Targeting an RNA-binding Protein Network in Acute Myeloid Leukemia

SUMMARY

RNA-binding proteins (RBPs) are essential modulators of transcription and translation frequently dysregulated in cancer. We systematically interrogated RBP dependencies in human cancers using a comprehensive CRISPR/Cas9 domain-focused screen targeting RNA-binding domains of 490 classical RBPs. This uncovered a network of physically interacting RBPs upregulated in acute myeloid leukemia (AML) and crucial for maintaining RNA splicing and AML survival. Genetic or pharmacologic targeting of one key member of this network, RBM39, repressed cassette exon inclusion and promoted intron retention within mRNAs encoding HOXA9 targets as well as in other RBPs preferentially required in AML. The effects of RBM39 loss on splicing further resulted in preferential lethality of spliceosomal mutant AML, providing a strategy for treatment of AML bearing RBP splicing mutations.

eTOC blurb

Using a CRISPR/Cas9 screen targeting RNA-binding domains of classical RNA-binding proteins (RBPs), Wang et al. uncover a network of interacting RBPs upregulated in acute myeloid leukemia (AML) and crucial for RNA splicing and AML survival, highlighting RBM39 as a central, targetable component of the RBP network.

Graphical Abstract

INTRODUCTION

Eukaryotic cells employ a wide range of mechanisms to regulate the fine-tuning of mRNA expression (Glisovic et al., 2008). These co- and post-transcriptional processes are orchestrated by interactions between RNA molecules and RNA-binding proteins (RBPs). RBPs are a diverse class of proteins containing unique RNA-binding domains (RBDs) (Cook et al., 2011). Classical RBPs encompass one or multiple conventional RBDs that include, but are not limited to, RNA recognition motifs (RRMs), K-homology (KH) domains, and DEAD-box domains, that have been extensively determined structurally and biochemically (Lunde et al., 2007). Considering that RBPs are key regulators of gene expression, alterations of these proteins are also implicated in several human genetic diseases, including multiple cancers (Cooper et al., 2009; Kapeli et al., 2017; Lukong et al., 2008).

Acute myeloid leukemia (AML) is an aggressive hematological malignancy with a dismal survival rate (<30% five-year overall survival rate), highlighting the need for improved therapeutic interventions (Dohner et al., 2015). Advances in next-generation sequencing have exposed the genetic and epigenetic heterogeneity in AML pathogenesis (Papaemmanuil et al., 2016). Analyses of the genomic landscape in myeloid leukemia patients have discovered several somatic genetic lesions, including frequent mutations in RBPs which serve as splicing factors (Yoshida et al., 2011). Although the mechanisms by which spliceosome mutations promote transformation are still being investigated, several studies have demonstrated that such mutations alter RNA binding preferences to promote altered splicing as key in the molecular pathogenesis of these mutations (Dvinge et al., 2016). AML cells bearing splicing factor mutations exhibit increased sensitivity to pharmacologic modulation of splicing, thereby providing a therapeutic strategy for patients harboring spliceosome mutations (Lee and Abdel-Wahab, 2016; Lee et al., 2016; Obeng et al., 2016; Seiler et al., 2018). This has led to an ongoing phase I clinical trial to determine the safety and efficacy of a modulator of core splicing function H3B-8800 in patients with refractory leukemias (Seiler et al., 2018).

In the past few years, there has also been a deeper understanding of “non-oncogene addiction” of RBPs in cancers, which is the scenario in which RBPs are essential for cancer cell survival but are not targeted by genetic mutations. A few such RBP dependencies have already been identified by studying RBPs that are upregulated in diverse tumors (Anczukow et al., 2012; Fox et al., 2016; Wurth et al., 2016). Despite recent progress in uncovering essential RBPs in leukemia maintenance (Kharas et al., 2010; Palanichamy et al., 2016), the promise of targeting RBPs therapeutically is still limited by a lack of systematic evaluation for required RBPs in cancer. Here, our goal was to use functional screens to dissect the roles of classical RBPs in AML.

RESULTS

Aberrant RBP expression in AML Patient Samples

Aberrant RBP expression has been commonly linked to promoting cancer progression through co- and post-transcriptional mechanisms. Analysis of datasets from The Cancer Genome Atlas (TCGA) uncovered numerous differentially expressed RBPs across various solid cancers, suggesting that RBPs are often dysregulated in tumors (Sebestyen et al., 2016). To define the extent to which RBP genes are altered in AML, we analyzed the mRNA expression of 484 classical RBPs in AML patients using publicly available data from the TCGA and Leucegene (Cancer Genome Atlas Research et al., 2013; Lavallee et al., 2015). Comparison of RBP mRNA expression in AML patients to normal human CD34+ human hematopoietic stem/progenitor cells identified approximately 51% (247/484) of the genes to be differentially expressed (Figure 1A and Table S1). Among the dysregulated RBPs in AML, we identified 107 RBP genes that were significantly upregulated (p<0.05), whereas 140 genes were found to be downregulated. These data support previous findings that aberrant expression of RBPs occurs ubiquitously in cancers (Sebestyen et al., 2016).

Figure. 1: A CRISPR Domain-Targeted Screen Identifies RBP Dependencies in AML.

(A) Volcano plot of differentially expressed classical RBPs (483 genes) in AML patient samples (n=195) compared to normal human CD34⁺ hematopoietic stem and progenitor cells (n=28). Blue vertical lines indicate log₂ fold-change (FC)= 0.5/−0.5 cutoff (p value > 0.05). (B) Schematic depicting pooled RBP-focused CRISPR screen. RBD, RNA-binding domains. RRM, RNA- recognition motifs; zf, zinc finger; DEAD, DEAD-box ATPase; KH, KH-homology; dsRBD, double-stranded RBD; PWI, PWI-motif; CSD, Cold-shock; La, La-motif; PUA, PseudoUridine synthase and Archaeosine transglycosylase; R3H, R-x3-H domain. (C) Scatter plot comparison of CRISPR RBP-domain screen in AML and T-ALL, T-cell acute lymphoblastic leukemia. Plotted is the log₂ fold change of sgRNA abundance (day 4/ day 20) for each cell line. Each dot represents the average of all sgRNAs targeting a RBP. Red dots indicate RBPs that are significantly overexpressed in AML and preferentially required in AML. (D) Scatter plot representation of CRISPR RBP-domain screen in AML and LUAD, lung adenocarcinoma. Red dots indicate RBPs that are significantly overexpressed in AML and preferentially required in AML. (E) Fold-change (day 4/day 20) in sgRNA abundance in pooled RBP-focused negative selection screen in MOLM-13 AML cells. Red dots indicate RBPs that are overexpressed in AML patients and exhibit greater than 3-fold depletion in CRISPR negative selection screen. Each dot represents the average of all sgRNAs targeting a RBP. (F and G) Screen validation of RBP candidates using a competition-based proliferation assay in MOLM-13 (F) and THP-1 (G) AML cell lines. Plotted are GFP percentages measured during 20 days in culture and normalized to Day 4. Negative control, sgRosa and two independent sgRNAs targeting each RBP are shown in the graphs (mean ± SD, n=3). See also Figure S1 and Table S1.

A CRISPR/Cas9 Domain-targeted Screen Reveals RBP Dependencies in AML

The majority of cancer dependencies can be predicted based on gene expression and therefore serve as clinically relevant biomarkers or therapeutic targets (Tsherniak et al., 2017). Based on our TCGA analysis, we hypothesized that several of these dysregulated RBPs may act as genetic vulnerabilities in AML and represent potential therapeutic targets. To investigate potential RBP dependencies in AML, we used the type II CRISPR (clustered, regularly interspaced, short palindromic repeat) system, which has been widely adopted in high- throughput methods to identify essential genes for cancer survival (Shalem et al., 2014; Wang et al., 2017). Given the established biochemical function of RBDs in RNA biology, we used a previously described single-guide RNA (sgRNA) domain-focused approach that enhances CRISPR/Cas9 negative selection by targeting functional protein domains (Shi et al., 2015). As a proof-of-principle, we evaluated domain essentiality by designing sgRNAs against the coding exons of three classical RBPs required to sustain cell growth based on a previous genome-wide CRISPR/Cas9 knockout screen (Hart et al., 2015). Individual sgRNAs were subcloned into a lentivirus-based GFP-tagged (LRG) sgRNA vector and cell proliferation was measured using competition assays. Notably, CRISPR scanning of the selected RBPs led to robust negative selection of sgRNAs that specifically targeted the RBDs in comparison to non-domain exon sgRNAs, which demonstrated modest depletion (Figure S1A). Based on these findings, we built a custom 2,900 sgRNA (~6–8 per gene) library against 490 RBPs comprised of well-defined RBDs based on previous studies (Gerstberger et al., 2014; Lunde et al., 2007) (Figure 1B). Pooled sgRNAs were subcloned into the LRG vector and next-generation sequencing was carried out to confirm optimal sgRNA representation (Figure S1B). Subsequently, we performed a loss-of-function pooled screen in a Cas9-expressing AML cell line, MOLM13 (MLL-AF9, FLT3^itd) to identify RBPs required for AML proliferation. Genomic DNA was harvested from cells on days 4 and day 20 post-transduction of sgRNA library and individual sgRNA read counts were evaluated by deep sequencing. Changes in sgRNA abundance were assessed by measuring the average fold change (Day 4/Day 20) of all sgRNAs targeting a given gene. This negative selection screen identified 71 RBPs that strongly dropped out with >3-fold depletion. These also included positive controls of domain-targeting sgRNAs against known chromatin regulators (BRD4, DOT1L, and KMT2D) required for AML survival and the essential gene, RPA3, thereby confirming the quality of our negative selection screen (Figure S1C). In parallel, we performed counter-screens in three other cancer cell lines, a T-cell acute lymphoblastic leukemia (CUTLL-1), a lung adenocarcinoma (A549) line and a melanoma line (501MEL), to delineate potential cancer lineage-specific RBPs. Pair-wise comparison of domain CRISPR/Cas9 screens revealed several RBP dependencies unique to each cell line (Figure 1C-D and Figure S1D). We narrowed our focus on RBPs that were > 2-fold depleted in our AML screen relative to T-ALL, melanoma, and lung adenocarcinoma (LUAD) to identify AML RBP candidates. Using this criterion, we uncovered 23 RBPs preferentially required in AML.

We next integrated the domain CRISPR screen with our transcriptome analysis in AML patients to identify RBPs that are both required for AML survival and dysregulated in expression in AML. From this analysis, we identified genes encoding 21 RBPs that were amongst the most highly depleted (>3-fold depletion) in our AML screen and significantly overexpressed in patient samples (p value < 0.05) (Figure 1E). Based on our CRISPR counter-screens, we found 8 of the 21 RBP candidates (RBM39, DHX37, PABPN1, ZFP36L2, TRA2B, SRSF10, HNRNPH1 and SUPT6H) to be selectively required and upregulated in AML (Figure 1C-D and Figure S1D). To further validate these findings, we evaluated four candidates (RBM39, DHX37, SUPT6H, HNRNPH1) by selecting the top sgRNAs for each gene and monitored their ability to inhibit AML cell growth using a competition-based assay. Our results confirmed that all sgRNAs demonstrated robust depletion in two independent AML cell lines, MOLM-13 and THP-1 (MLL-AF9, NRAS^mut) (Figure 1F-G). We also performed Gene Ontology (GO) analysis focusing on RBPs that were at least 5-fold depleted in our AML screen. Most of the genes significantly clustered into key RNA biological processes that included, mRNA export, splicing, and mRNA processing (Figure S1E). Finally, as some of the top-ranking depleted RBPs were also considered core fitness genes (Hart et al., 2015) (Figure S1F), we eliminated them, to increase the therapeutic significance of our findings.

RBM39 is Required to Sustain Leukemia Survival In Vitro and In Vivo

One of the top-scoring candidates that met the above criteria was RBM39, also known as CAPER-α, a protein previously characterized as interacting with U2AF65 and SF3B1 (Imai et al., 1993; Loerch et al., 2014; Stepanyuk et al., 2016). Our expression studies were able to show that, RBM39 was found significantly more highly expressed in AML patient samples when compared to normal hematopoietic cells (Figure 2A). RBM39 median expression in AML was highest among all other cancer subtypes in the TCGA. We did not observe any significant differences in RBM39 expression across molecular or cytogenetics AML risk groups, suggesting a potential unique requirement for RBM39 in this type of leukemia (Figure S2A-B). To further evaluate the requirement across different cancers for RBM39, we performed competition assays using RNAi or CRISPR/Cas9 to target RBM39 in various cancer cell lines. Our one-by-one validation revealed that AML cell lines, across a variety of molecular subtypes, were the most susceptible to growth inhibition upon loss of RBM39 (Figure 1F-G, and Figure S2C-E). Additionally, RBM39 suppression in AML led to marked induction of apoptosis, as determined by Annexin V staining (Figure S2F). In contrast, RBM39 was relatively more dispensable for growth in non-AML cell lines (Figure 1C-D, Figure S1D and Figure S2G). Given the evident requirement of RBM39 in AML pathogenesis, we also employed CRISPR-Cas9 “domain-scanning” to identify essential RBM39 protein domains for future drug discovery efforts. We designed sgRNAs with low off-target scores to perform a CRISPR scanning of RBM39 coding exons in two independent human AML cell lines. These include sgRNAs targeting the three RRM domains and the serine/arginine-rich (RS) region of RBM39, which are critical for pre- mRNA splicing (Prigge et al., 2009). Our results revealed strong depletion of sgRNAs that exclusively targeted RRM1 and RRM2 domains, whereas RRM3 and non-domain sgRNAs demonstrated lesser negative selection (Figure 2B and Figure S2H). Overall, these findings demonstrate that RBM39 relies on specific RNA-binding domains and supports a critical RBM39 dependency in AML.

Figure 2: RBM39 is Required to Sustain AML Growth In Vitro and In Vivo.

(A) Violin plot of RBM39 normalized expression in AML patients (red) and normal human CD34⁺ hematopoietic stem and progenitor cells (blue). Horizontal line inside the box represent the Mean, 25^th-75^th percentiles, showing all data points. Statistical analysis was performed using Wilcoxon Rank Sum test. (B) CRISPR mutagenesis of RBM39 exons in MOLM-13 AML cells using a competition-based assay. Green boxes represent annotated RNA-binding domains of RBM39. (C) Bioluminescent images of mice transplanted with MLL-AF9 Nras^G12D cells transduced with sgRosa (n=4) or two independent sgRbm39 (n=7/group). Representative images of 3 mice per sgRNA construct is shown. The same mice are depicted at each time-point. (D) Quantification of bioluminescent imaging in sgRosa negative control and two independent sgRbm39 at the indicated time points. Box-and-whisker plot, Min. to Max. show all points, 25^th-75^th percentiles, Median (horizontal line). Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (**p< 0.01, ***p< 0.001). (E) Flow cytometry analysis of GFP positive sgRNA- expressing leukemia cells in peripheral blood of MLL-AF9 Nras^G12D leukemia recipient mice at indicated time points. Box-and-whisker plot, Min. to Max. show all points, 25^th-75^th percentiles, Median (horizontal line). Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (*p< 0.05, **p< 0.01). (F) Kaplan-Meier survival curves of recipient mice transduced with sgRosa negative control and two independent Rbm39 sgRNAs are plotted. The p values were determined using Log rank Mantel-Cox test. Data with statistical significance are as indicated, *p< 0.05, **p< 0.01, ***p< 0.001. See also Figure S2.

We next examined the in vivo significance of Rbm39 in AML progression by using a previously established MLL-AF9 Nras^G12D mouse model (RN2) (Shi et al., 2015; Zuber et al., 2011a). We designed sgRNAs targeting the RRM domains of Rbm39 and confirmed knockout of Rbm39 protein in RN2 cells (Figure S2I), subsequently sgRNA-expressing RN2 cells were injected intravenously into sub-lethally irradiated recipient mice. Strikingly, we observed remarkable delay in leukemia progression in mice receiving Rbm39-deficient RN2 cells, as determined by bioluminescent imaging (Figure 2C-D). Moreover, FACS analysis of peripheral blood in late disease-onset mice detected low percentage of circulating leukemia cells harboring Rbm39 sgRNAs (Figure 2E and Figure S2J). These findings correlated with prolonged absolute survival (Figure 2F). Indel analysis by next-generation sequencing verified that sgRNA-positive cells mostly lacked CRISPR/Cas9 editing of Rbm39, suggesting that the mice succumbed to an outgrowth of unmodified alleles and substitution mutations (Figure S2K-L). Taken together, these observations support the notion that Rbm39 is required for AML growth in vivo.

Mapping of the RBM39 Proteome Identifies an Essential AML Splicing Network

To identify RBM39 interaction partners in AML cells, we performed immunoprecipitation followed by mass spectrometry (IP-MS) in the AML cell line MOLM-13 (Figure 3A and Figure S3A). We identified a total of 54 RBM39-interacting proteins that were at least 10-fold more enriched over IgG (Figure 3B and Table S2). Proteomic network analysis identified numerous proteins associated with the spliceosome complex and ribosome biogenesis (Figure 3C). This was in agreement with GO analysis, which identified proteins highly enriched in fundamental processes involved in RNA metabolism and ribosomal RNA processing (Figure S3B). In addition, we also found several RBM39 binding partners involved in chromatin and transcriptional regulation (e.g. histone chaperone, SUPT16, and the SWI/SNF-related subunit, SMARCA5), thereby suggesting possible roles for RBM39 in gene transcription. Moreover, our proteomics approach identified 31 RBM39-interacting RBPs that also appeared in our domain CRISPR screens. We found 15 of the 31 RBPs to exhibit strong essentiality (> 2-fold depletion) in our genetic screen in AML (Figure 3D). Among these RBM39 interacting partners, the splicing factors, SRSF10 and HNRNPH1 were identified in our CRISPR screen to be preferentially required in AML and are also highly expressed in AML patient samples (Figure 3D). Altogether, these data place RBM39 in an extended RBP network that is essential for leukemia maintenance and suggest that modulation of AML splicing can be therapeutically exploited by targeting the RBM39 interactome.

Figure 3: Mass Spectrometry of RBM39 Proteome Identifies an RBP Network Required in AML.

(A) Silver staining of endogenous RBM39 immunoprecipitation (IP) in MOLM-13 cells. Arrowhead represents predicted size of RBM39, asterisks denotes non-specific IgG heavy/light chain signals, and M= protein marker. (B) Overlay of peptide-spectrum match (PSM) counts between IgG control and RBM39 IP (left panel). Top RBM39-interacting partners enriched over IgG (right panel). (C) STRING network analysis of RBM39 protein interactome in AML. Black circle, RBM39; red circles, spliceosome complex; blue circles, ribosome, and green circles, ribosome biogenesis. (D) Plot represents data from MOLM-13 AML screen as shown in Figure 1E. Green denotes RBP genes identified in RBM39 IP-MS and overexpressed in TCGA AML patients. Dotted line represents 2-fold depletion. See also Figure S3 and Table S2.

RBM39 Loss Alters Splicing of mRNAs Essential for AML

Given the physical interaction of RBM39 with key core splicing factors and prior data identifying splicing changes upon RBM39 loss (Han et al., 2017; Uehara et al., 2017), we hypothesized that loss of RBM39 might disrupt splicing and expression of mRNAs preferentially required for leukemia growth. To this end, we initially evaluated changes in splicing by RNA-sequencing (RNA-Seq) of human AML cells upon RBM39 sgRNA deletion. We measured “percentage spliced in” (ΔPSI) values across five main types of AS events (Cassette exon (CE), alternative 5’ ss exon (A5E), alternative 3’ ss exon (A3E), mutually exclusive exons (MXE) and retained introns (RI)) in RBM39 knockout versus control MOLM-13 and THP-1 AML cells. This revealed that the predominant change in RNA splicing upon RBM39 depletion was a change in cassette exon event splicing (68–74% of differentially spliced events) across both AML cell lines with an FDR < 0.1 and ΔPSI > 10% (Figure 4A and Tables S3–4). Using the same parameters, we also found alterations in MXE (~6–9%), A5E (5–6%), A3E (8–10%), and RI (5–6%) events, however to a much lesser extent compared to CE (Figure 4A). Within cassette exons, we found approximate equal proportions of exon inclusion and exclusion with RBM39 genetic loss, which was also observed in other AS groups (Figure 4B).

Figure 4: RBM39 Loss Alters Splicing of mRNAs Essential for AML Cell Growth.

(A) Number of differentially spliced events in MOLM-13 and THP-1 cells treated with RBM39 small- guide RNA (sgRNA) versus control sgRNA. (B) Scatter plot of cassette exons promoted (red circles) and repressed (blue circles) in MOLM13 treated with RBM39 sgRNA versus control sgRNA. Splicing is quantified using a ‘percent spliced in’ value (PSI, or ψ value). Promoted and repressed cassette exons are defined as those whose inclusion levels are increased or decreased by Ψ ≥10%, respectively. Grey circles represent exons for where ΔΨ is <10%. (C) eCLIP analysis of RBM39 binding sites in MOLM-13 cells. Input-normalized peak signals are shown as log₂ fold change. Purple points indicate eCLIP enriched RBM39 peaks (FDR >5%, logFC>1) in biological replicates. (D) Genomic distribution of RBM39 eCLIP-seq peaks. (E) Metagene profile of RBM39-binding sites on exons throughout the transcriptome in MOLM-13 cells relative to 5’ and 3’ splice sites. Grey box represents exonic region. (F) Enriched motifs for RBM39 binding. The top ten motifs are shown. Inset, consensus sequence, deduced from the top 10 motifs. (G) Gene Ontology (GO) enrichment analysis of terms enriched in differentially spliced mRNAs in MOLM13 cells with RBM39 sgRNA versus control. Terms in red font are related to RNA processing and/or splicing. (H) Enrichment plots from Gene Set Enrichment Analysis (GSEA) of HOXA9 target genes that are differentially spliced upon RBM39 knockout versus control sgRNA (normalized enrichment score (NES) is inferred from permutations of the gene set and the false discovery rate (FDR)). (I) RNA-seq coverage plot of BMI1 and Sashimi plots of GATA2 in MOLM-13 cells treated with RBM39 sgRNA or control overlaid with anti-RBM39 eCLIP-seq tracks. Yellow highlighted reads in GATA2 Sashimi plots highlight exon skipping events with RBM39 sgRNA. See also Figure S4 and Tables S3–S5.

The direct impact of RBM39-RNA interactions on pre-mRNA splicing has not been explored in cancer nor is the precise role of RBM39 in splicing well defined. We thus performed anti-RBM39 enhanced UV cross-linking immunoprecipitation (eCLIP) using the MOLM-13 AML cell line to identify genome-wide RNA targets of RBM39 (Van Nostrand et al., 2016) (Figure S4A). eCLIP analysis identified 9,560 significant sequence clusters using a false discovery rate (FDR) < 0.05 and log (FC) > 1, which corresponded to 4,775 annotated transcripts (Figure 4C and Table S5). GO analysis of the top 200 enriched binding sites showed processes involved in RNA metabolism, cell cycle, and transcriptional regulation (Figure S4B). Approximately 79% of RBM39 binding sites mapped to exons (Figure 4D and Figure S4C), with a preferential occupancy of proximal exonic sequences near 5’ and 3’ splice sites throughout the transcriptome (Figure 4E and Figure S4D). In each instance, we identified that at least 50% of splicing events altered by RBM39 loss in MOLM-13 were also bound by RBM39 suggesting a direct role of RBM39 in splicing regulation of its mRNA targets (Figure S4E). Consistent with prior data suggesting an interaction of RBM39 with U2AF2 (Imai et al., 1993; Loerch et al., 2014; Stepanyuk et al., 2016), motif analysis showed highly pyrimidine-enriched sequences on mapped RBM39-binding sites (Figure 4F). In fact, ten of the top hexamer sequences enriched in RBM39 binding sites were pyrimidine rich, similar to results seen with U2AF2-RNA interactions previously (Shao et al., 2014). GO analysis of differentially spliced exons revealed that RBM39 loss was strongly enriched in processes related to RNA splicing, export, and metabolism as well as DNA replication and mitosis (Figure 4G). Moreover, gene set enrichment analysis (GSEA) of differential spliced mRNAs with RBM39 sgRNA treatment identified prominent downregulation of HOXA9 targets, known to be required for leukemogenesis of these MLL-rearranged AML cell lines (Figure 4H)(Faber et al., 2009). These prominently include aberrant splicing due to intron retention across several introns of BMI-1, a gene known to be required for leukemogenesis that was also identified as a direct RBM39 mRNA binding target by our eCLIP-Seq. Additionally, GATA2, an essential hematopoietic factor involved in maintaining AML transcriptional homeostasis was aberrantly spliced upon RBM39 loss (Katsumura et al., 2016; Yang et al., 2017)(Figure 4I). The unannotated splicing changes observed in BMI1 and GATA2 upon RBM39 depletion are predicted to result in mRNAs which would be subjected to nonsense-mediated decay (NMD). These studies suggest that the dependency of RBM39 is, at least in part, due to defective splicing of genes involved in pathways required for AML cell survival.

Potent Anti-Leukemic Effects of Pharmacologic RBM39 Degradation

An emerging field in drug discovery has been the characterization of small molecules that modulate E3 ligases to promote targeted protein degradation (Kronke et al., 2015; Lu et al., 2014). Interestingly, a class of compounds known as anticancer sulfonamides (including the drugs indisulam (also known as E7070), E7820, and chloroquinoxaline sulfonamide) were recently identified to selectively degrade RBM39 through an interaction with DCAF15, an adapter protein for the CUL4/Ddb1 E3 ubiquitin ligase (Han et al., 2017; Uehara et al., 2017). These compounds demonstrated an excellent safety profile in clinical trials and have previously been shown to exhibit some antitumor efficacy; however, overall response rates were low (Owa et al., 1999; Supuran, 2003), most likely because neither the mechanism of action nor potential biomarkers of response were known. To this end, we observed that DCAF15 was more highly expressed in AML patient samples than normal hematopoietic progenitors (Figure 5A), and that increasing indisulam concentrations led to a dose-dependent reduction in RBM39 and HOXA9 target genes (BMI-1 and MYB) protein levels in human leukemia cells (Figure 5B and Figure S5A-B). Indisulam exposure led to severely impaired G2/M cell cycle arrest and increased apoptosis after 48 hours of treatment in AML cells in vitro (Figure 5C-D, and Figure S5C). We additionally confirmed that the cellular effects of indisulam are dependent on DCAF15 expression by designing sgRNAs to knockout DCAF15 in AML-sensitive cells. Suppression of DCAF15 had little impact on AML cell viability at baseline. However, DCAF15 loss conferred robust resistance to treatment with indisulam, as demonstrated by the positive selection of AML cells expressing DCAF15 sgRNAs (Figure S5D). Moreover, we verified RBM39 as a direct target of indisulam by over-expressing a previously described RBM39 G268V mutation shown to confer resistance to indisulam (Figure S5E).

Figure 5: Pharmacological Inhibition of RBM39 Displays Broad Sensitivity Across Diverse AMLs.

(A) Violin plot of DCAF15 normalized counts from RNA-seq in AML patients (red) and normal human CD34⁺ hematopoietic stem and progenitor cells (blue). Horizontal line inside the box represent the Mean, 25^th-75^th percentiles, showing all data points. Statistical analysis was performed using Wilcoxon Rank Sum test. (B) Western blot of K562 whole cell lysates treated with escalated doses of indisulam after 48 hr. (C) Representative flow cytometry plots of Annexin V staining performed on MOLM-13 AML cells treated with 500 nM of indisulam or DMSO for 48 hr. (D) Representative flow cytometry plots showing EdU cell cycle analysis of MOLM-13 AML cells exposed to 500 nM indisulam or DMSO for 48 hr. (E) Bioluminescent imaging of mice transplanted with MOLM-13 luciferase cells and treated with vehicle or 25 mg/kg indisulam. Representative images of 5 mice/group are shown. The same mice are depicted at each time-point. (F) Quantification of bioluminescent signals in vehicle versus indisulam-treated leukemic mice at indicated time-points (n=6/group). Unpaired Student’s t test using Prism 7 (Graphpad). Box-and-whisker plot, Min. to Max. show all points, 25^th-75^th percentiles, Median (horizontal line). Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (**p< 0.01). (G) Flow cytometry analysis of human MOLM-13 cells in peripheral blood in DMSO versus indisulam groups. Gating was performed on human CD45⁺ in DMSO (n=4–5) and Indisulam (n=5) treated mice. Box-and-whisker plot, 25^th-75^th percentiles, Min. to Max. show all points, Median (horizontal line). Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (*p< 0.05, **p< 0.01). (H) Kaplan-Meier overall survival of indisulam-treated AML mice. Grey represents the exposure time (days) to indisulam. The p values were calculated using Log rank Mantel-Cox test. (I) Schematic of patient-derived xenograft (PDX) generation and treatment with indisulam (25 mg/kg/day). Patient 1 PDX 1 received 3 days of drug treatment. (J) Percentage of human CD45⁺ (hCD45) cells amongst total live bone marrow mononuclear cells by flow cytometry analysis of bone marrow aspirate pre- and post-treatment with indisulam in five primary AML PDX models from 3 patients. Red indicates SF3B1 mutant patients. (K) Representative flow cytometry plots of CD45 in Patient 2 PDX and Patient 3 PDX 2 bone marrow aspirates pre- and post-indisulam treatment. Percentage of hCD45⁺ cells is indicated. Data with statistical significance are as indicated, *p< 0.05, **p< 0.01, ***p< 0.001. See also Figures S5 and S6.

We next assessed the in vivo efficacy of indisulam in AML progression by transplanting AML cells expressing firefly luciferase into immune-deficient recipient mice. To resemble disease burden in patients, we waited until onset of disease, and subsequently mice were treated for 13 consecutive daily treatments of either vehicle or 25 mg/kg indisulam, a previously described dose (Han et al., 2017). In concert with in vitro findings, administration of indisulam led to strong anti-leukemic effects in vivo using two AML cell line xenograft models, MOLM-13 and OCI- AML3. Bioluminescent quantification and imaging of animals treated with indisulam showed marked delay in AML burden (Figure 5E-F and Figure S5F). Compared with vehicle control, FACS analysis of peripheral blood and bone marrow revealed a substantial reduction in MOLM- 13 cells and elevated apoptosis in animals treated with indisulam (Figure 5G and Figure S5G). Immunohistochemistry analysis of RBM39 expression in bone marrow and spleen further confirmed in vivo degradation of RBM39 in indisulam-treated mice (Figure S5H). Overall, indisulam administration significantly extended survival in both AML xenograft models (Figure 5H and Figure S5I) with a median survival time of 26 days in MOLM-13 indisulam-treated animals versus 15 days in vehicle control cohorts (Figure 5H). In addition, indisulam treatment to five patient-derived xenografts (PDX) generated from three distinct patients demonstrated reduced leukemia burden with indisulam treatment in each case (Figure 5I-K and Figure S5J). In contrast to the robust anti-leukemic effects of RBM39 degradation in vitro and in vivo, four weeks of indisulam treatment in normal C57/B6 mice resulted in no hematologic toxicities while causing on-target degradation of RBM39 (Figure S6A-H). Similarly, we have noticed no significant effects of pharmacologic degradation of RBM39 on human normal hematopoietic cells in vivo using immunocompromised mice transplanted with cord blood CD34+ cells (Figure S6I-Q). Overall, these findings demonstrate potent single agent leukemic effects of indisulam in vivo with preferential effects of RBM39 loss on AML over normal human hematopoietic cells.

Preferential Sensitivity of Spliceosomal Mutant AML to RBM39 Degradation

Recent genomic studies of AML and related myeloid malignancies have identified recurrent change-of-function mutations in RNA splicing factors highlighting a potential role for altered splicing in the pathogenesis of clonal hematopoietic malignancies. These mutations are conspicuously present in splicing factors which physically interact with RBM39 (including SF3B1 and U2AF1) and others (SRSF2) as a series of mutually exclusive, heterozygous point mutations at highly specific residues. Prior work has demonstrated that cancer cells bearing these mutations are genetically dependent on otherwise wild-type splicing catalysis for cell survival (Fei et al., 2016; Lee et al., 2016; Zhou et al., 2015). Similarly, cells bearing these mutations are preferentially sensitive to drugs that modulate core splicing catalysis by binding to SF3B1 (Lee and Abdel-Wahab, 2016; Obeng et al., 2016; Seiler et al., 2018; Shirai et al., 2017). Preferential sensitivity of splicing mutant hematopoietic cells over splicing wild-type counterparts has previously been demonstrated for the drugs E7107, sudemycin, and H3B-8800. And, as mentioned earlier, this has led to an ongoing phase I trial to identify the safety and maximum tolerated dose of H3B-8800 in patients with refractory myeloid leukemia (Lee et al., 2016; Seiler et al., 2018; Shirai et al., 2017). Given our data that RBM39 loss resulted in greatly altered splicing of factors required for RNA splicing, export, and catabolism, we hypothesized that spliceosomal mutant AML might be preferentially sensitive to sulfonamide-induced RBM39 degradation. To this end, we further investigated the in vitro selectivity of sulfonamides across a panel of genetically diverse AML cell lines. Measurement of cell viability upon sulfonamide exposure revealed broad anti-leukemic effects with potent inhibitory activity across many AML subtypes, with most cell lines exhibiting sub-micromolar sensitivity (Figure 6A). We found that leukemia cells bearing mutations in leukemia-associated mutations in splicing factors were amongst the most sensitive cells to sulfonamides. In addition, a number of AML cell lines without spliceosomal gene mutations also exhibit sensitivity to the sulfonamides. Evaluation of relative DCAF15 mRNA expression revealed that the highest and lowest relative levels of DCAF15 mRNA correlated with greatest and worst response to E7820 respectively (Figure 6A). The preferential effects of sulfonamides on leukemia cells bearing spliceosomal gene mutations was further confirmed in a series of isogenic AML lines (K562 and TF-1) engineered to express hotspot mutations in SF3B1, SRSF2, and U2AF1 from their endogenous loci (Figure 6A-B) and B-cell acute lymphoblastic leukemia (NALM-6 cells; Figure 6C) with exposure to E7820 and indisulam. In each instance, spliceosomal mutant cells were preferentially sensitive to growth inhibition to sulfonamides over spliceosomal wild-type cells. In isogenic cells, E7820 exposure led to similar dose-dependent degradation of RBM39 in leukemia cell lines (Figure 6D). Furthermore, shRNA knockdown of RBM39 in K562 cells knockin of SF3B1^K700E resulted in greater competitive disadvantage when compared to K562 parental cells, supporting the concept that spliceosomal mutant cells are preferentially sensitive to RBM39 loss (Figure 6E).

Figure 6: Preferential Effects of Sulfonamides on Spliceosomal Mutant AML.

(A) Log₁₀ of IC₅₀ values of a panel of human AML cell lines to E7820 annotated for spliceosomal gene mutation status (red versus black bars indicate spliceosomal mutant or wild-type cells respectively) as determined by the CellTiterGlo assay. IC₅₀ values are calculated from at least 4 technical replicates per experiment, each experiment was performed at least twice, and error bars represent standard deviation. Also shown are log₁₀ relative DCAF15 mRNA expression levels for each cell line. (B) Bar plots of IC₅₀ values of isogenic K562 cells with or without knockin of spliceosomal gene mutations to E7820 (left) or indisulam (right). IC₅₀ values calculated from four technical replicates per experiment; one representative experiment of three biological replicates is shown; error bars represent standard deviation. Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (*p< 0.05). (C) Bar plots of IC₅₀ values of isogenic NALM-6 cells with or without knockin of spliceosomal gene mutations to E7820 (left) or indisulam (right). IC₅₀ values calculated from one representative experiment of three biological replicates is shown; error bars represent standard deviation). Statistical analysis was performed using unpaired Student’s t test by Prism Graphpad (*p< 0.05). (D) Anti-RBM39 Western blot of isogenic K562 cells treated with DMSO or increasing doses of E7820 for 24 hr. Densitometry quantification of Western blot is shown on the right. (E) Competition assay of RBM39 shRNA knockdown in K562 parental vs. K562 SF3B1^K700E. Analysis was determined by unpaired Student’s t test. Data with statistical significance of mutant cells relative to control are as indicated, (**p< 0.01, ***p< 0.001).

RBM39 Degradation Targets an RBP Network Required for AML Survival

To understand the basis for the preferential effects of sulfonamides on spliceosomal mutant cells, we next evaluated RBM39 protein levels across a panel of isogenic AML cells with or without knockin of spliceosomal gene mutations. Degradation of RBM39 occurred in a comparable dose-dependent fashion across cell lines regardless of gene mutation status (Figure 6D), suggesting that a greater dependency on residual wild-type splicing function may be the basis for increased sensitivity of spliceosomal mutant cells to sulfonamides. Given that spliceosomal mutant cells are preferentially sensitive to alterations in splicing over their wildtype counterparts, we next performed RNA-sequencing of parental K562 and isogenic lines expressing SF3B1^K700E and SRSF2^P95H mutations treated with 1 mM of E7820 (which represents the IC₅₀ of parental K562 cells to E7820; Figure 7A). In parallel, we also carried out RNA-seq of the same cell lines treated with E7107, a small molecule that inhibits splicing through impeding binding of SF3B1 to the branch point (Cretu et al., 2018; Finci et al., 2018). Treatment with either E7820 or E7107 at the IC₅₀ of each drug in parental K562 cells resulted in increased cassette exon skipping and intron retention relative to DMSO treatment regardless of spliceosomal gene mutation status (Figure 7B-C). Interestingly, however, at equipotent nontoxic doses, E7820 resulted in greater changes in splicing within each cell type and across each category of splicing versus E7107 at this dose (Figure 7C-D and Figure S7A-D). Moreover, a greater number of differential splicing events were identified within each type of splicing in SF3B1^K700E cells treated with E7820 versus SF3B1 wild-type counterparts. These data suggest that at least one reason for the preferential effects of sulfonamides on SF3B1-mutant over wildtype cells is a heightened splicing response of SF3B1-mutant cells to RBM39 degradation.

Figure 7: RBM39 Degradation Targets an RBP Network Required for AML Survival.

(A) Schema of drug testing (6 hours) across isogenic K562 cells with or without heterozygous knockin of the SF3B1^K700E or SRSF2^P95H mutation at the endogenous locus for RNA-seq. Each experiment was perform in duplicates. (B) Violin plot of inclusion level differences of introns or cassette exons in SF3B1^K700E/WT cells treated with DMSO versus SF3B1^WT cells treated with DMSO (“DMSO” column) or SF3B1^K700E/WT cells treated with E7820 or E7107 versus the same cells treated with DMSO. Horizontal line inside the box represent the Mean, 25^th–75^th percentiles, points are “outliers” (out of the 99^th percentile line). Statistical analysis was performed using Wilcoxon Rank Sum test. (C) Bar plot of number of differential splicing events across parental, SF3B1^K700E/WT, and SRSF2^P95H/WT K562 cells. The numbers above each bar indicate number of differentially spliced events. (D) RNA-seq coverage plots of SUPT6H, HNRNPH1, SRSF10, and U2AF2 in the isogenic K562 cells from (A). See also Figures S7.

We also noticed that a number of differentially spliced events upon sulfonamide exposure involved mRNAs encoding RBPs identified in the CRISPR screen as upregulated and required for AML cell survival (Figure 1E). These included SUPT6H, hnRNPH, and SRSF10 (Figure 7D), where E7820 exposure resulted in intron retention that was most pronounced in spliceosomal mutant cells. In addition and in agreement with our previous findings, RBM39 degradation also resulted in enhanced aberrant splicing of the HOXA9 target genes BMI-1 and MYB and a number RBPs in spliceosomal mutant AML over WT counterparts, including aberrant splicing events in U2AF2 and RBM3 (Figure 7D and S7A-E). Although there was no mis-splicing of HOXA9 upon indisulam treatment (Figure S7F-H), we observed aberrant splicing and reduced expression of MYB, GATA2, and BMI1 (Figure S7I). These data are consistent with the prediction that mis-splicing of MYB, GATA2, and BMI1 upon RBM39 loss would result in NMD and indicate that RBM39 loss resulted in anti-leukemia effects by altering splicing and expression of HOXA9 target genes without impacting HOXA9 splicing or expression itself. Moreover, GSEA of differentially spliced events in response to E7820 also revealed downregulation of targets of MYC and PI3K-AKT-mTOR signaling as well as mRNAs involved in response to inflammation (Figure S7J-K) all known to be important in AML pathogenesis or progression. Overall, these data suggest that the presence of spliceosomal gene mutations as well as DCAF15 expression may serve as important predictors of response to RBM39 degradation in AML.

DISCUSSION

AML continues to have a dismal survival rate. This can largely be attributed to limited advances in treatment regimens that, for the last decades, have relied on the use of non-targeted cytotoxic drugs. However, recent FDA approvals of the first small molecule inhibitors targeting recurrent genetic lesions in AML (midostaurin, enasidenib) provides hope for molecularly targeted therapies in this challenging illness. Despite the success in the development of these small molecule inhibitors, the genetic heterogeneity of AML presents a current therapeutic barrier, in which available selective inhibitors target only a genetically defined subset of patients. However, it appears that AML cells rely on specific pathways that have not been subjected to genomic alterations for their survival. This has led to the discovery of broader targeted therapies that are currently in clinical trials for hematological malignancies (Kotschy et al., 2016; Zuber et al., 2011b).

From our focused CRISPR/Cas9 screen, we identified several dysregulated RBPs that are essential for AML maintenance. Many of these proteins have not been described previously as leukemia dependencies and warrant future research to elucidate their molecular mechanism in leukemia. Also, the direct comparison to identical CRISPR/Cas9 screens in T-ALL, melanoma and lung adenocarcinoma, gives us the ability to identify RBPs that play specific functions in AML cells. Here, we have focused our efforts on understanding the dependency of RBM39 in AML. Mechanistically, we showed that CRISPR-mediated deletion or pharmacological degradation of RBM39 causes altered splicing of HOXA9 target genes, an essential transcriptional network required in AML. This in turn consequently leads to a preferential targeting of AML cells compared to other cancers. Given the mechanistic role of RBM39 in splicing shown here and the requirement of the DCAF15 adapter protein for anti-cancer sulfonamide activity, we identified that both the presence of spliceosomal gene mutations and levels of DCAF15 expression are important predictors of response to sulfonamides. Currently, a regiment that includes indisulam in combination with standard chemotherapy is being investigated in phase II trials in patients with refractory or relapsed myeloid malignancies (Assi et al., 2018). While indisulam is given to patients by intravenous infusion, E7820 is an orally bioavailable drug that has been tested as an improved, highly-on target, second-generation anticancer sulfonamide in phase II clinical trials for solid tumors such as colorectal cancer (Milojkovic Kerklaan et al., 2016). The present study supports further clinical investigation of E7820 in patients with myeloid malignancies. These data also provide mechanistic support for expanded use of sulfonamides in clinical trials, as it identifies RBM39 as a key non-oncogenic addiction in AML, describes its mechanism of action, and offers valuable potential biomarkers and genetic predictors of response. The known safety of anticancer sulfonamides established through multiple phase I and phase II clinical trials to date provides an additional advantage over drugs targeting splicing by modulating SF3b function whose safety is not yet known. Indeed, we are currently planning a multi-center clinical trial using E7820, targeting AML patients that either carry spliceosomal mutations and/or express high levels of DCAF15, as biomarkers of response.

STAR METHODS

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell Lines and Cell Culture

All human and mouse leukemia cells were cultured in recommended media, typically RPMI medium with 10% FBS and 1% penicillin/streptomycin. The NALM-6 isogenic cell lines (NALM-6 cells engineered to express the single mutations SF3B1^K700E, SF3B1^K666N, SF3B1^H662Q from the endogenous SF3B1 locus) were cultured in RPMI/10% FBS and 1% penicillin/streptomycin. K562 isogenic cell lines (engineered to express SF3B1^K700E, SF3B1^K666N, SRSF2^P95H, or U2AF1^S34F mutations from each respective endogenous locus) were cultured in IMDM/10% FBS. TF-1 isogenic cell lines (engineered to express SF3B1^wild-type or SF3B1^K700E from the endogenous SF3B1 locus by Horizon Discovery Inc. as described previously) (Seiler et al., 2018) were cultured in RPMI/10% FBS and 1% penicillin/streptomycin with 5ng/mL human GM-CSF. Genotyping and variant allele frequency of spliceosomal gene mutations in SF3B1, SRSF2, U2AF1, and ZRSR2 in each cell line was performed using the MSKCC IMPACT assay (Cheng et al., 2015; Zehir et al., 2017). The adherent cell lines, A549, 501MEL and HEK293T cells were grown in DMEM medium with 10% FBS and 1% penicillin streptomycin. SK-MEL239 were cultured in RPMI, 10% FBS and 1% penicillin streptomycin. Cell lines transduced with retroviral Cas9 puromycin (Addgene plasmid no. 65655) were selected with puromycin (Sigma Aldrich) 48 hours after transduction. All transfections were performed in HEK293T cells using Polyethylenimine (PEI) reagent at 4:2:3 ratios of sgRNA construct: pVSVG: pPax2 in OPTI-MEM solution. Viral supernatant was collected 48 hrs and 72 hrs post-transfection. Spin infections were performed at room temperature at 1,800 RPM for 30 mins with polybrene reagent (1:2000 dilution) (Fisher Scientific).

Primary human samples

Studies were approved by the Institutional Review Boards of Memorial Sloan Kettering Cancer Center and conducted in accordance to the Declaration of Helsinki protocol. Primary human de- identified AML samples derived from whole peripheral blood or BM mononuclear cells were utilized. Mutational genotyping of each sample was performed by the MSKCC IMPACT assay as described previously (Cheng et al., 2015; Zehir et al., 2017). Cord blood was acquired from NY Blood Bank. Informed consent was obtained from all subjects to obtain the patient specimens used in the studies described. Specimens were obtained as part of the Memorial Sloan-Kettering Cancer Center Institutional Review Boar approved clinical protocol #06–107 to which all subjects consented. O.A-W is a participating investigator on this protocol.

Animals

8–10 weeks-old C57BL/6 male mice were purchased from Jackson Laboratory. 10 week-old NOD scid gamma as well as NSG-SGM3 (NSGS) female mice were obtained from Jackson Laboratory. Mice were bred and maintained in individual ventilated cages and fed with autoclaved food and water at NYU School of Medicine Animal Facility as well as Memorial Sloan Kettering Cancer Center. All animal experiments were done in accordance with approved protocols from the Institutional Animal Care and Use Committees, according to national and institutional guidelines. Kaplan-Meier survival curves were compared using the Wilcoxon Rank- Sum test via GraphPad Prism. All animal experiments were performed in accordance with protocols approved by the New York University Institutional Animal Care and Use Committee (IACUC).

METHOD DETAILS

sgRNA Library Design

All sgRNAs in this study were designed using http://benchling.com/. Most sgRNAs had an off-target score of 70 or higher. RBP domain sgRNA library was designed to target 490 well-annotated RBDs based on NCBI database annotation. For each RBP gene, we designed on average 6–8 sgRNA for a total of ~2,900 sgRNAs. Library also consisted of positive sgRNAs targeting RPA3, DOT1L, BRD4, and KMT2D and the negative controls, sgRosa. Individual sgRNA oligos was synthesized by Twist Bioscience (https://twistbioscience.com/) on a 12K array and amplified using array primers (Table S1). Using a Gibson Assembly master mix (New England Biolabs), we cloned sgRNAs into a lentiviral sgRNA GFP-tagged vector (LRG) (Addgene plasmid no. 65656). Gibson reactions were transformed using DH10B electrocompetent cells (Invitrogen) at 2 kV, 200 Ω, and 25 uF. Bacterial colonies were quantified to obtain ~70X coverage. Subsequently, library was deep-sequenced using Miseq to confirm sgRNA representation. All sgRNA sequences used in this study are provided in Table S1.

CRISPR Screen

Cas9-expressing cell lines were transduced with sgRNA library virus at a low MOI (~0.3). On Day 4 post-transduction, GFP percentage was assessed to determine infection efficiency and sgRNA coverage (~300–500X). Remaining 300–500X cells were placed back into culture after each passage until 20 days’ post-transduction. Genomic DNA (gDNA) of cells containing 300– 500X coverage were harvested on Day 4 and Day 20 using Qiagen DNA kit based on manufacturer’s protocol. For library construction, 200 ng of gDNA was amplified for 20 cycles using Phusion Master Mix. End repair products were generated using T4 DNA polymerase (NEB), DNA polymerase I (NEB) and T4 polynucleotide kinase (NEB). Subsequently, A-tail overhangs were added using Klenow DNA Pol Exo- (NEB). DNA fragments were then ligated with Illumina-compatible barcodes (Bioo Scientific) using T4 Quick Ligase (NEB) and amplified using pre-capture primers (5 cycles). Barcoded libraries were then sequenced using Mi-Seq instrument (150 cycles). For pooled CRISPR screen analysis, individual time-points for all samples were normalized using the formula (sgRNA read count/total read count) × 100,000. Subsequently, normalized reads were then used to calculate log₂ fold change (normalized read count Day 4/normalized read count Day 20).

CRISPR/Cas9 indel analysis

To quantify the spectrum of indel mutations in mouse Rbm39 sgRNAs, we transduced RN2 cells with sgRbm39s, followed by cell sorting of GFP+/sgRNA+ populations at day 3 and day 20 post-infection. Cells were then harvested for gDNA and PCR amplicon (ranging from 100–200 bps) were designed to flank the sgRNA recognition sequence. 200 ng of gDNA was amplified using 2× Phusion Master Mix, followed by end repair using T4 DNA polymerase (NEB), DNA polymerase I (NEB) and T4 polynucleotide kinase (NEB). Subsequently, A overhangs were added following end repair using Klenow DNA Pol Exo- (NEB). Adaptor barcodes (Bioo Scientific) were then ligated and products were purified by AMPure beads. Finally, precapture PCR cycles were performed and libraries were sequenced using Mi-Seq instrument (150 cycles). Indel analysis was performed using CRISPResso (http://crispresso.pinellolab.partners.org/).

Apoptosis and Cell cycle

To assess cell cycle, we used the 5-ethynyl-2’-deoxyuridine (EdU) incorporation using the Click-iT Plus EdU Alexa Fluor 647 kit and performed experiments as described in manufacturer’s protocol (Life Technologies). Apoptotic analysis was determined using APC Annexin V (BD bioscience) and performed according to manufacturer’s specifications and co-stained with 4’,6-Diamidino-2-Phenylindole, Dihydrochloride (DAPI) for DNA content. Both EdU and Annexin V stained cells were analyzed by flow cytometry and FlowJo software.

RNA-sequencing library preparation and sequencing

CRISPR-mediated RBM39 knockout cells were harvested day 6 post-transduction and wash with 1× PBS. To extract total RNA, we used RNeasy Plus Mini Kit (Qiagen) and QIAshredder (Qiagen), which were performed according to manufacturer’s protocol. We isolated poly(A) mRNA using magnetic isolation method (NEB) using ~1 ug of total mRNA. To generate RNA-sequencing libraries, we used NEXTflex Rapid Directional (Bioo Scientific) protocol and NEXTflex RNA-seq barcodes (Bioo Scientific), which were carried out according to manufacturer’s protocol. RNA-seq barcoded libraries were then sequenced using Hi-Seq 4000 (150 cycles paired end).

For analysis of sulfonamide vs. E7107 treated K562 cells, RNA sequencing was performed in biological duplicate for each cell line and drug treatment. RNA was extracted using Qiagen RNeasy columns. poly(A)-selected, unstranded Illumina libraries were prepared with a modified TruSeq protocol. 0.5X AMPure XP beads were added to the sample library to select for fragments of 400 bp, followed by 1X beads to select for fragments 4100 bp. These fragments were then amplified with PCR (15 cycles) and separated by gel electrophoresis (2% agarose). 300 bp DNA fragments were isolated and 101 bp reads were sequenced on the Illumina HiSeq 2000 in paired end mode. For analysis of sulfonamide vs. E7107 treated K562 cells, 100 million paired end reads of 100 bp were performed for each sample.

eCLIP library preparation

eCLIP studies were performed in duplicates by Eclipse Bioinnovations Inc (San Diego, www.eclipsebio.com) according to the published single-end seCLIP protocol [[can cite pubmed ID 28766298]] with the following modifications. 10 million Molm13 cells were UV crosslinked at 400 mJoules/cm2 with 254 nm radiation, and snap frozen. Cells were then lysed and treated with RNase I to fragment RNA as previously described. RBM39 antibody (A300–291A lot 001, Bethyl) was then pre-coupled to Protein G Dynabeads (Thermo Fisher), added to lysate, and incubated overnight at 4 deg C. Prior to immunoprecipitation, 2% of the sample was taken as the paired input sample, with the remainder magnetically separated and washed with lysis buffer only (as the standard high-salt eCLIP wash buffer gave poor immunoprecipitation yield). eCLIP was performed by excising the area from ~65 kDa to ~140 kDa. RNA adapter ligation, IP-western, reverse transcription, DNA adapter ligation, and PCR amplification were performed as previously described.

Western blotting

Cell lines were treated with the indicated dose of drug for 24 hours. Lysate protein concentration was measured with the BCA reagent and 10 mcg was loaded per lane onto 4–12% bis-tris protein gels. After transfer, PVDF membranes were probed with anti-RBM39 rabbit polyclonal (Atlas Laboratories) at 1:200, anti-RBM39 (Bethyl Laboratories) at 1:1,000, BMI-1 (Abcam), and MYB (Thermo Fisher Scientific) and visualized by standard methods. Western blot densitometry was performed using ImageJ.

Sulfonamide drug treatment IC₅₀ measurements

E7820 (molecular weight: 336.37 grams/mole) was provided by Eisai Pharmaceuticals. For in vitro experiments, E7820 was dissolved in DMSO to make a 100 micromolar stock solution, and this was then added to tissue culture media to the appropriate final concentration. Cell lines were plated in 96 well plates and exposed to the indicated sulfonamide compound at concentrations ranging from 1 micromolar to 1 nanomolar with a minimum of four technical replicates per concentration per cell line. Cell viability was measured with the CellTiter Glo reagent (Promega) as per manufacturer’s instructions. Absolute viability values were converted to percentage viability versus DMSO control treatment, and then non-linear fit of log(inhibitor) versus response (three parameters) was performed in GraphPad Prism v7.0 to obtain an IC₅₀ values. Experiments were performed at least in duplicate or triplicate.

qRT-PCR measurement of gene expression

RNA was extracted from the indicated cell lines and reverse transcribed into cDNA using the Verso cDNA synthesis Kit (ThermoFisher Scientific). Measurement of DCAF15 (Hs00384913_m1) gene expression was performed using Taqman probes (Life Technologies) with GAPDH (Hs02786624_g1) or 18S ribosomal RNA (Hs99999901_s1) as the housekeeping gene. Assay IDs for other primers: HoxA9 Hs00365956_m1, BMI1 Hs00180411_m1, MYB Hs00920556_m1, GATA2 Hs00231119_m1. Relative expression levels across cell lines were calculated using the Delta-delta Ct method as per standard procedures.

RBM39 cDNA overexpression

RBM39 was PCR amplified from MOLM-13 cDNA and cloned into MSCV Puro-IRES-GFP construct (Addgene #18751). RBM39 G268V retroviral overexpression plasmid was a gift from Eisai Pharmaceuticals, G268V CDNA was amplified and subsequently cloned into MSCV Puro-IRES-GFP plasmid. MOLM-13 cells were transduced with either empty vector, RBM39, or RBM39 G268V construct and high GFP+ expressing cells were sorted for in vitro indisulam experiments.

Immunoprecipitation

150 million MOLM-13 cells were harvested and washed twice with 1X PBS. Cell pellets were resuspended with Cytosolic Hyptonic Buffer (10 mM HEPES pH 7.9, 10 mM KCl, 1.5 mM MgCl2, and 0.1 mM EDTA) containing protease inhibitor (PI) and phosphatase inhibitors. Subsequently, cell pellets were resuspended in Nuclear Extraction Buffer (20 mM HEPES pH 7.9, 400 mM NaCl, 1.5 mM MgCl2, 0.4% Triton X-100, and 1 mM EDTA) with PI and phosphatase inhibitors. Supernatant was extracted and diluted with Equilibrium Buffer (20 mM HEPES pH 7.9, 10% glycerol and 1 mM EDTA). Cell extracts was incubated with 2 ug of antibody overnight. Next day, cell extract containing antibodies were incubated with protein A magnetic beads and washed with BC-100 (20 mM HEPES pH 7.9, 100 mM NaCl, 10% glycerol, 0.4% Triton X-100, and 1 mM EDTA) and resuspended with loading buffer and boiled for 10 mins at 95 degrees Celsius. Supernatant was taken and western blot analysis was performed. Silver staining was performed with SilverQuest (Invitrogen) according to manufacturer’s instruction.

IP-MS

High resolution full MS spectra were acquired with a resolution of 70,000, an AGC target of 1e6, with a maximum ion time of 120 ms, and scan range of 400 to 1500 m/z. Following each full MS twenty data-dependent high resolution HCD MS/MS spectra were acquired. All MS/MS spectra were collected using the following instrument parameters: resolution of 17,500, AGC target of 5e4, maximum ion time of 120ms, one microscan, 2 m/z isolation window, fixed first mass of 150 m/z, and NCE of 27. MS/MS spectra were searched against a Uniprot human + mammalian IgG database using Sequest within Proteome Discoverer.

Animal experiments

For in vivo Cas9 experiments, RN2 Cas9-expressing cells were transduced with sgRosa (negative control) (n=4) or sgRbm39 constructs (n=7/group). At day 2 post-transduction, sgRNA positive cells (GFP+) were sorted by FACS. One million leukemia-sgRNA expressing cells were intravenously injected into each sub-lethal irradiated (5.5 Gy) B6/SJL recipient mice. For indisulam trials, a 50 mg/ml indisulam (Sigma Aldrich) stock was diluted in 20% (2-Hydroxypropyl)-β-cyclodextrin (Sigma Aldrich) to obtain a final concentration of 25mg/kg. NOD scid mice were intravenously injected with 1 million MOLM-13-expressing luciferase cells. Upon disease onset as measured by bioluminescent imaging, we intraperitoneally injected once daily with either 25 mg/kg indisulam or vehicle (1% DMSO) (n=6/group) for 13 consecutive days. All whole-body bioluminescent imaging was performed by intraperitoneally injection of Luciferin (Goldbio) at a 50 mg/kg concentration and imaging was performed after 5 mins using an IVIS imager. Bioluminescent signals (radiance) were quantified using Living Image software with standard regions of interests (ROI) rectangles. AML patient-derived xenografts were generated from patient peripheral blood and/or bone marrow mononuclear cells and subsequently transplanted intrafemorally into NSGS mice and eventually treated with indisulam (25mg/kg/daily) for 12–15 days. In vivo analysis of indisulam toxicity in C57/B6 mice was performed by intraperitoneally injection for 4 weeks (5 days on/2 days off). Additionally, human CD34⁺ cells were purified by MACS CD34⁺ from mononuclear cells from cord blood and intrafemorally injected into NSG and eventually treated with two weeks of indisulam (25mg/kg/daily) or vehicle. hCD34⁺ were also purified from mice following treatment for Western blotting analysis. Indisulam was also obtained from Eisai Pharmaceuticals.

Immunohistochemistry

Immunohistochemistry was performed on bone marrow and spleen mice treated with vehicle (DMSO) or 25 mg/kg indisulam fixed, paraffin-embedded, 5-pm tissue sections collected on plus slides (Fisher Scientific, Cat # 22–042-924) and stored at room temperature. Polyclonal rabbit RBM39 antibody (Bethyl Laboratories Cat# A300–291A) was used for immunohistochemistry. Antigen retrieval conditions were determined followed by serial dilution of the primary antibody to determine optimum dilution/concentration. A nuclear staining pattern in was considered positive. Chromogenic immunohistochemistry was performed on a Ventana Medical Systems Discovery XT instrument using Ventana’s reagents and detection kits unless otherwise noted. In Brief, sections were deparaffinized online. Antigen retrieval was performed using Cell Conditioner 2 (Citrate) for 20 minutes. Endogenous peroxidase activity was blocked for all samples. CAPER was diluted 1:500 in Ventana antibody diluent (Ventana catalog number 760–219) and incubated for 3 hours at 37°C. Primary ant ibody was detected using goat anti-rabbit horseradish peroxidase conjugated multimer incubated for 8 minutes. The complex was visualized with 3,3 diaminobenzidene and enhanced with copper sulfate. Slides were washed in distilled water, counterstained with hematoxylin, dehydrated and mounted with permanent media. Negative controls consisted of diluent only tested with the study sections.

QUANTIFICATION AND STATISTICAL ANALYSIS

Splicing analysis

Significant changes in alternative splicing events and constitutive spliced exons and intron in response to CRISPR knockout or treatment depletion were quantified by MATS v4.0.2 (ref. MATS: a Bayesian framework for flexible detection of differential alternative splicing from RNA-Seq data https://doi.org/10.1093/nar/gkr1291). Analyses were restricted to events with 10 or more reads. Events were defined as Significant if (i) the FDR corrected p value was smaller than 0.1 and (2) absolute Inclusion Level Difference was larger than 10%. Gene set enrichment was performed using the fgsea R package (1.4.0) using the KEGG, GO and MsigDB specific signatures according to the manual. Specific examples of splicing and mis- splicing events were visualized with IGV (Broad Institute).

eCLIP data analysis

The eCLIP data was processed similarly as described previously (Van Nostrand et al., 2016) and is outlined shortly in the following. First, adapter sequences were trimmed from both reads of all read-pairs using cutadapt version 1.14. Then, all remaining reads longer than 16 bases were aligned against the human reference genome sequence hg19/GRCh37 using STAR version 2.5.0c. Only uniquely mapped reads were kept. Read-pair duplicates by position were removed using picard tools version 2.6.0. To identify binding sites, we first ran a custom script to identify clusters of overlapping reads that had a read-depth of at least 10 reads. Then, we calculated significant enrichments for all such identified clusters by comparing IP-samples versus input-samples using edgeR. More specifically, we ran bamutils count version 0.5.7 to counted stranded reads within all identified clusters for all samples. Using this output, we calculated differential coverage between IP-vs-input for each cluster with edgeR after normalizing for total sequencing depth per replicate (resulting in counts per million/CPM per cluster). Final binding sites were called by applying logFC > 2 and FDR < 0.05 thresholds between IP-vs-input.

Gene Ontology analysis

Gene set enrichment was performed using the fgsea R package (1.4.0) using the KEGG, GO and MsigDB specific signatures according to the manual.

AML TCGA analysis

Raw RNA-seq reads from 28 Bone marrow progenitor cell populations and 43 AML patient from the Leucegene data set were retrieved from the US National Center for Biotechnology Information (NCBI) sequence read archive (GEO accession numbers GSE74246, GSE63569, GSE49642). Gene expression quantification was performed using the TCGA pipeline method. Briefly, Reads were aligned with STAR (2.5.3) using the GRCh38.p7 (May 2017) human assembly (https://gdc.cancer.gov/about-data/data-harmonization-and-generation/gdc-reference-files) and gene level expression were quantified using HTSeq (ref. HTSeq—a Python framework to work with high-throughput sequencing data doi:10.1093/bioinformatics/btu638). Gene expression of 151 AML patients from the TCGA project quantified by HTSeq was retrieved from GDC TCGA-LAML data portal (https://portal.gdc.cancer.gov/proiects/TCGA-LAML). Additionally, 43 AML patient samples from Leucegene were obtained and included in our gene expression analysis (Lavallee et al., 2015a; Lavallee et al., 2015b; Lavallee et al., 2016; Pabst et al., 2016). Raw counts were transformed and normalized using the voom method (ref. RNA-seq analysis is easy as 1–2-3 with limma, Glimma and edgeR doi: 10.12688/f1000research.9005.2). We evaluated RBM39 expression across known risk groups of AML patients from the TCGA based on (i) the 2017 European LeukemiaNet (ELN) risk stratification system for AML (Dohner et al., 2017) and (ii) standard AML cytogenetic risk groups (Dohner et al., 2017).

Statistical analysis

Kaplan-Meier survival curve p values were performed using Log rank Mantel-COX test. For statistical comparison, we performed unpaired Student’s t test. Statistical analyses were performed using Prism 7 software (GraphPad). Data with statistical significance are as indicated, *p< 0.05, **p< 0.01, ***p< 0.001.

Data and Software Availability

Gene Expression Omnibus: all newly generated RNA-seq data were deposited under accession number GSE114558.

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Iannis Aifantis (ioannis.aifantis@nvumc.org).

Supplementary Material

2

Table S1, Related to Figure 1: Differential expression of AML patients versus normal hematopoietic stem cells.

3

Table S2, Related to Figure 3: Mass spectrometry of RBM39 immunoprecipitation in AML

4

Table S3, Related to Figure 4: Differential Spliced Events in MOLM-13 cells treated with anti-RBM39 sgRNA versus control sgRNA

5

Table S4, Related to Figure 4: Differential Spliced Events in THP-1 cells treated with anti-RBM39 sgRNA versus control sgRNA

6

Table S5, Related to Figure 4: eCLIP identification of RBM39 interacting RNAs.

7

Table S6, Related to STAR Methods: CRISPR/Cas9 RBP-focused library

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
RBM39/Caper-alpha	Bethyl Laboratories	Cat#A300–291A
RBM39	Atlas Labs (Sigma Aldrich)	Cat#HPA001591– 100UL
Beta-actin antibody	Sigma Aldrich	Cat#A5441-.5ML
Anti-MYB antibody	Thermo Fisher Scientific	Cat#05175MI
BMI1 antibody	Abcam	Cat#ab38295
Beta-Actin Peroxidase antibody	Sigma Aldrich	Cat#A3854–200UL
Mouse IgG	Santa Cruz	Cat#Sc-2025
Bacterial and Virus Strains
MegaX DH10B Electrocomp Cells	Life Technologies	Cat#C640003
Chemicals, Peptides, and Recombinant Proteins
Indisulam	Sigma Aldrich	Cat#SML1225
E7820	Eisai	N/A
(2-Hydroxypropyl)-β-cyclodextrin powder	Sigma Aldrich	Cat#H107
D-Luciferin, Potassium Salt	Gold Bio Technology	Cat#LUCK-500
Puromycin	Sigma Aldrich	Cat#P7255–25MG
Polyethylenimine (PEI)	Polysciences, Inc	Cat#23966
Polybrene transfection reagent	Fisher Scientific	Cat#TR1003G
Dimethyl sulfoxide (DMSO)	Fisher	Cat#D128–500
Critical Commercial Assays
Click-iT Plus EdU Alex Fluor 647	Thermo Fisher Scientific	Cat#C10340
APC Annexin V kit	BD bioscience	Cat#BDB559763
NEXTflex Rapid Directional RNA-Seq Kit	Bioo Scientific	Cat#5138–07
QIAamp DNA Mini Kit	Qiagen	Cat#51304
NEBNext Poly(A) mRNA Magnetic Isolation	New England Biolabs	Cat#E7490S
NEXTflex RNA-seq Barcodes-12	Bioo Scientific	Cat#NOVA-512912
RNeasy Plus Mini Kit	Qiagen	Cat#74136
SilverQuest staining	Invitrogen	Cat#LC6070
2x Phusion Master Mix	Thermo Scientific	Cat#F-548
Deposited Data
RNA-seq Raw data	This paper	GSE114558
RBM39 eCLIP Raw data	This paper	GSE114558
RBM39 splicing analysis	This paper	GSE114558
Experimental Models: Cell Lines
Human: MOLM-13	DSMZ	ACC 554
Human: K562	ATCC	CCL-243
Human: NALM6	ATCC	CRL-3273
Human: MV4–11	ATCC	CRL-9591
Human: HNT34	DSMZ	ACC 600
Human: NKM-1	JCRB	IFO50476
Human: TF-1	ATCC	CRL-2003
Human: TF-1 SF3B1K700E/Wl knockin cells	Horizon Discovery	N/A
Human: Monomac6	DSMZ	ACC 124
Human: HEL	DSMZ	ACC 11
Human: K052	JCRB	JCRB0123
Human: SET2	DSMZ	ACC 608
Human: UKE-1	Fiedler et al., Cancer 2000	N/A
Human: THP-1	ATCC	TIB-202
Human: OCI-AML3	DSMZ	ACC 582
Human: KG-1	ATCC	CCL-246
Human: HL-60	ATCC	CCL-240
Human: KASUMI-1	ATCC	CRL-2724
Human: NOMO-1	DSMZ	ACC 542
Human: A549	Thales Papagiannakopoulos Lab	N/A
Human: CUTLL1	Adolfo Ferrando Lab	N/A
Human: HEK293T	ATCC	CRL-1573
Mouse: MLL-AF9 NrasG12 (RN2) Cas9	Zuber et al., 2011a	N/A
Human: JURKAT	ATCC	TIB-152
Human: SK-MEL239	Eva Hernando Lab	N/A
Human: 501MEL	Eva Hernando Lab	N/A
Human: K562	ATCC	CCL-243
Human: K562 SF3B1K700E/W1, SF3B1K666N/W1, U2AF1S34F/wt/wt, SRSF2P95H/wt knockin cells	PUBMED ID: 30107174	N/A
Human: NALM6 SF3B1K700E/Wl, SF3B1K666N/W1, SF3B1H662QWT knockin cells	PUBMED ID: 30107174	N/A
Experimental Models: Organisms/Strains
Mouse: C57BL/6	N/A	N/A
NOD scid gamma	JAX	Cat#005557
Oligonucleotides
sgRNA sequence for RBP domain library, see Table S6	This paper	N/A
sgRNA sequences for RBP CRISPR scanning, see Table S6	This paper	N/A
In vivo mouse Rbm39 sgRNA, see Table S6	This paper	N/A
sgRNA for DCAF15, see Table S6	This paper	N/A
Quantitative PCR
Human DCAF15 probe set, Hs00384913_m1	Life technologies	Cat#4331182
Human 18S rRNA probe set, Hs99999901_s1	Life technologies	Cat#4448485
Human GAPDH probe set, Hs02786624_g1	Life technologies	Cat#4448485
Recombinant DNA
MSCV-Cas9 puro	https://www.nature.com/articles/nbt.3235	Addgene: 65655
psPAX2	N/A	Addgene:12260
pVSVG	N/A	Addgene:12259
LRG sgRNA vector	https://www.nature.com/articles/nbt.3235	Addgene:65656
Software and Algorithms
FlowJo V8.7	TreeStar (BD Biosciences)	https://www.flowjo.com/
Prism 7.0	GraphPad	https://www.graphpad.com
Living Image Software	Perkin Elmer	http://www.perkinelmer.com/product/li-software-for-spectrum-1-seat-add-on-128113
GSEA	Broad Institute	http://software.broadinstitute.org/gsea/
rMATS	Xing Lab, UCLA	http://rnaseq-mats.sourceforge.net/
ImageJ	NIH	https://imagej.nih.gov/ij/index.html
IGV	Broad Institute	http://software.broadinstitute.org/software/igv/
Other
PerkinElmer IVIS Lumina III XR	N/A	N/A
Illumina MiSeq	N/A	N/A
BD LSR FORTESSA	N/A	N/A
Illumina HiSeq 4000	N/A	N/A
QuantStudio 6 Flex Real-Time PCR system	N/A	N/A

Significance.

RNA-binding proteins (RBPs) regulate many aspects of transcription and translation and, as such, are thought to elicit cell- and tissue-type specific functions. Here through systematic evaluation of RBPs across several cancer types, we identify RBPs specifically required in individual forms of cancer. In so doing we identify a network of functionally and physically interacting RBPs upregulated in (AML) over normal hematopoietic precursors and required for AML maintenance. Pharmacologic degradation of one such RBP, RBM39, led to aberrant splicing of multiple members of this RBP network as well as of transcriptional regulators required for AML survival. These data therefore identify RBPs with cancer-specific roles and illuminate a therapeutic approach targeting RBPs required for AML maintenance.

Highlights.

• CRISPR/Cas9 domain screen reveals RBP dependencies in AML.

• RBM39 is required for AML maintenance through mis-splicing of HOXA9 target genes.

• Proteomic studies identify an essential RBP splicing network in AML.

• Pharmacologic RBM39 degradation leads to broad anti-leukemic effects.