SF3B1 inhibition disrupts malignancy and prolongs survival in glioblastoma patients through BCL2L1 splicing and mTOR/ß-catenin pathways imbalances

Glioblastoma is one of the most devastating cancer worldwide based on its locally aggressive behavior and because it cannot be cured by current therapies. Defects in alternative splicing process are frequent in cancer. Recently, we demonstrated that dysregulation of the spliceosome is directly associated with glioma development, progression, and aggressiveness. Different human cohorts and a dataset from different glioma mouse models were analyzed to determine the mutation frequency as well as the gene and protein expression levels between tumor and control samples of the splicing-factor-3B-subunit-1 (SF3B1), an essential and druggable spliceosome component. SF3B1 expression was also explored at the single-cell level across all cell subpopulations and transcriptomic programs. The association of SF3B1 expression with relevant clinical data (e.g., overall survival) in different human cohorts was also analyzed. Different functional (proliferation/migration/tumorspheres and colonies formation/VEGF secretion/apoptosis) and mechanistic (gene expression/signaling pathways) assays were performed in three different glioblastomas cell models (human primary cultures and cell lines) in response to SF3B1 blockade (using pladienolide B treatment). Moreover, tumor progression and formation were monitored in response to SF3B1 blockade in two preclinical xenograft glioblastoma mouse models. Our data provide novel evidence demonstrating that the splicing-factor-3B-subunit-1 (SF3B1, an essential and druggable spliceosome component) is low-frequency mutated in human gliomas (~ 1 %) but widely overexpressed in glioblastoma compared with control samples from the different human cohorts and mouse models included in the present study, wherein SF3B1 levels are associated with key molecular and clinical features (e.g., overall survival, poor prognosis and/or drug resistance). Remarkably, in vitro and in vivo blockade of SF3B1 activity with pladienolide B drastically altered multiple glioblastoma pathophysiological processes (i.e., reduction in proliferation, migration, tumorspheres formation, VEGF secretion, tumor initiation and increased apoptosis) likely by suppressing AKT/mTOR/ß-catenin pathways, and an imbalance of BCL2L1 splicing. Together, we highlight SF3B1 as a potential diagnostic and prognostic biomarker and an efficient pharmacological target in glioblastoma, offering a clinically relevant opportunity worth to be explored in humans.


Background
Gliomas are the most frequent (> 80%) primary malignant brain tumor in adults [55]. They are classified into low-grade (I and II) and high-grade (III and IV) gliomas based on integrated classic histological/molecular features [43]. Grade IV astrocytoma, the most prevalent glioma, known as glioblastoma multiforme (GBM), is one of the most devastating and malignant cancers [55] and its incidence has increased relevantly in recent years, while in other gliomas remained stable [57]. Despite significant advances in the knowledge of GBM pathophysiology, it remains an incurable disease with median survival after diagnosis of ~ 15 months [54,61]. Still, effective therapeutic targets are severely lacking, and, therefore, innovative therapeutic approaches are urgently needed [62].
Growing evidence indicates that defects in the alternative splicing process are frequent in cancer, which has gained important attention in the past 10 years [27,53]. Moreover, our group and others have demonstrated that the spliceosome, the cellular machinery controlling the splicing process, is drastically altered in GBM and different cancer types [7,23,31,34,67], leading to the appearance of aberrant/oncogenic splicing variants (SVs) from different genes [e.g., GFAP [47]/VEGF [26]/ TP53 [2]/BCL2L1 [72]/TP73 [23]]. Specifically, we have demonstrated that the dysregulation of the spliceosome is associated with GBM development/progression/ aggressiveness, which could potentially be considered as a source of novel diagnostic/prognostic-biomarkers and therapeutic targets to combat this devastating pathology [23].
The splicing-factor-3B-subunit-1 (SF3B1) is a core spliceosome component essential for splicing function [66]. SF3B1 gained importance due to many functionally deleterious mutations found in various cancer types [41] [i.e., myelodysplastic syndrome [29]/breast cancer [20]/prolactinomas [38]/uveal melanoma [32]/pancreatic ductal adenocarcinoma [4]], which are associated with patient poor-prognosis/survival. Additionally, we have recently found that SF3B1 is overexpressed and associated with malignant features in prostate cancer [30] and hepatocellular carcinoma [42], supporting that SF3B1 could represent a valuable therapeutic target in cancer. Accordingly, various drugs have now been designed to specifically target SF3B1, including pladienolide B, a selective inhibitor that disrupts the spliceosome assembly [16,35,48]. However, to the best of our knowledge, the oncogenic implication of SF3B1, its somatic mutations, and expression profile or its association with molecular features and clinical parameters have not been characterized in GBM, nor its putative therapeutic potential. Therefore, different human cohorts and a dataset from different glioma mouse models were analyzed to determine the mutation frequency as well as the gene and protein expression levels between tumor and control samples of the SF3B1, an essential and druggable spliceosome component. SF3B1 expression was also explored at the single-cell level across all cell subpopulations and transcriptomic programs. The association of SF3B1 expression with relevant clinical data (e.g., overall survival) in different human cohorts was also analyzed. Moreover, several functional and molecular endpoints were measured in different GBM cell models (human primary cultures and two cell lines) after SF3B1 blockade (using pladienolide B treatment). In addition, tumor progression and initiation in response to SF3B1 blockade were examined in two GBM xenograft mouse models. These analyses unveil SF3B1 as a potential biomarker being a novel pharmacological target in this devastating tumor.
characteristics were collected to perform clinical correlations. This study was approved by Reina Sofia University Hospital Ethics Committee and was conducted by the principles of the Helsinki Declaration. Written informed consent was obtained from all individuals.

Bioinformatic analysis of in silico cohorts for RNAseq and proteomic data
All the bioinformatic methodology was implemented in R language 3.5. Specifically: i) Rembrandt microarray (n = 219 GBM; n = 28 non-tumor) and CGGA (bulk-RNAseq data; n = 388) were interrogated through the GlioVis-Tools (http:// gliov is. bioin fo. cnio. es) ( Table  S2). ii) Single-cell RNAseq data of adult GBM were downloaded from Single-cell -Portal -Broad-Institute (GSE131928; total adult cells, n = 5528) [50] and analyzed using Seurat-packageV3 [60] . Filtering was performed removing cells with < 200 and > 8000 features and selecting cells with a percentage of mitochondrial genes over 0.9 (n = 5123 filtered cells were obtained; Fig. S1a-b). Data were normalized using LogNormalize-method and scaled with a factor = 10,000. PCA and UMAP methods were applied to perform cell clustering ( Fig. S1c-d). Top 10 markers were used to characterize each cluster (Table S3/ Fig. S1e). Transcriptional programs were classified using a relative meta-module score [log2(|SC1-SC2| + 1)] [50]. iii) Paired-end bulk-RNAseq data from EPed mouse models have been aligned against UCSC hg19 assembly using STAR2.7.0a. Normalization, count per gene associations, and differential expression analysis were achieved by Partek Flow ® software (Partek Incorporated, St. Louis, MO, USA). iv) CPTAC GBM Discovery Study proteomic data (n = 100 GBM; n = 10 Non-tumor; Table S4) were downloaded from https:// cptac-data-portal. georg etown. edu [19]. v) Genomics of Drug Sensitivity in Cancer (GDSC) database was used to determine the resistance and sensitivity of 100 compounds on 18 GBM cell lines (https:// www. cance rrxge ne. org) and combined with SF3B1 expression of GBM cell lines from Broad Institute Cancer Cell Line Encyclopedia (CCLE-https:// porta ls. broad insti tute. org/ ccle). vi) Group of patients for survival analyses were selected based on the cutoff points determined by survminer R package. vii) STRING database (https:// string-db. org) was used to determine the potential functional association between several genes correlated with SF3B1 (r > ± 0.800). Enrichment analysis was performed based on KEGG-Pathways Analysis (Table  S5). Reactome database was used to identify relevant pathways associated with SF3B1 expression and plotted using ggplot2 R package.

RNA isolation, quantitative real-time RT-PCR (qPCR), and customized qPCR dynamic array based on microfluidic technology
Total RNA from fresh non-tumor and tumor human samples and from GBM cell lines was extracted and DNase-treated, the concentration quantified, and the RNA retro-transcribed for qPCR analyses as previously described [9,23]. As recently reported [30,31], qPCR dynamic array based on microfluidic technology was implemented to determine the expression of SF3B1 simultaneously in human samples and cell lines. Specific primers for human transcripts including SF3B1, key GBM tumor markers, selected signaling pathway endpoints genes and 3 housekeeping genes were specifically designed with the Primer3 4.0.0 software (Table  S6). To control for variations in the efficiency of the retrotranscription -reaction, mRNA copy numbers of the different transcripts analyzed were adjusted by a normalization factor, calculated with the expression levels of 3 housekeeping genes [β-actin (ACTB), hypoxanthine-guanine phosphoribosyl-transferase (HPRT), glyceraldehyde 3-phosphate dehydrogenase (GAPDH); Table S6] and the GeNorm 3.3 software as previously reported [31,44].

GBM cell lines
U-87 MG and U-118 MG cells were obtained from the American Type Culture Collection (ATCC, #HTB-14/ #HTB-15, respectively) and cultured according to the supplier's recommendations. These cell lines were previously checked for mycoplasma contamination by PCR as previously reported [65].

Dose-response, IC 50 determination, and measurements of proliferation and migration rates
Proliferation assay was used to perform a dose-response [1 nM, 100 nM, and 10 μM; dose selected based on previously reported in vitro studies [30,67]] and IC 50 determination (at 48 h) of pladienolide B in GBM cell lines and primary-GBM cell cultures. Least-squares regression was used as a fitting method to IC 50 determination. As previously described [28], cell proliferation was analyzed using alamarBlue ™ assay (5,000 cells/well for cell lines and 10,000 cells/well for primary cell-cultures; Biosource International, #BUF012B), and migration using the wound-healing technique (150,000 U-118 MG cells/ well). For the migration assay, U-118 MG cells cultured under confluence were serum-starved for 24 h to achieve cell synchronization, and then, the wound was made using a 200 μl sterile pipette tip. Wells were replaced and cells were incubated for 6 h and 24 h with supplemented medium without FBS. Wound-healing was compared with the area just after the wound was performed. Three pictures were randomly acquired along the wound per well to calculate the area by ImageJ 1.8.0_172 software [58].

Apoptosis measurement
Apoptosis induction in response to pladienolide B treatment in GBM cell lines (5,000 cells/well onto white-walled multiwell luminometer plates) was performed by using Caspase-Glo ® 3/7 Assay (Promega Corporation, #G8091) as previously reported [23]. In addition, Cleaved-Caspase 3 protein level was identified by western blot (see below) after pladienolide B treatment.

Tumorspheres formation
Previously described assay [23] was carried out with both GBM cell lines (100 cells/well) cultured in a Corning Costar ultra-low attachment plate (#CLS3473) using D-MEM F-12 (Gibco, #11320033) with EGF (20 ng/μl) (#SRP3027) for 10 days (refreshing every 48 h, EGF and pladienolide B treatment) [23]. Additionally, tumorspheres formation was measured in U-87 MG and U-118 MG cells, pre-treated with pladienolide B (24 h and 48 h) before seeding the experiment. Photographs were taken to visualize and measure the area after 10 days of incubation with pladienolide B.

VEGF secretion
The VEGF Human ELISA Kit (ThermoFisher-Scientific, #KHG0112) was used to quantify VEGF secretion in response to pladienolide B in GBM cell lines, following the manufacturer's instructions and previously described methods [23].

Colony formation
Colony formation assay was performed in GBM cell lines. Briefly, 300 or 500 cells/well (6-well plate) of U-87 MG and U-118 MG were seeded, respectively. Cells were pre-treated with pladienolide B (for 24 h and 48 h) before seeding the experiment to evaluate its effect on tumor onset/formation. Then, cells were seeded, medium was replaced, cells washed with PBS 1x, and crystal violet 0.5% plus glutaraldehyde 6% was added and incubated 45 min at room temperature. Finally, cells were rinsed 3 times with distilled water and left to dry at room temperature. Colonies (particles per well) were measured by ChemiDoc-XRS+ System (Bio-Rad, Hercules, CA) and analyzed using ImageJ 1.8.0_172 software.

Preclinical mouse models, Micro-CT imaging, and Hematoxylin & Eosin examination
A preclinical xenograft mouse model to test pladienolide B in vivo was developed. 5-week-old ATHYM-Foxn1 nu/ nu mice (n = 6; Janvier-Labs) were injected subcutaneously with 3 × 10 6 of U-87 MG cells in both flanks [resuspended in 100 μl of basement membrane extract (Trevigen, #3432-010-01)]. Once the tumor was clearly measurable, each mouse received an intra-tumor injection (12 days after cell-inoculation) with 50 μl of pladienolide B (100 nM) into one flank and vehicle (1xDulbecco's Phosphate-Buffered Saline, Sigma-Aldrich, #D1408; used as control) into the other flank. Tumor growth was monitored every 2 days using a digital caliper. Eight days after injection, mice were sacrificed and each tumor was dissected, fixed, and sectioned for histopathologic examination after H&E-staining. Examination of mitosis number, vascular proliferation, and necrosis was performed by expert pathologists. Additional tumor pieces were placed in liquid nitrogen and then frozen at -80 °C until RNA or protein extraction using Trizol-reagent or SDS-DTT buffer, respectively, and as previously reported [23]. Micro-CT imaging using SkyScan1176 Bruker and software environment was used to show in vivo tumor location previous to dissection. Specifically, 2D analysis together with 3D imaging rendering was performed using VolView 3.4 software (KitWare Inc).
In addition, a preclinical xenograft mouse model was developed by inoculating U-87 MG cells previously pretreated with pladienolide B at 100 nM in vitro for 24 h and 48 h. Specifically, 5-week-old ATHYM-Foxn1 nu/nu mice were injected subcutaneously with 3 × 10 6 of U-87 MG pre-treated cells [n = 6 mice/condition (i.e., cells pretreated for 24 h or 48 h with pladienolide B in one flank and their corresponding vehicle-treated controls in the other flank)] using similar approaches described above. In this case, Slicer 4.11 software was used for 2D analysis together with 3D imaging rendering in the Micro-CT imaging. These experiments were performed according to the European Regulations for Animal Care under the approval of the university/regional-government research ethics committees.

SVs detection by end-point-PCR and by qPCR in response to pladienolide B treatment
End-point-PCR was developed using cDNA from GBM cell lines, primary-GBM cell cultures, and U-87 MG xenograft mouse models in response to pladienolide B vs. control condition to detect SVs of KLF6, CRK, MST1R, CASP2, RAC1, MCL1, BIRC5, SPP1, and BCL2L1 using specific primer pairs of each gene. Specifically, primer design for BCL2L1 was performed using primers with specific annealing in ExonIIa (forward-sequence) and ExonIII (reverse-sequence) neighboring the splicing event (Fig. 9g). Different sized amplicons were estimated and subsequently identified by agarose gel-electrophoresis (BCL2L1-xS: 305 pb; BCL2L1-xL: 494 pb). Details of the end-point PCR to detect splicing events have been previously reported [18]. Then, qPCR was performed using the same cDNA samples using specific primers for each SVs (BCL2L1-xS and BCL2L1-xL) to quantify individually both SVs and calculate the ratio BCL2L1-xS/ BCL2L1-xL. All primer sequences are included in Table  S6.

Antisense oligonucleotides design, transfection, and proliferation assay
Five different antisense oligonucleotides (ASOs; 18-24 bases) were designed based on different studies reporting how an ASO could be accurately designed to be stable inside the cell and identifying the most appropriated BCL2L1 region [5,24,40,45]. Briefly, these ASOs target different sequences of the ISS (Intron Splicing Silencer) region in the interexon region of Bcl-xL/Bcl-xS exon II and III, where the splicing is carried out. Phosphorothioate bonds binding all bases and four 5′ ends modified by 2′ O-methoxy-ethyl (2'MOE) residues were placed at the oligo's end. Then, 100 nmol DNA Oligo was synthesized by Integrated DNA Technologies, Inc. [5,45]. The ASO sequences are included in Table S7. For the cell proliferation assay, ASO1_BCL2L1 and ASO3_BCL2L1 were used. Briefly, 200,000 cells were transfected with 100 nM of each ASO individually using Lipofectamine ™ 2000 (Ther-moFisher-Scientific, #11668019) according to the manufacturer's instructions. Nuclease-free water was used as a control condition. After 48 h, cells were collected for validation of the transfection (SVs detection) and seeded for proliferation assays (see above). Pladienolide B was administered 24 h before SV detection.

Statistics
Data were evaluated for heterogeneity of variance by using the Kolmogorov-Smirnov test. Statistical differences were assessed by T-test, Mann-Whitney U test, or by 1-way ANOVA followed by Fisher's correction exact test. Correlations were studied by using the Pearson correlation test. All statistical analyses were performed using Prism software 8.0 (GraphPad Software, La Jolla, CA, USA). P-value < 0.05 were considered statistically significant. Data represent median (interquartile-range) or means±SEM. Plus symbol (+) indicates a tendency between conditions (+P > 0.05 < 0.1). Asterisks (*P < 0.05; **P < 0.01; ***P < 0.001) indicate statistically significant differences, and "ns" indicates not statistically significant differences, across different conditions.

Data availability
External bulk-RNAseq data analyzed in the present study are available in GlioVis-Tools (http:// gliov is. bioin fo. cnio. es) and single-cell RNAseq data in the single-cell Portal-Broad Institute (https:// singl ecell. broad insti tute. org). The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

SF3B1 is markedly overexpressed in human GBM samples compared to non-tumor brain samples
SF3B1 mRNA levels were analyzed in three different human cohorts (Table S2). Specifically, a marked SF3B1 overexpression was found in GBM compared to non-tumor brain tissues (control-tissues) in our cohort (n = 22 and 4, respectively; Fig. 1d; Fig. S2f), which was also corroborated in another well-characterized external patient cohort (Rembrandt; n = 219 and 28, respectively; Fig. 1d; Fig. S2g). Moreover, we also observed that the expression levels of SF3B1 found in the tumor samples of these two cohorts were comparable with those found in the CGGA database (n = 388 GBM-samples; control-samples are not available; Fig. 1d [8]. Notably, Receiver-Operating-Characteristic (ROC)-curve analyses revealed the capacity of SF3B1 levels to strongly discriminate between GBM vs. control-tissues, showing an

SF3B1 single-cell characterization in human intra-tumor cell populations
SF3B1 expression was analyzed at single-cell level (GSE131928; n = 5528), which includes tumor microenvironment (TME) and tumor-like cells (Fig. 1g). Clustering analysis and classification based on cellular markers uncovered three different TME cell populations (Microglia/Macrophages, CD2+ immune cells, and OPC-like cells) and three tumor-like cells (NPC-like cells, Astrocyte-like cells, and Cycling cells) (Fig. 1h top-panel, and Table S3), being all these cell populations directly associated with tumor progression and dissemination [25,50]. SF3B1 expression was homogeneously present across the different cell populations ( Fig. 1h bottom-panel, and Fig. 1i), being this expression virtually higher in TME vs. tumor-like cells ( Fig. 1i; Fig.  S1f). Likewise, SF3B1 was expressed in all transcriptional programs of GBM cells that recapitulate distinct neural cell states (NPC-like, MES-like, AC-like, OPC-like; Fig. 1j) [50], wherein a higher expression was found in neural progenitorlike program (with proliferative potential) than the other programs (Fig. 1k). Therefore, the ubiquitous expression of SF3B1 across all intra-tumor cell types/states suggests that SF3B1 might represent a potential and global pharmacological target against all GBM-populations.

SF3B1 expression is correlated with relevant oncogenic tumor markers in GBM samples
A strong association between SF3B1 expression and key tumor -markers of development/progression (VEGFA/ MKI67/EGFR/CDK4/PDGFRA) was found in GBM (CGGA-and Rembrandt-datasets), but not in the non-tumor samples (Rembrandt-dataset) (Fig. 1l). A robust correlation between SF3B1 expression and the most critical oncogenic spliceosome components [SRSF3/RBM22/PTPB1/RBM3 [23]] was also found in GBM (CGGA-and Rembrandt-datasets), but not in the non-tumor samples (with the exception of PTBP1; Rembrandt-datasets) (Fig. 1m). These data suggest a potential prognostic role of SF3B1 in GBM.

Sf3b1 overexpression is validated in electroporated (EPed)-glioma mouse -models
Sf3b1 overexpression was also corroborated in tumor samples from EPed mouse model vs. control samples Fig. 2 Sf3b1 is overexpressed in different electroporated (EPed)-glioma mouse models. a Generation of mouse models of GBM by plasmid DNA mix injection into the left lateral ventricle following mouse brain electroporation (adapted from [9]). b mRNA expression levels of Sf3b1 and, c ROC-curve analysis of Sf3b1, in the control and tumor samples of the EPed mouse model. d Correlation of Sf3b1 with different key prognostic biomarkers and relevant spliceosome components in GBM samples from these models. Asterisks (*P < 0.05; **P < 0.01) indicate statistically significant differences across different conditions. Plus symbol (+) indicates a tendency between conditions (+P > 0.05 < 0.1) from neural precursors [9] (Fig. 2a-b). ROC-curve analyses also supported the capacity of Sf3b1 levels to discriminate between tumor vs. control samples, showing an AUC of 0.96 (Fig. 2c). Moreover, Sf3b1 expression was also significantly correlated with key glioma/spliceosome -markers (Mki67/Pdgfra/Rbm22/Sf3b1; Fig. 2d).

SF3B1 protein levels are elevated in GBM -samples
Consistent with the mRNA results, IHC analyses of FFPE samples from our patient cohort (Table S2) revealed that nuclear SF3B1 protein levels were significantly elevated in GBM samples vs. non-tumor FFPE -samples (Fig. 3a). This drastic elevation was clearly observed in an available GBM -tissue vs. its non-tumor adjacent -tissue (Fig. 3b). Results were confirmed using CPTAC proteomic-data [19] (n = 100 GBM-samples vs. 10 control-tissues; Fig. 3c and Table S4). Moreover, ROC-curve analyses of SF3B1 protein levels confirmed its capacity to discriminate between GBM vs. control samples, showing an AUC of 0.99 (Fig. 3d). Additionally, we found a significant correlation between SF3B1 and MKI67 in GBM, but not in control tissues (Fig. 3e).

SF3B1 overexpression is associated with poor survival and prognostic in humans/mice and with drug-s resistance in GBM
High SF3B1 mRNA levels were strongly associated with a worse survival rate in GBM patients in our cohort (Fig. 4a), which was corroborated in two additional patient cohorts (Rembrandt-and CGGA-dataset;  Fig. 4b-c). Remarkably, a higher SF3B1 expression was found in human mesenchymal and classical GBM (both GBM -subtypes with poorer -survival) compared to control samples and/or to proneural GBM (GBM subtype with better-survival) in both Rembrandt (Fig. 4d) and CGGA (Fig. 4e) cohorts. Moreover, ROC-curve analyses reinforced the potential prognostic capacity of the SF3B1 overexpression levels to significantly discriminate between classical/mesenchymal -GBM and proneural -GBM in both external patient cohorts [Rembrandt (Fig. 4f ) and CGGA (Fig. 4g)]. Consistently, Sf3b1 expression levels were also elevated in mesenchymal-like -GBM vs. control samples from neural precursors from the EPed mouse model (Fig. 4h), being its expression in mesenchymal-like GBM also higher than in proneurallike GBM but this latter difference did not reach statistical significance (Fig. 4h). Additionally, the Genomics of Drug Sensitivity in Cancer (GDSC) dataset was explored to analyze the potential implication of SF3B1 in pharmacological resistance (Fig. S3a-b). These analyses revealed that the resistance to drugs targeting RTK signaling pathways, chromatin acetylation, DNA replication, cell cycle, and mTOR/PI3K signaling pathways was associated with SF3B1 expression, which unveils the potential implication Comparison of expression levels of SF3B1 and heatmaps generated using SF3B1 levels between control samples and proneural, mesenchymal, and classical GBM subtypes from the Rembrandt (d) and CGGA (e) datasets. ROC-curve analyses of SF3B1 comparing classical/mesenchymal GBM vs. proneural GBM samples in the Rembrandt (f) and CGGA (g) datasets. h SF3B1 expression levels (upper panel) and heatmap (lower panel) discerning between neural precursor cells, proneural and mesenchymal-like tumors from EPed mouse models. Asterisks (*P < 0.05; **P < 0.01; ***P < 0.001) indicate statistically significant differences across different conditions. Plus symbol (+) indicates a tendency between conditions (+P > 0.05 < 0.1) of the dysregulation of SF3B1 in these oncogenic pathways to confer drug resistance in GBM ( Fig. S3c; Tables S8-9).

Pharmacological inhibition of SF3B1 with pladienolide B decreases functional and molecular aggressiveness parameters in vitro in GBM cells
SF3B1 expression levels were significantly higher in U-87/U-118 MG cells compared with non-tumor brain tissues (Fig. S4a) and were slightly higher, but comparable, in U-87/U-118 MG cells and GBM samples, suggesting that both cell lines were appropriate GBM models to study SF3B1 functional role. Subsequently, doseresponse experiments indicated that 100 nM of pladienolide B was the most effective concentration reducing proliferation rate in U-87/U-118 MG cells (Fig. S4b) and in primary-GBM cell cultures (Fig. S4c) after IC 50 determination (Fig. S4d). Therefore, 100 nM-dose was selected for subsequent experiments.
Pharmacological SF3B1 blockade, which disrupts the spliceosome activity (Fig. 5a), significantly decreased proliferation rate in a time-dependent manner in both cell lines (Fig. 5b) and primary-GBM cell cultures (Fig. 5c), but not in primary non-tumor brain cell cultures (Fig. 5d), suggesting that pladienolide B effects are selectively exerted on GBM cells. In this sense, a positive correlation between SF3B1 expression levels in the primary GBM cells cultures and the percentage of reduction of pladienolide B on proliferation rate was found (Fig. S4e-f ). Therefore, we might speculate that pladienolide B is not effective in reducing proliferation rate in non-tumor cells due to the significantly lower expression levels of SF3B1 compared with GBM cells; however, further studies would be necessary to unequivocally corroborate this idea. Furthermore, pladienolide B treatment also reduced the migration rate in U-118 MG cells at 6 h and 24 h (Fig. 5e). Furthermore, a tumorsphere formation assay (used to quantify the proliferation capacity of cancer stem-like progenitor cells) revealed that SF3B1 blockade drastically decreased the number and area of tumorspheres in both cell lines (Fig. 5f ). Moreover, a decrease in VEGF secretion was observed after pladienolide B treatment in both cell lines (Fig. 5g). Capase3/7 luciferase-assay revealed that SF3B1 inhibition induced apoptosis in both cell lines (Fig. 5h), as also confirmed by an increase of cleaved-caspase 3 levels by western blot (Fig. 5i). All these results revealed that pladienolide B treatment affected different critical functional endpoints associated with the development, progression and aggressiveness of GBM cells (Fig. 5j).
Pladienolide B treatment also decreased the expression of key tumor progression markers and critical oncogenic spliceosome components [previously found to be correlated with SF3B1 in GBM samples (Fig. 1lm)] in both cell lines ( Fig. 5k and m, respectively) and primary-GBM cell cultures ( Fig. 5l and n, respectively).

In vivo pharmacological inhibition of SF3B1 with pladienolide B impairs GBM progression and vascularization
Pladienolide B intra-tumor administration in vivo reduced tumor volume and weight compared with control-treated tumors in a preclinical-xenograft U-87 MG GBM model (Fig. 6a-d). Tumor volume clearly showed that GBM progression in vivo was completely stopped in pladienolide B treated tumors vs. controltreated tumors (that rapidly continued their progression; Fig. 6b). Moreover, 2D-micro-CT images together with 3D-rendering confirmed these in vivo differences (Fig. 6e). Furthermore, mitosis number was decreased in pladienolide B treated tumors vs. control-treated tumors (Fig. 6f ). Additionally, pladienolide B treated tumors showed low levels of vascular proliferation (5/6 tumors) and absence of necrosis (all tumors) (Fig. 6g). As previously observed in vitro, pladienolide B administration in vivo significantly decreased various relevant  Fig. 6h-i). Therefore, all these in vivo results (Fig. 6) support the antitumor effects of SF3B1 blockade previously observed in vitro (Fig. 5).

Pre-treatment with pladienolide B in vitro affects the onset/formation of GBM tumors in vivo
Pre-treatment with pladienolide B in vitro for 24 h and 48 h in GBM cells was able to impair GBM onset/formation in an in vivo preclinical-xenograft U-87 MG GBM model (Fig. 7a-b). Specifically, average tumor volume and weight were impaired in vivo in the xenograft U-87 MG GBM model pre-treated with pladienolide B compared with control-treated tumors, being this effect more pronounced in pre-treated cells for 48 h vs. 24 h (Fig. 7b-e). 2D-micro-CT images together with 3D-rendering also confirmed these in vivo differences (Fig. 7f ). Additionally, we demonstrated that GBM cells pre-treated with pladienolide B in vitro were not able to undergo colony formation and tumorsphere formation (Fig. 7g and h, respectively). Altogether, these data demonstrate that treatment with pladienolide B is able to impair the capacity of GBM cells to onset tumor formation in vitro and in vivo.

SF3B1 expression is strongly associated with relevant components of cancer-related pathways in GBM
A specific analysis of highly correlated genes (r > ± 0.800; CGGA-dataset) using STRING-tool and KEGG-database  Table S5), which further supported the relevance of SF3B1 in tumor -physiopathology. Indeed, a more detailed enrichment analysis using Reactome-database was used to identify the main pathways associated with SF3B1 expression, which revealed that the splicing process and AKT-mTOR/ßcatenin signaling pathways were closely associated with SF3B1 (Fig. 8c).
Interestingly, pladienolide B treatment significantly decreased SF3B1 mRNA/protein levels in GBM cells (U-87/U-118 MG cells and primary-GBM cell cultures) and in the preclinical-xenograft GBM model (Fig. S5cd). Moreover, given the reported implication of SRSF1 splicing factor with AKT, mTOR, and Wnt/ß-catenin pathways [22,64,74] (Fig. 9d), we also interrogated the SRSF1-status [which was strongly correlated with SF3B1, Fig. 8 SF3B1 is strongly related to certain cancer-related pathways. a Functional association network of the significantly correlated genes with SF3B1 using the CGGA dataset. These significantly altered genes were analyzed using the STRING database, and (b) they are marked according to their KEGG pathways analysis. c Gene set analysis enrichment terms for the genesets within the Reactome pathways using SF3B1 correlated genes (cut-off r > ± 0.800) (See figure on next page.) Fig. 9 Pharmacological blockade of SF3B1 reveals AKT-mTOR and ß-catenin signaling pathways and BCL2L1 alternative splicing as major drivers of the pladienolide B antitumor effects in GBM. Heatmaps showing the western blot densitometric level (log2) of phosphorylated-proteins (a) and total-proteins levels (b) of several components of AKT/mTOR and ß-catenin pathways in GBM cells (U-87 MG and U-118 MG) after pladienolide B administration. c Images of western blot results showed in the previous heatmaps (a) and (b). d AKT-MTOR and ß-catenin pathways diagram showing the downregulated (in red) and upregulated (in the yellow box) components/processes after pladienolide B administration identified in this work. Expression levels of CCND1 (e) and MYC (f) as endpoints of AKT/mTOR and ß-catenin pathways in GBM cells (U-87 MG and U-118 MG), primary patient-derived GBM cells and in the preclinical-xenograft GBM model after pladienolide B administration. g BCL2L1 splicing variants produced by an alternative 5′ spliced site (A5SS) splicing event and associated with apoptosis and cell death pathway. h BCL2L1-xS/BCL2L1-xL ratio determined by qPCR in GBM cell lines (U-87 MG and U-118 MG), in the preclinical-xenograft GBM model, and in primary patient-derived GBM cells in response to pladienolide B treatment. i BCL2L1-xS/BCL2L1-xL ratio determined by qPCR in primary non-tumor brain cell culture after pladienolide B administration. PSI of BCL2L1 A5SS event in GBM cell lines (U-87 MG and U-118 MG) (j), in primary patient-derived GBM cells (k), and the preclinical-xenograft GBM model (l) in response to pladienolide B treatment. m Validation of designed antisense oligonucleotides (ASOs; ASO1_BCL2L1 and ASO3_BCL2L1) by determination of PSI of BCL2L1 A5SS event in GBM cell lines (U-87 MG and U-118 MG; n = 3). n Proliferation rate in GBM cells in response to control, pladienolide B, ASO1_BCL2L + pladienolide B, and ASO3_BCL2L1 + pladienolide B cells (n = 3). The % has been calculated with the control, ASO1_BCL2L1 and ASO3_BCL2L1 transfected cells (without pladienolide B treatment) of each condition. Asterisks and symbols (*P < 0.05; **P < 0.01; ***/###/ † † † P < 0.001) indicate statistically significant differences across different conditions (i.e.; pladienolide B vs. control; ASO1_BCL2L1+ pladienolide B vs. pladienolide B; ASO2_BCL2L1+ pladienolide B vs. pladienolide B, respectively). Plus symbol (+) indicates a tendency between conditions (+P > 0.05 < 0.1) MTOR, and CCTNB1 expression (Fig. S5e)]. Specifically, SRSF1 mRNA levels were decreased after pladienolide B administration in GBM cells in vitro (U-87/U-118 MG cells and primary-GBM cell -cultures) and the preclinical-xenograft GBM model (Fig. S5f ).

Changes in BCL2L1 splicing variants (SVs) expression profile as a potential driver of SF3B1 blockade antitumor actions
We next explored whether SF3B1 pharmacological blockade, using pladienolide B altered the splicing process of some critical genes implicated in GBM progression which have been previously reported to be associated with cancer-related signaling -pathways (i.e., KLF6/CRK/MST1R/ CASP2/RAC1/MCL1/BIRC5/SPP1/BCL2L1). Specifically, we performed a screening of the SVs of these genes in U-87/U-118 MG cells using end-point PCR-methodology (data not shown). Among them, only BCL2L1 showed an alteration in the SV-profile (Fig. 9g-l). More specifically, BCL2L1 has nine SVs, and two of them, Bcl-xL and Bcl-xS, are commonly reported to be associated with cancer [6,10], being Bcl-xL an anti-apoptotic and oncogenic, while Bcl-xS acts as a pro-apoptotic tumor suppressor (Fig. 9g). We studied the balance of these two BCL2L1variants, their regulation by PSI (Percent Spliced In index) analysis, and/or their presence by RT-qPCR. This latter analysis revealed that Bcl-xS/Bcl-xL ratio was elevated after SF3B1 inhibition with pladienolide B in GBM in vivo (i.e., preclinical-xenograft GBM model) and GBM cells in vitro (U-87/U-118 MG cells and primary-GBM cell cultures) (Fig. 9h), but not in an available primary non-tumor cell cultures (Fig. 9i). Particularly, anti-apoptotic Bcl-xL was significantly downregulated while pro-apoptotic Bcl-xS was upregulated after pladienolide B treatment in the preclinical-xenograft GBM model (Fig. S5g), GBM cells [U-87/U-118 MG cells (Fig. S5h) and primary-GBM cell cultures (Fig. S5i)], but not in primary non-tumor cell cultures (Fig. S5j). Likewise, PSI determination confirmed previous results since pladienolide B treatment reduced BCL2L1 alternative 5′ end splice site (A5SS) splicing event in U-87/U-118 MG cells (Fig. 9j), primary-GBM cell -cultures (Fig. 9k), and in the preclinical-xenograft GBM model (Fig. 9l). Additionally, the ability of SF3B1 to influence BCL2L1splicing was further substantiated by the fact that PSI values were directly correlated with SF3B1 expression in U-87/U-118 MG cells (Fig. S5k).
In order to corroborate that the alternative splicing dysregulation of BCL2L1 is a potential driver of pladienolide B-mediated antitumor effects, we designed, validated, and used different ASOs that might be able to revert the Bcl-xS/Bcl-xL splicing process observed in response to pladienolide B. First, we performed an initial screening using U-87 MG cells to optimize the ASOs transfection and to identify which of the 5 designed ASOs were efficient in inhibiting the pro-apoptotic Bcl-xS variant and in promoting the anti-apoptotic Bcl-xL variant in response to pladienolide B treatment. Among the 5 designed ASOs, only two (ASO1_BCL2L1 and ASO3_BCL2L1) were able to efficiently revert the splicing process of BCL2L1 after pladienolide B treatment (Fig. 9m). Specifically, as previously observed, pladienolide B treatment significantly downregulated the anti-apoptotic Bcl-xL variant while upregulated the proapoptotic Bcl-xS variant in intact (non-transfected) GBM cells (Fig. 9m). In contrast, the pro-apoptotic Bcl-xS variant was significantly inhibited while the anti-apoptotic Bcl-xL variant was upregulated in ASO-transfected GBM cells (ASO1_BCL2L1 and ASO3_BCL2L1) treated with pladienolide B (Fig. 9m). Similar results were then confirmed in both cell lines (U-87 MG and U-118 MG; n = 3) transfected with ASO1_BCL2L1 and ASO3_BCL2L1 and treated with pladienolide B (Fig. 9m). Therefore, these results revealed that the transfection with both ASOs in GBM cells treated with pladienolide B was able to revert the splicing process of BCL2L1 to the same level as control-treated cells. Then, we tested if this ASO-mediated inhibition of the BCL2L1 pro-apoptotic variant following pladienolide B treatment was able to reduce the antitumor effect of pladienolide B on GBM growth using a proliferation assay. Specifically, the results uncovered that the proliferation rate of ASO-transfected GBM cells (ASO1_BCL2L1 and ASO3_BCL2L1) in response to pladienolide B treatment was significantly blunted compared with non-transfected cells treated with pladienolide B (Fig. 9n). Particularly, ASO1_BCL2L1 seems to be more effective than ASO3_BCL2L1 in impairing the antitumor effect of pladienolide B which might be explained by the significant low PSI observed in the cells transfected by ASO3_BCL2L1 vs. ASO1_BCL2L1 (Fig. 9m). Altogether, these data suggest that alternative splicing dysregulation of BCL2L1 seems to be a potential driver of pladienolide B-mediated antitumor effects.

Discussion
Targeting the spliceosome machinery could become an innovative and successful therapeutic approach to treat incurable cancers like GBM. Indeed, the transcriptomic landscape of cancer cells makes them particularly vulnerable to pharmacological inhibition of splicing, which might have important therapeutic relevance in the near future as suggested by multiple ongoing clinical trials aimed to answer this question [7]. Specifically, various drugs have been designed to target SF3B1 (a central/essential core-component of the spliceosome) [16,35,48], making this spliceosome element the best candidate to study its translational oncogenic implication and therapeutic capacity in cancers wherein there are no successful treatments or cure. However, the data published so far focused on the potential oncogenic role and therapeutic effectiveness of the modulation of critical spliceosome components through pharmacological approaches is quite limited, fragmentary, and unclear [7,48,52]. To the best of our knowledge, the oncogenic implication and therapeutic capacity of SF3B1, its somatic mutations, and expression profile have not been characterized in GBM, neither its association with molecular features nor clinical parameters.
Herein, we demonstrated that SF3B1 dysregulation clearly affects several cancer hallmarks including apoptosis/proliferation/migration/angiogenesis/splicing pattern and signaling among others, and, of particular clinical relevance, that it could be associated with the development of drug resistance. Since splicing perturbations are common in cancer, including brain tumors [23], and are associated with mutations and/or altered expression of splicing machinery [23,34,70], we determined the SF3B1mut-frequency and whether these mutations were associated with glioma progression. Interestingly, SF3B1mut-frequency was low in glioma patients (~ 1%) compared to other cancer pathologies [wherein SF3B-1mut range from 5% in breast cancer to 81% in myelodysplastic syndromes [7,36,69]]. Moreover, no difference was observed in mean OS of glioma patients with SF3B-1mut compared to SF3B1wt, an observation that is not similar to previous data in chronic lymphocytic leukemia indicating that SF3B1mut are associated with rapid disease progression and unfavorable OS [69]. The low SF3B-1mut-frequency found in glioma patients might suggest that the potential SF3B1role in glioma pathogenesis could be exerted through altered expression levels rather than somatic mutations. In fact, we demonstrate for the first time a drastic SF3B1 overexpression (at mRNA/ protein levels) in different cohorts of GBMs vs. nontumor tissues, which was also confirmed in EPed-glioma mouse models vs. control samples. Moreover, bioinformatic analyses revealed a potential diagnostic capacity of SF3B1 levels to discriminate between GBM/gliomas vs. control tissues from humans and mice, suggesting that GBM/glioma curse with a global dysregulation of SF3B1 in different species. Furthermore, our data revealed a potential utility of SF3B1 as aggressiveness biomarker in GBM which is supported by the direct and strong association found between SF3B1 expression levels and relevant development/progression tumor -markers (e.g., MKI67/PDGFRA) [59] and different oncogenic spliceosome components, including SRSF3 (the most critical splicing machinery component in GBM recently identified by our group) [23], in human GBM and tumor -samples from EPed-glioma mouse models.
Most importantly, this study revealed that high SF3B1 expression is directly associated with a worse OS rate in GBM patients, certainly, the main clinical problem in this pathology. This finding was corroborated in two external patient cohorts with GBM (Rembrandt/ CGGA-dataset) and further supported by similar observations found in other tumor pathologies [3,4,32,38,49,63]. Similarly, a higher SF3B1 expression was observed in human and mouse classical and mesenchymal GBM (subtypes with poorer survival rate) compared to proneural GBM (subtype with better survival rate) or non-tumor samples, which reinforced the prognostic value and potential oncogenic role of SF3B1 [68]. To the best of our knowledge, this is the first report identifying the diagnostic and prognostic capacity of SF3B1 in human GBM, and in glioma mouse models with different prognoses, wherein these observations suggest a causal link between SF3B1 dysregulation and GBM aggressiveness. Notably, we also characterized SF3B1 expression at the single-cell level demonstrating SF3B1 was homogeneously expressed across all GBM cell populations/states, being higher in cells expressing a proliferative neural progenitors-like transcriptional program. These data are therapeutically important and have a potential translational/oncogenic implication since the current therapeutic strategies for GBM are not efficient at reducing tumor volume/ growth or augmenting survival rate, which is likely due, in part, to the resistance acquired by tumors, particularly by neural progenitors-cells, to different current drugs [51]. Therefore, our data showing that SF3B1, a druggable spliceosome component, is homogeneously overexpressed in all GBM cell populations/states offer a novel opportunity and therapeutic approach to treat GBM. Remarkably, GDSC-dataset analysis also unveiled a potential implication of SF3B1 dysregulation in different oncogenic pathways (e.g., mTOR-PI3K/cell cycle/DNA replication, etc.) to confer drug resistance in GBM, which further encourages the use of an SF3B1 specific-inhibitor in GBM.
Indeed, we demonstrate strong in vitro/in vivo antitumor actions of pladienolide B in GBM cells. Notably, SF3B1 blockade induced marked reductions in aggressiveness features of different GBM cell models [cell -lines and primary-GBM cell -cultures, i.e., inhibition of proliferation/migration/VEGF secretion, and increase of apoptosis]. Most notably, SF3B1 blockade strikingly decreased also GBM-stem/progenitor cells in terms of tumorspheres number and area, both relevant functional results that may help to explore the GBM onset and how to overcome the well-known GBM -resistance to different/current drugs [21,51]. It should be emphasized that our data also suggest that pladienolide B effects selectively impact on GBM cells and not non-tumor brain cells, which is clinically relevant and agrees with previous data in other cancer-types, where splicing inhibitors exert stronger, more selective actions on cancer cells than on non-transformed cells [14]. Moreover, we demonstrate that SF3B1 is also an effective target in GBM in vivo since pladienolide B treatment effectively blocks GBM progression of already established GBM tumors and the GBM onset/formation in preclinical GBM mouse models. Indeed, pladienolide B clearly blunted tumor volume compared to control tumors (which drastically continued their progression), and markedly decreased tumor-weight/mitosis-number of GBM cells and vascularization and necrosis in vivo. Furthermore, pharmacological SF3B1 blockade also decreased the expression of key tumor progression markers and critical oncogenic spliceosome components in GBM cells in vitro and in vivo. Remarkably, pre-treatment with pladienolide B in vitro (24/48 h) was also capable to impair the in vivo onset/formation of GBM possibly through disruption of GBM stemcell survival. Thus, all these robust in vitro/in vivo results, together with the extended OS observed in different human cohorts, unveiled an important pathophysiological role of SF3B1 in GBM. Although some aspects should be considered when using pladienolide B (e.g., specific concentration used, possible side effects in patients, etc), our data suggest that SF3B1 blockade could be a novel therapeutic avenue with relevant pathophysiological/clinicalpotential to combat this devastating disease.
We also interrogated the signaling mechanisms underlying the antitumor actions of SF3B1 blockadge in GBM in vitro and in vivo. Our data revealed, for the first time in GBM, a striking alteration in relevant routes closely associated with GBM progression and initiation, especially the AKT-mTOR and ß-catenin signaling pathways [12,37,46,56], in response to SF3B1 blockade. In support of the link between SF3B1 activity and AKT pathway, it has been recently reported that SF3B1 K700E mutation can modulate the expression of key components of the AKT pathway with resulting increases in the migration/invasion of breast cancer cells [39]. Specifically, we observed an overall downregulation in several critical points belonging to these pathways [i.e., total-protein -levels of AKT/MTOR/S6K1/CTNNB1/ TP53/HIF1A; phosphorylated-protein levels of AKT/ MTOR/S6K1/PDK1 and expression levels of CCND1 and MYC) and an upregulation of phosphorylated-TP53 levels, in different GBM models (in vitro and/or in vivo) in response to SF3B1 blockade. Moreover, our data indicate that pladienolide B inhibitory actions observed in AKT-mTOR/ß-catenin signaling pathways may likely be exerted through a significant down-regulation in SRSF1-levels, a relevant pro-oncogene overexpressed in GBM [1,13,23,71,74] which acts activating both signaling pathways simultaneously [22,64]. This idea is further supported by the fact that SRSF1 and SF3B1 are functionally connected since SRSF1 directly interacts with the U2-snRNP complex where SF3B1 takes part [15], and by our data indicating that SRSF1 expression is strongly correlated with SF3B1, MTOR, and CCTNB1 expression in GBM. Therefore, these data provide original, compelling evidence that SF3B1 is functionally linked, likely via SRSF1 modulation, to these well-known relevant pro-oncogenic pathways (AKT-mTOR/ßcatenin) in GBM, which further supports the pathophysiological relevance of SF3B1 and the antitumor actions of SF3B1-blockade in GBM. Interestingly, SF3B1 blockade suppressed SF3B1 -expression suggesting positive feedback that could enhance its antitumor effects. SF3B1 blockade also exerted important molecular actions involving the splicing modulation of two clinically relevant SVs of BCL2L1 (Bcl-xL and Bcl-xS) associated with cancer -development and known to play an oncogenic role and a tumor suppressor actions, respectively [6,10]. Specifically, SF3B1-blockade downregulated anti-apoptotic Bcl-xL, while upregulated the pro-apoptotic Bcl-xS, in GBM, both in vitro and in vivo, but not in non-tumor brain cell cultures. This idea was further corroborated by PSI analysis demonstrating a pladienolide B-induced reduction of BCL2L1 A5SS splicing -event in GBM in vitro and in vivo. In line with these data, previous reports found that several apoptosisregulatory genes, including BCL2-related genes, generate alternatively SVs with opposite activities, which is a biological program often employed by cancer -cells to escape from intrinsically programmed cell death and radiotherapy/chemotherapy-induced cytotoxicity [72]. In fact, our observations in Bcl-xL/xS, together with the data demonstrating the implication of SF3B1 dysregulation in different oncogenic -pathways that confers drug resistance in GBM (e.g., mTOR-PI3K/ cell cycle/DNA -replication, etc.), might be clinically relevant because it has been demonstrated that Bcl-xL is transcriptionally upregulated and associated with poor prognosis and chemoresistance in many cancers [6,10]. In this sense, it should be indicated that the use of two different ASOs that inhibited the pro-apoptotic Bcl-xS variant and promoted the anti-apoptotic Bcl-xL variant in response to pladienolide B treatment was able to significantly reduce the antitumor effect of pladienolide B on GBM cells. All these data demonstrate that changes in the splicing of BCL2L1 seem to be one of the main molecular mechanisms underlying the link between SF3B1 blockade and the significant decrease in GBM onset, GBM progression, and aggressiveness features observed in response to pladienolide B treatment.

Conclusions
Taken our evidences together, our results unveiled new conceptual and functional avenues in GBM, with potential clinical implications, by demonstrating that SF3B1 is an attractive therapeutic target in GBM since its inhibition impaired key pathophysiological processes in GBM -biology (i.e., proliferation/migration/tumorspheres formation/apoptosis, etc.) likely by modulating different oncogenic signaling pathways (AKT-mTOR/ß-catenin) associated with GBM survival/initiation/progression, and an imbalance of BCL2L1 splicing. Moreover, we found that SF3B1 overexpression in GBM is associated with key molecular and clinical features including overall survival, poor prognosis, and drug resistance. Therefore, these results point out SF3B1 as a potential diagnostic/ prognostic biomarker and a promising pharmacological target to treat patients with GBM, offering a clinically relevant opportunity that should be tested for use in humans. pathway in SF3B1 high-expression cell lines using the Genomics of Drug Sensitivity in Cancer dataset. Table S9. Drug sensitive hits together with its corresponding annotated targeted pathway in SF3B1 low-expression cell lines using the Genomics of Drug Sensitivity in Cancer dataset.