Signaling between Transforming Growth Factor β (TGF-β) and Transcription Factor SNAI2 Represses Expression of MicroRNA miR-203 to Promote Epithelial-Mesenchymal Transition and Tumor Metastasis*

Background: TGF-β promotes tumor metastasis by inducing SNAI2. Results: TGF-β induces SNAI2 and promotes EMT by repressing miR-203. Conclusion: The SNAI2 and miR-203 feedback loop plays integral roles in EMT and tumor metastasis. Significance: Our findings provide new insights into molecular mechanisms of EMT and tumor metastasis and identify a therapeutic target for tumor invasion and metastasis. TGF-β promotes tumor invasion and metastasis by inducing an epithelial-mesenchymal transition (EMT). Understanding the molecular and epigenetic mechanisms by which TGF-β induces EMT may facilitate the development of new therapeutic strategies for metastasis. Here, we report that TGF-β induced SNAI2 to promote EMT by repressing miR-203. Although miR-203 targeted SNAI2, SNAI2 induced by TGF-β could directly bind to the miR-203 promoter to inhibit its transcription. SNAI2 and miR-203 formed a double negative feedback loop to inhibit each other's expression, thereby controlling EMT. Moreover, we found that miR-203 was significantly down-regulated in highly metastatic breast cancer cells. The restoration of miR-203 in highly metastatic breast cancer cells inhibited tumor cell invasion in vitro and lung metastatic colonization in vivo by repressing SNAI2. Taken together, our results suggest that the SNAI2 and miR-203 regulatory loop plays important roles in EMT and tumor metastasis.

Tumor metastasis is the main cause of mortality in patients with solid tumors (1). Epithelial-mesenchymal transition (EMT) 2 has been considered as an early step in tumor metastasis (2). EMT is a process in which cells lose their epithelial features and acquire mesenchymal properties, including cytoskeleton remodeling, epigenetic changes, and increased invasiveness (3). A hallmark of EMT is the loss of E-cadherin (CDH1), a defining marker of epithelial cells. E-cadherin is a transmembrane glycoprotein that forms the core of adherens junctions between the adjacent epithelial cells and plays a critical role in cell-to-cell adhesion (4). Both in vitro and in vivo studies have shown that the loss of E-cadherin is linked to cancer progression and metastasis (5)(6)(7). A family of E-box-binding transcription factors, including SNAI1, SNAI2, ZEB1, ZEB2, and TWIST1, have been found to promote EMT and tumor metastasis by repressing CDH1 (8). Although SNAI2 is considered a homologue of SNAI1, several studies suggest that SNAI2 may play unique roles in tumor invasive growth and metastasis. For example, SNAI2 inhibited apoptosis by repressing p53-mediated transcription (9). SNAI2, but not SNAI1, was found to inversely correlate with the expression of E-cadherin in breast cancer cells (10). Clinical studies found that SNAI2 was highly expressed in metastatic breast cancer and correlated with poor prognosis of breast cancer (11). Moreover, overexpression of SNAI2 has been found to induce ZEB1 and/or ZEB2, suggesting that SNAI2 may cross-talk with other regulators of EMT (12). TGF-␤ is considered as the major upstream inducer of EMT, and increasing evidence implicates TGF-␤ signaling as a key effector of EMT in cancer progression and metastasis (13). However, the precise molecular mechanism by which TGF-␤ signaling induces EMT and how TGF-␤ signaling is regulated are still obscure. Therefore, understanding the TGF-␤ signaling network may help to develop novel therapeutic strategies for metastasis.
MicroRNAs (miRNAs) are small non-coding RNAs that regulate a variety of biological processes by modulating gene expression at the post-transcriptional level (14). Growing evidence shows that miRNAs play an important role in the control of cancer cell invasion and metastasis (15)(16)(17). For example, miR-335 was found to inhibit metastasis by targeting SOX4 and tenascin C (16). miR-10b induced by TWIST stimulated tumor cell migration and invasion (17). The miR-200 family, including miR-200b, miR-200a, miR-409, miR-200c, and miR-141, formed a feedback loop with ZEB1 and ZEB2 to regulate EMT and tumor metastasis (18,19). miR-203 inhibited breast cancer invasion by targeting SNAI2 (20). Interestingly, using a combination of bioinformatics and functional analyses, we found that TGF-␤ induced SNAI2 to suppress miR-203 in EMT. SNAI2 and miR-203 formed a double negative feedback loop to inhibit each other's expression, thereby controlling EMT and tumor invasive growth and metastasis. Repression of miR-203 was required for TGF-␤-induced EMT. Moreover, the induction of SNAI2 by TGF-␤ controlled the miR-200 and ZEB1/2 regulatory loop that is an important regulator of EMT, suggesting that SNAI2 plays an integral role in tumor invasion and metastasis. Our clinical analysis and experimental models demonstrate that the SNAI2 and miR-203 loop is an attractive therapeutic target for tumor metastasis.

EXPERIMENTAL PROCEDURES
Cell Culture and Viral Transduction-Hs-578T, MCF7, MDA-MB-468, MDA-MB-231, T47D, HEK293T, and MDCK cells were maintained in DMEM containing 10% FBS and antibiotics. BT-549 and BT-474 were maintained in RPMI 1640 medium containing 10% FBS and antibiotics. SUM159 was purchased from Asterand Co. and cultured according to the manufacturer's instructions. For TGF-␤ treatment, MDCK was treated with 5 ng/ml recombinant human TGF-␤1 (Pepro-Tech) for the different days. During TGF-␤ treatment, cells were split every 2 or 3 days depending on the cell confluence.
For viral transduction, retroviruses were generated by cotransfection of pQNCX2 EV or pQNCX2-SNAI2 with packaging plasmids into HEK293T cells as described previously (21,22). MDA-MB-468 cells were transduced with these virus particles and subsequently selected with 60 g/ml neomycin for at least 10 days. Lentiviruses overexpressing miR-203 were packaged and generated in 293T cells according to the manufacturer's instructions (System Biosciences). MDA-MB-231 and MDCK cells were infected and subsequently subjected to FACS for green fluorescent protein (GFP) to obtain the stable cell lines. The target sequence for SNAI2 shRNA was 5Ј-GCATTT-GCAGACAGGTCAAAT-3Ј, and the shRNA was subcloned into pLKO lentiviral vectors. The lentiviruses expressing SNAI2 shRNA were packaged and generated in 293T cells as described previously (20,21). A sponge insert containing 8ϫ tandem "bulged" (at positions 9 -12) miR-203 binding sites in PUC57 was purchased from Genescript. The pBabe-d2eGFP miRNA sponge backbone was made by digesting pBabe-d2eGFP-miR-9 (Addgene) with XhoI and ApaI (removing the miR-9 sponge). The miR-203 sponge inserts were subcloned into the pBabe-d2eGFP vector.
Plasmids and Luciferase Reporter Assays-SNAI2 ORF was amplified from the MDA-MB-231 complementary DNA (cDNA) library and then cloned into pQNCX2 with or without an HA tag in the C terminus. The full-length 3Ј-UTR of SNAI2 was amplified from the MDA-MB-231 cDNA library and then cloned into the pmiR-REPORT luciferase construct (Ambion). The promoter region of miR-203, miR-200b, and miR-200c was amplified from MCF7 genome DNA and subcloned into pGL3-Basic vector (Promega). Mutations in the above constructs were generated using the QuikChange site-directed mutagen-esis kit (Stratagene). The pmiR-203 construct and the corresponding empty vector were purchased from System Biosciences. All of these constructs were verified by DNA sequencing. For transient transfections, cells of 50% confluence in 24-well plates were transfected using Lipofectamine 2000 reagents according to the manufacturer's protocol (Invitrogen). For the 3Ј-UTR reporter assay, 10 ng of reporter plasmids together with 150 ng of control plasmids or pmiR-203 plasmids and 2 ng of pmiR-␤-gal plasmids (Applied Biosystems) were co-transfected into 293T cells. For the miRNA promoter-reporter assays, 50 ng of reporter plasmids together with 5-50 ng of pQNCX2-SNAI2 plasmids and 2 ng of ␤-pmiR-␤-gal plasmids were co-transfected into 293T cells. The total amount of DNA in each individual well was kept constant by adding pQNCX2 empty vector. Luciferase and galactosidase activities were measured 24 -48 h after transfection using the Dual-Luciferase reporter assay system (Promega) and GalactoStar kits (Tropix), respectively. All experiments were performed in triplicate with data pooled from at least two independent experiments.
Transfection of siRNA and miRNA and Real Time RT-PCR-For transient knockdown of SNAI2, 30 nM siRNA or control siRNA (Dharmacon) was transfected using RNAiMAX (Invitrogen) following the manufacturer's instructions. For transient overexpression of miR-203, 30 nM pre-miR-203 (Ambion) was transfected using RNAiMAX. In each case, total RNA and protein were collected for assay 3 days post-transfection. For functional assays, cells were plated for Matrigel invasion assays 2 days after transfection.
Total RNA from cultured cells was isolated using the mir-Vana miRNA isolation kit according to the manufacturer's instructions (Ambion). Mature miRNA expression was quantified using TaqMan miRNA assays according to the manufacturer's instructions (Applied Biosystems). miRNA expression was normalized to U6 small nuclear expression. Real time PCR was performed as described previously (21).
ChIP Assay and Human Affymetrix Microarray-ChIP assays were carried out using a ChIP assay kit based on the manufacturer's instructions (Upstate Biotechnology). Cells were incubated with a dimethyl 3,3Ј-dithiobispropionimidate HCl (Pierce) solution (5 mM) for 10 min at room temperature, and then formaldehyde was added. For each ChIP reaction mixture, 2 ϫ 10 6 cells were used. The resulting precipitated DNA samples were quantified by real time PCR with specific primer sets (supplemental Methods). The antibodies for ChIP assays were purchased from the following sources: anti-H3AC, anti-H3TriMeK4, anti-H3DiMeK9 (Millipore, CS200587), and anti-HA (Abcam, ab9110). Data are expressed as the percentage of input DNA.
For microarray analysis, we extracted total RNA using an miRNeasy kit based on the manufacturer's protocol (Qiagen). 5-g aliquots of total RNAs from each sample were transcribed to double-stranded cDNA using SuperScript II reverse transcriptase (Invitrogen) with an oligo(dT) primer and then used to generate single-stranded RNAs. The biotin-labeled RNAs were fragmented and hybridized with an Affymetrix Human Genome U133 Plus 2.0 Array. We scanned the arrays with the GeneArray scanner (Affymetrix) and used the robust multichip average method to normalize the raw data. Microarray data have been deposited in the Gene Expression Omnibus (GEO) under accession number GSE44239.
Matrigel Invasion Assays and Mouse Xenografts-Matrigel invasion assays were performed using BD BioCoat Matrigel invasion chambers as described previously (21). Briefly, 1 ϫ 10 5 cells suspended in 0.5 ml of serum-free medium were plated in the upper chamber. In the lower chamber, 0.75 ml of medium containing 0.1% FBS was added as a chemoattractant. After 24 -48 h, Matrigel was removed, and the cells that had invaded the lower chamber were stained with a HEMA-3 kit (Fisher). Cell numbers from four random fields were counted and averaged. Each group was performed in triplicates. All animal experiments were performed in accordance with a protocol approved by the UCLA Animal Research Committee. For lung metastasis, cells (1 ϫ 10 6 ) were injected into the tail vein of mice (five mice per group). After 2 months, mice were euthanized, the lungs were resected, and nodules were counted.
Bioinformatics and Statistical Analysis-To search genes coexpressed with CDH1, the normalized data of mRNA and miRNA were downloaded from the Genomics and Bioinformatics Group website and combined into a single database. The GeneNeighbours module in the GenePattern program (23) was used to perform the search for genes co-expressing with CDH1 in our combined database. For gene expression correlation analysis in clinical samples, two independent databases containing both mRNA and miRNA normalized expression data were used. The first one had 91 samples from different cancer types, and the second one had 139 samples from prostate cancer. The Spearman correlation coefficient was calculated between CDH1 and miR-203 in an individual data set using SPSS 17.0 software. To compare miR-203 expression in primary and metastatic breast cancers, the raw data of a published data set (E-TABM-971) (24) was downloaded from European Bioinformatics Institute ArrayExpress. The global median method in BRB-ArrayTools (Richard Simon and Amy Peng Lam, National Cancer Institute, National Institutes of Health, Bethesda, MD) was used to normalize the raw data. The above mentioned prostate data set was also used to compare the miR-203 expression between prostate primary and metastatic cancers. For relapse-free survival analysis, the prostate data set, which provide detailed patient follow-up information, was used again. Patients were dichotomized into two groups as above or below median for the miR-203 expression value. A Kaplan-Meier survival plot was performed for relapse-free survival, and the log rank test was applied to test the survival differences and calculate the p value using SPSS 17.0 software. TargetScan (25) and miRDB (26) were used to predict the target genes of miR-203.

miR-203 Is Repressed in TGF-␤-induced EMT-Because
CDH1 is suppressed by TGF-␤-SNAI2 signaling in EMT (8), it is possible that genes that were regulated by TGF-␤ might be co-expressed with CDH1. We interrogated the array data set composed of mRNA and miRNA of NCI60 cells using the GeneNeighbours module in the GenePattern program (23,27,28). The top 20 candidate genes, consisting of 13 coding mRNA genes and seven miRNA genes, were selected for further analysis (Fig. 1A). We found that the expression profile of miR-203 was closest to the expression of CDH1 among all miRNA genes in the NCI60 cells, including the miR-200 family. Correlation analysis showed a significant positive correlation between the expression of miR-203 and the expression of CDH1 at both mRNA and protein levels (supplemental Fig. S1, A and B). To confirm this finding, we examined the expression of miR-203 in eight breast cancer cell lines using quantitative real time PCR. Interestingly, the expression of miR-203 in four lowly invasive breast cancer cell lines (MCF7, MDA-MB-468, BT474, and T47D) with high CDH1 expression was several hundred times higher than in four highly invasive breast cancer cell lines (MDA-MB-231, BT549, Hs578T, and SUM159) with low CDH1 expression (supplemental Fig. S1C). Furthermore, to investigate whether this correlation also existed in human tumor tissues, we analyzed two published human tumor tissue expression data sets containing both mRNA and miRNA expression data (29,30). The first data set consisted of tumors from six different tissues, including breast, whereas the second included only prostate tumors. In both data sets, we found that the expression of miR-203 was positively correlated with the expression of CDH1 ( Fig. 1B and supplemental Fig. S1D). Altogether, these data indicated that miR-203 was co-expressed with CDH1 in both cancer cell lines and primary cancer tissues.
To test whether the expression of miR-203 could be suppressed upon TGF-␤ treatment, we used a classical EMT model system: the MDCK cell, which can be induced to undergo EMT in response to TGF-␤ exposure. TGF-␤ treatment in MDCK cells led to a gradual morphological change (Fig. 1C) that was accompanied by the loss of CDH1 and induction of CDH2, vimentin (VIM), and FN1 (Fig. 1, D and G). TGF-␤ also potently induced the expression of SNAI2, ZEB1, and ZEB2 but not SNAI1 (Fig. 1E). These hallmark shifts at the morphological and molecular levels indicate that EMT was induced in MDCK cells by TGF-␤. Importantly, we found that the level of miR-203 in MDCK cells was significantly reduced 2 days after TGF-␤ treatment, which is earlier than the reduction of CDH1 and later than the induction of SNAI2 (Fig. 1F). The miR-200 family was also down-regulated after TGF-␤ treatment (Fig. 1H). While our work was in progress, it was shown that loss of miR-203 was associated with cell invasion (31). To investigate whether the down-regulation of miR-203 was required for TGF-␤-induced EMT, we stably expressed miR-203 in MDCK cells. Interestingly, we found that overexpression of miR-203 prevented the morphology change induced by TGF-␤ (Fig. 1I). Western blot analysis showed that ectopic expression of miR-203 potently prevented the induction of SNAI2 and the inhibition of CDH1 mediated by TGF-␤. Interestingly, we found that miR-203 also inhibited the induction of ZEB1 induced by TGF-␤ (Fig. 1J). Taken together, these results suggest that the inhibition of miR-203 was required for TGF-␤-induced EMT.

miR-203 Targets the SNAI2-E-cadherin Axis and Inhibits
Cancer Metastasis-To investigate the mechanism through which miR-203 was co-expressed with CDH1 and regulated the TGF-␤-induced EMT, we searched for targets of miR-203 that might mediate this effect using two different searching programs, TargetScan (25) and miRDB (26). Meanwhile, we made a stable cell line (MB231/miR-203) that overexpressed miR-203 and compared the gene expression profile between MB231/EV and MB231/miR-203 cells using a microarray. We found that 177 genes were down-regulated in MB231/miR-203 cells compared with MB231/EV cells (only a Ͼ0.5-fold difference was considered). 25 among these 177 genes were found to be the predicted targets by two different programs (Fig. 2A). The most interesting predicted target by the searching program and by  the microarray analysis was SNAI2, which is one of the key CDH1 repressors. A 7-mer site recognized by miR-203 was found in the human SNAI2 3Ј-UTR that was evolutionarily conserved among diverse species (supplemental Fig. S2A). Consistent with the findings of Zhang et al. (20), our luciferase assay also confirmed that SNAI2 is a direct target of miR-203 (supplemental Fig. S2, B and C). We transiently expressed pre-miR-203 in both MDA-MB-231 and Hs578T cells in which SNAI2 is highly expressed and both CDH1 and miR-203 are repressed. Western blot analysis revealed that the SNAI2 proteins were significantly decreased after transient transfection of pre-miR-203 in both MDA-MB-231 and Hs-578T cells (Fig. 2, B and C). Consistently, real time RT-PCR showed that transient transfection of pre-miR-203 also restored CDH1 expression. To further examine whether the repression of SNAI2 by miR-203 mediated these effects, we made MDA-MB-231 cells stably expressing miR-203 alone (MB231/miR-203) or together with SNAI2. Although a significantly increased CDH1 expression accompanied by decreased SNAI2 was observed in MB231/miR-203, miR-203-induced CDH1 was abolished by the reintroduction of SNAI2 (Fig. 2D). Taken together, our results demonstrate that mi-R203 targets SNAI2 to regulate CDH1 expression.
The invasive growth of solid tumors is critical for distant metastasis. The hallmarks of EMT are the functional loss of CDH1 and increased invasion. SNAI2 can promote cell migration and invasion through suppression of CDH1 and an increase of matrix metalloproteinase expression (32). Interestingly, recent studies by Korpal et al. (33) showed that the miR-200 family promoted metastatic colonization by inducing E-cadherin and targeting the Sec23a-mediated tumor cell secretome, casting doubt on the potential therapeutic utility of miR-200s. They showed that overexpression of miR-200 promoted lung metastasis in a tail vein injection assay. To directly confirm the repressive role of miR-203 in metastasis, we examined whether the restoration of miR-203 could inhibit human breast cancer cell metastasis using the same model. Overexpression of miR-203 in both MDA-MB-231 and Hs578T cells significantly decreased their invasion as determined by Matrigel invasion assays (Fig. 2E). Also, the tail vein injection revealed that, unlike miR-200, miR-203 strongly inhibited MDA-MB-231 cell lung colonization in vivo (Fig. 2F). Interestingly, we noted that the restoration of SNAI2 only partially increased cell invasion in MB231/miR-203 cells by Matrigel invasion assays (Fig. 2G), suggesting that miR-203 may also inhibit invasion by targeting other molecules in addition to the SNAI2-E-cadherin axis. Two recent reports showed that miR-203 suppresses cell migration and invasion of breast cancer and squamous cell carcinoma through targeting LASP1 (34,35). Interestingly, we found that LASP1 is on the potential target gene list based on our microarray ( Fig. 2A).
Because TGF-␤ inhibited miR-203 in EMT, we examined whether miR-203 was required for maintaining the epithelial property. We used miRNA sponges to stably knock down miR-203 in MCF7 cells, which have high levels of miR-203. Western blot showed that knockdown of miR-203 by miR-203 sponges in MCF7 significantly increased the expression of SNAI2 and decreased the expression of CDH1 (Fig. 3A). The knockdown of miR-203 also increased the invasion capability of MCF7 (Fig. 3B). The knockdown of miR-203 in MDCK cells drastically induced morphological changes. Western blot analysis revealed that the knockdown of miR-203 resulted in the induction of VIM and SNAI2 and the loss of CDH1 (Fig. 3, C and D), suggesting that miR-203 is required for maintaining the epithelial property of MDCK cells.
SNAI2 Directly Inhibits miR-203 to Promote EMT-Increasing evidence has shown that a feedback circuit exists between miRNA and the transcription factor target to enhance the robustness of gene regulation in mammalian genomes (40). For unknown reasons, we were unable to efficiently knock down miR-203 in MDA-MB-468 cells. To investigate whether SNAI2 repressed miR-203 in a reciprocal regulatory circuit containing SNAI2 and miR-203, we overexpressed the full-length SNAI2 cDNA, lacking the 3Ј-UTR, in lowly metastatic MDA-MB-468 cells. Overexpression of SNAI2 in MDA-MB-468 cells (MB468/ SNAI2) resulted in a morphological change from a rounded compact shape to a loose spindle shape indicative of EMT. This was accompanied by a loss of CDH1 and gain of VIM as revealed by immunofluorescence staining (Fig. 4A). The reduction of CDH1 and induction of VIM were further confirmed by Western blot analysis (Fig. 4B). A dramatic decrease of miR-203 was observed in SNAI2-expressing MDA-MB-468 cells, indicating that miR-203 might be the target of SNAI2 (Fig. 4C).
Interestingly, overexpression of SNAI2 also induced the expression of endogenous SNAI2 as determined by real time RT-PCR with a specific set of primers targeting the 3Ј-UTR of SNAI2 mRNA (Fig. 4D). To determine whether endogenous SNAI2 could repress the expression of miR-203, we knocked down the expression of SNAI2 in both MDA-MB-231 and Hs578T cells using siRNA. Compared with scrambled siRNA, we found that the knockdown of SNAI2 potently stimulated miR-203 expression in both MDA-MB-231 and Hs578T cells (Fig. 4, E and F).
To investigate how TGF-␤ inhibited miR-203, we characterized the primary precursor transcript of miR-203. The miR-203 gene was localized on human chromosome 14 surrounded by several ESTs spanning around 2000 bp as annotated on the UCSC Genome Browser (36) (Fig. 5A). Six overlapping ESTs matched around 500 bp downstream of the miR-203 hairpin were found, and a canonical poly(A) signal, AATAAA, was found in EST BE551807, indicating that the 3Ј boundary of this primary transcript could be identified. A transcription start site (TSS) predicted by the Eponine method (37) was found close to the 5Ј terminus of the EST that matched the upstream region of the miR-203 hairpin. Comparative genomic analysis using the VISTA program (38) showed that the region around the predicted TSS is highly conserved among human, mouse, and rat ( Fig. 5A). Furthermore, the predicted TSS is also supported by the presence of an overlapping CpG island (39) that is known to co-localize with the TSS. Collectively, through our analysis, the primary transcript of miR-203 is around 2 kb in length with no significant open reading frame (ORF) in it.
To investigate whether SNAI2 directly repressed the transcription of miR-203, we cloned the putative promoter of miR-203, which spans from 700 bp upstream to 300 bp downstream of the predicted TSS, into a luciferase reporter construct. Cotransfection experiments showed that SNAI2 suppressed the luciferase activity of the reporter construct in a dose-dependent manner (Fig. 5B). The promoter region of miR-203 contains three putative binding sites for SNAI2 (E-boxes). Mutation of the first two E-boxes, but not the third E-box, significantly rescued the repression effect by SNAI2, indicating that these two E-boxes are responsible for SNAI2 binding (Fig. 5C). To further confirm this binding, we performed ChIP assays. Because of the lack of commercial ChIP-grade SNAI2 antibody, we made MDA-MB-468 cells stably expressing SNAI2-HA (MB468/ SNAI2-HA). ChIP assays using HA antibodies confirmed that SNAI2 bound to the paired E-box region near the TSS of miR-203 (Fig. 5D). These results suggest that SNAI2 can directly suppress the expression of miR-203 by binding to the paired E-box in the promoter region of miR-203. Taken together, miR-203 and SNAI2 could repress each other and constitute a double negative feedback loop during EMT.

SNAI2 Directly Inhibits miR-200-Recently
, several reports demonstrated that the miR-200 family was an important epithelial marker and was lost during the EMT process. The miR-200 family targets ZEB1 and ZEB2, and inhibition of the miR-200 family has been found to induce EMT through upregulation of ZEB1 and ZEB2 (18,19,(41)(42)(43). Interestingly, a recent report showed that SNAI2 also formed the feedback loop with miR-200 to regulate EMT in prostate cancer (44). Therefore, we examined whether SNAI2 also inhibited miR-200 in breast cancer cells. Real time RT-PCR showed a substantial reduction in miR-200 family members after the induction of EMT in MDA-MB-468 cells by SNAI2 (Fig. 4C). Consistently, the expression of ZEB1 and ZEB2 was increased in SNAI2expressing MDA-MB-468 cells (Fig. 6A). To examine whether SNAI2 could directly repress the miR-200 family, we performed both luciferase and ChIP assays. Luciferase assays showed that SNAI2 could repress the promoter activity of the miR-200 family in a dose-dependent manner (Fig. 6B). ChIP assays also showed that SNAI2 could bind to the promoter of the miR-200 family (Fig. 6, C and D). Our results suggest that these factors may act in concert to elicit the EMT process.
TGF-␤ Inhibits miR-203 through the Induction of SNAI2-Based on the aforementioned observation that SNAI2 could repress the expression of miR-203, we asked whether the downregulation of miR-203 by TGF-␤ was dependent on the induction of SNAI2. We stably knocked down SNAI2 in MDCK cells using shRNA. As shown in Fig. 7, A and B, knockdown of SNAI2 prevented the EMT induced by TGF-␤, suggesting that induction of SNAI2 by TGF-␤ is essential for TGF-␤-induced ⌭⌴⌻ in MDCK cells. The knockdown of SNAI2 also inhibited TGF-␤-mediated suppression of miR-203 (Fig. 6C). Luciferase assays also showed that the promoter activity of miR-203 was inhibited upon TGF-␤ treatment and that both mutation of the E-box and knockdown of SNAI2 could prevent this inhibition (Fig. 6D), suggesting that the integrity of the E-box in the promoter of miR-203 is important for the down-regulation of miR-203 by SNAI2 upon TGF-␤ treatment.
Because our result showed that the restoration of miR-203 in MDCK cells inhibited TGF-␤-induced EMT, we examined whether miR-203 could prevent the down-regulation of the miR-200 family by TGF-␤. Real time RT-PCR found that the ectopic expression of miR-203 suppressed the down-regulation of CDH1 and the miR-200 family induced by TGF-␤ (Fig. 7,  E-H). Taken together, our results suggest that the miR-203 and SNAI2 double negative feedback loop is a master regulator of EMT and tumor metastasis.
miR-203 Is Lost in Human Metastatic Tumor Samples-Our data established that the SNAI2-miR-203 double negative feedback loop regulated CDH1 expression and tumor invasion and metastasis in breast cancer cells. To determine whether loss of miR-203 was associated with human tumor metastasis, we compared the miR-203 expression between the primary and metastatic samples in a publicly available breast tumor data set (24). miR-203 expression was significantly decreased in human metastatic breast tumor samples compared with human primary breast tumor samples (Fig. 8A). Finally, we explored whether loss of miR-203 was involved in the metastasis of other tumors and associated with cancer progression and prognosis.
We interrogated a prostate tumor data set (30) that provided detailed clinical information and found that down-regulation of miR-203 also occurred in the metastatic prostate tumors (Fig.  8B). Interestingly, survival analysis showed that loss of miR-203 was significantly correlated with reduced recurrence-free survival (Fig. 8C).

DISCUSSION
In this study, we identified that SNAI2 and miR-203 form a new feedback regulatory loop that plays an important role in TGF-␤-induced EMT and tumor metastasis. Although miR-203 targeted SNAI2 (20), SNAI2 repressed miR-203 by directly binding to its promoter. The loss of miR-203 and gain of SNAI2 were associated with human breast cancer invasion and metastasis. Moreover, we demonstrated that repression of miR-203 was indispensable for EMT induced by TGF-␤. The SNAI2 and miR-203 loop also controlled the expression of miR-200 and ZEB1/2. Given the fact that miR-200 and ZEB1/2 have been found to play a critical role in EMT, our results suggest that the miR-203 and SNAI2 double negative feedback loop is a critical regulator of EMT and tumor metastasis.
A recent study showed that miR-203 could regulate EMT by forming a feedback loop with SNAI1 in the breast cancer cell line HTB129 (46). Overexpression of miR-203 down-regulated SNAI1 and promoted epithelial-like properties. In contrast, we found that miR-203 formed a feedback loop with SNAI2 in TGF-␤-induced EMT. SNAI2 was inversely associated with the expression of miR-203. This difference may be cell context-dependent. Nevertheless, these results suggest that miR-203 plays an important role in maintaining the epithelial properties of the cells. Interestingly, miR-203 was found to be significantly down-regulated in the mesenchymal component of endome- trial carcinosarcoma, which represents a bona fide example of EMT in vivo (47), and in claudin-low type breast cancer, which contains a majority of mesenchymal-like cells (48). Previously, the miR-200 family was found to inhibit EMT by targeting ZEB1 and ZEB2, suggesting that the miR-200 family has therapeutic value. On the contrary, Korpal et al. (33) recently found that the miR-200 family promoted metastatic colonization by inducing E-cadherin and targeting the Sec23a-mediated tumor cell secretome using the tail vein injection model. Their findings cast doubts on the potential therapeutic utility of miR-200s (33). Using the same model, we demonstrated that the restoration of miR-203 in human breast cancer cells could inhibit metastasis. Although both miR-200s and miR-203 inhibit EMT, our results suggest that miR-203 may be a more attractive target for tumor metastasis. Although EMT is required for the early steps of tumor metastasis, mesenchymal-to-epithelial transition plays a critical role in metastatic tumor growth (48). Therefore, miR-203 expression should be restored in metastatic tumors. However, analysis of publicly available breast tumor data sets revealed that miR-203 was significantly decreased in human metastatic breast tumors. Currently, we can provide an explanation for this inconsistency. It is possible that miR-203 may not be required for mesenchymal-to-epithelial transition during metastatic tumor growth. As shown in our microarray, miR-203 might target heparin-binding EGF, a potent mitogen for tumor growth (Fig. 2A). Wang et al. (34) showed that miR-203 inhibited baculoviral IAP repeat-containing protein 5 (BIRC5) to reduce cancer proliferation. Therefore, the down- regulation of miR-203 is important for metastatic tumor growth.
The inhibition of CDH1 by SNAI2 plays a critical role in EMT and tumor metastasis. However, repression of CDH1 could not fully explain tumor invasive growth and metastasis induced by SNAI2. Our studies demonstrated that the inhibition of miR-203 by SNAI2 is also critical for developmental EMT, providing novel insights into SNAI2-induced EMT. Given the pleiotropic roles of miRNAs, the inhibition of miR-203 by SNAI2 could affect a cohort of genes associated with tumor invasive growth and metastasis. Thus, although SNAI2 is a transcriptional repressor, it could directly up-regulate gene expression by inhibiting miR-203, thereby promoting tumor metastasis. Previous reports showed that overexpression of SNAI1/2 up-regulates ZEB1 by unknown mechanisms (12). A recent study showed that SNAI2, but not SNAI1, induces ZEB1 expression by directly binding to the promoter of ZEB1 (49). We found that SNAI2 also induced ZEB1 and ZEB2. However, given the fact that SNAI2 is a transcriptional repressor, it is unlikely that SNAI2 directly induces ZEB1 transcription. Interestingly, we demonstrated that the miR-200 family members, which are negative regulators of ZEB1 and ZEB2, were directly repressed by SNAI2, further amplifying EMT. Therefore, an alternative mechanism was that SNAI2 indirectly up-regulated ZEB1 and ZEB2 by repressing the miR-200 family. Moreover, we found that the restoration of miR-203 suppressed the downregulation of miR-200 and up-regulation of ZEB1 mediated by TGF-␤. Our results suggest that the SNAI2 and miR-203 regulatory loop may be in concert with miR-200 and ZEB1/2 to form a feed-forward loop to regulate EMT and gene expression (18,19,41,42). In summary, the SNAI2-miR-203 double negative feedback loop identified here provided new insights into the molecular mechanisms of EMT and tumor metastasis. This feedback loop holds promise as a therapeutic target for tumor invasion and metastasis.