MicroRNA-455-3p mediates GATA3 tumor suppression in mammary epithelial cells by inhibiting TGF-β signaling

GATA3 is a basic and essential transcription factor that regulates many pathophysiological processes and is required for the development of mammary luminal epithelial cells. Loss-of-function GATA3 alterations in breast cancer are associated with poor prognosis. Here, we sought to understand the tumor-suppressive functions GATA3 normally performs. We discovered a role for GATA3 in suppressing epithelial-to-mesenchymal transition (EMT) in breast cancer by activating miR-455-3p expression. Enforced expression of miR-455-3p alone partially prevented EMT induced by transforming growth factor β (TGF-β) both in cells and tumor xenografts by directly inhibiting key components of TGF-β signaling. Pathway and biochemical analyses showed that one miRNA-455–3p target, the TGF-β–induced protein ZEB1, recruits the Mi-2/nucleosome remodeling and deacetylase (NuRD) complex to the promotor region of miR-455 to strictly repress the GATA3-induced transcription of this microRNA. Considering that ZEB1 enhances TGF-β signaling, we delineated a double-feedback interaction between ZEB1 and miR-455-3p, in addition to the repressive effect of miR-455-3p on TGF-β signaling. Our study revealed that a feedback loop between these two axes, specifically GATA3-induced miR-455-3p expression, could repress ZEB1 and its recruitment of NuRD (MTA1) to suppress miR-455, which ultimately regulates TGF-β signaling. In conclusion, we identified that miR-455-3p plays a pivotal role in inhibiting the EMT and TGF-β signaling pathway and maintaining cell differentiation. This forms the basis of that miR-455-3p might be a promising therapeutic intervention for breast cancer.

GATA3 is a basic and essential transcription factor that regulates many pathophysiological processes and is required for the development of mammary luminal epithelial cells. Loss-offunction GATA3 alterations in breast cancer are associated with poor prognosis. Here, we sought to understand the tumor-suppressive functions GATA3 normally performs. We discovered a role for GATA3 in suppressing epithelial-to-mesenchymal transition (EMT) in breast cancer by activating miR-455-3p expression. Enforced expression of miR-455-3p alone partially prevented EMT induced by transforming growth factor ␤ (TGF-␤) both in cells and tumor xenografts by directly inhibiting key components of TGF-␤ signaling. Pathway and biochemical analyses showed that one miRNA-455-3p target, the TGF-␤induced protein ZEB1, recruits the Mi-2/nucleosome remodeling and deacetylase (NuRD) complex to the promotor region of miR-455 to strictly repress the GATA3-induced transcription of this microRNA. Considering that ZEB1 enhances TGF-␤ signaling, we delineated a double-feedback interaction between ZEB1 and miR-455-3p, in addition to the repressive effect of miR-455-3p on TGF-␤ signaling. Our study revealed that a feedback loop between these two axes, specifically GATA3-induced miR-455-3p expression, could repress ZEB1 and its recruitment of NuRD (MTA1) to suppress miR-455, which ultimately regulates TGF-␤ signaling. In conclusion, we identified that miR-455-3p plays a pivotal role in inhibiting the EMT and TGF-␤ signaling pathway and maintaining cell differentiation. This forms the basis of that miR-455-3p might be a promising therapeutic intervention for breast cancer.
The GATA family of zinc finger transcription factors, comprising GATA1-GATA6, specifically binds the (T/A)G-ATA(A/G) consensus DNA sequence with high affinity (1,2).
Among these transcription factors, GATA3 is the most highly expressed in the mammary epithelium and is expressed exclusively in the luminal epithelial cell population, where it not only specifies but also maintains luminal epithelial cell differentiation (3,4). In addition, mutations in the zinc finger domains of GATA3 that diminish or abolish DNA-binding ability have been commonly identified in the human breast (5), and GATA3 was recently found to be one of three genes (with PIK3CA and TP53) mutated in Ͼ10% of all breast cancers (6). We have previously shown that GATA3 coordinates a program of transcriptional repression composed of G9A and the MTA3-Mi-2/ nucleosome remodeling and deacetylase (NuRD) 2 complex, which represses a cohort of genes including ZEB2 and inhibits the invasive potential and metastasis of breast cancer (7). However, how GATA3 loss-of-function contributes to the expression of microRNAs (miRNAs), which are also important regulators of tumor progression and metastasis, is still poorly understood.
Transforming growth factor ␤ (TGF-␤) signaling is a multifunctional pathway that regulates the vast majority of physiological processes, including pluripotency, proliferation, migration, apoptosis, and differentiation in metazoan cells (8 -10). TGF-␤ and downstream signaling play a dual role in carcinogenesis (11), and the effects of this pathway depend on the cellular context, which is particularly apparent in tumors. TGF-␤ from the inflammatory tumor microenvironment mediates cancer cell apoptosis and tumor suppression (12) or induces epithelial-to-mesenchymal transition (EMT), which promotes the separation, migration, invasion, and metastasis of epithelial cancer cells (13) or cancer stem cell heterogeneity and drug resistance (14). As a common and potent inducer of EMT, TGF-␤ induces this process through both Smad and non-Smad signaling pathways (15)(16)(17). In the Smad pathway, phosphory-lated and activated Smad2 and Smad3 form a complex with Smad4 to promote the expression of the three families of EMTinducing transcription factors, namely Snail/Slug, ZEB1/2, and Twist (15,17,18). These transcription factors, which have historically been implicated in lineage determination, repress the expression of epithelial markers while activating the expression of mesenchymal markers (18). Furthermore, TGF-␤ can also promote EMT by activating the PI3K/Akt/mTOR pathway and small GTPases (19,20). In addition, it was noted that GATA3 inhibits the activity of TGF-␤ signaling and TGF-␤-induced EMT (21). However, how GATA3 regulates TGF-␤ signaling is largely unexplored. miRNAs, 19 -22 nucleotides in length, are members of a large family of small noncoding RNAs that regulate a variety of cellular processes (22) and have concordantly been found to be dysregulated in diseases, including cancer (23,24). As a class of post-transcriptional regulators, miRNAs modulate gene expression by binding complementary sequences in the 3Ј-UTR of mRNAs to either target them for degradation or inhibit their translation (22). Because the binding of miRNA to mRNAs is often achieved via imperfect complementarity, each miRNA can bind and regulate multiple protein-coding mRNAs and noncoding RNAs. Therefore, aberrant expression of a single miRNA can deleteriously affect the translation of multiple genes within a cell, leading to profound phenotypic effects (25). Moreover, miRNAs have established roles as oncogenes or tumor suppressors (23). Recent evidence indicates that some miRNAs, including miR-10b and miR-374a, act as oncogenes to promote breast cancer metastasis (26,27), whereas miR-146 and miR-17/20 exhibit the opposite function (28,29).
In this study, we addressed the issue of how GATA3 loss-offunction contributes to the expression of miRNAs in breast cancer. For this, we explored the molecular mechanism underlying the functional interplay among miR-455-3p, Smad2, ZEB1, and HDAC2 to investigate the role of miR-455-3p in EMT and breast cancer progression.

GATA3 directly induces miR-455-3p expression independent of ER␣ signaling
GATA3 is a transcription factor that has been functionally linked to estrogen receptor ␣ (ER␣) expression and activity in breast carcinoma; moreover, it is involved in a positive crossregulatory loop with ER␣, where each is required for the transcription of the other (31). Recently, Mair et al. (32) found that GATA3 interacts with the histone methyltransferases G9A and GLP independent of estrogen receptor signaling. Therefore, we investigated whether ER␣ plays a role in the regulation of miR-455-3p by GATA3. To this end, the putative promoter region (Ϫ2050 to ϩ500 bp) of miR-455-3p was analyzed using the JASPAR database (http://jaspar.genereg.net) 3 (79), and nine potential GATA3-binding sites were located; however, no ER␣binding sites were identified (relative profile score threshold ϭ 90%; Fig. 2A). To gain a deeper insight into the mechanism through which GATA3 stimulates miR-455-3p transcription, we next conducted a quantitative ChIP (qChIP)-based promoter-walk using MCF-7 cells and mapped the enrichment of GATA3 to two regions of the miR-455-3p promoter, specifically at positions #4 and #6, as indicated. This is consistent with previous results that GATA3 promotes its expression by directly binding the promoter region of miR-455-3p. Meanwhile, there was no significant ER␣ enrichment in this putative region (Fig. 2B). Next, the putative promoter sequence, containing all predicted GATA3-binding sites, was cloned into the luciferase reporter vector pGL4, and GATA3-dependent activity was assessed in ER-positive (MCF-7, MCF 10A, and T-47D) and ER-negative cell lines (MDA-MB-231, MDA-MB-436, and MDA-MB-468). As shown in Fig. 2 (C and D), promoter activity was activated by GATA3 in a dose-dependent manner in both ER-positive and ER-negative cell lines. These results indicated that GATA3 induces the transcription of miR-455-3p in an ERindependent manner. Furthermore, previous results indicated that GATA3 was enriched at the #4 (approximately Ϫ755 to Ϫ633) and #6 (approximately Ϫ1145 to Ϫ908) regions of the MIR455 promoter (Fig. 2, A and B), which contain two potential binding sites for GATA3 at site 1 (AGATTG, 644 bp upstream of the transcription start site) and site 2 (TGATAAGG, 926 bp upstream of the transcription start site). Luciferase experiments in MCF-7 cells or MDA-MB-231 cells showed that the deletion of site 1 and/or 2 diminished GATA3-mediated reporter induction, demonstrating that these sites are necessary and functional (Fig. 2, E and F).

miR-455-3p inhibits the proliferation and metastatic potential of breast cancer cells
As reported previously, GATA3 can maintain the differentiation of luminal epithelial cells in the mammary gland and inhibit the metastasis and proliferation of breast cancer (4, 7, 33-35). Therefore, we postulated that GATA3 might affect the A, MCF-7 cells were transduced with short hairpin RNAs against GATA3 ORF (shGATA3 #1) or the 3Ј-UTR (shGATA3 #2) together with FLAG-tagged GATA3; morphological alterations in these cells were examined by phase-contrast microscopy (left) and F-actin staining (middle). DAPI staining was used to detect nuclei (blue) and rhodamine-conjugated phalloidin for actin polymers (red). The epithelial and mesenchymal markers were detected by Western blotting (right). B, microarray analysis identified 44 up-regulated and 48 down-regulated miRNAs (-fold change Ͼ 2) upon knockdown of GATA3 in MCF-7 cells. C, the knockdown efficiency of GATA3 was verified by qPCR. D, verification of the microarray results by qPCR analysis of the indicated miRNAs in GATA3-depleted MCF-7 cells. E, the overexpression level of GATA3 was verified by qPCR. F, verification of the microarray results by qPCR analysis of the indicated miRNAs in MDA-MB-231 cells overexpressed with GATA3. C-F, the miRNA levels were normalized to those of U6. Error bars, S.D. of three independent experiments (*, p Ͻ 0.05; **, p Ͻ 0.01, two-tailed unpaired t test).

GATA3 regulates miR-455-3p to inhibit TGF-␤ signaling
proliferation and metastasis of breast cancer by regulating miR-455-3p. To verify this hypothesis, we performed 5-ethynyl-2Јdeoxyuridine (EdU) assays to examine the role of miR-455-3p in the proliferation of breast cancer cells. The less-differentiated MDA-MB-231 cells had a much lower proportion of EdU-labeled cells after transfection with miR-455-3p mimics, whereas the number of positively labeled cells in the differentiated MCF-7 cell line obviously increased upon treatment with miR-455-3p inhibitors (Fig. 3A). Next, to further investigate the role of miR-455-3p in the invasion and metastasis of breast cancer, we detected the expression level of epithelial/mesenchymal markers in miR-455-3p mimic-or miR-455-3p inhibitor-transfected MDA-MB-231 or MCF-7 cells by qPCR and Western blotting. miR-455-3p gain-of-function in MDA-MB-231 cells resulted in the induction of epithelial markers (E-cadherin, ␣-catenin, and ␥-catenin) and the reduction of mesenchymal markers (N-cadherin, fibronectin, and vimentin) at both the mRNA and protein level. Accordingly, miR-455-3p loss-of-function in MCF-7 cells was associated with an apparent reduction in epithelial markers and the significant induction of mesenchymal markers (Fig. 3, B and C). In addition, transwell invasion assays showed that forced expression of miR-455-3p significantly inhibited the invasive potential of MDA-MB-231 cells. Consistently, treatment with miR-455-3p inhibitors improved the invasive potential of MDA-MB-231 cells (Fig. 3D). However, the negative effect on invasion mediated by GATA3 was diminished by treatment with miR-455-3p inhibitors (Fig. 3E). Together, these results suggested that miR-455-3p can inhibit the proliferation, EMT, and invasion of breast cancer cells.
To investigate the role of miR-455-3p in tumor development and progression in vivo, MDA-MB-231 cells engineered to stably express firefly luciferase (MDA-MB-231-Luc-D3H2LN, Xenogen Corp.) were co-infected with lentiviruses harboring hsa-miR-455-3p. These cells were then orthotopically implanted into immunocompromised severe combined immunodeficiency (SCID) mice to assess tumorigenesis. Specifically, MDA-MB-231-Luc-D3H2LN cells were implanted into the abdominal mammary fat pad (n ϭ 6) of 6-week-old female SCID mice. The growth of tumors was monitored weekly through bioluminescence imaging using an IVIS imaging system(XenogenCorp.).Accordingly,orthotopictumorsweremeasured by quantitative bioluminescence imaging after 8 weeks. The results showed that, in the orthotopically implanted groups, forced expression of miR-455-3p resulted in a significant reduction in MDA-MB-231-Luc-D3H2LN tumor growth (Fig. 3F, top). The tumor specimens were surgically removed (Fig. 3F, bottom) and examined by in vitro bioluminescence imaging (Fig. 3G). Next, we analyzed published data down-loaded from the GEO database (http://www.ncbi.nlm.nih.gov/ geo/). The results supported the idea that miR-455-3p expression was decreased in breast carcinoma relative to that in adjacent normal tissues (Fig. 3H). To further extend our observations to a clinical and pathologically relevant context, we analyzed the expression of miR-455-3p and its correlation with the clinical outcomes of breast cancer patients. Kaplan-Meier survival analysis of the relationship between survival time and the expression of miR-455-3p was performed with an online tool (http://kmplot.com/analysis/) 3 (80). This showed that higher expression of miR-455-3p (p ϭ 0.02) was associated with improved survival in breast cancer patients when the influence of systemic treatment, endocrine therapy, and chemotherapy were excluded (Fig. 3I). In conclusion, these experiments demonstrated that miR-455-3p inhibits the tumorigenesis and metastasis of breast cancer cells in vitro and in vivo.

miR-455-3p directly targets Smad2, ZEB1, and HDAC2
To explore the molecular mechanism through which miR-455-3p regulates the cell cycle and TGF-␤ pathway, we predicted miR-455-3p targets using programs including DIANA, miRANDA, miRDB, miWalk, and TargetScan. It was found that miR-455-3p has 212 overlapping potential targets, including Smad2, ZEB1, HDAC2, and other genes closely related to tumorigenesis (Fig. 5A). As a key component of the TGF-␤ signaling pathway, phosphorylated Smad2, which is induced by TGF-␤, combines with Smad3 and Smad4 to form trimeric Smad complexes that translocate to the nucleus to regulate gene expression (36). The ␦EF1 family proteins, namely ␦EF1 (ZEB1) and SIP1 (ZEB2), act as EMT transcription factors and are activated by TGF-␤ signaling (37). Meanwhile, the epigenetic modifier HDAC2 is up-regulated by TGF-␤ signaling during endothelial-to-hematopoietic transition (38), and TGF-␤ also activates HDAC2 in the kidneys (39). We used the public algorithm TargetScan (http://www.targetscan.org/vert_72/) 3 (82) and identified the putative miR-455-3p-binding sites of these three targets, which were all conserved among Sus scrofa, Canis familiaris, Oryctolagus cuniculus, and Homo sapiens. To test whether Smad2, ZEB1, and HDAC2 are directly modulated by miR-455-3p, we performed reporter assays using HEK 293T cells, wherein the luciferase gene was driven by either WT or mutated 3Ј-UTR sequences. The results showed that luciferase activity was reduced with the WT 3Ј-UTRs, but not the mutant 3Ј-UTRs, of these targets when HEK 293T cells were transfected with miR-455-3p mimics. Consistently, treatment with miR-455-3p inhibitors led to the opposite effect on luciferase activity with WT and mutant 3Ј-UTRs (Fig. 5, B and C). However, neither mimics nor inhibitors of miR-455-3p had an effect on the luciferase activity of the other potential binding site of Smad2 3Ј-UTR at positions 7258 -7264 (data not shown).
In MCF-7 and MDA-MB-231 cells transfected with miR-455-3p mimics, Western blot analysis showed that the protein levels of ZEB1, Smad2, and HDAC2 were significantly decreased; in contrast, in cells transfected with miR-455-3p inhibitors, the protein levels of all three were partially increased (Fig. 5D). Similarly, the depletion of GATA3 in MCF-7 cells resulted in the induction of ZEB1, Smad2, and HDAC2, whereas their expression was depressed in MDA-MB-231 cells overexpressing GATA3 (Fig. 5E). Importantly, these results suggest that miR-455-3p binds the 3Ј-UTR of Smad2, ZEB1, or HDAC2 and that these genes are the downstream targets of miR-455-3p.

miR-455-3p inhibits TGF-␤-induced EMT
As stated, studies have shown that TGF-␤ is a common and potent inducer of EMT (15)(16)(17); thus, it was reasonable to speculate that miR-455-3p can inhibit EMT by disrupting the TGF-␤ signaling pathway. To this end, gain-of-function miR-455-3p experiments were performed, and morphological alterations and the expression of epithelial/mesenchymal markers were analyzed in MCF 10A nonmetastatic human mammary epithelial cells with TGF-␤1 treatment. After treatment, MCF 10A cells acquired a spindle-shaped morphology, and the actin filaments that regulate cell migration were reorganized (Fig.  6A). Further, phenotypic changes in MCF 10A cells were accompanied by the up-regulation of mesenchymal markers (N-cadherin, ZEB1, and fibronectin) and the down-regulation of epithelial markers (E-cadherin and ␣-catenin) at both the mRNA and protein level (Fig. 6, B and C). The increased levels of Smad2 and Smad3 phosphorylation and up-regulation of ZEB1 and ZEB2 verified the activation of TGF-␤ signaling. Of the components of NuRD, the metastasis-associated (MTA) family of proteins, including MTA1, MTA2, and MTA3, have been implicated in cancer progression and metastasis (40 -42). MTA3 inhibits EMT and breast cancer metastasis, and its expression is increasingly lost during breast cancer progression (43), whereas MTA1 promotes breast tumor progression, and its expression gradually increases during the process (44). Hence, we investigated whether the MTA family is involved in TGF-␤-induced EMT. However, we did not detect statistically changes in HDAC2, GATA3, MTA1, MTA2, and MTA3, which have been implicated in cancer progression and metastasis, in TGF-␤-stimulated MCF 10A cells (Fig. 6C).

Reciprocal regulation of the ZEB1/NuRD complex and miR-455-3p
It has been reported that the EMT-activator ZEB1 can repress expression of the miR-200 family, which also directly targets ZEB1 (46 -48). Accordingly, we examined whether ZEB1, which is also a target of miR-455-3p, could regulate the expression of miR-455-3p. Scanning the MIR455 promoter region using the JASPAR database identified ten putative binding sequences (relative profile score threshold ϭ 90; (Fig. 7A). This implied that ZEB1 might also directly suppress the transcription of MIR455. We previously reported that ZEB2 physically associates with the NuRD (MTA1/2) complex (7), but it was unclear whether ZEB1 could also interact with this complex. Immunoprecipitation assays with an anti-ZEB1 antibody followed by immunoblotting using antibodies against Mi-2, MTA1, MTA2, MTA3, HDAC1, HDAC2, RbAp46/48, or MBD2/3 indicated that all of these proteins were efficiently co-immunoprecipitated with ZEB1 in MDA-MB-231, MCF-7, and T-47D cell lines (Fig. 7B). In addition, GST pulldown with GST-fused MTA1, MTA2, MTA3, HDAC1, HDAC2, RbAp46, RbAp48, or MBD3 and in vitro-transcribed/translated ZEB1 revealed that ZEB1 interacts with MTA1 and MTA2, but not with MTA3 and other tested components of the NuRD com-plex (Fig. 7C). These experiments indicated that ZEB1 can also associate with the NuRD complex. Interestingly, HDAC2 is also a direct target of miR-455-3p, and thus, ZEB1 might interact with the NuRD complex to repress the expression of MIR455.
We next confirmed the role of ZEB1/NuRD in the transcription of MIR455. For this, quantitative ChIP analysis was performed using MDA-MB-231 cells and specific antibodies against ZEB1, MTA1, or HDAC2 to test ten promoter regions of MIR455. Results showed that ZEB1, MTA1, and HDAC2 were strongly enriched at the #4 and #7 regions of MIR455 (Fig.  7D). To further support the proposition that ZEB1 nucleates NuRD to form one protein complex at target promoters, sequential ChIP or ChIP/re-ChIP experiments were performed on the #4 and #7 regions of MIR455. For these experiments, soluble chromatin was first immunoprecipitated with antibodies against ZEB1, MTA1, or HDAC2. The immunoprecipitates were subsequently re-immunoprecipitated with appropriate antibodies. The results showed that the #4 and #7 promoter regions of MIR455 that were immunoprecipitated with antibodies against ZEB1 could be re-immunoprecipitated with antibodies against MTA1 or HDAC2 (Fig. 7E). Similar results were obtained when the initial ChIP was performed with antibodies against MTA1 or HDAC2 (Fig. 7E).
Next, we investigated whether ZEB1, MTA1, and HDAC2 directly suppress transcriptional activity of the MIR455 miRNA gene. Luciferase assays showed that the promoter activity of MIR455 was suppressed by ZEB1, MTA1, or HDAC2 overexpression in MCF-7 cells (Fig. 7F, top). In contrast, knockdown of either HDAC2, MTA1, or ZEB1 in MDA-MB-231 cells resulted in enhanced promoter activity when compared with that in controls (Fig. 7F, top). Moreover, gain-of-function and loss-of-function ZEB1, MTA1, or HDAC2 experiments were performed, and the expression of miR-455-3p was examined in MCF-7 or MDA-MB-231 cells. We found that the overexpression of either ZEB1, MTA1, or HDAC2 could down-regulate miR-455-3p expression in MCF-7 cells (Fig. 7F, bottom). Consistently, the individual depletion of ZEB1, MTA1, or HDAC2 resulted in the induction of miR-455-3p expression (Fig. 7F,  bottom). These results support the contention that a reciprocal feedback regulatory loop exists between GATA3 and the ZEB1nucleated repression program to regulate the expression of miR-455-3p, which ultimately inhibits TGF-␤ signaling to control EMT and the metastasis of breast cancer cells (Fig. 7G).

Discussion
It is widely recognized that the loss of or mutation in GATA3 is tightly linked to the progression of breast cancer (32,33) and that this cooperates with ER␣ (31). Because miRNAs are widely involved in the regulation of biological processes (22,49), the transcription factor GATA3 modulates its own biological func-

GATA3 regulates miR-455-3p to inhibit TGF-␤ signaling
tions by regulating the expression of miRNAs. As a transcription factor, GATA3 plays a dual role in regulating microRNA expression. On one hand, GATA3 increases the expression of miR-29b and miR-503, which promote cell differentiation and suppress tumor metastasis (21). On the other hand, GATA3 can also down-regulate levels of miR-183 to induce adipogenesis in 3T3-L1 cells (50). In this study, we first revealed that the depletion of GATA3 leads to multiple changes in miRNA expression, such as up-regulation of miR-134, miR-152, and miR-486 -5p and down-regulation of miR-455-3p, miR-200b-5p, and miR-7-1-3p; moreover, miR-455-3p was found to be the most dysregulated miRNA in these cells. Furthermore, GATA3 was found to significantly increase the transcription of MIR455 in either ER-positive or ER-negative cells, which indicates that GATA3 promotes the expression of miR-455-3p, independent of the function of ER␣. These results indicated that GATA3 has an additional mechanism that does not rely on ER␣ to regulate gene expression. miR-455-3p has been reported to suppress the expression of ROCK2, HOXB5, hTERT, and RUNX2, which are involved in multiple biological processes including renal fibrosis, chondrogenic development and differentiation, Alzheimer's disease, preeclampsia, and tumorigenesis (51)(52)(53)(54)(55)(56)(57). EMT has been described as the transformation of epithelial cells to mesenchymal cells and is associated with embryonic development, tissue repair, and carcinoma progression (58,59). During cancer pathogenesis, EMT is induced by pleiotropic signals and then triggers the dissociation of cells from primary carcinomas, which subsequently migrate and disseminate to distant sites (60). Here, we showed that miR-455-3p acts as an inhibitor of EMT in breast cancer cells and that miR-455-3p loss-of-function promotes the acquisition of cell invasive properties and the ability to metastasize. Simultaneous inhibition of miR-455-3p function in GATA3-positive cells was found to attenuate the inhibitory effect of GATA3 on cell metastasis and dedifferentiation. These results indicate that miR-455-3p represents an important node that inhibits breast cancer metastasis.
The TGF-␤ signaling pathway activates the expression of multiple transcription factors, such as ZEB1, ZEB2, and SNAIL, and induces the initiation of EMT (15,18,61). miRNAs usually play a role in transcriptional regulation by inhibiting the expression of target genes (22). As mentioned under "Results," miR-455-3p broadly inhibited TGF-␤-induced EMT by repressing the expression of Smad2, ZEB1, and HDAC2. Smad2 as well as Smad3 are phosphorylated and activated by TGF-␤ receptors and are referred to as R-Smads; these signaling molecules initiate TGF-␤ signal transduction and translocate to the nucleus to activate or inhibit target gene transcription (36,62,63). Further, ZEB1 is a potent promoter of EMT. Acting as a transcriptional repressor of E-cadherin and the microRNA-200 family and an activator of mesenchymal markers (47,48,64), ZEB1 positively affects cellular signaling pathways including TGF-␤ and the Wnt pathway to promote EMT and metastasis. Moreover, the chromatin remodeler HDAC2, which is a subunit of the NuRD complex, acts as a corepressor to repress the transcription of E-cadherin and miR-200 and regulate cancer metastasis (65)(66)(67). In collaboration with HDAC1, HDAC2 can also modulate TGF-␤ signaling and promote hematopoiesis in the hemogenic endothelium (38). Thus, the suppression of Smad2, ZEB1, and HDAC2 by the GATA3/miR-455-3p axis is of particular significance. Taken together, it is obvious that miR-455-3p inhibits the TGF-␤ signaling pathway, maintains cell differentiation, and inhibits EMT. Previous reports indicate that GATA3 can also inhibit the TGF-␤ signaling pathway (21); together with our results indicating that the signaling protein Smad2 and the effector protein ZEB1 are also targets of miR-455-3p, this suggests that GATA3 can inhibit this signaling pathway at the transcriptional and post-translational levels. This complex regulatory network would ensure that GATA3 can efficiently regulate the TGF-␤ signaling pathway.
The NuRD complex, which consists of Mi-2, MTA1, MTA2, MTA3, HDAC1, HDAC2, RbAp46, and MBD3, possesses histone deacetylation activity and primarily functions in transcriptional repression programs to regulate breast cancer metastasis (68). Our results showed that TGF-␤ could induce the expression of N-cadherin, fibronectin, ZEB1, and ZEB2, in addition to the phosphorylation of Smad2 and Smad3, while down-regulating levels of E-cadherin and ␣-catenin in MCF 10A cells. Interestingly, we did not observe statistically significant changes in the expression of MTA family proteins, including MTA1, MTA2, and MTA3, after TGF-␤1 treatment. However, these TGF-␤1-mediated alterations could be reversed by miR-455-3p treatment. Specifically, miR-455-3p dramatically restored the loss of E-cadherin and ␣-catenin and reversed the up-regulation of N-cadherin, fibronectin, and ZEB1, which respond to TGF-␤1 stimulation. It is worth noting that we only observed a decrease in Smad2 expression and phosphorylation, without changes in Smad3. These results demonstrate that miR-455-3p could not only repress the direct TGF-␤ signaling mediator Smad2, but also inhibit the downstream target gene of TGF-␤ signaling pathway ZEB1 and the epigenetic regulator HDAC2, thus disturbing the TGF-␤ signaling pathway at multiple levels. Therefore, miR-455-3p could be regarded as an inhibitor of the TGF-␤/Smad signaling pathway and a potential new therapeutic target for breast cancer.

GATA3 regulates miR-455-3p to inhibit TGF-␤ signaling
As previously described, GATA3 physically associates with the G9A/NuRD (MTA3) complex to repress ZEB2, which in turn suppresses the expression of G9A and MTA3 through the recruitment of G9A/NuRD (MTA1) (7). Here, we reported that GATA3, through activation of miR-455-3p, inhibits expression of the EMT activators ZEB1 and HDAC2 and maintains differentiation of the mammary epithelium. Upon GATA3 loss-offunction, due to down-regulation or mutation, which occurs frequently in breast cancer (3,69,70), the expression of ZEB1 and HDAC2 are derepressed; consequently, elevated ZEB1 and HDAC2 can promote EMT and metastasis. Surprisingly, we found that ZEB1 represses the expression of miR-455-3p through recruitment of the MTA1-NuRD complex. Apparently, a reciprocal feedback regulatory loop exists in which GATA3 loss-of-function in breast cancer leads to the elevated expression of ZEB1 and HDAC2, which in turn further downregulates miR-455-3p to maintain high levels of these markers; this promotes EMT and eventually leads to breast cancer metastasis. Meanwhile, these results showed that GATA3 and the ZEB1/NuRD (MTA1) complex occupy similar promoter regions of MIR455, which suggests possible competition between GATA3 and ZEB1 for the transcriptional regulation of miR-455-3p based on their abundance and binding to the promoter. In this loop, miR-455-3p acts as an integral component and represents a rate-limiting factor. Generally, GATA3 separately suppresses the expression of ZEB1 and ZEB2 at the transcriptional or post-translational level, thereby inhibiting breast cancer progression.
In summary, our study demonstrated that a reciprocal feedback loop between the GATA3/miR-455-3p and TGF-␤/ZEB1/ NuRD (MTA1) axes regulates mammary epithelial cell fate and dynamically regulates epithelial cell plasticity; further, dysfunction of this loop affects the fate of mammary epithelial cells and promotes breast cancer metastasis. Our research further revealed the molecular mechanism through which GATA3 loss-of-function affects the progression of breast cancer and enhances our understanding of the complexity of EMT hierarchical regulatory networks. Several studies have indicated that GATA3 is mutated or deleted in many types of diseases including tumors; however, it was not found to act as a classic tumor suppressor (5,(71)(72)(73). miRNAs are considered potential anticancer drugs (74,75), and thus the activation or expression of miR-455-3p in tumors exhibiting a loss of GATA3 function represents a new potential therapeutic approach.

Antibodies and reagents
The antibodies used in this study are listed in Table S2. Dynabeads Protein G was obtained from Invitrogen by Thermo Fisher Scientific (Waltham, MA), and the protease inhibitor mixture was from Roche Applied Science. GSH Sepharose TM 4B beads were purchased from GE Healthcare (Uppsala, Sweden). Recombinant Human TGF-␤1 protein was obtained from R&D Systems (Minneapolis, MN). The synthesized mimics/inhibitors of miR-455-3p and short hairpin RNAs were from GenePharma Co., Ltd. (Shanghai, China). The sequences are listed in Table S3. MCF 10A cells were cultured with the mammary epithelium growth factor medium kit supplemented with growth factors (cc-3150, Lonza, Switzerland). In addition to mammary epithelium growth factor medium, all of the other medium were supplemented with 10% fetal bovine serum, 100 units/ml penicillin, and 100 mg/ml streptomycin (Gibco, BRL); the complete medium for MDA-MB-436 cells also contained 10 g/ml insulin and 16 g/ml GSH. Transfections were carried out using Lipofectamine 2000 or Lipofectamine RNAiMAX Reagent (Invitrogen) according to the manufacturer's instructions.

siRNA transfection
Specific MTA1 siRNA and negative control siRNA were obtained from Dharmacon (Lafayette, CO), and siRNAs against HDAC2, ZEB1, SMAD2, and SMAD3 were purchased from Sigma-Aldrich. The transfection of siRNA was performed with Lipofectamine RNAiMAX Reagent (Invitrogen) according to the manufacturer's instructions. The sequences are listed in Table S3.

RT-qPCR
Total cellular RNA was isolated with TRIzol reagent following the manufacturer's instructions (Invitrogen). Any potential DNA contamination was avoided with RNase-free DNase treat- GATA3 regulates miR-455-3p to inhibit TGF-␤ signaling ment (Promega, Madison, WI), and cDNA was prepared with Moloney murine leukemia virus reverse transcriptase (Promega). Relative quantitation of all gene transcripts was performed using the ABI PRISM 7500 sequence detection system (Applied Biosystems, Foster City, CA) through the measurement of real-time SYBR Green fluorescence. The results were obtained using the comparative Ct method (2 Ϫ⌬⌬Ct ) with GAPDH or U6 as the internal control. These experiments were performed in triplicate, and the primer pairs used are listed in Table S4.

Plasmid construction and luciferase assays
The WT promoter or mutant promoters with GATA3-binding sequences in MIR455 deleted (nucleotides Ϫ2050 to ϩ500) were amplified by PCR and cloned into pGL4 (Promega). Then the pGL4-miR-455-3p-Luc reporter was co-transfected with a vector encoding Renilla luciferase (pRL-TK) and either a vector control or FLAG-GATA3 into ER-positive or ER-negative cells. Human WT and mutant Smad2, ZEB1, and HDAC2 3Ј-UTRs were amplified by PCR and cloned into the SpeI-HindIII sites of the pMIR-REPORT luciferase vector (Ambion, Grand Island, NY). The TGF-␤ reporter p3TP-Lux (J. Massagué, Memorial Sloan-Kettering Cancer Center, New York; plasmid 11767) was obtained from Addgene. The p3TP-Lux reporter was cotransfected with pRL-TK into MCF 10A cells. Three repeat sequences of the activin response element were artificially synthesized and inserted into the pGL4 vector (ARE-Luc). For the ARE-luc experiments, the cells were co-transfected with FLAG-FoxH1 (76). Then cells were serum-starved overnight and stimulated with 5 ng/ml TGF-␤1 (R&D Systems) for 24 h. HEK 293T cells treated with controls, miR-455-3p mimics, or miR-455-3p inhibitors were transfected with WT or mutant Smad2, ZEB1, or HDAC2 3Ј-UTR luciferase reporters together with a vector encoding Renilla luciferase. All luciferase assays were performed as described previously (77). In brief, 48 h after transfection, lysates were collected, and firefly and Renilla luciferase activities were detected using the Dual-Luciferase Reporter System and a GloMax luminometer according to the manufacturer's protocol (Promega). Firefly luciferase activity was normalized to that of Renilla luciferase. Each experiment was repeated in triplicate.

EdU assays
After respective treatments, MCF-7 and MDA-MB-231 cells were seeded in 6-well plates and subjected to EdU assays to detect proliferation based on fluorescence detection (Thermo Fisher Scientific). Before fixation, the cells were incubated in conditioned medium from the kit, and EdU assays were performed according to the manufacturer's instructions.

Cell invasion assays
Transwell invasion assays were performed using transwell chambers (BD Biosciences) coated with Matrigel. After appropriate treatment, cells were resuspended in serum-free medium, and then 2.5 ϫ 10 4 cells were seeded into the upper chamber in a volume of 500 l. The chamber was then transferred to a well containing 500 l of medium with 10% FBS at 37°C for 18 -20 h. Cells on the upper side of the chamber were removed by wiping the top of the membrane with cotton swabs. Cells on the lower side were then stained and counted. Three high-powered fields were counted for each membrane.

Immunoprecipitation
For immunoprecipitation assays, cellular extracts were harvested as described previously (78) and incubated with a specific anti-ZEB1 antibody or normal mouse/rabbit IgG antibody at 4°C overnight with constant rotation. Samples were then incubated with Dynabeads Protein G for 2 h at 4°C and, after washing with cell lysis buffer, were subjected to SDS-PAGE. Immunodetection was performed using the enhanced chemiluminescence (ECL) system (Amersham Biosciences) according to the manufacturer's instructions.

GST pulldown assays
The coding sequences of GST-fused proteins were amplified and cloned into the pGEX 4T-3 vector (Amersham Biosciences). GST-fused proteins were expressed in Escherichia coli (BL-21 strain) at the appropriate temperature, followed by ultrasonic cell disruption in cold PBS in the presence of a protease inhibitor mixture, and purified using GSH Sepharose 4B beads according to the manufacturer's instructions (GE Healthcare). The beads were co-incubated with the in vitrotranscribed/translated proteins, which were transcribed and translated using rabbit reticulocyte lysate (Promega), and then subjected to further washing with binding buffer (75 mM NaCl, 50 mM HEPES, pH 7.9) five times to prevent nonspecific interactions. The beads were resuspended in 30 l of 2ϫ SDS-PAGE loading buffer and then analyzed by Western blotting.

ChIP and re-ChIP
ChIP and re-ChIP experiments were performed with MCF-7 cells as previously described (68). In brief, the cells were crosslinked with 1% formaldehyde, sonicated, precleared, and incubated with 5-10 g of antibody per reaction, which was fol-  -tailed unpaired t test). G, graphic model as discussed under "Results." A feedback loop between GATA3 and the ZEB1-nucleated repression program is involved in regulating the expression of miR-455-3p to control EMT and the metastasis of breast cancer cells.

GATA3 regulates miR-455-3p to inhibit TGF-␤ signaling
lowed by the addition of Dynabeads Protein G. The beads were then washed in buffers of high and low salt concentrations, and DNA was eluted and purified with the QIAquick PCR purification kit. For re-ChIP assays, immune complexes were eluted from the beads with 20 mM DTT at 37°C for 30 min. The eluents were then diluted 30-fold with ChIP dilution buffer, which was followed by re-immunoprecipitation with the second antibodies. The final elution step was performed with 1% SDS solution in Tris-EDTA buffer, pH 8.0. Enrichment of the DNA template was analyzed by qPCR using primers specific for each target gene promoter. The primer pairs used are listed in Table S5.

Mouse xenograft models
MDA-MB-231 cells stably expressing firefly luciferase (Xenogen Corp.) were infected with lentiviruses encoding control plmiR-Control or plmiR-455-3p (GenePharma). These cells were inoculated into the left abdominal mammary fat pad (5 ϫ 10 6 cells) of 6-week-old female SCID mice. For bioluminescence imaging, mice were injected abdominally with 200 mg/kg D-luciferin in PBS. At 15 min after injection, mice were anesthetized, and bioluminescence was determined using a charge-coupled device camera (IVIS; Xenogen Corp.). Bioluminescence images were obtained with a 15-cm field of view, binning (resolution) factor of 8, 1/f stop, open filter, and an imaging time of 30 s to 2 min. Bioluminescence from relative optical intensity was defined manually. Photon flux was normalized to the background, which was based on relative optical intensity from a mouse not administered luciferin. The volume of tumors was measured using a Vernier caliper and calculated according to the formula, 1 ⁄ 6 ϫ length ϫ width 2 . Animal handling and procedures were approved by Tianjin Medical University Institutional Animal Care.

Statistical analysis
Results are reported as mean Ϯ S.D. unless otherwise noted. Comparisons were performed using two-tailed paired t tests based on a bidirectional hypothesis for continuous variables. Tumor data sets were downloaded from http://www.ncbi. nlm.nih.gov/geo, and GSE numbers are shown in the figures. Data for Kaplan-Meier survival analysis were from http:// kmplot.com/analysis. 3