FOXF2 deficiency promotes epithelial-mesenchymal transition and metastasis of basal-like breast cancer

Introduction Our previous clinical study demonstrated that the under-expression of FOXF2 is associated with early-onset metastasis and poor prognosis of patients with triple-negative breast cancer. In this study, we further characterized the role of FOXF2 in metastasis of basal-like breast cancer (BLBC) and underlying molecular mechanisms. Methods RT-qPCR, immunoblot, immunofluorescence and immunohistochemistry were performed to assess the expression of genes and proteins in cell lines and tissues. A series of in vitro and in vivo assays was performed in the cells with RNAi-mediated knockdown or overexpression to elucidate the function and transcriptional regulatory role of FOXF2 in breast cancer. Results We found that FOXF2 was specifically expressed in most basal-like breast cells. FOXF2 deficiency enhanced the metastatic ability of BLBC cells in vitro and in vivo. Additionally, FOXF2 deficiency induced the epithelial-mesenchymal transition (EMT) of basal-like breast cells. Furthermore, we identified that TWIST1 is a transcriptional target of FOXF2. TWIST1 was negatively regulated by FOXF2 and mediated the FOXF2-regulated EMT phenotype of basal-like breast cells and aggressive property of BLBC. Conclusions FOXF2 is a novel EMT-suppressing transcription factor in BLBC. FOXF2 deficiency enhances metastatic ability of BLBC cells by activating the EMT program through upregulating the transcription of TWIST1. Electronic supplementary material The online version of this article (doi:10.1186/s13058-015-0531-1) contains supplementary material, which is available to authorized users.


Introduction
Breast cancer is the most frequently diagnosed cancer among females [1]. Metastasis is the leading cause of death in breast cancer patients. To develop distant metastatic disease, the epithelial cancer cells need to undergo a morphological change into a mesenchymal phenotype to acquire migration and invasion ability [2]. Epithelialmesenchymal transition (EMT) is a program of switching polarized epithelial cells to migratory mesenchymal cells [2]. The aberrant activation of the EMT program is implicated in tumor progression, metastasis and acquisition of therapeutic resistance [2,3]. EMT-transcription factors (EMT-TFs) such as TWIST1, SNAIL1, SNAIL2, ZEB1, ZEB2, Goosecoid and FOXC2, can trigger the EMT program by directly or indirectly suppressing Ecadherin expression [4][5][6][7]. TWIST1 is a member of the basic helix-loop-helix (bHLH) transcription factor family [8] and plays a pivotal role in the regulation of embryogenesis, gastrulation and mesoderm formation during early embryonic development [9,10]. TWIST1 was found to be frequently activated in a wide array of human cancers and is associated with poor prognosis [3,11]. TWIST1 induces the EMT program by downregulating E-cadherin expression through indirect effects on the CDH1 promoter [3].
Breast cancer is a heterogeneous disease. Based on their gene expression profiles, breast cancers can be classified into distinct molecular subtypes: normal breast-like, luminal A, luminal B, human epidermal growth factor receptor 2 (HER2)-enriched, and basal-like [12]. Basal-like breast cancer (BLBC) is less likely to express estrogen receptor (ER), progesterone receptor (PR) and HER2, which are also characteristics of triple-negative breast cancer (TNBC). Thus, BLBC shares many features with TNBC, and the two terms are often used interchangeably [13]. In addition to the triple-negative marker status, BLBC is characterized by the expression of basal markers such as cytokeratins (CK) 5/6, CK14, CK17 and epithelial growth factor receptor (EGFR) in the clinic [12]. TNBC and/or BLBC is recognized as a particularly aggressive subtype and receives less benefit from targeted therapy [12]. Therefore, there is an urgent need to elucidate the molecular pathogenesis of TNBC and/or BLBC and develop effective systemic therapies, especially molecular-targeted therapy.
Recent reports have revealed that TNBC/BLBC is a group of heterogeneous tumors [14]. BLBC also can be divided into extraordinarily diverse basal-like A and basal-like B subtypes [15]. The basal-like A cells have either luminal-like or basal-like epithelial morphology, while the basal-like B cells appear poorly differentiated and possess more mesenchymal characteristics [15]. Thus, the basal-like B subtype is more aggressive than the basal-like A subtype [15]. Due to the heterogeneity of BLBC, it is important to identify the critical regulatory factors that are associated with aggressive phenotype of BLBC. It is well known that various embryonic and mesenchymal EMT-TFs, including SNAIL1 [16], SNAIL2 [17] TWIST1 [18], and Forkhead box (FOX) transcription factor superfamily members FOXC1 [19], FOXC2 [4] and FOXQ1 [20], contribute to the aggressive phenotype of BLBC.
The mesenchymal regulator FOXF2 belongs to the FOX transcription factor superfamily [21]. It is specifically expressed in the mesenchyme adjacent to the epithelium in organs derived from the splanchnic mesoderm [22], and plays an important role in tissue homeostasis through regulating epithelium-mesenchyme interaction to maintain epithelium polarity [22]. Our previous clinical study demonstrated that the under-expression of FOXF2 is associated with early-onset metastasis and poor prognosis of patients with TNBC, but not the prognosis of non-TNBC patients [23]. This result suggests that FOXF2 deficiency is involved in TNBC/BLBC metastasis through regulating EMT. Recent studies have indicated that FOXF2 is a potential tumor suppressor in both prostate cancer [24] and breast cancer [25]. However, the role of FOXF2 in breast cancer metastasis and the underlying molecular mechanisms remain largely unknown.
In this study, we identified FOXF2 as a novel EMTsuppressing transcription factor in BLBC and demonstrated that it directly represses the transcription of TWIST1. FOXF2 deficiency promotes metastasis of BLBC cells through transcriptional upregulation of TWIST1 and activation of EMT.

Lentiviral transduction of shRNA and transfection of interfering RNA and plasmids
To obtain the stable FOXF2-knockdown cells, cells were infected with the lentiviral vector expressing a short hairpin RNA (shRNA) targeting the sequence GTCCTC AACTTCAATGGGATT in the coding sequence region of human FOXF2 gene (shFOXF2; Sigma-Aldrich, St Louis, MO, USA; TRCN0000013959). A shRNA nontargeting human and mouse gene was used as control (shControl; Sigma-Aldrich; SHC002). The cells were selected in 2 μg/mL puromycin to establish stable expressing shRNAs cells. For the rescue experiment, the stable FOXF2-knockdown cells were transfected with shRNAresistant full-length FOXF2 plasmid (Genechem, Nanjin, China), which has three nucleotide mismatches within the target sequence of the shFOXF2 (GTCCTCAATTT TAACGGGATT). To silence the expression of TWIST1, the cells were transiently transfected with two independent small interfering RNAs (siRNAs) targeting the coding sequence region of human TWIST1 gene GGATCA AACTGGCCTGCAA (si-TW#1) and GAACACCTTTA GAAATAAA (si-TW#2; RiboBio Co., Guangzhou, China) or a negative control siRNA (si-Control) using Lipofectamine 2000 (Invitrogen) according to the manufacturer's specifications. The human full-length FOXF2 cDNA was amplified from a pEV-FOXF2 plasmid, which was a gift from Professor Peter Carlsson (University of Gothenburg, Department of Cell and Molecular Biology, Sweden), and then was subcloned into the pcDNA3.1-HA vector (Invitrogen). The human full-length TWIST1 cDNA (Genechem) was subcloned into the pcDNA3.1 vector (Invitrogen). The cells were transfected with pcDNA3.1-HA-FOXF2 (FOXF2), pcDNA3.1-TWIST1 (TWIST1) or vector control (Invitrogen) using Lipofectamine 2000.

Reverse transcription quantitative polymerase chain reaction
Total RNA isolation from cultured cells or tissues, reverse transcription (RT) reactions, quantitative PCR (qPCR), and the quantification of target gene expression were performed as previously described [23].

Cell invasion and migration assay
Cell invasion and migration ability in vitro were assessed using Matrigel-coated and non-Matrigel-coated transwell inserts (BD Biosciences) as described previously [26].

Xenograft tumor assay
Twenty female 4-to 6-week-old severe combined immunodeficiency (SCID) mice were randomly divided into two groups for the xenograft tumor assay. Then, 5 × 10 6 231-Luc cells infected with shFOXF2 (231-Luc-shFOXF2) or shControl (231-Luc-shControl) were washed and harvested in 0.1 mL PBS and injected into the left lower abdominal mammary fat pad of the SCID mice [27]. After 22 days, all mice were anesthetized using isoflurance and injected intraperitoneally with 150 mg/kg D-luciferin/PBS. The bioluminescent imaging was performed using a Xenogen IVIS system (Caliper Life Sciences, Hopkinton, MA, USA). The data of living image were expressed as photon flux (photons per second). The tumor volumes were measured by caliper at the indicated time and calculated using the formula (length x width 2 )/2 [28]. The survival time of mice was defined as the time interval between the day of cell injection and the time of death or sacrifice. The xenograft tumors, livers and lungs were harvested from xenograft mice at sacrifice. The tissues were prepared for paraffin-embedded tissue section and hematoxylin and eosin (H&E) staining or immunohistochemistry staining. Two independent experiments were performed with xenograft mice. In one experiment, the mice died naturally to allow survival time calculations. In the second experiment, the mice were sacrificed for dissection on day 21 after cell injection. The mouse xenograft tumor assays were performed in accordance with protocols approved by the Animal Ethics Committee of Tianjin Medical University Cancer Institute and Hospital (TMUCIH; Tianjin, China. No. 058).

Breast cancer tissue specimens
A total of 34 primary TNBC tissue specimens were obtained from patients who underwent breast surgery in TMUCIH. The use of these specimens was approved by the Institutional Review Board of TMUCIH, and written consent was obtained from all participants. All cases were followed up over 3 years, and 30 cases were followed up over 5 years. Disease-free survival (DFS) was defined as the time interval between primary surgery and any relapse (local-regional, contra-lateral and/ or distant), or the terminal time of follow-up without any relapse events.

Independent data sets for validation of gene expression levels in breast cancer cell lines and breast cancer tissues
The FOXF2 mRNA expression pattern in different subtypes of breast cell lines was validated using the Gene expression-based Outcome for Breast cancer Online (GOBO) data sets [29].

Statistical analysis
Data from in vitro and in vivo experiments were presented as mean ± standard deviation (SD). Student's t test was used to compare differences between the experimental group and the control group. Fisher's exact test was used to compare the difference of metastatic incidence between 231-Luc-shFOXF2 mice and the 231-Luc-shControl mice. Spearman's rank correlation was used to analyze the correlation between FOXF2 and TWIST1 mRNA levels in breast cancer tissues. Survival plots were created using Kaplan-Meier analysis, and logrank test was used to assess statistical significance. P <0.05 was considered statistically significant.

FOXF2 deficiency enhances migration and invasion of basal-like mammary epithelial cells and BLBC cells in vitro
FOXF2 mRNA and protein expression levels in various breast cell lines, including basal-like subtype cells MDA-MB-231, BT-549, MCF-10A, HBL100 and non-basal-like cells MCF-7, BT474, ZR-75-30, SKBR-3 and MDA-MB-453, were measured by RT-qPCR and immunoblot, respectively. The results showed that FOXF2 was highly expressed in most basal-like cell lines and was less expressed in non-basal-like cell lines ( Figure 1A and B). This expression pattern of FOXF2 in the basal-like (n = 14) and non-basal-like (luminal; n = 25) subtype breast cell lines was validated using GOBO data set analysis ( Figure 1C). The result confirmed that FOXF2 was frequently and highly expressed in mesenchymal/myoepithelium-like breast cells. Combining these results with our previous finding that the under-expression of FOXF2 is associated with early-onset metastasis and poor prognosis of TNBC patients, but not the prognosis of non-TNBC patients [23], we hypothesized that FOXF2 may play a role in the metastasis of BLBC.
To investigate the roles of FOXF2 in regulating the invasive potential of BLBC cells in vitro, the immortalized non-tumorigenic basal-like mammary epithelial cell line MCF-10A and the invasive BLBC cell lines MDA-MB-231 was selected to establish stable FOXF2-knockdown cells by infection with a lentiviral vector expressing a FOXF2 shRNA (shFOXF2) or a non-target shRNA control lentiviral vector (shControl). The knockdown efficiency of FOXF2 in these cells was shown in Figures 1D  and 2B. Boyden chamber transwell assays showed that FOXF2 knockdown significantly promotes migration and invasion of both the two cell lines ( Figure 1E). To exclude the off-target effects of the shFOXF2, we performed a rescue experiment by transfecting a shRNA-resistant fulllength FOXF2 in the stable FOXF2-knockdown cells. The results showed that the effect of shFOXF2 on migration and invasion of MCF-10A and MDA-MB-231 cells was reversed ( Figure 1E). Next, we transiently transfected pcDNA3.1-FOXF2 in the BT549 cells, a BLBC cell line with high invasiveness and less FOXF2 expression. The overexpressed FOXF2 in the cells was shown in Figures 1D  and 2B. We observed that FOXF2 overexpression significantly decreased the migration and invasion of BT549 cells ( Figure 1E). In contrast with the role of promoting cell migration and invasion, the cell growth curves showed that FOXF2 knockdown significantly suppressed the proliferation of MCF-10A and MDA-MB-231 cells and the transfection of shRNA-resistant full-length FOXF2 rescued the effect of shFOXF2 on the proliferation of MCF-10A and MDA-MB-231 cells ( Figure 1F). Conversely, FOXF2 overexpression in BT549 cells promoted the cell proliferation ( Figure 1F). In addition, we also tested the functions of FOXF2 on luminal breast cancer cells by ectopically expressing FOXF2 in MCF-7 cells (see Figure  S1A in Additional file 1). However, the exogenous FOXF2 did not significantly affect the migration, invasion and proliferation of MCF-7 cells compared to the control cells (see Figure S1B and C in Additional file 1). Collectively, these results suggest that FOXF2 deficiency enhances the metastatic potential of basal-like mammary epithelial cells and breast cancer cells but reduces cell proliferation.

FOXF2 deficiency enhances metastatic ability of BLBC cells in vivo
To confirm the role of FOXF2 in metastasis of BLBC cells in vivo, 231-Luc-shFOXF2 and 231-Luc-shControl cells were injected into the mammary fat pad of female SCID mice. The general physical condition of mice with 231-Luc-shFOXF2 cells (231-Luc-shFOXF2 mice) was weaker compared to that of mice with 231-Luc-shControl cells (231-Luc-shControl mice) by day 22 after injection ( Figure 3A). The weight of 231-Luc-shFOXF2 mice was  also lower than that of the 231-Luc-shControl mice ( Figure 3B). Consistent with the cell proliferation experiments in vitro, 231-Luc-shFOXF2 mice had smaller xenograft tumors than the 231-Luc-shControl mice ( Figure 3C).
At day 22 post injection, we performed in vivo bioluminescent imaging to monitor the metastasis using a Xenogen IVIS system. The bioluminescent imaging and metastatic photon flux analysis revealed that the 231- Luc-shFOXF2 mice suffered significantly more local invasion and metastatic dissemination than the 231-Luc-shControl mice, especially in the lung and liver metastases ( Figure 3D). The data revealed that 80% of 231-Luc-shFOXF2 mice (8/10) suffered metastasis, whereas, 20% of the control mice (2/10) suffered metastasis ( Figure 3E). Visible liver metastatic nodules were found in half of the 231-Luc-shFOXF2 mice (5/10), whereas none were found in the livers of the 231-Luc-shControl mice ( Figure 3F). H&E staining confirmed that more metastatic foci formed in the lungs and livers in 231-Luc-shFOXF2 mice than in the control mice ( Figure 3G). Moreover, there was more muscle invasion in 231-Luc-shFOXF2 mice than in the control mice. The tumors had blood vessel invasion in the xenograft tumors of 231-Luc-shFOXF2 mice but not in the tumors of the control mice ( Figure 3H). A Kaplan-Meier survival analysis revealed that the 231-Luc-shFOXF2 mice had a significantly shorter survival time than the 231-Luc-shControl mice ( Figure 3I). Taken together, these results indicate that FOXF2 deficiency enhances the metastatic ability of BLBC cells and leads to invasive and metastatic disease in vivo.

FOXF2 deficiency induces EMT phenotype of basal-like mammary epithelial cells and BLBC cells
During mouse embryogenesis, FOXF2 is expressed in the mesenchyme derived from the splanchnic mesoderm adjacent to the ectoderm-derived epithelium and plays an important role in tissue homeostasis through regulating epithelium-mesenchyme interactions to maintain the epithelium polarity [22]. This expression pattern suggests a potential role of FOXF2 in regulating EMT. Therefore, we hypothesized that FOXF2 deficiency in BLBC metastasis might be involved the EMT process. We next investigate whether FOXF2 deficiency induces EMT in BLBC cells. Compared with the MCF-10A-shControl cells, FOXF2 knockdown transformed MCF-10A cells from epithelial morphology into fibroblast-like shape and caused cell scattering, and the expression of shRNA-resistant full-length FOXF2 rescued the phenotype (Figure 2A). The detection of EMT markers revealed that, compared to the control cells, FOXF2knockdown cells had reduced level of the epithelial marker E-cadherin in MCF-10A cells (undetectable in the MDA-MB-231 cell lines). Additionally, the FOXF2knockdown cells had elevated levels of the mesenchymal markers vimentin in the MCF-10A, MDA-MB-231 cell lines ( Figure 2B). Immunofluorescence staining revealed that the epithelial markers E-cadherin (undetectable in MDA-MB-231) and β-catenin were lost from cell membranes of FOXF2-knockdown cells and that the mesenchymal markers vimentin and fibronectin were increased ( Figure 2C). The expression of shRNA-resistant fulllength FOXF2 reversed the effect of shFOXF2 on the expression of EMT markers ( Figure 2B and C). Immunohistochemistry staining confirmed that vimentin was increased in primary xenograft tumors of 231-Luc-shFOXF2 mice and metastases in lungs and livers ( Figure 2D). FOXF2 knockdown upregulated the expression of MMP1 and MMP9 in MCF-10A, MDA-MB-231, and the expression of shRNA-resistant full-length FOXF2 rescued the effect of shFOXF2 ( Figure 2E). Conversely, FOXF2 overexpression in BT549 cells resulted in a reversal of the EMT phenotype (Figure 2A, B and E). However, the ectopic expression of FOXF2 in MCF-7 cells did not significantly affect the morphology (see Figure S1D in Additional file 1) and the expression of E-cadherin and vimentin (see Figure S1E in Additional file 1). These observations indicate that FOXF2 functions as an EMT suppressor in basal-like breast cells. FOXF2 deficiency enhances metastatic ability of basallike mammary epithelial cells and BLBC cells through inducing an EMT phenotype.

FOXF2 deficiency activates EMT program of basal-like mammary epithelial cells and BLBC cells through upregulation of EMT-TFs
To further investigate the mechanisms of how FOXF2 deficiency induces EMT phenotype, we examined the mRNA expression of EMT-TFs including SNAIL1, SNAIL2, TWIST1, ZEB1, ZEB2 and GSC (coding gene of Goosecoid) in MCF-10A, MDA-MB-231 cells with or without FOXF2 knockdown. The results showed that FOXF2 knockdown upregulated TWIST1 and ZEB2 expression in both MCF-10A and MDA-MB-231 cells ( Figure 4A and B). Only ZEB1 and GSC were upregulated in MDA-MB-231 cells and SNAIL2 was induced in MCF-10A cells ( Figure 4A and B). And the expression of shRNA-resistant full-length FOXF2 rescued the effect of shFOXF2 ( Figure 4A and B). Conversely, FOXF2 overexpression in BT549 cells downregulated SNAIL2, TWIST1, ZEB1, and GSC expression ( Figure 4C). However, the ectopic expression of FOXF2 in MCF-7 cells did not significantly change the expression of these EMT-TFs (see Figure S1F in Additional file 1). These results indicate that FOXF2 deficiency induced the EMT phenotype through the activation of different EMT-TFs in the individual basal-like breast cell lines.

FOXF2 binds to TWIST1 promoter and suppresses its transcription
Because TWIST1 mRNA expression levels were commonly and negatively correlated with FOXF2 expression in the basal-like breast cell lines ( Figure 4A, B and C), we investigated whether TWIST1 is a transcriptional target of FOXF2. First, we performed a BLAST search for FOXF2 binding sites in the promoter region of the TWIST1 gene. We found a putative FOXF2 binding site AATAAATA in the TWIST1 promoter region from −2120 to −2113 relative to the transcription start site (TSS; Figure 5A). Subsequently, the binding of FOXF2 on the TWIST1 promoter region containing or lacking the putative binding site was verified by ChIP assay in MCF-10A cells. The results confirmed that FOXF2 can bind to the TWIST1 promoter region containing the putative binding site but cannot bind to region lacking the putative binding site ( Figure 5B). Therefore, TWIST1 is a candidate transcriptional target gene of FOXF2.
To further assess the regulatory activity of FOXF2 on the TWIST1 promoter, we performed a luciferase reporter assay by transfecting pGL3-TWIST1 (−2140 to +27) containing the putative binding site or pGL3-TWIST1 (−2057 to +27) lacking the putative binding site in MCF-10A cells with or without FOXF2 knockdown. The results showed that the reporter activity of pGL3-TWIST1 (−2140 to +27) was increased in MCF-10A-shFOXF2 cells compared to the control cells, and the expression of shRNA-resistant full-length FOXF2 rescued the effect of shFOXF2 ( Figure 5C). The reporter activity of pGL3-TWIST1 (−2057 to +27) was not significantly changed in FOXF2-knockdown cells compared to the control cells ( Figure 5C). Conversely, FOXF2 overexpression in BT549 cells decreased the reporter activity of pGL3-TWIST1 (−2140 to +27; Figure 5C). Furthermore, the role of FOXF2 in negatively regulating TWIST1 protein expression was confirmed using FOXF2 loss or gain of function in MCF-10A and BT549 cells by immunoblot ( Figure 5D) and immunofluorescence detection ( Figure 5E). Whereas, the ectopic expression of FOXF2 in MCF-7 cells did not significantly change the reporter activity of pGL3-TWIST1 (see Figure S1G in Additional file 1) and TWIST1 expression (see Figure S1E in Additional file 1). These results demonstrate that FOXF2 suppress the transcriptional activity of the TWIST1 promoter in basal-like breast cells. Cumulatively, these data demonstrate that TWIST1 is a downstream target of FOXF2 and that FOXF2 suppresses TWIST1 promoter activity and protein expression in basal-like breast cells.

FOXF2 deficiency induces EMT phenotype of basal-like mammary epithelial cells and aggressive property of BLBC cells via upregulating TWIST1 expression
To further investigate whether TWIST1 mediates the role of FOXF2 deficiency in inducing EMT in basal-like breast cells, we knocked down TWIST1 expression in MCF-10A-shFOXF2 cells by transiently transfecting two independent TWIST1 siRNAs (si-TW#1 and si-TW#2), respectively. The results showed that the FOXF2 knockdown-induced EMT phenotype of MCF-10A cells was reversed by silencing TWIST1 expression. Also, the expression of E-cadherin and vimentin ( Figure 6A) was reversed, and epithelial morphology was restored ( Figure 6B). Additionally, the cells showed decreased migration and invasion ( Figure 6C). Next, we transiently transfected pcDNA3.1-TWIST1 into MDA-MB-231 cells, and the overexpressed TWIST1 in the cells was shown in Figure 6D. We observed that TWIST1 overexpression resulted in elevated levels of the mesenchymal markers vimentin ( Figure 6D) and enhanced migration and invasion of the cells ( Figure 6E). These results suggested that TWIST may mediate FOXF2-regulated EMT phenotype and aggressive property.
To provide in vivo evidence to illustrate the effect of FOXF2 on metastasis was mediated by negatively regulating TWIST1 expression, we detected the TWIST1 protein expression in primary tumors formed by inoculation of MDA-MB-231-Luc shFOXF2 and MDA-MB-231-Luc shControl cells in SCID mice by immunohistochemistry staining. The result showed increased TWIST1 expression in xenograft tumor of MDA-MB-231-Luc shFOXF2 comparing to the control tumor ( Figure 6F). Furthermore, we detected the mRNA levels of FOXF2 and TWIST1 mRNA in 34 cases of human primary TNBC tissues by RT-qPCR. Consistent with the negative regulation of FOXF2 on the transcription of TWIST1, TWIST1 mRNA levels inversely correlate with the FOXF2 mRNA levels in TNBC tumor (Spearman's rho = −0.421, P = 0.013; Figure 6G). Then, the TNBC cases were divided into four groups based on FOXF2 and TWIST1 mRNA levels: FOXF2 high /TWIS-T1 low (n = 8), FOXF2 high /TWIST1 high (n = 8), FOXF2 low / TWIST1 low (n = 8) and FOXF2 low /TWIST1 high (n = 10). Kaplan-Meier analysis showed that the patients in the four subgroups had distinct DFS: the patients in FOXF2 high / TWIST1 low and FOXF2 low /TWIST1 high groups had the best and poorest DFS, respectively (P = 0.002 between both); the patients in the FOXF2 high /TWIST1 high and FOXF2 low /TWIST1 low groups had moderate DFS (P = 0.687 between both); FOXF2 high /TWIST1 low group had better DFS than FOXF2 low /TWIST1 low (P = 0.063) and FOXF2 high /TWIST1 high (P = 0.143) groups; FOXF2 low / TWIST1 high group had poorer DFS than FOXF2 high / TWIST1 high (P = 0.089) and FOXF2 low /TWIST1 low (P = 0.149) groups ( Figure 6H). This result suggests that TWIST1 low could contribute to the FOXF2 high -suppressed recurrence and metastasis of TNBC cells; conversely, TWIST1 high could facilitate the FOXF2 low -induced recurrence and metastasis. Taken together, these results demonstrated that TWIST1 functionally mediates the effect of FOXF2 on EMT and metastasis in BLBC cells.

Discussion
The FOX transcription factors have tissue-specific expression patterns and play critical roles in embryogenesis and tissue development through regulating tissuespecific gene expression and cell differentiation [30]. Recent evidence has demonstrated that the deregulation of FOX factors is correlated with carcinogenesis and tumor progression [21]. It is worth noting that several FOX transcription factors link to the biological characteristics of breast cancer. FOXA1 is highly expressed in luminal subtype breast cancer [31], and its deficiency promotes  the metastasis of this subtype cancer [32]. FOXC1 [19], FOXC2 [4], and FOXQ1 [20] are highly expressed in basal-like subtype breast cancer, but are less present in luminal subtype cancer. The ectopic expression of these mesenchymal FOX factors in normal mammary epithelial cells is able to induce EMT and lead to an aggressive phenotype. FOXF2 is a mesenchymal transcription factor and specifically expressed in the mesenchyme adjacent to the epithelium and controls mesenchymal differentiation to maintain the tissue homeostasis via epitheliomesenchymal interactions [22]. FOXF2 was found highly expressed in benign prostatic hyperplasia and the transition zone of the normal prostate, where less suffer prostate cancer [33,34]. In this study, we showed that FOXF2 is highly expressed in most basal-like breast cells. Since basal-like breast cells are characterized by the basal gene expression and with mesenchymal characteristics [15], the expression pattern of FOXF2 in basal-like breast cells is consistent with its nature of mesenchymal transcription factors. We also found that FOXF2 deficiency promoted the metastasis of BLBC cells, which is consistent with our previous report that the under-expression of FOXF2 is associated with early-onset metastasis and poor prognosis of TNBC patients [23]. Our finding indicates that FOXF2 may play a role in controlling the differentiation state of basal-like breast cells and inhibits the aggressive property of BLBC.
It is well known that a variety of embryonic and mesenchymal transcriptional factors function as EMT activators and are involved in the aggressive behavior of BLBC, including SNAIL1 [16], SNAIL2 [17], TWIST1 [18], and FOX transcription factor family members FOXC1 [19], FOXC2 [4] and FOXQ1 [20]. Most EMT-TFs were identified as activators of EMT and promoters of BLBC metastasis, while EMT suppressors are rarely identified. FOXA2 is one such rare EMT suppressor [35]. In this study, we showed that FOXF2 is a novel EMT-suppressing transcription factor in basal-like breast cells. FOXF2 deficiency enhanced the metastatic ability of basal-like mammary epithelial cells and BLBC cells by inducing an EMT phenotype. We observed that the depletion of FOXF2 results in E-cadherin and β-catenin loss from the cell membrane. It is known that during EMT progression, the loss of E-cadherin expression from adherent junction results in the release of βcatenin into the cytoplasm. In addition, we observed that the depletion of FOXF2 leads to a significant increase in MMPs expression. Mesenchymal cells and the cancer cells undergoing EMT secrete extracellular matrix (ECM)-degrading enzymes to facilitate invasion and motility [36]. Matrix metalloproteinase (MMP)2 [37], MMP3 [38] and MMP9 [39] are able to induce EMT, and they trigger a positive regulatory feedback loop that stabilizes EMT. Therefore, the translocation of β-catenin translocation and upregulation of MMPs caused by FOXF2 deficiency could lead to initiate vicious cycle of EMT.
TWIST1 is recognized as a critical modulator in EMT, and its overexpression is able to trigger EMT phenotype and tumor metastasis [3]. Thus, clarifying the regulatory mechanism of TWIST1 expression would provide insight into the regulatory pathway of TWIST1-induced EMT programming and aid in developing molecular targets to inhibit tumor metastasis. It has been reported that HIF-1 [40] and p65 [41] transcriptionally activate TWIST1 expression to mediate hypoxia-induced and tumor necrosis factor alpha (TNF-α)-induced EMT and tumor metastasis, respectively. miR-720 [42] and miR-106b [43] suppress TWIST1 expression by posttranscriptional regulatory mechanisms. In this study, we identified FOXF2 as a transcriptional suppressor of TWIST1. Therefore, we speculate that the two embryonic/mesenchymal transcription factors might act cooperatively to maintain tissue homeostasis through balancing the differentiation or dedifferentiation of mesenchymal/myoepithelial cells. FOXF2 deficiency could lead to more dedifferentiation of myoepithelial cells and an aggressive phenotype in BLBC via a TWIST1mediated pathway. This novel finding provided an unanticipated regulatory pathway of TWIST-induced EMT and metastasis in BLBC.
In EMT programming, pleiotropic EMT-TFs form an interaction network to enable them to work in concert to regulate the EMT phenotype. The zinc-finger transcription factors SNAIL1 [44], SNAIL2 [45], ZEB1 [46] and ZEB2 [47] directly bind to the E-boxes of the CDH1 promoter to repress its transcription. TWIST1 [3], Goosecoid [48], and FOXC2 [4] trigger EMT by indirectly suppressing CDH1 transcription. TWIST1 directly regulates SNAIL2 transcription, and SNAIL2 is an essential mediator of TWIST1-induced EMT [49]. TWIST1 can also induce FOXC2 and SNAIL2 expression [7]. SNAIL1 has been found to activate ZEB1 [2], FOXC2 and SNAIL2 expression [7]. In this study, we found that FOXF2 not only negatively regulated TWIST1 expression but also regulated the expression of SNAIL1 and SNAIL2, ZEB1, ZEB2 and GSC in basal-like cell lines. Interestingly, FOXF2 regulated TWIST1 expression in all the three cell lines, but only regulated SNAIL2 in MCF-10A and BT-549 cells, regulated ZEB1 and GSC in MDA-MB-231 and BT-549 cells, regulated ZEB2 in MCF-10A and MDA-MB-231 cells. Therefore, in addition to activating TWIST1 transcription, FOXF2 deficiency may induce EMT through orchestrating different EMT-TFs expression in the different basal-like breast cells.
In addition to BLBC, we also tested the effect of FOXF2 on luminal breast cancer cells. However, we observed that the ectopic expression of FOXF2 could not significantly change the phenotype of luminal breast cancer cells, even slightly enhance the migration and invasion, and slightly increase the transcription of TWIST1. Our results suggest that there are different functions of FOXF2 in basal-like and luminal cells. Based on our current results, we speculate that mesenchymal FOXF2 might be induced expression in luminal cells that had undergone transdifferentiation into basal-like cells, and when FOXF2 depletion during the progression of the cells adapted mesenchymal characteristics could drive these cells to possess more mesenchymal stem-like plasticity and aggressive phenotype by activating the transcription of target gene TWIST1 and upregulating the expression of other EMT-TFs. The regulatory mechanisms of the different expression pattern and functions of FOXF2 in basal-like and luminal breast cancer cells should be further deeply investigated. Since FOXF2 plays an essential role in the metastasis of BLBC cells, the exogenous administration of FOXF2 or interference of FOXF2-TWIST1 pathway may be a promising therapeutic strategy for improving the outcome of BLBC patients.

Conclusions
In conclusion, our study identified FOXF2 as a novel EMT-suppressing transcription factor in BLBC. FOXF2 deficiency enhances metastatic ability of BLBC cells via activating an EMT program by upregulating the transcription of TWIST1.

Additional file
Additional file 1: Figure S1. (G) Transcriptional activity of TWIST1 promoter in the indicated cells was assessed by a dual-luciferase reporter assay. Three independent assays were performed in triplicate. The data were expressed as mean ± SD. n.s., P >0.05.