Metastasis is the main cause of death in cancer patients and is widely claimed to be driven by epithelial-mesenchymal transition (EMT). While the role of EMT in cancer progression has been challenged several years ago,1 there have nevertheless been numerous studies conducted based on this underlying premise. This article reviews recent developments in the field, as well as presenting the argument that for most cancers, mesenchymal differentiation is not essential for invasion and metastasis.
Epithelial-mesenchymal transition (EMT) has been implicated as the critical event initiating cancer invasion and metastasis. After disseminating through the circulation, the malignant cells have been proposed to undergo subsequent mesenchymal-epithelial transition (MET) to form secondary tumors. However, strong evidence from human tumor specimens for this paradigm is lacking. In carcinomas, cancers derived from epithelial tissues, epithelial morphology and gene expression are always retained to some degree. While mesenchymal transdifferentiation may be involved in the pathogenesis of carcinosarcomas, even in these neoplasms, as well as in germ cell tumors capable of multilineage differentiation, the mesenchymal phenotype does not facilitate metastatic progression. Indeed, most cancers invade and travel through lymphatic and blood vessels via cohesive epithelial migration, rather than going through the EMT-MET sequence. EMT gene expression is also consistently associated with high histologic grade and while the transcription factors, Snail, Slug and Twist have traditionally been thought of as inducers of EMT, under certain conditions, they also mediate dedifferentiation and maintenance of the stem cell state. In various malignancies, including basal-like breast cancer and colorectal cancer, the genetically unstable, undifferentiated phenotype predicts early metastatic spread and poor prognosis. This article discusses some of the controversies surrounding differentiation and metastasis from a clinicopathologic perspective and presents evidence that the epithelial phenotype is maintained throughout the process of cancer metastasis.
EMT and Metastasis
Definitions of EMT are loose. It is generally agreed that during EMT, epithelial cells show loss of cell-cell adhesion by losing E-cadherin expression, reorganization of the cytoskeleton via switching from keratin to vimentin intermediate filaments, loss of apical-basal polarity, acquisition of a fibroblast-like (i.e., spindle) cell shape and increased motility.2, 3 According to this paradigm, EMT typically occurs at the tumor-stroma interface, induced by paracrine signals, particularly TGF-β, from stromal cells.2, 3 Several key pathways are activated within tumor cells including: receptor tyrosine kinases,4 TGF-β/SMAD,5 Wnt/β-catenin6 and NF-kB.7 Importantly, these pathways converge in the activation of specific transcription factors, including Snail, Slug and Twist, which cause transcriptional repression of E-cadherin, a defining step in EMT.2, 3, 8 The acquisition of a mesenchymal phenotype is stated to enable cancer cells to detach from the primary tumor, invade through the stroma and into blood vessels and lymphatics. Furthermore, EMT has been proposed to generate cancer stem cells.9 The mesenchymal-like cancer cells eventually extravasate out of capillaries or lymphatic vessels at a distant site, and go through a mesenchymal-epithelial transition (MET), to enable establishment of a secondary tumor.2, 3
Controversial Issues Surrounding the EMT Concept
In this paradigm, it is unclear whether the switch from the epithelial to the mesenchymal phenotype occurs directly or preceded by dedifferentiation—the reversion to a more primitive, uncommitted state, reflected by the expression of stem cell markers and loss of features of terminally-differentiated cells (Fig. 1). While dedifferentiation is a frequent finding in cancer cells, it is debatable whether this is a precursor for further mesenchymal differentiation (i.e., EMT). In the vast majority of carcinomas, metastatic cells may show aberrant expression of a few mesenchymal markers,10 but otherwise retain epithelial characteristics (e.g., columnar or cuboidal cell shape, apical-basal polarity, cohesive migration and keratin expression).11
The lack of rigorous criteria to define EMT may contribute to its ubiquitous usage in the literature, as observed changes in one or two genes are often labeled by some as “incomplete/partial EMT,”3 even though such a description may not be biologically accurate. The changes may be better explained as the cumulative effects of the intrinsic plasticity of epithelial cells, dedifferentiation and genetic instability.
While in the literal sense, EMT implies transdifferentiation—the complete conversion of a terminally-differentiated cell type to another terminally-differentiated cell type1 (Fig. 1), many researchers use the term loosely to be synonymous with cellular plasticity to describe reversible changes in gene expression.2, 3 This vague interpretation does not specify which genes or the extent of changes necessary to classify a process as EMT. It is well known that artificial cell culture conditions cause remodeling of the cytoskeleton to facilitate cell-matrix adhesion,1, 12 and so changes in cytoskeletal proteins, such as vimentin, N-cadherin, smooth muscle actin and fibroblast-specific protein 1 (FSP1/S100A4), like cell shape, are unreliable in vitro for assigning cell identity. Importantly, even in vivo, these genes may be expressed by certain epithelial, neural or inflammatory cells. Moreover, normal somatic stem/progenitor cells from various organ systems, (e.g., liver,13 kidney14 and breast9, 15) co-express epithelial and mesenchymal markers, reflecting their uncommitted state.
In malignancy, EMT gene expression is consistently associated with high histologic grade,16, 17 which begs the question whether these genes are a surrogate marker for the poorly-differentiated state rather than reflective of mesenchymal transdifferentiation per se. Histologic grading of most tumors incorporates degree of differentiation, reflected by cell morphology and tissue architecture, in addition to nuclear atypia and mitotic rate11 (Fig. 2). Nuclear atypia is due to aneuploidy combined with mutations in genes affecting nuclear structure (i.e., nuclear membrane constituents, scaffolding proteins, etc.),18 whereas frequent mitoses requires mutations in pathways regulating cell division and escape from cellular senescence.19, 20 These features are thus reflective of the degree of genetic instability, which results from dysfunctional DNA damage repair mechanisms (e.g., p53, BRCA1, BRCA2 and ATM).21–23 Due to genetic instability, many of the changes in cancer reflect the propagation of “passenger mutations,” of minimal or no functional significance for cancer cell growth or survival, rather than “driver mutations,” which directly contribute to malignant progression.24, 25 Such “passenger mutations” may result in the inappropriate expression of genes associated with particular cell lineages.10 Breast cancer has also been shown to express neuronal and melanocytic markers, in addition to mesenchymal markers.26, 27 These observations should not be over-interpreted to represent transdifferentiation as they may not necessarily confer any survival benefit to the tumor, but rather reflect a general state of chaotic gene expression.10, 26 Therefore, as opposed to being a directed program to initiate the metastatic cascade, the expression of mesenchymal markers in cancer may simply represent the reversion to a more primitive state in the context of a globally disordered genetic background.
Morphologic and Molecular Features of Migratory Cancer Cells
The EMT model has formed the basis of numerous studies in which spindle cell shape and mesenchymal marker expression has been presumed to confer metastatic potential. However, histomorphologic and gene expression analyses of tumor specimens from cancer patients question the validity of this supposition.
Most carcinomas invade and metastasize by cohesive (i.e., cohort or collective) migration, maintaining epithelioid cell shape and preserved cell-cell junctions. In surgical resection specimens, it is much more common to find multicellular tumor fragments, rather than single dissociated cells, within lymphatic spaces and blood vessels.11, 28 These tumor microemboli show clear evidence of epithelial differentiation, with formation of glandular architecture and mucin in the case of adenocarcinomas or spinous differentiation and keratinization in squamous cell carcinomas.11, 28 In some instances, nests of cancer cells invading through and destroying walls of vessels could be observed, consistent with cohesive migration, rather than EMT, during the intravasation process (Figs. 3b and 4b).
Metastasis is a highly inefficient process, with the majority of circulating tumor cells dying during the process.29 Cohesive migration may provide a survival advantage by protection of cells deep within the cluster from immune attack or shear forces during transit through the circulation.30 Indeed, animal studies from almost forty years ago have shown that mammary tumor cells injected intravenously as aggregates were more efficient in forming lung metastases than when a comparable number of dissociated cells were injected.31 Interestingly, similar findings were observed for fibrosarcoma cells, which form cell aggregates during hematogenous dissemination despite their mesenchymal derivation.32
In a recent clinicopathologic study of 4,000 thyroid cancers by Mete and Asa, both dissociated and cohesive sheets of cells were observed invading through vessel walls, and the presence of intravascular tumor microemboli closely correlated with distant metastasis.33 Interestingly, angioinvasive tumors with areas of dedifferentiation had a significantly higher risk of distant metastasis compared to purely well-differentiated tumors with vascular invasion.33
While the invasion of cancer cells into vascular lumina is well recognized, in some neoplasms, the situation may be reversed, that is, the “invasion” of blood vessels within the primary tumor, rather than the tumor cells themselves, may be responsible for the dissemination of malignancy. Nests of cancer cells become surrounded by proliferative sinusoidal blood vessels and passively enter the collecting drainage vein as endothelium-coated tumor emboli.34–36 This has been observed experimentally in mouse models, in which angiogenesis enhances spontaneous metastasis,34, 36 and clinically, in surgical resections of highly vascularized tumors, including most follicular thyroid carcinomas, as well as renal cell and hepatocellular carcinomas.35
In epithelial cancers, stromal invasion via single cell migration is seen more commonly in poorly-differentiated tumors, as in diffuse-type gastric adenocarcinoma, for example.11 There are exceptions, notably lobular carcinoma of the breast, a well-differentiated (low grade) malignancy, characterized by loss of E-cadherin early in its pathogenesis (Fig. 2c). However, single cell migration does not necessarily imply EMT. In lobular carcinoma, the malignant cells show no evidence of mesenchymal differentiation, maintaining distinctive epithelial morphology and lacking an EMT gene expression signature.37 Interestingly, studies have found that the lobular subtype portends a similar and perhaps even better prognosis compared to stage-matched ductal carcinomas.38, 39 Thus, single cell migration per se in a low grade malignancy does not seem to affect clinical outcome.
From Hematogenous Dissemination to Colonization
Measurement of circulating tumor cells from blood has identified cell aggregates in addition to dissociated cells.40–43 Current methods however may underestimate the amount of multicellular tumor fragments released into the circulation, as most would likely be arrested or dissociated upon encountering the fine capillary beds of the liver and lungs.32, 36, 44, 45 While dissociated human melanoma cells colonized the brain when injected directly into the internal carotid arteries of nude mice, when implanted subcutaneously, multicellular tumor emboli became entrapped within the pulmonary vasculature and only lung metastases developed.36 Earlier studies have demonstrated that colonization of distant organs often results from the intravascular growth of tumor cells attached to the endothelium, which eventually destroys the vessel wall, as opposed to extravasation involving MET46, 47 (Fig. 3c).
However, simply arriving at the destination is not enough; suitability of the microenvironment is an important factor in determining whether colonization occurs. This was elegantly demonstrated in a report by Tarin et al. describing cancer patients treated for malignant ascites using peritoneovenous shunts.47 While the shunts carried up to a billion viable cancer cells per week directly into the jugular vein, at autopsy, the lungs were free of metastasis in most patients despite being the organ with the first microcapillary bed encountered. Interestingly, some of these patients instead developed additional metastatic lesions in organs already colonized by tumor prior to the insertion of the shunt.47 These results demonstrate for the first time in humans that cancer cells have intrinsic preferences as to which organs to form metastases.
Clinicopathologic Studies and the EMT Concept
Numerous clinicopathologic studies have attempted to translate observations from cell culture experiments to human cancers by investigating the expression of EMT-related genes in primary patient tumors. However, these studies do not generally differentiate between cells deep within the tumor and those at the invasive edge, as immunohistochemical staining is usually performed on hypercellular areas within the tumor. Aberrant protein expression therefore reflects intrinsic characteristics of the cancer subtype (e.g., permanent loss of E-cadherin in lobular breast carcinoma), probably due to genetic mutations or stable epigenetic modifications and not a transient process as suggested by cell culture and animal experiments.
Similarly, with regard to cancer cells undergoing MET when they reach a distant organ, one would expect to find the rare hybrid lesion with some trace of mesenchymal differentiation. To provide strong evidence in clinical tumor tissue of metastasis occurring via sequential EMT-MET, it would be necessary to show the differential expression of mesenchymal markers in a subpopulation of cells with concordant histomorphologic features in both primary and secondary lesions. These cells should be located near vessels and the invasive front and show transition to epithelial areas, with some cells harbouring an intermediate phenotype. To date, such findings have not been reported.
The Immortalized Epithelial Progenitor Phenotype
Clonal evolution and cellular hierarchy
The malignant cells within an epithelial tumor exhibit considerable diversity with respect to morphology, gene expression, and importantly, tumorigenic/metastatic potential.48, 49 This phenomenon has been attributed to the accumulation of random genetic mutations and chromosomal abnormalities during tumor progression, leading to the emergence of heterogeneous subclones.50 The mutational profiles of cancer cells microdissected from different regions within a tumor and of parallel comparisons between primary and metastatic lesions are consistent with clonal evolution.51–56 Genetic instability generates cells with the capability for indefinite proliferation and able to survive through the process of stromal invasion, hematogenous dissemination and metastatic colonization.
The alternate explanation for intratumoral heterogeneity is that the cancer cells are organized in a cellular hierarchy, similar to normal tissue development. In normal tissue, rare, immortalized adult/somatic stem cells give rise to lineage-restricted progenitor cells. After several rounds of amplification and differentiation, these eventually become mature terminally-differentiated cells, which are non-proliferative. Importantly, the stem cell undergoes asymmetric division—one daughter cell is an identical copy of itself, while the other is a more differentiated progenitor cell with limited proliferation ability. Similarly, in certain malignancies, particularly leukemias and germ cell tumors, a minor subpopulation of immortalized “cancer stem cells,” capable asymmetric division, generates the entire community of phenotypically diverse malignant cells.57–59 Cellular hierarchy and clonal evolution are not mutually exclusive phenomena; a tumor may consist of multiple genetically distinct subclones varying in aggressiveness, each of which consist of cancer stem cells that could undergo epigenetic modifications to differentiate into phenotypically mature non-tumorigenic cancer cells (refer to Shackleton et al.49 for a more comprehensive discussion). Interestingly, in some instances, the non-tumorigenic population can regain tumorigenicity in response to signals from the supporting stroma.9, 60 Thus, the notion of malignant progression being driven by a fixed subpopulation of cancer stem cells is perhaps an over-simplification for most solid tumors.
Reactivation of stem cell self-renewal programs is a prerequisite for metastasis
Self-renewal and differentiation are regulated by common pathways (e.g., Myc, Notch and Wnt) and there is considerable evidence that the capacity for indefinite proliferation (immortality) and impaired differentiation are inextricably coupled.61–67 Poorly-differentiated, but not well-differentiated, malignancies of breast, bladder and brain have been shown to have gene expression signatures similar to human embryonic stem cells.68 However, while metastatic carcinoma cells are typically characterized by an immortalized epithelial progenitor phenotype, they do not acquire all features of normal stem cells. Clearly, the malignant undifferentiated state does not confer pluripotency, as multilineage differentiation is present only in certain rare tumor types, (discussed under Metastatic Behavior of Tumors with Multilineage Differentiation). Furthermore, not all tumors are hierarchically organized and as such, the immature immortalized carcinoma cell should not be confused with the concept of the “cancer stem cell.” It is likely that in high-grade malignancies, genetic instability facilitates the emergence and natural selection of hyperproliferative cancer cells that are not restrained by asymmetric division and normal maturation programs. The result is a neoplasm in which most, if not all, of the malignant cells are arrested in an undifferentiated state and potentially capable of forming tumor when relocated to another site.49, 69
Several studies using primary human tumor material support the association between impaired differentiation and tumorigenic/metastatic ability. In primary spheroidal cultures of colon carcinomas, cells characterized by high levels of Wnt activation, expression of stem cell markers, LGR5 and ASCL2, and lacking intestinal differentiation markers, MUC2 and CK20, were capable of self-renewal, as confirmed by serial transplantations into nude mice. In contrast, the more phenotypically mature cells were non-clonogenic.60 In human breast cancers, Pece et al. demonstrated that Grade 3 (poorly-differentiated) tumors consist of a four-fold increase in tumor-initiating cells relative to Grade 1 (well-differentiated) tumors.70 These findings are in accordance with the activation of stem cell self-renewal pathways in cancer cells with a poorly-differentiated phenotype, and partially explains the association of high tumor grade with poor prognosis, cancer recurrence and metastasis.11 Of course, as previously discussed, high tumor grade also reflects genetic instability, which generates resilient, rapidly dividing cancer cells with heightened survival and invasive capabilities.
Dedifferentiation and the tumor microenvironment
R. Weinberg's group showed that TGF-β signaling and activation of Twist or Snail in cultured normal and tumorigenic mammary epithelial cells induces “mesenchymal” marker expression along with mammary stem cell characteristics, namely the CD44high/CD24low expression pattern and increased mammosphere-forming and tumor-initiating abilities.9 These findings have led to the interpretation that EMT is a mechanism for generating cancer stem cells. However, it is possible that TGF-β, Twist or Snail may be inducing or maintaining a dedifferentiated progenitor-like state, independent of EMT. Accordingly, the study also showed that CD44high/CD24low cells isolated from normal breast tissues were positive for Snail and Twist. In normal breast, bipotential epithelial stem cells, which give rise to both luminal and myoepithelial cells, reside in the basal layer (i.e., basal cells) and share a similar gene expression profile with myoepithelial cells, namely, the expression of basal keratins, p63, vimentin and smooth-muscle markers, yet also express markers of luminal epithelial cells.15, 70, 71
That Twist, Snail and Slug are more likely responsible for maintaining the stem cell state than EMT in certain conditions is supported by the expression of these factors in normal stem cells of epithelial tissues, including lung,72 intestine,73 breast9 and endocrine pancreas.74 Ectopic expression of Snail and Slug causes activation of TGF-β and Wnt signaling,75–77 pathways implicated in stem cell maintenance and renewal. Indeed, TGF-β, at low concentrations, has been suggested to be involved in maintaining the primitive undifferentiated state in embryonic78 and hematopoietic stem cells.79 Furthermore, TGF-β signaling cross-talks with the Wnt pathway,80 which has been implicated in the regulation of mammary stem/progenitor cells.81 Therefore, instead of EMT being a novel mechanism to generate cancer stem cells, perhaps a more fitting and parsimonious conclusion may be that Snail, Twist or TGF-β signaling induces dedifferentiation or a luminal-to-basal transition in breast epithelial cells. This interpretation is supported by recent evidence that basal-like breast cancer originates from the luminal progenitor cell.82 As would be expected from its poorly-differentiated histology, basal-like breast cancer is clinically aggressive and defined by a gene expression profile similar to myoepithelial/basal stem cells83—often incorrectly interpreted as evidence of EMT. Epithelial morphology and marker expression is retained as cancer cells migrate collectively through the stroma and within the vasculature.84
While TGF-β has been presented thus far in the context of its effects on cancer cells, its major role is the induction of fibroblast proliferation and activation, manifested in the desmoplastic stromal reaction often observed within tumors.85 This reactive stroma could then interact with cancer cells to promote invasiveness, cell survival, proliferation and metastasis.60, 86–88
In colorectal cancer, single cells at the invasive front of the tumor, also known as “tumor buds,” have been shown to express markers of intestinal stem cells, including EpCAM and ABCG5, as well as nuclear localization of β-catenin.89 The latter indicates activation of Wnt signaling, which has been implicated in the self-renewal and migration of intestinal stem cells.77 There is experimental evidence that myofibroblasts at the tumor-stroma interface secrete factors that induce the dedifferentiation of mature colorectal cancer cells via activation of the Wnt pathway, resulting in enhanced tumorigenic potential.60 Tumor buds have not been shown to express any of the classic mesenchymal markers. Indeed, vimentin is often silenced by promoter methylation in colorectal cancer.90 These findings are consistent with budding tumor cells having an immortalized epithelial progenitor phenotype induced by stromal signals. It is worth noting that despite the presence of tumor buds, cohesive migration is often the predominant pattern observed in colorectal cancer, with supporting stromal cells not uncommonly identified within intravascular tumor microemboli (Fig. 3).
Metastatic Behavior of Tumors with Multilineage Differentiation
Carcinosarcomas are rare aggressive neoplasms found in various organ systems, including breast, lung, genitourinary tract and most commonly, the female genital tract. In the ovary or uterus, where they are otherwise known as Malignant Mixed Mullerian tumors, carcinosarcomas consist of high-grade epithelial and high-grade mesenchymal malignant cells (Fig. 4). Their pathogenesis may involve EMT, though this remains a controversial issue, as early transformation of a bipotential stem cell, with divergent differentiation along epithelial and mesenchymal lines remains a viable explanation. While some have been shown to result from the “collision” of two primary tumors,91, 92 in the vast majority of cases, there is evidence of a clonal relationship between the epithelial and mesenchymal components, with common genetic mutations and loss of heterozygosity of identical alleles.91–93
Irrespective of histogenesis, it is well established that Malignant Mixed Mullerian tumors behave as carcinomas, since metastatic lesions consist only of epithelial histology in the vast majority of cases.92, 94 Furthermore, only epithelial tumor cells are detected within lymphovascular spaces, thus diminishing the possibility that the mesenchymal component disseminates through the circulation to subsequently undergo MET at a distant site.92, 94
This contrasts with the corresponding phenomenon of sarcomatoid transformation in renal cell carcinoma. Similarly, genetic analyses suggest that the epithelial and mesenchymal populations are monoclonal, implying either EMT or divergence from a malignant pluripotential stem cell.95 In one study, over 95% of metastatic lesions consisted of either the epithelial or mesenchymal component, but not both.96 Strikingly, the likelihood of sarcomatoid metastasis was correlated with the proportion of the primary tumor consisting of sarcomatoid features. Though preliminary, these findings suggest that in this tumor type, metastatic potential does not depend on either the epithelial or mesenchymal phenotype. Rather, a pro-survival/pro-metastatic mechanism may be selected for in either epithelial or mesenchymal subclones by random mutational events, leading to clonal expansion of the affected component.
Germ cell tumors
Germ cell tumors of the ovaries and testes illustrate the importance of the stem/progenitor phenotype for metastasis. Embryonal carcinoma, the most undifferentiated germ cell tumor, is essentially made up of malignant embryonic stem cells. The presence of various tissue types in this tumor is evidence of pluripotentiality, an uncommon phenomenon in malignancy. Metastasis occurs early in the course of the disease, through lymphatic and hematogenous routes. In localized testicular tumors showing mixed histology, the proportion of tumor consisting of embryonal carcinoma is directly correlated with risk of metastasis and poor prognosis.97 Interestingly, embryonal carcinoma shows diffuse immunopositivity for keratins and may attempt to form glandular structures, suggestive of a predilection for epithelial differentiation,98 although it is unclear whether this has any pathobiologic significance.
The benign teratoma consists of mature differentiated tissue derived from all three germ layers. In contrast, the malignant immature teratoma (distinct from embryonal carcinoma) is defined by the presence of immature embryonal structures, often in the form of primitive neuroepithelial tissue. In ovary, metastatic behavior and prognosis is correlated with the extent of the immature component within the tumor, despite the presence of areas showing mesenchymal differentiation.99 The cure rate for germ cell tumors is high, exceeding 90% overall, with a 5-year survival of 50% in the worst prognosis cases (e.g., embryonal carcinoma with distant metastases), due to the excellent response of these tumors to modern chemotherapy regimens.100, 101 Given the poor prognosis and metastatic behavior of other poorly-differentiated malignancies, studying the unique chemosensitivity of germ cell tumors may yield insights into the development of new therapeutic strategies.
Metastases arise from highly resilient cancer cells that are able to proliferate indefinitely and survive through the harsh journey of vascular invasion, hematogenous dissemination and colonization of a foreign microenvironment. The accumulation of genetic/epigenetic changes and microenvironmental factors culminate into poorly-differentiated clones with aggressive behavior, yet retaining epithelial lineage restriction. Cohesive epithelial migration is the predominant mode of cancer cell transport and EMT is unlikely to contribute significantly. In accordance, in highly malignant neoplasms comprised of both epithelial and mesenchymal cancer cells, mesenchymal differentiation does not confer increased metastatic potential. Rather than overemphasis on mesenchymal markers, future studies should direct focus on the genes and pathways underlying genetic instability, dedifferentiation and immortalization. Importantly, histopathologic studies on surgically resected tumor specimens are essential for placing experimental molecular data into clinicopathological context and for interpreting the significance of the findings.