The Transcription Factor Elf3 Is Essential for a Successful Mesenchymal to Epithelial Transition

The epithelial to mesenchymal transition (EMT) and the mesenchymal to epithelial transition (MET) are two critical biological processes that are involved in both physiological events such as embryogenesis and development and also pathological events such as tumorigenesis. They present with dramatic changes in cellular morphology and gene expression exhibiting acute changes in E-cadherin expression. Despite the comprehensive understanding of EMT, the regulation of MET is far from being understood. To find novel regulators of MET, we hypothesized that such factors would correlate with Cdh1 expression. Bioinformatics examination of several expression profiles suggested Elf3 as a strong candidate. Depletion of Elf3 at the onset of MET severely impaired the progression to the epithelial state. This MET defect was explained, in part, by the absence of E-cadherin at the plasma membrane. Moreover, during MET, ELF3 interacts with the Grhl3 promoter and activates its expression. Our findings present novel insights into the regulation of MET and reveal ELF3 as an indispensable guardian of the epithelial state. A better understanding of MET will, eventually, lead to better management of metastatic cancers.


Introduction
The epithelial to mesenchymal transition (EMT) and the mesenchymal to epithelial transition (MET) has been known to scientists for decades. They are crucial for a variety of biological processes such as embryogenesis, development, and wound healing [1][2][3], and their aberrant regulation has deleterious outcomes [1,[4][5][6]. For example, tumor cells of epithelial origin can benefit from an EMT event and lose their adherent and polarized morphology and become motile and unpolarized [1,7]. These cells can now relocate and colonize sites in distant organs. To accomplish a metastatic status, relocated tumor cells need to undergo a MET event, after which, they would regain the polarized epithelial-like state, proliferate, and form secondary tumors [1,6,8].
Alongside the global changes in gene expression and cytoskeletal rearrangements, switching the expression of the adhesion molecules E-cadherin and N-cadherin is a hallmark of EMT and MET events [3,6,[8][9][10]. The downregulation of E-cadherin is crucial for a successful EMT [10,11], which is accomplished by a group of transcription factors including but not limited to, Snai1, Snai2, Zeb1, and Zeb2 [3,12], collectively dubbed 'The EMT Inducers'. These transcription factors bind to several E-boxes in the promoter of Cdh1 and cause the repression of E-cadherin expression [12][13][14][15]. The initiation of MET has been mostly explained as a result of the reduction of the levels of the EMT inducers [15,16]; however, MET is not merely the reverse process of EMT and is a fundamental program as well [2,17]. It involves a series of dynamic and tightly regulated events by either transcriptional and posttranscriptional mechanisms, in the center of which is the initiation and maintenance of Cdh1 expression and the presence of E-cadherin at the plasma membrane [18], which are essential steps for proper entry to MET [10]. MET is primarily described as a loss of the expression of transcription factors such as Snai1/2 and Zeb1/2 and gaining epithelial-like features via the initiation of Cdh1 and Epcam expression. Previous studies have shown that E-cadherin internalization and trafficking to and from the plasma membrane is crucial for maintaining epithelial dynamics [18,19] and controlling protein stability at the cell surface and cellular levels of the protein [15].
The hallmarks of transitions between epithelial and mesenchymal states include changes in the cadherin repertoire [20][21][22]. The Cdh1 promoter alone, including all known E-box elements, is insufficient to confer a strict cell-type specificity [23]. In contrast, intron 2 carries sufficient information for proper E-cadherin expression [24]. Although the second intron has emerged as an essential regulator of Cdh1 expression [23][24][25], very little is known about the molecular determinants controlling the transcriptional activity of the locus. Besides Klf4 [26,27], Grainy head-like 2 (Grhl2), a mammalian homolog of Drosophila grainy head, was shown to control epithelial differentiation in several tissues [28][29][30] and during uretic bud formation by regulating E-cadherin via binding to an element in intron 2 [31].
Moreover, the GRHL3 transcription factor plays a fundamental role during the initiation of MET [25]. It activates the expression of Hnf4α, and together they cooperate to activate the expression of Cdh1 through binding to enhancers in the second intron. Despite the abundance of publications on the regulation of Cdh1 expression, there still exists limited knowledge on the transcription factors involved in the regulation of Cdh1 expression. It is then plausible to assume that regulators of Cdh1 transcription are de facto regulators of MET; thus, a better understanding of how Cdh1 is transcriptionally regulated would shed light on the regulation of MET.
To gain insights into this, we searched for novel regulators of MET and hypothesized that regulators of MET could be regulators of Cdh1 transcription or at least have a strong correlation with its expression. Using an in silico approach, we first generated an epithelial-specific set of transcription factors with high correlation to Cdh1 expression. Moreover, we used the normal murine mammary gland epithelial cells (NMuMG) that undergo a reversible EMT by transforming growth factor β (TGFβ) treatment and identified Elf3 as a compelling regulator of MET and E-cadherin localization to the plasma membrane. Moreover, we provide evidence showing that ELF3 is a regulator of the transcription factor Grhl3.

Microarray Data Analysis and Correlation Study
Raw data (CEL files) of four gene expression datasets based on the GPL6246 platform (GSE130548, GSE77551, GSE55711, and GSE55072) were downloaded from the Gene Expression Omnibus database. Samples with distinct epithelial and mesenchymal states were selected for analysis. Datasets and samples included in the analysis are shown in Table S1. Data were RMA normalized using Bioconductor package oligo [32]. Normalized data were merged and transformed into z-scores using scale function in R [33]. The z-score transformation was performed both on the sample and on gene level [34]. Merged z-scores profile was used to evaluate spearman correlation between Cdh1 and all other probe sets present in the expression profile. Differential expression analysis between epithelial and mesenchymal samples was performed using the Limma package [35] to retrieve log fold changes and p-values. Gene to probe set annotation was performed using the annotate package [33] and mogene10sttranscriptcluster.db annotation library [36]. List of mouse transcription factors was retrieved from Animal TFDB3.0 [37]. Cdh1 correlation and differential expression profile of transcription factors were extracted from the performed analysis. The Ggplot2 package was used to draw the plots [38].

siRNA Knockdown and mRNA Expression Analysis
siRNAs against Elf3 were purchased from Qiagen for siElf3_2 and siEhf_1-4, and from Dharmacon for siElf3_10 and siElf3_12; sequences are available in Table S2. Cells were transfected with 50 nM siRNA using the Lipofectamine RNAi Max (13778150, Invitrogen, Waltham, MA USA) and collected 72 h later. Total RNA was isolated using the Nucleospin RNA II kit (740955.50, Macherey-Nagel, Düren, Germany) as recommended by the manufacturer. Total RNA (1 µg) was converted to cDNA using the Maxima First Strand cDNA Synthesis Kit (K1642, Thermo Scientific, Waltham, MA USA). cDNA was then diluted 1:20 and used for qPCR with the TaqMan™ Universal Master Mix II (4440040, Applied Biosystems, Waltham, MA USA) and Universal Probe Library probes (Roche, Basel, Switzerland). All qPCR data were normalized to the Gapdh gene. Primers and probes used are available in Table S3. Quantification was performed using the ∆∆Ct method.

Luciferase Reporter Assays
Cells in 48-well plates were transfected with plasmid DNAs using Fugene HD transfection reagent (E2311, Promega, Madison, WI, USA). Transfection was carried out with 100 ng DNA containing five ng pRL-TK (Promega) to normalize for transfection efficiency. We normalized to equal molarity of the plasmid to use equivalent copies of the reporter plasmid. The total DNA content was complemented with promoter-less plasmid DNA. 20 ng of Elf3 expression plasmid or empty vector (Mock) were transfected. Luciferase reporter activity was measured using Dual-Glo Luciferase Assay (E2940, Promega, Madison, WI, USA) in a Centro XS 3 LB 960 luminometer (Berthold Technologies, Bad Wildbad, Germany). Firefly luciferase reporter values were normalized to those of the Renilla luciferase control. Fold induction was calculated relative to the values of the empty vector.

Expression and Reporter Vectors
The Elf3 expression vector was prepared by amplifying the coding sequence of Elf3 by PCR, adding restriction sites for BamHI (5') and EcoRI (3'). The PCR product was inserted into a modified version of pcDNA4/TO/Myc-His A (Invitrogen, Waltham, MA USA) containing a sequence coding for a 3× FLAG epitope between HindIII and BamHI sites. The full sequence of the plasmid is available upon request.
The Grhl3 promoter (1 kb; between −1 and −1000 bp relative to the transcription start site) was amplified by PCR in 25 µL reaction volume containing 1× PCR buffer, 1.5 mM MgCl 2 , 200 µM dNTP, 10 pmol of each primer (Table S4), 5% DMSO, 1 U of Taq DNA polymerase (2900242, 5 PRIME), and 50 ng of C57BL/6J mouse genomic DNA. Thermal cycler conditions were an initial denaturation step at 95 • C for 5 min, 30 cycles of 95 • C for 30 s, 60 • C for 30 s, and 72 • C for 30 sec, followed by a final extension at 72 • C for 10 min. The amplified PCR products were subsequently inserted upstream of the luciferase reporter in pGL4.10 (Promega, Madison, WI, USA) using XhoI and HindIII restriction sites to produce the reporter plasmid named pGrhl3. The shorter versions of pGrhl3 were prepared by removing sequences flanked by XhoI/BsaI for pGrhl3-XB, XhoI/Psp1406I for pGrhl3-XP and Psp1406I/HindIII for pGrhl3-PH.

Chromatin Immunoprecipitation (ChIP)
ChIP was performed as reported previously [40] with the following minor modifications. NMuMG cells were grown and treated in 10 cm plates. Following 10 min cross-linking and lysis, cell lysates were sonicated in the S220 focused-ultrasonicator (Covaris, Woburn, MA, USA) for 30 min (60 s on/60 s off; Peak Power: 140 W; Duty factor: 15; Cycles/burst: 200; water temperature: 4 • C). Immunoprecipitation was performed at 4 • C for 16 h using an antibody against ELF3/ESE-1 (NBP1-30873, Novus Biologicals, Centennial, CO, USA) and a rabbit control IgG (2729S, Cell Signaling, Danvers, MA, USA). Quantitative real-time PCR was performed using the TaqMan Universal master mix II and Universal Probe Library probes (Roche, Basel, Switzerland) and amplification was carried out in the ABI PRISM 7500 Fast qPCR System (Thermo Scientific, Waltham, MA USA) with primers and probes depicted in Table S5 at 95 • C for 15 min, 50 cycles of 95 • C for 15 s and 60 • C for 1 min. Data were normalized to the input and presented as percentage of input.

SA-β-GAL Assay
NMuMG cells recovering from TGFβ3 were transfected with siCntrl or siElf3_10 as described before. Seventy-two hours later, the cells were washed with PBS, and SA-β-Gal activity was measured by a Senescence Detection Kit (K320, BioVision, Milpitas, CA, USA) according to the manufacturer's instruction. As a positive control, Huh7 cells were used. Induction of senescence was performed with 100 nM Doxorubicin (Dox; A4361, Applichem, Darmstadt, Germany). Huh7 cells were grown in appropriate media containing Dox for three days and an additional six days after Dox withdrawal. The presence of senescent cells was detected using an Olympus CKX41 microscope equipped with a DP25 camera.

RT-PCR
Total RNA was isolated using the Nucleospin RNA II kit (Macherey-Nagel, Düren, Germany) as recommended by the manufacturer. Total RNA (1 µg) was converted to cDNA using the Maxima First Strand cDNA Synthesis Kit (Thermo Scientific, Waltham, MA USA). cDNA was then diluted 1:20 and 2 µL was used for PCR amplification of target genes in a 50 µl reaction containing 0.2 µM of forward and reverse primers ( TGFβ3 treated cells were prepared as described before. Cells (10 × 10 6 ) were harvested, washed in FACS buffer (1 mM EDTA, 25mM HEPES and 1% FBS in PBS), and then stained with E-cadherin antibody (sc-8426, Santa Cruz Biotechnology, Dallas, TX, USA) and goat anti-mouse IgG (H + L) cross-adsorbed secondary antibody, Alexa Fluor 488 (A-11001, Thermo Fisher Scientific, Waltham, MA USA) for 30 min at 4 • C. Cells were washed with FACS buffer and stained with DAPI (final concentration: 0.1 µg/µL). Stained cells subsequently collected by BD FACSAria™ III and single E-cadherin-positive and E-cadherin-negative cells were collected in individual wells of a 96-well plate and allowed to grow into colonies.

Western Blot
NMuMG cells recovering from TGFβ3 were transfected with either siCntrl or siElf3_10 as described before. Seventy-two hours later, the cells were harvested and lysed in RIPA buffer. Protein concentration was determined by TaKaRa BCA Protein Assay Kit (T9300A, Takara, Kusatsu, Japan Protein samples were then diluted with Laemmli Sample buffer. After electrophoresis, proteins were transferred to a nitrocellulose membrane, blocked for 1 h in blotto (Tris-buffered saline containing 0.5% Tween 20 and 5% nonfat milk powder), and incubated with primary antibodies at 1:1000 dilution (Anti-E-cad: 610181, BD Biosciences and anti-ELF3: NBP1-30873, Novus Biologicals) overnight at 4 • C with shaking. Anti-GAPDH (sc-47724, Santa Cruz Biotechnology, Dallas, TX, USA) was used as a loading control. Proteins were detected with IRDye ® 800CW Goat anti-Mouse IgG and IRDye ® 680RD Goat anti-Mouse IgG (H + L) antibodies (LI-COR Biosciences, Lincoln, NE, USA) visualized on an Odyssey ® CLx Imaging System (LI-COR Biosciences, Lincoln, NE, USA).

Wound Healing Assay
For wound healing assay experiments, NMuMG cells recovering from TGFβ3 were transfected with either siCntrl or siElf3_10 as described before. Six hours later, cells were scratched with a sterile pipette tip and cultured for an additional 48 h. Microscopic evaluations in 24-h intervals were carried out using an Olympus CKX41 microscope equipped with a DP25 camera. Scratch width was determined by measuring the distance in pixels between the edges of the scratch in ImageJ. Several measurements were taken from each slide, and the average was calculated.

Statistical Analysis
Statistical significance was determined by performing the Student's t-test using the normalized values of each test sample compared to the normalized value of the control, using a 95% confidence interval; p-values less than 0.05 were considered significant. For Figure S2A One Way ANOVA test was performed. Data are presented as the mean of at least three independent experiments, done in triplicates. Luciferase reporter assays and ChIP experiments were carried out at least three times, qPCR was performed in triplicates. Error bars represent standard error of the mean.

The Transcription Factors Elf3 and Grhl3 Are Highly Correlated with the Epithelial State
Previously, we have reported that the transcription factor Grhl3 is essential for the progression of MET and we showed that it is a crucial regulator of the cell adhesion molecule E-cadherin as well as the transcription factor Hnf4α [25]. In this study, we wanted to reveal other transcription factors essential for the progression of MET. We reasoned that the expression of such transcription factor(s) would correlate with that of Cdh1. To this end, we analyzed the expression profiles from several experimental microarray datasets originating from different and unrelated studies but contained distinct epithelial and mesenchymal samples. We had to apply some criteria to be unbiased in our selection, as the NCBI GEO database includes hundreds of examples. One of the data sets used here is an expression profile of primary keratinocytes and E13.5 MEFs previously produced in our lab (GSE130548 prepared with the Affymetrix MoGene-1_0-st chip (GEO platform GPL6246)). Since we wanted to compare epithelial and mesenchymal samples across different studies, using samples coming from the same platform would make the analyses relatively comparable. Second, the selected study should contain both epithelial and mesenchymal samples. The distinction between epithelial and mesenchymal phenotypes within the same study was made based on the expression of Cdh1, Epcam, Fn1, and Vim genes. In addition to our dataset (six samples), we were able to identify three datasets (GSE77551, GSE55711, and GSE55072) with 14 samples, ending up with a total of 20 samples.
Correlation analysis of Z-score derived from 4 datasets revealed that the transcription factors Casz1, Zfp750, Elf3, Sp6, and Grhl3 as highly correlated with Cdh1 gene and highly expressed in the epithelial phenotype ( Figure 1A), while transcription factors such as Zeb1, Zeb2, Snai1, and Twist2 were found overexpressed in the mesenchymal phenotype. Top positively and negatively correlated genes are shown in Figure 1B. Genes highly correlated with Cdh1 were indeed associated with upregulation in epithelial cell type as indicated by positive fold change, on the other hand, negatively correlated genes elevated in mesenchymal phenotype elicited by negative log fold change. as the transcription factor Hnf4α [25]. In this study, we wanted to reveal other transcription factors essential for the progression of MET. We reasoned that the expression of such transcription factor(s) would correlate with that of Cdh1. To this end, we analyzed the expression profiles from several experimental microarray datasets originating from different and unrelated studies but contained distinct epithelial and mesenchymal samples. We had to apply some criteria to be unbiased in our selection, as the NCBI GEO database includes hundreds of examples. One of the data sets used here is an expression profile of primary keratinocytes and E13.5 MEFs previously produced in our lab (GSE130548 prepared with the Affymetrix MoGene-1_0-st chip (GEO platform GPL6246)). Since we wanted to compare epithelial and mesenchymal samples across different studies, using samples coming from the same platform would make the analyses relatively comparable. Second, the selected study should contain both epithelial and mesenchymal samples. The distinction between epithelial and mesenchymal phenotypes within the same study was made based on the expression of Cdh1, Epcam, Fn1, and Vim genes. In addition to our dataset (six samples), we were able to identify three datasets (GSE77551, GSE55711, and GSE55072) with 14 samples, ending up with a total of 20 samples. Correlation analysis of Z-score derived from 4 datasets revealed that the transcription factors Casz1, Zfp750, Elf3, Sp6, and Grhl3 as highly correlated with Cdh1 gene and highly expressed in the epithelial phenotype ( Figure 1A), while transcription factors such as Zeb1, Zeb2, Snai1, and Twist2 were found overexpressed in the mesenchymal phenotype. Top positively and negatively correlated genes are shown in Figure 1B. Genes highly correlated with Cdh1 were indeed associated with upregulation in epithelial cell type as indicated by positive fold change, on the other hand, negatively correlated genes elevated in mesenchymal phenotype elicited by negative log fold change. Microarray-based screen for epithelial-specific transcription factors coregulated with E-cadherin. One-thousand-four-hundred-and-sixty-nine transcription factors were analyzed for differential expression in epithelial compared to mesenchymal samples, and Spearman correlation was calculated between each transcription factor and E-cadherin across all data sets. The top 5 correlated genes with either the epithelial or the mesenchymal phenotypes are labeled. (B) The ranking plot is displaying the top 25 correlated transcription factors in the epithelial and in the mesenchymal states. The x-axis represents the differential expression of each transcription factor in epithelial versus mesenchymal samples. Microarray-based screen for epithelial-specific transcription factors coregulated with E-cadherin. One-thousand-four-hundred-and-sixty-nine transcription factors were analyzed for differential expression in epithelial compared to mesenchymal samples, and Spearman correlation was calculated between each transcription factor and E-cadherin across all data sets. The top 5 correlated genes with either the epithelial or the mesenchymal phenotypes are labeled. (B) The ranking plot is displaying the top 25 correlated transcription factors in the epithelial and in the mesenchymal states. The x-axis represents the differential expression of each transcription factor in epithelial versus mesenchymal samples.

Elf3 Is Essential for MET
To establish the relevance of the highly correlated transcription factors to MET, we utilized NMuMG cells (an established murine model for the study of EMT and MET). These cells will gain a mesenchymal phenotype once treated with TGFβ for three days, and after TGFβ withdrawal, the cells will revert to the epithelial phenotype in three days (MET). We first analyzed the expression levels of the top correlated genes in NMuMG cells by RT-PCR ( Figure S1A). While Casz1 and Zfp750 were barely detectable in NMuMG cells (only detectable when 20 times more cDNA was used), Elf3 had comparably higher RNA levels, and thus was selected for further analyses. We used siRNA mediated silencing of candidate genes at the onset of MET in NMuMG cells and followed the progression of MET both morphologically as well as by expression levels of epithelial and mesenchymal defining genes.
We first validated MET in NMuMG cells by single-cell sorting of cells treated with TGFβ for three days, as well as seeding TGFβ treated cells at a very low density (single-cell cloning) to rule out the possibility that remaining epithelial cells within a mesenchymal population could take over due to their faster proliferation rate rather than an actual MET ( Figure S1B). As a result, we were able to confirm that single mesenchymal cells were able to form colonies of epithelial cells after the withdrawal of TGFβ. Then, we examined the expression pattern of Elf3 in NMuMG cells during the progression of EMT and MET. We have noticed that unlike the Cdh1 expression pattern, Elf3 RNA levels were relatively stable ( Figure S1C). We then used several siRNAs targeting Elf3 and validated their knockdown efficiency in NMuMG cells by measuring Elf3 RNA levels. As a result, we noticed that all siRNAs used produced a similar phenotype with varying yet substantial knockdown efficiencies ( Figure S2A). In cells transfected with nontargeting siRNA, we did not notice any morphological changes; on the other hand, Elf3 targeting siRNAs caused a failure of MET evident by the enlarged and flattened cell morphology ( Figure S2B). Confocal images of siElf3 transfected cells revealed oversized cells, with actin present as stress fibers, absence of E-cadherin at the plasma membrane, and an increased Vimentin expression (Figure 2). These results confirmed the MET failure and suggested that this aberrant MET was, in part, due to the absence of E-cadherin from the plasma membrane.

Elf3 is Essential for MET
To establish the relevance of the highly correlated transcription factors to MET, we utilized NMuMG cells (an established murine model for the study of EMT and MET). These cells will gain a mesenchymal phenotype once treated with TGFβ for three days, and after TGFβ withdrawal, the cells will revert to the epithelial phenotype in three days (MET). We first analyzed the expression levels of the top correlated genes in NMuMG cells by RT-PCR ( Figure S1A). While Casz1 and Zfp750 were barely detectable in NMuMG cells (only detectable when 20 times more cDNA was used), Elf3 had comparably higher RNA levels, and thus was selected for further analyses. We used siRNA mediated silencing of candidate genes at the onset of MET in NMuMG cells and followed the progression of MET both morphologically as well as by expression levels of epithelial and mesenchymal defining genes.
We first validated MET in NMuMG cells by single-cell sorting of cells treated with TGFβ for three days, as well as seeding TGFβ treated cells at a very low density (single-cell cloning) to rule out the possibility that remaining epithelial cells within a mesenchymal population could take over due to their faster proliferation rate rather than an actual MET ( Figure S1B). As a result, we were able to confirm that single mesenchymal cells were able to form colonies of epithelial cells after the withdrawal of TGFβ. Then, we examined the expression pattern of Elf3 in NMuMG cells during the progression of EMT and MET. We have noticed that unlike the Cdh1 expression pattern, Elf3 RNA levels were relatively stable ( Figure S1C). We then used several siRNAs targeting Elf3 and validated their knockdown efficiency in NMuMG cells by measuring Elf3 RNA levels. As a result, we noticed that all siRNAs used produced a similar phenotype with varying yet substantial knockdown efficiencies ( Figure S2A). In cells transfected with nontargeting siRNA, we did not notice any morphological changes; on the other hand, Elf3 targeting siRNAs caused a failure of MET evident by the enlarged and flattened cell morphology ( Figure S2B). Confocal images of siElf3 transfected cells revealed oversized cells, with actin present as stress fibers, absence of E-cadherin at the plasma membrane, and an increased Vimentin expression (Figure 2). These results confirmed the MET failure and suggested that this aberrant MET was, in part, due to the absence of E-cadherin from the plasma membrane.  The enlarged cellular morphology was reminiscent of senescent cells, but we could not confirm this using the very well-established SA-β-Gal staining method ( Figure S2C), suggesting a different cause for the enlarged cells. We wanted to confirm these observations by studying RNA expression levels of genes relevant to the EMT and MET programs. We selected Ahr, Cebpα, Ovol2, Exosc9, and Ehf for their potential involvement in the process of maintaining epithelial integrity, regulating the dynamic switch between epithelial and mesenchymal states, and key roles in metastasis [41][42][43][44][45]. NMuMG cells undergoing MET were transfected with different siRNAs targeting Elf3, and gene expression levels were evaluated by qPCR analysis. There was a significant downregulation of several transcription factors such as Ahr, Cebpα, Ehf, and Ovol2 ( Figure 3A).
The enlarged cellular morphology was reminiscent of senescent cells, but we could not confirm this using the very well-established SA-β-Gal staining method ( Figure S2C), suggesting a different cause for the enlarged cells. We wanted to confirm these observations by studying RNA expression levels of genes relevant to the EMT and MET programs. We selected Ahr, Cebpα, Ovol2, Exosc9, and Ehf for their potential involvement in the process of maintaining epithelial integrity, regulating the dynamic switch between epithelial and mesenchymal states, and key roles in metastasis [41][42][43][44][45]. NMuMG cells undergoing MET were transfected with different siRNAs targeting Elf3, and gene expression levels were evaluated by qPCR analysis. There was a significant downregulation of several transcription factors such as Ahr, Cebpα, Ehf, and Ovol2 ( Figure 3A). On the other hand, the expression levels of some mesenchymal markers showed an evident increase in the absence of Elf3; we noticed a significant increase in the classical EMT inducers such as Snai1/2 and Zeb2 ( Figure 3B), which was accompanied by an increased expression of Cdh2 and Fn1 ( Figure 3B). In agreement with the data in Figure 2, RNA and protein levels of E-cadherin after Elf3 depletion were comparable to those of control cells, which can complete the MET transition ( Figure  3A and 3C). These results confirmed once more the MET failure, with a restricted appearance of the cells in a morphology reminiscent of sustained EMT. This mesenchymal state was also confirmed by the increased migratory capacity of Elf3 depleted cells. We observed a five-fold increased migration as compared to control cells in wound healing assay ( Figure 3D and Figure S3A). The observed impact on MET is not a typical feature of the depletion of transcription factors with a strong correlation with the epithelial phenotype. We affirmed this by using siRNAs targeting Ehf, another transcription factor from the Ets family, which is also found among the top 25 genes correlated with the epithelial phenotype. We found that the siRNA mediated depletion of Ehf did not affect MET progression nor the Cdh1 RNA levels ( Figure S3B, C, respectively). On the other hand, the expression levels of some mesenchymal markers showed an evident increase in the absence of Elf3; we noticed a significant increase in the classical EMT inducers such as Snai1/2 and Zeb2 ( Figure 3B), which was accompanied by an increased expression of Cdh2 and Fn1 ( Figure 3B). In agreement with the data in Figure 2, RNA and protein levels of E-cadherin after Elf3 depletion were comparable to those of control cells, which can complete the MET transition ( Figure 3A,C). These results confirmed once more the MET failure, with a restricted appearance of the cells in a morphology reminiscent of sustained EMT. This mesenchymal state was also confirmed by the increased migratory capacity of Elf3 depleted cells. We observed a five-fold increased migration as compared to control cells in wound healing assay ( Figure 3D and Figure S3A). The observed impact on MET is not a typical feature of the depletion of transcription factors with a strong correlation with the epithelial phenotype. We affirmed this by using siRNAs targeting Ehf, another transcription factor from the Ets family, which is also found among the top 25 genes correlated with the epithelial phenotype. We found that the siRNA mediated depletion of Ehf did not affect MET progression nor the Cdh1 RNA levels ( Figure S3B,C, respectively).

Elf3 Regulates the Expression of Grhl3
The results from Figure 3 revealed the effect of Elf3 silencing on genes previously associated with epithelial integrity or EMT. We also wanted to study the influence of Elf3 depletion on the genes highly correlated with the epithelial phenotype. For this, we examined the expression of the top 10 correlated genes from Figure 1 after Elf3 silencing at the onset of MET. We noticed a significant downregulation of Grhl3 expression and a modest downregulation of Tfcp2l1 ( Figure 4A). We were intrigued by the appearance of Grhl3 in the top 5 genes correlated with Cdh1 as well as its downregulation in response to Elf3 silencing because of its importance in MET regulation [25]. For this reason, we measured z-scores for transcription factors correlated with Grhl3, and we observed that it also correlates with genes having the highest correlation with the epithelial phenotype, such as Casz1 and Elf3 ( Figure S4A). We analyzed the bilateral correlation between Elf3 and Grhl3 in the datasets from Figure 1 and found a high correlation of 0.84, p-value < 2.22 × 10 −16 ( Figure S4B). We studied changes in Grhl3 RNA levels after modulating Elf3 expression in other cellular models and found that in primary keratinocytes, for example, Grhl3 was significantly down-regulated upon Elf3 depletion by siRNAs ( Figure 4B). Also, the overexpression of ELF3 in primary MEFs was effective in increasing the RNA levels of Grhl3 ( Figure 4C), suggesting a potential transcriptional regulatory relationship between the two genes.

Elf3 Regulates the Expression of Grhl3
The results from Figure 3 revealed the effect of Elf3 silencing on genes previously associated with epithelial integrity or EMT. We also wanted to study the influence of Elf3 depletion on the genes highly correlated with the epithelial phenotype. For this, we examined the expression of the top 10 correlated genes from Figure 1 after Elf3 silencing at the onset of MET. We noticed a significant downregulation of Grhl3 expression and a modest downregulation of Tfcp2l1 ( Figure 4A). We were intrigued by the appearance of Grhl3 in the top 5 genes correlated with Cdh1 as well as its downregulation in response to Elf3 silencing because of its importance in MET regulation [25]. For this reason, we measured z-scores for transcription factors correlated with Grhl3, and we observed that it also correlates with genes having the highest correlation with the epithelial phenotype, such as Casz1 and Elf3 ( Figure S4A). We analyzed the bilateral correlation between Elf3 and Grhl3 in the datasets from Figure 1 and found a high correlation of 0.84, p-value < 2.22e-16 ( Figure S4B). We studied changes in Grhl3 RNA levels after modulating Elf3 expression in other cellular models and found that in primary keratinocytes, for example, Grhl3 was significantly down-regulated upon Elf3 depletion by siRNAs ( Figure 4B). Also, the overexpression of ELF3 in primary MEFs was effective in increasing the RNA levels of Grhl3 ( Figure 4C), suggesting a potential transcriptional regulatory relationship between the two genes. Elf3 targeting siRNAs, after 72 hours, cells were collected, RNA was isolated, and changes in the expression of the top 10 genes correlated with the epithelial phenotype were measured by qPCR. The * below gene names indicate genes with efficient downregulation. (B) Primary mouse keratinocytes were transfected with siRNAs targeting Elf3. Seventy-two hours after transfection RNA was isolated, and the expression levels of Elf3 and Grhl3 were measured by qPCR. (C) MEFs were isolated from E13.5 embryos and transfected with ELF3 expressing plasmid. Seventy-two hours after transfection cells were collected in RNA lysis buffer, total RNA was isolated and converted to cDNA. qPCR was performed using Grhl3 specific primers. Relative expression was calculated using the ΔΔCt method relative to the vector. Data represent the averages of at least three independent experiments. Paired Student's t-test was used to calculate statistical significance. ** p-value = 0.005; **** p-value < 0.0001. Elf3 targeting siRNAs, after 72 h, cells were collected, RNA was isolated, and changes in the expression of the top 10 genes correlated with the epithelial phenotype were measured by qPCR. The * below gene names indicate genes with efficient downregulation. (B) Primary mouse keratinocytes were transfected with siRNAs targeting Elf3. Seventy-two hours after transfection RNA was isolated, and the expression levels of Elf3 and Grhl3 were measured by qPCR. (C) MEFs were isolated from E13.5 embryos and transfected with ELF3 expressing plasmid. Seventy-two hours after transfection cells were collected in RNA lysis buffer, total RNA was isolated and converted to cDNA. qPCR was performed using Grhl3 specific primers. Relative expression was calculated using the ∆∆Ct method relative to the vector. Data represent the averages of at least three independent experiments. Paired Student's t-test was used to calculate statistical significance. ** p-value = 0.005; **** p-value < 0.0001.

The Transcription Factor Elf3 Activates the Promoter of Grhl3
To address the potential transcriptional regulatory relationship between ELF3 and Grhl3, we analyzed the promoter region of Grhl3 (1 kb proximal to TSS) for conserved sequences and ELF3 binding sites. The promoter of Grhl3 was analyzed for conservation using the genome vista tools and aligned to several mammalian species. We were able to identify three conserved sequences from Mouse/Human and Mouse/Chimp alignments, and two conserved sequences from Mouse/Rhesus, Mouse/Cow, and Mouse/Dog alignments ( Figure S5A). The sequence of the proximal promoter region is presented in Figure 5A, and some promoter elements are highlighted, the Sp1 site appeared conserved in all comparisons ( Figure S5B). We identified a total of 17 ELF3 binding sites in the 1 kb region of Grhl3 promoter (green nucleotides in Figure 5A). This 1 kb sequence was then cloned in the pGL4.10 vector to measure the reporter activity in response to ELF3 induction ( Figure 5B). NMuMG cells were transfected with various extents of the promoter sequence; we observed that the Grhl3 promoter retained its activity in the absence of CNS2 and 3 but required the presence of CNS1 together with the TATA and Sp1 elements ( Figure 5B). Although the functionality of the TATA box per se cannot be deduced from these experiments, the presence of other crucial promoter elements cannot be ruled out Moreover, we found that ELF3 expression was sufficient to upregulate the reporter activity. The activation of the Grhl3 promoter by ELF3 was not confined to in vitro assays, as we were able to detect a significant enrichment of ELF3 at the Grhl3 promoter in cells undergoing MET ( Figure 5C). Of the 17 ELF3 binding sites in the Grhl3 promoter, six were present in the first conserved sequence (CNS1: 210 bp upstream of TSS), four of which were found conserved among different mammalian species ( Figure 5D and Figure S5D). The binding of ELF3 at the Grhl3 promoter was confined to CNS1 during MET; we were not able to detect similar binding using primers outside of this conserved sequence ( Figure S5C).

Discussion
The expression of the adhesion molecule E-cadherin is indispensable for the epithelial state, and its regulation is tightly and dynamically controlled during several developmental and physiological processes, during which EMT and MET programs alternate, and is characterized by the switching of cadherin gene expression. Positive regulation of Cdh1 expression is a limiting factor for an efficient MET, suggesting that regulators of Cdh1 are in fact regulators of MET and the epithelial state. Here, by using in silico methods and comparing several datasets with distinct epithelial and mesenchymal phenotypes, we defined an epithelial-specific set of transcription factors based on their correlation with Cdh1 expression. We identified several transcription factors with high correlation with the epithelial phenotype, such as Elf3, Grhl3, and Ovol2.
Elf3, a member of the Ets family of transcription factors, appeared to be essential for proper MET progression, and its depletion caused a severe MET defect marked by the absence of E-cadherin from the plasma membrane and also the downregulation of several epithelial genes in particular Grhl3. Closer examination of the relationship between Elf3 and Grhl3 led us to identify ELF3 as a regulator of Grhl3 expression. ELF3 was able to activate the Grhl3 promoter in reporter assays, and this was further verified as ELF3 was found to be present at the promoter of Grhl3 in cells undergoing MET.
The Ets family of transcription factors is one of the largest families of transcription factors, consisting of~30 members (27 in humans and 26 in mice) that can be expressed both ubiquitously and in a tissue-specific manner [46,47]. Ets transcription factors are characterized by conserved winged helix-turn-helix DNA binding domains called ETS domain that binds a typical DNA sequence 5 -GGA(A/T)-3 [46,48,49]. Members of this family are trans-acting phosphoproteins [50] that can act as upstream and downstream effectors of most signaling pathways, including MAP kinases, Erk1/2, p38, and JNK [51], thus playing crucial roles in a wide variety of biological processes such as differentiation, development, proliferation, apoptosis, tissue remodeling and the epithelial to mesenchymal transition with a suggested involvement in the mesenchymal to epithelial transition [52][53][54][55][56].  The E74-like transcription factor-3 Elf3 (also known as ESX, ESE-1, ERT, and JEN), which was first described by Oettgen and colleagues in 1997 [57], is the main Ets transcription factor which is expressed ubiquitously by epithelial-rich tissues, such as stomach, pancreas, uterus, colon, kidney, lung, mammary gland, and skin [58,59]. The observed expression of Elf3 in our EMT/MET model ( Figure S1C) was not restricted to the epithelial states (Vehicle and PT) as for Cdh1, but also detected in the mesenchymal cells. While this is contradictory to the high correlation scores between Elf3 and Cdh1 in Figure 1, it could be explained by a context-dependent expression of Elf3 in mesenchymal cells since previous reports indicated that it could also be induced in nonepithelial cells by the proinflammatory cytokines TNF-α and IL-1β [50,60]. This context-dependent expression of Elf3 could also imply a context-dependent function of ELF3, the interaction of ELF3 with the Grhl3 promoter was only detectable during MET, but not in mesenchymal cells.
Besides the critical roles of ELF3 in epithelial cell differentiation, gut development, inflammation, and apoptosis, it is also essential for the pathophysiology of cancer epithelial cells such as in breast, lung and prostate cancers [61][62][63][64]. It was demonstrated that ELF3 both activates the type II TGFβ receptor gene (TβR-II) in epithelial cells and binds TβR-II promoter in vivo [65]. For instance, ELF3 controls intestinal epithelial differentiation during development by altering the expression pattern of TβR-II [66]. ELF3 was also recently identified as a candidate transcriptional regulator involved in human urothelial cytodifferentiation together with GRHL3; its depletion in urothelial cells result in a declined expression of some transcription factors such as FOXA1 and GRHL3, which are known to be involved in urothelial differentiation [67]. It was reported that IRF6 and GRHL3 function downstream of RIPK4 to promote ELF3 gene expression in keratinocytes [68], and considering the data presented here, it is plausible to hypothesize a positive feedback regulatory circuit involving Grhl3 and Elf3. A recent publication also revealed that ZEB1 and ELF3 transcriptionally regulate IRF6 [69].
The role of Elf3 was also indicated in animal models. Elf3 knockout results in embryonic lethality in~30% of mice models at E11.5. Born Elf3 −/− mice develop a wasted phenotype characterized by watery diarrhea, lethargy, and malnourishment [66].
The studies mentioned above accentuated the importance of ELF3 in epithelial cells; this is in agreement with the findings presented here; Elf3 tops the highly correlated transcription factors with the epithelial phenotype. The fact that Grhl3 was also among the first factors correlated with the epithelial phenotype was not surprising; it holds a critical role in MET [25]. The close association between Elf3 and Grhl3 was reported in earlier studies [67], and hence, we deduced their regulatory relationship. The transcriptional relationship between Elf3 and Grhl3 was not limited to NMuMG cells since results obtained from MEFs and keratinocytes provided compelling proof.
Similar to the loss of Grhl3, loss of Elf3 during MET failed the epithelial transition. However, unlike Grhl3, lack of which resulted in downregulation of Cdh1, Elf3 depletion did not cause a reduction in Cdh1 mRNA or its protein levels. Instead, E-cadherin was absent from the plasma membrane, which is a significant obstacle preventing MET progression favoring a mesenchymal phenotype. The implications of a deregulated E-cadherin expression as well as its absence from the plasma membrane have been described before. Analysis of tumor samples suggested that aberrant expression of E-cadherin (or no expression) significantly correlates with poor prognosis and promoted metastasis [70][71][72]. However, the association of Elf3 with an aberrant E-cadherin expression is a new finding but with comparable observations from Drosophila. The Drosophila aop (anterior open, also named Yan) and Eip74EF (Ecdysone-induced protein 74EF) are two genes found in the FlyBase [73] as orthologs for several mammalian Ets transcription factors, such as Elf3, Ehf, and Elf5. In particular, aop was reported to control the border cell migration in the Drosophila egg chamber by altering shg (DE-Cad) expression and its localization [74]. Therefore, a dramatic level of evolutionary conservation of the Elf3 function cannot be ruled out.
In Drosophila, grainy head (grh) is driving postembryonic neuroblast function and proliferation by promoting DE-cadherin expression [18,75]. This transcriptional relationship between grh and shg is conserved; the mouse orthologs of grh, GRHL2, and GRHL3 have been reported to regulate the transcription of Cdh1 in mouse cells [25,31]. Interspecies functional conservation would then infer conserved regulation; this, in turn, would suggest conserved cis-regulatory elements, such as enhancers. Several reports have indicated that interspecies genome comparisons of non-coding DNA sequences could lead to correct predictions of regulatory sequences [25,40,76,77]. The presence of several conserved ELF3 binding sites in the promoter of Grhl3 together with the experimental evidence on the regulatory effect of ELF3 on Grhl3 is in support of the statement mentioned above.
Although the ETS transcription factors have a novel role in various diseases, the function of these transcription factors, particularly in cancer remains largely unclear. ELF3 depletion during MET was not limited to the aberrant expression of Cdh1; our data confirmed the role of ELF3 as a regulator of epithelial identity, reflected by the dramatic changes in the expression of selected EMT and MET relevant genes. The ligand-activated transcription factor Aryl hydrocarbon receptor (Ahr) promotes EMT and metastasis via its ligands l-kynurenine (l-Kyn) and d-kynurenine (d-Kyn) [45,78]. EHF has a regulatory function in epithelial cell differentiation and gaining stem cell-like properties via EMT [41]. The elevated expression of EHF/ESE-3 is associated with poor clinical outcomes and increased recurrence [41,79].
Cellular plasticity is another big puzzle during EMT/MET phenotypic switches. Numerous studies have shown that the expression levels of EMT inducers are linked to cell plasticity and stem cell phenotypes. There is no doubt that the dynamic nature of EMT/MET transitions is highly regulated by other transcription factors which can influence known master EMT regulators. Multilevel regulation of some transcription factors defines a harmonic balance between epithelial plasticity and stability in a context-dependent manner. OVOL2 was recently revealed as a transcriptional repressor of EMT through downregulating ZEB1 and TWIST in various cancer types such as in lung, breast, and colorectal cancers [44,80,81]. In a recent report, OVOL2 was identified as a MET inducing transcription factor in fibroblasts, which also enhanced the reprogramming of dermal fibroblasts into keratinocyte-like cell state [82]. In this report OVOL2 was identified following a comprehensive screening approach for transcription factors correlated with CDH1 in a large number of expression datasets. In support of our findings, it was interesting to find that besides Elf3 and Ovol2, similar transcription factors were also common between the two studies and present among the top correlated genes, such as Grhl3, Irf6, Grhl2, Grhl1, and Ehf, suggesting a conserved function of these transcription factors in EMT and MET regardless of the organism or cell type studied.
Considering the importance of EMT/MET during development and cancer, a better understanding of these fundamental processes will lead to a better comprehension of the common players in physiological and pathological contexts. In a recent study, Elf5 emerged as a lineage regulator of mammary gland development and as an inhibitor of EMT in breast cancer via repressing Snail2/Slug [83]. Similarly, Ovol2 was identified as an indispensable guardian of the epithelial differentiation besides its EMT suppressing ability [84]. Our findings present novel insights into the regulation of MET and seek to emphasize the remarkable role of Elf3 as a transcriptional gatekeeper of epithelial state. It is also not surprising that physiological and pathological EMT share common attributes in terms of transcriptional regulators. Altogether, our findings provide a broader perspective for the management of metastatic cancers.
In conclusion, we provided compelling evidence for the involvement of ELF3 in the regulation of MET by altering the localization of E-cadherin. Moreover, ELF3 regulates the expression of Grhl3 by binding to conserved sequences in its promoter. Loss of Elf3 prevented the MET initiation and preserved the mesenchymal state. Further studies are needed to establish the mechanistic relevance of these findings as well as their relevance to possible functions of Elf3 in the progression of metastasis.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4409/8/8/858/s1. Figure S1: The expression of Elf3 in NMuMG cells and during MET; Figure S2: The siRNA mediated silencing of Elf3 results in an impaired MET; Figure S3: Different influence of Elf3 and Ehf during the progression of MET; Figure S4: Genes correlated with Grhl3; Figure S5: Conservation of the Grhl3 promoter; Table S1: NCBI GEO accession numbers and phenotypes of samples used in correlation studies; Table S2: Sequence information of siRNAs used; Table S3: Primers and corresponding UPL probes used in qPCR experiments; Table S4: Sequences of