In vivo and In vitro Evidence for Transforming Growth Factor-β1-Mediated Epithelial to Mesenchymal Transition in Esophageal Adenocarcinoma

Esophageal adenocarcinoma is increasing in incidence (1). The majority of patients present with locally advanced or metastatic disease, and as a result, the outcome for these patients is poor with an overall 5-year survival rate of 13% (2).

Many molecular and phenotypic events underlie the development of invasion and metastases (3). These include the loss of proliferative control, enhanced cellular migration and invasion, extracellular matrix degradation, angiogenesis, lymphangiogenesis, and vascular invasion followed by the distant seeding of tumor cells in specific tissues, such as the liver or lung.

One of the mechanisms by which epithelial cells acquire the motile properties required for invasion is thought to be epithelial to mesenchymal transition (EMT; ref. 4). The epithelial state is characterized by the expression of the cell adhesion molecule E-cadherin (5) and cytokeratins, such as cytokeratin 18 (CK18; ref. 6). During EMT, epithelial cell-cell contact is decreased by the down-regulation of cytoskeletal components and the cell morphology becomes more fibroblast-like with up-regulation of mesenchymal markers, including α-smooth muscle actin (α-SMA; ref. 7) and vimentin (8). EMT promotes cellular motility, invasion, and cytoskeletal rearrangement in a range of tumor cells in vitro (9, 10). Immunohistochemical evidence for changes in the expression and localization of adhesion molecules and extracellular matrix proteins at the invasive margin of colorectal tumors suggests that EMT also occurs in vivo (6, 11).

Transforming growth factor-β1 (TGF-β1) is an important inducer of EMT (8). Several signaling pathways have been implicated in this process, including c-met, Src, Ras, integrin-linked kinase (ILK), and the SMADs (4). The effectors of these pathways act on the small GTPases Rho and Rac as well as acting directly on the transcriptional repressors of E-cadherin expression. The transcriptional repressors are zinc finger proteins, including snail, SMAD-interacting protein 1 (SIP1; ref. 12), slug, and the more recently described LIV1 [a member of a subfamily of Zrt-like, Irt-like proteins (ZIP) zinc transporters] also termed LZT (LIV1 subfamily of ZIP zinc transporters; ref. 13). In addition, it has been shown that activation of the AKT/mitogen-activated protein kinase (MAPK)–dependent pathways may act directly on the E-cadherin transcriptional repressors SIP1 and snail (12, 14).

As the molecular steps involved in EMT have been elucidated, it has also been recognized that the reversal of this process, so-called mesenchymal to epithelial transition (MET), may offer a potential therapy for the control of metastases. Bone morphogenetic protein 7 (BMP7), another member of the TGF-β1 superfamily, has been shown to reverse TGF-β1-induced EMT in both in vitro and in vivo models (15). In this context, BMP7 acts through the canonical SMAD pathway, recruiting SMAD1, SMAD5, and SMAD8 rather than SMAD2 and SMAD3 (16). More recent evidence in Xenopus suggests that it may also act via the MAPK pathway (17).

There is some preliminary in vitro evidence for the role of EMT in the pathogenesis of esophageal adenocarcinoma (18). This study was designed initially to find in vivo evidence for EMT by comparing the immunohistochemical characteristics of epithelial and mesenchymal markers at the invasive margin compared with the central portion of esophageal tumors. An esophageal in vitro model system for EMT was then developed to investigate the phenotypic, molecular, and functional effects of TGF-β with or without BMP7. This system was then used to examine the underlying mechanisms for EMT by analysis of candidate pathways (AKT, MAPK, SMAD2, SMAD3, SIP1, and snail activity) and obliteration of the effect using RNA interference (RNAi) of SMAD4.

Upper gastrointestinal tumor samples from 167 patients were obtained from surgical resection specimens collected at Bristol Royal Infirmary (Bristol, United Kingdom) following Central and South Bristol local research ethics committee approval. Three 7-μm frozen sections were taken from these specimens and stained with H&E. The original histologic diagnosis used for clinical management was confirmed by an independent, experienced, and expert upper gastrointestinal histopathologist (V.E.S.). Cases were selected based on a confirmed diagnosis of esophageal adenocarcinoma and the presence of both an invasive margin and a central tumor area on the same section. Several criteria to verify the presence of an invasion front were assessed. Firstly, H&E criteria were used to identify single cells, which appeared to be at the tumor edge. Samples with invasion fronts identified on H&E criteria were then subjected to immunohistochemistry using (a) MNF116, a pan cytokeratin marker (1:50, clone MNF116; DakoCytomation, Ely, United Kingdom) to identify epithelial cells, and (b) CD45, a hematopoietic cell marker (1:100, clones 2B11 and PD7/26; DakoCytomation) to confirm that cells identified as possibly invasive were not in fact inflammatory. Ten adenocarcinoma samples from 10 different patients that fulfilled these stringent criteria were selected, and a further thirty 7-μm sections of these samples were then cut. An additional H&E section was stained, and the remaining sections were used for immunohistochemistry. Tissue was fixed with acetone (100%) for 10 minutes and rehydrated with ethanol. Slides were blocked with 10% horse serum for 1 hour at room temperature and incubated with E-cadherin (1:2,500, clone 36; BD Biosciences, Franklin Lakes, NJ), CK18 (1:40, clone 5D3; Novocastra, Newcastle-upon-Tyne, United Kingdom), α-SMA (1:50, clone 1A4; Abcam, Cambridge, United Kingdom), vimentin (1:50, clone V9; Novocastra), TGF-β (1:40, clone TGFB17; Novocastra), or BMP7 (1:160, clone 164311; R&D Systems, Minneapolis, MN) antibodies overnight at 4°C. Slides were incubated with anti-mouse biotinylated secondary antibody (1:250; Vector Laboratories, Burlingame, CA). Antibody detection was achieved with anti-mouse horseradish peroxidase (DAKO Ltd., Ely, United Kingdom) followed by 3,3′-diaminobenzidine (Vector Laboratories) for 60 seconds and a hematoxylin counterstain. Each slide had at least three replicate sections for each antibody. Quantitative analysis of E-cadherin and α-SMA staining in the central tumor area compared with the invasive margin was done by two independent observers (J.R.E.R. and V.E.S.). The invasive front was identified and the area of central tumor most distant from this invasive front was then selected, and three independent fields were scored for staining intensity and for cellular localization of E-cadherin staining. The intensity of staining was scored as 0 (none), 1 (weak), 2 (mild), 3 (moderate), and 4 (strong) compared with a negative (no primary antibody) and positive control [E-cadherin (adjacent normal esophagus) and α-SMA (adjacent normal artery)]. The cellular localization of E-cadherin was classified as membranous, cytoplasmic, or mixed.

The ability of TGF-β1 treatment to induce EMT was assessed across a panel of esophageal cell lines: OE33, a junctional esophageal adenocarcinoma [European Collection of Animal Cell Cultures (ECACC), Porton Down, United Kingdom]; TE7, Barrett's adenocarcinoma (gift from T. Nishihira, Kurokawa County Hospital, Kurokawa, Japan); and KYSE-30, an esophageal squamous cell carcinoma (ECACC). BIC-1 (gift from D. Beer, University of Michigan, Ann Arbor, MI), a SMAD4-deficient esophageal adenocarcinoma, which is generally unresponsive to TGF-β1 (19, 20), was used as a negative control. TE7 and OE33 cells were maintained in RPMI 1640 (Sigma-Aldrich Co. Ltd., Dorset, United Kingdom) supplemented with 10% fetal bovine serum (FBS), 100 units/mL penicillin, 100 mg/mL streptomycin, and 2 mmol/L glutamine. BIC-1 cells were maintained in DMEM (Invitrogen, Paisley, United Kingdom) supplemented with 10% FBS, 100 units/mL penicillin, 100 mg/mL streptomycin, and 2 mmol/L glutamine. KYSE cells were maintained in DMEM/RPMI 1640 (1:1 mixture), both with 2% FBS, 100 units/mL penicillin, 100 mg/mL streptomycin, and 2 mmol/L glutamine.

EMT induction and reversal was undertaken using a modification of methods described by Okada et al. (21) and Strutz et al. (22). Cells were seeded into six-well plates or onto 8-mm glass coverslips in 24-well plates at 70% confluency and incubated in standard medium for 48 hours. Cells were then incubated in serum-free medium supplemented with 5 μg/mL transferrin, 5 μg/mL insulin, 5 × 10⁻⁸ mol/L hydrocortisone, and 10 ng/mL endothelial growth factor (EGF) at 37°C, 5% CO₂ atmosphere for 96 hours with TGF-β1 over the concentration range 0, 0.005, 0.5, and 5 ng/mL with daily replacement of the culture medium. These concentrations are within the physiologic range and lower than certain extracellular fluids (pleural) in acute infective and fibrotic states (8.1-39.6 ng/mL; ref. 23).

After 96 hours of TGF-β1 treatment, the medium was replaced with serum-free medium supplemented with 100 ng/mL recombinant human BMP7 (R & D Systems, Minneapolis, MN; ref. 24) in place of TGF-β1. The culture medium was changed daily for 48 hours. Control experiments were also performed, in which TE7 cells were pretreated with TGF-β1 and then cultured in serum-free medium supplemented with 5 μg/mL transferrin, 5 μg/mL insulin, 5 × 10⁻⁸ mol/L hydrocortisone, and 10 ng/mL EGF but not BMP7. All experiments were undertaken in triplicate on three occasions.

Following treatment with TGF-β1 with or without BMP7, cells on coverslips were fixed with 4% paraformaldehyde for 20 minutes followed by ice-cold 100% methanol for 5 minutes at −20°C, washed with PBS, and blocked with 10% horse serum in PBS for 30 minutes. Slides were incubated with a 1:5,000 dilution of E-cadherin monoclonal antibody (clone 36) for 1 hour at room temperature and then washed and incubated with a 1:1,000 dilution of FITC-conjugated anti-mouse IgG (Vector Laboratories, Cambridgeshire, United Kingdom). Nuclei were counterstained with TO-PRO-3 iodide (Invitrogen, Paisley, United Kingdom) and examined using an Axion LSM 10 laser confocal microscope (Carl Zeiss, Oberkochen, Germany). Quantification of fluorescence was undertaken using the LSM 5 software package (Carl Zeiss).

Phase-contrast microscopy was undertaken using MatTek 35-mm glass-bottomed dishes (MatTek Corporation, Ashland, MA) using a Zeiss Axiovert 200 M microscope (Carl Zeiss) at ×20 and ×40 magnifications. Images were captured using Volocity 3.5.1 software (Improvision, Coventry, United Kingdom).

Protein lysates were prepared with ice-cold lysis buffer [20 mmol/L Tris (pH 7.4), 1% (octylphenoxy)polyethoxyethanol (Igepal CA-630), 1% Triton X-100, 50 mmol/L NaF, 50 mmol/L NaCl, 1 mmol/L EDTA (pH 8), 30 mmol/L sodium pyrophosphate] containing protease inhibitors (1 Complete tablet per 50 mL; Roche, Mannheim, Germany). Following quantification (bicinchoninic acid protein assay, Sigma-Aldrich), 25 μg protein was separated by gel electrophoresis on 8% polyacrylamide gels and transferred to polyvinylidene difluoride membranes (GE Healthcare, Little Chalfont, United Kingdom). Membranes were incubated overnight at 4°C with the following antibodies: phosphorylated SMAD2/SMAD3 (1:500, Ser⁴³³/Ser⁴³⁵; Santa Cruz Biotechnology, Inc., Santa Cruz, CA) and total-SMAD2/SMAD3 (1:200, clone sc-7960; Santa Cruz Biotechnology). Western blot images were digitized using a Hewlett-Packard PSC1310 scanner (Hewlett-Packard, Bracknell, United Kingdom) for densitometry. Mean band densities were deduced (mean band intensity = absolute band intensity − background intensity) using Kodak 1D v3.5.4 software (Eastman Kodak, New Haven, CT). Data from three independent experiments were analyzed, and intensity relative to total SMAD2/SMAD3 was calculated.

Total RNA was isolated using Trizol reagent (Invitrogen). RNA (2 μg) was reverse transcribed, and 2 μL cDNA was amplified as follows: 30 cycles of 94°C for 30 seconds, 55°C for 30 seconds, and 72°C for 45 seconds for SMAD4 [5′-CCAGGATCAGTAGGTGGAAT-3′ (forward) and 5′-GTCTAAAGGTTGTGGGTCTG-3′ (reverse)] and 28 cycles of 94°C for 30 seconds, 60°C for 30 seconds, and 72°C for 45 seconds for glyceraldehyde-3-phosphate dehydrogenase [GAPDH; 5′-GCAGGGGGGAGCCAAAAGGG-3′ (forward) and 5′-TGCCAGCCCCAGCGTCAAAG-3′ (reverse)]. PCR products were analyzed on 1.4% agarose gels and stained with ethidium bromide.

Real-time PCRs were optimized by melt-curve analysis, and efficiency was calculated by serial dilution. cDNA (2 μL, diluted 1:5) was amplified in a 20 μL volume containing 10 μL of Sigma SYBR Green PCR Master Mix (Sigma-Aldrich) and 0.2 μmol/L final concentration of each primer. Primers used were designed using the PerlPrimer software package (http://perlprimer.sourceforge.net/). Triplicate reactions were done in an Applied Biosystems 7900HT thermal cycler using the conditions of initial enzyme activation of 2 minutes at 95°C followed by 40 cycles of 30 seconds at 95°C, 15 seconds at 58°C, and 15 seconds at 72°C. Following PCR, the threshold cycle (C_T) was obtained and relative quantities were determined for each sample normalized to GAPDH using the formula: relative transcript abundance = 10,000 / 2^{(CT gene − CT GAPDH)} (25). The primers used are described in Supplementary Table S1.

Two different, predesigned small interfering RNA (siRNA) duplexes of SMAD4 (Genbank accession no. NM_005359) were selected: siRNA1 (sense, 5′-GCCAUAGUGAAGGACUGUUtt-3′) targeting exon 7 and siRNA2 (sense, 5′-CCUAUGUUAUUUUGUGUACtt-3′) targeting exon 13 from Ambion, Inc., (Austin, TX). Both nonspecific control siRNA duplexes with a similar GC content as SMAD4 siRNAs (Ambion) and an oligonucleotide-free reaction were used. The siRNAs were transfected into human TE7 cells at a final concentration of 200 nmol/L using Oligofectamine (Invitrogen, Carlsbad, CA). Control experiments were also done using transfection reagent alone and an empty vector. Experiments were done in triplicate. To determine the efficiency and persistence of the siRNA knockdown, transfected cells were serially collected at 24-hour intervals from days 0 to 4 and mRNA levels of SMAD4 were assessed with real-time PCR.

For each of the following assays, cells were incubated with standard medium supplemented with 5 μg/mL transferrin, 5 μg/mL insulin, 5 × 10⁻⁸ mol/L hydrocortisone, and 10 ng/mL EGF.

Matrigel two-chamber invasion assay. Invasiveness of treated cells was assessed using Transwell chambers with a 6.5-mm polycarbonate filter (8-μm pore size; Becton Dickinson Labware, Oxford, United Kingdom) membrane as described previously (26). The filter was precoated with 10 mg/mL Matrigel (BD Biosciences) and 25 μg/mL plasminogen (Sigma-Aldrich) and then incubated for 1 hour. Cells (5 × 10⁴) were placed in the upper chamber in EMT medium. A chemoattractant was placed in the lower chamber (750 μL serum-containing standard medium), and the system was incubated for 24 hours. After incubation, the membrane was fixed with 100% methanol, stained with hematoxylin, excised from the Transwell, and mounted on a microscope slide. The slides were examined using an Olympus BX41 microscope (Olympus, Hamburg, Germany), and the absolute invading cell number was counted. At least five replicates were done, and the experiment was repeated on four occasions.

Slow aggregation assay. Cells were trypsinized, and 20,000 cells were transferred onto an agar gel (0.66%, w/v) in a 96-well plate in a total volume of 100 μL of EMT medium. After a 24-hour incubation, cells were examined using an inverted microscope (Zeiss HBO 50, Carl Zeiss). Aggregate formation was assessed and scored using the following schema: 2, no aggregates; 1, small aggregate; and 0, single large aggregate (27). The assessment was repeated 10 times for each set of conditions, and the experiments were repeated at least thrice.

In vitro wounding assays. Wound-healing experiments were done as described by Watanabe et al. (28) with modifications. Following TGF-β1 or BMP7 treatment, medium was removed and a standardized, artificial wound was made by mechanical denudation with a rotating tip fitted with a 10 μL pipette propelled by a small electric drill (MB186; Minicraft, Greenville, WI). The wounded cell monolayer was washed twice with PBS and incubated in EMT medium. Wound size was deduced using an eyepiece graticule previously calibrated using a hemocytometer grid at ×10 magnification using a Zeiss HBO 50 inverted microscope. Measurements of the longest and shortest perpendicular axis were taken at 0 and 24 hours, and the wound area was estimated by multiplying the product of the two axial measurements by π/4. The absolute change in wound size was deduced, and percentage healing at 24 hours was then calculated. Experiments were done in quadruplicate and repeated at least thrice.

Time-lapse cell tracking. Time-lapse cell tracking was undertaken with cells seeded on MatTek 35-mm glass-bottomed dishes using a Zeiss Axiovert 200 M microscope at ×40 magnification. Images were captured every 3 minutes over 7 hours, and movements were quantified using Volocity 3.6.0 software (Volocity, Coventry, United Kingdom).

Data were analyzed using GraphPad Prism and Microsoft Excel software. The Kruskal-Wallis test and ANOVA test were used to compare values between different groups, and the Student's t test was used to analyze single specific differences of biological interest. The Kruskal-Wallis test combined with the Dunn's multiple comparison test was used to identify multiple specific differences. All data were expressed as a mean (SE), and P < 0.05 was considered statistically significant.

Increased expression of mesenchymal markers at invasion front. Immunohistochemical analysis of the invasive component of adenocarcinoma specimens was compared with the central area of the tumor. There was down-regulation of the intensity of the epithelial staining with E-cadherin and CK18 at the invasive tumor margin (Fig. 1A). Furthermore, there was a redistribution of E-cadherin staining from a predominantly membranous pattern in the central area (77.8% of cells) to a predominantly cytoplasmic pattern at the margin (66.9% of cells; Fig. 1A). In contrast, there was a low intensity of both cytoplasmic α-SMA and vimentin staining in the center of the tumor compared with increased intensity of expression at the invasive margin (Fig. 1A). Quantification of E-cadherin and α-SMA staining intensity showed down-regulation of E-cadherin at the invasive margin (P = 0.0003.) and a contrasting up-regulation of α-SMA staining (P = 0.000001). Staining for TGF-β showed a predominantly stromal expression pattern in both the central and invasive tumor components with foci of increased uptake in the invasive front (Fig. 1A). BMP7 expression was almost completely absent at the invasive margin, whereas occasional darkly staining foci were shown in the central tumor area (Fig. 1A).

Effects of TGF-β and BMP7 on molecular and phenotypic characteristics. To develop an in vitro model, immunocytochemical evidence for EMT was sought in TE7, KYSE, OE33, and the SMAD4-deficient BIC cell line. The initial data suggested that convincing immunocytochemical and morphologic characteristics of EMT occurring after 4 days of treatment were unique to TE7 cells. Therefore, TE7 cells were selected for further study. TE7 cells treated with TGF-β1 over the concentration range 0.005 to 5 ng/mL resulted in a significant decrease in E-cadherin immunofluorescence (691.2 ± 36.4 fluorescent units without TGF-β1 compared with 206.6 ± 37.9 fluorescent units with 5 ng/mL TGF-β1; P < 0.0001, mean of three experiments assessing three independent areas of each slide; Fig. 2A,, i and ii; Supplementary Fig. S1A). Phase-contrast microscopy revealed a dose-dependent loss of the adherent phenotype with cellular elongation, decrease in cell-to-cell contact, and the induction of a fibroblast-like state (Fig. 2B,, i and ii; Supplementary Fig. S1B). When TGF-β treatment was followed by the replacement of TGF-β with its antagonist BMP7, there was a recovery of E-cadherin immunofluorescence (Fig. 2A,, iii; Supplementary Fig. S1C). The molecular changes were accompanied by a change from a mesenchymal phenotype back to an epithelial morphology as seen by the lack of pseudopodia and the return of cell-cell adhesion (Fig. 2B,, iii; Supplementary Fig. S1D). BMP7 treatment alone without prior TGF-β did not alter the immunofluorescent or morphologic cell characteristics (Fig. 2A  and B, iv).

Real-time reverse transcription-PCR (RT-PCR) was done to quantify these molecular changes. At all concentrations of TGF-β treatment (0.005-5 ng/mL), there was a 4-fold decrease in E-cadherin mRNA expression (P = 0.046, compared with untreated cells) and a 9-fold reduction in CK18 expression at the highest TGF-β1 treatment dose (P = 0.012; Fig. 3A). In contrast, expression of the mesenchymal markers α-SMA and vimentin increased by ∼2-fold following TGF-β1 treatment (P = 0.024; Fig. 3B). When TGF-β1 was followed by BMP7 treatment, the epithelial markers E-cadherin and CK18 were up-regulated by up to 8-fold (P = 0.05) and 2-fold, respectively (Fig. 3C). Under the same conditions, expression of the mesenchymal marker α-SMA no longer increased (Fig. 3D), whereas vimentin was suppressed at lower TGF-β1 doses (P = 0.037; Fig. 3D).

In control conditions, when TGF-β1 was removed from the medium and no BMP7 was added, there was no increase in E-cadherin above untreated levels by day 6, whereas α-SMA decreased and vimentin remained unchanged. In addition, there was no return to an epithelial morphology (data not shown).

Effects of TGF-β1 signaling via SMAD2/SMAD3 phosphorylation and SIP1 expression. Because TGF-β1-mediated SMAD signaling occurs through phosphorylation of the SMAD2/SMAD3 complex (29), the effect of TGF-β1 treatment on SMAD2/SMAD3 phosphorylation was examined by immunoblotting. This revealed an increase in phosphorylated SMAD2/SMAD3 protein levels, which was reduced by subsequent BMP7 treatment, whereas nonphosphorylated SMAD2/SMAD3 level did not change (Fig. 4A). Analysis of three independent experiments using band densitometry confirmed these findings (Fig. 4B). These changes were not seen when TGF-β1 was withdrawn without the addition of BMP7 (data not shown). Expression of the E-cadherin transcriptional repressor snail was not significantly altered in a dose-dependent manner by TGF-β1 treatment (data not shown), but SIP1 was potently induced (P = 0.013; Fig. 4C). Following BMP7 treatment, SIP1 was down-regulated at the mRNA level (P = 0.044; Fig. 4C). There was no up-regulation of MAPK and AKT expression at the mRNA or protein level for all concentrations of TGF-β1 treatment, suggesting that these pathways were not implicated (data not shown).

SMAD4 knockdown with siRNA prevents TGF-β1-induced EMT and abrogates EMT-associated invasive behavior. To confirm the role of canonical TGF-β1 signaling in this system, we then used siRNA against SMAD4. SMAD4 was selected because it forms a single, common point within the canonical TGF-β1 pathway (30). RNAi for SMAD4 resulted in a >95% knockdown, which persisted at 4 days (P = 0.0009; Fig. 5A,, inset). SMAD4 RNAi abrogated the TGF-β-induced suppression of E-cadherin expression (P = 0.002; Fig. 5B) and suppressed α-SMA expression at all doses of TGF-β treatment (P = 0.02; Fig. 5C).

Effects of TGF-β1-induced EMT in the development of invasive and migratory phenotypes. Treatment with 5 ng/mL TGF-β1 resulted in a significant increase in cell invasion (Matrigel two-chamber invasion assay; P = 1.63 × 10⁻⁶), a statistically significant decrease in aggregation as marked by an increase in the aggregation score (slow aggregation assay; P = 1.15 × 10⁻¹¹), an increase in healing (in vitro wounding assay; P = 3.2 × 10⁻⁶), and promotion of cell motility from 2.6 μm/s in untreated cells to 3.4 μm/s in cells treated with TGF-β1 at 5 ng/mL (time-lapse live cell imaging; P < 0.001; all results in Table 1). These effects were abrogated by the addition of BMP7 or SMAD4 RNAi for each assay: invasion (P = 7.56 × 10⁻⁶), aggregation (P = 1.43 × 10⁻⁷), healing (P = 0.003), and motility (P = 0.0081; Table 1).

These in vivo data show immunohistochemical evidence for EMT, which is associated with TGF-β1 expression, at the invasion front of esophageal adenocarcinoma. A TGF-β1-induced in vitro model of EMT has shown morphologic, molecular, and functional evidence for this process that was reversible by BMP7 and SMAD4 RNAi.

The differential expression of epithelial and mesenchymal markers has been recognized in several rare tumors of mixed phenotype (31, 32). More recently, this differential expression of epithelial and mesenchymal markers at the invasive margin has been described in colorectal and hepatocellular tumors (11, 33, 34), suggesting that EMT might be a feature of the invasive characteristic of epithelial tumors. In addition, changes suggestive of EMT have been described in cell lines, such as nonmalignant Madin-Darby canine kidney cells, and cell lines derived from the pancreas (35) and breast (36) cancers.

The molecular and phenotypic changes from an epithelial to a mesenchymal cell type seem to be functionally relevant because several studies have shown that EMT is important in cancer progression (4, 37). Loss of the classic epithelial marker E-cadherin is associated with poor outcome in several tumor sites, including non–small cell lung cancer, invasive ductal breast carcinoma (38), and gastric adenocarcinoma (39). In tumors of the esophagus and gastroesophageal junction, disturbances in E-cadherin expression have been correlated with increasing invasive capacity, dedifferentiation, and lymph node metastases (40). Loss of E-cadherin expression is also a feature in the transition from dysplastic Barrett's esophagus to invasive adenocarcinoma (41).

TGF-β1 normally has antiproliferative effects across a broad range of cell lineages. In cancer progression, the antiproliferative and homeostatic effects of TGF-β1 are frequently lost and invasion was promoted through an ability to induce EMT (42, 43). TGF-β1 has been implicated in the development of several gastrointestinal tract tumors, including colonic (44) and pancreatic (45) cancers. It has recently been shown that there is loss of the antiproliferative response to TGF-β1 in a panel of esophageal adenocarcinoma cell lines with promotion of invasive behavior (46). The possibility that TGF-β1 might promote invasion by the induction of EMT has not previously been investigated either in vivo or in vitro in the context of esophageal adenocarcinoma.

TGF-β1 can induce EMT by several mechanisms. It can occur by canonical TGF-β1 signaling (47), which alters the function of the E-cadherin transcriptional repressors snail and SIP1 directly (4, 39). Alternatively, EMT can be induced by components of the TGF-β1 signaling pathway acting either on ILK (48, 49) or by ILK-dependent phosphorylation of AKT (50). Both of these mechanisms act via alterations in the transcriptional activity of snail. In the in vitro model system used in the present experiments, TGF-β1 treatment activated the canonical TGF-β1 signaling pathway with induction of phosphorylated SMAD2/SMAD3 protein expression (Fig. 5A). In addition, TGF-β1 induced up-regulation of SIP1 but not snail expression, suggesting that SIP1-mediated E-cadherin repression was more relevant (Fig. 5B). SMAD4 RNAi abrogation of both expression and functional effects of TGF-β1 adds further support to the role of canonical SMAD signaling in this setting.

We have also shown that BMP7, another member of the TGF-β1 superfamily, potently reverses the molecular, phenotypic, and functional effects of TGF-β1-induced EMT in vitro (Figs. 2-4 and Table 1). This is consistent with observations in a murine model of chronic renal injury (renal fibrosis) in which systemic administration of recombinant human BMP7 led to repair of severely damaged renal tubular epithelial cells (15), a change associated with the induction of MET by the formation of epithelial cell aggregates in adult renal fibroblasts in vivo (24). However, the effects of BMP7 seem to be context specific. For example, in a hyperplastic cell line (BPH-1), BMP7 induced cell cycle arrest, and in two malignant cell lines (PC-3 and LNCaP), it induced EMT and apoptosis, respectively (51).

In summary, we have presented data to suggest that EMT may be relevant in esophageal adenocarcinoma. Our data are limited to immunohistochemical analysis of human tissues and a detailed functional analysis in a single cell line. In future, it would be useful to establish in vivo models of esophageal adenocarcinoma where evidence for EMT can be sought. Although such models are currently lacking, an in vivo approach would permit manipulation of TGF-β1 signaling and examination of the role of BMP7 in metastasis prevention with the potential of developing novel compounds for therapeutic purposes.

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.

We are greatful to Richard Hardwick for his general support. We thank Dr. Pierre Lao-Sirieix for his assistance in managing the references and Paul B. Savage for his assistance in the identification and cataloguing of the human samples used.