FOXA2-Interacting FOXP2 Prevents Epithelial-Mesenchymal Transition of Breast Cancer Cells by Stimulating E-Cadherin and PHF2 Transcription

FOXP2, a member of forkhead box transcription factor family, was first identified as a language-related gene that played an important role in language learning and facial movement. In addition, FOXP2 was also suggested regulating the progression of cancer cells. In previous studies, we found that FOXA2 inhibited epithelial-mesenchymal transition (EMT) in breast cancer cells. In this study, by identifying FOXA2-interacting proteins from FOXA2-pull-down cell lysates with Mass Spectrometry Analysis, we found that FOXP2 interacted with FOXA2. After confirming the interaction between FOXP2 and FOXA2 through Co-IP and immunofluorescence assays, we showed a correlated expression of FOXP2 and FOXA2 existing in clinical breast cancer samples. The overexpression of FOXP2 attenuated the mesenchymal phenotype whereas the stable knockdown of FOXP2 promoted EMT in breast cancer cells. Even though FOXP2 was believed to act as a transcriptional repressor in most cases, we found that FOXP2 could activate the expression of tumor suppressor PHF2. Meanwhile, we also found that FOXP2 could endogenously bind to the promoter of E-cadherin and activate its transcription. This transcriptional activity of FOXP2 relied on its interaction with FOXA2. Furthermore, the stable knockdown of FOXP2 enhanced the metastatic capacity of breast cancer cells in vivo. Together, the results suggested that FOXP2 could inhibit EMT by activating the transcription of certain genes, such as E-cadherin and PHF2, in concert with FOXA2 in breast cancer cells.


INTRODUCTION
Breast cancer is the most common malignant tumor and the leading cause of cancer death among women (1). Metastasis of breast cancer is one of the main reasons for poor survival in patients (2). One of the important mechanisms regulating the invasive behavior of cancer cells is the epithelialmesenchymal transition (EMT), which represents the conversion of differentiated epithelial cancer cells into migratory mesenchymal cancer cells, leading to cancer invasion, systemic cancer cell dissemination and metastasis (3). Moreover, EMT results in cancer cells avoiding cellular senescence and apoptosis, and participates in the generation and maintenance of cancer stem cells (4), highly consistent with the ability of metastatic cells to initiate new tumors (5). During the progress of EMT, the expression of epithelial markers such as the junction protein E-cadherin is lost and the expression of mesenchymal markers such as Vimentin is up-regulated in cancer cells (6). Gene expression profiling experiments of EMT suggest that many genes adjust in their expression during EMT (7), regulated by a network of signaling pathways from a variety of growth factors [i.e., epidermal growth factor (EGF) (8)] and multiple transcription factors (9,10).
In our previous studies, transcription factor FOXA2 was confirmed to inhibit EMT in breast cancer cells by regulating the transcription of EMT-related genes and the stable overexpression of FOXA2 abolished breast cancer cell metastasis in vivo (11). Thus, we intend to identify FOXA2interacting proteins from FOXA2-pulled down cell lysates with Mass Spectrometry Analysis in current studies. Interestingly, transcription factor FOXP2, another member of the FOX transcription factor family, has been found to interact with FOXA2. The FOX transcription factor family is widely distributed in various eukaryotes and contains more than 40 mammalian members, which possess a conserved DNA binding domain (DBD) known as Forkhead box/winged helix domain (12). The chromatin immune-precipitation experiment identifies the candidate FOXP2-binding sequence CAAATT as the most probable target for FOXP2 binding in chromatin (13). FOXP2 has been shown to both promote and more often inhibit the transcription of target genes (14). FOXP2 can interact with the co-repressors, such as C-terminal binding protein-1 (CtBP-1) that mediates transcriptional repression primarily through recruitment of histone deacetylases HDAC1/2 (15), to confer its transcriptional repressive properties (16,17). An increasing amount of evidence supports the repressor role of FOXP2 upon the transcription of its target genes, such as SRPX2/uPAR complex (18) and DLL3 (19), which are involved in oncogenic progression of different types of cancers. On the other hand, FOXP2 has also been reported to activate the transcription of genes, such as the protein-tyrosine kinase SYK that is described as a tumor suppressor in breast cancer cells (20). This transcriptional activation of FOXP2 is often explained by the differential affinity of FOXP2 for DNA binding sites or by the cofactors that interact with FOXP2.
While FOXP2 has first been reported to participate in speech and language development and neuronal development (21,22), the expression of FOXP2 is observed in multiple adult tissues, such as heart, lung, liver, ovaries, and gut (23,24). A growing number of evidences have linked FOXP2 to multiple cancers and its dysregulation may play a main role throughout cancer initiation and progression (25), even though it may act as either a tumor-suppressor or a tumor-stimulator depending on the type of cancers studied. For example, its expression is downregulated in breast cancer (26), hepatocellular carcinoma (27), and gastric cancers (28), in which FOXP2 plays roles as a tumorsuppressor. Conversely, overexpressed FOXP2 has been found in lymphomas (29), neuroblastomas (30), and prostate cancers (31), implicating a pro-oncogenic role of FOXP2 in these cancers. These differences may suggest alternative and tissue-specific roles for FOXP2 as a tumor suppressor or as an oncogene, depending on activated signaling pathways in certain types of cancer. The strong evidence of FOXP2 as a tumor-suppressor role comes from a breast cancer study, in which silencing FOXP2 through miRNA-mediated FOXP2 repression promotes cancer stem cell traits and metastasis in breast cancer cells (32).
In the current study, we identified that FOXP2 interacted with FOXA2, and the expression of FOXP2 was strongly correlated with the epithelial phenotype of breast cancer cells. The stable knockdown of FOXP2 expression promoted the mesenchymal phenotype of breast cancer cells, while the overexpression of FOXP2 inhibited the EMT of breast cancer cells. We confirmed that FOXP2 alone could activate the expression of tumor suppressor PHF2. Meanwhile, FOXP2 could endogenously bind to the promoter of E-cadherin and activate E-cadherin transcription, relying on its interaction with FOXA2. Furthermore, the stable knockdown of FOXP2 enhanced the metastatic capacity of breast cancer cells in vivo. Together, the results suggested that FOXP2 could inhibit EMT in breast cancer cells by activating transcription of certain genes, such as E-cadherin and PHF2.

Cell Culture
The

Mass Spectrometry (MS)
The Avi-FOXA2 DNA fragment was PCR amplified with the following primers: Forward 5′-GGA ATT CAT GTC CGG CCT GAA CGA CAT CTT CGA GGC TCA GAA AAT CGA ATG GCA CGA AAC TAG TAT GCA CTC GGC TTC CAG T-3′ and Backward5′-CCC AAG CTT TTA AGA GGA GTT CAT AAT-3′, using pCMV-FOXA2 (11) as the template. The PCR product was digested and cloned into EcoRI and HindIII sites of pCDNA3.1 plasmid to obtain pAvi-FOXA2. The BirA of E. coli was PCR amplified from E. coli genomic DNA with the following primers: Forward 5′-CGG ATC CAT GAA GGA TAA CAC CGT GCC ACT G-3′ and Backward5′-CGT CTA GAG GTA GAA GAG GTC AGA CTA CGC-3′. The PCR product was digested and cloned into BamHI and XbaI sites of the lentivirus plasmid vector (33) to obtain pEF1-BirA. MCF-7 cells were plated in 10 cm dishes and transfected with pAvi-FOXA2 (10 mg), pEF1-BirA (5 mg), or both. Forty hours later, the cells were collected, suspended in 100 ml of lysis buffer (150 mM NaCl, 1% NP-40, 50 mM Tris-HCl pH 7.4, protease inhibitor mixture) and incubated for 20 min on ice. The lysates were centrifuged for 15 min at 14,000 g at 4°C and the supernatant containing 200 mg of proteins was incubated with Streptavidin Resin (Sangon Biotech C006390, China) overnight. The resin beads were centrifuged and washed four times with the lysis buffer and subjected to SDS-PAGE, followed by staining with coomassie brilliant blue. Protein samples from SDS-PAGE gel were digested with 10 nmol of MS grade trypsin for 8 h at 37°C and used for mass spectrometry analysis with LTQ Orbitrap Velos Rro (Thermo Fisher Scientific, Swiss).

Clinical Data Analysis
Clinical breast cancer samples expressing FOXA2 (the BRCA data set, n=394) including both tumor and non-tumor tissue were collected from The Cancer Genome Atlas (TCGA) database (see Supplementary Table). The TCGA samples not expressing FOXA2 were excluded for the further analysis in the study. The analysis of FOXP2 levels in the four subgroups (Basal, Her2, LumA, and LumB) of breast cancer and the correlation analysis between FOXA2 and FOXP2 in the BRCA data set were executed by using ggstatsplot package through R project (http://www.rproject.org). For survival curve analysis, two FOXA2-related data sets were first extracted from the BRCA data set, in which the FOXA2 High subgroup (n=131) or the FOXA2 Low subgroup (n=131) contained either the top 1/3 or the bottom 1/3 of total samples respectively according to the levels of FOXA2 expression. FOXP2-related survival of patients in the FOXA2 High and FOXA2 Low subgroups was fitted by the "survfit" function, and Kaplan-Meier curves were drawn by the "ggsurv" function in the R package "survival". The cut-off value of FOXP2 high or low expression was determined by setting "minprop" parameter to 0.3.
Total RNA Isolation, Blood RNA Isolation, Quantitative Real-Time PCR (qPCR) Total RNAs from cells or tumors were prepared with Total RNA Kit I (Omega, USA) and blood RNAs from mice were isolated by the Blood RNA Kit (Omega, USA). The cDNAs were synthesized with M-MLV Reverse Transcriptase (Invitrogen, USA) from RNA samples (2 mg) to get 100ml cDNAs. Quantitative realtime PCR (qPCR) was performed by using SYBR Green (Transgen, China) in the realplex2 qPCR system (Eppendorf, Germany) with 1 ml cDNAs as templates in each reaction.

Lentivirus Construction and Infection
Three different FOXP2-specific shRNA fragments and a control fragment were cloned into AgeI and EcoRI sites of lentivirus shRNA interference plasmid pMAGic7.1 behind the U6 promoter to obtain pU6-shFOXP2#1, #2, #3, respectively. The sequences for FOXP2-specific shRNA and control fragments primers are as follows: Then, the constructed plasmid (pU6-shFOXP2#1, #2, #3, or pU6-shControl) was cotransfected into 293T cells with two packaging plasmids (pVSVG and D8.91) by calcium phosphate transfection to produce lentiviruses. Forty-eight hours post transfection, the medium of 293T was collected and the titration of the virus was measured by flow cytometry (Beckman Coulter, USA). MCF-7 cells were infected with the lentivirus (20 pfu/cell) containing either FOXP2-specific shRNA or shControl to establish cell lines stably expressing shFOXP2 or shControl. Because the lentiviral vector contained an EGFP expression cassette, the infected cells were screened by flow cytometry to obtain cell lines stably expressing EGFP that represented the stable expression of shFOXP2 or shControl. The MCF-7 cell line that stably expresses shFOXP2#2 was used in subsequent experiments.

Transwell Assays
Cell migration assays were performed by using Transwell migration chambers (8 mm pore size; Corning, USA) according to the vendor's instructions. Briefly, the cells were trypsinized and 1× 10 4 cells were plated into the insert of the well. 24 h later, the cells in the insert of each well were removed and the cells under the bottom of the well were stained by 0.1% hexamethylpararosaniline. All experiments were repeated three times. Representative photos were taken using a TE2000 microscope (Nikon Instruments Inc., Japan) (100×) or SMZ1500 stereomicroscope (Nikon Instruments Inc., Japan) (10×). The digital pixel densitometry from at least three different photos was measured with Image-J software (NIH, USA).

Chromatin Immunoprecipitation (ChIP) Assays
ChIP assays were performed as previously described (11). The following antibodies were used for immunoprecipitation: rabbit anti-FOXP2 (Cell Signaling Technology #5337, USA), rabbit anti-IgG (Millipore PP64, USA). For immunoprecipitation, 2 mg of each antibody was used. The ChIP DNA sample or 1% total input (5 ml) was used in qPCR with the following primers: E-

Electrophoretic Mobility Shift Assays (EMSA)
FAM-labeled double-strand DNA oligonucleotides were synthesized by Sangon (Shanghai) Co., Ltd, China, based on the sequence 5′-ACT GTT TGT AAA CAG GCT AAT A-3′ from PHF2 upstream region (-1766 bp to -1744 bp) and the sequence 5′-AAA AAT ACA AAC AAA ACA AA-3′ from E-cadherin upstream region (-706 bp to -686 bp), which contained the FOXP2 consensus binding sites. In the binding reactions, 10 mg of nuclear proteins isolated from FOXP2-expressing cells was incubated with 1 pmol of the FAM-labeled probe and 2 ml of 5×binding buffer (Beyotime, China) in a total volume of 10 ml for 30 min at room temperature. The reactions were resolved in 4% native polyacrylamide gel electrophoresis in 0.5×TBE. The dose chosen for the competitive experiments was in the 50× molar excess of the unlabeled oligonucleotides. The oligonucleotides mutated in FOXP2 binding sites (5′-ACT GTT CTG CCC ACC GCT AAT A -3′ for PHF2 or 5′-AAA CAT CCG CAC GCT ACA AA-3′ for E-cadherin) were also used as controls in EMSA experiments. For the supershift analysis, 1 mg of anti-FOXP2 (Cell Signaling Technology #5337, USA) was added to the binding reaction.

Transfection of siRNA
Human FOXA2 siRNA (Santa cruz, sc-35569) and control siRNA (Santa cruz, sc-37007) were purchased from Santa Cruz, USA. The siRNA transfection was performed with Lipofectamine 2000 (Invitrogen, USA) according to the manufacturer's instructions.

Animal Experiments
All animal experiments were conducted in accordance with institutional animal care and use guidelines, following approval by the Laboratory Animal Center of Hunan, China (Protocol No. SYXK [Xiang] 2008-0001). BalB/c nude mice (female, 4-week old) were purchased from Slac Experimental Animal Company (Changsha, China). To generate mouse models of metastasis of breast cancer cells in vivo, MCF-7 cells expressing EGFP or MCF-7 cells stably knockdown FOXP2 were injected into the tail vein of each mouse (2×10 5 cells/mouse). At day 1, day 30, or day 60 post injection, the mice (n= 6, each group) were sacrificed and blood samples or lung samples were collected. The total RNAs of blood samples or lung samples were isolated and the relative concentration of human tumor cells in the blood or in the lung was determined by qPCR for the mRNA levels of human specific CYCLOPHILIN over the mRNA levels of mouse specific Cyclophilin. The collected lung tissues were fixed overnight in 4% PFA and embedded in paraffin. Sections were stained with hematoxylin and eosin dyes.

Statistical Analysis
We used Microsoft Excel to calculate SD and determine statistically significant differences between samples and used GraphPad Prism to draw the bar graphs. The asterisks in each graph indicate statistically significant changes with P values calculated by Student T Test: *P < 0.05, **P < 0.01 and ***P < 0.001. P values <0.05 were considered statistically significant.

Identification of FOXA2-Interacting Transcription Factor FOXP2
To identify transcription factors interacting with FOXA2 proteins to regulate gene expression, we constructed eukaryotic expression vectors pEF1-BirA and pAvi-FOXA2 and established a biotin tagging system for FOXA2 transcription factor ( Figure S1). The potential FOXA2-interacting proteins were pulled down with streptavidin resin from MCF-7 cell lysates and separated by PAGE ( Figure 1A). The samples were analyzed by mass spectrometry and 28 putative FOXA2-interacting proteins were identified ( Table 1) after filtering out non-specific interactions. Some of the identified proteins, such as HDAC1, had been confirmed to interact with FOXA2 (34), proving the reliability of the analysis. Among the 28 proteins, only three proteins NCOA3 (35), CRTC3 (36), and FOXP2 were found to be transcription factors. FOXP2, as a member of the FOX transcription factor family and a suppressor of tumor metastasis in various human cancers (26,27), was chosen for subsequent experiments in this study. First, the lysates of MCF-7 cells transfected with pAvi-FOXA2 and pHis-FOXP2 plus pEF1-BirA  or not, were pulled down with streptavidin resin (binding to Avi-FOXA2) or Ni beads (binding to His-FOXP2). Following Western blotting, we confirmed that FOXA2 and FOXP2 interacted physically in breast cancer MCF-7 cells (Figures 1B, C). Subsequently, we mapped the region mediating FOXA2-FOXP2 interaction in the proteins, using different Flag-tagged fragments of the two proteins for co-imnunoprecipitation assays. FOXA2 protein was divided to three regions including 1-165aa (the Nterminal transcription activation domain), 166-324aa (the DNA binding domain), 325-463aa (the C-terminal transcription activation domain), and FOXP2 protein was divided to three regions including 1-196aa (the N-terminus), 197-408aa (the zinc-finger/leucine zipper motif), 409-633aa (the DNA binding domain plus the C-terminus). The 166-324aa region of FOXA2 and the 197-408aa region of FOXP2 were identified to participate in the interaction between the two proteins ( Figures 1D, E). The interaction of endogenous FOXA2 and FOXP2 was further confirmed with immunofluorescence assay, which showed that FOXA2 and FOXP2 co-localized in the nucleus of MCF7 cells ( Figure 1F). Then we analyzed the levels of FOXP2 expression in the clinical breast cancer samples (n=394) from TCGA (Supplementary Table) and found that the levels of FOXP2 in all four breast cancer subgroups (Basal, Her2, LumA, and LumB) were lower than that in the normal group ( Figure 1G). Based on this data set, we observed a weak correlation between the expressions of FOXA2 and FOXP2 in the samples (R=0.14, p=0.007), while a moderate correlation between their expressions was further found in the subgroup of basal breast cancer especially (n=98, R=0.34, p=0.001) (Figures 1H, I).
Together, these results confirmed that the FOXA2-FOXP2 interaction occurred in breast cancer cells, implicating the involvement of the two proteins in regulating the cancer cells.

FOXP2 Inhibited the EMT of Breast Cancer Cells
The expression levels of FOXP2 varied among different breast cancer cell lines ( Figure S2), in which the epithelial-type cell lines such as MCF-7 and HCC1397 exhibited higher levels of FOXP2 than that of the mesenchymal-type cell lines such as MDA-MB-231, MDA-MB-453, and MDA-MB-436. This variation of FOXP2 levels matched the different FOXA2 levels among the cell lines mentioned above ( Figure S2) and implicated that FOXP2 might regulate the epithelial phenotype of the cancer cells. The prediction was first supported by the EGF-induced EMT of MCF-7 cell model in which the epithelial MCF-7 cells acquired a mesenchymal morphology at day 6 post EGF treatment (Figure 2A). Both the protein and mRNA levels of FOXP2 were decreased by the EGF treatment, correlated to the decreased levels of the epithelial marker E-cadherin and the increased levels of the mesenchymal marker Vimentin ( Figures  2B, C). To test whether FOXP2 was required for the epithelial phenotype of breast cancer cells, we used RNA interference strategy to knock down the levels of FOXP2 in MCF-7 cells. The transfection of the three constructed FOXP2-specific shRNA expression plasmids (pU6-shFOXP2#1, #2, #3) resulted in a significant but similar decrease of FOXP2 mRNA and protein levels in MCF-7 cells, among them shFOXP2#2 performing slightly better ( Figure S3). For further analyzing the roles of FOXP2 in EMT with in vitro and further in vivo experiments, we infected MCF-7 cells with the lentiviruses expressing various FOXP2-specific shRNA (#1, #2, #3) and selected the FOXP2 stable-knockdown clones of MCF-7 cells through the lentiviral vector-expressed EGFP. FOXP2-knockdown MCF-7 cells exhibited the spindle-like morphology compared with control lentivirus-infected MCF-7 cells ( Figure S4). Consequently, the knockdown of FOXP2 also decreased the expression of Ecadherin and resulted in the increased expression of Vimentin in the cells (Figures 2D, E). In addition, the migration ability of FOXP2-knockdown cells was dramatically strengthened when measured by Transwell invasion test ( Figure 2F). To test whether FOXP2 abolished the phenotype of mesenchymal breast cancer cells, we transfected the FOXP2-expression plasmids into MDA-MB-231 cells. We found that the overexpression of FOXP2 resulted in the elevation of E-cadherin levels and the reduction of Vimentin levels ( Figures 2G, H) and    (Figures 3A, B). As predicted, the FOXP2 expression was induced during this CTx-induced MET progression ( Figures 3A, B). PHF2 (Plant homeodomain finger 2), a demethylase participating in epigenetic regulation of gene expression through demethylation of histone H3K9m3 (39), was also stimulated in this MET model ( Figures  3A, B). PHF2 was considered as a tumor suppressor because of its decreased levels in various tumor tissues (40) and could upregulate certain epithelial genes, leading breast cancer cells to acquire epithelial phenotypes (38). We noticed the correlation of the increased expression of FOXP2 and PHF2, and then found that the overexpression of FOXP2 alone was able to stimulate both the mRNA and protein levels of PHF2 dramatically in The asterisks indicate statistically significant changes: *P ≤ 0.05, **P ≤ 0.01, and ***P ≤ 0.001. Figure 3C). To analyze whether FOXP2 stimulated the PHF2 transcription directly, we scanned -2kb promoter region of human PHF2 gene with the FOXP2 DNA binding consensus sequence and found multiple putative FOXP2 binding sites at the regions -1,519 bp to -1,403 bp and -1,838 bp to -1,693 bp in this promoter. ChIP assays confirmed that FOXP2 bound to endogenous PHF2 promoter at regions around -1,838 bp to -1,693 bp but not on another tested region in the cells ( Figure 3D). EMSAs were performed to confirm the binding of FOXP2 at the PHF2 promoter with a FAM-labeled DNA probe synthesized from the PHF2 promoter region from -1,765 bp to -1,743 bp and nuclear extracts containing FOXP2 proteins. We found that the probe could form a DNA/protein complex in EMSAs with FOXP2, and the addition of either an unlabeled probe (50×) or FOXP2-specific antibody disturbed the formation of the FOXP2/DNA complex or resulted in a supershift complex, whereas the FAM-labeled mutated probe could not form the FOXP2/DNA complex ( Figure 3E). To further confirm FOXP2 stimulating the PHF2 promoter, we constructed a luciferase reporter plasmid containing the fragment of −2,000 bp to +60 bp of PHF2 promoter. This PHF2 promoter was activated by FOXP2 in a dosage-dependent manner in MCF-7 cells ( Figure  3F). These results demonstrated that FOXP2 might inhibit the EMT of breast cancer cells by directly stimulating the transcription of epithelial-driven gene PHF2, contrary to the general notion of FOXP2 mainly acting as a transcriptional repressor.

The Activation of the E-Cadherin Promoter by FOXP2 Relied on the Participation of FOXA2
The expression of the junction protein E-cadherin, as a typical epithelial cell marker, decreased during EMT and the decline of E-cadherin's functions promoted the EMT progression (41). In this study, we found that the levels of E-cadherin relied on FOXP2 in breast cancer cells (see Figure 2). We further investigated whether FOXP2 could stimulate the transcription of E-cadherin. In the dual-luciferase reporter assays, FOXP2 was able to activate the -733 bp to +60 bp region of E-cadherin promoter in MCF-7 cells ( Figure 4A), while the knockdown of FOXP2 resulted in the down-regulated transcriptional activities of the promoter ( Figure S6). When this promoter was scanned with the consensus FOXP2-DNA binding sequence, we identified multiple putative FOXP2 binding sites at the regions -692 bp to -681 bp and -316 bp to -305 bp. ChIP assays confirmed that FOXP2 bound to endogenous E-cadherin promoter at regions around -760 bp to -610 bp but not another tested region ( Figure 4B). Consistent with this finding, EMSAs showed that a FAM-labeled DNA probe, synthesized from the E-cadherin promoter region from -706 bp to -686 bp, could form a DNA/protein complex with FOXP2, and the addition of either an unlabeled probe (50×) or FOXP2-specific antibody disturbed the formation of the FOXP2/DNA complex or resulted in a supershift complex, whereas the FAM-labeled mutated probe could not form the FOXP2/DNA complex ( Figure 4C). These results suggested that the E-cadherin promoter was activated by FOXP2. To test whether the interaction between FOXP2 and FOXA2 affected the transcription of E-cadherin, we performed an additional EMSA experiment, in which the nuclear extracts containing overexpressed FOXP2 only, overexpressed FOXA2 only, or overexpressed FOXP2 and FOXA2 together, were used. We found that FOXP2 and FOXA2 together formed complexes binding to the DNA probe but FOXA2 alone could not bind to the probe ( Figure 4D), implicating that the mechanism of FOXP2 stimulating the E-cadherin transcription might involve its interaction with FOXA2. This speculation was further supported by the evidence that the knockdown of FOXA2 could significantly decrease the activation of FOXP2 on Ecadherin promoter ( Figure 4E). Furthermore, the simultaneous knockdown of FOXP2 and FOXA2 synergistically inhibited the expression of E-cadherin in MCF-7 cells (Figures 4F, G). When tested in MDA-MB-231 cells possessing low endogenous levels of FOXA2, FOXP2 showed a weaker stimulation on E-cadherin promoter than that in MCF-7 cells ( Figure S7). The overexpression of FOXP2 and FOXA2 together synergistically activated the expression of E-cadherin in MDA-MB-231 cells ( Figures 4H-J), resulting in a noticeable morphological change of the cells at the same time ( Figure S8). Together, these observations suggested that FOXP2 could bind to the Ecadherin promoter and stimulate the transcription of Ecadherin through the FOXP2-FOXA2 interaction in breast cancer cells.

Stable Knockdown of FOXP2 Activated Migration Capability of Breast Cancer Cells In Vivo
To further investigate the functions of FOXP2 on the EMT of breast cancer cells in vivo, we generated a metastasis model in nude mice via the tail vein injection of the FOXP2-specific shRNA#2 lentivirus-infected MCF-7 cells (shFOXP2) or control lentivirus-infected MCF-7 cells (shCon) (n=12 for each group). The mice were sacrificed at days 1, day 30, and day 60 post-injection, in which the total RNAs of the blood of each mouse were isolated from day 1 and day 30 groups (n=4) and the total RNAs of the lung tissue of each mouse were isolated from day 60 groups (n=4). The relative concentration of circulating human tumor cells in the blood or the lung tissue were determined by the mRNA levels of human-specific CYCLOPHILIN (Gene Symbol: PPIA) compared with those of mouse-specific Cyclophilin (11,42). The stable knockdown of FOXP2 in MCF7 cells increased the number of circulating tumor cells in the blood of tested mice of day 30 groups ( Figure 5A). An elevated metastasis and tumor formation were found in the lungs of shFOXP2 cell-injected mice at day 60 ( Figure 5B). The knockdown of FOXP2 enhancing the lung metastasis of the cells was further observed by EGFP fluorescence intensity in the isolated tissues ( Figure S9A, B). The amount of metastasized tumor cells in the lungs of day 60 groups was also measured by the mRNA levels of human-specific CYCLOPHILIN over mouse-specific Cyclophilin and showed a significant increase in the FOXP2 knockdown group ( Figure 5C). Compared to the controls, the shFOXP2 cells produced obvious tumors in the lung tissue of day 60 groups with H&E staining ( Figure 5D). To analyze the levels of selected genes from the grafted human cancer cells, we designed species-specific qPCR primers for FOXP2, E-cadherin, and PHF2 of human and mouse. Using the total RNAs of lung tissues of day 60 groups, we found that there was no significant difference in the expression of endogenous mouse FOXP2, E-cadherin, and PHF2 between the two groups ( Figure S10). At the condition that the tail veininjected shFOXP2 cells produced more tumors in lung tissues, we observed lower mRNA levels of exogenous human FOXP2, Ecadherin, and PHF2 in the harvested samples of the shROXP2 groups than that of the control groups ( Figure 5E), suggesting that the knockdown of FOXP2 and following down-regulated levels of E-cadherin and PHF2 resulted in elevated abilities of metastasis of the cancer cells. Together, these results determined that inhibition of FOXP2 could enhance the metastasis of breast cancer cells in vivo. This was supported by the further analysis of the clinical relevance of FOXP2 expression from TCGA data. The levels of FOXP2 were significantly down-regulated in the group of invasive breast carcinomas compared to normal breast tissue ( Figure 6A). To further analyze the effect of FOXP2 on the survival of breast cancer patients and test whether its effect relied on the levels of FOXA2, we built two FOXA2-related subgroups of clinical samples, in which the FOXA2 High subgroup (n=131) or the FOXA2 Low subgroup (n=131) contained either the top 1/3 or the bottom 1/3 of the collected BRCA data set (n=394) respectively according to the levels of FOXA2 expression. In the FOXA2 High subgroup, we found that the patients with high levels of FOXP2 showed a better performance on the survival than the patients with low levels of FOXP2 (p=0.03) ( Figure 6B). However, no significant difference in the survival probability was found between the patients with high or low levels of FOXP2 in the FOXA2 Low subgroup ( Figure 6B). The results confirmed that the FOXP2-improved survival in breast cancer patients was correlated with the high levels of FOXA2. normal epithelial characteristics of breast by selectively inhibiting ZEB2 expression (45). Our previous research found that FOXA2 inhibits EMT in breast cancer cells by stimulating Ecadherin transcription and repressing ZEB2 transcription (11). In this study, FOXA2-interacting FOXP2 was confirmed to act as an EMT suppressor during metastasis in breast cancer cells, adding a new Fox transcription factor to the list of EMT regulators and also providing a further support of FOXA2's roles on EMT inhibition. Most studies revealed the predominant repressor role of FOXP2 on the transcription of target genes through interacting with co-repressors such as CtBP-1 (16,17). For example, FOXP2 was found to bind directly to the promoters of SRPX2 and uPAR, yielding a marked inhibition of their promoter activity (18). Both SRPX2 and uPAR were identified as the stimulators of EMT progression in multiple types of cancers including breast cancer (46)(47)(48)(49), providing a clue to explain the role of FOXP2 as an EMT repressor ( Figure 6C). On the other hand, FOXP2 was also shown to activate gene transcription (14,20), even though its mechanism was not well described. In this study, we found that FOXP2 could bind to certain promoters and stimulate the transcription of its target genes such as PHF2 and E-cadherin in breast cancer cells ( Figure 6C). FOXP2 was found to directly bind to and activate the promoter of PHF2 that stimulated the epithelial phenotype of cancer cells (see Figure 3). We also provided evidence that the transcriptional activation by FOXP2 could be mediated by FOXA2 at the promoter of E-cadherin. FOXP2, through its zinc-finger/leucine zipper motif (197-408aa), interacted with the DNA binding domain of FOXA2 (166-324aa) (see Figure 1) but not the FOXA2 N-terminal and C-terminal transcriptional activation domains that stimulated the transcription of target genes (50). The analysis of clinical samples showed that the FOXP2-improved survival in patients depended on the high expression of FOXA2 in breast cancer (see Figure 6B), implicating involvement of FOXA2 for FOXP2 performing its functions. Together, we proposed that FOXP2 could act as a transcriptional activator through its interaction with FOXA2 in breast cancer cells. Furthermore, FOXP2's zincfinger/leucine zipper motif was close to its transcriptional repressor region (422-426aa) that mediated the recruitment of its co-repressor CtBP-1 (16,51). We wondered whether FOXA2 binding to FOXP2's zinc-finger/leucine zipper motif might prevent the interaction between FOXP2 and CtBP-1 and consequently limit the transcription-repressing abilities of FOXP2. This assumption needs to be investigated in the future. The expression of FOXP2 itself was also regulated in cancer cells. In this study, we noticed that the expression of FOXP2 was down-regulated during the EGF-induced EMT process (see Figure 2) and up-regulated during the Cholera Toxin-induced MET process (see Figure 3). Even though dissecting the regulatory mechanisms of FOXP2 expression was out of the range of the current study, some published literatures provided clues for explaining the variation of FOXP2 levels during EMT or MET process. One of the FOXP2 down-regulation mechanisms during the EGF-induced EMT might be mediated by the key mesenchymal transcription factor TWIST (10), whose levels were stimulated by EGF treatment in cancer cells (52). TWIST was found to increase the levels of multiple microRNAs such as the miR-199a-214 cluster, which in turn converged on and repressed the expression of FOXP2 (26,32). This provided at least one possible mechanism of FOXP2 down-regulation in EGF-induced EMT progression in the cancer cells. The decrease of TWIST levels during the MET progression (38) might also explain the up-regulation of FOXP2 in MET of the Cholera Toxin-induced cancer cells. Furthermore, there was a bona fide binding site on the promoter of FOXP2 for the tumor suppressor p53 (25), whose expression increased in MET (53), implicating that the transcription of FOXP2 in MET might be directly stimulated by p53. Interestingly, PHF2, which was confirmed to be stimulated by FOXP2 in this study, could act as a coactivator of p53 via the demethylation of histone H3K9m3 at the promoters of p53 target genes (39), providing a possible positive feedback loop for the up-regulation of FOXP2 during the MET progression in the cancer cells. After all, detailed regulatory mechanisms of FOXP2 expression were worthy to be further studied.

DISCUSSION
FOXP2 expressed in multiple adult tissues (23,24) and might act as either a tumor-suppressor or a tumor-stimulator depending on the type of cancers studied (25), presenting a challenge for targeting FOXP2 as a universal target in cancer therapy. This study provided evidence that FOXP2 activated the transcription of its target genes in the maintenance of epithelial characteristics for breast cancer cells. The in vivo data showed that the FOXP2-knockdown resulted in the gain of mesenchymal traits in epithelial MCF-7 cells and elevated the migration capabilities of the cells in blood circulation. FOXP2 therefore might be identified as a suppressor of breast cancer metastasis. Interestingly, as a protein interacting with FOXA2, FOXP2 showed the highest correlation with FOXA2 in the basal subtype (see Figures 1H, I), which was the most lethal subtype with a high degree of metastasis ability associated with mesenchymal characteristics (54), implicating that the two The FOXP2-improved survival in breast cancer patients relied on the participation of FOXA2. The FOXA2 High subgroup (n=131) or the FOXA2 Low subgroup (n=131) contained either the top 1/3 or the bottom 1/3 of the collected BRCA data set (n=394), respectively. FOXP2-related survival of patients in the FOXA2 High and FOXA2 Low subgroups was fitted by the "survfit" function, and Kaplan-Meier curves were drawn by the "ggsurv" function in the R package "survival". (C) The roles of FOXP2 in epithelial-mesenchymal transition (EMT) of breast cancer cells. FOXP2 prevented EMT of breast cancer cells by regulating the transcription of multiple EMT-related genes: FOXP2 could bind to certain promoters and stimulate the transcription of genes such as PHF2 and Ecadherin. The transcriptional activation by FOXP2 could be mediated by FOXA2. FOXP2 could also repress the transcription of certain genes such as SRPX2 and uPAR, through recruiting co-repressors such as CtBP-1.
proteins played important roles together in this subtype of breast cancer. In previous studies, we confirmed that FOXA2 abolished the mesenchymal traits of basal-like MDA-MB-231 cells by stimulating E-cadherin and repressing ZEB2 through recruiting a transcriptional co-repressor TLE3 to ZEB2 promoter (11). In this study, we found that FOXP2 and FOXA2 together synergistically activated the transcription of E-cadherin in MDA-MB-231 cells and resulted in the cells losing their mesenchymal morphology. However, the overexpression of FOXP2 in MDA-MB-231 cells did not affected either the expression of ZEB2 or the FOXA2-mediated repression of ZEB2 promoter ( Figure S11), suggesting that various mechanisms existed for FOXP2 and FOXA2 to regulate the transcription of their different target genes. The future analysis of RNA-seq and ChIP-seq of the two factors compared in different subtype cells of breast cancer would provide a foundation to describe the detailed picture of FOXP2 and FOXA2 on regulating EMT progression and metastasis of breast cancer cells. Together, this study confirmed that in concert with FOXA2, FOXP2 acted as a tumor-suppressor through inhibiting EMT of breast cancer cells.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The human tissues were obtained and studied in strict adherence to the protocol approved by the Hunan University College of Biology Review Board. The animal study was reviewed and approved by Laboratory Animal Center of Hunan, China (protocol no. SYXK [Xiang] 2008-0001).

AUTHOR CONTRIBUTIONS
YXL: conceptualization, methodology, investigation, formal analysis, writing-original draft, writing-review, and editing. TC: conceptualization, methodology, investigation, resources, and formal analysis. MG: resources and investigation. YL: resources and investigation. QZ: resources. GT: methodology and data curation. LY: methodology, data curation, writing-original draft, and project administration. YT: conceptualization, formal analysis, supervision, writing-original draft, writing-review and editing, and project administration. All authors contributed to the article and approved the submitted version.