Overlaid Transcriptional and Proteome Analyses Identify Mitotic Kinesins as Important Targets of Arylsulfonamide-Mediated RBM39 Degradation

Abstract

Certain arylsulfonamides (ArSulf) induce an interaction between the E3 ligase substrate adaptor DCAF15 and the critical splicing factor RBM39, ultimately causing its degradation. However, degradation of a splicing factor introduces complex pleiotropic effects that are difficult to untangle, since, aside from direct protein degradation, downstream transcriptional effects also influence the proteome. By overlaying transcriptional data and proteome datasets, we distinguish transcriptional from direct degradation effects, pinpointing those proteins most impacted by splicing changes. Using our workflow, we identify and validate the upregulation of the arginine-and-serine rich protein (RSRP1) and the downregulation of the key kinesin motor proteins KIF20A and KIF20B due to altered splicing in the absence of RBM39. We further show that kinesin downregulation is connected to the multinucleation phenotype observed upon RBM39 depletion by ArSulfs. Our approach should be helpful in the assessment of potential cancer drug candidates which target splicing factors.

Implications:

Our approach provides a workflow for identifying and studying the most strongly modulated proteins when splicing is altered. The work also uncovers a splicing-based approach toward pharmacologic targeting of mitotic kinesins.

Introduction

Recent techniques that use small molecules to co-opt natural mechanisms for regulating protein stability are exciting innovations because they enable drugs that mimic genetic techniques such as RNAi or CRISPR/Cas9 (1, 2). The ubiquitin proteasome pathway (UPP) is one of the primary natural mechanisms for the controlled degradation of proteins. Within the UPP the key molecular signal for protein degradation is the sequential transfer of ubiquitin (Ub) proteins (typically K48-linked Ub 4–6mers) onto client substrates. By binding the surface of an E3 ligase substrate receptor, certain small molecules can act as molecular glues that favor the association and degradation of non-native substrates, while potentially inhibiting the association of natural substrates. Here we study the effects of arylsulfonamide (ArSulf) molecular glues, which promote degradation of the splicing factor RBM39 by directing it to the E3 ligase substrate receptor DCAF15 (Fig. 1A).

Figure 1.

A, ArSulfs cause the degradation of the splicing factor RBM39 by recruiting it to the DCAF15 cullin-ring ligase; loss of RBM39 causes splicing defects. B, The arylsulfonamide compounds proven to degrade RBM39.

Indisulam (Fig. 1B) is an antitumor ArSulf, which was discovered and tested in clinical trials more than a decade ago (3–7), but whose mode-of-action was only recently clarified in two independent studies (8, 9). Several structural studies have since shown that ArSulfs create a new protein–protein interaction by bringing together DCAF15 and the splicing factor RBM39 (Fig. 1A; refs. 10–12). Because DCAF15 is an E3 ligase adaptor protein, this binding makes RBM39 susceptible to ubiquitination and subsequent degradation in a mechanism that is reminiscent of the more established immunomodulatory drugs (IMiD) thalidomide and lenalidomide. The ArSulfs chloroquinoxaline sulfonamide (CQS), tasisulam, and E7820 (Fig. 1B) are structurally related to indisulam and have all been shown to degrade RBM39. Although numerous Phase II studies were conducted by Esai to treat colorectal cancer using indisulam (E7070) or E7820, these have not led to further trials or market approvals (NCT00165854, NCT00165867, NCT00309179, NCT01133990, NCT01347645). The same is true of other trials in patients with relapsed AML (NCT01692197) and high-risk MDS, whereas the discovery of increased lethality of spliceosomal mutant AML upon drug treatment (13) has led to an ongoing trial in patients with relapsed AML, MDS, or CMML with mutated splicing factors (NCT05024994). A recent report has shown that neuroblastoma cells have high levels of DCAF15 which could make them particularly interesting to target with the ArSulfs (14).

Recent structural and biochemical efforts have identified the second of the two RNA-recognition motifs (RRM2) within RBM39 to be responsible for the DCAF15/ArSulf/RBM39 interaction (10–12, 15, 16). Improving the ArSulf molecular glues could take three forms: enhancing DCAF15 binding, increasing RBM39 degradation capacity, or expanding the glue interaction to target other RRM-motif bearing proteins. Such improvements are hamstrung by the pleiotropic effects induced in the wake of RBM39 degradation. For example, ArSulfs directly cause DCAF15 inhibition (17), RBM39 degradation, as well as downstream transcriptome and proteome effects that result from aberrant splicing (18). Methods to analyze each of these mechanisms independently will be critical in understanding RBM39 biology, to redeploy DCAF15 against new targets, and to use ArSulfs as cancer therapeutics. In the following study, we dissected the poly-pharmacolologic effects of ArSulfs by combining various cellular and bioinformatic analyses. Our work uncovers RBM39 as a regulator of the arginine and serine rich protein RSRP-1, which is likely involved in the spliceosome (19), as well as the mitotic kinesin proteins KIF20A and KIF20B (20, 21). We also demonstrate that kinesin downregulation is the likely cause of the multinucleation phenotype observed when HCT116 cancer cells are treated with ArSulfs, and identify KIF20A as a likely contributor to ArSulf toxicity in sensitive cells.

Materials and Methods

Chemical synthesis, generation HCT-116^DCAF15-/− cells, RNA-IP experiments, and TMT labeling with analysis are described in the Supplementary Information.

Cell lines and cell culture

All cell lines were purchased from ATCC or DSMZ and used as received. During culturing they were tested for mycoplasma using the Mycostrip Mycoplasma Kit (Invivogen). The adherent HCT-116 cells were grown in DMEM with 10% FCS and 1% penicillin/streptomycin. The Calu-6 cells were grown in MEM with 10% FCS and 1% penicillin/ streptomycin. The Kelly cells were grown in RPMI 10% FCS and 1% penicillin/ streptomycin. The human leukemia cells Mv4;11, MOLM-13, K-562 (kind gifts from J. Schwaller, University Basel) were grown in RPMI with 10% FCS and 1% penicillin/ streptomycin. The cells were kept in the an incubator at 37°C with 5% CO₂. The HCT-116^DCAF15-/− cells were generated as described previously (9).

Cell treatments

All cell treatments were done in standard 12-well plates. HCT-116 cells were seeded at 170k cells per well and K562, Mv4;11, MOLM-13 cells at a density of 250k cells per well. After 1 day, the cells were treated with compounds for the indicated period of time. The adherent cells were washed thrice with ice cold PBS after which they were scraped from the wells and centrifuged at 8.5k × g for 15 minutes at 4°C. The suspension cells were collected by centrifugation (12k × g for 5 minutes), and washed by resuspending them in cold PBS and collectinon by centrifugation (12k × g for 5 minutes) three times. The PBS was removed and the cell pellets were stored at −80°C and either used for LC/MS analysis or Western blotting.

The compounds ABT-751 (MedKoo Biosciences, Inc.), Nocodazole (MedChem Express), and GSK-923295 (MedChem Express) were purchased.

The cells for RNA extraction and qPCR were not washed but directly collected in 1 mL TRI Reagent (TR 118) and stored at −80°C.

Western blot analysis

The cell pellets were lysed in 50 μL lysis buffer (RIPA buffer, supplemented with Roche cOmplete protease inhibitor and 100 mmol/L orthovanadate). The lysate was incubated on ice over 30 minutes with periodic vortexing and subsequently centrifuged at 16.9k × g for 15 minutes at 4°C. Protein concentrations were measured using DC Protein Assay. The lysates were incubated in loading buffer (5×, SDS, DTT, bromophenol blue, glycerol) at 98°C for 5 minutes. The lysates (15 μg per lane) were seperated on a 8% or 11% tris-glycine SDS-polyacrylamide gel and then transferred on nitrocellulose membranes (Amersham Protran 0.45 NC) using a discontinuos CAPS buffer system with the BioRad Trans-Blot Turbo System. The blots were blocked using Odyssey Blocking Buffer (TBS) and probed wiith antibodies for RBM39 (Sigma-Aldrich, HPA001591, 1:5k dilution), -tubulin (Abcam, ab7291, 1:10k dilution), p70 S6 kinase (SCBT, sc-8418, 1:500 dilution), PCNA (SCBT, sc-56, 1:500 dilution), p53 (SCBT, sc-126, 1:500 dilution), Cyclin A1 (SCBT, sc-271682, 1:500 dilution), Cyclin D1 (SCBT, sc-8396, 1:500 dilution), p-H3 (CST, 3377T, 1:1k dilution), Cyclin-B1 (GeneTex, GTX100911, 1:500), MCM2 (Proteintech, 10513–1-AP, 1:1k dilution), KIF20A (Proteintech, 67190–1-lg, 1:1k dilution), GAPDH (Proteintech, 10494–1-AP, 1:10k dilution). The blots were further developed using anti-mouse (Licor, IRDye 680RD or 800CW Goat anti-mouse IgG, 1:10k dilution) or anti-Rabbit secondary antiboy (Licor, IRDye 680RD or 800CW Goat anti-Rabbit IgG, 1:10k dilution) and the bands were visualized using Licor Odyssey CLx imager. The images were processed and the bands quantified using Licor Image Studio software.

Sample preparation for TMT and PRM LC/MS-MS analysis

Cells were lysed in 50 μL lysis buffer [1% sodium deoxycholate (SDC), 0.1 M TRIS, 10 mmol/L TCEP, pH 8.5] using strong ultra-sonication (10 cycles, Bioruptor, Diagnode). Protein concentration was determined with a Pierce BCA Kit (Thermo Fisher Scientific) using an aliquot of the sample. Aliquots of 50 μg of total proteins were then concentrated by heating for 10 minutes at 95°C, followed alkylation with 15 mmol/L chloroacetamide for 30 minutes at 37°C. Digestion was then performed by incubation with sequencing-grade modified trypsin (1/50, w/w; Promega) for 12 hours at 37°C. The resulting tryptic digests were acidified (pH < 3) using TFA and desalted using iST cartridges (PreOmics) according to the manufacturer's instructions. Samples were dried under vacuum and stored at −20°C until further use.

PRM LC/MS-MS assay generation

In a first step, parallel reaction-monitoring (PRM) assays (22) were generated from a mixture containing 50 fmol of each proteotypic heavy reference peptide of the target protein RSRP1 (AATEEASSR, ALGTTNIDLPASLR, MELLEIAK, TVYPEEHSR; JPT Peptide Technologies GmbH), KIF20A (DEKIEELEALLQEAR, EHSLQVSPSLEK, ESLTSFYQEEIQER; JPT Peptide Technologies GmbH), KIF20B (FGDFLQHSPSILQSK, TLNVLFDSLQER; JPT Peptide Technologies GmbH). Peptides were subjected to LC/MS-MS analysis using a Q Exactive Plus Mass Spectrometer fitted with an EASY-nLC 1000 (both Thermo Fisher Scientific) and a custom-made column heater set to 60°C. Peptides were resolved using a RP-HPLC column (75 μm × 30 cm) packed in-house with C18 resin (ReproSil-Pur C18–AQ, 1.9 μmol/L resin; Dr. Maisch GmbH) at a flow rate of 0.2 μL/min using Buffer A [0.1% formic acid (v/v) in H₂O] and Buffer B [0.1% formic acid (v/v) in MeCN/20% H₂O (v/v)] and a linear gradient 5% to 45% (60 minutes; B). Further instrument settings, experimental parameters, and data analysis workflows have been previously reported and we followed these with minimal changes (23).

PRM LC/MS-MS analysis

Peptide samples for PRM analysis were resuspended in 0.1% aqueous formic acid, processed, and measured according to a previously published protocol (23). The normalized ratios were transformed from the linear to the log-scale, normalized relative to the control condition. For RSRP-1 the median ratio among peptides was reported. The treatments and measurements for KIF20A and KIF20B quantification were conducted in biological quadruplicates (n = 4) and the mean ratios amongst the peptides and the replicates were reported. Statistical analysis was performed using t test.

Code

The python scrpt used to map the eClip scores onto the proteomics data has been deposited to GitHub https://github.com/Gillingham-Lab/eCLIP-Proteomics.

qRT-PCR measurement of gene expression

RNA was extracted from the treated cells by using TRI Reagent (MRC, TR118) according to the manufacturer's protocol and reverse transcribed using the standard protocol for SuperScript III Reverse Transcriptase (Thermo Fisher Scientific), whereby Oligo(dT)15 Primers (Promega) were used. Measurement of the gene expression was done using PowerUP SYBR Green Master Mix (Thermo Fisher Scientific) in triplicate (n = 3 biological triplicate) on an Applied Biosystems StepOnePlus Instrument with GAPDH as the housekeeping gene. The data were quantified using StepOne Software v2.3 and the C_t values were used for averaging across the biological replicates. Primer sequences provided in Supplementary Information.

siRNA

The siRNA-mediated knockdowns of DCAF15 (J-031237–18, Horizon Discovery), RBM23 (J-016689–11, Horizon Discovery), and RBM39 (D-011965–02, Horizon Discovery) in HCT-116 cells were done using the reverse transfection protocol for RNAiMax (Thermo Fisher Scientific), whereby a solution of siRNA (3 μL, 10 μmol/L) diluted in 197 μL Opti-MEM I (Gibco) was directly added to 2 μL of RNAiMAX in the well. The mixture was left to incubate at rt for 5 minutes after which a suspension of 200k cells in 0.8 mL DMEM supplemented with 10% FCS was added. The cells were left to incubate at 37°C with 5% CO₂ for 72 or 96 hours.

Flow cytometry

For the cell-cycle assays 500k cells were seeded in a 6-well plate format in 2 mL growing medium. For cell synchroization, a double thymidine block was conducted by treating the cells the next day with 40 μL thymidine (100 mmol/L in PBS) and left to grow in the incubator for 16 hours. Hereafter, the growing medium was removed and the cells washed with warm PBS once. The cells were left too grow in fresh medium for another 8 hours and then treated with 40 μL thymidine again. Once they had grown for another 16 hours, the medium was removed again, the cells washed with warm PBS. Subsequently, they were grown (released) and treated in fresh medium for the indicated time points. The growing medium was removed and 0.05% Trypsin was added to detach the cells. The trypsin was neutralizied by the addition of growing medium and the cells collected via centrifugation (200 × g at 4°C for 5 minutes). The pellets were washed twice with PBS, resuspended in cold PBS and fixed with 70% ethanol. The fixed cells were collected by centrifugation (8000k × g at 4°C for 5 minutes), resuspended in staining buffer [0.1% (v/v) Triton; 0.005% propidium iodide (PI); 0.02% DNAse-free RNAse A]. The cells were incubated at 37°C for 15 minutes after which they were measured on a BD LSR Fortessa Analyzer and the data analyzed with FlowJo software (version 10.5.3; TreeStar).

In the cell viability measurements, 500k cells were seeded in a 6-well plate format in 2 mL growing medium and left to grow in the incubator for 16 hours. Hereafter, the growing medium was replaced for fresh medium and the cells were treated for the indicated timepoints. The growing medium was collected, cells were washed with warm PBS, and 0.05% Trypsin was added to detach the cells. The trypsin was neutralizied by the addition of growing medium and the cells collected via centrifugation (200 × g at 4°C for 5 minutes). Cell viability was assessed using the Pacific Blue Annexin V Apoptosis Detection Kit with PI (BioLegends) following manufacters's protocol. The pellets were washed twice with cold Cell Staining Buffer and resuspended in Annexin V Binding Buffer. Cells were incubated with Pacific Blue Annexin V and PI at room temperature for 15 minutes after which they were measured on a BD LSR Fortessa Analyzer and the data analyzed with FlowJo software (version 10.5.3; TreeStar).

Immunofluorescence

HCT-116 cells were seeded on coverslips in a 6- or 12-well plate format in 2 mL resp. 1 mL growing medium. To overexpress KIF20A, the cells were susequently transient-transfected with 2 μg FLAG-KIF20A plasmid (Genescript; OHu24388D) using the transfection protocol for Turbofect (Thermo Fisher Scientific). They were treated with DMSO, E7820, or GSK-923295 at the indicated concentrations for the indicated time. The cells were washed twice with warm PBS, fixed using 3.7% paraformaldehyde in PBS and were permeabilized with 0.1% (v/v) Triton in PBS. The coverslips were blocked in 10% BSA in PBS at RT for 1 hour and then incubated with antibodies for CENPE-E (Abcam; ab5093, 1:500 dilution), RBM39 (Bethyl Labs; A300–291A; 1:500 dilution), FLAG (Sigma-Aldrich; F1804; 1:500), and α-tubulin (GeneTex, GTX112141, 1:250 dilution or Abcam, ab7291, 1:500 dilution or SCBT, sc-53030, 1:400 dilution) at RT for 1 hour. The coverslips were further developed using anti-Mouse (Thermo Fisher Scientific; Alexa Fluor 594, 1:200 dilution), anti-Rabbit secondary antibody (Thermo Fisher Scientific; Alexa Fluor 488, 1:200 dilution), and anti- Rat secondary antibody (Thermo Fisher Scientific; Alexa Fluor 647, 1:200 dilution) at RT for 1 hour. Finally, they were incubated with DAPI (0.2 μg/mL) for 5 minutes and mounted on slides using ProLong Diamond Antifade Mountant. The images were collected at 0.2 μm z-sections with a ×60 1.35 NA oil objective using a DeltaVision Core system (Applied Precision) with a Coolsnap HQ² camera (Roper). Images were deconvolved with SoftWoRx (Applied Precision) and maximum-intensity two-dimensional projections were assembled using FIJI (ImageJ, NIH). For the quantification, 100 cells were analyzed for each treatment condition and classified into mitotic, nonmitotic, and multinucleated cells. Each quantification was conducted in a biological triplicate (n = 3).

Data availability

The TMT-proteomics data and the eClip scores are available within the article and its supplementary data files.

The dataset used for the eClip scores is available from the Gene Expression Omnibus (GEO) at GSE114558.

Results

Effect of RBM39 degradation on cell cycle progression

Early reports suggested that the ArSulfs, Indisulam, and CQS lead to a G₁ cell-cycle arrest in murine lymphoma cells and the human HCT-116 colorectal cancer cells but on the other hand induced both a G₁–S and G₂–M arrest in non–small lung cancer cells A549 (4–6, 24, 25). The hypothesis in these early studies was that the G₁ arrest was caused by the upregulation of TP53 and the cyclin-dependent kinase inhibitor CDKN1A (aka p21; ref. 6), although this was never directly connected to any ArSulf mechanism of action. G₂–M arrest was apparent, though, in many of the reported cancer cell lines upon longer drug exposure (13). By analyzing data downloaded from depmap.org, we found that indisulam-mediated cell death strongly correlates with other mitotic poisons such as tubulin disruptors, as well as inhibitors of aurora kinase, polo-like kinase, and mitotic kinesins (Fig. 2A). In contrast, no correlation was found upon analysis for molecules that are known to cause G₁ cycle arrest, such as mTOR, EGFR, or CDK4 inhibitors (Supplementary Table S1).

Figure 2.

RBM39 degradation causes changes in the cell cycle. A, The PRISM dataset reveals that indisulam toxicity across hundreds of cancer cell lines correlates with the toxicity profile observed with mitotic poisons. B, Cell cycle was determined using flow cytometry. The data were analyzed using FlowJo 10.5.3 and the population quantified using the Watson (pragmatic) model. Asynchronous cells were treated with DMSO or E7820 (5 μmol/L) for 24 hours. C, Western blot analysis of anti-phospho-histone H3 (Ser10), anti-cyclin B1, anti-RBM39, and anti-tubulin of HCT-116 cells treated with vehicle (DMSO), E7820 (5 μmol/L), Tasisulam (5 μmol/L), ABT-751 (5 μmol/L), Nocodazole (325 nmol/L), or GSK-923295 (325 nmol/L) for 6 hours (right), 24 hours (middle), or 48 hours (left). The numbers on the Western blot indicate the signal intensities of the bands relative to the tubulin bands, normalized to DMSO treatment as quantified with Image Studio Lite.

Hence, we took a closer look at the cell-cycle effects. In particular, we used flow cytometry to explore the effects of ArSulf exposure on cell-cycle progression in HCT116 cells and the lung cancer cell line Calu-6. In asynchronously grown cells, 24-hour treatment leads to a near tripling of cells (9.8%→27.7%) in G₂–M (Fig. 2B; Supplementary Fig. S1A–S1C), with E7820 inducing the most pronounced effect. This is in line with E7820 inducing the strongest RBM39-DCAF15 binding and RBM39 degradation as we and others have observed (10, 12). Hence, we chose E7820 for further analysis. In HCT116 cells synchronized in the S-phase, we observed that after 6 hours the phases of the E7820- and DMSO-treated cells were indistinguishable (Supplementary Fig. S1D), although Western blotting confirmed RBM39 degradation (Supplementary Fig. S1E). After 12 hours, however, the DMSO cells had completed a second cell cycle whereas the E7820 cells still had a substantial fraction at G₂–M (Supplementary Fig. S1D). At 48 hours, we observed high toxicity and apoptosis in the E7820-treated samples (Supplementary Fig. S2B–S2D) across both the aforementioned cell lines and neuroblastoma Kelly cells, which have high DCAF15 levels. The latency period for toxicity is consistent with the mechanism of action of E7820, because rapid RBM39 loss is then followed by splicing changes and slow proteome adaptation. These results suggest that E7820 does not impact the cell cycle immediately upon RBM39 degradation, but on second or subsequent cycles, where G₂–M arrest is observed, ultimately leading to halted proliferation or cell death.

The microtubule inhibitors nocodazole, ABT-751, and the CENPE inhibitor GSK-923295 cause mitotic arrest. Mitotic markers such as cyclin B and phosphorylated histone H3 (Ser10) showed a moderate increase at 24 hours, but then dropped after 48 hours (Fig. 2C). This contrasted to the known microtubule inhibitors nocodazole, ABT-751 as well as the CENPE inhibitor GSK-923295 (which cause mitotic arrest) where the levels of the mitotic markers remained high. Control experiments in HCT116^DCAF15-/− cells confirmed the involvement of DCAF15 (Supplementary Fig. S2A). Furthermore, proliferation markers PCNA and MCM2 were downregulated in the surviving cells (Supplementary Fig. S2E). Taken together these results suggest that the ArSulf E7820 first leads to a G₂–M arrest, followed by cell death or nonproliferating cells. The results with E7820 contrast sharply with the other mitotic inhibitors; for example, ABT-751 gives a strong and durable G₂–M arrest, with little change in markers over 48 hours (Supplementary Fig. S2E), indicating a different mode of action. Degrading a splicing factor such as RBM39 induces downstream effects on gene transcription and protein expression. Hence, to identify the most critical molecular effects of ArSulf treatment, an ideal work-flow would measure direct protein degradation, direct splicing targets, as well as the downstream changes on the proteome. We present here a multi-omics approach that helped us identify several new strongly modulated splicing targets of RBM39 upon ArSulf treatment.

RSRP-1 splicing is altered by RBM39

To assess the overall proteome effects of the ArSulfs (Fig. 1B), we performed quantitative TMT proteomics experiments upon exposure to ArSulfs in colorectal cancer cell line HCT116 as well as in two AML cell lines, MV4;11 and MOLM-13. In each cell line, the data confirmed the exquisite selectivity of the ArSulfs for RBM39 degradation (Fig. 3A; Supplementary Fig. S3A, S3D, and S3E; Supplementary Fig. S7A). Among the consistently downregulated proteins was also RBM23—whose second RRM domain (RBM23_RRM2) shares near perfect sequence homology with that of RBM39 (RBM39_RRM2). RBM39 and RBM23 degradation caused by ArSulfs leads to extensive splicing changes (9, 13, 16). RBM23 degradation, however, has been shown to contribute little to ArSulf pharmacology (16). Furthermore, TMT-proteomics of cells upon siRNA-mediated knockdown of RBM39, RBM23, and DCAF15 (Supplementary Fig. S3F) showed that indeed only RBM39 silencing caused big changes to the proteome, indicating neither RBM23 degradation nor DCAF15 inhibition as likely drivers of ArSulf impact. Hence, to identify the precise mechanism of ArSulf cellular activity, we needed to identify those proteins most affected by RBM39 degradation. In our proteomics data, we consistently found the Arg/Ser-rich protein 1 (RSRP1) strongly upregulated (Fig. 3A; Supplementary Fig. S3A and S3D). RSRP1 is not particularly well characterized but has been shown to interact with the splicing kinases CLK1 and CLK2 in yeast 2 hybrid experiments (26, 27) as well as high throughput affinity purification mass spectrometry (AP-MS) in HEK293T cells (28). More recently, RSRP1 was shown to be directly involved in splicing and interacting with key members of the spliceosome (19), which purportedly promoted a mesenchymal (MES) phenotype in glioblastoma (GBM). Interestingly, in the same study, RBM39 was identified as a protein interactor from AP/MS experiments; it was unclear from this study, however, whether RSRP1 RNA directly interacted with RBM39. We hypothesized that RSRP1 might be a splicing target of RBM39 and decided to examine its regulation to build a workflow for characterizing RBM39 splicing targets. We assumed that the loss of RBM39 favors a particular protein coding transcript of RSRP-1 leading to its upregulation. In our hands commercial RSRP-1 antibodies performed poorly in Western blotting and global proteomics quantitation was unpredictable because RSRP-1 is a low abundance protein and therefore often not observed. As such, so we validated RSRP-1 upregulation in HCT-116 cells by parallel reaction monitoring (PRM) using four unique heavy peptides for RSRP-1. Indeed, RSRP-1 was upregulated for each ArSulf in comparison with DMSO control (Fig. 3B; Supplementary Fig. S4A) in a DCAF15 dependent manner (Supplementary Fig. S4B). We could further confirm the upregulation in Calu-6 and Kelly cells (Supplementary Fig. S4C–S4D). To validate that RSRP1 is a direct RBM39 target, we first analyzed the RBM39 eCLIP data (previously generated in one of our labs; ref. 13) and found that RSRP1 transcripts were significantly enriched by RBM39 immunoprecipitation (Fig. 3C). To understand how RSRP1 was being regulated, we designed several primer pairs that could distinguish alternately spliced RSRP1 transcripts by qPCR (Supplementary Fig. S5A for the full set of primers used). Notably, we identified that a retained intron distinguished a coding transcript (RSRP1–202) from a transcript (RSRP1–204) leading to nonsense-mediated decay (NMD; Fig. 3D). We postulated that aberrant splicing due to low RBM39 levels leads to the retained intron and increases the abundance of coding relative to noncoding transcripts. Using the Ensembl genome browser we designed primer pair E, which analyzes a region unique to noncoding RSRP-1 transcripts, such as RSRP1–204 (Fig. 3D; Supplementary Fig. S5B). Indeed, qPCR indicated that all the ArSulfs targeting RBM39 lead to downregulation of this region (Fig. 3E). These changes were also observed in K562 human leukemia cells, Kelly cells, and Calu-6 cells treated with ArSulfs (Supplementary Fig. S5E–S5G), but were absent in HCT116^DCAF15-/− control cells (Supplementary Fig. S5D). Taken together, these results support that RBM39 degradation changes alternative splicing decisions, ultimately causing upregulation of the RSRP-1 protein.

Figure 3.

RSRP1 is upregulated by RBM39 degradation. A, Change in protein levels relative to vehicle treatment (DMSO) in HCT-116 cells treated with CQS at 5 μmol/L for 6 hours quantified by TMT labeling and LC/MS-MS analysis, versus q value (Benjamini–Hochberg corrected P value from Bayes moderated t statistics). B, Change in RSRP-1 protein levels relative to vehicle treatment (DMSO) in HCT-116 cells treated with the sulfonamides at 5 μmol/L for 24 hours and quantified by taking the mean of the level of three unique peptides (each individual peptide is shown as a different colored dot) measured in PRM experiments. Individual dots represent mean levels of the peptides from biological quadruplicates (n = 4). Statistical analysis was performed using t test (*, P < 0.0332; **, P < 0.0021; ***, P < 0.0002). C, eCLIP data show strong enrichment relative to input of RSRP-1 transcripts after RBM39 immunoprecipitation. D, Comparison of noncoding RSRP1 transcript (202) and coding RSRP1 transcript (204) shows missing intron, which was analyzed by primer pair E (full transcript list Supplementary Fig. S5A). E, Change in amplicon levels from primer pair E relative to vehicle treatment (DMSO) in HCT-116 cells treated with the sulfonamides at 5 μmol/L for 6 hours. Values represent mean C_t values ±SEM from biological triplicates (n = 3, dots show individual values). Statistical analysis was performed using parametric unpaired t test with Welch correction (*, P < 0.0332; **, P < 0.0021; ***, P < 0.0002; for an example individual measurement see Supplementary Fig. S5B).

Splicing of kinesin proteins KIF20A and KIF20B is altered upon RBM39 degradation

Inspired by the RSRP-1 findings, we wondered whether there might be a more general way to identify direct transcriptional targets of RBM39 degradation. Hence, we overlayed the eCLIP targets for RBM39 (13) with proteins identified in TMT proteomics datasets after ArSulf treatment. In this way we could find proteins whose levels changed significantly upon ArSulf treatment, and whose transcripts seemed to have a direct interaction with RBM39. Two kinesin motor proteins KIF20A and KIF20B were such cases: consistently downregulated after exposure to the ArSulfs and highly enriched in the eCLIP data (Fig. 4A; Supplementary Fig. S4E). We decided to take a closer look at both proteins. As with RSRP-1, using PRM (as well as Western blotting), we could validate the ArSulf and DCAF15 dependent downregulation of both KIF20A and KIF20B using 3 and 2 unique peptides respectively in HCT-116, HCT-116^{DCAF15−/−}, and Kelly cells. (Fig. 4B–D; Supplementary Fig. S4H–S4K). In Calu6 cells, we could replicate the KIF20B downregulation did not observe significant changes in KIF20A protein levels, most likely due to its high basal expression of KIF20A. Furthermore, RNA-IP experiments for RBM39 showed enrichment of KIF20A, pointing to it being a direct splicing target of RBM39 (Supplementary Fig. S4F–S4G).

Figure 4.

KIF20A and KIF20B are downregulated by RBM39 degradation. A, Plot of proteins changes quantified from proteomics as in Fig. 3A with their scores from RBM39 eCLIP enrichment (as colors). As shown in Fig. 3A proteomics was done in Mv4;11 cells treated with indisulam at 3 μmol/L for 8 hours. Strongly up- or downregulated sequences with high eCLIP scores are likely to be transcriptional targets of RBM39, such as KIF20A or KIF20B. B, Change in KIF20A protein levels relative to vehicle treatment (DMSO) in HCT-116 (left) or HCT-116^{DCAF15−/−} (right) cells treated with the sulfonamides at 5 μmol/L for 6 hours (left) and quantified by taking the mean of the level of three unique peptides measured in PRM experiments. Individual dots represent mean levels of the peptides from biological quadruplicates (n = 4). Statistical analysis was performed using t test (*, P < 0.0332; **, P < 0.0021; ***, P < 0.0002). C, As for B but for KIF20B using two unique peptides. D, Western blot analysis of anti-KIF20A, anti-RBM39, and anti-tubulin of HCT-116 cells treated with vehicle (DMSO), E7820 (5 μmol/L) for 6 hours. The numbers on the Western blot analysis indicate the signal intensities of the bands relative to the tubulin bands, normalized to DMSO treatment as quantified with Empiria Studio. E, Comparison of noncoding KIF20B transcript (204) and coding KIF20B transcript (201) shows retained intron which was analyzed by primer pair B, whereby the coding transcript was analyzed by primer pair A (full transcript list Supplementary Fig. S5A). RBM39 loss leads to intron retention and preferential formation of a noncoding transcript in KIF20B splicing. F, Change in amplicon levels from primer pair A relative to vehicle treatment (DMSO) in HCT116 cells treated with the sulfonamides at 5 μmol/L for 6 hours. Values represent mean C_t values ±SEM from biological triplicates (n = 3, dots show individual values). Statistical analysis was performed using parametric unpaired t test with Welch correction (*, P < 0.0332; **, P < 0.0021; ***, P < 0.0002; for an example individual measurement see Supplementary Fig. S5B). G, As in E for primer pair B. H, As in D comparison of the coding transcript and noncoding transcripts show retained introns which were analyzed by primer pair A and B (full transcript list Supplementary Fig S5A). RBM39 loss leads to intron retention and preferential formation of a noncoding KIF20A transcript after splicing. I, As in E for primer pair A. J, As in E for primer pair B.

Upon analyzing the four transcripts of KIF20B listed in the Ensembl genome browser, three of them were coding and the other one retained an intron, making it noncoding. We hypothesized that RBM39 loss was causing retention of this intron, leading to a preference for the noncoding transcript. Indeed, qPCR analysis showed a significantly higher retention of the intron in ArSulf treated HCT-116, K562, Calu-6, and Kelly cells (Fig. 4E and F; Supplementary Fig. S4G), along with a slight downregulation of the coding transcripts (Fig. 4E–G; Supplementary Fig. S5E–S5G).

KIF20A has seven transcripts, of which only four are coding. Two of the remaining noncoding transcripts (KIF20A-203, KIF20A-206) retain an intron and the last one skips an exon, leading to NMD. We decided to analyze these transcripts (KIF20A-203 and KIF20A-206) more closely in the region of the retained introns (Fig. 4H). qPCR with corresponding primers identified KIF20A-203 as being upregulated upon ArSulf treatment, whereas the other was unaffected in both HCT116 and K562 cells (Fig. 4I and J; Supplementary Fig. S5E). Like with KIF20B we could again observe transcriptional changes in Calu-6 and Kelly cells (Fig. 5F and G).

Figure 5.

KIF20A and KIF20B downregulation leads to multinucleated cells. A, Immunofluorescence images of HCT-116 cells treated with DMSO or E7820 (5 μmol/L) for 48 hours and stained for DNA (blue), RBM39 (red), FLAG (cyan), and α-Tubulin (green). For clarification purposes, the images shown for the different channels correspond to different z-slices. B, Amount of multinucleated and mitotic cells observed in vehicle treated (DMSO) and E7820 treated HCT116 cells at 5 μmol/L for 48 hours. Values represent mean values ±SD from counting 100 cells per treatment in biological triplicates (n = 3, dots show individual values). Statistical analysis was performed using parametric unpaired t test with Welch correction (*, P < 0.0332; **, P < 0.0021; ***, P < 0.0002). Example images showed in Supplementary Fig. S3A. C, As for A but in HCT116 cells overexpressing FLAG tagged KIF20A. More example images shown in Supplementary Fig. S3B. D, As for B but in HCT116 cells overexpressing FLAG tagged KIF20A.

Importantly, the changes in both KIF20B and KIF20A were phenocopied by siRNA-mediated RBM39 knockdown in HCT116 cells (Supplementary Fig. S5C) and are absent upon ArSulf treatments of HCT116^DCAF15-/− (Supplementary Fig. S5D) cells as well as in HCT116 cells, where RBM23 is knocked-down with an siRNA (Supplementary Fig. S5C). Collectively, these results show that RBM39 directly mediates KIF20A and KIF20B splicing. Pharmacologic loss of RBM39 leads to aberrant splicing, ultimately altering the protein levels of both kinesins. KIF20A and KIF20B changes seemed likely contributors to ArSulf toxicity since these are important mitotic proteins often upregulated in cancer (29, 30). In summary, by combining proteomics and eCLIP analysis, we identified RSRP-1, KIF20A, and KIF20B as transcriptional targets of RBM39 that are strongly modulated at the protein level, providing a workflow that broadly explores the proteome impact of targeting splicing factors.

ArSulf treatment phenocopies KIF20A and KIF20B downregulation

Kinesin motor proteins play essential roles across each stage of cell division (20, 31, 32) and are upregulated in nearly all types of cancer (33). Whole proteome screens revealed that KIF20A and KIF20B are upregulated in mitosis (34–37), most likely in the G₂–M phase and cytokinesis/telophase transitions (38). Both KIF20A (a.k.a. MKLP2) and KIF20B (a.k.a. MPP1) belong to the Kinesin-6 family and are involved in the organization and regulation of the mitotic spindle and cytokinesis (21, 32). Inhibiting KIF20A with the small molecule Paprotrain or by siRNA-mediated knockdown has resulted in varying effects on cancer cells (39), leading to attenuated cell growth or apoptosis (40). A consistent finding, however, is that inhibiting KIF20A and KIF20B activity leads to failure in cytokinesis, resulting in multinucleated cells (32, 41–43). If ArSulf treatment was causing toxicity by altering kinesin splicing, we should also observe increases in multi-nucleated cells upon treatment. Indeed, exposure for 24 hours to E7820 resulted in a measurable increase in multinucleated cells, but the effect after 48 hours was more dramatic (Fig. 5A; Supplementary Fig. S6A), with 20% of cells showing multinucleation. As would be expected in cells experiencing such mitotic stress, the 48 hours time-points also showed high cytotoxicity. In addition, cells remaining after 48 hours showed significantly reduced mitosis (Fig. 5B). To rescue the induced phenotype, we overexpressed FLAG tagged KIF20A by transient transfection (Fig. 5C; Supplementary Fig. S6B and S6C) and observed fewer multinucleated and mitotic cells (Fig. 5D). We observed, however, that the overexpression of KIF20A also sometimes gave rise to condensed nuclei and altered microtubule networks (Supplementary Fig. S6B). In summary, lowering of KIF20A and KIF20B protein levels after pharmacologic RBM39 degradation is a feature observed in five cell lines covering three indisulam-sensitive cancer lines originating from different organ systems (colon carcinoma (HCT116), leukemia [MV4;11, MOLM-13, KBM7 (data from others; ref. 44)], and neuroblastoma [IMR-32 (data from others; ref. 7)]. We further demonstrate that multinucleation and cytotoxicity are downstream effects of these changes in HCT116: a cell line of particular ArSulf-sensitivity.

Discussion

From previous studies (8, 9, 14) it is clear that the ArSulfs are potent DCAF15-dependent RBM39 degraders but that various types of cancer cells have shown different sensitivities. In general, DCAF15 levels in cells correlate with sensitivity to ArSulfs (depmap.org; refs. 14, 15), however little can be said in general about the targets giving rise to the toxicity. With the help of functional genomics HOXA9 targets were shown to be misspliced in the absence of RBM39 in AML and splicesome mutations increased the sensitivity to RBM39 loss (13). Recent studies show preferential sensitivity of neuroblastoma cells upon ArSulf treatment and that factors from key pathways like cell cycle (e.g., CDK4) or DNA repair are being misspliced and downregulated as a result of RBM39 degradation (7, 14). It was further suggested the increased sensitivity could stem not only from high DCAF15 levels but from the need for timely mRNA processing in Myc-driven neuroblastoma. A more recent preprint used a CRISPR dropout screen in the lung cancer cell line A549 to identify the Serine/Arginine (SR) protein kinase 1 (SRPK1), which is specific to the SR family of splicing factors as a synthetic lethal interaction of ArSulf treatments (45). Taken together, these findings suggest that the splicing targets of RBM39 and any possible crosstalk amongst splicing factors need to be looked at carefully to identify cell lines or cancers that would be particularly intolerant of RBM39 downregulation.

Exploiting RBM39 degradation as a therapeutic strategy will require a precise understanding of the proteins most affected by its degradation. Although resistance screens towards ArSulfs gave rise to most mutations in the key α-helix containing the RRM2 domain of RBM39 (8–10, 12), the splicing activity of the mutants remains to be shown and does not preclude that the splicing targets also contribute to resistance. On this note, recently, mutations outside of this domain were identified to confer resistance to ArSulfs (46). Moving forward with the inhibition or degradation of RBM39 as a therapeutic strategy will require defining its key splicing targets, whether in wildtype or resistant mutants. We have outlined a methodology using a multi-omics approach which can help disentangle the multifaceted effects of RBM39 degradation.

We identified RSRP-1 (a largely uncharacterized protein most likely involved in the splicesome; ref. 19) as a gene whose protein levels are altered by RBM39-mediated splicing. Expanding our approach, we identified two kinesin proteins which are downregulated and are important for proper cytokinesis. eCLIP data, transcript analysis, and qPCR of RBM39-IPs show that changes in splicing programs are consistent with a direct physical interaction between RBM39 and KIF20A and KIF20B transcripts. Cell-cycle analysis shows that a larger proportion of cells treated with ArSulfs remain in G₂–M. Typically, however, KIF20A and KIF20B are upregulated in G₂–M as was evident from proteomics data (Supplementary Fig. S7A) of nocodazole, ABT-751, and GSK-923295 treated cells (molecules that cause G₂–M arrest via different mechanisms). Furthermore, the observed cellular phenotype of, for example, CENPE inhibitors is quite different than ArSulfs (Supplementary Fig. S7B), although they both lead to G₂–M arrest. Hence, the unusual downregulation of KIF20A and KIF20B is again indicative of a direct splicing effect and accentuates the importance of identifying downstream targets of RBM39 degradation. Of the three strongly altered targets identified, KIF20A is likely the most consequential in cancer. We can infer this from our own data (Fig. 5) as well as from comparing expression levels (Supplementary Fig. S8A) as well as overall survival in cancer (Supplementary Fig. S8B and S8C) for RSRP1, KIF20A, and KIF20B using data in publicly available datasets (analyzed with the GEPIA tool; ref. 33). Although KIF20A and KIF20B are strongly overexpressed (Supplementary Fig. S8A) in most cancers (as might be expected from proteins involved in cell division), only KIF20A is predictive of survival in certain tumors (Supplementary Fig. S8B–S8C). Elevated KIF20A levels have been associated (47) with poor outcomes in breast, cervical, and pancreatic (48) cancer. A recent meta-analysis has pointed to reduced overall survival with overexpressed KIF20A as a robust observation (49), although this was a purely correlative analysis. The accumulated evidence on KIF20A in cancer has led to industrial medicinal chemistry efforts to develop a direct KIF20A inhibitor, resulting in DIACC2010 (50). Although the structure of DIACC2010 has yet to be disclosed, the molecule has shown promising in vivo efficacy data in preclinical models of leukemia. Our results suggest that an alternative approach to KIF20A targeting would be to target RBM39 with ArSulfs.

Modulating alternative splicing of disease-causing proteins (51) or targeting splicing mutated cancers (52) is an emerging area of therapeutic development. In this vein the ArSulfs represent an excellent starting point because they profoundly alter splicing decisions. With the present analysis workflow in hand, it would be interesting to examine ArSulfs in diseases where targeting faulty splicing (53) or inducing exon skipping (54) have delivered therapeutics. More generally, we believe this workflow will be helpful for finding molecular glues that have downstream effects on transcription, such as transcription factors or splicing factors. Projecting forward from our findings, SLAM-seq (55) coupled with proteomics would offer a global comparison of transcriptome/proteome changes for any target protein in response to treatment with protein degrading molecules.

Supplementary Material

Supplementary Data

Supplementary Data

Supplementary Data Table 1

Supplementary Data Table 1

Supplementary Data

eClip Scores

Supplementary Data Table 2

Supplementary Data Table 2

Supplementary Data

TMT Proteomics Datasets

Supplementary Figure S1

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S2

Supplementary Figure S3

Supplementary Figure S3

Supplementary Figure S4

Supplementary Figure S4

Supplementary Figure S5

Supplementary Figure S5

Supplementary Figure S6

Supplementary Figure S6

Supplementary Figure S7

Supplementary Figure S7

Supplementary Figure S8

Supplementary Figure S8

This article is featured in Highlights of This Issue, p. 753

Authors' Disclosures

O. Abdel-Wahab reports grants from Nurix Therapeutics, Loxo/Lilly, and grants from Minovia Therapeutics; other support from Envisagenics and Harmonic Discovery Inc. outside the submitted work. D. Gillingham reports personal fees from Forx Therapeutics and Innosuisse outside the submitted work. No disclosures were reported by the other authors.

Authors' Contributions

S. Coomar: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. P. Mota: Investigation, methodology, project administration, writing–review and editing. A. Penson: Resources, data curation, software, formal analysis, visualization, methodology, writing–review and editing. J. Schwaller: Formal analysis, supervision, methodology, writing–review and editing. O. Abdel-Wahab: Data curation, software, formal analysis, validation, visualization, methodology, writing–review and editing. D. Gillingham: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.