Introduction

Cold-shock proteins (CSPs) are a family of multifunctional DNA/RNA binding proteins that contain a highly conserved nucleic acid binding domain called the cold shock domain. YBX1 is a pleiotropic DNA and RNA binding protein that modulates translation, RNA-stability, mRNA splicing, transcription or cell signaling depending on cell type and genetic background [1,2,3,4,4B). However, genes that were differentially expressed following YBX1 deletion did not show relevant YBX1 binding. Notably, YBX1-DNA binding to a specific gene may be associated with a repressive function, since we detected a trend for YBX1-bound genes to show increased expression following YBX1 deletion (Supplementary Fig. 4C).

YBX1 mediates translation in a transcript-dependent manner

In order to generate a global view on the functional properties of YBX1 in AML, we performed a genome-wide CRISPR-Cas9 screen in MOLM13 cells comparing the genetic vulnerabilities of YBX1-knockout and control cells (Fig. 5A). This functional genomics approach enabled us to screen in an unbiased manner for cellular networks that are specifically affected by YBX1 loss. As expected, genetic deletion of YBX1 reduced cellular proliferation thus providing the required selective pressure to conduct the screen (Fig. 5B). Following Next-Generation Sequencing, alignment and quantification of each guide-RNA barcode to the respective guide library, p values and corresponding beta-scores were calculated for each gene (Fig. 5C). Positive beta scores represent an enrichment of guides targeting a certain gene over time, typically being interpreted as a tumor-suppressor-like function, while negative beta-scores represent selective dependencies resulting in out-competition. The beta score of each gene in the non-targeting (NT) control condition was then subtracted from the respective score in the YBX1-knockout condition to generate a Δbeta-score that reflects differential dependency (Fig. 5D). When performing GSEA for REACTOME-terms on the ranked list of Δbeta-scores, the top 15 enriched terms reflected pathways and functions associated with translational initiation and elongation (Fig. 5E). Most genes associated with these terms represent functional dependencies in the NT-control condition, since translation mediators and ribosomal subunits are important housekee** genes but lose this specific gene-dependency in the YBX1-knockout setting (Supplementary Fig. 5A). This finding suggests that YBX1 exerts its function via these molecules. Of note, among the top differential dependencies, several targets had previously been identified as protein binding partners of YBX1 [16], highlighting the power of functional genomic screening for the identification of functional molecular networks (Fig. 5F).

Fig. 5: YBX1 modulates translation in a transcript-dependent manner.
figure 5

A Schematic depicting the experimental procedure to conduct the genome-wide CRISPR-Cas9 screen. MOLM13 cells were transduced YBX1-sgRNA1 or non-targeting control in the p.LKO5-GFP vector system. Cells were subsequently sorted for GFPhigh expressing cells and expanded to facilitate genome-wide screening. Lentiviral transductions with 2 paired human whole genome-libraries (H1/H2) [25] were performed at low multiplicity of infection (MOI) and cells were selected with puromycin for 3 days. The screen was carried out for 12 days after puromycin selection was completed and the baseline DNA sample was harvested. B Growth curve of MOLM-13 cells with a YBX1-KO (blue line) or empty vector control (gray line) during the CRISPR-Cas9 screen for the estimation of the selective pressure applied during the screen. C Volcano-plots showing the distribution of genes being enriched (positive beta-scores) or depleted (negative beta-scores) in the genome-wide CRISPR-Cas9 screen in cells harboring a YBX1-KO compared to control. D Dot-plot of ranked differential CRISPR-Cas9 screening hits between the YBX1-KO and control condition. The X axis shows the gene-rank, on the Y axis the differential beta scores (Δbeta-scores) are plotted. E Top15 REACTOME-terms showing differential enrichment in the CRISPR-Cas9 screening dataset between the YBX1-KO and the control condition. F Dot-plot of ranked differential CRISPR-Cas9 screening hits between the YBX1-KO and control condition. Highlighted as red or blue dots are the genes among the top 500 differentially enriched or depleted genes in the YBX1-KO condition, that have previously been identified as binding partners of YBX1 in IP-Mass spec [16]. G Schematic depicting the experimental procedure underlying the polysomal RNA profiling in MOLM-13 harboring a YBX1-KO or non-targeting control (sgLUC). MOLM13 cells were transduced with YBX1-sgRNA (or NT-control), selected with puromycin and expanded for 14 days prior to polysomal fractionation followed by RNA-sequencing. H Volcano-plot of differentially expressed genes in RNAseq 14 days after knockout of YBX1 by CRISPR-Cas9 (YBX1-sgRNA1). Left: DEGs in YBX1-KO cells compared to control when sequencing the total cellular mRNA content. Right: DEGs in YBX1-KO cells when sequencing mRNAs that are bound to polysomal chains. I Correlation between the magnitude of gene-loss from the polysomal fractions (Y axis: log2FC) upon YBX1-KO and the dependency of the respective genes from the cancer dependency map portal (X axis: CERES gene effect, depmap.org). The red dotted line marks the arbitrary cutoff of genes that are generally considered functional dependencies. Genes that are labeled in red are genes that have been identified as differentially expressed hits in proteome screening. J Pie chart showing the proportion of genes that were found to be lost from polysomes and are known binding partners of YBX1 (blue) in relation to genes that have not been shown to bind to YBX1 (gray). K Correlation between the magnitude of gene-loss from the polysomal fractions (Y axis: log2FC) upon YBX1-KO and the differential dependency of the respective genes in our CRISPR-Cas9 screen in YBX1-KO vs. control cells. Genes that are labeled in green are genes that have been identified as RNA-targets of YBX1 in iCLIPseq [2].

Full size image

In order to assess for the ability of YBX1 to influence translation of mRNAs, we performed transcriptomic profiling from purified ribosomal fractions (Fig. 5G, Supplementary Fig. 5G, 6). Recruitment of mRNAs to polysomal chains is a major mechanism to increase the output of protein synthesis per mRNA molecule and is therefore considered a crucial determinant of translation efficiency. Consistent with our observations from RNA-sequencing (day 7), the number of DEGs in the bulk RNAseq-sample appeared rather limited (Fig. 5H, left panel). Genes showing reduced expression were predominantly translation initiation factors with EIF4B showing the strongest reduction on the protein level (Supplementary Fig. 5B–D). In contrast, we observed a large number of genes being differentially expressed within the polysomal fractions (Fig. 5H, right panel). The number of DEGs detected after polysomal fractionation was about 20-fold increased, compared to bulk mRNA and some genes showed a high magnitude of change. Of note, forced expression of EIF4B as the single initiation factor that was consistently and strongly affected by YBX1-ko on the total RNA and protein level was not sufficient to rescue the competitive disadvantage of YBX1-inactivation (Supplementary Figure 5E,F), suggesting a direct impact of YBX1 on polysomal transcript recruitment. Relevant YBX1-targets on polysomes were validated on the protein level by Western blot (Supplementary Fig. 5H). To identify candidates that are lost from polysomes and represent relevant functional dependencies, we integrated the magnitude of loss from the polysomes of each significantly down-regulated gene (adjusted p < 0.05, fold change >1.5) with the CERES gene effect score from genome-wide CRISPR-Cas9 screens (Broad-Institute, Achilles-portal). 153/747 (20.5%) of genes lost from the polysomes were shown to be functional dependencies identified by CRISPR-Cas9 editing (CERES-score < −0.5) (Fig. 5I). Importantly, a number of those genes, including cell cycle mediators and ribosome subunits showed decreased expression in global proteome analysis (Fig. 5I, highlighted in red). Furthermore, 30% (n = 226) of genes that were lost from the polysomes represent RNA-binding targets of YBX1 in iCLIP-sequencing analyses (Fig. 5J) [2]. Finally, we aimed to understand how genes that are lost from polysomes are associated with YBX1-dependent functional pathways. Therefore, we integrated the magnitude of loss from polysomes with the respective functional dependencies (Δbeta-scores; Fig. 5K). Here, relevant targets could be identified that were differentially recruited to polysomes and also enriched following CRISPR-Cas9 editing. Several of these targets, including MYC, were also CLIP-targets of YBX1. GSEA showed significant loss of the MYC target gene signature (Fig. 6A). Using iCLIP-sequencing it had been demonstrated, that YBX1 is consistently bound to MYC-transcripts, establishing MYC as a high confidence mRNA-binding partner of YBX1 (Fig. 6B). Importantly, genetic inactivation of YBX1 did not affect MYC-transcript abundance in bulk RNA-sequencing (Fig. 6C). In contrast, MYC mRNA was significantly lost from the polysomal mRNA fraction upon YBX1-deletion demonstrating an involvement of YBX1 in the recruitment of MYC transcripts to polysome chains (Fig. 6C). Consequently, using two different sgRNAs that reduce YBX1 expression to a different extent, gene-dose dependent reduction in MYC expression could be confirmed (Fig. 6D). Likewise, MYC was a prominent dependency in MOLM13 cells, an effect that was significantly reduced following genetic deletion of YBX1 (Fig. 6E). The fact that MYC was identified as a relevant driver of YBX1 dependent gene expression and YBX1 is binding to MYC mRNA, indicates its role as a direct downstream effector. In line with a recent report demonstrating IGF2BP-family proteins as being critical for YBX1-binding to its target mRNAs [5], IGF2BP2-knockout was shown to mediate resistance to YBX1-inactivation (Fig. 6F).

Fig. 6: YBX1 regulates MYC by modulating polysomal recruitment.
figure 6

A Geneset enrichment analysis showing a loss of a MYC target gene signature in MOLM-13 cells harboring a knockout of YBX1 (sgRNA1/2, day 7 after lentiviral transduction). B Representative iCLIPseq tracks over the MYC gene from 3 independent replicates, showing binding of YBX1 to MYC in breast cancer [2]. C Bar graph depicting gene expression changes of MYC in total RNAseq (gray) and polysomal RNAseq, showing that a reduction in MYC-expression appears to be restricted to the polysomal fractions rather than to the unfractionated total RNA. D Western Blot showing the protein expression of MYC in relation to the knockout efficiency of YBX1 using 2 different sgRNAs in MOLM-13-Cas9 cells. E Bar graph visualizing the beta scores of MYC in control MOLM-13 cells compared to YBX1-KO cells. F Bar graph visualizing the beta scores of IGF2BP2 in control MOLM-13 cells compared to YBX1-KO cells. G Schematic showing the proposed model of YBX1-action in AML by binding to its target RNAs and mediating their recruitment to polysomal chains to drive productive translational output.

Full size image

Taken together, we propose, that YBX1 associates with target mRNAs, including MYC, and thereby modulates translational output by recruitment of relevant mRNAs to polysomal chains (Fig. 6G). Protein expression of MYC (among other mediators of cell cycle progression and cellular homeostasis) appears to be stabilized through YBX1 due to its preferential recruitment to polysomes. Furthermore, YBX1 may indirectly influence translation by regulating the availability of ribosomal building blocks and translation mediators (Fig. 6G). Therefore, genetic inactivation of YBX1 impacts the translational output of transcripts on the protein level and thereby selectively modulates protein abundance of oncogenic drivers and influences proliferative capacity and cell competition in AML.

Discussion

Identification of therapeutic targets that are tractable vulnerabilities and selective dependencies in cancer while being dispensable for normal tissues represent the ideal prerequisite for the development of cancer therapies. Cold shock protein YBX1 has been identified as a pan-cancer dependency in publicly available CRISPR-Cas9-screens and several studies in different tumor entities [1,2,3,4,5, 15]. Conversely, genetic inactivation of YBX1 had no deleterious effects on normal hematopoiesis [16], making it a potentially interesting therapeutic target for cancer therapy. Consistent with recent reports [5], we have shown, that YBX1 is required for development and maintenance of human and murine AML in vitro and in vivo. Even though the cold shock domain as a common structural component of the CSPs is conserved among the family members, only YBX1 showed a potent phenotype in leukemia as well as in other cancers.

Mechanistically, we demonstrate that deletion of YBX1 in AML shows minor impact on mRNA abundance, while having significant effects on the cellular proteome. Moreover, no relevant increase in mis-spliced isoforms could be found in AML cells after deleting YBX1, clearly distinguishing the apparent mechanisms in AML from our previous findings in MPN, where YBX1 was acting as a relevant splicing factor [16]. Using an unbiased multi-omics screening approach, we found that YBX1 mediates translation of specific transcripts in AML, which is in line with previous reports [1, 4, 15]. To the best of our knowledge this is the first report describing a global CSP regulatory network using functional genomics. Taken together, our data provide strong evidence for YBX1 acting as a cancer-specific modulator of translation in AML, while leaving total mRNA levels largely unaffected.

A recent report published by Feng and colleagues [5] complements our findings by providing novel insights into how YBX1 binds its target mRNAs in leukemia cells. YBX1 appears to bind to methylated (m6A) transcripts via IGF2BP-family of proteins to facilitate RNA binding and stabilization. In line with this claim, we find that deletion of IGF2BP2 confers resistance to YBX1-inactivation in our CRISPR-Cas9 screen. Structurally, the cold-shock domain seems to be required for both IGF2BP- and mRNA-binding of YBX1. Consistent with our findings, Feng et al. report an impact of YBX1 on MYC expression and show that its expression can rescue the phenotype evoked by inactivation of YBX1.

In contrast to our findings the authors assume that regulation of RNA stability represents a major mechanism of YBX1-action in AML, similar to findings described in breast cancer [2]. This assessment is based on experimental data showing that shRNA-mediated knockdown of YBX1 can affect RNA abundance [5]. However, when we conducted parallel RNA-sequencing comparing RNAi- and CRISPR-mediated genetic inactivation of YBX1 (to rule out a potential bias) we found regulation of RNA-stability exclusively in RNAi- but not CRISPR-treated AML cells. In RNAi-treated samples, we observed high numbers of DEGs, including MYC, BCL2 and MCL1, consistent with findings described by Feng and colleagues (Supplementary Fig. 5I). Absence of these findings in cells treated with CRISPR-Cas9 technology indicate that an intracellular defense and stress response when using RNAi may influence gene expression changes. Therefore, we assume that the mechanism of action and kinetics of YBX1 inactivation substantially influence experimental results.

Taken together, our data and the findings presented by Feng et al. establish YBX1 as a selective genetic vulnerability in leukemia without major restrictions towards specific genetic subtypes.

Of note, a novel small molecule, SU056, was recently reported to directly bind and inhibit YBX1 [32]. SU056 demonstrated activity in ovarian cancer models in vitro and in vivo and showed favorable biochemical and pharmacologic properties. The availability of this compound will allow direct targeting of YBX1 in pre-clinical models and may facilitate translation into early clinical trials in AML.