SRSF2 Regulates Alternative Splicing to Drive Hepatocellular Carcinoma Development

Alternative splicing is a regulated posttranscriptional process that controls gene expression as well as generates proteomic diversity (1). This involves trans-acting splicing factors and cis-regulatory RNA elements located within alternative exons and/or flanking introns (2). The majority of splicing factors are RNA-binding proteins, which bind to RNA elements, positively or negatively influence splice site selection, and determine alternative exon usage (3). In humans, the vast majority of protein-coding genes are alternatively spliced, leading to diverse protein isoforms with distinct functions (4, 5).

Aberrant alternative splicing events are commonly observed in cancer cells and have been implicated in many types of cancer (6). In recent years, genome-wide studies have extensively increased the number of alternative splicing events altered in cancers, and many of these events were associated with multiple aspects of tumor cell functions, such as cell-cycle control, cytoskeletal organization, migration, and cell–cell adhesion (7, 8). Importantly, these cancer-specific splice variants are often upregulated in tumors, which contribute to tumor cell survival and cancer progression, and also predict survival in patients with cancer. Altered alternative splicing can occur in several ways, including mutation of RNA sequence elements or/and deregulation of RNA-binding proteins. In fact, deregulation of splicing regulators such as SRSF1, SRSF10, RBFOX2, MBNL1/2, and QKI proteins has been observed and accounts for hundreds of altered alternative splicing events present in multiple cancer types (7, 9–12).

SR proteins and hnRNP proteins are well-characterized RNA-binding proteins that play critical roles in splicing regulation, and changes in their expression can dramatically affect alternative splicing profiles in cells (13–15). SRSF2 is one of the SR proteins that are composed of RRM domain and RS domain. SRSF2 frequently binds to splicing enhancer sequences (ESE) and acts as a general splicing activator. SRSF2 also plays additional roles in transcription activation, RNA stability, mRNA transport, and mRNA translation (16, 17). Heterozygous mutations in SRSF2 occur frequently in patients with myelodysplastic syndromes (MDS) and generally are associated with adverse prognosis. Two recent studies demonstrate that the most frequent mutation occurs in SRSF2 affects its binding affinity and causes abnormal splicing of key hematopoietic regulators (18, 19). These findings further emphasize the importance of SRSF2-regulated alternative splicing events in disease progression.

Hepatocellular carcinoma (HCC) is one of the most prevalent human cancers worldwide, with 55% of cases (and deaths) occurring in China. Our previous work has demonstrated that hepatocyte-specific inactivation of SRSF2 resulted in acute liver failure and early death in mice, underscoring the critical role of SRSF2 in liver development (20). However, little has been known about the clinical relevance and prognostic implications of SRSF2 in patients with HCC.

In the current study, we analyzed the expression of SRSF2 and evaluated its clinical relevance using an HCC tissue array. Upregulation of SRSF2 occurs in a high frequency in HCC samples and appears to predict patients' adverse prognosis, as higher expression of SRSF2 is associated with the lower survival time of patients with HCC. RNA-sequencing (RNA-seq) analysis and motif analysis revealed that SRSF2 modulates both exon activation and repression in highly position-dependent manner. Mechanically, SRSF2 binding within the alternative exon is associated with activation, whereas binding in the region of the flanking constitutive exon is associated with repression.

Importantly, we observed that SRSF2 can stimulate expression of cancer-associated splicing variants in clinical HCC samples. These splicing variants play critical roles in cancer cell survival and tumorigenic potential. We further demonstrated that SRSF2 mediates tumorigenesis of HCC cells through controlling expression of cancer-specific splice variants. Taken together, our study highlights the biologic significance of SRSF2 in HCC development and identifies SRSF2 as a key regulator for controlling cancer-related splicing events and thus maintaining oncogenic phenotypes of hepatoma cells.

Twenty pairs of primary HCC and their corresponding normal tissues were obtained from patients with HCC treated at Zhongshan Hospital (Shanghai, China) after their written informed consent, which are kindly provided by Drs. Xiaodong Zhu and Huichuan Sun (Zhongshan Hospital). Fresh samples were snap-frozen in liquid nitrogen and used for RT-PCR analysis. Tissue microarray chips containing 90 pairs of primary HCC and matched normal tissues were purchased from the National Engineering Center for BioChips in Shanghai, China. And they were used for immunohistochemical (IHC) analysis, IHC score analysis, and Kaplan–Meier survival analysis.

SRSF2 antibodies for IHC staining were from Sigma (#HPA049905), whose specificity has been demonstrated in the immunostaining of SRSF2 in normal and cancer tissues from the Human Protein Atlas (http://www.proteinatlas.org). In brief, paraffin sections were deparaffinized with xylene and rehydrated, followed by antigenic retrieval. The sections were then treated with 3% hydrogen peroxide to quench the endogenous peroxidase activity, followed by incubation with anti-SRSF2 antibodies. All sections were counterstained with hematoxylin.

IHC evaluation was performed independently by 2 pathologists from the Department of Pathology, Zhongshan Hospital. The staining was scored according to the staining intensity and the percentage of cells stained. The staining intensity was scored on a scale of 0 to 3: 0 (no staining), 1 (weak staining ∼ light yellow), 2 (moderate staining ∼ yellowish brown), and 3 (strong staining ∼ brown). IHC scores were calculated as the product of staining intensity multiplied by the percentage of stained cells. Tumors with IHC scores ≥ 1.5 were classified as SRSF2-high expression group; tumors with IHC scores < 1.5 were classified as SRSF2-low expression group.

Human liver cancer Huh7 cells, SK-Hep-1 cells and human embryonic kidney 293T cells were purchased from Cell Bank of Type Culture Collection of Chinese Academy of Sciences (Shanghai, China). 293 GPG cells were obtained from Dr. Lixing Zhan at the Institute for Nutritional Sciences (Shanghai, China). All cell lines were initially obtained several years ago and have been tested to confirm lack of mycoplasma contamination; however, no additional authentication has been performed. α-Tubulin, β-actin, and hnRNPK antibodies were purchased from Santa Cruz Biotechnology. SRSF2 antibodies for Western blotting were from Millipore. GCH1 and HA tag antibodies were from Abcam. STK39 antibodies were from Cell Signaling Technology. Ki-67 antibodies were from BD Biosciences. SRSF10 antibodies were from Manley's laboratory as previously described (15). Growth factor–reduced Matrigel was from BD Biosciences.

RT-PCR and Western blotting were performed as previously described (10, 21). In vivo CLIP of RNA bound to SRSF2 proteins was performed as previously described (10). In brief, hemagglutinin (HA)-tagged SRSF2, HA-SRSF10, HA-hnRNPK plasmids, or empty vector controls were transiently transfected into 293T cells, and ultraviolet cross-linking was performed followed by immunoprecipitation using Magna RIP kit (Millipore).

Cell proliferation, colony survival assay, and xenograft tumor formation assay were performed as previously described (10).

Minigenes were constructed by amplifying genomic sequences spanning exons 5 to 7 of GCH1 gene or exons 11 to 13 of MKNK1 gene, which were then cloned into PCDNA3.1 vectors, respectively. Deletion or add-in mutant derivatives were made on the basis of the minigene plasmids. Cells were transfected with siRNAoligos using Lipofectamine RNAiMAX (Invitrogen). SRSF2 overexpression was generated with pCMV-Tag2B plasmids. sh-SRSF2 retrovirus was generated with pSiREN-puro retroviral vectors in 293 GPG cells. sh-SRSF2, sh-GCH1-L and sh-STK39-L lentivirus were generated with PLKO.1-puro lentiviral vectors in 293T cells. And lentivirus expressing GCH1-L/S or STK39-L/S isoforms were generated with pCDH-puro lentiviral vectors in 293T cells. Huh7 cells were infected with retroviruses or lentivirus and selected for puromycin resistance.

Motif analysis for SRSF2 proteins were performed as previously described (14).

Total RNAs isolated from Huh7 cells transfected with si-SRSF2 or control siRNA were subjected to paired-end RNA-seq using Illumina HiSeq 2000 system according to the manufacturer's instruction. Reads mapping and data analysis for differentially regulated exons between 2 samples were carried out as previously described (14). The raw sequence data have been submitted to Gene Expression Omnibus with accession number GSE78705.

Oncomine data analyses were performed as previously described (22). RNA-seq data from a total of 341 cases of human liver HCC samples and 50 normal samples was downloaded from The Cancer Genome Atlas (TCGA) database. Ratio for GCH1 E6 IN/EX (inclusion/exclusion) was calculated on the basis of the number of reads supporting inclusion or exclusion events. Expression of SRSF2 was calculated by counting the number of reads that falling into SRSF2 gene. After log₂-transformed ratio of GCH1 E6 IN/EX and expression levels of SRSF2, a linear analysis was performed between these 2 transformed variables.

All data presented as histograms refer to a mean value ± SD of the total number of independent experiments. Statistical analysis was performed by the Student t test at a significance level of P < 0.05. Survival curves were calculated by Kaplan–Meier methods, with comparisons performed using the log-rank test. For TCGA database HCC samples, statistical analyses were performed using SAS software.

Our previous results have shown that hepatocyte-specific inactivation of SRSF2 resulted in severe liver injury in mice (20). Now we wanted to test whether SRSF2 is implicated in carcinogenesis of liver cancers. To this end, we first analyzed mRNA levels of SRSF2 in human HCC tissues using published data sets from Oncomine (23). In silico analysis of 2 independent datasets demonstrated that SRSF2 was expressed at high levels in HCC tissues than in normal tissues (Fig. 1A). Then we examined protein levels of SRSF2 in a HCC tissue array by IHC. The array included cancer tissue samples and corresponding normal tissues from 90 patients with HCC. The SRSF2 antibodies from Sigma are suitable for IHC, which has been demonstrated in many previous studies. As shown in Fig. 1B, we observed that 62 cases exhibited strong immunopositivity (69%), 18 cases exhibited moderate-to-strong immunopositivity (20%), and 10 cases exhibited weak immunopositivity (11%). Statistical analysis revealed that SRSF2 was significantly upregulated in the cancer tissues compared with normal tissues (Fig. 1C). Importantly, we observed that SRSF2 protein levels were correlated with the higher histologic tumor grade (Supplementary Table S1). Moreover, Kaplan–Meier survival analysis showed that patients with low levels of SRSF2 expression had significantly longer median survival than patients with high SRSF2 expression (Fig. 1D, P = 0.006). Taken together, these observations strongly indicate that upregulation of SRSF2 was closely associated with progression and poor prognosis in patients with HCC.

To investigate the role of SRSF2 played in tumors, we first designed 2 independent siRNAs targeting against SRSF2 and examined the effects of SRSF2 knockdown in 2 hepatocarcinoma cell lines. Western blot analysis confirmed that siRNA-mediated transient elimination of SRSF2 (si-SRSF2) was very efficient in both Huh7 and SK-Hep-1 cell lines (Fig. 2A). Strikingly, SRSF2 knockdown induced significant growth inhibition in both cell lines in comparison with cells treated with control siRNA (si-NC; Fig. 2B and C). SRSF2 knockdown also severely impaired colony-forming abilities of both cell lines in vitro (Fig. 2D and E). Consistently, stable expression of SRSF2-shRNA in Huh7 cells dramatically decreased their tumorigenic potential compared with control cells expressing luci-shRNA (Supplementary Fig. S1A and S1B). Moreover, overexpression of SRSF2 in Huh7 cells could increase their colony-forming efficiency, further confirming its protumorigenic function in tumor cells (Supplementary Fig. S1C and S1D). Finally, to examine the role of SRSF2 in tumorigenicity, Huh7 cells stably expressing SRSF2-shRNA#1 were injected into the flanks of 4-week-old nude mice. Knockdown of SRSF2 induced a complete loss of tumor formation in all the injections (Fig. 2F). Overall, these data demonstrated that loss of SRSF2 dramatically represses cell growth in vitro and decreases the tumorigenic potential of hepatocarcinoma xenografts in mice.

It is well established that SRSF2 acts as a critical splicing regulator, next we performed RNA-seq using RNA extracted from SRSF2 knockdown and wild-type (WT) Huh7 cells to identify its regulated alternative splicing events (see Materials and Methods; Supplementary Table S2). We found that SRSF2 knockdown resulted in alterations in 966 splicing events, including 706 cassette exons, 79 alternative 5′splice sites, 74 alternative 3′ splice sites, 66 mutually exclusive exons, and 41 retained introns (Fig. 3A). Significantly, many of SRSF2-affected splicing events involved great changes either in inclusion or exclusion of alternative exons (Fig. 3B). Moreover, these SRSF2-affected targets were functionally associated with cancer-related functions, such as cell-cycle control, DNA repair, chromatin modification, and cell division (Fig. 3C).

Next we selected 120 targets for experimental validation, based mainly on their P values. Among them, 89 splicing events were successfully confirmed, reflecting −74% validation rate (Fig. 3D). Representative examples of 10 validated events were shown in Fig. 3E and F. SRSF2 knockdown could activate isoform switches for which the change often occurs in the most abundant isoform. This is the case both for alternative exon-containing mRNA variants from genes such as RBCK1, FDPS, GCH1, STK39, and CNKSR2 (Fig. 3E) and for alternative exon-lacking isoforms from genes such as LAS1L, CARM1, KAT2A, MKNK1, and SETD5 (Fig. 3F). In addition, overexpression of SRSF2 could switch their splicing in the opposite direction (Supplementary Fig. S2A and S2B). The effects of overexpression were relatively modest, likely reflecting high abundance of SRSF2 in cells. Significantly, introduction of SRSF2 in knockout cells recovered WT splicing patterns for the majority of splicing events (Supplementary Fig. S2C–S2E). These data provided strong evidence that SRSF2 functions in control of endogenous transcripts, and as investigated below, can function both positively and negatively in regulating exon inclusion.

It is well documented that SRSF2 regulates splicing by binding to ESE, next we investigated whether the distribution of SRSF2-binding motifs differs between SRSF2-included and -excluded alternative exons. To this end, we simply searched the 89 validated alternative splicing targets for UCCA/UG or UGGA/UG sequences, which were described as conserved sequence motifs for SRSF2 proteins (18). SRSF2-included exons show a predominant enrichment of the TCCAG within the exons themselves over the flanking constitutive exons (Fig. 4A). However, SRSF2-excluded exons are associated with the enrichment of the TCCTG only in the flanking constitutive exons (Fig. 4B).

Next we wanted to examine whether SRSF2 binds to those exons containing sequence matches for SRSF2 in vivo, we transiently overexpressed HA-tagged SRSF2 cDNA in 293T cells. To rule out the possibility that HA antigen might cause experimental artifacts, we also transfected cells along with HA-vector, HA-SRSF10 cDNA, and HA-hnRNPK cDNA for controls (Supplementary Fig. S3A). In vivo CLIP was performed and then RT-PCR analysis was carried out with primer pairs specific for the indicated exons (Fig. 4C and D). We first analyzed 5 individual cases that displayed SRSF2-dependent exon inclusion, such as FDPS, GCH1, RBCK1, STK39, and DLG1. Significantly, binding of SRSF2 was observed within all the 5 alternative exons when compared with empty vector control (Fig. 4C, compare lanes 5 and lane 6, also with lanes 1 and 2). However, both HA-SRSF10 and HA-hnRNPK displayed weak or negligible binding to the alternative exons (Fig. 4C, compare lanes 7–8 with lane 6). More importantly and consistent with the motif analysis, affinity of SRSF2 was predominant within the alternative exons over their downstream or upstream exons, with the exception of DLG1 exon 5 (E5) and STK39 E13 (Fig. 4C). Then, we analyzed another 5 SRSF2-excluded alternative splicing events, such as MKNK1, KAT2A, LAS1L, CARM1, and FAM76A. Only with the exception of FAM76A E6, binding of SRSF2 to each alternative exon was much weaker than within the downstream or upstream constitutive exons (Fig. 4D, compare lanes 5 and 6, also with lanes 1 and 2). In contrast, either HA-SRSF10 or HA-hnRNPK did not show a strong preference for any of these constitutive exons (Fig. 4D, compare lanes 7–8 with lane 6). Consistent with the binding affinity, knockdown of SRSF10 or hnRNPK has no or limited effects on the splicing of the above-mentioned genes (Supplementary Fig. S3B–S3D). Taken together, both motif analysis and in vivo CLIP assay strongly indicated that binding of SRSF2 to the alternative exon was associated with exon inclusion, whereas binding of SRSF2 to the flanking constitutive exons was associated with exon exclusion.

To investigate whether binding of SRSF2 to the alternative exon is associated with exon inclusion, we generated a minigene reporter plasmid containing gDNA fragment of GCH1 exons 5–7 (Fig. 5A). We then cotransfected this minigene with si-SRSF2 into 293T cells and analyzed splicing by RT-PCR. Consistent with its endogenous splicing pattern, the GCH1 E6 was almost fully included in WT cells, whereas SRSF2 knockdown significantly decreased its inclusion, indicating that inclusion of GCH1 E6 was SRSF2-dependent (Fig. 5C, compare lane 1 with lane 2).

Careful observation revealed that several potential SRSF2 motifs were present in the exon 6 of GCH1 (Fig. 5A, bottom). We next examined, in more detail, the role of these motif elements in exon inclusion. To this end, we introduced a fragment deletion into the minigene construct (Fig. 5B). As shown in Fig. 5C, the F2 deletion almost completely abolished exon inclusion, which is similar to the effects caused by SRSF2 deficiency (Fig. 5C, compare lanes 5 and 6). In contrast, the other F1 deletion displayed no effect on exon exclusion and still remained responsive to SRSF2 deficiency (Fig. 5C, compare lanes 3 and 4). Then, we made 2 other constructs in which 6 copies of TCCAG (SRSF2-binding motif) or random sequences were inserted in the middle exon of the ΔE6F2 plasmid (Fig. 5B). When SRSF2-binding motifs were present in the ΔE6F2 plasmid, exon inclusion was restored nearly to the WT levels, compared to controls (Fig. 5C, compare lanes 7 and 9). However, the differential effects on exon inclusion between the 2 inserted sequences were completely diminished after SRSF2 depletion (Fig. 5C, lanes 8 and 10).

Next, we wanted to investigate the possible role of downstream or upstream SRSF2-binding motifs in the regulation of exon exclusion. We selected an exon from the MKNK1 gene, which displayed exon exclusion in the cells but showed increased inclusion on SRSF2 depletion (Fig. 5D). This effect was reproduced in the transient transfection assay with a MKNK1 minigene construct (Fig. 5F, compare lanes 1 and 2). The potential SRSF2-binding motifs were observed within the F1 or F2 fragment of downstream constitutive exon 13 (Fig. 5D, bottom). Consistently, the deletion of F1 significantly increased exon inclusion, which is even greater than effects caused by SRSF2 depletion (Fig. 5E and F, compare lanes 3 and), whereas the deletion of F2 had little effects on exon inclusion (compare lanes 5 and). Importantly, insertion of SRSF2-binding motifs in the exon 13 partially restored the WT splicing pattern of the ΔE13F1 substrate, further demonstrating that SRSF2-dependent binding motifs within the constitutive exon resulted in exon skipping (Fig. 5E and F, compare lanes 7 and 9).

Next, we asked whether SRSF2 mediates its effects in HCC via regulating alternative splicing. To this end, we compared expression patterns of SRSF2-regulated alternative splicing events between 20 HCC samples and their paired normal tissues by RT-PCR. Although there were no obvious changes in the splicing patterns of many transcripts examined, 5 revealed significant differences. Strikingly, increased inclusion of GCH1 E6 occurred in 15 of 20 HCC samples compared with control tissues. And increased inclusion of STK39 E13 was detected in 12 patients. On the other hand, increased exclusion of KAT2A E7 and increased inclusion of CNKSR2 E9 and TERF1 E6 were also detected in the 8 tumor samples (Fig. 6A).

To test whether these splice variants are functional in tumor cells, we designed isoform-specific siRNAs. For simplicity, we designated the splicing variant composing of an alternative exon as L isoform and the excluded one as S isoform. As shown in Fig. 6B, treatment of individual siRNA in Huh7 cells significantly decreased L-variant levels of genes GCH1, STK39, CNKSR2, and TERF1 or S-variant levels of KAT2A compared with control siRNA. Importantly, knockdown of these cancer-related variants for GCH1, STK39, or TERF1 dramatically repressed growth rate of both Huh7 and SK-Hep-1 cell lines, whereas knockdown against KAT2A or CNKSR2 has relatively weak effects on cell proliferation (Fig. 6C and Supplementary Fig. S4A). Consistently, cancer-related variants of GCH1, STK39, and TERF1 also play critical roles in colony formation of both Huh7 and SK-Hep-1 cells, whereas the other 2 variants have more effects on Huh7 cells than on SK-Hep-1 cells (Fig. 6D and Supplementary Fig. S4B). On the other hand, specific knockdown of the short variant of GCH1, STK39, CNKSR2, and TERF1 or the long variant of KAT2A has no obvious effects on cancer cell proliferation (Supplementary Fig. S4C–S4E). These results strongly indicated that HCC-associated splice variants could play important roles during cancer development.

Because increased inclusion of GCH1 E6 and STK39 E13 occurs in a high frequency in patients with HCC, we next aimed to investigate in detail how the function of the GCH1-L and STK39-L variants is involved in cancers. We first introduced GCH1-L/S and STK39-L/S into Huh7 cells (Supplementary Fig. S5A). In accordance with knockdown effects, overexpression of both L isoforms significantly increased colony forming efficiency of Huh7 cells, whereas both S isoforms have no obvious effects (Fig. 6E), further confirming their protumorigenic functions in hepatoma cells. Next, we wanted to characterize the tumorigenic capacity of GCH1-L–depleted cells and STK39-L–depleted cells using xenograft tumor formation in nude mice. To this end, we constructed 4 stable expression of shRNA in Huh7 cells, with 2 specifically targeted against GCH1-L and another 2 against STK39-L, and all shRNAs were shown to inhibit tumor cell proliferation in vitro (Supplementary Fig. S5B–S5F). Consistent with in vitro observation, knockdown of GCH1-L and STK39-L significantly decreased both the tumor growth rate and tumor size (Fig. 6F and G). Meanwhile, Ki-67 staining further demonstrated decreased proliferation in the xenografts derived from Huh7/shGCH1-L cells and Huh7/sh-STK39-L cells (Fig. 6H). Taken together, these data strongly indicated that both GCH1-L and STK39-L variants promote tumorigenic potential of hepatoma cells both in vitro and in mice.

Considering their functional significance in cells, we next asked whether GCH1-L and STK39-L could mediate the effects of SRSF2 in cancer cell proliferation. We first stably introduced GCH1-L or STK39-L into Huh7 cells, followed by transient transfection with si-SRSF2; however, re-expression of both L isoforms only slightly improved the survival of SRSF2-depleted cells (Supplementary Fig. S6A–S6E). This finding is not surprising, given that either GCH1 or STK39 is one among many splicing targets of SRSF2 in hepatoma cells.

Next, we generated a relatively low-level expression system by lowering levels of endogenous SRSF2 with shRNA-semi, in which only half amount of SRSF2 mRNA was depleted compared with control cells (Fig. 7A). Consistent with SRSF2 levels in cells, approximately half amount of GCH1 E6 was observed in the skipping status, compared with almost 100% skipping of exon 6 in the SRSF2-depleted cells. Likewise, skipping of STK39 E13 was observed to certain extent compared to controls (Fig. 7B). We then stably expressed GCH1-L and STK39-L in these shRNA-semi cells. As shown in Fig. 7C and D, SRSF2 knockdown led to decreased numbers of survival colonies in shRNA-semi cells, but overexpression of both L isoforms could significantly increase their tumorigenic potential whereas overexpression of the S isoforms has no obvious effects. Taken all observations together, these findings indicate that re-expression of both GCH1-L and STK39-L could significantly rescue survival defects of shRNA-semi cells, further underscoring the importance of SRSF2-mediated alternative splicing events in tumorigenesis.

To explore whether there is a positive correlation between SRSF2 expression levels and SRSF2-mediated splice variants in clinical HCC samples, RNA-seq data from a large HCC cohort were downloaded from TCGA database for further analysis (See Materials and Methods).

Given the lower sequencing depths, there is not enough reads coverage for analyzing STK39, KAT2A, TERF1, and CNKSR2 splice variants. However, consistent with our data, increased ratio of GCH1 E6 inclusion versus exclusion was indeed observed in tumor samples compared with normal tissues (Fig. 7E). More importantly, SRSF2 was again identified as differentially expressed between patients with HCC and controls. And increased inclusion of GCH1 E6 was positively associated with the expression of SRSF2 in these HCC samples (Fig. 7F), both of which are related with a higher tumor grade (Supplementary Tables S3 and S4). Together, these results strongly indicated that SRSF2 is implicated in HCC, largely reflecting its ability to control expression of cancer-related splice variants, for example, by activating inclusion of GCH1 E6 during cancer development.

In this study, our data showed that SRSF2 is frequently upregulated in clinical HCC samples, and its overexpression is significantly associated with poor prognosis of patients with HCC. SRSF2 controls cancer-associated splicing events to generate different protumorigenic protein isoforms. SRSF2 and its regulated isoforms stimulate cancer cell proliferation both in vitro and in mice. Thus, we concluded that SRSF2 regulates tumorigenesis of hepatoma cells through controlling expression of cancer-associated splice variants and further confirmed that aberrant alternative splicing is a major contributor to cancer development.

In recent years, transcriptome analysis and genome-wide studies of splicing regulators have revealed that they can modulate both exon activation and repression in highly position-dependent manner, although this positional effect exerted by individual regulators appears to be fundamentally distinct (14, 24). Here, we demonstrated that SRSF2 also displays positional effects in modulating exon inclusion in vivo. Motif analysis, CLIP assay, and functional studies conducted on a model minigene provided strong evidence that when binding within an alternative exon, SRSF2 acts as an activator to promote exon inclusion. On the other hand, we also observed that SRSF2 can repress exon inclusion in vivo and has provided mechanistic insights into this negative regulation as well. However, in some cases, we observed that the final splicing outcomes are not directly linked to the binding locations of SRSF2 within the pre-mRNA. This probably reflects the combinatory control of alternative splicing in vivo, which emphasizes cooperative and/or competitive effects between the SR proteins and other regulatory proteins on the splicing (13).

Knockdown of SRSF2 demonstrated its essential role in cancer cell survival, which is fully in line with its critical functions in the mouse liver (20). More importantly, alterations in several alternative splicing events caused by deregulated expression of SRSF2 contribute to tumor cell proliferation. The most prevalent HCC-associated alternative splicing changes were identified as the GCH1 transcripts. GCH1 encodes GTP cyclohydrolase, which is a key enzyme to produce the essential enzyme cofactor, tetrahydrobiopterin. It was previously reported that specific GCH1 polymorphisms in humans prevent the upregulation of GCH1 upon stimulation and thus are associated with longer survival times of patients (25). In this study, we observed that GCH1 was subjected to alternative splicing regulation and that SRSF2-regulated GCH1 splicing plays a critical role in tumorigenesis. Moreover, our data demonstrated that GCH1-L isoform plays a causative role in cancer progression and therefore could be a prime target for therapeutic purpose.

SRSF2 mutations on its P95 proline residue frequently occur in 20% to 30% of patients with MDS but rarely were observed in de novo patients with acute myeloid leukemia (AML; refs. 26, 27). In this study, SRSF2 was observed to be upregulated in tumor samples, and its overexpression is significantly associated with poor prognosis of patients with HCC. After sequencing gDNA extracted from clinical HCC samples, we did not detect any hot spot mutations in the SRSF2 gene (data not shown). This strongly indicated that deregulated expression of SRSF2 proteins could be a major determinant in HCC progression. Future studies will be required to investigate how the regulatory splicing networks involving SRSF2 proteins in the carcinogenesis of HCC.

In summary, we observed that upregulation of SRSF2 is significantly associated with higher tumor grade and poor prognosis of patients with HCC. Our findings further demonstrated that SRSF2 increases proliferation and tumorigenic potential of hepatoma cells by specifically controlling cancer-related splicing events.

No potential conflicts of interest were disclosed.

Conception and design: C. Luo, W. Wu, Y. Feng

Development of methodology: C. Luo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Luo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Luo, W. Wu

Writing, review, and/or revision of the manuscript: Y. Feng

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Cheng, Y. Liu, L. Chen, L. Liu, N. Wei, Z. Xie

This project was supported by grants from the National Natural Science Foundation 31570818 to Y. Feng and 31370786 to Y. Feng and 31601170 to N. Wei and 31400677 to W. Wu, and also from the "Personalized Medicine-Molecular Signature-based Drug Discovery and Development," Strategic Priority Research Program of the Chinese Academy of Sciences (XDA12010100).

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.