The T Box Transcription Factor TBX2 Promotes Epithelial-Mesenchymal Transition and Invasion of Normal and Malignant Breast Epithelial Cells

The T box transcription factor TBX2, a master regulator of organogenesis, is aberrantly amplified in aggressive human epithelial cancers. While it has been shown that overexpression of TBX2 can bypass senescence, a failsafe mechanism against cancer, its potential role in tumor invasion has remained obscure. Here we demonstrate that TBX2 is a strong cell-autonomous inducer of the epithelial-mesenchymal transition (EMT), a latent morphogenetic program that is key to tumor progression from noninvasive to invasive malignant states. Ectopic expression of TBX2 in normal HC11 and MCF10A mammary epithelial cells was sufficient to induce morphological, molecular, and behavioral changes characteristic of EMT. These changes included loss of epithelial adhesion and polarity gene (E-cadherin, ß-catenin, ZO1) expression, and abnormal gain of mesenchymal markers (N-cadherin, Vimentin), as well as increased cell motility and invasion. Conversely, abrogation of endogenous TBX2 overexpression in the malignant human breast carcinoma cell lines MDA-MB-435 and MDA-MB-157 led to a restitution of epithelial characteristics with reciprocal loss of mesenchymal markers. Importantly, TBX2 inhibition abolished tumor cell invasion and the capacity to form lung metastases in a Xenograft mouse model. Meta-analysis of gene expression in over one thousand primary human breast tumors further showed that high TBX2 expression was significantly associated with reduced metastasis-free survival in patients, and with tumor subtypes enriched in EMT gene signatures, consistent with a role of TBX2 in oncogenic EMT. ChIP analysis and cell-based reporter assays further revealed that TBX2 directly represses transcription of E-cadherin, a tumor suppressor gene, whose loss is crucial for malignant tumor progression. Collectively, our results uncover an unanticipated link between TBX2 deregulation in cancer and the acquisition of EMT and invasive features of epithelial tumor cells.

TBX2 is a member of the evolutionary conserved T box transcription factor family [11,12], a class of master regulators of embryogenesis that share a T box DNA binding domain and comprise many disease genes [13]. TBX2 acts mainly as a transcriptional repressor [14,15,16] that has been shown to recruit Histone Deacetylase 1 (HDAC1) to target gene promoters [2]. During embryogenesis, TBX2 is fundamental to the regulation of cell fate decisions, cell migration, and morphogenesis in a variety of organs including the limbs, heart, kidney, nervous system, and eyes [17,18,19,20,21,22] -albeit through mechanisms that remain poorly understood. TBX2 is also prominently expressed in the embryonic mammary glands with a restricted expression in breast mesenchymal cells, which give rise to the stroma [23,24]. Although early embryonic lethality of Tbx2 knockout mice, due to severe heart defects, has precluded analysis of TBX2 function in mammopoiesis [18], adult heterozygous Tbx2 mouse mutants exhibit mild mammary gland branching defects, suggesting that TBX2 may be required for normal mammary gland morphogenesis [24].
TBX2 is also implicated in cell cycle regulation [9,25], whereby overexpression of TBX2 in different in vivo model and cell culture systems has shown to both promote [9,10,21,26], as well as attenuate cell proliferation [27,28]. Inappropriate activation of TBX2 in cancer is thought to contribute to early tumor progression by its ability to over-ride senescence and therefore maintain tumor growth [9]. Senescence is a permanent G1 growth arrest induced by DNA damage or oncogenic insult that represents a failsafe mechanism against cancer [29]. TBX2 has been shown to suppress senescence through both p53-dependent [2,9,15,30] and p53-independent mechanisms [10,26]. Consequently, TBX2 can cooperate with transforming oncogenes (c-Myc, Ras) or the loss of tumor suppressor genes (p53, Rb) in cellular transformation [9,31,32]. Moreover, overexpression of TBX2 in human lung and skin cancer models, although inhibitory to cell growth, has been shown to promote the resistance of tumor cells to the anti-cancer drug cisplatin [28]. Whilst the anti-senescence activity of TBX2 has been extensively studied, it has remained unclear whether TBX2 can also contribute to tumor invasion, as the clinical association of TBX2 gene amplification with invasive epidermal tumors would suggest.
There is increasing evidence that aberrant activation of the embryonic morphogenetic program, termed the epithelial-mesenchymal transition (EMT), is crucially involved in tumor cell invasion [33]. During EMT, adherent epithelial cells lose polarity, undergo a major reorganization of the cytoskeleton and acquire a fibroblastic (mesenchymal), highly motile phenotype [34]. EMT is potently activated by TGFß [35], and at the transcriptional level by a growing list of embryonic transcription factors (TFs) [33]. These include the E-cadherin repressing Zinc finger proteins Snail (SNAI1) and Slug (SNAI2) [36,37], ZEB1 (hEF1) and ZEB2 (SIP1) [38,39]; the basic helix-loop-helix proteins TWIST1/2 [40,41,42]; the homeodomain proteins Goosecoid, LBX1 and SIX1 [43,44,45]; and the winged-forkhead transcription factor FOXC2 [46]. Virtually all of these TFs have also been implicated as drivers of oncogenic EMT and breast cancer metastasis [47,48,49,50]. EMT is increasingly viewed as a significant clinical problem in cancer, as EMT is thought to promote an aggressive cancer stem cell phenotype [51,52], therapy resistance [53], and tumor recurrence [54], thereby contributing to poor disease outcome. Thus, there is an urgent need for the identification and characterization of the genes involved in this process.
Through ectopic expression of TBX2 in normal mammary epithelial cells and RNAi-mediated silencing of endogenous TBX2 overexpression in malignant human breast carcinoma cell lines, we demonstrate that TBX2 acts as a strong cell-autonomous inducer of EMT. We found that TBX2 directly represses E-cadherin transcription and promotes malignant tumor progression by imparting an aggressive mesenchymal tumor phenotype. These findings, together with a significant correlation between high TBX2 expression levels in primary tumors and reduced metastasisfree survival of breast cancer patients, suggest that TBX2 may be an attractive new target for anti-metastatic cancer therapies.

TBX2 Efficiently Induces EMT in Mammary Epithelial Cells
We were intrigued by previous observations that during mouse embryonic development, Tbx2 is exclusively expressed in mesenchymal cells surrounding the mammary epithelial anlagen [23,24], suggesting it may regulate mesenchymal cell specification in the breast. Yet, studies examining TBX2 expression in a small number of human breast cancers have reported TBX2 mRNA and protein overexpression primarily in the epithelial compartment of tumors with little or no expression in stromal cells [7,55]. We therefore hypothesized that the apparent misexpression of TBX2 in breast epithelial cells during carcinogenesis may confer mesenchymal properties to these cells. To test this hypothesis, TBX2 was stably introduced into murine HC11 and human MCF10A cells, two spontaneously immortalized but otherwise normal mammary epithelial cell lines, which we found lack endogenous TBX2 expression ( Figure 1B, 1D and Figure S3A, S3B). To avoid clonal selection bias, several individual polyclonal cell cultures expressing pCDNA3-TBX2 plasmid (+TBX2) or pCDNA3 vector (+vector) alone were established. Each of the TBX2-expressing HC11 and MCF10A cell derivatives (n = 3 per line) showed a dramatic change in cell morphology from the earliest passages onwards ( Figure 1A, 1C). While vector-transfected HC11 or MCF10A cells had a typical epithelial cell structure, HC11+TBX2 cells were abnormally enlarged and stretched out with lamelopodia-like migratory protrusions ( Figure 1A). In addition, MCF10A+TBX2 cells clearly displayed a spindle-shaped, fibroblastoid, and scattered morphology ( Figure 1C). Analysis of EMT marker protein expression by Western Blot and immunofluorescence revealed that the TBX2-induced morphologic changes were due to EMT ( Figure 1B, 1D, and 1E). Protein levels of the epithelial adherence junction proteins E-cadherin and ß-catenin were decreased in HC11+TBX2 and MCF10A+TBX2 cells, whereas mesenchymal markers, Vimentin and N-cadherin, were markedly upregulated as compared to the respective vector control cells ( Figure 1B, 1D, and 1E). Furthermore, whereas in confluent HC11+vector control cells, E-cadherin, ß-catenin, and the tight junction protein ZO1 were detected primarily at cell-to-cell junctions, TBX2-expressing HC11 cells at the same cell density exhibited a reduced and disrupted immunostaining for these epithelial cell adhesion molecules at the cell periphery ( Figure 1E). A breakdown of epithelial adhesion complexes was further evident by a significant reduction in mRNA levels for E-cadherin, ß-catenin, ZO1, and the desmosomal component Desmoplakin by 50-70% respectively in TBX2-overexpressing HC11 cells as determined by quantitative realtime PCR (qPCR) ( Figure 1F). Of mesenchymal markers analyzed, N-cadherin, and the extracellular matrix metalloproteinase Mmp3 were most significantly upregulated in these cells ( Figure 1F). A similar switch from epithelial to mesenchymal marker gene expression was also evident in MCF10A+TBX2 cells (Figure S1A, S1B). Thus, TBX2 efficiently induced morphologic and molecular changes characteristic of EMT in mammary epithelial cells.

TBX2 Promotes Mammary Epithelial Cell Motility and Invasiveness
We next tested whether ectopic expression of TBX2 promotes any behavioral changes associated with EMT, such as increased motility and gain of invasiveness. Parental and vector-expressing HC11 or MCF10A cells have a low propensity to migrate and invade extracellular matrix ( Figure 2). However, both TBX2overexpressing HC11 and MCF10A cells exhibited a significant increase in cell motility in ''in vitro scratch'' assays, which was visible as early as 4-8 hours after an experimentally induced wound and became more pronounced between 8-32 hours, leading to a complete wound closure by MCF10A+TBX2 cells at 24 hours (Figure 2A, 2B). The increased movement of TBX2expressing mammary epithelial cells was not due to increased proliferation since the assay was done in low serum-containing medium (see Methods), in which these cells were growth-retarded (data not shown). To examine the role of TBX2 in the regulation of cell invasiveness, we performed Transwell matrigel invasion assays ( Figure 2C, 2D). Both TBX2-expressing HC11 and MCF10A cells, showed an approximately four-fold increase in their abilities to invade through the matrigel layer towards serumcontaining media ( Figure 2C, 2D). Thus, TBX2 plays a central role in the acquisition of cell motility and invasiveness of breast epithelial cells through induction of EMT.

TBX2 is Induced during TGFß-mediated EMT of Primary Human Mammary Epithelial Cells
We further examined expression of endogenous TBX2 in a cellular model of inducible EMT. We have previously shown that treatment of primary finite-lifespan human mammary epithelial cells (HMEC) with the cytokine TGFß efficiently induces EMT, as well as a hierarchy of known EMT-associated transcription factors [56]. While TBX2 was not expressed in untreated HMEC ( Figure 3A and 3C), both TBX2 mRNA and protein were upregulated upon EMT induction by TGFß coinciding with a decrease in epithelial and acquisition of mesenchymal marker expression ( Figure 3A-3C). A significant induction of TBX2 mRNA expression occurred as early as 6 hours (,1.8 fold; p,0.001) upon TGFß stimulation and further increased to 3.3-4.3 fold by 9-12 days (p,0.001) ( Figure 3B). Notably, the rapid increase of TBX2 mRNA levels, accompanied by a reduction in epithelial E-cadherin and an increase in mesenchymal Vimentin mRNA expression ( Figure 3B), occurred well before any morphological changes of EMT became visible at 3 days of TGFß treatment [56]. Moreover, TBX2 protein specifically localized to the nucleus of TGFß-treated HMEC ( Figure 3A), where it is thought to be active as a transcription factor. These data indicate that TBX2 is also part of the endogenous EMT program of primary HMEC and further implicate TBX2 in EMT induction of breast epithelial cells.

TBX2 Expression in Human Breast Tumors Correlates with EMT Features and Increased Disease Recurrence
EMT and increased invasiveness are key features of epithelial tumor cells as they progress into malignant metastatic cancer cells [33]. We therefore asked whether the EMT-inducing and proinvasive abilities of TBX2 that we observed in normal breast epithelial cells, could also play a role in human breast cancer.
Since little is known about TBX2 expression in human breast tumors, we performed a comprehensive meta-analysis of TBX2 expression using an integrated gene expression database that encompassed 1107 primary human breast tumors from six published datasets [57]. We found that TBX2 was variably expressed across the different molecular subtypes of breast cancer [58] ( Figure S2A), largely independent of estrogen receptor (ER) status and tumor grade ( Figure S2B, S2C), with the highest expression levels among the rare, aggressive 'claudin-low' subtype of breast cancer and lowest in basal tumors ( Figure S2A). Interestingly, the 'claudin-low' subtype of breast tumors has recently been shown to be enriched for EMT features [59,60]. Another dataset showed that TBX2 was higher in poor prognosis metaplastic breast cancers ( Figure S2D), which like the 'claudinlow' group exhibit an EMT gene signature and are highly metastatic [60,61]. Expression of TBX2 was not significantly higher in the 'claudin-low' tumors of this dataset, but the numbers of tumors (n = 13) were much lower than in the in the 'six study' analysis (n = 34) [57].
A similar tumor subtype distribution of TBX2 overexpression was observed in a panel of a total of 20 human breast carcinoma cell lines using Western blot and qPCR analyses ( Figures S3 and  S4). We confirmed endogenous TBX2 overexpression in three ER-positive luminal tumor lines (MCF7, MDA-MB361, and BT-474) ( Figure S3A-C) that have previously been shown to harbor TBX2 gene amplifications [6,9,10]. Moreover, the ER-negative basal breast carcinoma cell line SUM52 [62] displayed abundant TBX2 protein and mRNA expression, as well as modest gene amplification levels ( Figure S3A-3C), consistent with previous Fluorescence In Situ Hybridization (FISH) data [6]. In addition, we identified TBX2 overexpression in two metaplastic tumor-derived breast carcinoma cell lines, MDA-MB-435 and MDA-MB-157 [63,64], which we found to express higher median levels of TBX2 than most other breast tumors ( Figure S2D). Both of these cell lines exhibit mesenchymal gene signatures [64], and have an increased invasive, metastatic potential [63,65]. Comparative genomic hybridization array (aCGH) analysis showed no significant increases in TBX2 gene copy numbers in MDA-MB-435 and MDA-MB-157 ( Figure S3C), suggesting that overexpression of TBX2 in these tumor cell lines is not due to gene amplification. TBX2 gene amplification has previously been reported in aggressive BRCA1-related breast cancers [6,7]. However none of the four basal-subtype ER-negative BRCA1-deficient breast carcinoma cell lines (HCC1937, MDA-MB-436, SUM149, and SUM1315) we studied expressed TBX2 protein at detectable levels or exhibited TBX2 gene amplification ( Figure S4A, S4B). Furthermore, TBX2 was not expressed in any normal-derived human breast epithelial cell line ( Figures S3 and 1B, 1D).
The prognostic significance of TBX2 in human breast cancer was examined next by meta-analysis. High TBX2 transcript levels were found to be significantly associated with a shorter time to recurrence-free survival ( Figure 4A, 4B). This was more significant for ER-positive tumors and similar results were seen in the completely independent 295-sample NKI dataset [66] ( Figure 4A, 4B). Overall, these results are compatible with the notion that TBX2 is activated in certain primary breast cancers correlating with an EMT signature and reduced metastasis-free survival.

Silencing of TBX2 Leads to Mesenchymal-epithelialtransition and Impedes the Invasiveness of Human Breast Cancer Cells
To elucidate the potential role of TBX2 in malignant tumor progression, we employed RNA interference strategies to inhibit TBX2 in the metastatic breast carcinoma cell lines MDA-MB-435  . TBX2 overexpression in primary human breast tumors is correlated with reduced metastasis-free survival. Kaplan Meier analysis demonstrates that TBX2 mRNA overexpression is associated with shortened recurrence-free survival (A) in a meta-analysis of six combined published microarray datasets comprising 1107 primary human breast tumors [57], and (B) in an NKI study of 295 women with early-stage invasive breast carcinomas [66]. In both of these datasets, the optimal cut-point value of TBX2 expression was used to divide the samples into high (above median; green) and low (below median; blue) TBX2 expression. A batch correction was performed on the six-study set (see Methods). High TBX2 expression is particularly associated with poor prognosis in estrogen receptor-positive (ER+) breast tumors (right panels). ER status was determined by immunohistochemistry (IHC). P values are indicated. doi:10.1371/journal.pone.0041355.g004 of epithelial E-cadherin and ZO1 was enhanced, whereas transcription of mesenchymal genes (N-cadherin, Vimentin, Fibronectin, MMP3) was reduced in TBX2-depleted MDA-MB-435 and MDA-MB-157 breast cancer cells ( Figure 5D, 5E). Immunofluorescence analysis further confirmed that TBX2 inhibition resulted in the reexpression of E-cadherin protein in MDA-MB-435 cells, which are normally devoid of this epithelial marker ( Figure 5F) [35]. Moreover, epithelial ß-catenin and ZO1 not only were increased in levels, but also properly localized to the cell membrane in TBX2-depleted MDA-MB-435 tumor cells ( Figure 5F). In contrast, mesenchymal markers (N-cadherin, Vimentin) were drastically reduced and, for N-cadherin, mislocalized to the cytoplasm in MDA-MB-435-TBX2 knockdown cells ( Figure 5F).
Inhibition of TBX2 further led to reduced tumor cell migration in both MDA-MB-435 and MDA-MB-157 breast carcinoma cell lines ( Figure 6A, 6B). Moreover, tumor cell invasion rates in Transwell matrigel invasion assays were markedly decreased ( Figure 6D, 6E). Additionally, whereas control non-target shRNA expressing MDA-MB-435 tumor cells formed spheroids with extensive protrusions in three-dimensional (3D) Matrigel cultures, reflective of their invasive nature, knockdown of TBX2 resulted in the formation of round, non-invasive spheres ( Figure 6C). Thus, loss of TBX2 in malignant breast carcinoma cells abolished tumor cell invasion and lead to the restitution of a more differentiated epithelial phenotype. All experiments were done with at least two independent cell clones from each stably shRNA transfected cell line and three independent polyclonal cultures of cell lines transiently transfected with siRNAs.  Figure 7C). Moreover, mRNA expression of the cell cycle inhibitor p21 CIP1/WAF1 , which is a direct transcriptional target of TBX2 in senescence bypass [2,30], was not significantly altered in TBX2 knockdown cells ( Figure 7D), nor did these cells exhibit any signs of senescence (data not shown). These data suggest that TBX2 is crucial for the formation of metastases independent of effects on tumor cell growth. Taken together, our TBX2 inhibition studies in malignant breast carcinoma cell lines reinforce the notion that TBX2 promotes malignant tumor progression by imparting a highly invasive mesenchymal phenotype on breast epithelial tumor cells.

TBX2 Represses E-cadherin Transcription
Loss of E-cadherin is an important hallmark of EMT, directly contributing to transformation and metastatic tumor progression [67,68]. Studies in melanoma cells have raised the possibility that TBX2 may be implicated in E-cadherin regulation, but failed to detect a requirement of TBX2 for E-cadherin expression in vivo [69]. Since we found overexpression of TBX2 to consistently reduce endogenous E-cadherin levels in normal mammary epithelial cells, and, conversely, inhibition of TBX2 to lead to enhanced E-cadherin mRNA expression in metastatic breast cancer cell lines (Table 1, Figures 1F and 5D, 5E), we revisited the question whether TBX2 could directly repress E-cadherin at the promoter level.
Chromatin immunoprecipitation (ChIP) analysis was performed to determine direct in vivo binding of TBX2 to the endogenous Ecadherin gene in mammary epithelial cells ( Figure 8A, 8B). We used three different primer sets; one covering the most proximal promoter region of E-cadherin, including the initiator element (InR: TGGTGT in mouse and AGTGGC in human at +1 to +6 each), which has previously been shown to be bound by a recombinant TBX2 DNA binding domain in vitro [69]; one set spanning a conserved half T-box factor binding site (Half T-site: AGGTGTTA at 2682 in mouse and TCACACCT at 2645 in human) [69]; and one primer set in a distal region (21299/ 21119) devoid of potential TBX2 binding sites ( Figure 8A). PCR amplification of these genomic sequences using immunoprecipitated chromatin from HC11+vector or HC11+TBX2 cells demonstrated that TBX2 specifically bound to the proximal Ecadherin promoter ( Figure 8B), which contains the InR element ( Figure 8A) that can serve as a putative TBX2-binding site [15,69].
To test the functional significance of in vivo TBX2 occupancy of this E-cadherin promoter region, we transiently co-transfected HC11 cells with a luciferase reporter construct (pEcad-luc) containing the proximal human E-cadherin promoter (2108 to +125) [70] and with increasing concentrations of pCDNA3-TBX2 expression plasmid ( Figure 8C). As compared to empty vector control (basal), co-expression of TBX2 led to a significant (p,0.05) 3.7 fold reduction in the activity of this promoter in a dosedependent manner. Together, these results indicated that TBX2 directly represses transcription of E-cadherin by binding to its proximal promoter in vivo.

Discussion
In this study, we have identified the embryonic transcriptional repressor and anti-senescence factor TBX2 as a novel potent inducer of EMT that directly represses E-cadherin transcription and promotes an aggressive, mesenchymal breast tumor phenotype ( Figure 9). Since TBX2 is aberrantly amplified with high prevalence in a number of aggressive human epidermal cancers, or, as we found, can be induced by TGFß (Figure 9), a promoter of metastatic tumor progression [35], these findings are of potential high clinical relevance. Unlike other EMT-inducing TFs [33], TBX2 has not previously been implicated in the cell-autonomous regulation of EMT induction during embryogenesis. Our results therefore uncover a novel paradigm of TBX2 function that may also be relevant for its role in normal development.
During embryogenesis, TBX2 is expressed after gastrulation [11,23], and among other roles, controls mesenchymal cell specification in the limb buds underlying posterior digit identity [17], specification and morphogenesis of mesoderm-derived cardiomyocytes in the valve forming regions of the heart [18,71], pro-mesonephros identity during kidney development [19], and cell migration processes during brain and eye morphogenesis [20,22]. However, the consequences of inappropriate gain-of-function of this morphogenetic TF in epithelial tumor cells during carcinogenesis have been poorly explored. By modeling aberrant gain-of-function of TBX2 in breast cancer through ectopic expression of TBX2 in non-malignant mammary epithelial cell lines (HC11, MCF10A), we demonstrated that TBX2 alone is sufficient to induce EMT and cell invasion.
Since EMT has been implicated in tumor recurrence [53,54], and we found high levels of TBX2 expression to be significantly correlated with disease recurrence in a meta-analysis of over 1,000 primary human breast tumors (Figure 4), we further investigated whether the EMT-inducing ability of TBX2 could play a role in malignant tumor progression. Congruent with our TBX2 overexpression results in normal breast epithelial cells, RNAi-mediated silencing of TBX2 in two aggressive mesenchymal human breast carcinoma cell lines with endogenous TBX2 overexpression (MDA-MB-435, MDA-MB-157) led to the restitution of a differentiated epithelial tumor phenotype. This was most evident by reexpression of E-cadherin, ß-catenin, and ZO1, and the concomitant loss of mesenchymal N-cadherin, Vimentin, and Fibronectin expression ( Figure 5). Importantly, we found that inhibition of TBX2 also abolished tumor cell migration, invasion and profoundly diminished the capacity of MDA-MB-435 cancer cells to form pulmonary metastases in a xenograft in vivo mouse model. These effects appeared to be independent of the anti-senescence activity of TBX2, as cell proliferation and expression of the growth control gene p21 CIP1/WAF1 were unchanged in MDA-MB-435-shTBX2 cells (Figure 7). The observed pro-invasive activity of TBX2 was likely due to specific induction of MMP3, which unlike other matrix metalloproteases (MMP2, MMP9) was most consistently upregulated by TBX2 in our cellular systems (Figures 1, 4, S1; data not shown). Together, these results suggest that TBX2 is strongly implicated in malignant tumor progression by promoting an aggressive mesenchymal tumor phenotype.
Interestingly, we found that TBX2 is induced by the EMTpromoting cytokine TGFß, which is often excessively produced by invasive breast cancer cells and has been associated with metastatic tumor progression [35]. Because of the rapid kinetics of TBX2 induction in HMEC upon TGFß treatment, it is tempting to  speculate that TGFß signaling controls TBX2 expression at the transcriptional level. This notion is further supported by studies demonstrating that BMPs, which are TGFß homologues, positively regulate Tbx2 expression during cardiogenesis via functional SMAD binding sites in the Tbx2 gene promoter [72]. Thus, apart from chromosome 17q23 amplification, TBX2 induction by TGFß may represent a novel mechanism underlying the aberrant overexpression of TBX2 in invasive cancers (Figure 9).
Through ChIP analysis and cell-based promoter-reporter assays, we further demonstrate that TBX2 binds directly to and represses the activity of the proximal E-cadherin promoter, indicating that E-cadherin is a direct TBX2 target gene. Given that Ecadherin acts as a tumor suppressor, whose loss is causally implicated in EMT and metastatic tumor progression [67,68], transcriptional repression of E-cadherin by TBX2 may provide a possible mechanistic explanation for the observed EMTinducing and pro-metastatic activities of TBX2 in breast cancer cells. We note that there was little correlation between TBX2 expression and E-cadherin status in established breast cancer cell lines ( Figure S3) [90], a finding, which has also been reported for the EMT-inducing transcription factor LBX1 [44]. This could suggest that efficient repression of E-cadherin by TBX2 might require cooperation with other transcriptional repressors that may not be present in all TBX2-overexpressing breast tumor cell lines. Moreover, established tumor cell lines do not necessarily reflect the situation in primary breast tumors, in which EMT is histopathologically difficult to detect due to its transient nature [73]. In discrepancy with our results, previous studies have indicated that TBX2 knockdown in MCF7 breast carcinoma and human melanoma cell models did not diminish tumor cell invasion [69,74] or repress endogenous E-cadherin expression [69]. However, TBX3, a close TBX2 homolog with similar oncogenic functions [75] not yet implicated in EMT regulation, exerted these effects in both of those systems [69,74]. A possible explanation for these conflicting results could be that invasion of the inherently low-invasive MCF7 cell line in the previous study was induced by phorbol esters [74], which might have altered  Alternatively, there could be cell-type specific differences in the activities of TBX2 and TBX3 that may depend on phosphorylation status or the differential availability of protein partners. TBX2 has been shown to be phosphorylated by external stimuli that alters its cellular localization and, thereby, its activity as a transcription factor [12]. Furthermore, senescence suppression by TBX2 in human cells depends on the physical interaction of TBX2 with tumor suppressive TFs, for example EGR1 in breast cancer cells [10], and PML in fibroblasts [26]. Interestingly, recent studies have indicated that transcriptional repression of the senescence-associated tumor suppressor p14ARF by TBX2 [9,15], requires the association of TBX2 with NRAGE, a protein that is released from the cell membrane upon EMT induction [76]. To rule out the possibility that TBX3, which is primarily overexpressed in luminal E-cadherin-positive breast cancer lines and estrogen receptor-positive breast tumors [77,78], contributed to the pro-invasive effects elicited by TBX2, we investigated TBX3 expression in our TBX2-dependent breast epithelial model systems ( Figure S5). We found a reciprocal expression of these T-box factors, with TBX3 downregulated in TBX2-expressing HC11 cells but upregulated in TBX2-depleted HC11+TBX2 and MDA-MB-435 tumor cells ( Figure S5A, S5B), which is reminiscent of the mutually exclusive expression patterns of TBX2/3 in normal mammary gland tissues [23,79], Thus, the TBX2-induced EMT phenotypes were not due to a possible interference by TBX3 but rather may reflect the poorly explored role of TBX2 as a mesenchymal and baso-myoepithelial transcription factor in breast development [23,24]. In contrast, TBX3 functions as a master regulator of mammary epithelial cell fate [80,81], and in the adult breast is specifically expressed in differentiated luminal breast epithelial cells [79]. Significantly, a role of TBX2 in oncogenic EMT and malignant breast cancer progression is further suggested by our finding that TBX2 mRNA levels in clinical human breast cancer specimen were highest in rare EMT-enriched tumors of the 'claudin-low' and metaplastic breast tumor subtypes. These tumors represent one of the most aggressive and treatment-resistant forms of breast cancer [59,60,61]. Moreover, we found high TBX2 transcript levels to be significantly associated with reduced metastasis-free survival of breast cancer patients, which is in keeping with the correlation of TBX2 gene amplification data with poor clinical outcome [6,8]. Since tumor tissues used for expression profiling are subjected to histology and only included if they contain a reasonable percentage of tumor cells it is unlikely that the observed correlations are due to expression of TBX2 in tumor associated stroma. It was perhaps surprising that TBX2 expression was predictive of poor prognosis but independent of ER status and that basal subtype and high-grade breast tumors had slightly lower average levels of TBX2 expression ( Figure S2). However, these findings are consistent with TBX2 having a similar pattern of gene expression across subtypes as other EMT-inducing TFs, e.g. Twist, ZEB1, ZEB2, and SNAI2 (Slug) [60]. Furthermore, recent clinical population studies have shown that even breast tumors in the lowest risk category (ER+, early stage, small tumor size, nodenegativity) and despite adjuvant treatment can have relatively high relapse rates [82,83]. Thus, TBX2 may prove to have a unique value as a novel prognostic marker.
Collectively, our work suggests that TBX2 is a key driver of malignant tumor progression through induction of EMT and tumor cell invasiveness. Although previous mouse developmental genetic studies have indicated that TBX2 can indirectly promote EMT of endocardial cells during cardiac valve formation through induction of paracrine TGFß2 signaling in surrounding valveforming myocardium [72], our work is the first to demonstrate that TBX2 can also activate EMT in a cell-autonomous manner. Further experiments are under way to identify the signaling mechanisms, potential interacting partners, and target genes of TBX2 in EMT induction and epithelial tumor invasion. Finally, our discovery that TBX2, an established anti-senescence factor, is a strong inducer of EMT lends further support to the notion that EMT and senescence bypass may rely on some of the same molecular mechanisms [41,42]. We anticipate our studies to be a starting point for evaluating TBX2 as a new marker for breast cancer diagnosis and potential target for anti-metastatic cancer drug development.

Ethics Statement
All experiments including the use of mice were approved by the University of Miami IACUC (protocol number 10-226, Institutional assurance number for the University of Miami A-3224-01). For the studies we describe, there are no suitable alternative approaches, and care was taken to minimize animal distress.

Cell Lines
The HC11 mouse mammary epithelial cell line [84] (kindly provided by Dr. Kermit Carraway, University of Miami) was grown in complete growth medium (RPMI containing 10% FBS, 1% penicillin-streptomycin, 1 mg/ml EGF (Invitrogen) and 5 mg/ ml insulin (Sigma-Aldrich). Human MCF10A mammary epithelial cells were obtained from the American Type Cell Collection (ATCC) and grown in HuMEC medium (Invitrogen). Linearized pCDNA3 or pCDNA3-TBX2 expression plasmids [85] (kindly provided by Dr. Roni Bollag, Medical College of Georgia, Augusta, GA, USA) were introduced into cells by Lipofectamin 2000 transfection (Invitrogen) and stable transfectants were selected in 200-300 mg/ml G418 for 9-11 days. Primary HMEC were from Cambrex Bio Science and cultured in HuMEC medium. TGFß induction experiments using 5 ng/ml of recombinant TGFß1 (R&D Systems) were performed as described [56]. All human breast carcinoma cell lines, except the SUM lines (Asterand), were from ATCC and cultured according to the distributors' recommendations. Specifically, MDA-MB-435 and MDA-MB-157 cells were grown in DMEM plus 10% FBS and 1% Penicillin-Streptomycin, hereafter referred to as complete growth medium. All cell lines were maintained in a 5% CO 2humidified incubator at 37uC.

Immunofluorescence
Fluorescence immunocytochemistry on cultured cells was performed as previously described [56]. Quantitative Real-Time PCR (qPCR) qPCR analysis was performed as previously described [56,86] using SYBR Green PCR Master Mix (New England Biolabs) and a CFX96 Real Time PCR thermocycler (Biorad). Sequences of qPCR primers are shown in Table S1. Triplicate samples were performed and average C t values were normalized to the values of GAPDH.

In vitro Scratch Assay
Scratch assays were performed as described [87] with the following modifications. 1610 6 cells were plated on a 60 mm dish and cultured in complete growth media 24 h before the assay. When cells reached confluence, a p200 pipette tip was used to scrape a straight line through the cell monolayer. Cells were washed once with growth medium to remove cell debris and to smoothen the edge of the scratch. The culture medium was then replaced with growth medium with a lower FBS content (5% FBS) to minimize cell proliferation during the duration of the assay. Images were taken at different time points after scratch on a Leica DMIL inverted microscope using a Q-Imaging digital camera (Micropublisher) and analyzed using ImageJ software (NCBI). Acquisition of same field images was achieved by making reference points using an ultra fine tip marker.

Transwell Migration and Matrigel Invasion Assays
Cells (1-2610 4 ) were resuspended in 100 ml of serum-free growth medium and plated into uncoated or Matrigel-coated 8mm transwell filter inserts (Corning) of 24-well plates in triplicates. For invasion assays, filters were pre-coated with 10 ml of matrigel (BD) diluted 1:4 in ice-cold serum-free medium and allowed to solidify for 1 h at 37uC before use. The bottom wells contained 500 ml of growth medium with 10% FBS as chemoattractant. After incubation of cells for 16 h for migration or for 48 h for invasion assays, cells on the upper surface of the filter were removed with a cotton swab and cells on the bottom side were fixed in 100% methanol and stained with 1% Toluidine Blue in 1% Borax. Cells were counted in three random fields on microscopic images taken at 10x or 40x magnification.

3-D Matrigel Assays
Single cell suspensions of 2610 4 MDA-MB-435 cells in 100 ml of complete growth media containing ice cold Matrigel (BD) (1:1) were plated in triplicates on a 96-well plate. Plates were incubated at 5% CO 2 , 37uC for 30 min to allow the matrigel to solidify, after which 100 ml of complete media was added to each well. The culture media was changed every 3 days. Ten to 14 days after plating, pictures were taken under bright field at 20X magnification using a Leica DMIL inverted microscope.

Tail Vein Metastasis Assays
Six week-old Nude (nu/nu) mice (Charles River Laboratories) were inoculated with 1610 6 MDA-MB-435 tumor cells/mouse (in 150 ml of PBS) via tail vein injection. Forty days after tumor cell inoculation, animals were euthanized, and lungs were inflated with India ink, as described [88]. Surface lung nodules and micrometastases were scored in a genotype-blinded fashion using a Leica MZ16 stereomicroscope. Lungs were then paraffin-embedded, and 5 mM sections were stained with hematoxylin-eosin.

Cell Proliferation Assays
MDA-MBA-435 cells (1.5610 5 ) were seeded in triplicates on 12 well plates on day 0. Cells were grown under subconfluent conditions in complete growth medium containing 1 mg/ml puromycin and split at a ratio of 1:3 every 3 days. Cell numbers were counted at the time of passaging over a total period of 9 days.

Chromatin Immunoprecipitation (ChIP)
Cells were grown to 80% confluence and cross-linked with 1% formaldehyde at RT for 10 min. ChIP assays were performed as in [86], except that sonication of cell lysates was performed for 15 pulses of 10 sec with 1 min interval each on ice at 20% power on a Misonix sonicator. Sheared chromatin was immunoprecipitated with 5 mg of antibodies to TBX2 (AB4147; Upstate-Millipore and sc-17880x; Santa Cruz), anti-acetyl Histone 3, or normal rabbit IgG (Upstate-Millipore). PCR primers for amplification of different regions of the mouse E-cadherin/Cdh1 promoter are listed in Table S1.

Luciferase Reporter Assays
HC11 were seeded in 12-well tissue culture plates at a density of 2610 5 the day prior to transfection. Cells were cotransfected using Lipofectamine 2000 (Invitrogen) with 500 ng of pCDNA3 or pCDNA3-TBX2 expression vectors, 500 ng of pGL2Basic-EcadK1 luciferase reporter construct (containing wild type human E-Cadherin/CDK1 promoter sequences from 2108 to +125/ Addgene) [70], and 25 ng of pRL-CMV Renilla plasmid (Promega), which served as normalization control. Forty-eight hours after transfection, cells were harvested and subjected to a Promega Dual Luciferase assay using a Veritas Luminometer.

Analysis of Published Gene Expression Datasets
Microarray data representing a total of 1107 primary breast tumors from six previously published Affymetrix studies were downloaded from repositories (E-TABM-158, GSE7390, GSE4922, GSE1456, GSE2990, GSE2034) and integrated as described previously using a mean-batch centering method [57]. The NKI [66] and [61] datasets were retrieved from http:// microarray-pubs.stanford.edu/would_NKI/explore.html and NCBI GEO (GSE10885). The x-tile method was used to determine the optimal cut-point in Kaplan Meier analysis while correcting for the use of minimum P statistics [89].

Statistical Analysis
All other data represented in graphs were analyzed by two-sided Student's t or ANOVA tests using GraphPad Prism software. Pvalues of ,0.05 were considered significant. Note that SUM52 is listed as luminal ER+ in the Neve et al. dataset [90] but has been re-classified as basal-subtype triplenegative [62]. Furthermore, TBX2 is overexpressed in the highly invasive basal subgroup metaplastic breast tumor cell lines MDA-MB-157, MDA-MB-435, and weakly in basal subgroup medullary (asterix) HCC1569 tumor cells. (B) qPCR analysis quantifies TBX2 mRNA expression levels in the tumor cell lines shown in (A). Values were normalized to GAPDH mRNA levels and represent fold change as compared to normal human mammary epithelial cells (HMEC). Error bars represent the mean 6 SEM (n = 3; Student t-test). (C) Comparative genomic hybridization array (aCGH) analysis shows relative gains and losses of the chromosomal region of TBX2 (17q23) in the selected breast cancer cell lines from two published aCGH studies. Dark grey = [90]; light grey = [91]. Not all cell lines were represented in both studies, however the relative gains/losses for the TBX2 region between the two studies was significantly correlated (Pearson, R = 0.6, p = 0.001) across the overlapping breast tumor cell lines.