Tumor-associated myoepithelial cells promote the invasive progression of ductal carcinoma in situ through activation of TGFβ signaling

The normal myoepithelium has a tumor-suppressing nature and inhibits the progression of ductal carcinoma in situ (DCIS) into invasive ductal carcinoma (IDC). Conversely, a growing number of studies have shown that tumor-associated myoepithelial cells have a tumor-promoting effect. Moreover, the exact role of tumor-associated myoepithelial cells in the DCIS-to-IDC development remains undefined. To address this, we explored the role of tumor-associated myoepithelial cells in the DCIS-to-IDC progression. We developed a direct coculture system to study the cell-cell interactions between DCIS cells and tumor-associated myoepithelial cells. Coculture studies indicated that tumor-associated myoepithelial cells promoted the invasive progression of a DCIS cell model in vitro, and mechanistic studies revealed that the interaction with DCIS cells stimulated tumor-associated myoepithelial cells to secrete TGFβ1, which subsequently contributed to activating the TGFβ/Smads pathway in DCIS cells. We noted that activation of the TGFβ signaling pathway promoted the epithelial-mesenchymal transition, basal-like phenotypes, stemness, and invasiveness of DCIS cells. Importantly, xenograft studies further demonstrated that tumor-associated myoepithelial cells enhanced the DCIS-to-IDC progression in vivo. Furthermore, we found that TGFβ-mediated induction of oncogenic miR-10b-5p expression and down-regulation of RB1CC1, a miR-10b-5p-targeted tumor-suppressor gene, contributed to the invasive progression of DCIS. Our findings provide the first experimental evidence to directly support the paradigm that altered DCIS-associated myoepithelial cells promote the invasive progression of DCIS into IDC via TGFβ signaling activation.

Ductal carcinoma in situ (DCIS) 2 is an early stage, non-invasive breast cancer. If untreated, DCIS tumors progress into invasive ductal carcinoma (IDC), a highly aggressive disease that is more difficult to treat than DCIS. Whereas there are ϳ50,000 cases of DCIS diagnosed every year, there are 250,000 new cases of IDC (1). This indicates that there may be a significant number of patients whose DCIS goes undiagnosed or untreated, allowing it to progress to IDC. Whereas early detection of DCIS allows for earlier treatment, the most common methods of DCIS treatment are lumpectomy and radiation. These patients have an 8% chance of their disease recurring as IDC within 5 years due to the failure of therapies (2). Therefore, there is a critical clinical need for effective, targeted DCIS therapies that eradicate the disease and prevent transition to IDC. Further investigations into the molecular mechanisms underlying DCIS and DCIS progression to IDC are of crucial importance.
The hallmarks of the DCIS-to-IDC progression are loss of the myoepithelial layer and the basement membrane and the invasion of tumor cells into the stromal and fat tissues. Although global profiling of genetic and gene expression alterations in DCIS and IDC has revealed a high level of similarity between them, researchers have so far failed to identify the key mechanisms driving the invasive transition (3)(4)(5). However, growing evidence suggests that the breast cancer tissue microenvironment, composed of myoepithelial cells, stroma, fat tissue, and extracellular matrix, is a key factor in promoting the DCIS-to-IDC transition (6,7).
The mammary gland is composed of ductal and alveolar architecture, both of which consist of an epithelial cell layer surrounded by a layer of myoepithelial cells. The myoepithelium produces and is in contact with the basement membrane and has important regulatory roles in normal mammary gland development and function. Myoepithelial cells are necessary for the maintenance of luminal epithelial cell polarity and for the induction of ductal branching and differentiation during mammary gland development (6). In vitro and in vivo studies have indicated that myoepithelial cells are natural tumor suppressors due to their negative effects on tumor cell growth, invasion, angiogenesis, and the invasive transition of a xenograft DCIS model (8 -10). Due to these tumor-suppressive roles, myoepithelial cells are commonly described as "gatekeepers" of the tumor. Through unknown mechanisms, the myoepithelial cell layer is lost during the progression from DCIS to IDC (11). It is likely that the loss of myoepithelial cells is a critical step in the transition to an invasive carcinoma.
Whereas normal myoepithelial cells have been demonstrated to be tumor-suppressive, several studies have identified specific phenotypic changes in tumor-associated myoepithelial cells that result in functional differences compared with normal counterparts (12)(13)(14). Tumor-associated myoepithelial cells have been reported to lose laminin-1 expression, resulting in the abrogation of their ability to maintain the cell polarity of breast epithelial cells (12). A genome-wide gene expression profiling study revealed that tumor-associated myoepithelial cells undergo significant changes in the gene expression pattern and overexpress oncogenic chemokines (e.g. CXCL12 and CXCL14) that bind their respective receptors on epithelial tumor cells and promote tumor proliferation and invasiveness (13). Moreover, myoepithelial cells in a subset of preinvasive DCIS tumors were shown to overexpress integrin ␣v␤6, which elicits tumor promoter activity via activation of TGF␤ and MMP9 (14,15). This molecular change in DCIS-associated myoepithelial cells could predict recurrence in breast cancer patients (15). When normal myoepithelial cells were bioengineered to mimic DCIS-associated myoepithelial cells, they acted as tumor promoters and enhanced in vitro and in vivo tumorigenicity of invasive breast cancer cell lines (e.g. MDA-MB-231 and MCF7) (15). Analysis of 169 DCIS samples also found that ϳ8% of DCIS cases have an aberrant molecular alteration (CK5 ϩ p53 ϩ ) in the myoepithelial layer and that this was increased in basal-like breast tumors (10). These findings support a model wherein DCIS-associated myoepithelial cells gradually lose their tumor-suppressive features and become tumor promoters, enhancing the invasive progression of DCIS via collaboration with neighboring stromal cells to degrade the basement membrane. However, currently there are no experimental data directly supporting the oncogenic role of tumorassociated myoepithelial cells in the in vivo transition from DCIS to IDC.
Aberrant activation of the epithelial-mesenchymal transition (EMT) has been known to contribute to the invasive and metastatic progression of cancers (16). Induction of the EMT by the TGF␤/Smads pathway was recently shown to be a critical underlying mechanism in the invasive transformation of HER2positive DCIS with 14-3-3 overexpression (17). Furthermore, the TGF␤-activated EMT has been shown to enhance the stemcell property (stemness) of normal and cancerous breast cells (18). These findings imply that the activation of the TGF␤/ EMT is a potential pathological mechanism driving the invasive progression of DCIS into IDC. Nevertheless, it remains unknown whether the tissue microenvironment plays a regulatory role in the TGF␤/EMT-related mechanism involved in the DCIS-to-IDC transition.
MicroRNAs (miRNAs) are small noncoding RNA molecules that regulate gene expression by binding to perfect or imperfect complementary sequences at the 3Ј-UTR of target mRNAs, leading to mRNA degradation, inhibition of their translation, or both (19). Because miRNAs are regulators of gene expression, dysregulation of miRNAs has been shown to play critical roles in tumorigenesis (20). Recently, genome-wide deep sequencing analysis identified several differentially expressed miRNAs in invasive breast carcinomas compared with in situ carcinomas (21). Therefore, miRNAs have been thought to be critically involved in regulation of the DCIS-to-IDC transition (22). Moreover, many studies have demonstrated the interplay between miRNAs and cellular signaling pathways (e.g. TGF␤ signaling) and the crucial regulatory roles of miRNAs in the EMT (23)(24)(25). Despite these advances in miRNA studies, the roles of miRNAs in the invasive progression of DCIS into IDC remain largely unknown.
In this study, we investigated the interaction between human DCIS cells (MCF10DCIS) (10,26) and tumor-associated myoepithelial cells (Hs578bst) isolated from the myoepithelial hyperplasia tissue peripheral to the breast carcinosarcoma (27). Our studies showed that in vitro coculture of Hs578bst cells with MCF10DCIS cells activates the TGF␤/Smads pathway and the EMT in MCF10DCIS cells. We further revealed that coculture with MCF10DCIS cells triggers tumor-associated myoepithelial cells to synthesize and secrete more TGF␤1, which is the primary contributor to activation of TGF␤ signaling in MCF10DCIS cells. TGF␤/Smads signaling activation induced the EMT and promoted basal-like phenotypes, stemness, and the migratory as well as invasive abilities of MCF10DCIS cells. Importantly, this altered myoepithelial cell model promoted the invasive progression of MCF10DCIS xenograft tumors in vivo. Through this coculture cell system, we identified miR-10b-5p as a downstream mediator of TGF␤ signaling and further found that miR-10b-5p-dependent signaling contributes to the invasive progression of DCIS cells induced by coculture with tumor-associated myoepithelial cells. Our studies for the first time provide experimental evidence to directly support the hypothesis that altered DCIS-associated myoepithelial cells play a tumor-promoting role in the invasive progression of DCIS into IDC.

Coculture with tumor-associated myoepithelial cells enhances the basal-like/EMT phenotypes and invasiveness/stemness of MCF10DCIS cells
To examine the effects of tumor-associated myoepithelial cells on human DCIS cells, we used the tumor-associated myoepithelial cell line Hs578bst and the human DCIS cell line MCF10DCIS.COM (MCF10DCIS). MCF10DCIS is derived from MCF10A, a non-malignant immortalized human mammary epithelial cell (HMEC) line, and reproducibly forms comedo DCIS-like lesions in immunodeficient mice. MCF10DCIS-derived xenograft tumors spontaneously progress to invasive tumors after 4 -8 weeks, making them an ideal model for the study of DCIS tumors (10,26). When designing our model system, we opted to culture both cell lines directly in

The role of myoepithelial cells in DCIS
the same cell culture dish so that the cell populations would not be separated by a membrane (as in a transwell plate). To track the cell lines and isolate them post-coculture for analysis, we established a stable GFP-expressing MCF10DCIS cell line (hereafter referred to as "MCF10DCIS-GFP").
After MCF10DCIS-GFP cells were cocultured with tumorassociated Hs578bst myoepithelial cells for 4 days, we performed live FACS to isolate GFP-positive MCF10DCIS-GFP cells for gene expression and phenotype studies (supplemental Fig. S1). We observed that following coculture, MCF10DCIS cells exhibited a spindle-like shape as opposed to the polygonal shape classically seen in non-cocultured counterparts. This phenotypic change suggested that these cells may have undergone the EMT (supplemental Fig. S2). To further examine this possibility, we performed quantitative RT-PCR (qRT-PCR) analysis of a panel of EMT-related genes (FOXC2, SLUG, SNAIL, TWIST1, TWIST2, vimentin, ZEB1, ZEB2, and E-cadherin) on sorted, cocultured MCF10DCIS-GFP cells compared with sorted, non-cocultured counterparts. FOXC2, SLUG, SNAIL, TWIST1, TWIST2, ZEB1, and ZEB2 are well-known transcription factors involved in driving the EMT, and vimentin up-regulation with E-cadherin down-regulation are two well-known hallmarks of the EMT (28,29). In line with morphological changes, the expression levels of EMT transcription factors (FOXC2, SNAIL, ZEB1, and ZEB2) and vimentin were elevated, and E-cadherin expression was decreased in cocultured MCF10DCIS cells (Fig. 1A). In addition, immunofluorescence staining data showed that the membranous staining of E-cadherin was lost in cocultured MCF10DCIS cells (supplemental Fig. S3). These results support our observation that coculture with tumor-associated myoepithelial cells promotes the occurrence of the EMT in MCF10DCIS cells.
Activation of the EMT in cells is known to enhance basal-like and stem-cell phenotypes (18). Therefore, we performed FACS analysis of CD44 (the basal-like cancer stem-cell marker), CD49f (the stem/progenitor cell marker), and EpCAM (the luminal stem/progenitor cell marker) on non-cocultured and cocultured MCF10DCIS-GFP cells. Coculture with Hs578bst cells significantly increased the CD44 ϩ CD49f ϩ cell population in MCF10DCIS-GFP cells (69.4% versus 32.1% of the noncocultured control), whereas the EpCAM ϩ cell population was dramatically decreased in cocultured MCF10DCIS cells (38.4% versus 83.5% of the non-cocultured control) (Fig. 1B). These FACS data indicate that tumor-associated myoepithelial cells promote basal-like and stem-cell phenotypes while concurrently inhibiting the luminal feature of MCF10DCIS cells, consistent with the enhanced EMT phenotype (supplemental Figs. S2 and S3 and Fig. 1A).
Induction of the EMT is one of the key mechanisms that promote invasiveness and metastasis of cancer cells (16). To examine whether the enhanced basal-like/EMT phenotype caused by coculture with tumor-associated myoepithelial cells promotes the invasiveness of MCF10DCIS cells, we performed transwell-based migration and invasion assays. As shown in Fig.  1C, cocultured MCF10DCIS-GFP cells exhibited higher migratory and invasive activities than non-cocultured control cells. The EMT has also been shown to enhance stem-cell features (18). To investigate the effect of coculture on the MCF10DCIS cancer stem cell (CSC) population, we conducted stem-cell sphere formation assays. As expected, coculture with Hs578bst cells promoted MCF10DCIS CSC self-renewal when compared with non-cocultured control cells (Fig. 1D). This result indicates that coculture with Hs578bst leads to an increase in the DCIS cancer stem-cell population.
To further understand whether these observed coculture events are specific to tumor-associated myoepithelial cells, we examined MCF10DCIS-GFP cells that were cocultured with MCF10A cells as a non-myoepithelial control. As shown in supplemental Fig. S4, coculture with MCF10A had no significant impact on the expression of EMT-programming genes, basal-like, luminal, invasiveness, and stemness features of MCF10DCIS cells. These data suggest that the oncogenic effects induced by coculture in MCF10DCIS cells are specifically derived from the interaction with Hs578bst myoepithelial cells.

Coculture induces increased TGF␤1 secretion from tumorassociated myoepithelial cells and activation of TGF␤/Smads signaling in MCF10DCIS cells
We next sought to identify the underlying mechanism by which coculture with tumor-associated myoepithelial cells triggers the EMT and invasive progression of DCIS cells. Because TGF␤ signaling is a well-known driver of EMT, we explored whether it is involved in this interaction. We first examined the levels of secreted TGF␤1 in the conditioned media from cocultured and non-cocultured cells. Using an ELISA, we found that there were significantly higher levels of secreted TGF␤1 in the conditioned media from cocultures compared with the conditioned media from non-cocultured cells ( Fig. 2A). To decipher which cell line was producing the increased levels of TGF␤1, we performed qRT-PCR analysis of TGF␤1 mRNA expression in sorted GFP-positive MCF10DCIS-GFP and GFP-negative Hs578bst cells compared with their respective non-cocultured control cells. Coculture significantly up-regulated TGF␤1 mRNA expression in Hs578bst myoepithelial cells, but not in MCF10DCIS-GFP cells (Fig. 2B). These findings suggest that coculture elicits increased production and secretion of TGF␤1 by tumor-associated myoepithelial cells.
To examine whether TGF␤1 is necessary for the observed effect of coculture on MCF10DCIS cells, we performed TGF␤1 knockdown studies using two distinct TGF␤1 siRNAs. As shown in Fig. 2C, transfection of siTGF␤1-1 siRNA into Hs578bst completely abolished the increased levels of TGF␤1 observed in Hs578bst cells after coculture. Another TGF␤1 siRNA (siTGF␤1-2) also eliminated Ͼ 90% of induced TGF␤1 mRNA levels in Hs578bst cells after coculture (Fig. 2C). Consistent with qRT-PCR data, transfection of siTGF␤1-1 into Hs578bst cells 24 h before coculture completely abolished the increased levels of secreted TGF␤1 seen in conditioned media of cocultures, and siTGF␤1-2 also eliminated Ͼ 90% of increased TGF␤1 secretion (Fig. 2D). These knockdown data demonstrate that the coculture-induced increase in secreted TGF␤1 is due to the increased TGF␤1 expression and secretion by Hs578bst cells. We next examined the impact of the increased concentration of secreted TGF␤1 on TGF␤ signaling pathway activation in MCF10DCIS cells through Western blot analysis of phospho-Smad2. We found that TGF␤1 knockdown in Hs578bst cells suppressed ϳ60 -70% of the coculture-induced Smad2 phosphorylation (Fig. 2E). This demonstrates that the coculture-mediated increase in TGF␤1 secretion from tumor-associated myoepithelial cells is the predominant con-tributor to coculture-induced activation of the TGF␤/Smads pathway in MCF10DCIS cells. Moreover, when we treated MCF10DCIS cells with recombinant TGF␤1, we observed identical results as those seen in cocultured cells. (Fig. 2F). These data validate that TGF␤1 is the primary effector of the coculture-mediated phenotypic changes. In addition to affect-

The role of myoepithelial cells in DCIS
ing MCF10DCIS cells, TGF␤ signaling was also activated in Hs578bst cells themselves (supplemental Fig. S5).
To further confirm whether TGF␤ receptors (TGF␤Rs) of MCF10DCIS cells mediate the coculture effect of myoepithelial cell-secreted TGF␤1, we treated MCF10DCIS cells with SB431542, a potent, specific inhibitor of TGF␤ type I receptors during their coculture with Hs578bst cells (30). Treatment with SB431542 completely abolished the coculture-mediated increase in Smad2 phosphorylation in MCF10DCIS cells (Fig.  2G). We also examined whether SB431542 treatment affects . Experiments were performed in triplicate. C, analysis of TGF␤1 knockdown by siRNA in cocultured Hs578bst myoepithelial cells. qRT-PCR analysis of TGF␤1 expression was performed on sorted non-cocultured and cocultured Hs578bst cells with or without TGF␤1 knockdown by siRNA. Two distinct TGF␤1 siRNAs (siTGF␤1-1 and siTGF␤1-2) that target two different sequence regions of TGF␤1 mRNA and control siRNA (siControl) were included in knockdown experiments. Experiments were performed in triplicate. D, ELISA analysis of TGF␤1 in conditioned media from three different cultures, including Hs578bst cultures (the non-cocultured control) and cocultures of MCF10DCIS cells with siControl-, siTGF␤1-1-or siTGF␤1-2-transfected Hs578bst cells. Experiments were performed in triplicate. E, TGF␤1 knockdown in Hs578bst myoepithelial cells impairs coculture-induced activation of TGF␤ signaling in MCF10DCIS cells. Western blot analysis of phospho-Smad2 (p-Smad2), total Smad2, and ␣-tubulin was performed on MCF10DCIS-GFP cells cocultured with or without siControl-, siTGF␤1-1-, or siTGF␤1-2-transfected Hs578bst cells. Non-cocultured and cocultured MCF10DCIS-GFP cells were sorted based on their GFP positivity before they were subjected to Western blot experiments. F, Western blot analysis of phospho-Smad2, total Smad2, and ␣-tubulin in mock-treated (control) and TGF␤1-treated MCF10DCIS-GFP cells. G, TGF␤ receptors are responsible for coculture-induced activation of TGF␤ signaling in MCF10DCIS cells. Western blot analysis of phospho-Smad2, total Smad2, and ␣-tubulin was performed on sorted non-cocultured (control), cocultured, and cocultured ϩ SB431542-treated MCF10DCIS-GFP cells. H, the TGF␤/Smads pathway is activated in MCF10DCIS cells cocultured with Hs578bst cells under a transwell-based coculture setting. Western blot analysis of phospho-Smad2, total Smad2, and ␣-tubulin was performed on non-cocultured and cocultured MCF10DCIS cells with or without SB431542 treatment. I, coculture with Hs578bst myoepithelial cells activated TGF␤ signaling in MCF10A, but not in MDA-MB-231 cells. MCF10A and MDA-MB-231 cells were cocultured with Hs578bst cells for 4 days under a transwell setting before they were subjected to Western blot analysis of phospho-Smad2, total Smad2, and ␣-tubulin. Their respective non-cocultured cells were included as controls. The arrow indicates total Smad2. Western blot data were quantified as described under "Experimental Procedures." The quantitative phospho-Smad2 and total Smad2 data were normalized to their respective ␣-tubulin. **, p Ͻ 0.01; ***, p Ͻ 0.001; ns, not significant. myoepithelial TGF␤1 synthesis and secretion. As shown in supplemental Fig. S6, SB431542 treatment did not cause significant changes in the coculture-induced increases in myoepithelial TGF␤1 synthesis and secretion. This indicates that the inhibitory effect of SB431542 resulted from suppressing activation of MCF10DCIS TGF␤Rs, not from affecting myoepithelial TGF␤1 production. These findings together demonstrate that coculture with tumor-associated myoepithelial cells triggers activation of the TGF␤/TGF␤Rs/ Smads pathway in MCF10DCIS cells, mainly initiated by increased TGF␤1 secretion from tumor-associated myoepithelial cells.
To further understand how this cell-cell interaction occurs, we performed a coculture experiment using the transwell model. This restricted the cocultured cell interactions to those mediated by secreted factors. Similar to the direct coculture results, transwell-based coculture with Hs578bst myoepithelial cells induced activation of TGF␤ signaling in MCF10DCIS cells ( Fig. 2H), although at a lesser extent when compared with the direct coculture system (Fig. 2, E and G). Consistently, expression of EMT-programming genes was induced in MCF10DCIS cells under this transwell-based coculture with Hs578bst cells (supplemental Fig. S7). This result suggests that secreted cytokines/exosomes are involved in the interaction between DCIS cells and tumor-associated myoepithelial cells and contribute to coculture events.
To extend our findings into additional cell models, we performed transwell-based coculture studies on the immortalized HMEC line MCF10A and the metastatic triple-negative breast cancer cell line MDA-MB-231. Western blot data showed that coculture with Hs578bst cells activated TGF␤ signaling in MCF10A cells, but not in MDA-MB-231 cells (Fig. 2I). Moreover, expression of EMT-programming genes was also induced in cocultured MCF10A cells (supplemental Fig. S8). These findings suggest that these observed coculture events only occur in the cell-cell interaction between tumor-associated Hs578bst myoepithelial cells and HMEC/DCIS, but not invasive/metastatic breast cancer cells. Because invasive breast cancer cells were unaffected by coculture, these data may suggest that the observed coculture effects promote the invasive progression in early stage, non-invasive breast cancers. To further confirm that the observed coculture-induced activation of TGF␤ signaling in DCIS cells is specific to tumor-associated myoepithelial cells, we examined cocultures of MCF10DCIS-GFP and MCF10A. Indeed, coculture with MCF10A failed to activate this event (supplemental Fig. S4E). Consistent with our previous data from supplemental Fig. S4, this Western blot result suggests that the coculture-mediated changes result from specific interactions with Hs578bst tumor-associated myoepithelial cells.

Coculture-enhanced basal-like/EMT phenotypes and invasiveness/stemness of MCF10DCIS cells are due to activation of TGF␤ signaling
To investigate whether myoepithelial cell-secreted TGF␤1 contributes to observed phenotypic changes to MCF10DCIS cells, we performed the aforementioned expression profiling and FACS studies on MCF10DCIS-GFP cells cocultured with or without either control or TGF␤1-knockdown Hs578bst myoepithelial cells. The data showed that TGF␤1 knockdown in Hs578bst cells by either siTGF␤1-1 or siTGF␤1-2 significantly impaired the coculture-induced increase in expression of EMT-programming genes and down-regulation of E-cadherin expression (Fig. 3A), correlating with the effect of myoepithelial TGF␤1 knockdown to impair TGF␤ signaling activation in MCF10DCIS cells (Fig. 2E). Consistently, coculture with TGF␤1-knockdown Hs578bst cells also had a diminished effect to increase the basal-like/stem (CD44 ϩ CD49f ϩ ) cell population and decrease the luminal epithelial (EpCAM ϩ ) cell population in MCF10DCIS cells when compared with coculture with control Hs578bst cells (Fig. 3, B and C). These results demonstrate that TGF␤1 secreted from tumor-associated myoepithelial cells is the predominant factor triggering the EMT induction of cocultured MCF10DCIS cells.
To examine whether coculture-induced TGF␤ signaling activation contributed to the EMT in MCF10DCIS cells, we performed FACS analysis of CD44/CD49f/EpCAM and EMT gene expression profiling on cocultured MCF10DCIS-GFP cells treated with or without the TGF␤ receptor inhibitor (SB431542) and on non-cocultured MCF10DCIS-GFP cells treated with or without recombinant TGF␤1. The results indicated that the complete inhibition of TGF␤ signaling by SB431542 treatment fully abolished the coculture-induced increase in the CD44 ϩ CD49f ϩ basal/stem-cell population in MCF10DCIS, cells, whereas the percentage of luminal EpCAM ϩ cells in cocultured MCF10DCIS cells reverted to the non-cocultured value (Fig. 4A). Consistently, TGF␤1-treated MCF10DCIS cells exhibited phenotypes similar to cocultured cells, including increased CD44 ϩ CD49f ϩ and decreased EpCAM ϩ cell populations (Fig. 4B).
To unravel the role of TGF␤ signaling in the coculture-induced EMT of MCF10DCIS cells, we performed qRT-PCR analysis to examine a panel of EMT-related genes (FOXC2, SNAIL, vimentin, ZEB1, and ZEB2) in non-cocultured, cocultured, cocultured ϩ SB431542-treated, and TGF␤1-treated MCF10DCIS-GFP cells. We found that expression of these five EMT genes was up-regulated in cocultured and TGF␤1-treated MCF10DCIS cells when compared with control cells and that EMT gene induction was completely abolished by TGF␤ receptor inhibitor treatment (Fig. 4C). To confirm these results, we performed Western blot analysis of vimentin and E-cadherin protein expression. We observed increased vimentin and decreased E-cadherin protein levels in cocultured and TGF␤1-treated MCF10DCIS cells and found that treatment with SB431542 inhibited these expression changes (Fig. 4D). These results reveal that the induction of the EMT in DCIS cells by coculture with tumor-associated myoepithelial cells is attributable to activation of TGF␤ signaling in DCIS cells.
To determine the role of TGF␤ signaling in the cocultureenhanced invasiveness/stemness of MCF10DCIS cells, we performed migration, invasion, and stem-cell sphere formation assays on non-cocultured, cocultured, cocultured ϩ SB431542treated, and TGF␤1-treated MCF10DCIS-GFP cells. As expected, the coculture-induced enhancement of migration and invasion in MCF10DCIS cells was recapitulated by TGF␤1

The role of myoepithelial cells in DCIS
treatment. Treatment with the TGF␤ receptor inhibitor (SB431542) completely abolished these enhanced phenotypes (Fig. 5, A and B). Increased CSC sphere formation by cocultured MCF10DCIS cells was also inhibited by treatment with the TGF␤ receptor inhibitor (Fig. 5C). Similar to these outcomes, TGF␤1 treatment led to increased formation of MCF10DCIS CSC spheres (Fig. 5D). These findings indicate that the tumorassociated myoepithelial cell-induced increases in invasiveness and stemness result from activation of TGF␤ signaling in DCIS cells.

Tumor-associated myoepithelial cells promote the invasive progression of MCF10DCIS cells in vivo
We next investigated whether tumor-associated myoepithelial cells could promote the invasive progression of DCIS cells in vivo. We performed a xenograft tumor study wherein we transplanted MCF10DCIS cells with or without Hs578bst myoepithelial cells. To examine and track the fate of co-injected tumor-associated myoepithelial cells during the development of MCF10DCIS xenograft tumors, we created stable GFP-expressing Hs578bst cells (hereafter referred to as "Hs578bst-GFP cells") and co-transplanted them with MCF10DCIS cells into the mammary fat pads of nude mice for the development of xenograft tumors. MCF10DCIS xenografts have been shown to spontaneously progress into invasive tumors as early as 4 weeks after transplantation (10). Therefore, we restricted our xenograft study to 4 weeks, so we could specifically examine the effect of tumor-associated myoepithelial cells on in vivo DCIS tumors. The study showed that co-transplantation with Hs578bst-GFP cells promoted MCF10DCIS tumor growth in the mammary fat pads of nude mice (Fig. 6A). To examine the histological status of the tumor, we performed immunohistochemistry (IHC) analysis of ␣ smooth muscle actin (␣-SMA), a protein marker for detecting both myoepithelial and stromal cells, on xenograft tissue sections. The IHC result showed that xenograft tumors developed from only MCF10DCIS cells exhibited the typical DCIS architecture, wherein DCIS tumor lesions were surrounded by a myoepithelial cell layer and stroma (Fig. 6B, left). In contrast, xenograft tumors developed from MCF10DCIS cells co-transplanted with Hs578bst-GFP cells manifested the invasive tumor phenotype, wherein tumor cells broke through the myoepithelial cell layer and invaded into stroma (Fig. 6B, right). To determine whether co-transplanted tumor-associated myoepithelial cells activated TGF␤ signaling in DCIS cells in vivo, we performed IHC analysis of phospho-Smad2 in xenograft tumors. We found that phospho-Smad2 staining was mainly detected in the myoepithelium/ stroma and almost negative (Յ5% positivity) in DCIS tumor , and EpCAM was performed on mock-treated (control) and TGF␤1-treated MCF10DCIS-GFP cells as described in A. Cells were treated with 5 nM recombinant TGF␤1 for 3 days before FACS analysis. C and D, the enhanced EMT in cocultured MCF10DCIS cells is attributable to activation of the TGF␤ pathway. C, qRT-PCR analysis of five EMT-programming genes was performed on non-cocultured (control), cocultured, cocultured ϩ SB431542-treated, and TGF␤1-treated MCF10DCIS-GFP cells after GFP-positive cells were sorted. Experiments were performed in triplicate. *, p Ͻ 0.05 versus the non-cocultured control data set; #, p Ͻ 0.05 versus the cocultured data set. D, Western blot analysis of vimentin, E-cadherin, and ␣-tubulin was performed on non-cocultured (control), cocultured, cocultured ϩ SB431542-treated, mock-treated (control), and TGF␤1-treated MCF10DCIS-GFP cells. In the cocultured Western blot data (left), non-cocultured and cocultured MCF10DCIS-GFP cells were sorted based on their GFP positivity before they were subjected to Western blot experiments. The quantitative vimentin and E-cadherin data were normalized by their respective ␣-tubulin.
lesions of control xenograft tumors (Fig. 6C, left and right), which is consistent with the previously reported finding that the TGF␤ pathway was activated in myoepithelial cells of MCF10DCIS xenograft tumors (10). In contrast, Ͼ 50% of tumor cells in xenograft lesions formed from MCF10DCIS cotransplanted with Hs578bst-GFP had significant phospho-Smad2 staining (Fig. 6C, middle and right). Using RNA iso-lated from tumor cells from the resected xenograft tumors, we performed qRT-PCR and found that the expression of EMT-related genes was induced in xenograft tumor cells developed from MCF10DCIS co-transplanted with Hs578bst-GFP compared with those from MCF10DCISonly xenografts (Fig. 6D). These findings indicate that cotransplantation of tumor-associated myoepithelial cells To distinguish the co-injected Hs578bst-GFP myoepithelial cells from host myoepithelial cells, we performed anti-GFP IHC analysis on xenograft tumor tissue and confirmed the presence of GFP-positive Hs578bst myoepithelial cells in the xenograft tumors generated from transplanted MCF10DCIS cells mixed with Hs578bst-GFP cells (supplemental Fig. S9). Moreover, Hs578bst-GFP cells could also be detected in mammary gland tissue 4 weeks after they were transplanted into nude mice alone (supplemental Fig. S10). These data indicate that Hs578bst-GFP cells can be successfully transplanted into the mammary gland in vivo.

The role of myoepithelial cells in DCIS TGF␤-induced miR-10b-5p up-regulation contributes to tumor-associated myoepithelial cell-triggered invasive phenotypes in MCF10DCIS cells
miRNA dysregulation has been shown to play a critical role in tumorigenesis. To identify the miRNAs involved in the TGF␤-mediated DCIS-to-IDC transition, we performed miRNA expression profiling analysis on non-cocultured and cocultured MCF10DCIS cells. Using miRNA PCR arrays that detect 84 breast cancer-related miRNAs, we identified 25 up-regulated (Ն 2-fold) and 21 down-regulated (ՅϪ2-fold)

Figure 7. Tumor-associated myoepithelial cells activate the TGF␤/miR-10b-5p axis in DCIS cells and up-regulation of miR-10b-5p expression contributes to the coculture-enhanced EMT, invasiveness, and stemness of DCIS cells.
A, PCR array profiling of miRNA expression in control and cocultured MCF10DCIS cells. miRNA expression data were plotted into a two-dimensional dot plot. Up-regulated (Ն 2-fold) and down-regulated (Յ Ϫ2-fold) miRNAs in sorted cocultured MCF10DCIS-GFP cells compared with the non-cocultured control are indicated. TGF␤-regulated miRNAs are depicted in the dot plot. B, qRT-PCR analysis of miR-10b-5p expression was performed on the sorted cell set as described in the legend to Fig. 4C. C, qRT-PCR analysis of miR-10b-5p expression was performed on RNA samples as described in the legend to Fig. 6D. D, qRT-PCR analysis of four EMT-programming genes was performed on MCF10DCIS cells stably overexpressing the control scramble, miR-10b, or miR-10b sponge RNA. Expression bar graph data shown in B-D were plotted based on triplicate experiments. E, inhibition of miR-10b by the sponge RNA partially suppresses coculture-promoted migratory and invasive activities of MCF10DCIS cells. Migration and invasion assays were performed on cocultured MCF10DCIS-GFP cells overexpressing the control scramble or miR-10b sponge RNA. Non-cocultured MCF10DCIS-GFP cells overexpressing the control scramble RNA served as a control. Non-cocultured and cocultured cells were sorted based on their GFP positivity before they were subjected to migration and invasion assays. Quantitative bar graph data (n ϭ 3) were plotted as described in the legend to Fig. 1C. F, inhibition of miR-10b by the sponge RNA partially suppresses coculture-promoted CSC self-renewal of MCF10DCIS cells. Stem-cell sphere formation assays were performed on non-cocultured and cocultured cells as described in E. Quantitative CSC sphere formation data were plotted based on triplicate experiments. Error bars, S.D. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. miRNAs in cocultured MCF10DCIS cells (Fig. 7A). Among these differentially expressed miRNAs, five miRNAs (miR-10a-5p, miR-10b-5p, miR-145-5p, miR-199a-5p, and miR-199a-3p) have been shown to be positively regulated, and three miRNAs (miR-200b-3p, miR-200c-3p, and miR-205-5p) have been shown to be negatively regulated by TGF␤ signaling (Fig. 7A) (31)(32)(33)(34)(35). These miRNA expression profiling data are consistent with our prior findings demonstrating that the TGF␤ pathway is activated in cocultured MCF10DCIS cells (Fig. 2). In a recent publication (36) studying the CSC population of MCF10DCIS cells, we found that expression of miR-10b was up-regulated in aggressive cell clones isolated from CSCs compared with nonaggressive clones. Gain-and loss-of-function studies showed that miR-10b up-regulation enhanced the tumorigenic features (e.g. stemness and invasiveness) of aggressive MCF10DCIS cell clones (36). In the current study, we identified miR-10b-5p upregulation in cocultured MCF10DCIS cells and postulated that TGF␤-mediated miR-10b dysregulation contributes to the invasive progression of DCIS cells.
To test this hypothesis, we first examined whether miR-10b-5p up-regulation in cocultured MCF10DCIS cells was due to activation of TGF␤ signaling. miR-10b-5p expression in noncocultured, cocultured, cocultured ϩ SB431542-treated, and TGF␤1-treated MCF10DCIS cells was analyzed using qRT-PCR. As shown in Fig. 7B, miR-10b-5p up-regulation by coculture was recapitulated by TGF␤1 treatment and abolished by treatment with the TGF␤ receptor inhibitor. We next analyzed miR-10b-5p expression in tumor cells of xenografts generated from MCF10DCIS cells co-transplanted with or without Hs578bst cells. In line with the in vitro data, the in vivo data demonstrated that tumor-associated myoepithelial cells induced up-regulation of miR-10b-5p expression in MCF10DCIS xenograft tumor cells (Fig. 7C). These data indicate that coculture-induced up-regulation of miR-10b-5p expression in MCF10DCIS cells is due to activation of TGF␤ signaling in MCF10DCIS cells. To reveal whether miR-10b-5p is involved in regulating the expression of EMT-programming genes, qRT-PCR analysis was performed on MCF10DCIS cells overexpressing miR-10b or the miR-10b sponge, which inhibits miR-10b function. Overexpression of miR-10b selectively upregulated the expression of FOXC2 and ZEB1, whereas their expression was inhibited by overexpression of the miR-10b sponge (Fig. 7D). These findings indicate that miR-10b is involved in regulating a subset of EMT-programming genes.
We next investigated whether coculture-mediated up-regulation of miR-10b-5p contributes to the invasiveness/stemness of MCF10DCIS cells. We cocultured control scramble RNAexpressing and miR-10b sponge-overexpressing MCF10DCIS-GFP cells with Hs578bst and then subjected them to migration, invasion, and stem-cell sphere formation assays. miR-10b inhibition partially suppressed the coculture-enhanced migration, invasion, and CSC self-renewal of MCF10DCIS cells (Fig. 7, E  and F). These data suggest that miR-10b is a downstream target of the TGF␤ pathway in MCF10DCIS cells, which mediates the effect of coculture-activated TGF␤ signaling to promote the invasiveness and stemness of DCIS cells.

miR-10b-5p targets and inhibits RB1CC1 expression to promote the migration of MCF10DCIS cells
To unravel how the TGF␤/miR-10b-5p signaling axis enhances the invasive progression of DCIS cells, we identified putative gene targets of miR-10b-5p using PicTar, TargetScan, and Miranda algorithms (37)(38)(39). Through in silico analysis, we found 15 predicted miR-10b-5p target genes that are functionally implicated in the regulation of cell adhesion, cell cycle, and cell survival. To validate whether they are genuine miR-10b-5p targets, we performed an RNA pull-down study using the biotin-conjugated miR-10b-5p RNA as described previously (40). Five genes were significantly enriched (Ն 2-fold) by miR-10b-5p (Fig. 8A). Among these five genes, RB1-inducible coiled-coil 1 (RB1CC1, also known as FIP200) is particularly relevant, because it has been reported that RB1CC1 is a potential tumor-suppressor gene in breast cancer (41).
To further verify the RNA pull-down of RB1CC1, we analyzed the RB1CC1 3Ј-UTR using a luciferase reporter. Co-transfection of the miR-10b expression plasmid with the wild-type RB1CC1 3Ј-UTR reporter plasmid led to ϳ70% suppression of reporter activity (p Ͻ 0.01, n ϭ 3), whereas miR-10b had no significant effect on the activity of the mutated RB1CC1 3Ј-UTR reporter with mutations at the miR-10b-5p recognition site (Fig. 8B). Consistent with the reporter data, miR-10b overexpression significantly suppressed ϳ65% of RB1CC1 expression, whereas inhibition of miR-10b using the inhibitory sponge modestly increased (ϳ1.3-fold) RB1CC1 protein levels (Fig.  8C). These data together demonstrate that RB1CC1 is a genuine target of miR-10b-5p.
To analyze the role of miR-10b-5p-mediated RB1CC1 downregulation, we utilized siRNA to knock down RB1CC1 in MCF10DCIS cells. Western blot analysis confirmed the high knockdown efficiency of two distinct RB1CC1 siRNAs (supplemental Fig. S11). Using wound-healing assays, we found that RB1CC1 knockdown increased the migration of MCF10DCIS cells compared with the control (Fig. 8D), suggesting that down-regulation of RB1CC1 contributes to the effect of miR-10b-5p on promoting the migratory activity of MCF10DCIS cells.

RB1CC1 expression is down-regulated in the invasive progression of MCF10DCIS cells in vitro and in vivo and in triple-negative breast cancer
After identifying RB1CC1 as the downstream target of miR-10b-5p, we further examined whether coculture or TGF␤1-mediated miR-10b-5p up-regulation in MCF10DCIS cells (Fig. 7B) results in the down-regulation of RB1CC1 expression. As expected, the Western blot data showed that RB1CC1 expression was decreased in MCF10DCIS cells cocultured with Hs578bst cells compared with non-cocultured control cells. Moreover, this down-regulation was rescued when cocultured MCF10DCIS cells were treated with the TGF␤ receptor inhibitor (Fig. 9A, left). Consistently, TGF␤1 treatment also resulted in RB1CC1 down-regulation (Fig. 9A, right). These findings suggest that the TGF␤/miR-10b-5p/RB1CC1 signaling axis occurs in cocultured MCF10DCIS cells.

The role of myoepithelial cells in DCIS
To investigate whether these in vitro results could be recapitulated in DCIS tumors in vivo, we analyzed RB1CC1 protein expression in MCF10DCIS xenograft tumors with or without co-transplantation of tumor-associated myoepithelial cells using IHC. Consistent with the in vitro results shown in Fig. 9A, IHC staining of RB1CC1 was decreased in MCF10DCIS xenograft tumors with co-transplantation of Hs578bst myoepithelial cells compared with that of MCF10DCIS only tumors (Fig. 9B). Therefore, our studies suggest that the TGF␤/miR-10b-5p/RB1CC1 signaling axis The expression values were normalized to actin. The housekeeping gene GAPDH was included in the analysis as a negative control. B, overexpression of miR-10b-5p inhibits the luciferase expression of the wild-type, but not the mutated, RB1CC1 3Ј-UTR reporter. A map for the predicted miR-10b-5p targeting site in the 3Ј-UTR of the RB1CC1 mRNA is shown on the left. A DNA fragment with mutations in the miR-10b seeding site of RB1CC1 3Ј-UTR was used to construct the mutated reporter and its RNA sequence is shown under the map with its wild-type and miR-10b-5p sequences. HEK-293T cells were transfected with the wild-type or mutated RB1CC1 3Ј-UTR reporter plasmid DNA along with either the control scramble or the miR-10b expression plasmid. All cell samples were also co-transfected with the Renilla expression plasmid, which was used as a transfection efficiency control. Dual-Luciferase assays were performed on transfected cells 24 h after transfection. The measured luciferase activity values were normalized by Renilla activity values to make a quantitative bar graph (shown on the right). Error bars, S.D. of the data set (n ϭ 3). C, miR-10b-5p negatively regulates endogenous RB1CC1 expression. Western blot analysis of RB1CC1 and ␣-tubulin was performed on MCF10DCIS-GFP cells overexpressing the control scramble, miR-10b, or miR-10b sponge RNA. The quantitative RB1CC1 data were normalized by their respective ␣-tubulin. D, knockdown of RB1CC1 enhances the migratory activity of MCF10DCIS cells. Wound-healing assays were performed for 16 h to measure the migratory activity of RB1CC1-knockdown cells compared with that of siControl-transfected cells. Two different RB1CC1 siRNAs (siRB1CC1-1 and siRB1CC1-2) were utilized in the knockdown experiment. Their knockdown efficiency is shown in supplemental Fig. S11. The representative pictures of scratched wounds for three different siRNA-transfected cells are shown on the left. The cell pictures were analyzed using ImageJ to measure wound closure percentage as described under "Experimental procedures." The wound closure percentage data (n ϭ 3) were plotted to make a bar graph (shown on the right). Error bars, S.D. *, p Ͻ 0.05; **, p Ͻ 0.01.
contributes to the invasive progression of in vivo DCIS tumors.
Given that MCF10DCIS cells are a premalignant basal-like DCIS tumor model, they spontaneously progress into basallike/triple-negative invasive breast cancer in immunodeficient mice (10,26). Therefore, we predicted that RB1CC1 expression would be down-regulated in triple-negative breast cancers (TNBCs) compared with other breast cancer subtypes. To test this prediction, we performed in silico analysis of RB1CC1 expression in breast cancers using the Oncomine cancer gene expression database (http://www.oncomine.org) 3 (42). By analyzing two different breast cancer data sets (from Esserman et al. (43) and Tabchy et al. (44)), we found that RB1CC1 expression was significantly down-regulated in TNBCs compared with other non-TN breast cancers (Fig. 9C). These in silico data suggest that RB1CC1 down-regulation, potentially through the TGF␤/miR-10b-5p axis, is relevant in the development of TNBCs.

Discussion
In this study, we have unraveled a novel role for tumor-associated myoepithelial cells in the progression of DCIS to IDC. Using in vitro coculture studies, we show that tumor-associated myoepithelial cells promote the EMT and basal-like phenotypes of DCIS cells through activation of TGF␤ signaling. Interestingly, coculture-activated TGF␤ signaling also results in increased CD44 ϩ and decreased EpCAM ϩ DCIS cells, suggesting that there is an increase in basal-like and stem cell-like characteristics and a decrease in the luminal cell properties of DCIS cells. This result is consistent with the previous finding that activation of the TGF␤ pathway is required for maintaining basal-like CD44 ϩ breast cancer cells (45). Moreover, enhanced migration, invasion, and CSC self-renewal observed in DCIS cells cocultured with tumor-associated myoepithelial cells also relied on TGF␤ activation. These in vitro data suggest that tumor-associated myoepithelial cells promote the invasive progression of DCIS cells through enhancing their basal-like/EMT phenotypes, invasiveness, and stemness. Importantly, our in vivo study shows that tumor-associated myoepithelial cells activate TGF␤ signaling and the EMT in xenograft tumors and enhance the progression of in situ xenograft tumors to invasive tumors. To our knowledge, this is the first experimental evidence supporting the hypothesis that tumor-associated myoepithelial cells drive the progression of DCIS to invasive carcinomas. Our studies also highlight the critical role of the TGF␤/ EMT pathway in the transition of DCIS to IDC. Because we used the basal-like DCIS cell model in our study, our findings are in line with previous clinical observations that the EMT increases during the progression of in situ to invasive basal-like breast cancer (46). Moreover, our investigation reveals that interaction with DCIS cells increases myoepithelial cell production and secretion of TGF␤1, which subsequently contribute to activation of the TGF␤/Smads pathway in DCIS cells. How DCIS cells stimulate this change in tumor-associated myoepithelial cells is unclear. However, our transwell-based coculture studies suggest that secreted factors (e.g. cytokines, exosomes) from these two cell types are involved in this interaction. Given that TGF␤1 knockdown in tumor-associated myoepithelial cells did not completely eliminate coculture-induced activation of TGF␤ signaling in DCIS cells and that treatment with the TGF␤R inhibitor did, it is likely that other TGF␤ family ligands (e.g. TGF␤2 and TGF␤3) secreted from tumorassociated myoepithelial cells (and/or from DCIS cells) are involved in this coculture event. Further investigations are needed to reveal this mechanism. Isolation and culture of primary DCIS-associated myoepithelial cells is highly difficult, because the myoepithelial cell layer is eliminated during the DCIS-to-IDC progression. This challenge restricts the functional studies of tumor-associated myoepithelial cells in DCIS research. The studies of the cell-cell interaction between MCF10DCIS and Hs578bst cells in our current report for the first time reveal the oncogenic role of tumor-associated myoepithelial cells in promoting the invasive transformation of DCIS cells.
Two recent studies have raised a debate about the role of the EMT in invasive and metastatic processes of cancer cells (47,48). Through studies of mouse models of pancreatic ductal adenocarcinoma with deletion of Snail or Twist, two key transcription factors responsible for the EMT, Zheng et al. (47) found that Snail-or Twist-induced EMT was not required for the invasive progression and metastasis of pancreatic ductal adenocarcinoma. However, their studies could not rule out the possibility that other EMT-inducing factors may compensate for the loss of Snail or Twist to induce invasion and metastasis. By employing in vivo EMT-tracing systems, Fischer et al. (48) reported that the EMT is not necessary for lung metastasis of two mouse models of breast adenocarcinoma resembling the human luminal breast cancer subtype. Nevertheless, their studies reveal that EMT tumor cells were chemoresistant and significantly contributed to recurrent lung metastasis formation after chemotherapy (48). Given that the animal breast tumor models used in their studies are the luminal subtype, their findings suggest that the EMT is not required for the invasive progression of luminal breast cancers. However, our study of basallike MCF10DCIS cells suggests that basal-like DCIS is potentially transformed to become invasive by the EMT induced by interaction with tumor-associated myoepithelial cells. Our findings highlight a novel myoepithelial cell-based mechanism that potentially contributes to the high rate of the invasive transition of basal-like/triple-negative DCIS observed in animal and clinical studies. Both our study and that of Fischer et al. (48), therefore, suggest that the importance of the EMT in invasion and metastasis of breast cancer may be molecular subtype-specific.
Our mechanistic studies have further revealed that TGF␤ signaling activation dysregulates the oncogenic TGF␤/miR-10b-5p signaling axis within DCIS cancer cells. miR-10b is an oncogenic microRNA that is a well-known metastatic promoter in breast cancer and participates in Twist-mediated EMT programming (49). miR-10b targets the tumor suppressor HOXD10 and inhibits its expression, which results in the upregulated expression of migration-promoting and pro-metastatic factors (49). Therapeutic inhibition of miR-10b has been shown to suppress metastasis of in vivo mammary tumors in animals (50,51). miR-10b expression has been found to be preferentially up-regulated in basal-like/triple-negative breast cancer cell lines and in advanced/metastatic breast cancers (52,53). Recently, miR-10b was shown to be overexpressed in breast cancer stem cells (BCSCs) and to be required for self-renewal of BCSCs (54). The oncogenic role of miR-10b in BCSCs is found to result from its inhibitory effect on PTEN expression, leading to activation of the oncogenic PI3K/AKT pathway (54). We have found that miR-10b-5p expression is aberrantly up-regu-lated in aggressive cell clones isolated from MCF10DCIS cells and that this dysregulation is required for aggressive and invasive phenotypes of cell clones (36). Here, we identified miR-10b-5p as a critical TGF␤-induced microRNA implicated in the TGF␤-activated EMT, migration, invasion, and CSC self-renewal in DCIS cells. Importantly, interaction with tumor-associated myoepithelial cells results in up-regulation of miR-10b-5p expression in DCIS cells in vitro and in vivo, suggesting that the TGF␤/miR-10b-5p axis functionally contributes to promoting the invasive progression of DCIS.
Through the screening conducted in our miR-10b-5p pulldown experiment, we identified RB1CC1 as a novel target of miR-10b-5p. RB1CC1 knockdown impairs the migratory ability of DCIS cells, suggesting that miR-10b-mediated down-regulation of RB1CC1 contributes to the oncogenic function of miR-10b-5p. Multiple lines of evidence have shown that RB1CC1 is a potential tumor-suppressor gene. It has been reported that about 20% of primary breast tumors contained truncating mutations at RB1CC1 gene loci (41). RB1CC1 is found to activate expression of tumor-suppressor genes (e.g. RB1 and p16) and stabilize p53 protein levels, resulting in the activation of these tumor suppressor pathways (55-58). In addition, RB1CC1 inhibits ␤-catenin-mediated transcription via promoting ␤-catenin ubiquitination (59). In this study, we have revealed a novel TGF␤/miR-10b-5p/RB1CC1 signaling axis that is dysregulated in DCIS cells through aberrant activation of TGF␤ signaling by the interaction with tumor-associated myoepithelial cells. We revealed the tumorigenic roles of this dysregulated axis in vitro and showed that it is a part of the tumor-associated myoepithelial cell-mediated promotion of in vivo DCIS progression to IDC. These findings suggest that dysregulation of the TGF␤/miR-10b-5p/RB1CC1 signaling axis is involved in driving the invasive transition from DCIS to IDC, at least in basal-like breast cancer development. Furthermore, our in silico analysis indicates that down-regulation of RB1CC1 expression is common in TNBC, suggesting that the TGF␤/ miR-10b-5p/RB1CC1 signaling axis is relevant to the development of TNBC. Although our findings highlight that dysregulation of RB1CC1 is potentially involved in the invasive development of in situ breast tumors, future in vivo studies will be needed to reveal the exact role of RB1CC1 in the DCIS-to-IDC transition.
In summary, our studies provide new insights into the oncogenic roles of tumor-associated myoepithelial cells and the TGF␤ pathway in the progression of in situ to invasive tumors. Our findings support the hypothesis that altered tumor-associated myoepithelial cells have a tumor-promoting effect on DCIS progression to IDC. Our studies reveal that the aberrantly increased secretion of TGF␤1 from tumor-associated myoepithelial cells critically contributes to this oncogenic effect. Moreover, our investigations reveal that dysregulation of microRNA expression in DCIS cells by the DCIS-myoepithelial interaction is potentially implicated in driving the transition of DCIS to IDC. Further investigations are needed to unveil the underlying mechanism for how interaction with DCIS cells elicits alterations in tumor-associated myoepithelial cells and increases their secretion of TGF␤1, which is a potential target for preventive therapy.

Cell culture
The tumor-associated myoepithelial cell line Hs578bst, the immortalized HMEC line MCF10A, and the metastatic breast cancer cell line MDA-MB-231 were purchased from the American Type Culture Collection (Manassas, VA). The human DCIS cell line MCF10DCIS was purchased from Asterand USA (Detroit, MI). These cell lines were cultured according to the manufacturer's instructions. The coculture of MCF10DCIS and Hs578bst cells was conducted at a ratio of 1:2 for 4 days. Cocultured cells were treated with SB431542 (10 M) (Sigma-Aldrich) to study the role of the TGF␤ pathway. Transwell coculture was performed in a transwell plate containing 24-mm inserts with the 0.4-m polyester membrane (Corning Inc.).

Establishment of GFP-labeled cell lines
To establish stable GFP-expressing MCF10DCIS cells, we transfected MCF10DCIS cells with the pcDNA3.1-GFP expression plasmid using the FuGENE HD transfection reagent (Promega, Madison, WI) and selected stable GFP transfectants using G418 (0.4 mg/ml). To establish GFP-expressing Hs578bst cells, we infected Hs578bst cells with lentiviral particles carrying the GFP expression cassette, which were prepared as described previously (60). After viral transduction, GFP-expressing Hs578bst cells were isolated by cell sorting using a FACSAria II cell sorter (BD Biosciences).

Cell sorting and FACS analysis
We isolated GFP-expressing MCF10DCIS and GFP-negative Hs578bst cells from cocultures using a FACSAria II cell sorter (BD Biosciences) for functional and gene-expression studies. FACS analysis was performed using a FACSAria II cell sorter (BD Biosciences) as described previously (61). Cells were stained with the following antibodies from Biolegend (San Diego, CA): BV421-conjugated anti-CD44, phycoerythrin-conjugated anti-CD49f, and allophycocyaninconjugated anti-EpCAM.

qRT-PCR analysis
qRT-PCR analysis of mRNA/miRNA expression was performed as described previously with normalization to ␤-actin for mRNAs or to U6 small nuclear RNA for miRNAs (61). For miR-10b-5p expression measurement, we used the miScript primers and miScript-related reagents (Qiagen, Chatsworth, CA) for RT and qRT-PCR assays. The sequence information of primers used in expression analyses of EMT-programming and miR-10b target genes can be found in supplemental Table S1.

Migration and invasion assays
Transwell-based migration and invasion assays were implemented as described previously (61). Live migrated or invaded GFP-expressing MCF10DCIS cells were directly photographed using an Eclipse Ti-E fluorescent microscope (Nikon, Melville, NY) without fixation and staining.

Stem-cell sphere formation assay
We used a modified version of stem-cell sphere formation assays (61) in the stem-cell experiments. In brief, we plated 1000 cells equivalently into 96 wells of an ultralow attachment 96-well plate (Corning). Therefore, ϳ10 cells were plated on each well for stem-cell sphere formation in the serum-free mammosphere medium containing EGF, insulin, and B27 for 7 days. This modified method would have effectively avoided cell aggregation during sphere formation. This assay generated 1-2 spheres/well in the 96-well plate. Spheres with size Ͼ 50 m were counted.

Immunofluorescent staining analysis
Immunofluorescent staining analysis of E-cadherin was performed as described previously (62).

ELISAs
We measured the concentration of TGF␤1 in conditioned media using an ELISA kit (R&D Systems, Minneapolis, MN). The ELISAs were implemented according to the manufacturer's instructions.

Western blot analysis
Protein expression was examined by Western blotting using rabbit polyclonal antibodies for detecting phospho-Smad2 (Ser-465/467) (Cell Signaling Technology, Danvers, MA), Smad2 (Cell Signaling Technology), and RB1CC1 (Protein Technologies, Tucson, AZ). Mouse monoclonal antibodies (Thermo Fisher Scientific, Waltham, MA) were used to detect E-cadherin, vimentin, and ␣-tubulin. Protein expression was detected by chemiluminescence (ECL, Amersham Biosciences). Expression of ␣-tubulin was used as a protein-loading control. Western blot data were quantified by densitometric analysis of autoradiograms, using a computerized densitometer (Typhoon System, Molecular Dynamics, Inc., Sunnyvale, CA).

Xenograft assay
For xenograft studies, 1 ϫ 10 6 MCF10DCIS cells were injected into the fourth mammary fat pad of 6 -9-week-old female nude mice alone or together with the 2-fold excess of GFP-labeled Hs578bst-GFP cells in 50% Matrigel (BD Biosciences). Tumors were allowed to grow for 4 weeks. The length and width of tumors were measured weekly with a caliper to calculate tumor volume using the formula, V ϭ 1 ⁄ 2(length ϫ width 2 ) (63). After 4-week growth, xenograft tumors were isolated and then either digested for tumor cell isolation and following RNA purification or formalin-fixed and paraffin-embedded for IHC analysis. For RNA purification, xenograft tumors were digested into single cells using collagenases, and GFP-negative tumor cells were purified and separated from GFP-positive Hs578bst myoepithelial cells by using a cell sorter. Xenograft tumor experiments were performed according to the animal protocol approved by the institutional animal care and use committee of the University of Maryland School of Medicine, which is in accordance with the guidelines established by the United States Public Health Service.

Immunohistochemistry assay
The IHC analysis was performed using the avidin biotin peroxidase complex method as described previously (64). The antibodies used in IHC experiments include anti-phospho-Smad2

The role of myoepithelial cells in DCIS
(Cell Signaling Technology), anti-GFP (Cell Signaling Technology), and anti-RB1CC1 (Protein Technologies).

PCR array-based profiling of microRNAs
miRNA expression profiling was carried out using the breast cancer miScript miRNA PCR arrays according to the manufacturer's instructions (Qiagen, Chatsworth, CA). Real-time PCR was performed using the SYBR Green PCR Master Mix (Qiagen) and the Bio-Rad CFX 1000 real-time PCR machine (Bio-Rad) according to the user manual of the miScript miRNA PCR array system. Expression values of miRNAs were calculated as described previously (61).

Biotin-conjugated miR-10b-5p pull-down assay
The RNA pull-down assay was performed on MCF10DCIS cells transfected with either biotin-conjugated scramble or miR-10b-5p RNA as described previously (40). The miRNA transfection was performed with 20 nM miRNA using the Lipofectamine RNAiMAX transfection reagent (Thermo Fisher Scientific) according to the manufacturer's instructions. Biotinconjugated scramble and miR-10b-5p duplex RNAs were obtained from Sigma-Aldrich.

Reporter plasmid DNA construction and dual reporter assay
To construct the human RB1CC1 3Ј-UTR reporter plasmid, a fragment (1.35 kb) of the 3Ј-UTR of the human RB1CC1 gene was amplified by PCR using the MCF10DCIS cDNA and primers 5Ј-AGC CGT ATC ATG GAA TAA GAA AGT A-3Ј (forward) and 5Ј-ATC ATT TGT TCA TTT ATT CAT TAA CTT GTC-3Ј (reverse). The amplified DNA was cloned into the NheI and XhoI sites of the pSGG vector. The wild-type reporter plasmid was then used as a template to generate a mutant reporter plasmid construct with mutations at the seed site recognized by miR-10b-5p using the site-directed mutagenesis system (Thermo Fisher Scientific) and primers 5Ј-ATG TTC TTT AAC TAT ATA CTA TGT AAC AGC CTA AAC AGT GTT ATG TAG AAT AGA ATT GTG-3Ј and 5Ј-CAC AAT TCT ATT CTA CAT AAC ACT GTT TAG GCT GTT ACA TAG TAT ATA GTT AAA GAA CAT-3Ј. The mutant that contains two point mutations ACAG(G to C)(G to C)t was confirmed by sequencing. The dual reporter assay that measures the enzymatic activities of firefly and Renilla luciferases expressed from the reporter DNA was performed using the Dual-Luciferase reporter assay system (Promega) according to the manufacturer's instructions.

Wound-healing assay
For the wound-healing assay, cells were seeded in 6-well plates and cultured until confluent. A 200-l pipette tip was used to make a straight scratch, simulating a wound. Cell wound images were taken by an inverted phase-contrast microscope at 0 and 16 h for examining wound healing. The percentage of wound closure was calculated as described previously (67).

In silico analysis of RB1CC1 expression in breast cancer
The Oncomine cancer microarray database (http://www. oncomine.org) 3 (42) was used to perform in silico expression analysis of RB1CC1 in breast cancer.

Statistical analysis
Statistical analysis was performed by Student's t test. p values Ͻ 0.05 were considered significant. Data (n ϭ 3) are presented as mean Ϯ S.D. Data were analyzed using GraphPad Prism software version 6.0 (GraphPad Software, Inc., La Jolla, CA).