Ionizing Radiation Predisposes Nonmalignant Human Mammary Epithelial Cells to Undergo Transforming Growth Factor β–Induced Epithelial to Mesenchymal Transition Free

Transforming growth factor β1 (TGFβ) is a tumor suppressor during the initial stage of tumorigenesis, but it can switch to a tumor promoter during neoplastic progression. Ionizing radiation (IR), both a carcinogen and a therapeutic agent, induces TGFβ activation in vivo. We now show that IR sensitizes human mammary epithelial cells (HMEC) to undergo TGFβ-mediated epithelial to mesenchymal transition (EMT). Nonmalignant HMEC (MCF10A, HMT3522 S1, and 184v) were irradiated with 2 Gy shortly after attachment in monolayer culture or treated with a low concentration of TGFβ (0.4 ng/mL) or double treated. All double-treated (IR + TGFβ) HMEC underwent a morphologic shift from cuboidal to spindle shaped. This phenotype was accompanied by a decreased expression of epithelial markers E-cadherin, β-catenin, and ZO-1, remodeling of the actin cytoskeleton, and increased expression of mesenchymal markers N-cadherin, fibronectin, and vimentin. Furthermore, double treatment increased cell motility, promoted invasion, and disrupted acinar morphogenesis of cells subsequently plated in Matrigel. Neither radiation nor TGFβ alone elicited EMT, although IR increased chronic TGFβ signaling and activity. Gene expression profiling revealed that double-treated cells exhibit a specific 10-gene signature associated with Erk/mitogen-activated protein kinase (MAPK) signaling. We hypothesized that IR-induced MAPK activation primes nonmalignant HMEC to undergo TGFβ-mediated EMT. Consistent with this, Erk phosphorylation was transiently induced by irradiation and persisted in irradiated cells treated with TGFβ, and treatment with U0126, a MAP/Erk kinase (MEK) inhibitor, blocked the EMT phenotype. Together, these data show that the interactions between radiation-induced signaling pathways elicit heritable phenotypes that could contribute to neoplastic progression. [Cancer Res 2007;67(18):8662–70]

An emerging concept in cancer biology is that carcinogens can compromise tissue integrity by eliciting altered phenotypes and tissue composition (reviewed in ref. 1). Intercellular and extracellular signals are critical to the suppression of neoplastic growth, whereas disruption of cell-cell and cell-matrix interactions is implicated, if not required, for neoplastic progression. Indeed, reversion of the malignant phenotype by modulating extracellular signals suggests that cancer cells are susceptible to signals from the microenvironment (2).

We have postulated that ionizing radiation (IR) alters cell phenotypes, which in turn contribute, directly or indirectly, to carcinogenesis (3). IR activates multiple signaling pathways depending on the cell type, radiation dose, and cell status (reviewed in ref. 4). IR also affects the activity or abundance of proteases, growth factors, cytokines, and adhesion proteins that are involved in tissue remodeling (reviewed in ref. 5). We have shown that transforming growth factor β1 (TGFβ) is activated following IR, and that it, in turn, mediates cellular and tissue radiation responses (6, 7). Although TGFβ is considered to be a potent tumor suppressor during the initial stage of tumorigenesis, principally through its ability to cause growth arrest and apoptosis, numerous reports show that TGFβ can switch to tumor promoter during neoplastic progression (reviewed in ref. 8). We have shown that the progeny of irradiated nonmalignant human mammary epithelial cells (HMEC) cultured with TGFβ exhibit compromised morphogenesis, polarity, and growth control when cultured in reconstituted basement membrane (9).

Some epithelial tumors, particularly those that overexpress TGFβ (10), exhibit mesenchymal characteristics and are more aggressive. Several lines of evidence have led researchers to link this morphologic shift during carcinogenesis to the physiologic process of epithelial to mesenchymal transition (EMT). EMT is characterized by loss of epithelial cell polarity, loss of cell-cell contacts, and acquisition of mesenchymal markers and phenotypic traits that include increased cell motility (reviewed in ref. 11). Although the clinical relevance of EMT in late-stage tumor progression is controversial, it is generally agreed that EMT can occur during cancer progression (11). Approximately 18% of breast cancers exhibit evidence of EMT (12). EMT has recently been reported during tumor recurrence after therapy (13). TGFβ has been particularly targeted as a mediator of EMT during neoplastic progression, but analysis of normal epithelial and cancer cell lines showed that TGFβ alone rarely induces EMT (14).

In the present studies, we found that a single exposure to IR sensitizes HMEC to undergo TGFβ-mediated EMT and exhibit all the classic hallmarks of EMT. Neither irradiation nor TGFβ alone was sufficient to elicit EMT, although TGFβ activity was elevated in irradiated cells. Gene expression profiling revealed a specific signature in double-treated cells associated with mitogen-activated protein kinase (MAPK) signaling. As previously reported, IR induces transient activation of the MAPK pathway, but exposure to low concentrations of TGFβ maintains this pathway activity, which is required for maintenance of EMT. We postulate that the pathway signaling interactions between radiation-induced MAPK and TGFβ elicit the heritable EMT phenotype.

Cell culture. HMEC were cultured in serum-free medium as previously described for HMT-3522 S1 (S1; passages 55–60; ref. 15), MCF10A (American Type Culture Collection), 184 HMEC (184v; passage 7–10; ref. 16) and HMT-3522-S2 (S2; ref. 17). About 0.4 ng/mL of recombinant human TGFβ (R&D Systems) was added at the time of plating or of irradiation. HMEC were irradiated 4 to 5 h (S1, MCF10A and S2) or 10 to 14 h (184v HMEC) post-plating using either 160 kV X-ray or ⁶⁰Co γ-radiation with a total dose of 2 Gy. Control plates were sham irradiated. In some experiments, the medium was supplemented with 10 μg/mL TGFβ pan-specific neutralizing antibody (R&D Systems) or with an equivalent concentration of nonspecific mouse immunoglobulin G (IgG). When noted, some cultures were treated with 10 μm U0126 (Cell Signaling) or DMSO.

Reagents. Antibodies to phosphorylated Erk1/2 (Thr²⁰²/Tyr²⁰⁴), phosphorylated MAP/Erk kinase (MEK; Ser^217/221), Erk1/2 and MEK1/2 were from Cell Signaling. Phalloidin was from Molecular Probes. E-cadherin, β-catenin, and N-cadherin antibodies were purchased from BD Biosciences, and vimentin-clone VIM 13.2, β-actin, and fibronectin antibodies were from Sigma. ZO-1 antibody was obtained from Zymed. SMAD2/3 antibody was purchased from Santa Cruz Biotechnology.

Immunofluorescence. Cells were grown on LabTek eight-well chamber slides, fixed with 80% methanol for 10 min (N-cadherin, ZO-1) or methanol/acetone (vimentin, β-catenin) or 4% paraformaldehyde (fibronectin, SMAD3, phalloidin), followed by treatment with 0.1% Triton (N-cadherin, SMAD3) for 10 min. For E-cadherin staining, cells were either fixed with 80% methanol or extracted with CSK buffer (15), followed by 4% paraformaldehyde fixation. Nuclei were counterstained with 0.5 ng/mL 4′,6-diamidino-2-phenylindole (DAPI).

TGFβ bioassay. Mink lung cells transfected with the plasminogen activator inhibitor-1 promoter/luciferase (PAI-L) reporter were used to analyze TGFβ in the conditioned media as previously reported (18).

Protein analysis. Cells were lysed as previously described (15) or in buffer containing 50 mmol/L Tris-HCl (pH, 7.5), 150 mmol/L NaCl, 0.5 mmol/L MgCl₂, 0.2 mmol/L EGTA, 1% Triton X-100, protease inhibitor mixture (20 μL/mL; Sigma), 1 mmol/L Pefabloc, and 50 mmol/L glycerol-2-phosphate (Sigma) or phosphatase inhibitor cocktail set II (Calbiochem). Extraction and immunoprecipitation of soluble and insoluble pools of E-cadherin were done as previously described (19). Proteins were separated on 4% to 15% SDS-PAGE gels and transferred to Immobilon-P (Millipore) or nitrocellulose (Amersham Life Science) and probed with indicated primary antibodies. In some instances, detection was accomplished using chemiluminescence of secondary antibodies labeled with horseradish peroxidase followed by densitometry analysis of films. Alternatively, detection was done using the Odyssey system (LI-COR) as previously described (20). Target proteins were normalized to β-actin for loading.

Motility and invasion assays. Confluent HMT3522 S1 or MCF10A cultures were grown in medium containing epidermal growth factor (EGF). The cultures were scratched with a pipette tip and washed to remove detached cells. The invasion assay was done as previously described (21). The number of invading cells was the average of three wells from two duplicate experiments.

RNA purification. Cells were washed with PBS, denatured in TRIzol, scraped, and subjected to chloroform extraction. After centrifugation, the upper phase was precipitated with an equal volume of isopropanol. RNA precipitates were resuspended in RNase-free water and further purified on RNeasy columns (Qiagen). RNA quality was assessed on an Agilent Bio-Analyzer. The data set analyzed by microarray included biological duplicates for each treatment in two independent experiments and three sham-treated samples.

Microarray processing and analysis. Microarray data were generated at the Lawrence Berkeley National Laboratory Molecular Profiling Laboratory¹

using a high-throughput, automated GeneChip system (Affymetrix). Briefly, target preparation, HT_HG-U133A array plate hybridization setup, washing and staining were done on an Affymetrix robotic system (GCAS) using version 2.1 protocols. Scanning (protocol version 2.2.09) was done using a CCD-based high-throughput scanner (Affymetrix). Samples were analyzed and clustered with the (UNO) One Color Genetraffic software version 3.2-12 (Iobion Informatics LLC, Stratagene). Genes whose expression was specifically altered by treatment were defined as those in which the dye ratio was more than 1.75-fold (|mean log₂ ratio| > 0.8) from baseline in at least three out of the four treated samples compared with the three sham samples. Significance analysis tests (P < 0.05) were done using Excel between sham samples and either IR, TGFβ, or TGFβ + IR samples. Microarray data can be accessed as series GSE8240 in the National Center for Biotechnology Information/GEO database.²

Real-time PCR. Total RNAs were pretreated with amplification grade DNase I (Invitrogen) and then primed with random hexamers to generate cDNAs using a Superscript III first-strand synthesis kit according to the manufacturer's instructions (Invitrogen). About 1 μL of nondiluted cDNAs was then amplified with Lightcycler FastStart DNA master SYBR Green I (Roche Applied Science) using a Light Cycler (Roche Applied Science). TCF8, fibroblast growth factor 2 (FGF2) and CDH1 specific primers optimized for SYBR Green real-time PCR were purchased from SuperArray Bioscience Corporation.

Image acquisition, processing, and analysis. Imaging was done as previously described (20). Colony segmentation from phase images was done using live wire tool from Medical Image Processing, Analysis, and Visualization (MIPAV; ref. 22). Area and shape factors were calculated using Matlab (MathWorks Inc.) and DIPimage (image processing toolbox for Matlab, Delft University of Technology).

The progeny of irradiated HMEC undergoes EMT in response to TGFβ. We previously reported that E-cadherin immunofluorescence significantly decreased in double-treated (IR + TGFβ) HMT3522 S1 cultured in Matrigel compared with control or single-treated cells and failed to establish acinar morphology typical of HMEC (9). EMT is one means of disrupting morphogenesis, e.g., during wound healing. To examine the possibility that EMT underlies the response to radiation and TGFβ, we compared the epithelial characteristics of HMEC in traditional monolayer culture. HMEC were irradiated with 2 Gy shortly after attachment or treated with low concentrations of TGFβ (400 pg/mL), or double treated, and cultured to ∼80% confluence. As observed previously, E-cadherin and β-catenin immunolocalization in monolayer HMEC was significantly reduced in double-treated cells (Fig. 1A). The function of an epithelium as a barrier requires the establishment of tight junctions, evident by lateral, apical localization of ZO-1. ZO-1 immunofluorescence showed distinct punctate localization at cell borders of control and irradiated cells. TGFβ treatment somewhat altered ZO-1 staining, such that it appeared as a coarser, punctate-like pattern at the cell boundaries (Fig. 1A). However, ZO-1 staining completely disappeared in double-treated cells.

Notably, double treatment resulted in spindle-shaped, elongated cells, a phenotype which was examined by phalloidin staining of actin. Actin localized mainly at cell-cell borders of control and irradiated HMEC with very few stress fibers. TGFβ alone elicited a mild change in actin organization. Double-treated cells displayed dramatic actin reorganization consisting of longitudinal stress fibers, characteristic of fibroblasts (Fig. 1A).

In light of the loss of epithelial markers in irradiated HMEC cultured with TGFβ, we investigated whether there was a concomitant gain of mesenchymal markers, fibronectin, vimentin, and N-cadherin (Fig. 1B). Immunofluorescence analysis of the double-treated group revealed increased fibronectin deposition compared with sham and single-treated cells. Accumulated fibronectin was arrayed perpendicular to the cell edge and in randomly oriented networks. The intermediate filament protein vimentin dramatically increased following double treatment. Consistent with the shift to an elongated morphology, vimentin filaments were organized in a longitudinal meshwork. N-cadherin is normally present mainly in neuronal and muscle tissues, but is aberrantly expressed in epithelial tumors where it has been proposed to promote migration and invasiveness of carcinoma cells (23). N-cadherin immunofluorescence was dramatically increased following double treatment in a pattern reciprocal to E-cadherin loss.

Loss of epithelial markers and gain of mesenchymal markers in double-treated cells were accompanied by epithelial to mesenchymal cell shape change. Phase microscopy of 184v HMEC (Fig. 1C), HMT3522 S1, and MCF10A (data not shown) showed that untreated HMEC and irradiated cells displayed typical cuboidal appearance of epithelial cells. TGFβ elicited a modest shape change. In contrast, the morphology of all three HMEC following double treatment shifted to spindle shaped. The morphologic response was observed even when TGFβ was added 48 h post-IR and was not reversible upon removal of TGFβ from the culture media for 48 h (data not shown). Taken together, the concomitant increased expression of mesenchymal markers, loss of epithelial markers, and morphologic response of the progeny of irradiated HMEC to TGFβ suggest that cells have undergone EMT.

TGFβ alters the cytoskeletal associated E-cadherin/β-catenin complexes in irradiated HMEC. Although E-cadherin immunofluorescence was significantly reduced only in double-treated cells (Fig. 1A), Western blot analysis of total E-cadherin protein expression levels were comparable in TGFβ-treated and double-treated groups (Fig. 2A). β-Catenin protein levels followed the same pattern. This observation suggested that immunofluorescence was a function of protein localization rather than abundance. In epithelial cells, E-cadherin and β-catenin are associated with the cytoskeleton at intercellular junctions that are resistant to detergent extraction (24). We tested whether the difference between immunofluorescence and immunoblotting was due to solubility by using differential detergent extraction followed by E-cadherin immunoprecipitation. Although TGFβ reduced the soluble pool of E-cadherin regardless of irradiation, the insoluble, cytoskeletal associated E-cadherin was significantly decreased only in double-treated cells (Fig. 2B). The distribution of β-catenin was similarly affected. Thus, TGFβ alters the cytoskeletal associated E-cadherin/β-catenin complexes only in irradiated HMEC.

Functional consequence of the response of irradiated HMEC to TGFβ. During physiologic processes and in cancer, EMT is often accompanied by the acquisition of cell motility and invasiveness (11). We analyzed the migratory properties of double-treated HMEC using a monolayer scratch assay and time-lapse videomicroscopy (Fig. 3A). Control or irradiated cells moved as a sheet as evidenced by a smooth edge and failed to fill the gap during the 40-h observation period. Consistent with their intermediate cytoskeletal and cell-cell adhesion phenotype, TGFβ-treated HMEC began to close the wound after 20 h, but the cells still maintained sheet formation and, by 40 h, had not closed the gap. In contrast, irradiated HMEC cultured with TGFβ moved as single, spindle-shaped cells and closed the gap by 40 h by forming multicellular bridges of spindle-shaped cells between the two areas. Thus, the tissue-specific requirement for epithelia to maintain cell interactions as a sheet of cohesive cells is disrupted when irradiated HMEC are exposed to TGFβ, which leads to the independent single cell migration characteristic of mesenchymal cells.

The above observations led us to further investigate the invasive properties of HMEC by invasion assay through a membrane coated with Matrigel, a basement membrane-type matrix. HMT 3522 S1 did not invade in this assay under any condition. Therefore, we used HMT 3522 S2, which are derived from further propagation of HMT 3522 S1 in EGF-free media (21). Cells were cultured as described above and then trypsinized and plated into the Boyden chamber assay in standard media. Few control HMT 3522 S2 cells were able to invade. The progeny of irradiated cells showed greater invasion than treatment with TGFβ alone when compared with the control. In contrast, irradiation of S2 cells and subsequent culture in the presence of TGFβ increased the number of invading cells to more than 100-fold compared with sham (Fig. 3B).

There is ongoing discussion in the literature as to what criteria define true EMT in vitro and how to distinguish the EMT phenotype from a scattering phenotype (11). Both phenotypes exhibit disruption of cell junctions, fibroblast-type morphology, and enhanced motility. However, it has been suggested that they can be distinguished by reversibility because the EMT phenotype, but not the scattering phenotype, persists after withdrawal of the stimulus (11). As mentioned above, TGFβ withdrawal did not reverse the morphologic shift, consistent with EMT. Because the progeny of irradiated HMEC cultured with TGFβ exhibits disrupted morphogenesis in Matrigel (9), we used morphogenesis as a stringent test of persistence. Single- and double-treated HMEC cells were trypsinized and replated in Matrigel for analysis of acinar morphogenesis without further TGFβ stimulation. Replated double-treated cells formed larger colonies than control or single-treated cells, which was measured by increased size and irregularity of colony shape (Fig. 3C). We concluded that the phenotype elicited by IR and TGFβ resulting in dysplastic morphogenesis (i.e., three-dimensional growth deregulation and disorganization) and the persistence of these morphologic alterations in the absence of additional TGFβ is further evidence of EMT.

Irradiated HMEC produce active TGFβ. Latent TGFβ is activated in the irradiated mouse mammary gland in vivo (6), which could increase the susceptibility of irradiated cells to undergo TGFβ-mediated EMT by decreasing a threshold response. Nuclear translocation of receptor-phosphorylated SMAD2 or SMAD3 (R-SMAD) results from liganded TGFβ receptor (8). Immunofluorescence analysis showed that R-SMAD nuclear localization increased by 30% in irradiated HMEC 6 days after irradiation compared with control cells (Fig. 4A). These data suggest that IR elicits persistent TGFβ signaling. To determine whether IR increased TGFβ production or activation, we analyzed the conditioned medium (CM) from irradiated cells using a TGFβ bioassay (18). Bioassay of irradiated and control HMEC CM indicated similar levels of latent TGFβ. Active TGFβ, however, was undetectable in either CM (data not shown). Because the sensitivity of detection in CM for the bioassay is ∼0.2 ng/mL, we reasoned that low levels of TGFβ could be functional, although undetectable by bioassay. If so, then the CM should give rise to a similar phenotype in unirradiated HMEC seen in response to the addition of exogenous TGFβ. To investigate this hypothesis, we fed nonirradiated HMEC with 50% CM from either irradiated or nonirradiated HMEC cultures. CM from irradiated cells caused a significant decrease in E-cadherin protein levels when compared with cells treated with CM from nonirradiated cells. This response was reversed by adding TGFβ neutralizing antibody to the cells grown in the presence of IR-treated CM (Fig. 4B). To confirm that this biological activity produced by irradiated cells was indeed TGFβ, irradiated cells were grown in the absence or presence of TGFβ neutralizing antibody. When irradiated HMEC were grown in the presence of TGFβ neutralizing antibody, there was a significant reversal of the decrease in E-cadherin protein abundance compared with cells grown with control IgG control antibody (Fig. 4C).

Based on the above data, we estimated that irradiated cultures produce ∼100 pg/mL of active TGFβ. We then tested whether EMT was a function of the additive effect of exogenous (400 pg/mL) and endogenous TGFβ (∼100 pg/mL) by treating HMEC with graded concentrations of TGFβ. However, MCF10A treated with up to 1,200 pg/mL TGFβ did not undergo EMT (data not shown). Although IR induced TGFβ activity in cultured HMEC as it did in the mouse mammary gland (6), it was insufficient to disrupt HMEC acinar morphogenesis as shown in our previous study (9). Moreover, the progeny of irradiated HMT 3522 S1, MCF10A, and 184v HMEC exhibited a modest response in terms of epithelial morphology, E-cadherin expression, or invasion. Thus, chronic TGFβ activation by irradiated HMEC is insufficient to drive the EMT phenotype.

Genetic programs underlying EMT in double-treated HMEC. To comprehensively describe the altered epithelial cell phenotype following double treatment, we compared transcript expression levels in sham, IR-, TGFβ-, and double-treated HMEC. The expression profile of the progeny of irradiated HMEC was similar to control HMEC. A total of 43 genes were differentially expressed by TGFβ-treated HMEC that included 28 up-regulated genes and 15 down-regulated genes, with a >1.75-fold change in expression compared with sham samples (Supplementary Table). We identified 10 genes that constituted the double treatment signature. Expression was significantly increased for five genes and decreased for the five other genes after double treatment compared with TGFβ alone.

The five significantly up-regulated genes in double-treated samples were the growth factor FGF2/basic fibroblast growth factor, the serine/cysteine proteinase inhibitor, SERPINA1, the interleukin 1 receptor-like protein, IL1RL1, the transcription factor TCF8/ZEB-1, and the zinc transporter SLC39A8 (Fig. 5A and B). The five down-regulated genes in double-treated HMEC were the Wnt pathway transcription repressor secreted frizzle-related protein 1 (SFRP1), the cystein-rich protein, CRIP2, the aldo-keto reductase, AKR1C2, the cadherin EGF LAG seven-pass G-type receptor 2(CELSR2), and the aminopeptidase O, C9orf3. Although transcript levels of 5 of these 10 genes (SERPINA1, SLC39A8, CRIP2, AKR1C2, and CELSR2) were higher in TGFβ-treated samples than in sham samples, they were below the 1.75-fold cutoff used to define TGFβ-responsive genes. Changes in gene expression in double-treated HMEC were validated for TCF8 and FGF2 by real-time reverse transcription-PCR (Fig. 5B). For comparison, real-time RT-PCR of E-cadherin (CDH1) was decreased in both TGFβ and double-treated cells, consistent with the protein levels shown in Fig. 2A.

The EMT signature genes have roles in cell-matrix interactions, cell motility/invasion, inflammation, and disruption of adherens junctions. The products of TCF8/ZEB1 and FGF2 have been described to repress E-cadherin transcription and to promote EMT (25, 26). Some of the TGFβ-responsive and EMT-associated signature genes identified here have been previously described (27, 28). An extensive bibliography search showed that 5 of the 10 EMT signature genes are directly or indirectly associated with the Erk/MAPK pathway, suggesting a specific functional role of Erk/MAPK activation in inducing EMT upon double treatment. Several reports suggest that an independent stimulus must synergistically proceed or at least accompany the induction of EMT. The multiple downstream effectors of Ras, such as MAPK, may be a requirement to induce a premalignant state and endow epithelial cells with an increased rate of proliferation (29).

Persistent activation of Erk/MAPK in double-treated HMEC. Consequently, we examined the possible involvement of the Erk/MAPK signaling cascade in the mediation of TGFβ-induced EMT in irradiated HMEC by determining the activation of Erk1/2 and upstream activators, MEK1/2. The total amounts of Erk or MEK were similar following all treatments. Phosphorylated-Erk1/2 was modestly induced by TGFβ, but was significantly increased in irradiated HMEC treated with TGFβ (Fig. 6A , top). Consistent with this, MEK1/2 activation, as indicated by phosphorylation, was also increased upon double treatment.

There is a significant body of literature that suggests that Erk activation is required to initiate the EMT program (30, 31). IR can induce a rapid, ligand-independent activation of Erk/MAPK in cancer cells (32). To investigate whether activation of Erk1/2 by IR or TGFβ versus double treatment was indeed instrumental in driving EMT, we monitored MEK/Erk activation at early time points (4–48 h post-IR). At 8 h, irradiated cells exhibited greater Erk1/2 activation when compared with TGFβ or double treatment. From 12 to 24 h, MAPK pathway activation increased following IR or TGFβ or double treatment compared with controls. However, at 48 h post-IR, double-treated cells exhibited greater Erk activation compared with control and to single-treated HMEC (Fig. 6A , bottom).

To determine if persistent Erk activation was required for the EMT phenotype in double-treated HMEC, we used U0126, a specific MEK small-molecule inhibitor. Treatment with U0126 dramatically decreased Erk activation in HMEC (Fig. 6B). Treatment with U0126 also restored E-cadherin and cytoskeletal rearrangement (Fig. 6C) and reduced cell migration in the wound closure assay (Fig. 6D). These data suggest that IR and TGFβ collaborate to increase Erk activation, which is necessary for establishment and maintenance of EMT (7 days). We propose that transient IR-induced Erk activation is sustained in the presence of additional TGFβ, and that this event predisposes nonmalignant HMEC to undergo TGFβ-mediated EMT.

Here, we report that the progeny of irradiated HMEC are dramatically sensitized to undergo TGFβ-induced EMT. IR and TGFβ cooperated to induce a phenotypic transition that occurred in the progeny of cells irradiated once and persisted even in the absence of TGFβ. This resulted in increased motility, enhanced invasion, and disrupted epithelial morphogenesis and was accompanied by a distinct pattern of gene expression.

TGFβ has long been considered as both a positive and a negative effector of mammary tumorigenesis, acting early as a tumor suppressor but later as a stimulator of tumor invasion (8). Overexpression of constitutively active TGFβ can induce EMT during tumor progression in vivo (33), and the overexpression of TGFβ has been associated with poor prognosis of many human cancers (8). The phenotypes of breast cancer micrometastases in lymph nodes and the bone marrow have been interpreted as evidence that EMT occurs in primary tumors (34). However, in vivo verification of EMT has been controversial, perhaps due to the transient and reversible nature of the process and to the lack of analytic tools that distinguish carcinoma cells undergoing EMT from neighboring stromal fibroblasts.

In vitro studies of EMT were originally described in cells of murine origin, frequently containing ras mutations. It has been reported in a very limited number of human cell lines and very rarely in nonmalignant cell lines (35, 36). The in vitro studies in which TGFβ alone is capable of inducing EMT have either been done in serum-containing medium, which provides various growth factors commonly associated with wounding, including TGFβ, FGF, and EGF, and/or using high TGFβ concentrations (2.5–10 ng/mL; refs. 36, 37). Expression profiling of genetic program underlying EMT of HaCat keratinocytes stimulated with TGFβ to undergo EMT identified 80 EMT-related targets (27). Our expression profiling study distinguishes between genes regulated by TGFβ without concomitant EMT in nonmalignant HMEC and genes that are differentially expressed under conditions resulting in EMT.

We identified 10 genes specifically associated with EMT, five of which were not induced at all by TGFβ alone. Transcriptional repression of E-cadherin could be mediated by the transcription factor TCF8/ZEB1 (25). FGF2, whose transcript expression is significantly increased upon double treatment in this EMT model, has also been linked to E-cadherin repression and EMT (25, 26). Low E-cadherin immunoreactivity in breast cancer is associated with poor prognosis (38), whereas restoration of E-cadherin reverts the invasive phenotype of cancer cells (39). Irradiated HMEC redistribute E-cadherin from an insoluble to a soluble pool when cultured with TGFβ, which is accompanied by a significant increase in N-cadherin. The cadherin switch from E- to N-cadherin frequently accompanies pronounced tissue reorganization in normal and pathologic conditions (40). Studies on cancer cell lines indicate that N-cadherin is linked to a more malignant and invasive behavior (23). Consistent with this, double-treated HMEC were significantly more motile. Although double-treated HMT-3522 S1 cells were not invasive, double treatment induced invasive behavior in HMT-3522 S2 cells, which are an EGF-independent strain of the HMT-3522 progression series.

Neither chronic TGFβ signaling induced by irradiation nor supplementation with low TGFβ concentrations was sufficient to induce the EMT phenotype in any of the three HMEC that were examined in our studies. EMT occurred only in irradiated HMEC in the presence of TGFβ, even when TGFβ was added 2 days post-IR. This suggested that a IR-induced event was a prerequisite for TGFβ-mediated EMT that was sustained for 48 h. IR initially induces a transient activation of Erk/MAPK via a ligand-independent mechanism (32). Although the involvement of Erk signaling in TGFβ-induced EMT is controversial (41), many studies have shown a requirement for overexpression/mutational activation of elements of the Ras/Raf/Erk pathway for TGFβ-mediated EMT (41–43). Furthermore, the importance of enhanced Erk activation in the induction of EMT is also supported by studies in which Erk activity induced by Ras/TGFβ (41), EGF/TGFβ (31), HGF/ErB2 (44), and Akt (45) was found to be critically involved in EMT. Our data suggest that a model in which IR-induced Erk/MAPK signaling is sustained by TGFβ is required for establishing and maintaining EMT and is essential for the functional response, i.e., enhanced migration.

Cancer radiotherapy is primarily limited in many organs by the risk of developing fibrosis (46). Neilson et al. showed EMT as a significant source of fibrosis in a kidney ligation model (47). An interesting implication from our study is that normal epithelia may undergo EMT in response to irradiation and TGFβ, which could contribute to fibrosis following radiotherapy. If so, this would lend further credence to the potential application of TGFβ inhibitors in radiotherapy.

Based on studies in mouse mammary gland, we proposed that the action of radiation as a carcinogen is augmented by its ability to modulate signaling from the microenvironment (48). Tumorigenesis is increased 4-fold when unirradiated preneoplastic mammary epithelial cells are transplanted to the mammary stroma of a host irradiated with 4 Gy (3). IR induces abundant TGFβ activation (6). Our current and earlier studies (9) have shown that irradiated nonmalignant HMEC undergo EMT only if they are exposed to additional TGFβ, as might be derived from the stroma in intact tissues. If moderate radiation doses can prime preneoplastic cells to undergo EMT, it could accelerate cancer progression. Interestingly, a recent study shows that conventional renal cell carcinomas of the Ukrainian patients living in the radiation-contaminated areas exhibited significantly higher levels of TGFβ expression compared with similar tumors in populations in uncontaminated areas (49). These tumors were characterized by decreased or abnormal distribution of fibronectin, laminin, and E-cadherin/β-catenin, suggesting that chronic low-level radiation exposure in humans might indeed shift tumors toward a mesenchymal phenotype. Nonetheless, there is little evidence to support these events occurring in normal epithelia in vivo after moderate doses. Furthermore, additional studies from our laboratory indicate that TGFβ is instrumental in mounting a DNA damage response (7, 20) and in inducing apoptosis of genomically unstable cells.³

The complexity of radiation effects mediated by TGFβ will require further study to determine whether it plays a proximal role in suppressing or promoting radiogenic carcinogenesis.

Grant support: National Aeronautics and Space Administration Specialized Center for Research in Radiation Health Effects; the Low Dose Radiation Program of the U.S. Department of Energy Office of Biological Effects Research; U.S. Department of Defense DAMD17-00-1-0224 (A.C. Erickson); and the Office of Health and Environmental Research, Health Effects Division, U.S. Department of Energy (contract 03-76SF00098).

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 thank Samuel Haile, Megha Gupta, James Chen, Haleh Sakkaki, and Howard Park for their experimental assistance. We thank also Jeremy Semeiks, Heidi Feiler, and Lakshmi Jakkula for processing the microarray samples as part of the HTA core facility (Lawrence Berkeley National Laboratory).