Microarray analysis of gene expression during epithelial–mesenchymal transformation



One of the most fundamental biological processes in development, as well as a primary mechanism for tumor metastasis, is epithelial–mesenchymal transformation (EMT). To gain a greater understanding of this transition, we have obtained a genomic profile of the critical stages before and during this rapid change in morphology in the developing mouse palate. By isolating the medial edge epithelium of each palatal shelf, we were able to obtain pure gene expression data without contamination from surrounding mesenchymal cells. Our results support the important role of the TGF-β/Smad signal transduction pathway in the stimulation of EMT by means of up-regulation of the EMT-inducing gene, LEF-1. We document changes in gene expression profiles during palatal adherence and subsequent transformation of the medial edge epithelial seam that suggests a high number of LEF-1 target genes promote cellular transformation to mesenchyme. These include genes involved in cell adhesion, polarity, cytoskeletal dynamics, migration, and intracellular signaling. This knowledge of the changes in gene expression levels during palatogenesis should lead to a better understanding of the mechanisms of EMT. Developmental Dynamics 234:132–142, 2005. © 2005 Wiley-Liss, Inc.


Epithelial–mesenchymal transformation (EMT) is a fundamental mechanism in many embryological processes. EMT is associated with loss of cell–cell adhesion, cellular elongation, and invasion of the underlying extracellular matrix (ECM) by the new mesenchymal cells. The epithelial cells form filopodia-like structures along the developing leading ends as they break away from the surrounding cells and digest the basal lamina. The newly formed mesenchymal cells that arise from the epithelium migrate through the ECM to specific locations based on their developmental programs. This process is mimicked by tumor cell metastasis, but here the mesenchyme-like cells lack specific developmental potentials (Hay, 1995; Birchmeier et al., 1996).

It is widely believed that the molecule responsible for the signaling of almost all significant EMT genes, including those expressed in colon carcinomas (Korinek et al., 1997), is lymphoid enhancer factor-1 (LEF-1). The LEF-1 transcription factor is well known to (directly or indirectly) affect the expression of EMT genes, including the cell adhesion molecule E-cadherin, actin, vimentin, and many others such as cyclins (Hay, 1995; Kim et al., 2002). A common mechanism by which LEF-1 activates the transcription of EMT genes is by forming a complex with the intracellular protein, β-catenin. These complexes will then translocate to the nucleus and either activate or repress transcription, depending on the function of the protein being produced (Hay and Zuk, 1995; Hecht and Kemler, 2000).

The major signal transduction pathway known as the transforming growth factor-beta (TGF-β)/Smad pathway has been shown to stimulate EMT by means of the Smad2/4/LEF-1 transcription complex, rather than the β-catenin/LEF-1 complex (Labbe et al., 2000; Nawshad and Hay, 2003). TGF-β has a vast number of functions within the cell, which include proliferation, survival, mitotic regulation, apoptosis, and EMT. The actual function depends upon which member of the TGF-β family is activated, as well as responding cell specificity. In the case of EMT, usually either TGF-β1 (cancer) or TGF-β3 (development) is involved (Yue and Mulder, 2001).

In palate development, single palatal shelves come together at their medial edges and adhere within 12 hr (in vitro), forming a medial edge epithelial (MEE) seam. Autocrine TGF-β3 signal transduction then causes the MEE cells to undergo transformation. These cells break away from the seam between 24 and 36 hr, forming small islands of epithelia that are in the process of transforming into mesenchyme. At 60 hr, the palate is completely confluent and will continue its development as a single organ (Fitchett and Hay, 1989; Griffith and Hay, 1992; Nawshad and Hay, 2003).

TGF-β3 (secreted from the basal surface of the epithelium) interacts directly with a membrane serine–threonine kinase receptor called TBRII on the basal surface, causing its activation and leading to the recruitment of another receptor called TBRI (ALK5) to form a two receptor complex (Derynck et al., 1997). This action triggers endocytosis around the receptor complex, through molecules like Rab5, for the purpose of easily recruiting other cytoplasmic proteins (Penheiter et al., 2002). During early endocytosis, a lipid-based molecule within the endosomal membrane called phosphotydialinositol-2 phosphate (PI2P) is phosphorylated by PI3-kinase (PI3K) to form PI3P. This action signals the Smad anchor for receptor activation (SARA) to recruit a particular R-Smad protein (Smad2). SARA will then bind with PI3P through its phospholipid binding FYVE domain. The receptor complex phosphorylates Smad2, causing its release and association with the co-Smad, Smad4. When the Smad2 binds with Smad4, this new Smad complex moves into the nucleus (Wu et al., 2000; Yue and Mulder, 2001; Shi and Massague, 2003), where it will up-regulate LEF-1 transcription and, thus, increase LEF-1 protein levels (Nawshad and Hay, 2003).

In the palate system, recent studies have shown that antisense β-catenin/γ-catenin ODN has no effect on palate EMT, proving that LEF-1 can induce gene transcription without it (Nawshad and Hay, 2003). It has now been proved that LEF-1 can be activated by Smads instead of β-catenin, forming complexes that translocate to the nucleus and affect gene expression (Labbe et al., 2000; Nishita et al., 2000). During the first wave of TGF-β signaling, the phospho-Smad2/4 complex up-regulates gene expression levels of LEF-1 (Nawshad and Hay, 2003). Upon transcription and translation of LEF-1, a Smad2/4/LEF-1 complex is formed similar to the β-catenin/LEF-1 complex (which is not activated in the palate), which targets specific EMT genes (Fig. 1). This dual role of pSmad2/4 for both LEF-1 gene expression and LEF-1 protein activation is the essential mediator of TGF-β3–induced palatal EMT.

Figure 1.

A proposed schematic diagram of intracellular signaling during epithelial–mesenchymal transformation (EMT) in palate medial edge epithelial cells. As palatal adherence commences, an unknown signal causes up-regulation of members of the TGF-β signal transduction pathway. Upon synthesis, these proteins signal the activation and up-regulation of the LEF-1 master gene. This homeobox transcription factor is then stimulated by the Smad2/4 complex to activate genes that cause EMT.

Our objective is to obtain a genomic profile of the molecular mechanisms that bring about physiological changes in MEE cells during EMT. This study represents the first genomic analysis of isolated epithelial cells transforming to mesenchyme in vitro.


Demonstration of Palatal EMT

Hematoxylin and eosin staining was performed on palates in various stages of development. At 12 hr (Fig. 2A), the palatal shelves attach to form a medial edge epithelial seam. At 24 (Fig. 2B) and 36 hr (Fig. 2C), EMT causes the seam to break away into small islands of epithelial cells that are in the process of transforming to mesenchyme. At 60 hr (Fig. 2D), EMT is complete, resulting in total palatal confluence.

Figure 2.

Hematoxylin and eosin staining of tissue sections during mouse palate development. A–D: A steady progression of epithelial–mesenchymal transformation (EMT) in the medial edge epithelial cells (arrows) is shown at 12 (A), 24 (B), 36 (C) hr, with total conversion of the seam to mesenchyme and confluence at 60 hr (D).

Gene Chip Comparisons

A genomic readout of gene chips from each period of development (Fig. 3A–C), matched against the murine genome, shows 5,375 (0 hr), 4,510 (12 hr), and 4,567 (24 hr) genes flagged as present (high mRNA levels). Expression intensities of these genes are characterized by low (green), moderate (yellow), and high (red) levels. Graphic analysis of major expression changes between each developmental period, provides an accurate interpretation of the complexity of the EMT process. In each bow tie scatter plot graph (Fig. 3D–F), twofold increases or decreases in expression between each period are compared. Between 0 and 12 hr (Fig. 3D), 2,088 genes have a twofold increase, while 1,301 have a twofold decrease. Between 12 and 24 hr (Fig. 3E), 831 genes are increased by twofold, while 1,523 genes are decreased by twofold. As an overall process between 0 and 24 hr (Fig. 3F), we witness twofold up-regulations in 1,256 genes, and twofold down-regulations in 1,200 genes.

Figure 3.

Genomic profiling and regulation during each stage of development. A–C: A three-dimensional scatter plot graph analysis of low (green), moderate (yellow), and high (red) expression levels of genes flagged as present at 0 (A), 12 (B), and 24 (C) hr. D–F: A two-dimensional scatter plot comparison demonstrates twofold up-regulations and down-regulations between 0 and 12 (D), 12 and 24 (E), and 0 and 24 (F) hr.

The gene analysis software configured a normalized value for each gene to provide the optimal interpretation for expression changes. These values were computed by the Microarray Suite 5.0 (Affymetrix) software program, based on a standard algorithm, which allows for comparisons of each gene chip to be analyzed with a common average signal intensity. Genes flagged as absent were assigned a value of zero to show proper comparisons in graphic analysis. A high number of standardized controls are present in each gene chip, with probability values assigned to each probe based on hybridization quality assessed by these controls. Genes with probability (P) values of less than 0.05 were used for individual gene analysis, to ensure a 95% probability of correct data.

The TGF-β Signaling Pathway

A steady up-regulation of mRNA levels of the endogenous TGF-β3 ligand is shown over the course of development, with the largest increase between 0 and 12 hr. The EMT master gene LEF-1, which is not expressed at 0 hr, is shown to be activated with continually increasing mRNA levels as a result of TGF-β signaling (Nawshad and Hay, 2003). Other important TGF-β signaling molecules involved in this activation (Rab5, SARA, Smad2, and Smad4) show significant increases (Fig. 4A) as a result of MEE seam formation (12 hr), then assume a similar expression level while approaching EMT (24 hr). Smad3 was determined to be absent in each stage (data not shown), confirming that in the palate system, Smad2 is the necessary R-Smad for EMT (Cui et al., 2003). Validation of the expression of Smad2, Smad3, Smad4, and LEF-1 was performed by means of reverse transcriptase-polymerase chain reaction (RT-PCR; Fig. 4B). Protein up-regulations of TGF-β3, LEF-1, Smad2, and Smad4 were confirmed by immunohistochemistry of MEE cells from 0- and 12-hr time points (Fig. 4C).

Figure 4.

Progressive expression of genes necessary for transforming growth factor-beta (TGF-β) signal transduction. Steady up-regulation of the TGF-β3 and lymphoid enhancer factor-1 (LEF-1) genes are observed as development progresses. A: Important signaling factors in the TGF-β pathway such as Rab5, SARA, Smad2, and Smad4 peak at 12 hr before being regulated to a lower fixed signal level at 24 hr, which remains much higher than the control levels (0 hr; P values < 0.05). B: Reverse transcriptase-polymerase chain reaction validation confirms these up-regulations and the absence of Smad3. C: Confirmation of expression increases of TGF-β3, LEF-1, Smad2, and Smad4 between 0 and 12 hr in cultured medial edge epithelium (MEE) cells are demonstrated by immunofluorescence.

Importance of Palatal Adherence

Palatal adherence plays a crucial role in initiating TGF-β signaling leading to transformation of the MEE seam, as the data in Figure 4 illustrate. Because the signaling pathways that causes these gene up-regulations (Fig. 4A, 12 hr) are not yet known, we have focused on reporting changes in expression of genes that are known to be associated with adherence and EMT (Fig. 5A). Along with the TGF-β signaling factors, molecules necessary for cellular adhesion (β-catenin, E-cadherin, VE-cadherin) are increased during adherence. Increased β-catenin expression was confirmed by RT-PCR (Fig. 5B). The formation of desmosomes has long been associated with palatal adherence and has been shown to be activated by cadherins (Vanderburg and Hay, 1996; Sun et al., 1998b). Therefore, we see the expected up-regulation in genes associated with their formation (desmocollin, desmoglein, plakophilin, vinculin). Integrin-β1 (Int-B1), an extracellular matrix receptor associated with mesenchymal cell invasion, is also up-regulated. Activation of focal adhesion kinase (FAK), an important integrin signaling protein, is also observed. The expression level of Fos, a molecule know to be involved in LEF-1 activation (Eger et al., 2000), is increased at 12 hr. Jun, an important molecule in MAP kinase signaling, which forms the AP-1 transcription factor (Whitmarsh and Davis, 1996), does not appear to have a role in adherence or EMT, as it is down-regulated.

Figure 5.

Expression changes in genes associated with palatal adherence. A: Significant alterations were observed from 0 to 12 hr in genes associated with epithelial cell adherence (β-catenin [B-cat], E-cadherin [E-cad], VE-cadherin [VE-cad]), desmosome formation (desmocollin [Dscn], desmoglein [Dsgn], plakophilin [Pkpn], vinculin [Vcln]), cellular invasiveness (integrin-β1 [Int-B1], focal adhesion kinase [FAK]), and mitogen-activated protein kinase [MAPK] signaling (Fos, Jun; P values < 0.05). B: Increased β-catenin expression during adherence was confirmed by means of RT-PCR.

EMT Target Genes

Between 12 and 24 hr, the loss of adherent morphology by the MEE seam is observed and the acquisition of the mesenchymal phenotype begins in the component epithelial cells. Our results suggest that several molecular mechanisms are involved in this process. It appears that MEE cells are not able to divide during palatal EMT, as we observe that the expression levels of many key molecules involved in the cell cycle are inhibited (Fig. 6A), including cyclins (cyclin B1, B2, C, D2, E2, G1, G2, I), cyclin dependent kinases (cdk8), and others (cdc6). This finding was accompanied by the activation of growth (growth arrest specific 7) and cyclin-dependent kinase inhibitors (p15, p19, p21, p57).

Figure 6.

Cell cycle and apoptotic inhibition. A: Cell cycle inhibition was demonstrated with down-regulation of several cell cycle genes (Cdc6, CDK8, cyclin B1, B2, C, D2, E2, G1, G2, I) along with activation of cell cycle inhibitors (growth arrest specific 7 (GAS7), p15, p19, p21, p57). B: Reducing the possibility that apoptosis is occurring during the epithelial–mesenchymal transformation (EMT) phase of palatogenesis (12–24 hr), we have shown the inhibition of many key apoptotic genes (caspase 3, 7, 8, 9, Pcd2, death associated protein [DAP]) as well as up-regulation of antiapoptotic genes (defender against cell death 1 [Dad1], Birc2; P values < 0.05). A palatal section stained with biotin-16-UTP (terminal deoxynucleotidyl transferase–mediated deoxyuridinetriphosphate nick end-labeling [TUNEL]) confirms antiapoptosis during EMT of the medial edge epithelium (MEE) seam. C: Arrows demonstrate apoptotic bodies near the periphery but not in the seam (arrowhead). D: Reverse transcriptase-polymerase chain reaction analysis confirms down-regulation of caspases 3 and 7.

Although it has been suggested that apoptosis is the cause of disappearance of the MEE seam (Cuervo and Covarrubias, 2003), our results show otherwise (Fig. 6B), as we observe steady down-regulation of major apoptotic genes (caspase 3, 7, 8, 9, Pcd2, death associated protein [DAP]) and up-regulation of anti-apoptotic genes (Birc2, defender against cell death 1 [Dad1]). Presence of caspases at 12 hr may be associated with cell death and sloughing of the periderm component of the MEE seam, which is necessary for palatal adhesion (Fitchett and Hay, 1989). Validation of this profile was confirmed by biotin-16-UTP (terminal deoxynucleotidyl transferase–mediated deoxyuridinetriphosphate nick end-labeling [TUNEL]) staining of palatal sections during EMT, showing no apoptotic bodies within the transforming seam (Fig. 6C). RT-PCR analysis provided additional validation of caspase 3 and caspase 7 expression (Fig. 6D).

Extracellular matrix (ECM) is an essential factor involved in cell migration, invasion, and integrin signaling. We have shown that many of these genes are up-regulated during EMT (Fig. 7A), including fibronectin and collagen precursors (procollagen 1a1, 4a2, 6a1, 6a2, 7a1, 9a2, 15, 18a1), as well as the ECM glycoprotein fibulin (probe 1 and 2).

Figure 7.

Probable epithelial–mesenchymal transformation (EMT) (Smad2/4/LEF-1) target genes. An assessment of gene expression levels in cells that are approaching transformation. A,B: Extracellular matrix production (fibronectin [Fbn1], procollagen [Procol 1a1, 4a2, 6a1, 7a1, 9a2, 15, 18a1], fibulin [probe 1 and 2]; A) and increases in cytoskeletal gene expression (smooth muscle actin [SMA], γ2-actin, F-actin cap2b, β-filamin, talin, vimentin), along with down-regulation of the epithelial marker keratin 1-13 (probe 1 and 2; B) were observed as necessary elements for cellular elongation and migration. C: Other factors involved in integrin (integrin-linked kinase [Ilk], Src, integrin-β5, Akt1) and LEF-1 (cyclin D1) signaling, cellular invasion (matrix metalloproteinase-11, -12, -14), transforming growth factor-beta (TGF-β) signaling (ski), and intercellular adhesion (intercellular adhesion molecule [Int. Ad. Mol.], E-cadherin), were shown to have significant changes during the early stages of EMT (P values < 0.05). D: Loss of the epithelial marker E-cadherin and expression of mesenchymal markers vimentin and fibronectin (arrows) are shown in the MEE seam at 24 hr by means of immunohistochemistry. E-cadherin repressors such as Snail, Slug, and Twist do not appear to have a role in palatal EMT, as reverse transcriptase-polymerase chain reaction analysis demonstrates no up-regulation (E).

There are many cytoskeletal proteins that are necessary for pseudopod formation and cellular elongation, during acquisition of the mesenchymal morphology. We observe that most of these genes are up-regulated between 12 and 24 hr. They include smooth muscle actin (SMA), gamma-actin, F-actin cap2b, beta-filamin, talin, and vimentin. Also, the epithelial intermediate filament keratin 1-13 is down-regulated at the same time (Fig. 7B). Other genes involved in EMT (Fig. 7C) and cell migration, such as integrin-linked kinase (ILK; Novak et al., 1998), src, and integrin-β5 are shown to be up-regulated. Akt, which has been associated with EMT (Bakin et al., 2000), shows a significant increase between 12 and 24 hr. Although the cell cycle appears to be inhibited during EMT, cyclin D1 is up-regulated, which is not surprising considering that it is already known to be a gene that is activated by LEF-1 (Shtutman et al., 1999). Matrix metalloproteinases (MMP-11, 12, and 14), which digest the basal lamina, are also up-regulated. The TGF-β signaling inhibitor molecule ski (Prunier et al., 2003) is down-regulated, and many epithelial markers, such as intercellular adhesion molecule and E-cadherin are also inhibited. The presence of mesenchymal proteins such as vimentin and fibronectin, along with loss of E-cadherin, was confirmed by immunohistochemistry of palate sections at 24 hr (Fig. 7D). RT-PCR analysis of Snail, Slug, and Twist (E-cadherin repressors typically involved in EMT) demonstrated that they do not up-regulate during the EMT stage of palatogenesis (Fig. 7E).


Microarray analysis of varying stages of development makes use of the most modern technology to interpret cell signaling. Levels of mRNA detected by gene chip probes give a relatively accurate account of gene expression. Although these probes may not be as accurate for individual gene analysis as real-time RT-PCR or Northern blotting, this system is a powerful tool for assessing changes in the expression patterns of related genes. The major disadvantage is that it is highly unlikely that all genes of interest will be detected by the probes. Unfortunately, only those genes that are detected can be analyzed. Some genes that are known to be activated based on previous experiments may be absent on a genome wide screen. Also, there is a possibility that a lack of expected results could be caused by a defective probe within the gene chip. At best, one can only hope to establish a series of functional patterns within molecular association of genes present on any given chip.

Nevertheless, in our experiments, we were fortunate enough to obtain expression data for most of the EMT genes that are of interest. With this information, we can confirm that TGF-β signaling plays a significant role during the transformation of palate MEE cells into mesenchyme. In our previous studies, we have demonstrated that the EMT master gene (LEF-1) reaches peak expression at 36 hr during palate development when it orchestrates EMT under the influence of TGF-β3 (Nawshad and Hay, 2003). Unfortunately, we were unable to obtain a pure sample of MEE cells at this later time period due to difficulty of seam isolation at 36 hr using the Dispase method. The approach of using laser capture microdissection to obtain the remaining epithelium at 36 hr was attempted. However, although the laser capture method is sufficient for analysis of single genes of interest, we found that quality and purity of the total genomic RNA was sacrificed, possibly due to the fixation technique. The Dispase method, using living cells, was absolutely essential for maintaining the RNA quality required for optimal results in gene chip hybridization.

It should be noted that our graphic analyses are not scaled for total expression intensity. They do not represent the actual mRNA levels produced within the cells, but instead are based on a standard algorithm computed from averages of signal intensity, number of present genes, mRNA levels, and fold changes. These new normalized signal values present the most effective way to graphically visualize expression changes (personal communication; Affymetrix). Genes flagged as absent during initial analysis were given a value of zero to allow for appropriate comparisons. Also, only genes with P values less than 0.05 were used for individual gene analysis to ensure quality of data.

We were able to establish significant patterns in the changes taking place in many genes, which we expect continue as EMT progresses. Although the epithelial marker E-cadherin is present in all three stages of development, it should be noted that this was expected, as fully transformed mesenchymal cells lacking E-cadherin were not collected for analysis. We witness an increase in E-cadherin between 0 and 12 hr, which is undoubtedly a necessary element for palatal seam adherence. As time progresses from 12 to 24 hr, we begin to see a decrease in E-cadherin which should continue, as it appears to be fully inhibited upon completion of transformation (Sun et al., 1998). Surprisingly, the microarray analysis showed that other known repressors of E-cadherin such as Snail, Slug, and Twist, do not appear to up-regulate during EMT in the palate system (data not shown). In fact, our RT-PCR results confirm that Slug and Twist down-regulate during the EMT stage (12–24 hr). Questions still remain as to whether or not E-cadherin is directly inhibited by LEF-1 or through association with other molecules such as snail or ILK. Vimentin, the major mesenchymal intermediate filament, shows a steady increase from 12 to 24 hr and would be expected to continually increase through the peak of LEF-1 expression. Due to the high amount of extracellular matrix production, we assume that integrin signaling plays an important role in the migration of the transformed cells, leading to palatal confluence.

Although some studies on EMT insist that the Rho mitogen-activated protein kinase (MAPK; Bhowmick et al., 2001; Kaartinen et al., 2002), Notch (Zavadil et al., 2004), and NF-κB (Huber et al., 2004) signaling pathways play an important role in EMT, no true gene expression pattern could be found to support these hypotheses in our results. The data of Nawshad and Hay (2003) also appear to rule out a role for RhoA and MEK in palatogenesis, suggesting that the TGF-β/Smad–dependent pathway is the primary signaling for palatal EMT. Traditional Wnt isoforms also appeared to be absent from our array results, but since LEF-1 has been proven to be activated by Smads, rather than β-catenin, Wnt likely has no function in palatal EMT.

We were also able to establish some primary insights about the importance of palatal adherence. Previous experiments have shown that the addition of exogenous TGF-β3 to single palates could not transform their medial edge epithelium into mesenchyme (Sun et al., 1998a). Of interest, our microarray results indicate that endogenous levels of many important TGF-β signaling molecules are quite low in single palatal shelves and are probably insufficient to trigger gene expression to bring about EMT. Subsequent to adherence, the genes necessary to promote EMT up-regulate (Fig. 4C). These new findings could lead to identification of signal transduction pathways responsible for up-regulation of genes that facilitate TGF-β signaling.

While observation of the genetic regulation of molecules necessary for transformation to mesenchyme was the primary objective for this study, we were also able to document cellular preparation for post-EMT activities. Production of extracellular matrix proteins like fibronectin and collagen are necessary for interacting with integrin membrane receptors. These interactions cause focal adhesion formation with cytoskeletal proteins, leading to cell invasion and migration (Giancotti and Ruoslahti, 1999; Critchley, 2000). Our results show the up-regulation of many of the molecules involved in integrin signaling and focal adhesion formation. At 12 hr, we see increases in gene expression levels of integrin-β1, FAK, and vinculin, and at 24 hr, we begin to see up-regulations in talin, filamin, actin filaments, MMPs, src, and ILK. These results show an orderly progression of increased expression of genes needed to invade the extracellular matrix and migrate to bring about palatal confluence.

A major area for the future study of the role of TGF-β signaling in EMT is to determine which genes are targeted by the Smad2/4/LEF-1 complex versus which are targeted by only Smad2/4. Because TGF-β signal transduction occurs continually throughout EMT, both transcription factor complexes may target separate sets of genes. For instance, studies have shown that Smads directly target cyclin-dependent kinase inhibitors (p15, p21, and so on; Robson et al., 1999; Sherr and Roberts, 1999), whereas expression changes of molecules like E-cadherin and vimentin are considered to be (directly or indirectly) LEF-1–dependent during EMT (Kim et al., 2002). Because we have now obtained a significant list of target genes, we will undertake further studies to isolate individual genes affected by each targeting complex. However, although Smad2/4 may be responsible for mechanisms such as cell cycle inhibition, Smad2/4/LEF-1 is likely to be the primary transcription complex for the induction of EMT.


Dissection, Induction of Transformation, and Harvesting Cells

Fourteen-day-old mouse embryos were extracted from pregnant CF-1 mice (Charles River Laboratories) and kept in Hanks' buffer salt solution (Invitrogen) before removal of palates. Single palatal shelves were gently dissected from each embryo and placed on an organ culture dish lined with a 4% agarose/TSO gel. Single palates were then paired and placed together with their adjacent medial edge epithelia touching. An additional 100 μl of TSO medium (DMEM/F12 + 10% fetal bovine serum [FBS] + 1% penicillin/streptomycin + 0.1% gentamicin) was added to each culture dish to maintain a standard nutrient level. These full palates were incubated at 37°C for periods of 12 and 24 hr. Others were incubated for 12, 24, 36, and 60 hr, before being fixed with Bouin's solution, embedded in paraffin, sectioned, and stained with antibodies, hematoxylin and eosin, or TUNEL. Upon completion of each incubation period (12 and 24 hr), selected palates were removed, briefly washed with phosphate buffered saline (PBS), and placed in a separate culture dish containing Dispase II (2.4 U/ml; Roche Diagnostics). The palates were then incubated for 15 min at 37°C to promote Dispase activity in separating epithelium from mesenchyme. The resulting enzyme activity produced distinct differences in color and consistency between epithelium and mesenchyme, making it quite easy to visualize and isolate the fused medial edge epithelial seam. Palates were placed in F-12 medium (Sigma) containing 10% FBS. Each seam was removed and placed into a 1.5-ml microcentrifuge tube containing 350 μl of lysis buffer RLT (Qiagen) supplemented with 1% β-mercaptoethanol (Sigma). Palatal seams from six embryos were pooled together for each RNA sample. This process was then repeated with MEE from unfused single palates to obtain a control sample (0 hr). Samples of seam cells were cultured and anatomically analyzed by means of phase-contrast microscopy to ensure that all cells possessed an epithelial morphology, to avoid contamination by the underlying mesenchyme.

RNA Extraction and Purification

The 0-, 12-, and 24-hr samples were each homogenized in the lysis buffer by vortexing at maximum speed for 1 min. Total RNA isolation was performed on each sample using the RNeasy spin column RNA purification kit (Qiagen) and protocol. The 2-μl aliquots of each purified sample were diluted in 98 μl of 10 mM Tris buffer (pH 7.0) and subjected to ultraviolet spectrophotometry. Detection at 260 nm ensured adequate quantity of total RNA as well as purity (ratio: 1.8–2.1) when contrasted with 280 nm. Quality was also ensured by observing a graph of absorbance vs. wavelength plotted against a standard curve from a pure RNA control.

RNA Amplification and Biotin Labeling

The mRNA of each sample was isolated and amplified using the Riboamp RNA amplification kit (Arcturus) according to the manufacturer's protocol. The process of amplification is designed to produce single-stranded cDNA from the starting mRNA (isolated by a poly-dT primer in a reverse transcription reaction), followed by synthesis of double-stranded cDNA. An in vitro transcription (IVT) reaction was then performed, producing many copies of cRNA. A second round of amplification was carried out through the step of double-stranded cDNA synthesis. At this point, biotin labeling of RNA during the IVT reaction was achieved using the ENZO Bioarray high yield labeling kit (Affymetrix). Ultraviolet spectrophotometry was performed using aliquots from each sample after each round of amplification to ensure quantity, quality, and purity of the amplified RNA. Production of approximately 15 μg of biotin-labeled cRNA was necessary for gene chip hybridization.

Fragmentation, Hybridization, and Fluidics

Upon accumulation of 15 μg of labeled cRNA, 8 μl of a 5× fragmentation buffer was added to break down the cRNA to 35–200 base fragments, following the Affymetrix eukaryotic sample and array processing protocol provided in the Gene Chip Analysis technical manual. Each 40-μl fragmented sample, along with the corresponding murine U74A gene chips (Affymetrix), was then submitted to an Affymetrix core facility, where hybridization of the RNA to the chip probes and fluidics (washing and staining) were completed using the standard Affymetrix gene chip analysis protocol. Sample processing was completed using the Gene Chip Hybridization Oven 320 (Affymetrix) and Gene Chip Fluidics Station 400 (Affymetrix).

Scanning and Data Analysis

Completed chips were analyzed for expression intensity using an Affymetrix Gene Array Scanner (Hewlett-Packard). Raw data were compiled with Microarray Suite 5.0 software (Affymetrix). Extensive analysis of the raw data obtained from the hybridized chips was achieved using Genespring 6.0 software (Silicon Genetics).

Cell Culture and Antibodies

Medial edge epithelia from each stage of development were cultured using F-12 medium supplemented with 10% FBS and 1% penicillin/streptomycin, then used for immunohistochemical analysis. Blocking serum from rabbit or goat were applied at concentrations of 10% + 1% BSA in PBS. Primary antibodies were used for TGF-β3, LEF-1, Smad2, Smad4 (Santa Cruz), E-cadherin (Zymed), vimentin (Sigma), and fibronectin (Transduction Laboratories), at concentrations recommended by the manufacturers. Corresponding fluorescein- and rhodamine-conjugated secondary antibodies (Pierce) were used at concentrations of 1:250.

In Situ End Labeling (TUNEL)

This method was used to detect the apoptotic bodies by replacing UTP for RNA labeling with biotin-16-UTP, the substrate for SP6, T3, and T7 RNA polymerases. This preparation was used as a substrate for RNA polymerases and for RNA labeling to replace UTP in transcription. Linearized template DNA was transcribed with the corresponding polymerases using ATP, GTP, CTP, UTP, and biotin-16-UTP. Labeled RNA was subsequently detected with the streptavidin–alkaline phosphatase conjugate for nucleic acid detection.

Each section was deparaffinized in xylol and dehydrated in absolute, 90% and 70% ethyl alcohol, before digestion with 0.5% trypsin (Sigma) in 0.1 M HCl for 5 min at 37°C. The probe mixture for biotin-16-UTP: Buffer (DNA Buffer 100 μl, dH2O 900 μl) Probe: (dATP 10 μl, dCTP 10 μl, dGTP 10 μl, biotin-16-UTP 16 μl, and Klenow 3 μl) was processed by adding the whole probe mixture into 975 μl of DNA buffer. Incubation with streptavidin–horseradish peroxidase complex (Dako) diluted 1:100 in PBS for 1 hr at 25°C, was followed by incubation with 0.05 mg/ml diaminobenzidine in PBS containing 1% H2O2 (Zymed) for 2–4 min.

Between each step, sections were washed three times with PBS buffer (pH 7.3). All incubations were performed in a humidified chamber. Sections were counterstained in Mayer's hematoxylin. Controls included (1) omission of probe, replacing with PBS buffer; (2) replacement primary probe with an irrelevant specificity but of the same epitope and concentration; and (3) sections of mouse liver (hepatocytes) were used as a positive control for biotin-16-UTP (Schulte-Hermann et al., 1995).


Reverse transcription of total RNA and subsequent PCR was performed using the SuperScriptIII RT-PCR kit (Invitrogen) according to the manufacturer's protocol. Reactions were induced using a PTC-100 thermal cycler (MJ Research, Inc.), with 30 cycles per sample. Samples were run on 1.5% agarose gels containing ethidium bromide (1.5 μl) and visualized with a ChemiDoc XRS Imager (Bio-Rad). Sequences of primers used are as follows: Smad2 (sense, 5′-TCACAGTCATCATGAGCTCAAGG-3′; antisense, 5′-TGTGACGCATGGAAGGTCTCTC-3′); Smad3 (sense, 5′-GAGTAGAGACGCCAGTTCTACC-3′; antisense, 5′-GGTTTGGAGAACCTGCGTCCAT-3′); Smad4 (sense, 5′-TTCCAGCCTCCCATTTCC-3′; antisense, 5′-TAAGGCACCTGACCCAAA-3′); LEF-1 (sense, 5′-AGACACCCTCCAGCTCCTGA-3′; antisense, 5′-CCTGAATCCACCCGTGATG-3′); β-catenin (sense, 5′-AGCCGAGATGGCCCAGAAT-3′; antisense, 5′-AAGGGCAAGGTTCGAATCAA-3′); caspase3 (sense, 5′-TGGGCCTGAAATACCAAGTC-3′; antisense, 5′-CACCCCCAATCATTCCTCTA-3′); caspase7 (sense, 5′-GGATCCGAACGATGACCGATGATCAG-3′; antisense, 5′-AAGCTTGTGAGCATGGACACCATAC-3′); Snail (sense, 5′-CAGCTGGCCAGGCTCTCGGT-3′; antisense, 5′-GCGAGGGCCTCCGGAGCA-3′); Slug (sense, 5′-GCCTCCAAAAAGCCAAACTA-3′; antisense, 5′-CACAGTGATGGGGCTGTATG-3′); Twist (sense, 5′-CGGGTCATGGCTAACGTG-3′; antisense, 5′-CAGCTTGCCATCCTTGGAGTC-3′); GAPDH (sense, 5′-GACCCCTTCATTGACCTCAAC-3′; antisense, 5′-CTTCTCCATGGTGGTGAAGA-3′).


The data of the microarray results have been submitted to Gene Expression Omnibus (GEO), NCBI, NIH (accession no. GSE2537) and can be accessed from their Web site at (http://www.ncbi.nlm.nih.gov/geo). We thank Trent Rector of the Biopolymers facility at Harvard Medical School: Department of Genetics for his assistance in gene chip hybridization, ultraviolet spectrophotometry, and data analysis. E.D.H. was funded by the US Department of Public Health.