Epithelial—mesenchymal and mesenchymal—epithelial transitions in carcinoma progression (2024)

EMT in General/Development

Recognition of epithelial to mesenchymal transition (EMT) is relatively new in oncology. Researchers observing the morphology of various tumors noted epithelial and mesenchymal components of tumors and called these regions of metaplasia (Kahn et al., 1978; Ishikawa et al., 1979). It was not until 1987 (Krug et al., 1987) that the phrase “epithelial to mesenchymal transition” was used (in reference to a cellular change elicited by extracellular matrix (ECM)) and not until 1995 (Hay, 1995) that this phenomenon was officially characterized. The number of publications in which this phrase has been used has risen 15-fold since 2001–91 in 2006 alone, making EMT currently one of the hottest medical science topics.

So what is EMT? An EMT is a culmination of protein modification and transcriptional events in response to a defined set of extracellular stimuli leading to a long term, albeit sometimes reversible, cellular change. Core elements of EMT include reduction of cell–cell adherence via the transcriptional repression and delocalization of cadherins (adherens junctions), occludin and claudin (tight junctions), and desmoplakin (desmosomes). The cadherin supporting molecule β-catenin is often lost from the cell membrane and translocates to the nucleus to participate in EMT signaling events (Klymkowsky, 2005). Circumferential F-actin fibres of the cytoskeleton are replaced by a network of stress fibers, at the tips of which ECM adhesion molecules (including integrins, paxillin, focal adhesion kinase) localize. These changes potentially allow cells to separate, lose the apico-basal polarity typical of epithelial cells, and gain a more variable cell shape and changeable cell adhesions, all of which facilitate cell movement. Expression of epithelial intermediate filaments, containing cytokeratins, is typically reduced and the equivalent mesenchymal filament protein vimentin increased. Matrix metalloproteases (MMPs) such as MMP-1, -2, -3, -7, and -14 are frequently upregulated, potentially enabling cells to detach from each other (via E-cadherin cleavage) and to penetrate the basem*nt membrane. ECM synthesis changes from basal lamina to interstitial forms. These genetic alterations, along with changes in cellular shape to a more elongated, fibroblast-appearance with front-back polarity signify EMT (Hay, 1995). Crucially, cells following EMT show increased motile potential.

These changes are in some cases reversible, which enables cells in the developing embryo to respond to changing environmental cues. The following early developmental EMT and mesenchymal to epithelial transition (MET) events (Fig. 1) are summarized from various excellent reviews (Newgreen and Erickson, 1986; Duband et al., 1995; Hay, 1995; Summerbell and Rigby, 2000; Chaffer et al., 2007).

In the initial EMT, the primary mesenchyme is formed from the upper epiblast epithelium resulting in a three-layered blastocyst consisting of the ectoderm, mesoderm (primary mesenchyme), and the endoderm (Fig. 1, steps 2 and 3). Following on from these changes is the formation of the first somite (somitogenesis) via MET of the medial primary mesenchyme. More lateral mesenchyme also forms loose mesodermal epithelia of the body wall (Fig. 1, step 4). A second wave of EMT then occurs from part of the ectoderm, forming neural crest cells which migrate to form the peripheral nervous system (Fig. 1, step 5). This is coordinated with EMT from the ventral somite to form the sclerotome mesenchyme, or the secondary mesenchyme. Further EMT occurs progressively from the more dorsal part of the somite to give rise to muscle, and connective tissue of dermal layers of the skin. This is also coordinated with EMT of the loose intermediate or nephric mesoderm which later undergoes MET to form kidney epithelia (Fig. 1, step 6).

However, EMT is not restricted to early developmental events. EMT-like changes are thought to be a feature of palatal formation, which is the coming together of ectodermal layers and fusion of underlying branchial arch mesenchymal tissues, resulting in a separation of the nasal and oral cavities. Whether this fusion is a result of EMT is speculative, as some have argued that apoptosis or epithelial cell migration occurs, or that all of these processes occur simultaneously (Shuler, 1995; Dudas et al., 2007; Nawshad et al., 2007). Other examples of regulated EMT in the adult is placental formation (Vicovac and Aplin, 1996), and the production of fibroblasts during inflammation (Iwano et al., 2002) and wound healing (Desmouliere, 1995).

Many of these processes are under the strict control of growth factors and downstream transcription factors. For example, the Wnt family of growth factors in neural crest EMT trigger the canonical Wnt signaling pathway involving LEF and β-catenin, resulting in Snail2 signaling (Vallin et al., 2001). Fibroblast growth factor (FGF)-2 in Xenopus (Monsoro-Burq et al., 2003) and transforming growth factor (TGF)-β family members bone morphogenic protein (BMP) -4 and -7 (Liem et al., 1995) also induce neural crest cell EMT. Growth factors epidermal growth factor (EGF), hepatocyte growth factor (HGF), insulin-like growth factors (IGFs), FGF, and tumor necrosis factor (TNF)-α upregulate Snail1 and/or Snail2 in EMT via signaling through their corresponding receptor tyrosine kinases, which in turn can activate phosphoinositide-3-kinase (PI3K) and Ras, which then activate mitogen-activated protein kinase (MAPK) leading directly to Snail1/2 upregulation (Christiansen and Rajasekaran, 2006). Snail2, and other members of the Snail family such as Snail1, induce EMT by repressing the transcription of the key adherens junction protein E-cadherin (Batlle et al., 2000; Bolos et al., 2003). This is often a preliminary step triggering EMT as this releases β-catenin to participate in Wnt pathway signaling (Nelson and Nusse, 2004). Snail family proteins repress E-cadherin transcription by binding the E-box E-pal element in the E-cadherin promoter, as is the mechanism of action of other E-cadherin repressors involved in developmental EMT such as Twist (Kang and Massague, 2004; Yang et al., 2004).

The Snail1 and Snail2 genes are highly hom*ologous, and in certain circ*mstances can replace each other functionally: for example, the consequences of Snail2 knockdown in avian embryonic neural crest can be avoided by transfection of Snail1 (Sefton et al., 1998). However, physiologically, they have somewhat distinct roles. Snail1 is essential for mouse gastrulation (Ip and Gridley, 2002), whereas Snail2 knockout mice are viable and fertile (Jiang et al., 1998). Snail2 appears to be essential for normal breast tubule network development (Come et al., 2004). Other transcription factors which play a role in development include the Forkhead family of genes (FoxD3) and the SoxE gene family, the latter which are important in not only neural crest EMT but also EMT leading to the formation of astrocytes and chondrocytes (reviewed in Kalluri and Neilson, 2003).

As eloquently stated by Lappin and colleagues (Lappin et al., 2002) “The poetic rhythm of the developmental programme does not always appear to resonate with the same harmony in the adult form” and EMT in disease is no exception. Similar genetic and molecular players are involved in pathological EMT as in development but appear to lack the intricate coordination. Dysregulated responses such as diabetic nephropathy (Kalluri and Neilson, 2003) and cataracts (de Iongh et al., 2005) involve the over production of fibroblasts which overtake functioning tissue—these bear the signature of EMT. Many epithelial tumors hijack EMT to become metastatic (Thiery and Sleeman, 2006), and examples of this in various carcinoma contexts will be discussed in the following sections.

Evidence of EMT in Breast Cancer

Early studies showed that breast cancer cell lines with increased invasiveness in vitro, and metastatic potential in vivo, exhibited expression of the mesenchymal intermediate filament protein vimentin (Thompson et al., 1992). Further analysis also showed reduced cytokeratin levels, and reduced or absent components of the various cell:cell adhesion complexes such as E-cadherin, desmoplakin, and ZO-1 (Sommers et al., 1992; Thompson et al., 1992; Sommers et al., 1994; Gilles et al., 1999). Reduction or absence of E-cadherin expression is often accompanied by reciprocally increased expression of N-cadherin (Nieman et al., 1999), and indeed N-cadherin has been shown to promote breast cancer cell invasiveness in various studies (Nieman et al., 1999; Hazan et al., 2000), and N-cadherin status predicted invasiveness better than lack of E-cadherin (Nieman et al., 1999). Since N-cadherin and vimentin are both well-regarded markers of EMT, the conclusion that EMT facilitates invasiveness in breast cancer cell remains valid, and the importance of examining a panel of markers is emphasized. The relevance of the mesenchymal state, and the utility of multiple markers, was reinforced by the study of Zajchowski et al., who found that vimentin, as well as other mesenchymal gene products such as integrin α3, TIMPs-2 and -3, MT1–MMP, PAI-1, Osteonectin/SPARC, thrombospondin-1, collagen (VI)α1, and collagen (I)α2, were part of a 24 gene signature predicting invasiveness derived from a Clontec Atlas 588 element gene array analysis of non-invasive and invasive human breast cancer cell lines (Zajchowski et al., 2001). Epithelial gene products cytokeratins 18 and 19 and plakoglobin were also included in the signature, and increased levels of these predicted non-invasiveness. The invasive signature correlated strongly with colony morphology in the Matrigel outgrowth assay, consistent with a mesenchymal phenotype.

Such studies have recently been expanded upon, with good concordance and increased robustness. Charafe-Jauffret et al. (2006) interrogated 31 breast cancer cell lines with Affymetrix U133 Plus 2.0 arrays. Neve et al. (2006) studied 51 human cell lines with Affymetrix HG-U133A chips, and also integrated CGH and proteome analysis. Each of these provides an exceptional database of gene expression amongst cell lines, and each identifies a subset of mesenchymal-like cell lines. There is reasonable concordance between these two recent studies and earlier studies, with some exceptions. Several cell lines which have long been known to be invasive/mesenchymal are confirmed as BasalB/mesenchymal (e.g., BT-549, Hs578T, MDA-MB-157, MDA-MB-231, MDA-MB-435 (note caveat on melanoma nature Rae et al., 2004)), MDA-MB-436 and SUM159PT. Several N-cadherin-negative lines have clustered as BasalB, most notably MDA-MB-231 as mentioned above, but also SUM149 and SUM1315. These cell lines appear poorly invasive in early studies (Nieman et al., 1999), but more recently SUM149 have been reported to be highly invasive (Zajchowski et al., 2001; Neve et al., 2006). In the main, cell lines clustered to BasalB (distinct from BasalA or luminal) in the Neve study correspond to the group designated by Charafe-Jaufrett et al. as mesenchymal (compared to luminal or basal), the exceptions being HCC38, MCF10A, and SUM149PT which were classified as basal, and HCC1500 which was classified as luminal. BasalB/mesenchymal cells were shown in both studies to resemble “basal” cells in the expression of mesenchymal gene products, but lacked a number of cytokeratins seen in basal cells. In general, good concordance can be seen between the data obtained by different groups in relation to these cell lines, and there is clearly a robustness to the new gene expression profiles.

Examination of classical EMT markers in the Neve dataset, shown in Figure 2, reveals reasonable concordance of mesenchymal gene products and EMT drivers in the mesenchymal cell lines. Comparison of various EMT-regulating factors (E-cadherin repressors, see section 1) suggests roles for Snail2 and Twist, but not Snail1, in the mesenchymal phenotype. Importantly, there are exceptions in these expression patterns of differing degrees, with the tightest associations seen with Snail2. Although a specific set of gene products was found in these studies to distinguish mesenchymal breast cancer lines from basal lines (supplementary Table 4 in Charafe-Jauffret et al., 2006), they lack the commonly used indices of EMT because these are expressed by both cell types (basal/basalA and mesenchymal/BasalB). Further analysis of these distinguishing features may be quite revealing, but will require sophisticated approaches to discriminate tumor cells which have undergone EMT from host mesenchyme/stroma when analyzing whole tumors. A comprehensive survey by Lacroix and Leclercq (2004) is a valuable resource for information on the origins and behaviors of these cell lines. Although clinical breast cancer specimens can be classified into luminal and Basal subtypes, the substratification of the Basal subtype is not seen (Neve et al., 2006). It is possible that BasalB/mesenchymal cell lines represent a minority of cells in a given breast cancer (perhaps likely to be more common in basal cancers) undergoing further dedifferentiative/transdifferentiative change, through EMT. These may be limited in number, and also may migrate away from the primary mass. The molecular signatures associated with BasalB/mesenchymal lines overlaps strongly with that of BasalA/basal, and is distinguishable more by a lack of certain basal cytokeratins, etc. This will make it very difficult to distinguish a subset of BasalB-like cells in a predominately BasalA tumor.

It is interesting to note that several “normal” human mammary cell lines have clustered with BasalB in the Neve et al. (2006) dataset, such as HBL100 (note—some questions have arisen on the provenance of HBL100 due to the presence of Y chromosome, reviewed in (Thompson et al., 2004), MCF10A and MCF12A. Whether this reflects a plasticity of such normal cells to respond to cell culture conditions with a mesenchymal shift, or whether this phenotype exists in vivo in the normal epithelium and adapts best to cell culture, remains to be seen. What is clear, however, is that the predictions based on gene expression are corroborated by biological activity. MCF10A cells have been shown to have a high constitutive invasiveness (Zajchowski et al., 2001). Although somewhat mesenchymal, they do not exhibit a fully mesenchymal phenotype in vitro (Nagaraja et al., 2006). Importantly, Gilles et al. (1999) showed that vimentin expression is induced during migration of MCF10A cells in a monolayer wound assay, and that the abrogation of vimentin expression slowed the rate of migration. As with PMC42 cells, the MCF10A cell EMT is stimulated by EGF (Gilles et al., 1999). Cell lines derived from normal human mammary tissue such the 184 series (Thompson et al., 1994; Giunciuglio et al., 1995) and oncogene-transformed MCF10A cells also overexpress vimentin and show reduced E-cadherin (Giunciuglio et al., 1995), and are EGF-dependent/responsive.

In addition to the observations in human mammary cell lines, rat and mouse mammary cell lines have been shown to undergo EMT in response to various stimuli, especially TGFβ family members. This has been observed in HC-11, ED-11 cell-derived EpH4 cells after Ras transfection (EpRas cells), NMuMG cells, 4T1 cells, and comma-1D cell-derived ScP-2 cells from the mouse, and in the LA7 rat mammary tumor line (reviewed in Kokinnos et al., 2007). Perhaps the most striking example of the existence and functionality of cancer EMT in a mouse mammary system are the studies of Neilson and co-workers, using the fibroblast specific protein-1 (FSP-1)/S100A4 promoter to mark and manipulate the mesenchymal state in transgenic, erbB-2-induced mouse mammary tumors (Xue et al., 2003). FSP-1/green fluorescent protein expressing mammary cancer cells were shown to exit the primary tumor, and importantly, were shown to lose their mesenchymal nature when forming macrometastases in the lungs. Selective killing of mesenchymal cells through an FSP-1-driven suicide cassette reduced the number of metastases, as did global depletion of FSP-1/S100A4. Further functional evidence of EMT in whole animal systems is needed, and it is these systems in which we should try to better understand and control EMT, working towards therapeutic goals.

EMT in the PMC42 System

PMC42 is a unique human breast carcinoma cell line originally established from a pleural effusion (Whitehead et al., 1983a). Cytological studies showed that they could develop into eight different morphological types including spindle-shaped cells, cells with prominent vesicles or cysts, and syncytium-forming cells. Cells grew in monolayer and also as cords or organoids that resembled ductal cells (Whitehead et al., 1983b; Monaghan et al., 1985). PMC42 cells had characteristics of both secretory and myoepithelial cells, where secretory and lipid granules as well as parallel actin cables were detected in different cellular subtypes. The heterogeneity of PMC42 cells suggested they were either derived from a stem cell compartment in the breast or had acquired stem cell characteristics. Mitogenic responses to estrogen, progesterone, insulin, hydrocortisone, and EGF were also observed (Whitehead et al., 1984). When cloned in agar, PMC42 cells retained the capacity to differentiate into different morphological subtypes and showed the same morphological diversity as the original culture, confirming their stem cell characteristics (Whitehead et al., 1983a).

PMC42-LA is a well-differentiated, stable epithelial variant of the parental PMC42 cell line (Ackland et al., 2003b). It forms hollow organoids and expresses the milk protein β-casein when cultured on Engelbreth-Holm-Swarm-Sarcoma-derived basem*nt membrane matrix (EHS matrix) in the presence of lactogenic hormones (Ackland et al., 2001). Similar to the parental line, PMC42-LA is heterogeneous, consisting of subpopulations expressing both luminal epithelial and myoepithelial characteristics. PMC42-LA cells predominantly express the major luminal epithelial markers cytokeratins 8 and 18 and E-cadherin (Ackland et al., 2001). The presence of cytokeratins 8/18 is characteristic of a well-differentiated breast carcinoma. These cytokeratins are present in the relatively well-differentiated breast cancer lines including MCF-7 line but absent in the mesenchymal MDA-MB-231 line. PMC42-LA is heterogeneous with respect to some markers including vimentin, which suggests some plasticity with respect to phenotype (Ackland et al., 2003a).

A significant feature of PMC42-LA cells is that exposure of cells to EGF induces an EMT (Ackland et al., 2003a), including downregulation of E-cadherin and increased motility. This is typical of other systems in which EMT has been induced, such as the rat bladder cancer cell line NBT-II and the mouse mammary ScP-2 cell lines, and is also consistent with the lack of E-cadherin seen in a range of human breast carcinoma cell lines that showed in vitro invasiveness (Sommers et al., 1994; Nieman et al., 1999), as discussed above. In PMC42-LA cells, N-cadherin was expressed in untreated cells but was detected in increased amounts after treatment with EGF. Co-expression of E- and N-cadherin in untreated PMC42-LA cells is different from previous studies of other breast cancer cell lines, in which N-cadherin was selectively expressed by E-cadherin–negative invasive human breast carcinoma cell lines and predicted invasiveness (Nieman et al., 1999; Hazan et al., 2000). Co-expression of E- and N-cadherin by PMC42-LA cells under untreated conditions may restrain these cells from their EMT-like responses associated with N-cadherin, and the relative levels of expression of E-cadherin and N-cadherin may be more significant than the absolute levels of expression of these cell-adhesion molecules. The rapid loss/relocalization of E-cadherin following EGF treatment in these cells may allow N-cadherin to promote motility.

The stroma of the mammary gland plays an important role in directing differentiation of the two main cell lineages of the mammary gland: luminal and epithelial. Stromal fibroblasts produce ECM proteins and secrete many growth factors and hormones, including HGF, IGF-I, IGF-II, EGF, TGF-α, TGF-β1, -β2, and -β3, FGF-7, -2, and –10, and interleukin-6 (Bhowmick et al., 2004). In PMC42-LA cells, fibroblasts induced expression of the major myoepithelial markers α-smooth muscle actin and cytokeratin 14 (Lebret et al., 2006). Stromal-epithelial interactions in the mammary gland are important in cancer progression. The reactive stroma of the tumor environment differs from the stroma of the normal mammary gland, showing changes in fibroblast proliferation and remodeling of the ECM (Dvorak, 1986). During cancer progression, changes in the phenotype of fibroblasts occur and normal mammary fibroblasts (NMFs) are replaced by cancer-associated fibroblasts (CAFs) which express α-smooth muscle actin (Lazard et al., 1993) and fibroblast activation protein (Scanlan et al., 1994).

A role for CAFs, but not NMFs, in inducing changes consistent with EMT has been found in PMC42-LA cells where cells were grown in co-culture with NMFs or CAFs or with conditioned medium (Lebret et al., 2006; Lebret et al., 2007). In response to wounding, migration of PMC42-LA cells was increased in CAF conditioned medium, directly implicating a role for CAFs in inducing cell motility. This was associated with a dramatic upregulation of vimentin relative to cells cultured in NMF conditioned medium, and an elongated morphology stretching in the direction of wound closure.

CAFs also induce a number of other changes characteristic of EMT in PMC42-LA organoids. These include a reduction in membranous E-cadherin and altered localization of β-catenin away from the cell junctions to the cytoplasm and nucleus (Lebret et al., 2006). Furthermore, while NMFs induced vimentin-positive cells layered on the outside of organoids, vimentin-positive cells were localized throughout the organoids in cells cultured in CAF conditioned medium (Lebret et al., 2006). This may indicate that fibroblasts have the capacity to alter the organization of epithelial cells within a 3-D structure and future work may provide insights into changes in cellular organization in EMT.

EMT in Prostate Cancer

The involvement of EMT in prostate cancer progression is an enticing proposition given the altered expression of various lineage markers in pathological specimens (Lawrence et al., 2007). Decreased E-cadherin expression is correlated with various indices of prostate cancer progression including grade, local invasiveness, dissemination into the blood, and tumor relapse after radiotherapy (Loric et al., 2001; Mason et al., 2002; Ray et al., 2006). In contrast, markers of a mesenchymal phenotype including N-cadherin, OB (osteoblast)-cadherin, and WAP-type four disulfide core/ps20 proteins (WFDC-1) are all upregulated by tumor cells (Tomita et al., 2000; McAlhany et al., 2004; Jaggi et al., 2006). Increased levels of the extracellular domain of N-cadherin have also been detected in the serum of prostate cancer patients (Derycke et al., 2006). Importantly, the functional importance of decreased E-cadherin levels has also been demonstrated in prostate cancer cells with its inverse correlation with cellular motility and protease expression (Chunthapong et al., 2004). These changes in epithelial and mesenchymal markers and the loss of prostatic glandular architecture are consistent with the general dedifferentiated phenotype of aggressive prostate cancer cells, although decisive evidence for EMT remains elusive.

The proof of principal for EMT in prostate cancer has so far emerged from studies using in vitro and in vivo models of prostate cancer progression. Various factors that are altered in the prostate cancer microenvironment through increased production by tumor cells or the cancer-associated stroma are candidates for eliciting EMT. Many of these molecules have also been shown to cause EMT in other systems. For example, HGF and EGF can induce EMT in DU145 cells (Lu et al., 2003; Wells et al., 2005). The action of EGF is due to calveolae-dependant endocytosis of E-cadherin followed by transcriptional downregulation by Snail. Inhibition of EGF signaling restores E-cadherin levels (Yates et al., 2007). The transcription factor Twist similarly represses E-cadherin expression as well as upregulating N-cadherin levels in prostate cancer cell lines (Kwok et al., 2005; Alexander et al., 2006). In contrast, loss of the epithelium-specific transcription factor prostate-derived ETS factor (PDEF), which is downregulated by TGFβ, induces EMT in PC3 cells (Gu et al., 2007). In addition, over-expression of prostate specific antigen (PSA) and kallikrein-related peptidase 4 (KLK4), both potential activators of pro-EGF and latent TGFβ2, results in EMT in PC3 cells (Veveris-Lowe et al., 2005; Whitbread et al., 2006). While PSA and KLK4 are part of normal prostatic secretions, they leak into the tumor microenvironment due to the disruption of glandular architecture during cancer progression, suggesting a link between tissue architecture and EMT. Also emphasizing the relationship between tumor dedifferentiation and EMT, the hedgehog and BMP-7 developmental signaling pathways are reactivated in aggressive prostate cancer and can induce EMT (Karhadkar et al., 2004; Yang et al., 2005). BMP-7 is also abundant in the bone microenvironment which is of interest given recent findings that ARCaP prostate cancer cells undergo EMT when grown as tumors in mouse bone (Xu et al., 2006).

The majority of EMT studies in prostate cancer have used PC-3 and DU145 human prostate cancer cells. These cell lines already have a partially interconverted phenotype with altered cytokeratin profiles and heterogeneous expression of cell adhesion complexes. The androgen receptor is also absent or lowly expressed in PC-3 and DU145 cells. Intriguingly, the cadherin profile and invasiveness of prostate cancer cells reportedly correlates with androgen-independence (Jennbacken et al., 2006). Negative cross talk between androgen receptor and a number of EMT-related signaling pathways has also been described. Therefore, it is likely that perturbation of the androgen receptor axis has a permissive effect on EMT as aggressive prostate cancer cells exhibit increased plasticity and loose their luminal epithelial phenotype, including androgen receptor expression, during tumor progression. Further analysis of different cell lines and signaling pathways will help foster the development of a coherent model of EMT in prostate cancer.

MET in Cancer Progression and Metastasis

MET has been recognized in a number of mesenchymal tumors. Focal epithelial differentiation to form nests of glandular epithelium is observed in a subset of human synovial sarcomas. In these tumors, the SYT transcriptional coactivator gene is fused with SSX1 and interferes with Snail. This results in derepression of the E-cadherin promoter and subsequent E-cadherin expression and transition to an epithelial phenotype (Saito et al., 2006). In addition E-cadherin protein levels may be downregulated by dysadherin, a cancer-associated cell surface molecule expressed in some synovial sarcomas (Ino et al., 2002; Izumi et al., 2007). Rare epitheloid variants of mesenchymal neoplasms in bone have been described (Deyrup and Montag, 2007), and are typically associated with the upregulation of a number of epithelial markers including E-cadherin.

In epithelially derived tumors, failure of MET is central to the development of the pediatric kidney malignancy Wilms' tumor (Li et al., 2002). During development, epithelial nephrons develop via MET. Key genes, such as transcription factor Pax-2, are expressed during embryological MET but are switched off during terminal differentiation. A number of these genes are re-expressed in renal tumors. Both primary and metastatic ovarian carcinomas express E-cadherin, in contrast to normal ovarian surface epithelium which rarely expresses E-cadherin (Auersperg et al., 1999). Indeed overexpression of E-cadherin in ovarian surface epithelial cells induced a number of epithelial characteristics and markers associated with malignant transformation and tumor progression, indicative of a key role of MET in ovarian carinogenesis.

In colon cancer, constitutive activation of the Wnt signaling pathway is a key contributor to tumorigenesis (reviewed in Fodde and Brabletz, 2007). Accumulation of nuclear β-catenin was observed at the invasive front and in tumor cells migrating into stroma (reviewed in Hlubek et al., 2007), consistent with an EMT. In contrast, in the remainder of the primary tumor, and in metastases, heterogeneous intracellular distribution was detected.

Progression of solid tumors involves spatial and temporal occurrences of EMT, whereby tumor cells acquire a more invasive and metastatic phenotype. Subsequently, the disseminated mesenchymal tumor cells must undergo the reverse transition, MET, at the site of metastases, as metastases recapitulate the pathology of their corresponding primary tumors. Initiation of tumor growth at the secondary site is the rate-limiting step in metastasis. This suggests that cellular plasticity, the ability to undergo EMT and subsequently MET in the appropriate microenvironments, is a key feature of a successful metastatic cell.

The importance of the epithelial phenotype in the formation of secondary tumors has been demonstrated using the human transitional carcinoma of the bladder cell line TSU-Pr1. In vivo selection of these cells from metastastic tumors following seeding into the systemic circulation (Chaffer et al., 2005) showed an association of epithelial state with tumor formation (Chaffer et al., 2006). Consistent with this, the more mesenchymal parental cells displayed a more invasive phenotype in vitro and were detected as micrometastasis in lungs following orthotopic inoculation, but failed to form secondary tumors. Furthermore reversal of the epithelial phenotype, via knockdown of FGF receptor-2, inhibited the ability of the cells to form secondary tumors (Chaffer et al., 2006).

In prostate cancer, coculture of DU145 prostate cancer cells with hepatocytes, modeling liver metastasis, resulted in the re-expression of E-cadherin (Yates et al., 2007). This is consistent with findings in clinical material, in which membranous E-cadherin was detected in hepatic metastasis using immunohistochemistry, and vimentin was absent in the tumor cells. In the Dunning prostate cancer model, mapping of FGF receptor-2 splice variant expression demonstrated the occurrence of METs (indicated by co-expression of E-cadherin and FGF receptor-2(IIIb)) in primary tumors, typically where the tumor cells were in contact with the stroma (Oltean et al., 2006). In lung micrometastases, tumor cells in contact with the lung parenchyma and adjacent to blood vessels also frequently underwent MET.

Thus, malignant progression is based on dynamic processes, which cannot be explained solely by irreversible genetic alterations, but must be additionally regulated by the tumor environment. Indeed tumor expression of the hom*obox gene Cdx2 is modulated by grafting in different tissues and transiently downregulated in invasive cells (Benahmed et al., 2007). As transfected Cdx2 can inhibit β-catenin/TCF4-mediated transcriptional activation of target genes, including p14 (ARF) and cyclin D1 (probably through indirect mechanisms) (Saegusa et al., 2007) this provides a link to the dysregulated Wnt signaling pathway observed in colon cancer. Similarly, frizzled-7 (FZD7) is necessary for MET in the LIM1863-Mph colorectal carcinoma model as the loss of FZD7 results in the persistence of a mesenchymal state (increased SNAI2/decreased E-cadherin) (Vincan et al., 2007). Moreover, FZD7 is also required for migration of the LIM1863-Mph monolayer cells. This suggests that FZD7 orchestrates either migratory or epithelialization events depending on the context.

Thus, targeting MET may provide a novel strategy to inhibit the development of solid tumor metastases by trapping disseminated tumor cells in a state of micrometastasis.

Clinical and Translational Perspectives

Despite the abundance of evidence for EMT in embryonic development, in vitro, and in model systems, solid evidence of EMT in clinical carcinoma has been slow to accumulate. This has led to controversy and healthy debate (Tarin et al., 2005; Thompson et al., 2005), and remains an important issue for the field. A number of factors which may account for this differential have been raised, including (i) possibility of incomplete EMT, resulting in a hybrid state; (ii) the likelihood that EMT in carcinoma would generate small numbers of cells at the tumor edge which would immediately divorce themselves from the collective; (iii) the production of EMT molecules in other scenarios, such as Snail family members in relation to cell survival, vimentin in myoepithelial cells, EMT markers such as vimentin and N-cadherin in the tumor parenchyma of certain breast cancer subsets; and (iv) the apparent requirement for disseminated carcinoma cells to “re-epithelialise” in order to generate a viable metastasis. Thus, the two major repositories available for sampling—primary tumor and metastatic disease—probably should lack evidence of EMT, and EMT may rather be harder to sample as it is likely to be a transient state.

Firmer evidence is nonetheless still needed for EMT in the clinical setting and emerging technologies may help solve this dilemma (Gao et al., 2007). Perhaps the best to date are the immunohistochemical observations in provocative locations. Brabletz et al. (1998) have demonstrated altered distribution of β-catenein to the nucleus in colon carcinoma cells at the invasive front, and the return to a more epithelial pericellular presentation in metastases. Vimentin expression in cervical carcinoma parenchyma is associated with invasive carcinomas and lymph node metastases (Gilles et al., 1996). Recently, careful histochemical analysis linked vimentin expression in breast cancer to dedifferentiation and genomic disarray, rather than myoepithelial nature or EMT (Korsching et al., 2005). The authors suggested that breast cancers expressing vimentin should be considered as a unique subgroup of cancers. Overall, vimentin in breast cancers cannot be solely attributed to EMT or to direct myoepithelial histogenesis. Instead, vimentin positive breast cancers may be derived from breast progenitor cells with a bilinear (glandular, myoepithelial) differentiation potential. The data from Korsching et al. (2005) suggest that vimentin expression in mammary glandular tissues is not an indicator only of EMT, but the data does confirm that vimentin is expressed in invasive breast cancers. The recent generation of antibodies against Snail suitable for immunohistochemistry have allowed the rigorous analysis in breast cancer (Come et al., 2006) and carcinomas of the upper gastrointestinal tract (Rosivatz et al., 2006), where is it much more widely expressed than would be expected for EMT, and clearly plays a role in the epithelial/carcinoma component. In contrast, Snail1 was found to be selectively expressed at the invasive front of colon carcinoma (Franci et al., 2006). Importantly, Snail1 was central to the signature found associated with recurrence in a mouse mammary tumor model, suggesting an important role in the survival of minimal residual disease (Moody et al., 2005). Thus, traditional EMT markers are found in breast cancers in scenarios other than EMT. Their prevalence in “basal” subtype may indicate a propensity for these tumors to undergo EMT, and this would be consistent with the poorer prognosis associated with the subclass (van 't Veer et al., 2002). New markers of EMT, perhaps specific to the carcinoma context, are required to better identify EMT in clinical cases.

Mesenchymal derivatives of carcinoma cells show a number of attributes which would favor metastasis, such as survival after separation from the collective as individual cells, increased migratory and invasive potential, increased survival in suspension and resistance to apoptosis in response to hormone ablation and chemotherapy. These are not only biological observations, since molecular links between EMT and survival have been reported. Sustained expression of mesenchymal traits would assist in extravasation at the bone marrow, and possibly also survival in bone marrow. Indeed, early dissemination may be the most likely place to look for evidence of EMT, and some indications support the possibility that EMT may manifest in circulating tumor cells (CTCs) and micrometastases. For example, abnormal E-cadherin expression in primary prostate cancer is correlated with the presence of CTCs (Loric et al., 2001). It has long been recognized that CTCs show reduced expression of specific cytokeratins (Pantel et al., 1994). More recently, Pantel and co-workers have shown that cell lines derived from breast cancer micrometastases stably express the mesenchymal marker vimentin (Willipinski-Stapelfeldt et al., 2005). The possible roles of EMT in escape and survival in the blood stream as CTCs, and establishment as micrometastases, are summarized in Figure 3. The further re-epithelialisation to form a clinically overt metastasis is well supported by histological and immunohistological observations, but has rarely been modeled. The data obtained in the TSU-Pr1 bladder cancer progression series described above (Chaffer et al., 2006) are consistent with an MET requirement for successful macrometastasis. Confirmation of these transitions in clinical material is urgently needed, since it may offer new opportunity in the targeting latent micrometastases, and in targeting the re-epithelialisation process which appears germane to metastatic relapse.