Misregulation of Pre-mRNA Alternative Splicing in Cancer

Alternative Splicing Patterns in Cancer

Alternative splicing patterns in cancer cells reflect those found in normal cells. Global analysis of more than 15,000 cancer-specific splice variants in 27 types of cancer shows that the average number of cancer-specific splice variants per gene is smaller than that of tissue-specific splice variants in 35 normal tissues (1.51 vs. 1.99; ref. 9). This is expected, because tissue-specific splice variants are required for generating the necessary proteomic complexities of human tissues, and the splice variants have undergone extensive natural selection during the course of evolution (16, 17). Cancer-specific splice variants, which may bestow survival advantages to cancer cells, often result in rapid death of the human subject that harbors the cancer, and therefore are selected against rather than selected for at the organismal level. Regardless, cancer-specific alternative splicing includes all of the five main alternative splicing patterns observed in normal tissues: cassette exons, alternative 5′ splice sites, alternative 3′ splice sites, intron retention, and mutually exclusive exons (Fig. 1), suggesting that cancer cells and differentiated cells use fundamentally similar splicing mechanisms. To illustrate the alternative splicing patterns that cancer cells use to gain survival advantages, we describe below an exemplary set of functionally important alternative splicing events (Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Examples of alternative splicing patterns in cancer. The main alternative splicing patterns in cancer cells include cassette exons (skipping of one exon, skipping of multiple exons, and exon inclusion), alternative 5′ splice sites, alternative 3′ splice sites, intron retention, and mutually exclusive exons. Specific examples discussed in the text are shown.

Misregulation of Alternative Splicing in Cancer by Regulatory Splicing Factors

Splicing regulation is essentially the process of selecting splice sites in pre-mRNA transcripts. This process is generally directed by cis-acting regulatory sequences and trans-actingRNA-binding proteins (RBP; ref. 15). Well-studied RBPs include two families: serine/arginine-rich (SR) proteins and heterogeneous nuclear ribonucleoproteins (hnRNP; refs. 15, 51–55). SR proteins bind to exonic splicing enhancers (ESE) and intronic splicing enhancers (ISE) and usually promote exon inclusion. In contrast, hnRNPs bind to exonic splicing silencers (ESS) and intronic splicing silencers (ISS) and in most cases lead to exon skipping (15, 51–55). Thus, one major mechanism of alternative splicing misregulation is through alterations in the levels and activity of RBPs. Below we describe a few RBPs that have been implicated in misregulation of alternative splicing in cancer.

SRSF1 (formerly known as ASF/SF2) is perhaps the best-known SR protein, and is involved in both constitutive and regulated splicing as well as other cellular processes. It is upregulated in various human tumors, and its induced overexpression leads to transformation of mammary epithelial cells and immortal rodent fibroblasts, suggesting that it may be a proto-oncogene (56, 57). SRSF1 affects alternative splicing of many target pre-mRNAs, some of which are known to contribute to tumorigenesis. It binds to an ESE in exon 12 of RON and promotes the skipping of exon 11 to produce ΔRON, which enhances cancer invasiveness (20). Overexpression of SRSF1 leads to inclusion of exon 12a of BIN1 (a tumor-suppressor gene), and the resulting isoform loses tumor-suppressor activity due to its inability to interact with MYC (56). In addition, SRSF1 promotes the production of isoform 2 of S6K1 through alternative splicing, and overexpression of this isoform is able to transform NIH3T3 cells (56). SRSF1 also modulates alternative splicing of MNK2 pre-mRNA: overexpression of SRSF1 results in production of the MNK2b isoform, promoting MAPK-independent phosphorylation of the eukaryotic initiation factor eIF4E, which enhances cap-dependent translation and may contribute to oncogenic transformation (56). It has also been shown that SRSF1 can interact directly with mTOR to facilitate phosphorylation of the translation inhibitor 4E-BP, leading to 4E-BP release from eIF4E and activation of translation (58). Recently, Anczukow and colleagues (57) found that SRSF1 stimulates production of isoform BIM γ1, which lacks exons 2 and 3 of BCL2L11 (a proapoptotic BCL-2 family member), and concomitantly downregulates production of BIN1+13, a BIN1 isoform that includes exon 13. Expression of BIM γ1 increased acinar size and decreased apoptosis, whereas expression of isoform BIN1+13 did the opposite. Therefore, it was proposed that BIM γ1 upregulation and BIN1+13 downregulation combined contribute to SRSF1-induced tumorigenesis (57). Recently, it was shown that SRSF1 is regulated by MYC: MYC directly binds to two noncanonical E-boxes in the SRSF1 promoter and activates its transcription. Knockdown of MYC downregulates SRSF1 expression in lung cancer cell lines (59).

SRSF3 (formerly SRp20) is another serine/arginine-rich protein that has been implicated in misregulation of alternative splicing in cancer. It is overexpressed in human cervical, lung, breast, stomach, skin, bladder, colon, liver, thyroid, and kidney cancers (60). Overexpression of SRSF3 leads to transformation of rodent fibroblasts, suggesting that it is a proto-oncogene (60). Knockdown of SRSF3 results in apoptosis of a variety of cancer cells (60, 61). It was shown that knockdown of SRSF3 led to skipping of exon 8 of homeodomain-interacting protein kinase-2 (HIPK2), an antioncogene that induces tumor cell apoptosis. Full-length HIPK2 binds to an E3 ubiquitin ligase (Siah-1) and is constantly degraded, whereas the isoform HIPK2 Δe8, which lacks 27 amino acids, loses the ability to bind to Siah-1 and therefore is resistant to protein degradation. HIPK2 Δe8 still retains antioncogenic activity and therefore induces apoptosis (61). Tang and colleagues (62) provided evidence that SRSF3 also regulates alternative splicing of p53. Binding of SRSF3 to exon i9 of p53 inhibits production of isoform p53β. Downregulation of SRSF3 induces p53β production and promotes cellular senescence. Finally, Wang and colleagues (63) identified an SRSF3-binding site in exon 10 of PKM pre-mRNA and showed that knockdown of SRSF3 resulted in an approximately 20% switch from PKM2 to PKM1 and reduced lactate production.

hnRNP A1 and hnRNP A2 are two structurally and functionally related hnRNPs that likely play a role in cancer. Both proteins are overexpressed in a wide variety of cancers (42, 64, 65). RNA interference (RNAi)–mediated knockdown of hnRNP A1 and A2 (A1/A2) together results in apoptosis in cancer cells, but not in normal cells, suggesting that the two proteins are important for cancer cell growth (64). Golan-Gerstl and colleagues (65) showed that overexpression of hnRNP A2 in NIH3T3 cells induces skipping of RON exon 11 and production of ΔRON. Knockdown of RON inhibited the hnRNP A2-mediated transformation. A recent genome-wide analysis of alternative splicing events using high-throughput sequencing of RNA isolated by cross-linking immunoprecipitation (HITS-CLIP) showed that hnRNP A1 and A2 each potentially regulates more than 2,000 alternative splicing events (66). One important finding is that A1/A2 share many (one third) common targets, one of which is PKM. As mentioned above, deregulation of PKM alternative splicing is known to be critical for glucose metabolism in tumor cells. David and colleagues (42) showed that A1/A2 together with another hnRNP protein, PTB, bind to sequences flanking exon 9 of PKM pre-mRNA, repress E9 inclusion, and promote E10 inclusion. Expression levels of A1/A2 and PTB were found to correlate perfectly with the ratios of PKM2/PKM1 in a number of normal brain and glioma samples. Knockdown of A1/A2 and PTB, or of MYC, which drives their expression, results in switching from PKM2 to PKM1 (42). Consistent with these findings, another study showed that A1/A2 and PTB knockdown leads to a decrease of lactate production in a glioblastoma cell line (43). Chen and colleagues (67) further showed how mechanistically the protein levels of A1/A2 and PTB determine the outcome of PKM alternative splicing: at high levels, A1/A2 and PTB predominantly bind to sites in and around exon 9 to repress E9 inclusion, but when their levels are reduced, their binding shifts to sites flanking exon 10, preventing E10 inclusion.

PTB (also known as hnRNP I) is an hnRNP that binds to polypyrimidine-rich intronic elements and in most cases represses the inclusion of the regulated exon (68). It has been shown that PTB is upregulated in ovarian cancer and gliomas (42, 69, 70). Knockdown of PTB suppresses ovarian tumor cell growth and invasiveness in vitro (70). However, overexpression of PTB in immortalized or normal cells does not enhance proliferation, anchorage-independent growth, or invasion (71), suggesting that PTB may play a necessary, but not transforming, role in tumorigenesis. PTB is known to regulate several alternative splicing events that are relevant to cancer. Binding of PTB to an ISS element in the FGFR1 transcript leads to skipping of the α exon and production of isoform FGFR-1β (72). This truncated receptor has higher affinity for FGF-1 (73), and might facilitate malignant progression of astrocytic tumors (74). PTB also regulates alternative splicing of USP5, a deubiquitinating enzyme whose knockdown can lead to accumulation of p53 (75). Two USP5 isoforms can be generated by use of alternative 5′ splice sites in exon 15. In glioblastoma, high levels of PTB inhibit the proximal 5′ splice site and use of the distal 5′ splice site produces USP5 isoform 2. Switching isoform 2 to isoform 1 using antisense oligonucleotides inhibited growth and migration of two glioblastoma cell lines (69). Genome-wide studies of PTB-regulated alternative splicing events in HeLa cells using HITS-CLIP showed that PTB not only represses but also activates exon inclusion, depending on its binding site location with respect to the regulated exons (68, 76). The functional significance of these PTB-regulated alternative splicing events has not been examined, although one of the PTB targets is an ISS element upstream of exon 9 of PKM (68). As discussed earlier, PTB binds to this ISS and, together with hnRNP A1/A2, regulates PKM alternative splicing (42). The fact that tumor cells overexpress and recruit three different hnRNP proteins to regulate PKM alternative splicing reinforces the importance of producing PKM2 in tumor cells.

hnRNP H has recently been implicated in oncogenesis through the misregulation of alternative splicing of both IG20/MADD and RON pre-mRNAs (77). hnRNP H is upregulated in gliomas and binds to an ESS in exon 16 of IG20/MADD, leading to skipping of exon 16 and production of the MADD isoform, which is necessary and sufficient for cell survival (77). RNAi-mediated knockdown of hnRNP H reverses alternative splicing, producing the exon 16–containing IG20 isoform, and results in cell death of both U373 glioma and HeLa cells, possibly through IG20-triggered caspase-8 activation (78). In addition, hnRNP H binds to a similar ESS in exon 11 of RON and leads to skipping of exon 11 and production of ΔRON, which promotes cell invasiveness (77).

In the above examples, misregulation of alternative splicing occurs in the absence of genetic mutations and, in many cases, without changes in the overall levels of the alternatively spliced transcripts. Switching from one isoform to another is regulated by the levels and activity of RBPs, either individually or in combination. A recent proteomic study revealed as many as 860 RBPs in humans (79). However, only a couple dozen are well studied. A genome-wide analysis shows that each of six tested RBPs binds to multiple sites and more than half of all alternative splicing events are regulated by multiple RBPs (66). It remains a difficult challenge to determine how these hundreds of RBPs cooperate and coordinately regulate the tens of thousands of normal alternative splicing events that are required for tissue differentiation (16, 17, 80). Any misregulation in this process may generate aberrant alternative splicing that leads to serious consequences, such as cancer.

Mutations in the Core Splicing Machinery and Cancer

All of the above studies describe how changes in the intracellular levels of splicing regulatory proteins can contribute to cancer. What was lacking, however, were any examples of mutations in genes encoding splicing proteins that either cause or contribute to neoplastic transformation. A number of recent studies, though, have identified mutations affecting components of the core splicing machinery that play critical roles in neoplasia. Given the importance of these findings to our appreciation of the role of splicing in cancer, we discuss these studies in some detail.

In 2011, Yoshida and colleagues (81) reported recurrent somatic mutations in the genes encoding components of the RNA splicing machinery in myelodysplastic syndromes (MDS), a diverse group of myeloid neoplasms characterized by an abnormality in myeloid blood cell production and propensity of progression into acute myeloid leukemia. The most frequently mutated genes encode splicing factors SF3B1, U2AF1, SRSF2, and ZRSR2. Mutational frequencies for SF3B1 are particularly high in refractory anemia with ring sideroblasts (RARS) and RARS with thrombocytosis (RARS-T), ranging from 64% to 83% (81–83). SF3B1 is also frequently mutated in chronic lymphocytic leukemia (CLL; refs. 84–86), as well as in uveal melanoma (87, 88). All SF3B1 mutations are heterozygous, and none are nonsense mutations or introduce a frameshift (Fig. 2A). The mechanism through which the splicing factor mutations misregulate RNA splicing and subsequently lead to disease is still unknown. Here, we offer our perspectives.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Splicing factor mutations and aberrant splicing patterns. A, distribution of mutations in SF3B1. B, aberrant splicing with defects in 3′ splice site (ss) recognition. Outcomes of defects in 3′ splice site recognition or utilization are diagrammed.

SF3B1, U2AF1, SRSF2, and ZRSR2 are all involved in the selection of splice sites at the 3′ end of introns. Mutations in these genes most likely reflect defects in 3′ splice site recognition during RNA splicing. As shown in Fig. 2B, defects in 3′ splice site recognition (but with normal 5′ splice site recognition) can result in two 5′ splice sites competing for one 3′ splice site, an alternative splicing pattern that resembles alternative 5′ splice sites. A frequent outcome of alternative 5′ splice sites is the selection of the 5′ splice sites proximal to the downstream 3′ splice sites (89). As a result, the final mRNA product often has retained introns (Fig. 2B). In support of this speculation, Yoshida and colleagues (81) showed that expression of mutant U2AF1 results in large-scale (∼5%) intron retention in HeLa cells. Because introns are rich with stop codons, retained introns frequently introduce into the mRNA premature termination codons that activate nonsense-mediated mRNA decay (NMD), which was observed in the HeLa cells expressing mutant U2AF1 (81). It must be noted that not every intron is retained by 5%, but instead some introns are retained whereas others are not. For example, intron 59 of BIRC6 pre-mRNA is mostly retained, but no retention is found in intron 58 or 60 of the same gene. This “all-or-nothing” (technically “more-or-less”) splicing pattern implies that there is intron sequence specificity for mutant U2AF1. Because most of the mRNAs with retained introns are rapidly degraded by NMD, it is difficult to identify sequence conservation in those degraded introns. Therefore, NMD inhibitors may be useful to help identify the targets of the mutated splicing factors.

Several studies have begun to examine the role of SF3B1 in MDS. Visconte and colleagues (90) showed that knockdown of SF3B1 in K562 cells resulted in retention of introns. However, SF3B1 knockdown did not produce ring sideroblasts, possibly because K562 cells are not able to differentiate along the erythroid lineage. Indeed, in healthy human bone marrow cells, ring sideroblast formation was induced by meayamycin, an SF3B1 inhibitor (90). However, this SF3B1 “haploinsufficiency” hypothesis cannot explain the absence of nonsense and frameshift mutations in SF3B1, which suggests that the mutated protein likely maintains structural integrity, but with altered function. RNA sequencing analysis of samples from one healthy donor and 2 patients with SF3B1 mutations revealed that 130 genes were differentially expressed, of which 94% (an unusually high percentage) had lower expression in patients (90). One explanation for this is that some introns in these genes are retained and the mRNA is rapidly degraded through NMD. Nevertheless, none of these downregulated genes is involved in mitochondrial function or related to the ring sideroblast phenotype. This may reflect the choice of control. Visconte and colleagues used total bone marrow cells from a healthy donor as a control, which contains a mixture of all types of blood cells with unknown cell ratios, making a complex gene expression profile. When purified CD34⁺ cells were used as a control, Papaemmanuil and colleagues (83) found that key genes in the mitochondrial pathways are downregulated in patients with MDS carrying SF3B1 mutations. In particular, the mitochondrial gene ABCB7 is consistently downregulated in patients with RARS, suggesting that it may be a key mediator of ineffective erythropoiesis of RARS (91). Indeed, Nikpour and colleagues (92) recently showed that reduced expression of ABCB7 in normal bone marrow markedly reduced erythroid differentiation and growth with accumulation of mitochondrial ferritin, a phenotype similar to that observed in intermediate RARS erythroblasts. It remains to be determined how SF3B1 mutations result in downregulation of ABCB7, although it is possible that intron retention followed by NMD contributes.

The link between the splicing gene mutations and clonal expansion of hematopoietic stem cells remains unclear. Expression of mutant U2AF1 leads to death, rather than promoting growth, of both HeLa cells and TF-1 cells in vitro (81). This unexpected result might reflect the fact that the outcome of splicing defects may depend on certain cellular contexts, as knockdown of SF3B1 in K562 does not induce the ring sideroblast phenotype, whereas inhibition of SF3B1 in bone marrow cells does (90). Another unexpected result is that mutant U2AF1 impairs the reconstitution capability, rather than promoting clonal expansion, of mouse CD34⁺ cells (81). A possible explanation for this stems from the perhaps unexpected differences between mouse and human alternative splicing. Recent studies have shown that mouse alternative splicing is drastically different from human alternative splicing (16, 17). Even though human and mouse splicing factors are almost identical (SF3B1 and SRSF2 are 100%, U2AF1 is 96%, and ZRSR2 is 82% identical between human and mouse), only one quarter of human alternative splicing events were observed in mice (16). Therefore, using mouse models to study human diseases that reflect changes in alternative splicing, or aberrant splicing more generally, may be misleading.

Several other lines of evidence also suggest that the effects of spliceosomal gene mutations may be dependent on cellular contexts. For example, patients with MDS carrying SF3B1 mutations generally have a favorable prognosis, whereas SF3B1 mutations in CLL correlate with poor overall survival and resistance to chemotherapy (85, 86, 93–95). Unlike in adult MDS, spliceosomal mutations are rare in pediatric MDS and juvenile myelomonocytic leukemia (96). In uveal melanoma, SF3B1 mutations are frequent (87, 88), but none of the 85 cutaneous melanomas had an SF3B1 mutation (97). This cellular-context dependency of effects provides an opportunity for developing antitumor drugs: tumor cells and normal cells are known to have different cellular contexts; therefore, modulating the activities of spliceosomal proteins will likely yield different, even opposing, effects. Indeed, spliceosome modulators such as sudemycins, pladienolide B, FR901464, and its derivative spliceostatin A (SSA) have potent toxicity to tumor cell lines, but display little toxicity to normal cells (98–100). It has been shown that FR901464 and SSA bind to the SF3b complex and promote retention of intron 1 of p27, a cyclin-dependent kinase (CDK) inhibitor. Translation of the intron 1–containing pre-mRNA leads to production of a C-terminal truncated protein isoform p27*, which is resistant to proteasomal degradation and inhibits CDK2 kinase activity, thereby inhibiting cell growth (101). SSA treatment also leads to intron retention in VEGF and results in reduction of VEGF levels (possibly by NMD), inhibiting cancer cell angiogenesis (102). Although the exact mechanism of selective tumor cytotoxicity remains to be fully explored, one explanation is that growth of cancer cells often relies on oncogenic protein isoforms (arising from alternative splicing), which are lacking in normal cells.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Authors' Contributions

Conception and design: J. Zhang, J.L. Manley

Writing, review, and/or revision of the manuscript: J. Zhang, J.L. Manley

Grant Support

This work is supported by a grant (R01GM048259; to J.L. Manley) from the NIH. J. Zhang is the recipient of an American Brain Tumor Association Basic Research Fellowship in Honor of Denise Kimball.