FOXC1 Regulates FGFR1 Isoform Switching to Promote Invasion Following TGFβ-Induced EMT

Epithelial-to-mesenchymal transition (EMT) is a biologically important process whereby epithelial cells alter their morphology and adopt properties of mesenchymal cells (1, 2). EMT was originally observed and described as a transient transdifferentiation where cells from an epithelium lose characteristic marks, such as E-cadherin (CDH1) expression as well as their epithelial sheet morphology and gain mesenchymal properties such as N-cadherin (CDH2) expression and increased migratory properties (1). During development, EMT is a key process in generating tissues during gastrulation, somitogenesis and neural crest cell formation (3). EMT has also been implicated in the progression of organ fibrosis and wound healing in adult tissues (4, 5). Finally, EMT events are thought to drive metastasis in a number of cancers, including breast basal cell carcinoma, hepatocellular carcinoma, and pancreatic tumors (6, 7).

A key component to EMT is the characteristic decrease in E-cadherin levels resulting in the loss of epithelial adherens junctions. In concert with the loss of E-cadherin, the progression of EMT is marked by an upregulation of N-cadherin, common to mesenchymal cells, and results in what is termed the "cadherin switch" (8). Epithelial cells alter their adherens junction composition through this cadherin switch, reorganize their cytoskeleton, and lose apical–basal polarity in favor of the front-rear polarity of mesenchymal cells (2). Cells that have upregulated levels of N-cadherin also upregulate the intermediate filament vimentin that is required for cytoskeletal rearrangement and adoption of the spindle-like cell morphology. Multiple transcription factors including Snail1 (encoded by the Snai1 gene), Twist1, and ZEB1 and 2 have been demonstrated to directly downregulate E-cadherin expression and induce EMT events (6, 9).

Another key factor in the initiation of EMT is TGFβ1 (2). These secreted factors are members of a larger growth factor family, which includes the well-identified morphogens TGFβ, bone morphogenetic proteins, and the activin/inhibin subfamily. These factors act physiologically as paracrine or autocrine signals and are known to contribute to embryonic development, tissue homeostasis, as well as tumor suppression and metastasis (2). TGFβ1 acts primarily through the binding to type II TGFβ receptors, which in turn phosphorylate and activate type I receptors. These activated type I receptors then initiate a signal transduction cascade resulting in the phosphorylation of SMAD2 or SMAD3 proteins. Once phosphorylated, SMAD2/3 protein can associate with SMAD4 and translocate to the nucleus and regulate gene expression. In response to TGFβ1–SMAD2/3 signaling, expression of many transcription factor genes that initiate EMT, such as Snail1 (Snai1) and Zeb1/2, is rapidly induced (2, 10).

The forkhead box transcription factor FOXC1 is required for the development and formation of tissues derived from neural crest and paraxial mesenchyme, including the anterior segment of the eye, the meninges, the axial skeleton, and craniofacial skeleton (11–13). As much of the formation of these tissues is driven by EMT events, it stands to reason that FOXC1 may function in EMT. More recently, elevated expression of FOXC1 was detected in basal cell–like breast carcinomas and other metastatic cancers that underwent EMT (14–17). Furthermore, reduction of FOXC1 expression in some cancer cell lines can lead to a reversal of a mesenchymal phenotype (15). However, in these experiments, the cells had undergone EMT events prior to manipulating FOXC1 levels. Thus, little information is known regarding the role for FOXC1 in response to physiologic induction of EMT events. To assess this question, we utilized the TGFβ1 induction of EMT in the mouse mammary epithelial cell line NMuMG to investigate the role of Foxc1 in EMT events. We found that expression of Foxc1 was indeed induced during EMT by TGFβ1 treatment though SMAD activity. However, loss of Foxc1 function through RNAi did not prevent the induction of EMT events in response to TGFβ1. Instead, we found that Foxc1 regulated the specificity of the mesenchymal cell phenotype. Loss of Foxc1 expression led to a reduction of Fgfr1-IIIc expression and promoted a less invasive myofibroblast mesenchymal cell phenotype. Finally, we demonstrated that reexpression of Fgfr1 was able to rescue the reduced invasive properties in Foxc1 knockdown cells. Together, our results suggest that Foxc1 participates in a maturation of a migratory mesenchymal phenotype by promoting FGFR receptor isoform switching.

Namru murine mammary gland NMuMG cells (ATCC) were cultured in DMEM and 10% FBS at 37°C and 5% carbon dioxide. The cells were subcultured twice weekly once they reached approximately 80% confluence in the flasks. No testing or authentication for these cell lines was performed. Cells were discarded after 10 to 15 passages following cryorecovery. For experiments using TGFβ1 stimulation, 2 × 10⁵ cells were plated in a 35-mm plate and incubated for 1 day. TGFβ1 (R&D Systems) was added to the media up to a final concentration of 5 ng/mL. shEGFP and shFOXC1 MB231 were generated and cultured as described previously (16).

RNA was harvested from NMuMG cells RNeasy Mini Kit (Qiagen) and quantified by spectrophotometry (NanoDrop 1000). Total RNA (500 ng) from each experimental group was used to generate cDNAs from a reverse transcription reaction with random primers (Qiagen). qRT-PCR was performed on a 1:25 dilution of the cDNA samples using a Kapa SYBR Fast qPCR Master Mix (Kapa Biosystems) and reactions run on a Bio-Rad CFX-96 touch thermocycler. Gapdh and β-actin (Actb), and hypoxanthine-guanine phosphoribosyltransferase (Hrpt) were used as internal controls. Primer sequences are listed in Supplementary Table S1. Statistical analysis was conducted using Bio-Rad CFX-Manager Software (Version 3.0 1215.0601).

Luciferase reporter vectors corresponding to the human and mouse Foxc1 promoters have been described previously (16, 18). The pcDNA3-Flag-Smad4 vector (19) was a gift from Aristidis Moustakas, Ludwig Institute for Cancer Research, Uppsala, Sweden (Addgene plasmid # 80888) and Smad3 vector was obtained from DNASU (HsCD00330086; ref. 20). To generate Fgfr1-reporter, a 490-bp DNA fragment corresponding to mm10; chr8:25,514,384-25,514,679 was generated as a gblock gene fragment (Integrative DNA Technologies) and cloned into pGL4.23-luc2/minP vector (Promega). An additional gblock fragment was synthesized whereby the putative FOXC1-binding site was mutated on the basis of binding sequence preferences described previously (21). NMuMG cells were transfected with Mirus TransIT LT1 (3 μL Mirus:1 μg DNA). Briefly, 4 × 10⁴ cells were plated onto a single well of a 24-well tissue culture plate. Each well was transfected with 50 ng of luciferase reporter vector, 150 ng of each expression vector and 1 ng of pRL-TK Renilla control. Each transfection was performed in triplicate, and each experiment was repeated at least three times.

Chromatin immunoprecipitation (ChIP) was performed as described previously (18, 22, 23) with the following modifications. NMuMG cells were treated with TGFβ1 for 24 hours. Chromatin was sheared on ice for 15 cycles of 30 seconds at 30% intensity on a Branson Sonifier with 1-minute intervals between sonication cycles. Sheared chromatin was then incubated overnight with 5 μg, anti-SMAD3 (9520, Cell Signaling Technology), anti-FOXC1 (C18, sc-21396 Lot # J0206; Santa Cruz Biotechnology), or IgG. Primers to amplify Foxc1-regulatory regions have been described previously (18). The mouse Fgfr1 upstream regulatory region was amplified using the following primers: FGFR1 ChIP 1F 5′-TGT CCT CCG TCT CCG AGA AT-3′; FGFR1 ChIP R-5′-GAG GGA GGG GCA GAA TCT TG-3′. Primers targeting the Fgfr1 coding region (Supplementary Table S1) were used as a negative control. ChIP assays were performed from three independent chromatin preparations.

Stable shRNA cells lines were created by transducing NMuMG cells with lentiviral vectors containing shRNAs targeting mouse Foxc1 or EGFP as described previously (23). The Foxc1 shRNA (TRCN0000085449) and the EGFP shRNA were obtained from the RNAi Consortium and have been demonstrated to reduce Foxc1 levels in mouse embryonic stem cells (24). To establish stable cell lines, NMuMG cells were selected with 2 μg/mL of puromycin for 2 weeks. Approximately 100 colonies were pooled and monitored for Foxc1 expression. Foxc1 shRNA viral transduction was performed with two separate lentiviral preparations to generate two independent Foxc1 knockdown cells lines (shFOXC1 v1 and shFOXC1 v2). Both lines behaved in a similar manner and data presented are indicative of shFoxc1 v1 cells. A pool of four small interfering double-stranded RNAs with sequences complimentary to the Foxc1 transcript were obtained from Dharmacon (siGenome Foxc1). Cells were seeded at a density of 2 × 10⁵ cells per well in a 6-well plate. siRNA for Foxc1 and control nonspecific siRNAs were transfected 24 hours later with Dharmafect-1 transfection reagent, and TGFβ1 (5 ng/mL) was added to the cells 24 hours posttransfection. RNA was harvested for qRT-PCR analysis 24 hours after TGFβ1 treatment.

Cells were grown in TGFβ1 for 2 days, and protein was extracted for immunoblot analysis as described previously (23, 25). Membranes were incubated with goat-anti FOXC1 at a concentration 1:100 overnight at 4°C or with mouse anti-β-tubulin (G8, sc-55529, Lot # E2412; Santa Cruz Biotechnology) at 1:5,000. Blots were visualized on a LI-COR Odyssey imager using IR-DYE700 donkey-anti-goat IgG and IR-DYE800 donkey-anti-mouse IgG secondary antibodies.

NMuMG (shFOXC1 and shEGFP) cells were cultured at 2 × 10⁵ per well on sterile coverslip. The next day, cells were treated with TGFβ1 (5 ng/mL) for 24 hours. Cells were fixed for 20 minutes in 4% formaldehyde, followed by washes in PBS + 0.05% Triton X-100 (PBS-X) and blocked with 5% BSA for 15 minutes. Coverslips were incubated with the following primary antibody for 1 hour: E-cadherin [Cell Signaling Technology (24E10); 1:100]. Cells were washed with PBS-X and then incubated with secondary antibodies. DAPI was added to visualize nuclei and phalloidin-488 to detect polymerized actin. Coverslips were mounted onto glass slides with Prolong Gold mounting medium (Thermo Fisher Scientific). For cell viability assays, shEGFP or shFOXC1 cells grown on coverslips were either left untreated or treated with TGFβ1 (5 ng/mL) and Fgf2 (30 ng/mL) and incubated with ReadyProbes Cell Viability Imaging Kit (Molecular Probes) NucBlue Live and NucGreen Dead reagents before mounting onto slides. Slides were visualized at room temperature on Leica DRME fluorescent microscope using a 40× objective. Images were captured and pseudo-colored using Northern Eclipse Imaging software. Micrographs were prepared using CorelDraw 16.

RNA was isolated from shFOXC1 or shEGFP treated with TGFβ1 (5 ng/mL) for 24 hours from three independent treatment procedures. Fifteen micrograms of RNA was used for library construction and sequencing on an Illumina HiSeq2500. Sequencing and bioinformatic analysis was performed by Otogentics. Datasets were mapped against mouse reference genome (mm10) with tophat, and differential expression analysis was conducted using cufflinks and cuffdiff. An FDR of 0.01 was used to identify differences in gene expression between shFOXC1 and shEGFP samples.

HA-tagged FOXC1 was cloned into the pLNCX2 retroviral vector. The myristoylated FGFR1 vector pWZL-Neo Myr Flag FGFR1 (20) was a gift from William Hahn and Jean Zhao, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, (Addgene plasmid # 20486). Retrovirus was generated as described previously (23). After viral transduction, cells were selected with G418 (500 μg/mL). Over 100 colonies from two 60-mm plates were pooled for further analysis.

Cells were seeded in a 6-well plate at 2 × 10⁵ cells per well and treated with 5 ng/mL TGFβ1 and 30 ng/mL FGF2 (R&D Systems) along with 100 μg/mL heparin for 48 hours. Cells that did not receive growth factor treatment were used as controls. After treatment, cells were resuspended in serum-free media, and 2.5 × 10⁴ cells were placed into growth factor–reduced Matrigel invasion chambers (Corning). Each insert was placed into a 24-well plate containing 0.75 mL DMEM + 10% FBS. For FGFR inhibition cell invasion experiments, cells were treated with DMSO or PD161570 (40 nmol/L) for 24 hours before plating onto invasion chambers. Cells were incubated for 24 hours. Cells were removed from the topside of the insert with a cotton swab, and the membranes were fixed in 4% paraformaldehyde and stained with hematoxylin and eosin. Each transwell assay experiment was performed in triplicates per experimental variable and was performed twice.

For ChIP, luciferase reporter assays and invasion assays, one-way ANOVA was performed with SigmaPlot (Systat Software Inc.) Version 13 build 13.0.0.83. For qRT-PCR, statistical calculations were performed with Bio-Rad CFX manager Version 3.0 build 3.0.125.0601.

To determine whether Foxc1 expression was regulated during EMT induction, we utilized the well-established NMuMG cell model. These epithelial cells will undergo a transition to a mesenchymal phenotype in response to the EMT inducer, TGFβ1 (10). We detected changes in cytoskeletal architecture characteristic of EMT in NMuMG cells after treatment with TGFβ1 (5 ng/mL) for 24 hours (Fig. 1A). Foxc1 mRNA was elevated by 8-fold after 48 hours of induction (Fig. 1B). We next conducted a time course experiment to determine how soon following TGFβ1-induced EMT, Foxc1 mRNA levels rose. Expression of Foxc1 mRNA gradually increased and reached maximum levels 48 hours following EMT induction (Fig. 1C). In contrast, Snai1 mRNA levels were rapidly elevated 4 hours after TGFβ1 treatment. Expression of the epithelial marker E-cadherin declined over time following TGFβ1 treatment (Fig. 1C).

Next, we wished to determine whether the expression of Foxc1 was regulated by TGFβ1-dependent SMAD pathways. We utilized Foxc1 luciferase reporters containing the human (∼890 bp upstream of the transcription start site) (16) and mouse (∼1,100 bp upstream; ref. 18) promoter sequences and examined whether their activity was stimulated by TGFβ1 treatment. We observed that both the mouse and human promoters were activated by 24-hour TGFβ1 treatment (Fig. 1D). Changes in gene expression in response to TGFβ1 are executed through a specific subset of SMAD transcription factors (SMAD 2 or 3) specific to this cytokine along with a common SMAD (SMAD4). We examined the mouse promoter sequence for putative TGFβ1–SMAD binding sites using the CIS-BP database (26). Five SMAD3-binding sites were located along the length of 1,200-bp mouse Foxc1 reporter (Fig. 1E). We utilized previously described 401 bp and 140 bp mouse Foxc1 promoter deletions (18) that contain two and one SMAD3-binding site, respectively (Fig. 1E). We tested for direct promoter regulation by SMAD3 in these reporters by cotransfecting SMAD3 and 4 expression vectors. As indicated in Fig. 1E, all Foxc1 luciferase reporters demonstrated activation in response to the SMAD3 and 4. However, the extent of the activation was dependent on the number of SMAD3-binding sites, with five sites in the 1,228 bp demonstrating the highest levels of activation. To confirm that TGFβ1-regulated SMADs contributed to binding of the Foxc1 promoter, we performed ChIP assays. We utilized primers targeting putative SMAD3-regulatory elements located approximately 800 bp from transcription start site (24). We found that SMAD3 was able to bind to this region in NMuMG cells treated with TGFβ1 (Fig. 1F).

Given the increase in Foxc1 mRNA levels following TGFβ1-induced EMT, we asked whether decreasing Foxc1 levels would prevent EMT induction. To test this idea, we transiently transfected NMuMG cells with Foxc1 siRNA or stably transduced cells with Foxc1 shRNAs resulting in a 50% and 80% reduction of Foxc1 mRNA levels, respectively (Supplementary Fig. S1A; Fig. 2A). Very little FOXC1 protein was detected in the stably expressing Foxc1 shRNA cells (Fig. 2B). As a control, we transduced cells with a vector expressing shRNA targeting EGFP. When Foxc1 knockdown cells were treated with TGFβ1 for 24 hours, we detected no differences in expression profiles characteristic of EMT. Levels of Snail1, vimentin, and N-cadherin mRNA were increased in both control (shEGFP) and Foxc1 (shFOXC1) knockdown cells following TGFβ1 treatment (Fig. 2C and D; Supplementary Figs. S1 and S2). E-cadherin mRNA levels decreased following TGFβ1 treatment in both the shEGFP and shFOXC1 NMuMG knockdown cells. We examined whether the morphologic phenotype characteristic of EMT was affected in response to reduced Foxc1 levels. In control and Foxc1 knockdown cells, E-cadherin protein was distributed at the cell membrane in a manner characteristic of epithelial cells. We determined the distribution of actin though phaloidin-488 staining. In untreated cells, the actin distribution overlapped that of the E-cadherin immunofluorescence in both control and Foxc1 knockdown cells (Fig. 2E; Supplementary Fig. S3). Upon TGFβ1 treatment, the intensity of E-cadherin staining was greatly reduced in both control and Foxc1 knockdown cells. Moreover, we detected the redistribution of actin in to stress fibers characteristic of EMT following TGFβ1 treatment in both control and Foxc1 knockdown cells (Fig. 2E; Supplementary Fig. S3). Together, these data suggest that although Foxc1 mRNA expression is elevated following TGFβ-induced EMT, reducing Foxc1 levels appears to have no effect on the induction of this cellular event.

Given that Foxc1 expression was induced in response to TGFβ1-induced EMT and the elevated expression in mesenchymal cells suggests that Foxc1 may play other roles in EMT. We sought to determine what genes were differentially expressed in shFOXC1 versus shEGFP cells treated with TGFβ1. Three independent RNA samples were analyzed by RNA sequencing (RNA-seq). From this analysis, 660 genes were differentially regulated (451 downregulated; 209 upregulated) with an FDR of 0.01 (Supplementary Tables S2 and S3). We next validated whether shFOXC1 knockdown did indeed lead to changes in gene expression by qRT-PCR from independent biological replicates treated with and without TGFβ1. All genes we analyzed were differentially expressed in shFOXC1 cells treated with TGFβ1 compared with control shEGFP cells (Fig. 3A and data not shown). As the RNA-seq experiments only compared expression between shEGFP and shFOXC1 cells treated with TGFβ1, we did not know whether the expression of the genes themselves was altered in response to TGFβ1. Many of the genes that we identified (Lefty1, Jam2, Tead2, and Slc4a11) were induced by TGFβ1 and that activation was reduced in shFOXC1 cells (Fig. 3A). Of note, we did not detect any change in expression of known inducers or markers of EMT (such as Snail1, Slug, Zeb, E-cadherin, or N-cadherin), nor did we identify any changes in Foxc2 expression (a known regulator of EMT; ref. 27) in our RNA-seq analysis. Furthermore, Gene Ontology analysis did not reveal any enrichment of genes implicated in EMT events (data not shown).

From the RNA-seq data, we noted a decrease in the expression of Fgfr1 mRNA and increase in the expression of α-smooth muscle actin (aSMA; Acta) in our TGFβ1-treated shFOXC1 cells (Supplementary Tables S2 and S3). It has been demonstrated that TGFβ1 treatment can result in an FGFR receptor switch whereby expression of Fgfr2-IIIb isoform in epithelial is downregulated and Fgfr1-IIIc expression is elevated as cells transition to mesenchymal cells (28). Moreover, this isoform receptor switch is required for mesenchymal cells to further differentiate into active migratory mesenchymal cells. Indeed, we observed that the increase in Fgfr1 expression was reduced in TGFβ1-treated shFOXC1 cells compared with shEGFP control cells, while levels of αSMA were elevated in the shFOXC1 in response to TGFβ1 (Fig. 3B). We also noted that Fgfr2 expression was downregulated by TGFβ1 in both shEGFP and shFOXC1 cells (Fig. 3B). Using primer specific to Fgfr1-IIIc isoform, we confirmed that expression of the IIIc isoform was induced in our assays (Fig. 3B).

Next, we asked whether FOXC1 was directly regulating expression of the Fgfr1 promoter. As indicated in Fig. 4A, we identified a region approximately 4 kb upstream of the first exon of the mouse Fgfr1 gene that was conserved with the human FGFR1 gene and contained a putative FOXC1 DNA-binding site (GTAAATAA; refs. 21, 29). Using ChIP analysis, we confirmed that FOXC1 binding to this site was enriched compared with a nonregulatory region in the Fgfr1 coding sequence (Fig. 4B and C). Next, we created an Fgfr1 luciferase reporter consisting of this regulatory region along with the wild-type and mutant Foxc1-binding sites (Fig. 4D). We found the expression of Foxc1 alone was sufficient to activate luciferase expression in the wild-type Fgfr1 reporter but not the mutant (Fig. 4D). Together, these data indicate that Foxc1 acts to directly regulate expression of Fgfr1 in response to TGFβ1-induced EMT events.

As reduction of Foxc1 levels resulted in decreased Fgfr1 expression, we wished to determine whether increased Foxc1 expression resulted in an increase in Fgfr1 expression as well. We generated stably transfected NMuMG cells expressing HA-tagged FOXC1 (HA-FOXC1; Fig. 5A). We found that elevated expression was sufficient to increase expression of Fgfr1-IIIc (Fig. 5B). This elevated Foxc1 expression led to a concomitant decrease in Fgfr2 mRNA levels. However, increasing levels of Foxc1 in NMuMG cells did not lead to any changes in the expression of Snai1or N-cadherin in the absence or presence of TGFβ1, suggesting that Foxc1 was not sufficient to induce EMT, nor able to enhance TGFβ1-mediated EMT (Fig. 5B). Levels of αSMA were also unaffected when Foxc1 levels were elevated. Finally treating HA-FOXC1 cells with TGFβ1 for 24 hours resulted in a heightened increase in Fgfr1-IIIc mRNA expression and a marked reduction in Fgfr2 mRNA levels (Fig. 5B).

The Fgfr2-IIIb to Fgfr1-IIIc receptor switching that occurs during EMT alters the cell responsiveness to the FGF2 ligand and promotes the differentiation of mesenchymal cells to an invasive fibroblast state (28). We wished to test whether reduction in Foxc1 levels promoted the myofibroblast phenotype and thus a less migratory cell type in response to FGF2 treatment. Untreated shEGFP and shFOXC1 cells were treated with TGFβ1 and FGF2 and then seeded onto transwell membranes for cell invasion assays. As indicated in Fig. 6A, very few untreated shEGFP or shFOXC1 cells migrated through the membrane. TGFβ1 and FGF2-treated shEGFP cells displayed a robust invasive phenotype; however, TGFβ1 and FGF2-treated shFOXC1 cells demonstrated impaired invasion capacity (Fig. 6A). Thus, reducing Foxc1 levels resulted in a marked decrease in TGFβ–FGF2 induced invasive potential in NMuMG cells. Using cell viability assays, we determined that the reduced invasion of the shFOXC1 cells was not due to a reduction in the number of viable cells (Fig. 6B). Next, we sought to determine whether restoring Fgfr1-IIIc expression in shFOXC1 cells could rescue the migratory phenotype we observed in these cells. We created stable shEGFP and shFOXC1 cell lines expressing myristoylated human FGFR1 (Fig. 6C) (20). When these cells were tested in invasion assays, we noted that expressing FGFR1 in shFOXC1 cells was capable of increasing the invasive capacity of these cells (Fig. 6D). Together, these results demonstrate that increasing FGFR1 expression can overcome the inhibition of migration in the absence of Foxc1. Finally, we tested whether the invasive migratory phenotype is enhanced in FOXC1-overexpressing cells. Treatment of HA-FOXC1 cells with TGFβ1 and FGF2 led to an increase in migratory cells that was reduced when treated with the FGFR1 inhibitor PD161570 (Fig. 6E). Together, these data support the notion that FOXC1 regulates invasive cell migration of NMuMG cells in an FGFR1-dependent manner.

EMT mechanisms are thought to contribute in part the highly migratory properties of some cancer cells. It has been demonstrated that levels of FOXC1 mRNA and protein are elevated in breast cells that display a migratory mesenchymal phenotype (15, 16, 30). MCF7 cells are derived from luminal breast carcinoma and possess a noninvasive epithelial phenotype. In contrast, MB231 cells display hallmarks of a mesenchymal cell, including elevated N-cadherin expression and an increased invasive capability. As demonstrated in Fig. 7A, FOXC1 expression is elevated in MB231 compared with MCF7 cells. In addition, levels of FGFR1 are also elevated in the MB231 cells. We next determined whether FGFR1 levels were reduced in shFOXC1-expressing MB231 cells. As indicated in Fig. 7B, expression of FGFR1 mRNA was diminished when FOXC1 levels were reduced. We did observe a decrease in N-cadherin mRNA levels but no changes in Snail1 and vimentin levels (Fig. 7B). Finally, we sought to determine whether the increased migratory and invasive phenotype of MB231 cells would be affected by use of the FGFR1 inhibitor PD161570 (31). We observe that control shEGFP cells possess an inherent active migratory property compared with shFOXC1 cells (Fig. 7C). We also find that treatment with PD161570 reduces the invasive capacity of shEGFP MB231 cells but has little effect on the noninvasive shFOXC1 MB231 cells.

Elevated expression of FOXC1 has been attributed to regulation of EMT in many human cancer cells (14–17). However, the exact role for FOXC1 in EMT events has yet to be determined. In many cases, it has been reported that FOXC1 induces EMT possibly through the regulation of SNAI1 expression (32). We investigated the role of Foxc1 in the TGFβ1-mediated EMT in nontransformed mouse NMuMG cells. Although we did observe an activation of Foxc1 mRNA expression in response to TGFβ1-induced EMT, this increase in Foxc1 expression occurred 12 to 24 hours posttreatment, well after the induction Snai1 mRNA (Fig. 1B). We also observed that EMT events occurred when Foxc1 levels were reduced by siRNA or shRNA following treatment with TGFβ1. Levels of Snail1, N-cadherin, and vimentin mRNA were still upregulated in response to TGFβ1 in Foxc1 knockdown cells. As well, E-cadherin and Fgfr2 expressions were concomitantly downregulated in TGFβ1 in Foxc1 knockdown cells, indicative of a loss of the epithelial phenotype. The actin cytoskeleton reorganized from a cortical distribution in untreated epithelial cells to form actin stress fibers characteristic of EMT events. Finally, we also determined that overexpression of FOXC1 was not sufficient to drive EMT events, as HA-FOXC1 cells did not ectopically induce the expression of mesenchymal markers (Fig. 5B), and these cells maintained epithelial cell morphology. Together, these findings suggest that in mouse epithelial cells, FOXC1 does not function in the initiation of EMT events but may play other roles in maintaining mesenchymal cell properties.

We determined that Foxc1 expression was activated during TGFβ1-mediated EMT events through SMAD proteins binding to the Foxc1 promoter. Foxc1 mRNA was induced in a time-dependent manner, with appreciable expression detected after 12- to 24-hour TGFβ1. Furthermore, we demonstrated the expression of TGFβ1-regulated SMAD proteins was sufficient to activate expression of a mouse Foxc1 promoter reporter vector. Through ChIP analysis, we were able to demonstrate binding of SMAD3 to a region in the Foxc1 promoter (∼800 bp upstream of the transcription start site), although there are likely additional regions under SMAD2/3 regulation. Regulation of FOXC1 expression has been documented in additional experimental systems. Zhou and colleagues demonstrated that FOXC1 expression was activated in response to TGFβ1-induced growth arrest in SKOV3 ovarian cancer cells (33). In contrast, Hoshino and colleagues reported the Foxc1 expression was repressed in JygMC(A) and JygMC(B) breast cancer cells following TGFβ1 treatment (34). It is important to note that the cellular conditions described in these two reports did not promote EMT; rather, treatment with TGFβ1 resulted in growth arrest and protection apoptosis. This induction of Foxc1 expression appears to be specific to EMT induction rather than in epithelial cells versus mesenchymal cells, as we do not observe any increases in Foxc1 mRNA levels when mesenchymal cell such as mouse 10T1/2 fibroblasts are treated with TGFβ1 (Hopkins and Berry, unpublished observations). Expression of FOXC1 may be differentially regulated by TGFβ1 in specific cell contexts. We recently reported that expression of Foxc1 and promoter activity was activated in response to bone morphogenetic protein (BMP) signaling in early induction of osteoblast differentiation but was repressed by the same factor at later stages of differentiation (18). Thus, in different cellular contexts, additional factors act in concert with TGFβ1/BMP signaling networks to positively or negatively regulate Foxc1 expression.

Our results indicating that FOXC1 is not required for the induction of EMT events in response to TGFβ1 treatment in contrast to other findings suggest a role for FOXC1 as an initiator of this process. These differences may reflect the cell types used in each study. Our experiments focused on the induction of EMT through physiologic means, notably TGFβ1 treatment, in untransformed epithelial cells. Whereas many of other studies reduced FOXC1 by siRNA in cancer cells lines that had undergone EMT and demonstrated a loss of expression of mesenchymal markers, our data indicate that FOXC1 may function in the specification of the mesenchymal cell type but does rule out a role in the maintenance of mesenchymal properties. The reduction of FOXC1 expression in cancer cells may lead a reversion back to the epithelial state as low FOXC1 levels may be insufficient to maintain the mesenchymal cell phenotype. For example, we note that reducing FOXC1 expression levels in MB231 cells led to a modest reduction of N-cadherin expression and reduced invasion capacity. We did not observe any changes in E-cadherin levels (data not shown). Alternatively, the reversion of the mesenchymal phenotype observed when FOXC1 expression is knocked down in cancer cell lines may result from altering the abundance of FGFR2-IIIb to FGFR1-IIIc ratios to favor the epithelial state. More notably, however, we did not observe any induction of EMT marker gene expression when Foxc1 levels were elevated in NMuMG cells. Finally, enforced expression of FOXC1 in human MCF10A breast epithelial cells was not sufficient to activate expression of mesenchymal cells markers such as SNAIL1, TWIST and vimentin, nor sufficient to downregulate E-cadherin levels (35).

To ascertain the functional roles for Foxc1 in EMT events, changes in gene expression in cells with reduced Foxc1 levels treated with TGFβ1 were compared with cells with unaltered Foxc1 expression. Using an unbiased screen, we identified a number of genes that displayed differential expression when Foxc1 levels were reduced. In the RNA-seq analyses we compared expression between shEGFP and shFOXC1 cells treated with TGFβ1, and identified over 600 genes with either reduced or elevated expression in shFOXC1 compared with the shEGFP control (Supplementary Tables S2 and S3). We identified that Fgfr1, Tead2, Jam2, and Slc4a11 expression to be upregulated in response to TGFβ1 treatment, and this induction was attenuated when Foxc1 levels were reduced. Given that Foxc1 expression was also activated by TGFβ1 raises the notion that these genes may be under the regulatory control of FOXC1 or may participate in common pathways. Tead2 regulates the localization of Hippo pathway factors TAZ and YAP to promote TGFβ1-induced EMT in NMuMG cells (36). Jam2 encodes for a junction adhesion molecule and is located between tight junctions of endothelial cells (37). A role for this gene in EMT has yet to be elucidated.

Most notably from our RNA-seq experiments, we observed that expression of Fgfr1 levels was reduced and expression of αSMA was elevated in shFOXC1 cells treated with TGFβ1. When cells undergo TGFβ1-induced EMT, Fgfr genes undergo a receptor switch (28). In the epithelial cell state, levels of Fgfr2-IIIb are elevated compared with Fgfr1-IIIc, and subsequently, this ratio reverses with Fgfr1-IIIc being the predominant receptor type expressed in response to EMT induction. Each FGFR has varying affinity for distinct FGF ligands (38–40). In particular, FGFR1-IIIc binds to FGF2 at higher affinity than FGFR2-IIIb; thus, in response to FGF2 stimulation, mesenchymal cells can further differentiate into either myofibroblasts or activated fibroblasts (28). Myofibroblasts are characterized by elevated expression of αSMA and a reduced migratory phenotype, whereas activated fibroblasts exhibit aggressive migratory and invasive properties and low levels of αSMA expression. Both cell types, however, display elevated expression of mesenchymal markers, such as Snail1, vimentin, and N-cadherin (28). Our data suggest that FOXC1 regulates this receptor switch from FGFR2-IIIb to FGFR1-IIIc through the activation of Fgfr1-IIIc expression in response to TGFβ1. In Foxc1 knockdown NMuMG cells, we observed reduced expression of Fgfr1-IIIc and elevated levels of αSMA when cells were treated with TGFβ1. Furthermore, these cells displayed a reduced three-dimensional migratory phenotype in response to FGF2 treatment that could be rescued by reexpression of FGFR1. We observed elevated Fgfr2 mRNA expression in shFOXC1 NMuMG cells and reduced Fgfr2-IIIb mRNA level expression in Foxc1-overexpressing cells, suggesting that Foxc1 may participate in the negative regulation of Fgfr2-IIIb expression in EMT. It should be noted that the downregulation of Fgfr2-IIIb levels in response to TGFβ1 treatment still occurs in shFOXC1 NMuMG cells, suggesting that additional factors participate in this process. An examination of upstream regulatory elements in Fgfr2 gene did not reveal putative FOXC1-binding sites, suggesting an indirect regulatory control. We also detected the binding to and the activation of the Fgfr1 promoter by FOXC1 through ChIP and luciferase reporter assays. Overexpression of Foxc1 in NMuMG cells was sufficient to induce expression of Fgfr1-IIIc and increase migration in transwell assays. Together from our results, we propose a function for Foxc1 in the activation of Fgfr1-IIIc expression to promote in mesenchymal cell specification toward a highly invasive and migratory activated fibroblast-like cell type during EMT events.

Elevated expression levels of FGFR1 are characteristic of many aggressive metastatic cancers that have undergone EMT (41–43). Basal-like cell breast cancers display high levels of FGFR1 along with elevated FOXC1 expression (44, 45). Induced FGFR1 overexpression can promote EMT phenotype in prostate adenocarcinoma cells (46). In addition to cancer metastasis, EMT is an important contributor to tissue fibrosis. In kidney cells, activation of FGF2 signaling can promote EMT and fibrogenesis (47). Furthermore, therapeutic strategies to target FGFR1 activities look promising in the treatment of certain types of cancers, and the inhibition of FGFR1 activity can promote the reversion of the mesenchymal phenotype back to the epithelial cell state (48, 49).

In summary, the elevated expression of FOXC1 in many cells that have undergone EMT has implicated FOXC1 in this cellular process. However, a clear role for FOXC1 in regulating EMT has yet to be determined. We provide evidence that FOXC1 is not required for initiation of EMT events but rather participates in the specification of mesenchymal cell phenotype through regulation of FGF receptor switching from FGFR2-IIIb to FGFR1-IIIc in response to TGFβ1-induced EMT. Elevated expression of FOXC1 has been proposed as a prognostic biomarker in many aggressive cancers. Many of these cancers have elevated FGFR1 levels themselves; thus, treatments based on FGFR1 inhibitors may prove to be beneficial in FOXC1-elevated cancers.

No potential conflicts of interest were disclosed.

Conception and design: F.B. Berry

Development of methodology: F.B. Berry

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Hopkins, M.L. Coatham, F.B. Berry

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Hopkins, M.L. Coatham, F.B. Berry

Writing, review, and/or revision of the manuscript: A. Hopkins, M.L. Coatham, F.B. Berry

The authors thank R. Lavy for critically reading this manuscript. M.L. Coatham is a recipient of a studentship from the Graduate Program in Maternal and Child Health. F.B. Berry holds the Shriners Hospital for Children Endowed Chair in Pediatric Scoliosis Research.

Funding for this research was provided by grants from the Canadian Institutes of Health Research (MOP 114921; awarded to F.B. Berry) and from the Edmonton Civic Employees Charitable Assistance Fund.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.