Mutations in the RNA Splicing Factor SF3B1 Promote Tumorigenesis through MYC Stabilization

Mutations in genes encoding RNA splicing factors occur across a variety of hematologic malignancies and solid tumors, and have provided evidence for the importance of aberrant splicing to tumorigenesis (1). Mutations in the gene encoding the core RNA splicing factor SF3B1 are the most common across cancer types and occur as “hotspot” heterozygous point mutations. SF3B1 hotspot mutations are among the most common genetic alterations in patients with myelodysplastic syndromes (MDS; refs. 2, 3), chronic lymphocytic leukemia (CLL; refs. 4, 5), and uveal melanoma (UVM; refs. 6–8), and are also seen in 1% to 3% of many solid tumors, including breast-invasive carcinoma (BRCA) and skin cutaneous melanoma (SKCM; ref. 1). Analysis of the effects of SF3B1 mutations on RNA splicing has consistently identified that these mutations promote usage of aberrant branchpoint residues, which most commonly manifests in transcripts bearing aberrant intron-proximal 3′ splice site (3′ss; refs. 9, 10). To date, however, very few individual misspliced transcripts that functionally link mutant SF3B1 to malignant transformation have been identified. In addition, multiple distinct residues of SF3B1 are affected by hotspot mutations, and many of these mutations exhibit cancer lineage specificity. For example, mutations at the R625 residue of SF3B1 are strongly associated with uveal, acral, and mucosal melanomas (6, 8), whereas mutations at position K700 are present across myeloid leukemias, CLL, and epithelial cancers (11). However, mechanistic differences between distinct hotspot mutations in SF3B1 are not well defined, nor is the basis for the cancer specificity of SF3B1 mutant alleles understood.

It is expected that functional changes associated with SF3B1 mutations are the result of aberrant splicing events in key genes leading to deregulation in activity of regulatory networks. For example, recent work identified robust aberrant splicing of the mRNA encoding the bromodomain protein BRD9 across cancer types and across individual hotspot mutations in SF3B1 (12). However, the consequences of SF3B1 mutations on dysregulated gene expression, which poorly overlap with the mRNAs misspliced by mutant SF3B1 (9, 10), are not well defined. To this end, here we performed an analysis of the impact of mutant SF3B1 on the activity of gene-regulatory networks, which may be distinct from aberrant splicing changes associated with mutant SF3B1. In so doing, we elucidate the effects of SF3B1 hotspot mutations across cancer lineages at the level of both mRNA splicing and expression. This effort identified a striking effect of SF3B1 mutations on post-translational regulation of multiple proteins with very well-established roles in tumorigenesis via inactivation of a specific regulatory subunit of the protein phosphatase 2A (PP2A) complex. The partial inactivation of PP2A by aberrant splicing of a regulatory subunit illuminated a potential therapeutic strategy for SF3B1-mutant malignancies via reactivation of residual PP2A activity.

To provide a more complete understanding of how SF3B1 mutations alter splicing globally, we performed a comprehensive pan-cancer analysis using computational approaches that enable the identification and quantification of alternative 3′ss with enhanced sensitivity. To this end, we collected RNA sequencing (RNA-seq) data from 98 tumors and 12 isogenic cell lines harboring SF3B1 hotspot mutations in the region encoding the C-terminal HEAT repeat domains together with 102 randomly chosen SF3B1 wild-type (WT) controls (Fig. 1A; Supplementary Table S1; Methods). Hotspot mutations mostly clustered into two specific residues within SF3B1′s HEAT repeat domains (Fig. 1A). The most frequent hotspot mutation, K700E, was the most common mutation in CLL, MDS, acute myeloid leukemia (AML), and BRCA, whereas mutations at position R625 were mostly restricted to UVM and SKCM. All the E902K/G variants were exclusively observed in bladder urothelial carcinoma (BLCA), whereas K666 mutations were present across multiple tumor types. The association of specific SF3B1 mutational hotspots with distinct tumor types suggests the potential for allele-specific splicing abnormalities and unique pathogenic roles of each SF3B1 mutational hotspot within each lineage of cancer. To examine this possibility, unsupervised hierarchical clustering was applied to the identified aberrant 3′ss (Fig. 1B; Supplementary Table S2). Unsupervised clustering clearly separated SF3B1 mutant from SF3B1 WT samples, with the exception of all The Cancer Genome Atlas (TCGA) BLCA (E902K/G) as well as one sarcoma case (TCGA SARC, D894H), suggesting that these specific mutations do not lead to aberrant 3′ss usage. In addition, the clustering dendrogram substructure revealed a difference in aberrant 3′ss usage between hematologic malignancies and solid tumors, as well as obvious tumor lineage specificity (Fig. 1B; Supplementary Fig. S1A; χ² test; P value = 4.33e-51). For instance, patients with SF3B1-mutated CLL clustered together despite coming from three independent cohorts and further grouped with another cluster consisting of AML, MDS, K562 cells, and Nalm-6 cells, indicating greater similarity in aberrant 3′ss usage among hematologic malignancies. Meanwhile, UVM and SKCM formed a stable subgroup, exhibiting a different splicing pattern from TCGA BRCA patients. At the same time, unsupervised clustering also demonstrated a distinct preference for cryptic 3′ss associated with each mutational hotspot (Supplementary Fig. S1A; χ² test; P value = 1.40e-16). Consistent with our previous observation, clustering by rows demonstrated that splicing events split into two groups, mainly dominated by events associated with K700E (CLL and others) and R625 (UVM and SKCM) mutations. In addition, we applied a principal component analysis to the same data matrix of aberrant 3′ss usage, which indicated a more distinct separation of samples harboring mutations affecting the K700 and R625 residues of SF3B1 (Supplementary Fig. S1B).

Most cryptic 3′ss were associated with a modest change in Percent-Spliced-In (PSI) between SF3B1-mutant and WT tumors, usually less than 10% (10). As previously reported, the distance from cryptic 3′ss to canonical site showed a striking peak around 10 to 30 bp upstream exon (ref. 9; Supplementary Fig. S1C). Further evaluation of cryptic 3′ss usage specific to each mutational hotspot suggested that the distribution of distances from the canonical to the cryptic 3′ss varies among the different hotspot mutations with altered sequence motif associated with the aberrant 3′ss (Fig. 1C; Supplementary Fig. S1D and S1E). For example, SF3B1^R625 mutations seem to be associated with a shorter (<25 bp) and more specific insertion, whereas K666N induces usage of longer inclusions. Accordingly, the sequence contexts associated with the cryptic 3′ss exclusive to SF3B1^R625 are characterized by an upstream adenine enrichment, which was not observed in SF3B1^K666N-specific cryptic 3′ss (Supplementary Fig. S1D). We next compared the cryptic 3′ss with differential usage between SF3B1^K700E (MDS, BRCA, and CLL) and SF3B1^R625 (UVM and SKCM; Fig. 1D; Supplementary Methods). This analysis identified a list of cryptic 3′ss events that are probably specific to different SF3B1 mutational hotspots, suggesting that K700E and R625 positions may have a differential preference of 3′ss usage.

We extended the above pan-cancer analyses of the effects of mutant splicing factors to established hotspot mutations in SRSF2 and U2AF1. We identified 52 SRSF2-mutant samples (including 29 SRSF2^P95-mutant samples), as well as 59 U2AF1-mutant samples (5 U2AF1^Q157-mutant and 32 U2AF1^S34F-mutant samples; Supplementary Fig. S2A; Supplementary Table S3). Analysis of the effects of these mutations on splicing identified that most of the alternative splicing events in SF3B1-mutant samples were classified into alternative 3′ss, whereas the predominant events in SRSF2- or U2AF1-mutant samples were cassette exons (Supplementary Fig. S2B), which suggested the uniqueness of mutant SF3B1′s role in promoting cryptic 3′ss usage. Next, unsupervised hierarchical clustering was performed on all SF3B1-, SRSF2-, and U2AF1-mutant samples by combining the previously identified significantly differential splicing events (Supplementary Fig. S2C). Consistent with our previous observation, SF3B1 hotspot mutations were separated from SRSF2 and U2AF1 with the exception of very few R625 mutants (UVM). In contrast, SRSF2- and U2AF1-mutant samples had less distinguishable splicing patterns from one another. Overall, these pan-cancer analyses suggested the singularity of the aberrant splicing pattern in SF3B1 mutants.

Given the splicing specificity of SF3B1 hotspot mutations, we performed gene set enrichment analysis (GSEA) against curated MSigDB gene sets to identify tumor- and SF3B1 mutation–specific cellular processes. When we compared all the SF3B1-mutant and WT samples across all 12 tumor types, significantly enriched gene sets included those involved in nonsense-mediated decay (NMD; core enriched genes: a set of ribosomal genes), DNA repair (POLL, ERCC8, POLB, etc.), and RNA polymerase II initiation (CDK7, TAF9, TAF12, etc.; Fig. 1E; Supplementary Table S4). We also identified differential enrichment of signaling pathways depending on SF3B1 mutation and tumor types, including activation of NOTCH signaling in SF3B1-mutant CLL (13) and heme metabolism and NF-κB signaling (core enriched genes: proteasome genes) in patients with SF3B1-mutant MDS (Supplementary Table S4; refs. 14, 15). In addition, activation of multiple oncogenic pathways such as the ERK/MAPK, mTOR, and apoptosis-related pathways was suggested in SF3B1-mutant tumorigenesis (Supplementary Table S4). In contrast, we identified Gene Ontology terms (enriched with misspliced genes) associated with RNA processing, cellular metabolism, and organophosphate biosynthesis in SRSF2-mutant samples, and macromolecule catabolism, cellular metabolism, establishment of protein localization, and metabolism in U2AF1-mutant samples (Supplementary Fig. S3A–S3C).

To understand the functional role of mutant SF3B1 in promoting tumorigenesis from the view of transcriptional and post-translational interaction networks, we adopted a well-assembled transcriptional molecular interaction network developed in human B cells (16) and performed a regulatory network analysis on patients with SF3B1-mutated CLL. In total, 721 differentially expressed genes between SF3B1-mutant and WT samples from the International Cancer Genome Consortium (ICGC) cohort (Supplementary Table S5) were used as input to test the enrichment of each transcription factor (TF) network target. When taking into account the direction of transcriptional changes, the top master regulator and the only statistically significant TF was MYC (hypergeometric test, q value = 1.76 × 10⁻⁷ for positive regulation, q value = 4.12 × 10⁻³ for negative regulation; Fig. 2A; Supplementary Methods). This direction-specific association was validated from an independent dataset of 294 patients with CLL from the ICGC CLLE-ES project (Fig. 2B). We further confirmed this using a preranked GSEA as well as unsupervised clustering (Fig. 2C and D). When repeating the analysis on additional cancer types including UVM, SKCM, and BRCA, similar enrichment of the MYC transcriptional program was detected in TCGA BRCA samples with SF3B1^K700E mutations, whereas it was absent in UVM and SKCM samples with SF3B1^R625 mutations, or AML, lung adenocarcinoma, pancreatic adenocarcinoma, and uterine corpus endometrial carcinoma samples with SRSF2/U2AF1 hotspot mutations (Supplementary Table S6). Taken together, these observations suggested that tumors harboring SF3B1^K700E mutations (and potentially mutations affecting H662, K666, and G742) activate the MYC transcriptional program.

We next sought to verify the effects of MYC activation by mutant SF3B1 in the B-cell lineage in vivo. To achieve this, we crossed Cd19-cre^+/− mice with conditional Sf3b1^K700E/+ knock-in mice (17) to generate Cd19-cre^+/− Sf3b1^K700E/+ mice. Detailed analysis revealed that there are no robust changes in spleen weight, blood cell counts, lineage commitment, and B-cell maturation in Sf3b1^K700E/+ mice compared with WT counterparts (Supplementary Fig. S4A–S4M). Next, Cd19-cre^+/− Sf3b1^K700E/+ mice were crossed to mice transgenic for c-Myc driven by the IgH enhancer (Eμ-Myc mice; ref. 18) to generate control (Cd19-cre^+/−), Sf3b1^K700E-mutant (Cd19-cre^+/− Sf3b1^K700E/+), Eμ-Myc (Cd19-cre^+/− Eμ-Myc^Tg/+), and Sf3b1^K700E Eμ-Myc double-mutant (Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+) mice (Supplementary Fig. S5A). Although Cd19-cre^+/− Sf3b1^K700E/+, Cd19-cre^+/− Eμ-Myc^Tg/+, and Cd19-cre^+/− control mice did not develop B-cell malignancies within 300 days, 4 of 8 Sf3b1^K700E/+ Eμ-Myc^Tg/+ double-mutant mice developed B-cell leukemia/lymphoma (Fig. 3A; Supplementary Fig. S5B). In serial transplantation into sublethally irradiated recipients, mice transplanted with Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+ double-mutant cells developed severe anemia, thrombocytopenia, and splenomegaly, resulting in significantly shorter survival compared with single-mutant or control mice (Fig. 3B–H). These data provide the first evidence that SF3B1 mutations contribute to tumorigenesis in vivo.

To understand the molecular mechanisms for MYC activation across SF3B1-mutant cells and our mouse model, we next analyzed RNA-seq data from patients with CLL, isogenic Nalm-6 cells, and sorted splenic B cells from recipient mice (Supplementary Fig. S5C–S5E). This revealed a significant overlap in aberrant 3′ss events in SF3B1-mutant CLL samples, Nalm-6 cells, and murine B cells (Fig. 4A–D; Supplementary Fig. S5F and S5G; Supplementary Tables S7–S9). Most of these aberrant 3′ss were validated by RT-PCR in both samples from patients with CLL and isogenic Nalm-6 cells and resulted in decreased protein levels (Fig. 4E and F; Supplementary Fig. S6A–S6F). Given that MYC activation was observed specifically in SF3B1-mutant CLL and BRCA where mutations predominantly affect the K700 residue, we hypothesized that aberrant 3′ss usage specific to SF3B1^K700E-mutant cancers might be responsible for MYC activation. Interestingly, mRNA expression levels of MYC were not elevated in SF3B1-mutant CLL, BRCA, or Nalm-6 cells compared with SF3B1^WT counterparts, and SF3B1-mutant samples did not harbor MYC amplification (Supplementary Fig. S7A–S7E). These observations led us to hypothesize that MYC activation in SF3B1^K700E-mutant cancers might occur via post-transcriptional and/or post-translational mechanisms. Among genes whose product has previously been shown to result in increased stability of MYC protein (19–21), PPP2R5A was the only gene whose aberrant splicing events were specifically observed in SF3B1^K700E-, SF3B1^H662D-, and SF3B1^K666N-mutant cells and were validated in human and mouse SF3B1-mutant cells (Fig. 4F; Supplementary Fig. S6D–S6F). In addition, mis-splicing in PPP2R5A was more robust in SF3B1^K700E-mutant CLL compared with that in SF3B1^R625-mutant UVM (Supplementary Fig. S7F). These results led us to focus on the roles of PPP2R5A loss in MYC activation in more detail.

PPP2R5A is a regulatory B subunit of the major serine/threonine PP2A protein complex (which consists of a core trimer between the scaffolding A subunit, catalytic C subunit, and a regulatory B subunit; reviewed in ref. 22). PP2A complex containing PPP2R5A dephosphorylates serine 62 (S62) of MYC, resulting in an unstable, singly threonine 58 (T58)–phosphorylated form of MYC that is a substrate for ubiquitination by FBXW7 and degradation by the 26S proteasome (19). As expected, aberrant PPP2R5A transcript was confirmed to result in NMD (Fig. 4G and H), and a decrease in PPP2R5A mRNA and protein expression was observed, which was associated with MYC upregulation in SF3B1^K700E-mutant CLL, Nalm-6 cells, and mouse splenic B cells (Fig. 4I–K; Supplementary Fig. S7G–S7J; Supplementary Tables S10 and S11). Among genes encoding regulatory B subunit family members, PPP2R5A was specifically misspliced and downregulated in SF3B1-mutant tumors compared with WT counterparts across cancers (Fig. 4L; Supplementary Fig. S8A–S8G). The half-life of PPP2R5A protein was also diminished in SF3B1-mutant cells (Supplementary Fig. S8H), although the mechanism for this is unknown. Consistent with the dramatic increase of the S62-phosphorylated form of MYC in SF3B1-mutant cells compared with that in WT cells (Fig. 4L; Supplementary Fig. S8I and S8J), SF3B1^K700E expression significantly decreased the rate of MYC protein degradation following inhibition of protein synthesis with cycloheximide (Fig. 4M). In contrast, there was no difference in MYC protein synthesis rate between SF3B1 WT and mutant cells (Supplementary Fig. S8K). Importantly, restoration of PPP2R5A expression in SF3B1-mutant cells almost completely reduced phosphorylation levels of MYC at S62, leading to protein degradation (Fig. 4N). To further evaluate the role of MYC phosphorylation, we utilized S62-mutant forms of MYC that either mimic (S762D) or prevent (S62A) MYC S62 phosphorylation. When we overexpressed HA-tagged MYC WT, S62A, or S62D in SF3B1^WT and SF3B1^K700E cells, S62A mutations significantly shortened MYC half-life, whereas S62D mutations prolonged MYC half-life in both SF3B1^WT and SF3B1^K700E cells. These data are consistent with our observations that the differential status of S62 phosphorylation of MYC between SF3B1^WT and SF3B1^K700E cells altered MYC protein stability (Supplementary Fig. S8L).

We were further interested in potential BCL2 involvement in the context of MYC activation in mutant SF3B1–mediated tumorigenesis program because (i) the serine 70 (S70)–phosphorylated form of BCL2 is another target of PPP2R5A (23), (ii) phosphorylation at S70 is required for BCL2′s antiapoptotic function (24), and (iii) BCL2 has been shown to play a role in blocking MYC-induced apoptosis (25, 26). We first verified that the S70-phosphorylated BCL2 is increased in SF3B1-mutant cells compared with WT cells, which was canceled by PPP2R5A restoration (Fig. 4L and N). Intriguingly, mRNA expression of proapoptotic BH3-only BCL2 family genes such as BAD, BIK, BMF, and BBC3 (PUMA) was decreased in SF3B1-mutant Nalm-6 cells (Fig. 5A–E; Supplementary Table S10), suggesting that S70-phosphorylated BCL2–mediated antiapoptotic function coupled with transcriptional downregulation of proapoptotic BCL2 family might contribute to the antiapoptotic effect of MYC. Consistent with this, SF3B1-mutant cells were more resistant to apoptotic stimulus, such as that induced by a topoisomerase inhibitor, camptothecin (CPT; Fig. 5F–H). To evaluate the role of BCL2 phosphorylation, we utilized S70-mutant forms of BCL2 that either mimic (S70E) or prevent (S70A) BCL2 S70 phosphorylation (23). Expression of BCL2 S70E cDNA significantly protected SF3B1^WT cells from CPT-induced apoptosis and enhanced cell growth, whereas BCL2 S70A expression had minimal effects compared with BCL2 WT. On the other hand, expression of BCL2 S70A induced more apoptosis upon treatment with CPT in SF3B1^K700E cells, resulting in inhibition of cell growth (Fig. 5I; Supplementary Fig. S9A–S9C). Next, we investigated how BCL2 phosphorylation enhances antiapoptotic functions. Protein half-lives of BCL2 WT, S70A, and S70E were almost identical, suggesting that BCL2 phosphorylation does not affect protein stability of BCL2 (Supplementary Fig. S9D). As BCL2 is known to bind BAK and antagonize the effects of BAK (which is required for mitochondrial permeabilization during apoptosis), we compared the binding affinity of either WT, S70A-mutant, or S70E-mutant BCL2 protein with BAK by immunoprecipitation assays following treatment with CPT for 5 hours. This revealed that BCL2 S70E had higher binding affinity to BAK, whereas BCL2 S70A had reduced affinity compared with BCL2 WT protein. These data suggest that BCL2 phosphorylation status at S70 regulates binding with BAK (Fig. 5J). Levels of cytochrome C release in the isogenic cell lines above were analyzed by intracellular flow cytometry following treatment with CPT for 5 hours. Most likely as a result of BCL2 activation, cells with BCL2 S70A expression demonstrated a significant increase in cytochrome C release compared with cells with BCL2 WT or S70E expression, which is consistent with our previous data that expression of BCL2 S70E cDNA significantly protected cells from CPT-induced apoptosis and enhanced cell growth, whereas BCL2 S70A expression had minimal effects compared with BCL2 WT (Fig. 5K). Overall, these findings suggest that the mutant SF3B1–PPP2R5A axis regulates post-translational modification of BCL2 at S70 and plays an important role in the antiapoptotic phenotype of SF3B1-mutant cells.

Given that numerous missplicing events occur in the setting of mutant SF3B1, we next sought to evaluate the link among mutations in SF3B1, PPPR5A missplicing and expression, and MYC activity in more detail. We first evaluated the effects of restoring PPP2R5A levels in Sf3b1^K700E Eμ-Myc double-mutant tumors. We expressed an empty vector or Ppp2r5a cDNA in Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+ tumors and transplanted these cells into recipients (Fig. 6A). Recipients transplanted with Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+ cells with restored PPP2R5A expression had lower white blood cell (WBC) counts and less severe thrombocytopenia, splenomegaly, and tumor cell infiltration in multiple organs compared with control mice (Fig. 6B–F). In addition, serial transplantation of Sf3b1^K700E/+ Eμ-Myc^Tg/+ tumor cells revealed that PPP2R5A overexpression significantly prolonged survival (Fig. 6G). Conversely, depletion of PPP2R5A in SF3B1^WT Nalm-6 cells enhanced phosphorylation of both MYC and BCL2, and conferred growth advantage in vitro (Supplementary Fig. S10A–S10C). Similarly, genetic depletion of Ppp2r5a in Sf3b1^WT Eμ-Myc^Tg/+ tumors enhanced disease progression in vivo (Supplementary Fig. S10D–S10K). Overall, these data strongly suggest that alterations in PPP2R5A levels due to aberrant splicing by mutant SF3B1 promote MYC-driven tumorigenesis.

The molecular and functional consequences of loss of PPP2R5A-related MYC and BCL2 dephosphorylation suggested that activation of PP2A complex containing the PPP2R5A regulatory B subunit may be exploited for antitumor therapy. For this purpose, we assessed the therapeutic potential of the FDA-approved oral PP2A activator FTY-720 (also known as fingolimod; refs. 27–29). SF3B1-mutant Nalm-6 cells were more sensitive to FTY-720 treatment than their SF3B1^WT counterparts, experiencing growth inhibition at lower concentration compared with SF3B1^WT cells in vitro (Fig. 7A and B; Supplementary Fig. S11A–S11D). Similar results were obtained with other PP2A activators including the recently reported DT-061 (30, 31) as well as perphenazine (32) in vitro (Supplementary Fig. S11E–S11H). Moreover, both S62-phosphorylated MYC and S70-phosphorylated BCL2 decreased in a dose-dependent manner upon treatment with FTY-720 (Fig. 7C). MYC protein half-life was shortened with FTY-720 treatment, which was accompanied by downregulation of gene expression of MYC target genes (Supplementary Fig. S11I and S11J). In contrast, sensitivity to FTY-720 was diminished in SF3B1-mutant cells with PPP2R5A overexpression (Supplementary Fig. S11K–S11M). Finally, to assess the efficacy of FTY-720 in vivo, we serially transplanted Cd19-cre^+/− Sf3b1^+/+ or Sf3b1^K700E/+-mutant Eμ-Myc^Tg/+ tumors into sublethally irradiated CD45.1⁺ recipient mice and treated these mice with either vehicle or FTY-720 (Fig. 7D). Similar levels of PP2A activation from in vivo FTY-720 treatment were seen in recipients of both Sf3b1^WT and Sf3b1-mutant tumors (Supplementary Fig. S12A and S12B). Sf3b1^K700E/+ Eμ-Myc tumors were more aggressive compared with Sf3b1^+/+ tumors as evidenced by elevated CD45.2⁺ B cells in peripheral blood (PB), increased tumor burden in spleen and liver, increased CD45.2 chimerism across organs, and worsened anemia and thrombocytopenia, consistent with the results above. However, FTY-720 treatment in vivo selectively improved these phenotypes in recipients of Sf3b1^K700E/+ Eμ-Myc tumors, diminished MYC expression and induced apoptosis to a greater extent in Sf3b1^K700E/+ Eμ-Myc tumors than in Sf3b1^WT Eμ-Myc tumors, and prolonged their survival, whereas the effects on Sf3b1^+/+ Eμ-Myc tumors were minimal (Fig. 7E–M; Supplementary Fig. S12C–S12G).

Despite abundant genetic data that suggest that mutations in SF3B1 represent oncogenic drivers in malignancy, clear biological contribution of mutations in SF3B1 to tumorigenesis has not been demonstrated. In fact, evaluation of the role of mutant versus WT alleles of SF3B1 has failed to identify a role of these mutations in cancer maintenance and instead has highlighted the requirement of the WT allele in cells with heterozygous expression of SF3B1 mutants (33). Here, through combined evaluation of the effects of the SF3B1 mutation on splicing, gene expression, and TF regulatory networks across cancer types, we identify a novel mechanism by which the mutant SF3B1–mediated cotranscriptional alterations in RNA splicing contribute to stabilization and activation of oncogenic MYC. Activation of the MYC pathway was molecularly conserved across human and mouse B-cell malignancy models and identifies a clear contribution of aberrant splicing driven by mutant SF3B1 to a biological mechanism of tumorigenesis.

The high frequency of change-of-function mutations in SF3B1 in cancer and the reliance of SF3B1-mutant cells on otherwise WT splicing have resulted in efforts to target RNA splicing for therapeutic means in cells bearing mutations in SF3B1 and other splicing factors. However, the clinical viability of targeting RNA splicing in patients is unclear, and further efforts to dissect mechanistic dependencies in SF3B1-mutant cells are needed. To this end, we identify that a specific enzymatic activity, PP2A phosphatase activity, is altered in SF3B1-mutant cancers. PP2A's substrate specificity, function, and enzymatic activities are regulated in part by expression of specific regulatory B subunits. Consistent with this, loss of specific regulatory B subunits has been repeatedly shown to result in cellular transformation. Interestingly, the specific subunit of PP2A targeted for decay in SF3B1-mutant cells, PPP2R5A (also known as B56α), is well known to regulate human cell transformation from systematic RNAi screens for oncogenic PP2A regulators (34), studying PP2A targets of polyoma T antigens (which block interaction of B56 subunits with the PP2A holoenzyme; ref. 35), and genetic depletion studies in mice (36). In fact, mice hypomorphic for Ppp2r5a exhibit spontaneous skin tumorigenesis, lymphoid cell expansion, and increased clonogenic potential of hematopoietic precursors (36). Despite these abundant data highlighting PPP2R5A as a critical tumor-suppressive subunit of PP2A, however, a genetic basis for the loss of PPP2R5A in human cancers was absent until this current study.

One additional observation from this study was that distinct hotspot mutations in SF3B1 have unique effects on RNA splicing and gene expression. For example, samples with mutations affecting the H4–H8 HEAT repeat domains of SF3B1 had much clearer effects on aberrant intron proximal 3′ss usage than samples with mutations at the D894 and E902 residues. Although the precise mechanistic basis for this is not yet clear, amino acid residues within HEAT repeats H4–H8 of SF3B1 are clustered in a small area and are exposed to solvent, suggesting that these mutations may disrupt SF3B1 interactions with other spliceosomal proteins or SF3B subunits (37) to induce aberrant branchpoint usage. In contrast, D894 and E902 residues are located in the H11 domain, which may not directly affect RNA interaction. These computational predictions will be important to evaluate in future functional studies which will likely require structural and biochemical evaluation of each mutant form of SF3B1 independently.

Although prior work has highlighted MYC and BCL2 as critical substrates of PPP2R5A-containing PP2A, it is important to acknowledge that there are additional known substrates for this specific PP2A complex. Future work comprehensively defining the full spectrum of substrates of PPP2R5A-containing PP2A and how these are altered in SF3B1-mutant cancers may therefore be very important. In addition, given the striking redundancy in B subunits of PP2A, where 15 genes encode 26 B subunits potentially assembling >100 distinct PP2A complexes, it will be important to determine how loss of PPP2R5A affects other PP2A complexes in SF3B1-mutant cells. Nonetheless, the striking redundancy in PP2A complexes allows for pharmacologic interventions to reactivate residual PP2A activity in SF3B1-mutant tumors. FTY-720, the therapeutic modality used here, activates PP2A by binding SET, a negative regulator of PP2A, and directly causing its displacement from and assembly of active PP2A holoenzymes (38). Our data highlight an important novel therapeutic approach targeting the impact of mutant SF3B1 on post-translational modification and stabilization of oncogenic MYC. The specific compound used here is far from the only means to reactivate PPP2R5A-containing PP2A, however, and future efforts to target SF3B1-mutant tumors by complimentary means of reactivating via allosteric activation of its phosphatase activity may also be fruitful. Finally, besides PPP2R5A, recent studies identified robust alternative splicing changes in BRD9 (12) and DVL2 (13) that also play important roles in tumorigenesis of SF3B1-mutant cancers. Aberrant 3′ss usage in these two genes was also observed in the current study (Supplementary Fig. S13A and S13B). The genetic effects of restoring PPP2R5A expression in SF3B1-mutant tumors here, together with recent studies above, continue to highlight the need to study the impact of individual missplicing events generated by mutant SF3B1 to promote disease development.

For complete experimental details and computational analyses, see also Supplementary Methods.

RNA-seq data from patient samples or isogeneic cell lines with or without SF3B1 hotspot mutations were collected from Columbia University Medical Center (CUMC), Memorial Sloan Kettering Cancer Center (MSKCC), and public portal resources including TCGA, ICGC database of Genotypes and Phenotypes (dbGaP), and Gene Expression Omnibus (GEO). Detailed sample information is shown in Supplementary Table S1.

Six patients with CLL with SF3B1^K700E mutation and six WT cases were identified from CUMC. This study is approved by the Institutional Review Boards of Columbia University and performed in accordance with the Declaration of Helsinki. All patients gave their informed written consent. Poly-A pulldown was performed to enrich mRNA from total RNA samples, and libraries were prepared using the Illumina RNA TruSeqv2 PE100 60M (2 × 100 bp) and sequenced using the Illumina HiSeq2000 instrument at the Columbia Genome Center. Real-time analysis (Illumina) was used for base calling and bcl2fastq (version 1.8.4) for converting BCL to fastq format, coupled with adaptor trimming.

For TCGA RNA-seq data, we downloaded aligned bam files from the cancer genomics cloud in November 2017 (cgc.sbgenomics.com). This cohort includes 16 UVM, 10 BRCA, seven BLCA, five SKCM, three AML, two kidney renal clear cell carcinoma, two liver hepatocellular carcinoma, two prostate adenocarcinoma, two sarcoma, and two thymoma samples. In the meantime, an equal number of WT samples was randomly chosen with the closest name ID of TCGA barcode for each mutated sample.

The ICGC CLLE-ES cohort consists of six cases of SF3B1^K700E, one SF3B1^G742D, one SF3B1^H662D, one SF3B1^K666N, and nine randomly chosen WT samples. Raw fastq files were downloaded from the ICGC Data Portal (https://dcc.icgc.org) with the project name “CHRONIC LYMPHOCYTIC LEUKEMIA - ES.” Controlled access to these data is approved with application number DACO-1023313. The sequencing data are paired-end 2 × 76 bp libraries.

In addition, we collected 2 patients with AML (K666N, R625C) as well as two WT samples from Leucegene datasets. Raw fastq files were downloaded from the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA; accession number GSE67039). Nine samples of SF3B1-mutant Nalm-6 cell lines were also employed from GEO accession GSE72790, together with three WT samples. Three cases of SF3B1^K700E K562 cells and three WT cell lines were downloaded from GEO accession GSE95011 (39). Data from six SF3B1-mutant samples (four K700E, one K666T, and one K666R) and six patients with WT MDS were obtained from GEO accession GSE85712 (17). Another cohort of 19 SF3B1^K700E MDS samples with the same number of randomly chosen WT samples were added from GSE114922 (40). Six SF3B1^K700E and one SF3B1^K666E case of patients with CLL were accessed from dbGaP under the project of The UC San Diego CLL Study (phs000767.v1.p1).

Collectively, we have established transcriptomic characterization of pan-cancer SF3B1 mutations on 110 samples covered by 12 tumor types.

In order to annotate novel 3′ss that are not present in the current genome, we adopted the splice junction read output by STAR alignment (41). This preprocess is motivated by DeBoever and colleagues (10) to enable a more sensitive detection of novel sites, specifically designed for 3/5′ss identification. To this end, all the bam files were transformed back to .fastq format (samtools) and realigned using a splice junction database (10). Counts of junction reads from SJ.out.tab file (STAR output) were merged into a unique matrix, with each row indicating one splice junction and column for each sample. We further filtered out junctions with a read coverage of < 50 summed across all samples. Then, for each junction, we defined it as a novel splice site if it only has one end (either 3′ or 5′) annotated in the provided known genome, whereas the other end is not. Corresponding canonical junction is found if it shares the same annotated end of the cryptic junction and the other end also annotated. If this kind of canonical junction did exist, we calculated the distance between the novel end and the corresponding canonical end. If there was more than one distance, we utilized the minimum. After adding the information of transcript's strand and relative position to the canonical site, we determined whether a cryptic site is on 3′ss or 5′ss, upstream or downstream. For mouse RNA-seq, we reproduced this essential preprocess using a standard mm10 genome (Ensembl: Mus_musculus. GRCm38.75). Junctions with both ends unannotated were discarded from the analysis.

Next, with a clear relationship of each cryptic junction to its canonical one, we calculated the PSI using raw read counts for each sample, transforming the reads matrix into a PSI matrix with values between 0 and 1. Then, for two groups of interested comparison, t test was applied for all novel cryptic splice site using PSI, rather than read counts, followed by BH multiple test correction. The difference of averaged PSI between two groups was also adopted as an important metric for potential differential splicing. A bar plot of log₂ distance in base pairs was used to demonstrate cryptic 3′ splicing pattern on significantly differentially used novel sites.

To compare aberrant 3′ss usage between human CLL and mouse B cells, mouse genes were mapped into human symbols. A hypergeometric test was used to evaluate the significant level of overlapped common genes.

Studies were approved by the Institutional Review Boards of CUMC and MSKCC and conducted in accordance with the Declaration of Helsinki protocol. Written informed consent was obtained from all participants.

All animals were housed at MSKCC. All animal procedures were completed in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committees at MSKCC. The number of mice in each experiment was chosen to provide 90% statistical power with a 5% error level. Generation and genotyping of the Sf3b1^K700E/+ conditional knock-in mice on C57BL/6 background were as described (17). Eμ-Myc mice have been described previously (18). Complete blood count analysis was performed on PB collected from submandibular, using a Procyte Dx Hematology Analyzer (IDEXX Veterinary Diagnostics).

Freshly dissected femurs and tibias were isolated from Cd19-cre^+/−, Cd19-cre^+/− Sf3b1^K700E/+, Cd19-cre^+/−Eμ-Myc^Tg/+, and Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+ CD45.2⁺ mice. Bone marrow (BM) was flushed with a 3-cc insulin syringe into cold PBS supplemented with 2% BSA to generate single-cell suspensions. BM cells were pelleted by centrifugation at 1,500 rpm for 4 minutes, and red blood cells were lysed in ammonium chloride–potassium bicarbonate lysis buffer for 3 minutes on ice. After centrifugation, cells were resuspended in PBS/2% BSA, passed through a 40 μm cell strainer, and counted. For transplantation experiments, 1.0 × 10⁶ BM cells from Cd19-cre^+/−, Cd19-cre^+/− Sf3b1^K700E/+, Cd19-cre^+/−Eμ-Myc^Tg/+, and Cd19-cre^+/− Sf3b1^K700E/+ Eμ-Myc^Tg/+ Cd45.2⁺ mice were transplanted via tail-vein injection into 8-week-old sublethally irradiated (450 cGy) CD45.1⁺ recipient mice. For each bleeding, whole blood cell counts were measured on an automated blood analyzer.

FTY-720 (Sigma Aldrich) was dissolved in water and administered intraperitoneally (3 mg/kg/day). The dose was optimized and determined by multiple pilot tests to minimize weight loss (such that all mice maintained ≥85% of their weight after 2 weeks of treatment). Protein lysates extracted from the spleen of treated mice were assayed to evaluate the efficacy of FTY-720 to activate PP2A complex in vivo, using the PP2A Immunoprecipitation Phosphatase Assay Kit as directed by the manufacturer (Millipore Sigma).

mRNA stability assays were performed as previously described (42). Briefly, anti–UPF1 shRNA- or control shRNA lentivirus–infected Nalm-6 SF3B1^K700E/+ knock-in cells were generated using puromycin selection (1 μg/mL) for 7 days, and shRNAs against UPF1 were expressed using doxycycline (2 μg/mL) for 2 days. GFP (shRNA)-positive cells were FACS sorted, treated with 2.5 μg/mL actinomycin D (Life Technologies), and harvested at 0, 2, 4, 8, and 12 hours.

This was done as previously described (43). Briefly, newly synthesized proteins were labeled using the Click-iT Protein Labeling Kit (Invitrogen). Isogenic Nalm-6 cells were cultured in fresh medium for 24 hours, washed twice with PBS, and resuspended in methionine-free RPMI 1640 medium (Gibco) supplemented with 10% FCS for 30 minutes, at which point the methionine analog l-azidohomoalanine (AHA; Invitrogen) was added (50 μmol/L) to allow incorporation of AHA into nascent proteins. MG-132 (Sigma, M7449; 5 μmol/L) was added at 4 hours before the methionine replacement. Protein was extracted from the cells, and 150 μg of total protein was used in the cross-linking of AHA-labeled nascent proteins to alkyne-derivatized biotin in Click-iT Protein Reaction Buffer (Invitrogen) according to the manufacturer's instructions. Biotin–cross-linked nascent protein was captured overnight with streptavidin-coated Dynabeads (M-280, Invitrogen) and eluted. The whole volume of AHA-labeled, biotin–cross-linked, streptavidin-precipitated protein was separated by SDS–PAGE.

Statistical significance was determined by (i) unpaired two-sided Student t test after testing for normal distribution, (ii) one-way or two-way ANOVA followed by Dunnett, Tukey, Sidak, or Dunnett multiple comparison test, or (iii) Kruskal–Wallis tests with Uncorrected Dunn test where multiple comparisons should be adjusted (unless otherwise indicated). Data were plotted using GraphPad Prism 7 software as mean values, with error bars representing SD.

The data that support the findings of this study are available from the corresponding author upon reasonable request. The RNA-seq data have been deposited in NCBI SRA under accession number GSE116391.

A.R. Mato is an advisory board member of AbbVie, Genentech, J and J, Pfizer, AstraZeneca, Loxo, Sunesis, and Pharmacyclics; is an advisory board/DSMB of TG Therapeutics; is DSMB of Celgene; and reports receiving other commercial research support from AbbVie and Celgene. G. Fabbri is an employee of AstraZeneca and has stock ownership and/or stock options or interests in AstraZeneca. O. Abdel-Wahab is a consultant at H3 Biomedicine, Foundation Medicine, Merck, and Janssen, and has a consultant/advisory board relationship with Envisagenics Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Z. Liu, A. Yoshimi, O. Abdel-Wahab, R. Rabadan

Development of methodology: Z. Liu, A. Yoshimi, R. Rabadan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Liu, A. Yoshimi, H. Cho, S.C.-W. Lee, M. Ki, L. Bitner, H. Shah, B. Liu, O. Abdel-Wahab

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Liu, A. Yoshimi, J. Wang, H. Cho, S.C.-W. Lee, T. Chu, A.R. Mato, P. Ruvolo, O. Abdel-Wahab, R. Rabadan

Writing, review, and/or revision of the manuscript: Z. Liu, A. Yoshimi, H. Shah, A.R. Mato, P. Ruvolo, O. Abdel-Wahab, R. Rabadan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Cho, M. Ki, G. Fabbri, L. Pasqualucci, O. Abdel-Wahab

Study supervision: R. Rabadan

We thank Esther A. Obeng and Benjamin L. Ebert for sharing Sf3b1^K700E conditional knock-in mice, and Goutham Narla, Michael G. Kharas, and Piro Lito for technical advice. Z. Liu is supported by NIH grant P01CA087497. A. Yoshimi is supported by grants from the Aplastic Anemia and MDS International Foundation (AA&MDSIF), the Lauri Strauss Leukemia Foundation, and the Japanese Society for the Promotion of Science (JSPS) Overseas Research Fellowships. A. Yoshimi and S.C.-W. Lee are supported by the Leukemia and Lymphoma Society Special Fellow Award. S.C.-W. Lee is supported by the NIH/NCI (K99 CA218896) and the American Society of Hematology (ASH) Scholar Award. O. Abdel-Wahab is supported by grants from NIH/NHLBI (R01 HL128239), the Department of Defense Bone Marrow Failure Research Program (W81XWH-16-1-0059), the Starr Foundation (I8-A8-075), the Henry & Marilyn Taub Foundation, the Edward P. Evans Foundation, the Leukemia and Lymphoma Society, and the Pershing Square Sohn Cancer Research Alliance. R. Rabadan is supported by NIH grants (R01 CA185486, R01 CA179044, U54 CA193313, and U54 209997) and the NSF/SU2C/V-Foundation Ideas Lab Multidisciplinary Team (PHY-1545805).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.