DNA Methylation-regulated miR-193a-3p Dictates Resistance of Hepatocellular Carcinoma to 5-Fluorouracil via Repression of SRSF2 Expression*

Background: Chemoresistance prevents effective therapy of hepatocellular carcinoma (HCC). Results: Genomic and mechanistic studies suggested the role of miR-193a-3p via SRSF2 mediates up-regulation of the proapoptotic splicing form of caspase 2 in HCC 5-FU resistance. Conclusion: We identify a novel molecular mechanism underlying 5-FU resistance in HCC. Significance: These molecular events identified provide a set of prognostic markers for future rational 5-FU therapy in HCC. Chemoresistance prevents effective cancer therapy and is rarely predictable prior to treatment, particularly for hepatocellular carcinoma (HCC). Following the chemoresistance profiling of eight HCC cell lines to each of nine chemotherapeutics, two cell lines (QGY-7703 as a sensitive and SMMC-7721 as a resistant cell line to 5-fluorouracil (5-FU) treatment) were systematically studied for mechanistic insights underpinning HCC 5-FU chemoresistance. Genomic profiling at both DNA methylation and microRNA (miR) levels and subsequent mechanistic studies illustrate a new mechanism for how DNA methylation-regulated miR-193a-3p dictates the 5-FU resistance of HCC cells via repression of serine/arginine-rich splicing factor 2 (SRSF2) expression. In turn, SRSF2 preferentially up-regulates the proapoptotic splicing form of caspase 2 (CASP2L) and sensitizes HCC cells to 5-FU. Forced changes of miR-193a-3p level reverse all of the phenotypic features examined, including cell proliferation, cell cycle progression, and 5-FU sensitivity, in cell culture and in nude mice. Importantly, the siRNA-mediated repression of SRSF2 phenocopies all of the miR-193a-3p mimic-triggered changes in QGY-7703. This newly identified miR-193a-3p-SRSF2 axis highlights a new set of companion diagnostics required for optimal 5-FU therapy of HCC, which involve assaying both the DNA methylation state of the miR-193a gene and the expression of miR-193a-3p and SRSF2 and the relative level of the proapoptotic versus antiapoptotic splicing forms of caspase 2 in clinical samples.

Chemoresistance prevents effective cancer therapy and is rarely predictable prior to treatment, particularly for hepatocellular carcinoma (HCC). Following the chemoresistance profiling of eight HCC cell lines to each of nine chemotherapeutics, two cell lines (QGY-7703 as a sensitive and SMMC-7721 as a resistant cell line to 5-fluorouracil (5-FU) treatment) were systematically studied for mechanistic insights underpinning HCC 5-FU chemoresistance. Genomic profiling at both DNA methylation and microRNA (miR) levels and subsequent mechanistic studies illustrate a new mechanism for how DNA methylationregulated miR-193a-3p dictates the 5-FU resistance of HCC cells via repression of serine/arginine-rich splicing factor 2 (SRSF2) expression. In turn, SRSF2 preferentially up-regulates the proapoptotic splicing form of caspase 2 (CASP2L) and sensitizes HCC cells to 5-FU. Forced changes of miR-193a-3p level reverse all of the phenotypic features examined, including cell proliferation, cell cycle progression, and 5-FU sensitivity, in cell culture and in nude mice. Importantly, the siRNA-mediated repression of SRSF2 phenocopies all of the miR-193a-3p mimictriggered changes in QGY-7703. This newly identified miR-193a-3p-SRSF2 axis highlights a new set of companion diagnostics required for optimal 5-FU therapy of HCC, which involve assaying both the DNA methylation state of the miR-193a gene and the expression of miR-193a-3p and SRSF2 and the relative level of the proapoptotic versus antiapoptotic splicing forms of caspase 2 in clinical samples.
Cancer is a complex disease with extensive genetic and epigenetic defects; the archives of both are rapidly updated by the second generation sequencing-based genome-wide analyses at the DNA sequence (1), DNA methylation state (2), and protein and RNA levels (3). Among the epigenetic entities, which refer to the DNA sequence-independent mechanisms underlying the cross-cell generational transmission of gene expression memory, DNA methylation (the addition of the methyl group at the cytosine ring of the 5Ј-CpG-3Ј sequence) is the best characterized and is regarded as a promising molecular indicator for the existence and/or prognostic state of cancer (4,5). Micro-RNAs (miRs) 3 are small noncoding RNAs that regulate the expression of the protein-coding genes, including the genes controlling the DNA methylation state of the genome (6), at both RNA stability and translational levels in a sequencing-specific manner. The DNA methylation state of the promoter region of a subset of miR genes negatively correlates with their transcription state (7)(8)(9). Aberrant miR expression has also been regarded as a promising biomarker for cancer detection (10,11).
Hepatocellular carcinoma (HCC) is one of the most aggressive and common malignancies, ranking second highest in terms of cancer mortality rate worldwide and in mainland China (12) (see the World Health Organization GLOBOCAN Web site). Surgical removal of tumors followed by systemic chemotherapy is a preferred treatment for patients with localized disease (13,14). Patients with advanced disease are routinely treated by transarterial chemoembolization and systemic chemotherapy using doxorubicin, cisplatin, interferon, or 5-fluorouracil (5-FU), despite the uncertain clinical benefits (14,15). New molecular agents and antibodies targeting the defective signaling pathways have also been attempted for treating HCC, so far with limited success (14,16). Therefore, there is a compelling need for a robust diagnostics for early detection and (stratification) staging of the disease to maximize clinical benefits and minimize the toxicity and cost of nonsurgical treatments. For better mechanistic understanding of the chemoresistance of HCC cancer, we carried out a chemoresistance profiling of eight HCC cell lines to each of nine conventional chemotherapeutics and performed both genome-wide analysis and mechanistic studies of a pair of HCC cell lines that differ drastically in chemoresistance to 5-FU but to none of the rest drugs analyzed. The systematic studies targeting a DNA methylation-regulated miR-193a-3p as the first candidate of several dozen informative defects were studied in detail for both its role in HCC 5-FU resistance and the mechanistic details in both cell culture and a tumor xenograft nude mouse model.
The thiazolyl blue tetrazolium blue (MTT)-based cell proliferation assay was carried out as follows. 5 ϫ 10 3 to 1 ϫ 10 4 cells in triplicate were cultured into each well of a 96-well plate (in triplicate) for 72 h, followed by a 3-h 37°C incubation after the addition of 10 l (5 mg/ml) of MTT salt (Sigma). The A 570 nm reading was obtained, and the mean and S.D. of the triplicate experiments were calculated and plotted (19). To determine the IC 50 , cells in triplicate were subjected to a series dilution of each drug for 72 h, followed the MTT-based analysis. The relative A 570 nm readings (mean and S.D.) in each drug-treated set and in the no-drug mock sample were calculated (percentage) and plotted against the logarithm of the drug concentration. The linear regression parameters were determined for each curve, and the IC 50 value was extrapolated (20)

Luciferase Reporter Assay
The seed sequences (miR-193a-3p, 5Ј-AATT TGGGTCTTT-GCGGGCGAGATGAT-3Ј and 5Ј-CTAGATCATCTCGCCCG-CAAAGACCCA-3Ј; miR-127-3p, 5Ј-AATTCTGAAGCTCAG-AGGGCTCTGATT-3Ј and 5Ј-CTAGAATCAGAGCCCTCTG-AGCTTCAG-3Ј) were annealed and inserted into the EcoRI and XbaI sites of the CMV promoter-driven firefly luciferase reporter, pCDNA3.1-luc, to make miR-193a-3p-luc and pGL3-miR-127-3p-luc (as an unrelated control) luciferase reporter constructs, respectively. The analysis mediated with Lipofectamine TM 2000 (Invitrogen) transfection/reporter was performed as described previously (21). The cells at 60% confluence were transfected in duplication by 25 ng luciferase reporter, 50 ng carrier plasmid DNA together with 5 ng of CMV-Renilla luciferase (to control transfection efficacy) luciferase activities of the transfected cells were measured after 24 h with a dual luciferase reporter system (Promega, Madison, WI) by MiniLumat LB 9506 (Berthold, Germany). The S.D. of the ratio of the firefly luciferase over the Renilla luciferase activity of the duplicates was plotted against the tested constructs. All of the experiments were carried out at least three times, and the result from one representative experiment was presented (21). The miR-193a-3p mimic and antagomir (Ribobio, Guangzhou, China) transfection was performed at dose of 10 and 50 nM, respectively.
For the cell cycle profiling, cells were seeded in 6-well plates at 40% confluence and incubated at 37°C for 24 h before transfection with the mock (non-related mimic or antagomir), miR-193a-3p mimic, or antagomir with miR-193a-3p-luc reporter/ CMV-Renilla reporter constructs. The transfected cells were collected and fixed by 70% ethanol at Ϫ20°C for 24 h, stained with 50 g/ml propidium iodide (Sigma), and analyzed on a fluorescence-activated cell sorter (FACS) according to the manufacturer's instructions (BD Biosciences). All flow cytometry experiments were performed at least three times, and a representative experiment is shown.
RNA-Total RNA was isolated using TRIzol reagent (Invitrogen). Complementary DNA synthesis was performed using a primeScript RT reagent kit (Tiangen Biotech Co., Ltd., Beijing, China) for the SYBR Green-based real-time PCR analysis in the Rotor-Gene TM 6000 system (Qiagen) of SRSF2, E2F1, ABCC8, and CASP2S/L. The expression of miR-193a-3p (Ribobio) was assessed using the 2 Ϫ ⌬⌬Ct method with RNA content normalized to the U6 RNA (Ribobio) and to ␤-actin for the other genes and plotted (22). The primers for qRT-PCR analysis of SRSF2, E2F1, ABCC8, CASP2S/L, and ␤-actin expression were synthesized by Invitrogen (supplemental Table S1).
DNA-Genomic DNA was isolated by a standard phenol/ chloroform purification method, qualified by electrophoresis on an agarose gel, and visualized with ethidium bromide. For DNA methylation analysis, the primers for methylation-specific PCR (MSP) and bisulfite sequencing PCR (BSP) were designed according to MethPrimer (supplemental Table S1) The methylation status of the miR-193a gene locus was determined by both MSP and BSP as described previously (23).
The CpG methylome was established by probing the heavily methylated DNA fraction enriched by the methylated DNAbinding domain of the rat Mecp2 (24) to the Nimblegen customized CpG island array (0.38 million probes) and validated by bisulfite sequencing as described previously (25). Solexa sequencing-based miR expression profiling was provided by BGI (Shenzhen, China).

In Vivo Study
After transfection with 50 nM miR-193a-3p or mock antagomirs (Ribobio) for 24 h, SMMC-7721 cells were collected and subcutaneously injected into the right and left flanks (1.25 ϫ 10 6 cells/point) of each nude mouse (10 total), respectively. On day 12 after cell injection, five mice intraperitoneally received 5-FU (75 mg/kg mouse body weight) once per week. The remaining five mice received phosphate-buffered saline (PBS) as a mock treatment control. Also, at the same time, the right sides of all mice tumors were intratumorally injected with 50 nM miR-193a-3p antagomir, and the left sides were injected with mock antagomir (according to the manufacturer's instructions (Ribobio)). To monitor the tumor growth, the volume was measured with a Vernier caliper on days 12, 18, 25, and 32 and calculated as volume ϭ W 2 ϫ L ϫ 0.5 (where W and L represent the largest and next largest tumor diameters (cm)) and then plotted (26). Mice were humanely sacrificed on day 32, and the tumors were weighed and photographed. Immunostaining (27) was performed for Ki67, SRSF2, and E2F1 expression in both tumors of mouse number 4 (PBS-treated) and mouse number 6 (5-FU-treated).
Expressions of SRSF2, E2F1, and Ki67 proteins were measured by immunochemical analysis on 5-mm slices of formalinfixed paraffin-embedded tumor xenografts in nude mice. Antigens were retrieved by pretreating dewaxed sections in a microwave oven at 750 watts for 5 min in a citrate buffer (pH 6) and processed with the Super Sensitive Link-Labeled Detection System (Biogenex, Menarini, Florence, Italy). The enzymatic activity was developed by using 3-amino-9-ethylcarbazole (Dako, Milan, Italy) as a chromogenic substrate. Following counterstaining with Mayer hematoxylin (Invitrogen), slides were mounted in aqueous mounting medium (glycergel, Dako). The relative level of each protein was calculated as the density score (mean intensity/lasso, TMAJ density score) that was determined by using TMAJ software (available from the Johns Hopkins Medical Institute Web site), and the percentage of the TMAJ reads of mock over the antagomir-transfected tumor was calculated and plotted.

Statistical Analysis
Data are presented as means, and error bars indicate the S.D. or S.E. All statistical analyses were performed with Excel (Microsoft, Redmond, WA) or Prism (GraphPad Software Inc., La Jolla, CA). Two-tailed Student's t test, a one-way analysis of variance or Mann-Whitney U test was used to calculate statistical significance. A p value of Ͻ0.05 was considered to be significant.

DNA Methylation-regulated miR-193a-3p Expression Correlates 5-FU Resistance of SMMC-7721 and QGY-7703-The
chemoresistance of eight HCC cell lines was determined by IC 50 profiling: QGY-7703, SMMC-7721, YY-8103, PLC, HepG2, BEL-7402, Hep3B, and FOCUS to gemcitabine, 5-FU, cisplatin, vinorelbine, docetaxel, mitomycin, paclitaxel, ironotecan, and epirubicin, respectively. The IC 50 difference among HCC cell lines to vinorelbine, docetaxel, mitomycin, paclitaxel, and epirubicin was no more than 17-fold (Fig. 1A). However, the IC 50 to 5-FU of QGY-7703 was ϳ105.2-fold (IC 50 ϭ 0.39 versus 41.01 g/ml) lower than that of SMMC-7721, whereas their IC 50 to each of the remaining therapeutics differed little (Fig. 1, B and C). HepG2 was the second 5-FU-sensitive HCC cell line (IC 50 ϭ 1.45 g/ml, 27.3-fold lower than SMMC-7721; supplemental Fig. S1A) in this group, with the other five members even more resistant to 5-FU than SMMC-7721 (the data of four cell lines are shown in supplemental Fig. S1B). Therefore, the genomic profiling at both DNA methylation and miR expression levels was carried out in both QGY-7703 and SMMC-7721 for the DNA methylation regulated miRs having a key role in the 5-FU resistance of HCC cells. Among 39 differentially expressed miRs that were identified by a Solexa sequencing-miRomic analysis, 4 miR-193a-3p was one of the most differentially expressed between SMMC-7721 and QGY-7703: 11.21-fold higher in former than in the latter by Solexa sequencing (Fig. 1D) and 4.44-fold higher by qRT-PCR-based validation (Fig. 1E). The miR-193a-3p level was also low in another 5-FU-sensitive HCC cell line, HepG2 (the second 5-FU-sensitive cell line), but high in all other 5-FU-resistant cell lines: PLC, FOCUS, Hep3B, and BEL-7402 (supplemental Fig.  S1C). We also established the methylomic profiles of both cell lines by probing a customized Nimblegene 0.38 million oligonucleotide CpG island array with the heavily methylated DNA fraction from a methylated DNA-binding domain-based capture procedure (28). The miR-193a gene was at the top of the most differentially methylated CpG islands (Ͼ3000) 4 : hypermethylated in QGY-7703 and hypomethylated in SMMC-7721 (methylation index, ϩ1.22 versus Ϫ2.35; Fig. 1F), negatively correlating with the miR-193a-3p expression. We further determined the methylation state of the miR-193a gene by BSP and MSP of these two HCC cell lines and the other five HCC cell lines and showed that miR-193a was hypermethylated in both 5-FU-sensitive cell lines (HepG2 and QGY-7703) but unmethylated or barely methylated in all five of the 5-FU-resistant cell lines, including SMMC-7721 (supplemental Fig. S1, D-F). The DNA methylation state dependence of miR-193a transcription was finally confirmed by showing that 5-aza-2Ј-deoxycytidinemediated demethylation raised the steady state level of miR-193a-3p in QGY-7703 (supplemental Fig. S2). It is evident that the 5-FU sensitivity of HCC is tightly linked to the DNA methylation-regulated miR-193a-3p expression in HCC.
Both SRSF2 and E2F1 Genes Are Bona Fide Targets of miR-193a-3p-To determine whether differentially expressed miR-193a-3p do have a functional impact, we transfected both cell lines with each of three luciferase reporter constructs: CMV-luc (empty vector control) and its derivatives miR-193a-3p-luc (the testing construct) and miR-127-luc (a nonspecific control). The latter two have each of the cognate target sequences inserted at the 3Ј-end of the luciferase genes, respectively ( Fig. 2A). Correlating well with the level of miR-193a-3p, in comparison with each of other reporter, the luciferase activity of the transfected miR-193a-luc was no more than 0.29-fold in SMMC-7721 (a miR-193a-3p high expressing cell line) (Fig.  2B), but was similar in QGY-7703 (a miR-193a-3p low expressing cell line) (Fig. 2C).
To identify the target gene(s), expression of which is repressed by miR-193a-3p at the post-transcriptional level, we performed Western blot analysis of 10 proteins in both HCC cell lines. The following five proteins were selected from 141 bioinformatically predicted miR-193a-3p target genes (see the TargetScan and EMBL-EBI Web sites): E2F transcription factor 1 (E2F1), E2F transcription factor 6 (E2F6), tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein,

miR-193a-3p Dictates HCC 5-FU Resistance
-polypeptide (YWHAZ), serine/arginine-rich splicing factor 2 (SRSF2), and myeloid cell leukemia sequence 1 (BCL2-related; MCL1). We also selected thymidylate synthetase for its role in 5-FU metabolism, protein-tyrosine kinase 2 (PTK2) for its role in cell motility and survival, histone deacetylase 1 (HDAC1) for its involvement in protein acetylation regulation, and both proliferating cell nuclear antigen and CDC5 cell division cycle 5-like (CDC5L) for their functional coupling with SRSF2 that is suggested bioinformatically (see the FunCoup Web site). Both SRSF2 and E2F1 (0.02 versus 1 and 0.31 versus 1, respectively;  1.14; Fig. 2E). Taken together, the miR-193a-3p reg-ulation of HCC sensitivity to 5-FU is probably achieved by its regulation of both SRSF2 and E2F1 expression at both mRNA stability and translation levels. The lack of an intercellular difference in thymidylate synthetase level suggested that the 5-FU metabolistic process may not impact the 5-FU resistance of HCC cells in this study.

SRSF2-mediated Elevation in the Ratio of Proapoptotic/Antiapoptotic Forms of Caspase 2 Transcripts Underlies 5-FU Sensitivity in HCC Cells
-It has been suggested that relative levels of the proapoptotic/antiapoptotic splicing forms of apoptotic genes rather than the absolute level of either reflects the apoptotic propensity and therefore chemoresistance of cancer cells (30). As a key component of the spliceosome (31), SRSF2 preferentially up-regulate the proapoptotic splice forms of apopto-FIGURE 2. E2F1 and SRSF2 are bona fide miR-193a-3p targets. A, the constructs used in this study. CMV promoter-driven construct (CMV_luc) and its derivatives (miR-193a-3p-luc and miR-127-luc) with the target sequences (filled triangle) were inserted downstream of the luciferase gene. Each firefly luciferase reporter construct was co-transfected into SMMC-7721 (B) and QGY-7703 (C) with a Renilla luciferase reporter (to control for transfection efficacy). The firefly luciferase activities (mean Ϯ S.D. (error bars)) were standardized with the Renilla luciferase activities in each transfection, and then the relative activities of each over CMV_luc (set arbitrarily as 1) were plotted. D, protein levels by Western blot analysis were digitized with ␤-actin (arbitrarily as 1). E, mRNA levels (2 Ϫ ⌬⌬Ct) by qRT-PCR analysis were standardized with ␤-actin (set arbitrarily as 1), and the ratios of E2F1 and SRSF2 mRNA to ABCC8 mRNA in SMMC-7721 versus QGY-7703 were calculated and plotted (**, p Ͻ 0.01).

miR-193a-3p Dictates HCC 5-FU Resistance
tic genes, including CASP8 and FADD-like apoptosis regulator (C-Flip); caspase 8 and 9, apoptosis-related cysteine peptidase (Caspase 8 and 9); and apoptosis regulator Bcl-X (Bcl-x) (29). To test this notion, we quantified the steady-state level of both proapoptotic (CASP2L) and antiapoptotic (CASP2S) forms of caspase 2 mRNAs in both cell lines by qRT-PCR. Positively correlating with the SRSF2 level, the CASP2L level in QGY-7703 was 1.5-fold higher than that in SMMC-7721, and the level of proapoptotic and antiapoptotic CASP2S was 0.25-fold lower than in SMMC-7721 cells. Taking both measurements into consideration, the ratio of the CASP2L/CASP2S mRNA level in QGY-7703 was 2.125-fold higher than in SMMC-7721 (Fig. 5A). In light of all that is known so far, this ratio is very likely to be a most robust indicator of the 5-FU resistance of HCC. To determine the 5-FU impact in this system, we treated both cell lines by 4 g/ml (a low dose) and 300 g/ml (a high dose) for 1 day, respectively. Neither cell lines showed any sickness under a microscope even after a high dose of 5-FU treatment. As expected, both RNA and protein of SRSF2 were elevated by a larger margin than E2F1 in both QGY-7703 and SMMC-7721 (Fig. 5, B, C, and F). Compatible with the extent of the SRSF2 up-regulation by a low dose of 5-FU, CASP2L RNA (14.62-and 7.38-fold the untreated level in QGY-7703 and SMMC-7721, respectively) was more drastically up-regulated than CASP2S (1.46-and 1.97-fold the untreated levels) in both cell lines (Fig. 5, D and E). However, both CASP2S and CASP2L mRNAs were raised by a high dose 5-FU by a similar extent in both cell lines (Fig. 5, D and E). We also conclude that the intercellular difference (3.125-fold) of the CASP2L/CASP2S mRNA level was preferentially enlarged by 5-FU treatment, particularly for a low dose.
Antagomir-mediated Suppression of miR-193a-3p Inhibited Tumor Growth and Potentiated 5-FU Sensitivity of SMMC-7721 in Nude Mice-We then compared the tumorigenicity of QGY-7703 and SMMC-7721 in nude mice. Whereas subcutaneous injection of 7.5 ϫ 10 6 QGY-7703 cells/point produced no visible tumor mass 2 weeks after the injection, 1.25 ϫ 10 6 SMCC-7721 cells/point (one-fifth of the dose of QGY-7703) resulted in visible tumor growth (not shown). This indicated that miR-193a-3p expression also involves the HCC tumorigenicity.
SMMC-7721 cells were batch-transfected with the miR-193a-3p antagomir and mock antagomir, and both SRSF2 and E2F1 protein levels were determined (Fig. 6A). Cells were subcutaneously injected into the right and left flanks of nude mice, respectively. Five mice were subjected to 5-FU or PBS administration on days 18 and 25, respectively. At the same time, the antagomir and mock were respectively intratumorally injected into the corresponding sides of all 10 mice (Fig. 6B). The tumor volume of the antagomir-transfected tumor was significantly smaller than that of the mock antagomir-transfected one on day 12 (0.05 versus 0.2 cm 3 , p ϭ 0.032), on day 25 (0.13 versus 0.45 cm 3 , p ϭ 0.039), and on day 32 (0.65 versus 1.03 cm 3 ) (Fig.  6C), and the tumor mass of the antagomir treated were lighter than that of the mock treated mice on day 32 (0.71 versus 1.14 g, Fig. 6D). 5-FU administration showed little effect on tumor  (Fig. 6C). Tumor weight on day 32 was 1.14 versus 0.82 g (Fig. 6D). However, 5-FU had a significant impact on tumor growth in the antagomir-treated SMMC-7721 cells: tumor volume on day 32, 0.65 versus 0.29 cm 3 (p ϭ 0.0427) (Fig. 6C); tumor weight on day 32, 0.71 versus 0.37 g (p ϭ 0.0084) (Fig. 6D). The further confirmation of the key role of miR-193a-3p in 5-FU resistance of HCC came from the immunostaining analysis of SRSF2, E2F1, and Ki67 (an indicator for cell proliferation) in the tumor session from 5-FU-treated (number 6) and PBS-treated (number 4) mice. As expected, antagomir treatment raised the SRSF2 level by 20% in tumor xenografts on day 32, regardless of 5-FU administration (Fig. 6, F and G). However, a 20% elevation of E2F1 was evident only in 5-FU/antagomir-treated and not in any other cases, suggesting different roles and mechanisms between E2F1 and SRSF2 in the dictation by miR-193a-3p of HCC 5-FU resistance.

DISCUSSION
Our knowledge about cancer has been rapidly expanded for last few decades. However, our ability to control this threatening disease has only been marginally improved, because a patient's survival today is only marginally prolonged compared with 40 years ago. A lack of effective measures for early detection and/or accurate stratification of the disease have been largely blamed. Requirement of companion diagnostics by the United States Food and Drug Administration for approval of newly developed drugs will soon be introduced (32), highlighting the increasing appreciation of the urgent need for robust cancer diagnostics. Among all of the biological indicators evaluated, two epigenetic identities, DNA methylation and miRs, have garnered a great deal of attention recently, for their greater potential to achieve better detection and staging of cancer in clinic. Akin to the protein-coding genes, some miR genes are transcriptionally repressed by the hypermethylated states of their promoters, particularly for those collocated with the (PBS-treated) and mouse 6 (5-FU-treated) were determined by immunostaining. G, the level of each protein was quantified as the mean density of the whole area in each slide by using TMAJ software, and then the relative density of mock-versus antagomir-treated mice was calculated and plotted (*, p Ͻ 0.05; **, p Ͻ 0.01).

miR-193a-3p Dictates HCC 5-FU Resistance
CpG-rich regions in the genome. On the other hand, the expression of DMNT3B, a key regulator of DNA methylation in mammalian cells, is also subjected to regulation by the miR-29 cluster at the level of translation (33). It is desirable, therefore, to define the molecular details for HCC chemoresistance (5-FU in this study) from both DNA methylation and miR perspectives rather than from each independently.
In this study, we carried out IC 50 profiling of eight HCC cell lines to nine common chemotherapeutics and chose QGY-7703 as a 5-FU-sensitive and SMMC-7721 as a resistant cell line for the genomic screening at both DNA methylation and miRomic levels for the corresponding differences that are specific to HCC 5-FU resistance (Fig. 1, A and B, and supplemental  Fig. S1A). Tight association of DNA methylation-regulated miR-193a-3p expression with HCC 5-FU resistance was first suggested by genomic studies (Fig. 1) and then independently confirmed by qRT-PCR of miR-193a expression and DNA methylation analysis (BSP and MSP) in this pair of HCC cell lines ( Fig. 1 and supplemental Figs. S1A and 2) and another 5-FU-sensitive cell line, HepG2, and four more resistant cell lines, PLC, BEL-7402, Hep3B, and FOCUS (supplemental Fig.  S1, A-F). Therefore, the miR-193a-3p-centered mechanistic insights detailed by the systematic studies of QGY-7703 and SMMC-7721 are truly specific to 5-FU resistance of HCC. Forced reversion of the miR-193a-3p level in both cell lines has turned around phenotypic features tested: the expression of the target genes (Figs. 2 and 3), cell proliferation, cell cycling profile, and 5-FU sensitivity ( Fig. 4 and supplemental Fig. S3). The conclusion was further supported by the observation from the in vivo studies that repression of the miR-193a-3p level by antagomir technologies repressed tumor growth and sensitized SMMC-7721 cells to 5-FU (Fig. 6). For the key mediators downstream of miR-193a-3p, we compared both protein and mRNA levels of five bioinformatically predicted miR-193a-3p targets in both cell lines, SRSF2 and E2F1, but none of the others met the expectation: higher in QGY-7703 and lower in SMMC-7721 cells (Fig. 2). Importantly, siRNA-mediated repression of SRSF2 phenocopied all of the biological changes in the miR-193a-3p mimic-transfected QGY-7703 (Fig. 4). Up-regulated expression of SRSF2 was also evident in the SMMC-7721-derived tumor xenograft by the antagomir transfection (Fig. 6, F  and G). Thus, SRSF2 relays miR-193a-3p regulation (upstream regulator) by raising the ratio of proapoptotic versus antiapoptotic splicing of the caspase 2 to dictate the HCC tumorigenicity and 5-FU resistance (Fig. 7).
Deregulation of the E2F transcription factors is a hallmark of cancer, often indicative of a poor prognosis (34). E2F1 controls transcription of multiple genes, each directly or indirectly involved in the regulation of G 1 -S phase entry (cell proliferation) or apoptosis of cells (35,36). E2F1 activation triggered by DNA damage often accompanies with the favorable changes in its chemical modification state and its interaction state with the retinoblastoma protein (37). As an E2F1 target, SRSF2 transcription is also up-regulated in the stressed cell and consequently raises the ratio of the proapoptotic to the antiapoptotic splicing form of the apoptotic genes (29). In this study, we showed for the first time that both SRSF2 and E2F1 are bona fide targets of miR-193a-3p (Fig. 2). Therefore, repression of SRSF2 expression by miR-193a-3p consists of both direct and indirect components. The E2F1-mediated component explains well why the SRSF2 always responds more profoundly than E2F1 to the change of miR-193a-3p level in cells (Figs. 2-6 and supplemental Fig. S3). The genotoxic insults may also trigger both phosphorylation and acetylation and changes in SRSF2 activity (38), representing an E2F1-independent pathway for SRSF2 activation in the DNA-damaged cells. SRSF2 is a key member of a serine/arginine-rich protein family that regulates constitutive and alternative pre-mRNA splicing for extensive diversities in gene expression that are evident across tissues, developmental stages, and diseases (39). Multiple forms of protein are often generated from a single primary transcript, having different and even opposite functions (39). The eminent examples in the cancer field are proapoptotic versus antiapoptotic isoforms of the caspase genes (29) as well as prooncogenic versus antioncogenic isoforms of BRCA1 (40) and CD44 (41). The aberrant RNA splicing in cancer has been attributed to the mutations in cis-motifs to both genetic defects (42) and aberrant expression (43) of the protein components of the mRNA splicing machinery. The stressed cells often have a high level of both RNA splicing activity (44) and the proapoptotic form of the apoptotic genes (29). Consistent with this, the ratio of the proapoptotic/antiapoptotic forms of the caspase 2 transcript in 5-FU-sensitive QGY-7703 is higher than that in resistant SMMC-7721 (Fig. 5) and rose in the 5-FU-treated HCC cells (Fig. 5). Therefore, in comparison with other events, this ratio is a better indicator for both 5-FU resistance and tumorigenicity of HCC.
Several molecular events have been implicated in the 5-FU resistance of cancer cells. For instance, the association of the FIGURE 7. A model for mechanism of action. Expression of miR-193a-3p is suppressed by the hypermethylated state of its promoter (1Ϫ). miR-193a-3p suppresses both E2F1 and SRSF2 expression at the post-transcriptional levels (2Ϫ and 3Ϫ). The function of E2F1 is activated by 5-FU treatment (5ϩ), which activates the transcription and therefore activity of SRSF2 (4ϩ). SRSF2 preferentially up-regulates the proapoptotic splicing form (6ϩ) to the antiapoptotic splicing form of caspase 2 transcripts, potentiating cell death (7ϩ). It is quite possible for the involvement of the other genes under SRSF2 regulation at the alternative mRNA splicing level to be in control of 5-FU chemoresistance and the in vivo tumor growth (8Ϫ). FEBRUARY 17, 2012 • VOLUME 287 • NUMBER 8 JOURNAL OF BIOLOGICAL CHEMISTRY 5647 overexpressed astrocyte elevated gene-1 (AEG-1) with 5-FU resistance of HCC cells was ascribed to the ability of AEG-1 to activate the transcription the genes that catabolize 5-FU (45). Other indicators for resistance of cancer cells to 5-FU include dysregulated Ep-CAM (46), Hsp27, and Hsp40 proteins (47). miR-21 represses hMSH2 expression (26), which was reported to be overexpressed in colorectal cancers that are refractory to 5-FU therapy. Taken together, we showed for the first time that miR-193a-3p regulates the 5-FU resistance of HCC, and its transcription is repressed by the hypermethylated promoter state. Both SRSF2 and E2F1 are true targets of miR-193a-3p, and SRSF2 is the key mediator to dictate the 5-FU resistance in HCC. Both tumorigenicity and 5-FU resistance of HCC cells are directly associated with the ability of SRSF2 to elevate the ratio of the proapoptotic/antiapoptotic forms of the caspase 2 transcript. Finally, the observation detailed in this report suggests a list of novel candidates as prognostic indicators for rational 5-FU therapy of HCC: DNA methylation state or the levels of miR-193a-3p, SRSF2, E2F1, and the proapoptotic form of caspase 2.