Mutations in spliceosome genes and therapeutic opportunities in myeloid malignancies

Abstract Since the discovery of RNA splicing more than 40 years ago, our comprehension of the molecular events orchestrating constitutive and alternative splicing has greatly improved. Dysregulation of pre‐mRNA splicing has been observed in many human diseases including neurodegenerative diseases and cancer. The recent identification of frequent somatic mutations in core components of the spliceosome in myeloid malignancies and functional analysis using model systems has advanced our knowledge of how splicing alterations contribute to disease pathogenesis. In this review, we summarize our current understanding on the mechanisms of how mutant splicing factors impact splicing and the resulting functional and pathophysiological consequences. We also review recent advances to develop novel therapeutic approaches targeting splicing catalysis and splicing regulatory proteins, and discuss emerging technologies using oligonucleotide‐based therapies to modulate pathogenically spliced isoforms.


| INTRODUCTION
In 1977, Sharp, Roberts and colleagues discovered that eukaryotic genes are not contiguous but rather "split" by intervening sequences known as introns that are later removed to produce mature messenger RNAs by a macromolecular structure called the spliceosome. 1,2 One reason introns may have evolved is to diversify the number of messenger RNA species, and subsequently proteins, that can be produced by a single gene through alternative splicing. 3 As with many other essential cellular processes, cancer cells have co-opted alternative splicing to promote their survival and response to therapy. Many studies have revealed global dysregulation of splicing in cancer. [4][5][6][7] For example, synonymous mutations occurring in consensus splice sites can alter intron recognition leading to intron retention and tumor suppressor inactivation. 8,9 Additionally, genes that encode for regulators of pre-mRNA splicing are often overexpressed in cancer and may presumably enhance processing of transcripts that are important for cancer cell growth and survival. [10][11][12] A more extensive review of this literature can be found here. 13,14 Within the last decade, somatic mutations in genes encoding splicing factors themselves have been discovered at high frequency in patients with hematologic malignancies as well as in epithelial tumors, albeit less commonly. [15][16][17][18][19][20][21][22][23] Approximately 60% of patients with myelodysplastic syndromes (MDS) 15,16,18 or chronic myelomonocytic leukemia (CMML), and~55% of secondary acute myeloid leukemia (s-AML) 24 have mutations in genes encoding components of the spliceosome. 15,25,26 The most common mutations occur in SF3B1, SRSF2, U2AF1, and ZRSR2 and they tend to be mutually exclusive with one another. 15 In chronic lymphocytic leukemia (CLL), mutations in SF3B1 occur in~15% of patients. Spliceosomal mutations have also been discovered in a number of solid tumors including breast cancer, 27 pancreatic cancer, 28 lung cancer, 21,29 and uveal melanoma. 19,22,30 With the exception of ZRSR2 mutations, somatic mutations in SF3B1, SRSF2, and U2AF1 cause characteristic changes in pre-mRNA splicing that are distinct from loss-of-function, and will be discussed in greater detail in this article. These mutations are presumed to contribute to oncogenic transformation, but the underlying mechanisms remain elusive and are currently an area of intense research. The high frequency of spliceosomal mutations in myeloid malignancies has generated enthusiasm for their therapeutic targeting and preclinical studies have provided evidence that they can serve as an Achilles' heel in these cancers when using small molecule spliceosomal inhibitors. 31,32 There is anticipation that a deeper understanding of the oncogenic mechanisms of these mutations might shed light on other therapeutic avenues as well. Last, given the widespread aberrant splicing observed in cancers without spliceosome factor mutations, there is optimism that modulating the activity of the spliceosome may have broader therapeutic applicability in a larger group of cancer patients.

| SPLICEOSOME MUTATIONS IN CLONAL HEMATOPOIESIS AND MYELOID MALIGNANCIES
In 2011, several groups reported high frequency of splicing factor mutations in MDS, most frequently in SF3B1, SRSF2, U2AF1, and ZRSR2. 15,16,18 This striking finding was accompanied by the observation that these mutations were generally mutually exclusive from one another, suggesting they may share overlapping function, and/or are synthetically lethal when coexpressed. Taken together, spliceosomal mutations appears to occur predominantly in MDS patients. Subsequently, whole exome sequencing of 200 de novo AML samples in the Cancer Genome Atlas (TCGA) project 33 failed to detect any mutations in splicing factors. However, more recent studies have detected spliceosomal mutations occurring in AML using targeted sequencing panels with deeper coverage. 34 It is possible that splicing factor mutations were not detected among the TCGA cohort due to their low frequency in de novo AML and with the small number of patients they were simply not included or potentially these were missed by mutation calling algorithms due to the high GC content seen in splicing factor genes and the lower depth of sequencing by whole exome sequencing.
Indeed, a more recent study including 1540 patients with AML performed targeted sequencing of 111 genes and cytogenetic analysis and classified 11 subgroups, 34 including~18% of AML patients with mutations in chromatin modifiers (ASXL1, STAG2, BCOR, MLL PTD , EZH2, and PHF6) and spliceosome genes (SF3B1, SRSF2, U2AF1, and ZRSR2). The chromatin-spliceosome group was the second largest subgroup (following NPM1-mutant AML,~27%), and is generally composed of older patients with lower white blood cell counts, lower percentage of blasts, decreased responsiveness to chemotherapy, and an overall poorer survival. The Bayesian statistical model used to derive these subgroups was also applied to the TCGA dataset and found relatively equivalent frequency of subgroups, suggesting a biologically relevant distinction of these classes. The older patients with chromatin-spliceosome gene mutations appear to be genetically and biologically different from many other subclasses of AML and do not benefit from current treatment paradigms.
Genetically, the chromatin-spliceosome subgroup of AML resembles a mutation pattern more commonly seen in MDS. It is possible that AML patients that have chromatin-spliceosome mutations may have had a prodromal MDS period even if they did not necessarily meet the formal criteria for AML with myelodysplasia-related changes (AML-MRC). 35 In fact, a study of 194 patients with rigorously defined s-AML found that mutations in SF3B1, SRSF2, U2AF1, ZRSR2, ASXL1, EZH2, BCOR, or STAG2 was >95% specific for the diagnosis of s-AML. 24 When mutations in these genes were found in de novo AML, they conferred the same poor prognosis as seen in s-AML. Furthermore, mutations in these genes have also been detected in elderly individuals with clonal hematopoiesis. [36][37][38] Thus, either as a response to a stressor (genetic or environmental) that accumulates with age or as a phenomenon of aging itself, acquisition of mutations in chromatin modifiers and splicing factors predispose individuals to further development of MDS and/or AML.
A pair of recent studies using large cohorts of individuals with clonal hematopoiesis to attempt to define the factors that associate with progression to leukemia identified that in addition to TP53 mutations, spliceosomal mutations were associated with high risk of progression to AML. 39,40 These studies suggest that spliceosomal mutations are among the first mutations to occur in hematopoietic stem cells and because of their very high risk association with leukemia may be disease-initiating mutations. In fact, genetically engineered mouse models expressing splicing factor point mutations from the endogenous mouse locus provide more evidence to support this hypothesis. Therefore, targeting these mutations might provide the best means to eradicate disease-initiating cells.
In this review, we will focus on mutations affecting core components of the spliceosome and how they may be targeted for therapeutic applications. This includes mutations in SF3B1, SRSF2, U2AF1, and ZRSR2. The first three (SF3B1, SRSF2, and U2AF1) are all components of the major spliceosome and have the unique features of: (a) always occurring as heterozygous change-of-function mutations and (b) generally occurring in a mutually exclusive manner. Mutations in ZRSR2 do not always follow the same pattern; likely because ZRSR2 is not required for major splicing and is primarily a component of the minor spliceosome. These mutations are usually seen as loss-of-function and can, on rare occasions, co-occur with other splicing factor mutations. In the following sections, we will review basic mechanisms of splicing, how mutations in splicing factors affect normal splicing and the potential functional role of these mutations in myeloid neoplasms. We will then discuss the potential for targeting these mutations or reversing their effects with splicing modulators in cancer.

| BACKGROUND ON SPLICING
RNA splicing is a highly coordinated process that removes the intronic portions from the pre-mRNA and subsequently ligates the protein coding sequences (exons) in the majority of eukaryotic protein coding genes. It is estimated that as many as 95% of human multiexon genes undergo alternative splicing, which can significantly increase the diversity and function of the human proteome. 3 Splicing represents a critical posttranscriptional mechanism for regulating gene expression, and is orchestrated by a large, dynamic group of ribonucleoprotein complexes known as the major and minor spliceosome. The major spliceosome, which is responsible for removing the majority of human introns, consists of five small nuclear ribonucleoproteins (snRNPs): U1, U2, U4, U5, and U6, while the U5, U11, U12, U4atac, and U6atac snRNPs make up the minor spliceosome. 41,42 Significant work completed over the last few decades has deepened our understanding of the biochemical composition, regulation and activation of splicing catalysis, and more recently high-resolution structures of various stages of the spliceosome life cycle have provided detailed insights into this process. [43][44][45][46] In addition to removing intronic sequences, pre-mRNA splicing has evolved to couple with other key regulatory pathways involved in gene regulation (reviewed here [47][48][49] ).
Splicing catalysis is initiated when defined cis-elements in the pre-mRNA interact with various trans-acting factors to assemble the spliceosome complex. The catalytic core of the spliceosome is assembled de novo in a series of regulated steps and conformational changes. The These splicing modulators are critical in ultimately dictating the choice of splice site usage, and are extensively reviewed here. 50

| EFFECTS OF SOMATIC MUTATIONS IN SPLICING FACTORS
One of the most surprising findings from cancer genome sequencing efforts was the identification of recurrent somatic mutations in genes encoding pre-mRNA splicing factors in both hematologic malignancies including MDS, AML, and CLL, [15][16][17][18]20,51 and in epithelial cancers such as uveal melanoma, lung adenocarcinoma, breast cancer, and pancreatic ductal adenocarcinoma. 19,21,22,27,30,52

| SF3B1 mutations
In hematologic malignancies, SF3B1 mutations are commonly found in MDS, 16,53 AML, myeloproliferative neoplasms (MPN) and in some MDS/MPN overlap syndromes, 54 and in~10%-15% of CLL patients. 17,20 Mutations in SF3B1 are specifically enriched in a subtype of MDS previously known as refractory anemia with ring sideroblasts (RARS), characterized by anemia and dysplastic erythroblasts with abnormal iron accumulation in the mitochondria 55 causing a "ring" of blue granules to appear around the nucleus upon Prussian blue staining. RARS has been renamed MDS with ring sideroblasts (MDS-RS) and is generally associated with a favorable clinical course. 16,53 Interestingly, SF3B1 mutations are so common in MDS-RS that the WHO classification criteria have recently been revised to allow diagnosis of MDS-RS with as low as 5% ring sideroblasts in the presence of mutant SF3B1. 35 Most of the mutations in SF3B1 are clustered near the HEAT repeat domains 4 to 7 (HR4-HR7), with the most frequently mutated residues being K700 and K666 in MDS and CLL; while mutations in the R625 position are the most commonly occurring allele in uveal melanoma. The functional relevance of these distinct mutations to disease subtypes still remains unclear, and is an interesting area of focus for future studies.
SF3B1 is a component of the U2 snRNP that binds to the branchpoint during the formation of Complex A, and is predicted to be ubiquitous in recognizing the majority of 3 0 splice sites. 56 Transcriptomic analyses have revealed that the major splicing defect associated with SF3B1 mutations, regardless of cellular or disease origin, is the preferential usage of cryptic 3 0 splice sites approximately 10-30 nucleotides upstream of the canonical 3 0 splice site ( Figure 2). This is distinct from loss-of-function of SF3B1 that causes inefficient splicing catalysis. [57][58][59] The region around the cryptic 3 0 splice site coincides with an enrichment of adenosines that also appear to have stronger base-pairing affinity with the cognate U2 snRNA relative to the region around the canonical BPS. Structural analyses of the human and yeast spliceosome in the RNA-bound B-activated Complex suggest that cancer-associated mutations in SF3B1 change the charge and shape of the corresponding amino acid residues that results in direct disruption of the local interaction with pre-mRNA. 46

| SRSF2 mutations
SRSF2 mutations are found commonly in~50% of CMML,~15% of MDS,~20% of s-AML patients, and are often associated with poor prognosis and a higher risk of transformation to acute leukemia. 26 75 and is generally associated with poor prognosis, and increased risk of leukemic transformation. It is also found in a subset of pancreatic ductal adenocarcinomas 28 and nonsmall cell lung adenocarcinomas. 21 U2AF1 normally recognizes the AG-dinucleotide at the 3 0 splice site during the early steps of splicing catalysis in a sequence-specific manner. 76 Mutations in U2AF1 are concentrated on two distinct amino acid residues, S34 and Q157, both of which are located within the two distinct zinc finger domains.

RNA-seq analyses from human patient samples revealed that U2AF1
mutations induces aberrant splicing of~5% of total transcripts predominantly via aberrant cassette exon skipping or inclusion, and to a lesser extent, alternative 3 0 splice site usage. The exact splicing pattern mediated by mutant U2AF1 is dictated by consensus sequences around the AG-dinucleotide at the 3 0 splice site. The S34 mutation generally promotes cassette exon inclusion if the nucleotide immediately preceding the AG-dinucleotide is C/A over T (ie, the "−3" signature), whereas the Q157 mutation preferentially includes exons containing G over A immediately after the AG-dinucleotide (ie, the "+1" signature; Figure 2). 29  and isogenic cancer cells lines. 90 Less is known about the effects of defective U12 intron splicing driven by Zrsr2 mutations on hematopoiesis in vivo due to the lack of Zrsf2 knockout mice. In vitro, the effect of ZRSR2 depletion by RNAi altered erythroid and myeloid differentiation potential in human CD34 + cord blood cells, and resulted in reduced proliferation of leukemia cell lines. 83 Moreover, a series of genetic and pharmacologic studies have revealed that both hematologic and epithelial malignancies that carry spliceosomal mutations are highly dependent on the wildtype allele for survival, 31,32,67,[90][91][92] and that expression of multiple spliceosome gene mutations using rigorous isogenic models can induce synthetic lethality. 73  immune signaling. 73,93,97,98 Further studies will be required to systematically address these fundamental questions, and more importantly, to identify the key players and pathways that can potentially be exploited for therapeutic purposes.

| THERAPEUTIC IMPLICATIONS
Prior to the discovery of spliceosome gene mutations in cancer, naturally derived compounds that modulate pre-mRNA splicing were being tested as anticancer agents in cancer cell lines. 99,100 These studies showed that cancer cells in general are sensitive to splicing inhibition. from Streptomyces spp.) originally for use as pesticides and antibiotics.
These compounds were identified to have antitumor properties by causing cell cycle arrest in the G1 and G2/M phases and it was not until later that the SF3B component of the U2 snRNP was identified as the primary target. [110][111][112] Biochemical assays identified that pladienolides and spliceostatin A inhibit splicing by abolishing the interaction between the U2snRNP/SF3B complex to the branchpoint region of the pre-mRNA. 113,114 The discovery that a mutation in SF3B1 R1074H conferred resistance to pladienolides demonstrated that SF3B1 was a direct target of this class of splicing inhibitory compounds. 115 This was validated in a more recent study that identified a series of acquired mutations in SF3B1 (K1071 and V1078) in addition to R1074H, and in PHF5A Y36C that also conferred resistance to pladienolides. 116 The crystal structure of pladienolide B bound to the human SF3B complex was recently solved and confirmed that, in addition to SF3B1, pladienolide B also interacts with PHF5A, a PHDfinger-like domain containing member of the SF3B complex. 117 Structural analysis showed that pladienolide B fits into a hinge area that prevents the transition to a closed confirmation and precludes recognition of the branchpoint adenosine. This work is consistent with previous RNA sequencing data that showed intron retention as the major splicing alteration upon exposure to small molecule spliceosome inhibitors. 31,32 Last, structural analysis also suggested that because of the unique binding pocket and related shape of all the cur- were considered to be related to optic neuritis and potentially related to E7107 toxicity. Even at doses that pharmacodynamically affected splicing, there was only a single partial remission in both of the trials, and an additional 16 patients (35%) had stable disease. Therefore, subsequent efforts focused on developing potentially safer drugs or finding subsets of patients that might respond to lower doses.
In the presence of splicing factor mutations, the idea that cells become dependent on basal splicing and are thus more sensitive to pharmacologic splicing perturbation is an appealing therapeutic proposition. This hypothesis has been tested in vitro and in vivo in preclinical models and in clinical trials. 31,67,90-92 As discussed above, simultaneous preclinical efforts to discover the biology underlying splicing factor mutations in hematologic neoplasms identified that Srsf2 P95H/+ mutant hematopoietic cells required the wildtype allele of Srsf2 to survive. In a murine leukemia model driven by MLL-AF9 fusion oncogene, Srsf2 P95H/+ leukemias were more sensitive to E7107 than Srsf2 +/+ wildtype leukemias. 31 Similarly, increased sensitivity to spliceosome inhibitors were also observed in Sf3b1 K700E/+ and U2af1 S34F/+ murine hematopoietic cells in vivo, 67,92 and in SRSF2-mutant CMML patientderived xenograft model. 32 These studies provided proof-of-concept that a therapeutic window might exist for SF3B inhibitors to achieve the requisite splicing inhibition to selectively kill spliceosomal mutant cells. This coincided with the development of H3B-8800, an orally bioavailable derivative of pladienolide B that showed structural similarity to E7107 but with less potency. Further preclinical testing demonstrated that H3B-8800 also selectively affected splicing factor mutant myeloid neoplasms by causing enhanced retention of GC-rich introns in splicing factor mutant cells that are enriched in genes encoding spliceosomal proteins themselves. 32 Ongoing efforts will be described later but a deeper understanding of the molecular targets of known inhibitors and their effects on splicing will help set the stage for this discussion. A phase 1 clinical trial of H3B-8800 has recently been initiated targeting patients with relapsed/refractory myeloid neoplasms (MDS, CMML, and AML) that carry splicing factor mutations (NCT02841540).

| Targeting RNA splicing regulatory proteins
In addition to targeting the core spliceosome using small molecule inhibitors, splicing modulation can also be achieved by targeting pro-  gene mutations may also confer differential sensitivity to PRMT5 and/or type-I PRMT inhibitors.

| Targeting RBPs: anticancer sulfonamides and RBM39
Two elegant studies focused on identifying the molecular targets of anticancer sulfonamides have revealed a potential therapeutic applica-  146 The morpholino oligonucleotides also lack the negatively charged backbone of traditional ASOs that may interact nonspecifically with other components of the cell and thus may be less toxic as a result.
Despite these recent technological advances and clinical achievements in neuromuscular disorders, the utility of ASOs in spliceosome mutant leukemias remains challenging, primarily due to our insufficient understanding of the key mis-spliced targets responsible for disease initiation and/or progression. Future work focusing on systematic identification of pathogenic isoform alterations will be key to exploit the full potential of this therapeutic approach.

| CONCLUSIONS AND FUTURE DIRECTIONS
The field of targeting splicing as an anticancer therapeutic is in its nascence. The major idea leading the field so far and currently being tested in an early phase clinical trial is the use of spliceosomal modulation to induce synthetic lethality in splicing factor mutant malignancies. Whether drugs that target the SF3B complex can achieve this with minimal on-target toxicity remains to be seen. Regardless, more therapies that globally perturb splicing via alternative mechanisms could be developed into therapeutic approaches. As discussed above, inhibiting regulatory proteins such as splicing regulatory kinases (eg, CLKs and SRPKs) and PRMTs are currently being explored as potential therapeutic avenues. Another new tactic involves using molecules that link splicing proteins to E3-ubiquitin ligases, targeting them for proteasomal degradation. These can be either pre-existing drugs or specifically designed proteolysis-targeting chimeras, which are currently being used extensively as drug-discovery tools to determine the function of proteins or the result of destroying the protein, but are quickly being developed as pharmaceutical agents. As discussed above, the discovery that the sulfonamide compound indisulam promotes the recruitment of the spliceosome-associated protein RBM39 to the CUL4-DCAF15 E3-ubiquitin ligase 136 and its degradation is associated with aberrant splicing and anticancer activity is quite exciting. Further developments in understanding the oncogenic mechanisms of splicing factor mutations may lead to novel approaches as well. For example, increased R-loop formation is observed in spliceosome mutant leukemias and this has been shown to sensitize them to ATR inhibition. 147 Moreover, unbiased genetic and chemical screening approaches in the context of spliceosome mutant cancers will be crucial for uncovering novel biological features, as well as uncovering therapeutically actionable targets.
The rapid development of ASO therapies in the treatment of neuromuscular diseases has paved way for this class of therapeutic approach. As delivery systems improve and safety in humans is further established, these therapies will expand in cancer clinical trials and using them to cause splice switching seems likely to be among the