GE Healthcare) and analyzed using Cell Profiler.

Cells were fixed, after treatment, in 2.5% glutaraldehyde/0.1 M phosphate buffer (pH 7.4) at room temperature for 30 min and were then collected and in Eppendorf tubes and incubated overnight at 4°C. Cell pellets were subsequently rinsed in PBS, fixed in 2% osmium tetroxide (TAAB, Berks, England)/0.1 M phosphate buffer (pH 7.4) at 4°C for 2 hours, dehydrated in ethanol/acetone, and embedded in LX-112 (Ladd, Burlington, Vermont, USA). Ultrathin sections (~50 to 60 nm) were cut with Leica EM UC 6 (Leica Wien, Austria), incubated in uranyl acetate and lead citrate, and examined with a Hitachi HT 7700 (Tokyo, Japan) at 80 kV. Digital images were acquired with a Veleta camera (Olympus Soft Imaging Solutions, GmbH, Münster, Germany). Transmission EM was performed in EMil Core Facility at KI.

RNA-seq and CLIP-seq DE analysis was performed using the R packages DESeq2 (v1.25.7) (62) and DEXSEq (v1.31.0) (51) and the bash-scripting software for transcript quantification Salmon (v0.14.1) (63). Polysome profile RNA-seq DE analysis was performed using the R packages deltaTE (45). Proteomics analysis was achieved via the R package DEqMS (77). Graphs were produced in ggplot2 (v3.2.0) (78) or GraphPad Prism (v8.0.1) and graphical illustrations in Adobe Illustrator.

Synthesis of eIF4A3 chemical inhibitors (4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazin-1-yl)(6-bromopyrazolo[1,5-a]pyridin-3-yl)methanone and 3-(4-(4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazine-1-carbonyl)-5-methyl-1H-pyrazol-1-yl)benzonitrile (compounds 52a and 52b, respectively) were prepared as racemic compounds according to (79). Both compounds were separated into enantiomers using preparative supercritical fluid chromatography (SFC). Both enantiomers were collected for each compound and were analyzed by LC-MS and nuclear magnetic resonance. All fractions were analyzed by chiral SFC and 52b and 52a by optical rotation. Analytical data were in accordance with (79). Compound 52a was separated into enantiomers using a CHIRALPAK ID column (10 mm by 250 mm) eluting with 55% CO₂ and 45% MeOH, a flow of 15 ml/min, and a temperature of 45°C. The first eluting fraction was assigned as 52a, (R)-(4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazin-1-yl)(6-bromopyrazolo[1,5-a]pyridin-3-yl)methanone, and the second fraction was assigned as 52b (R)-(4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazin-1-yl)(6-bromopyrazolo[1,5-a]pyridin-3-yl)methanone. Compound 53 was separated into enantiomers using an YMC Chiral Cellulose SB-column (10 mm by 250 mm) eluting with 70% CO₂ and 30% MeOH, a flow of 15 ml/min, and a temperature of 45°C. The first eluting fraction was assigned as 53a (S)-3-(4-(4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazine-1-carbonyl)-5-methyl-1H-pyrazol-1-yl)benzonitrile, and the second fraction was assigned as 53b (R)-3-(4-(4-(4-bromobenzoyl)-3-(4-chlorophenyl)piperazine-1-carbonyl)-5-methyl-1H-pyrazol-1-yl)benzonitrile. Compound specificity was described in (79).

All tissues used for immunohistochemical analysis were archival specimens routinely fixed in buffered formalin and embedded in paraffin. The information about cohorts of human tissue and tumor specimens used in the present study can be found in our published studies on DNA damage response markers on these tissues, as follows: urinary bladder (normal tissue, n = 12; early Ta-T1 lesions, n = 222; and invasive T2-T4 tumors, n = 230) (80), colon (normal mucosa, n = 10; adenomas, n = 19; and colon carcinomas, n = 15) (81), uterine cervix (normal tissue, n = 3; dysplasia, n = 4; squamous carcinoma in situ, n = 5; and invasive squamous carcinoma, n = 15) (82), and brain tissue and glioblastomas (normal brain, n = 6; and glioblastoma, n = 43) (83). The tumors were classified and staged following the official international guidelines for grading from the World Health Organization. The studies were approved by the relevant ethical committees in Denmark. To detect the eIF4A3 protein and its staining patterns in the above listed human tissue and tumor specimens, we used our well-established sensitive immunohistochemical staining protocol (80–82, 84). Standard deparaffinization of the archival formalin-fixed, paraffin-embedded tissue sections was followed by antigen unmasking in the citrate buffer (pH 6, for 15 min in a microwave). After overnight incubation with the primary antibody, samples were processed for the indirect streptavidin-biotin-peroxidase method using the VECTASTAIN Elite Kit (Vector Laboratories) and nickel sulfate–based chromogen enhancement detection as previously described (83), without nuclear counterstaining (80–82, 84). The primary antibody against eIF4A3 used here was a rabbit polyclonal antibody used at 1:250 dilution (Santa Cruz Biotechnology, sc67369). For negative controls, sections were incubated with nonimmune rabbit serum. For positive controls, a rabbit antibody against human phospho-histone H2A.X (Ser¹³⁹) was used. Results were evaluated by a senior oncopathologist, and data were expressed in three scoring categories based on the percentage of positive tumor cells with defined subnuclear expression patterns of the eIF4A3 protein as follows: category (I) pan-nuclear, with eIF4A3 detected with similar intensity in both nucleolus and nucleoplasm in 95% or more cells; category (II) eIF4A3 preferentially accumulated in nucleoli over nucleoplasm in at least 6% cells (6 to 100% of cells per lesion); or category (III) eIF4A3 being selectively excluded from nucleoli in at least 6% cells (6 to 100% of cells per lesion), respectively.

To determine the mutational status of eIF4A3 in cancer, we combined TCGA data describing eIF4A3 mutations in multiple cancer types with point mutations presented by Bono et al. (43). Correlation of eIF4A3 mRNA levels and TCGA cancer patient survival rates was accomplished with LinkedOmics (17). Comparison of eIF4A3 mRNA expression between cancer and normal tissue was performed with GENT2 (Gene Expression database of Normal and Tumor tissues 2) (16). For the division of cancer cell lines in low- and high-RiBi groups, we used RNA-seq expression data from DepMap (v.20Q3) (18). For each gene of a curated (RiBi-related) gene list (based on Reactome pathways: RHSA73864 and RHSA72312 excluding the histones; table S1D), the median was calculated, and cell lines were characterized as high RiBi when the number of the genes scoring higher than each median was bigger than the number of the genes scoring lower than each median and vice versa. Compound cytotoxicity data were extracted from PRISM primary repurposing drug screen (DepMap, 19Q4).

RNA-seq and proteomics experiments were performed in biological triplicates, and statistical analysis followed the tests provided in the relevant DE analysis packages in R. quantitative RT-PCR–, image-, survival-, and apoptosis-associated experiments were performed a minimum of three times with independent biological samples. Parametric (N > 100) or nonparametric t test (N < 100) or analysis of variance (ANOVA) was used for statistical testing. Sample size (N) and P (or FDR) values are provided in the corresponding figures or figure legends.