Guidelines and definitions for research on epithelial–mesenchymal transition

Epithelial–mesenchymal transition (EMT) encompasses dynamic changes in cellular organization from epithelial to mesenchymal phenotypes, which leads to functional changes in cell migration and invasion. EMT occurs in a diverse range of physiological and pathological conditions and is driven by a conserved set of inducing signals, transcriptional regulators and downstream effectors. With over 5,700 publications indexed by Web of Science in 2019 alone, research on EMT is expanding rapidly. This growing interest warrants the need for a consensus among researchers when referring to and undertaking research on EMT. This Consensus Statement, mediated by ‘the EMT International Association’ (TEMTIA), is the outcome of a 2-year-long discussion among EMT researchers and aims to both clarify the nomenclature and provide definitions and guidelines for EMT research in future publications. We trust that these guidelines will help to reduce misunderstanding and misinterpretation of research data generated in various experimental models and to promote cross-disciplinary collaboration to identify and address key open questions in this research field. While recognizing the importance of maintaining diversity in experimental approaches and conceptual frameworks, we emphasize that lasting contributions of EMT research to increasing our understanding of developmental processes and combatting cancer and other diseases depend on the adoption of a unified terminology to describe EMT.

Moreover, EMT is known to be activated during cancer pathogenesis and tissue fibrosis. The reverse process, known as mesenchymal-epithelial transition (MET), also occurs frequently during development. A salient characteristic of EMT occurring in vivo, whether during normal development or in a pathological context, is that the transition from an epithelial to a mesenchymal state is often incomplete, resulting in cells that reside in interme diate states that retain both epithelial and mesenchymal characteristics. Importantly, these intermediate states can be diverse, depending on the biological context 3 .
The EMT research field has grown explosively over the past 20 years. More than half of all articles on EMT have been published in the past 5 years alone (Fig. 2), and half of those have reported on studies of EMT in the context of cancer biology. The growing complexity and diversity of the EMT literature has resulted in vague and often confusing definitions of EMT and associated nomenclature. Cell biologists have traditionally focused on the microscopically visible and profound changes in cell-cell interactions, cell motility, cytoskeletal organi zation, cell proliferation and resistance to various stress ors that occur during EMT. Molecular biologists have focused on changes in the activity of EMTassociated transcription factors (EMTTFs) and in aspects of their regulation, often involving various chromatin  modifications to orchestrate changes in EMTassociated gene expression. Cancer biologists have often empha sized the acquisition of various malignancyassociated cell phenotypes, notably invasiveness, as well as dis semination and different degrees of cell responsive ness to various therapeutic modalities. However, these contextspecific, or even researchcommunityspecific, criteria to define or refer to EMT do not reflect the current experimental shortcomings in the abil ity to identify EMT events. As we learn more about EMTassociated changes, it is becoming apparent that there is a great diversity of EMT phenotypic manifes tations. Thus, narrow definitions have become inap propriate or inaccurate, and a more encompassing definition is required in order to describe this complex cell biological programme.

Purpose of this Consensus Statement
The use of the term EMT in research areas as diverse as developmental biology, cell biology, tissue homeostasis and disease (notably cancer and fibrosis) has created discrepancies in data interpretation and persistent disa greements about whether the process studied is EMT 4-9 , largely because the plasticity and heterogeneity of EMT programmes have been insufficiently considered. EMT was originally described as an important process in embryonic development during which epithelial cells underwent a phenotypic transformation to mesenchy mal cells. These early analyses and subsequent valida tions in cell culture described the phenotypic changes associated with EMT using a limited set of molecular markers. However, the later identification of EMT as a crucial programme in cancer progression 10,11 indicated that EMT involves more than the originally identified developmental EMT programmes and that, clearly, many different variations of the EMT programme exist that cannot be accurately defined by those limited sets of markers.
As the complexity of EMT events and EMT regula tors in both development and cancer becomes increas ingly appreciated, there is a need for the community of researcher experts on EMT to agree on a number of key points. These include: definitions of major EMTrelated terms, a description of EMTassociated phenomena, a description of the diverse versions of EMT that can occur in different contexts, the contextdependent func tion and activity of EMT regulators, and the relation ship between the core and noncore EMT functions of EMTTFs. In the context of cancer, there is also a need to further consider the contributions of genetic alterations, the complex input of changing tumour environments and the EMTlike changes that occur in nonepithelial cancers such as melanoma, sarcoma and leukaemia. Provoked by the passionate town hall discussions that took place during the 2017 and 2019 meetings of the EMT International Association (TEMTIA), TEMTIA proposes the following guidelines to define the EMT programme, its phenotypic plasticity and the resulting multiple intermediate epithelial-mesenchymal states. By building such a consensus on EMTrelated concepts,  we aim to eliminate semantic problems in the EMT debate and facilitate genuine crossdisciplinary discus sion of the roles of EMT in both normal development and pathological conditions.

A brief history of EMT
Modern EMT studies began with research aimed at understanding tissue morphogenesis during develop ment, cell behaviour in culture and carcinoma invasive ness in cancer progression. Elizabeth Hay recognized the importance of EMT in embryogenesis 1 and began to discuss the concept of "epithelial-mesenchymal trans formation" in the late 1970s. EMT was subsequently observed in the context of neural crest formation 12,13 , heart valve formation 14 and Müllerian duct regression 15 , as well as in epithelial tissue explants in vitro 16 . This 'epithelial-mesenchymal transformation' process was alternately referred to as 'epithelial-mesenchymal tran sition' , to distinguish it from the process of neoplastic transformation commonly used by the cancer research community. 'Epithelial-mesenchymal transition' became the term of use after the first TEMTIA meeting, which brought the field together in 2003. Many observations originating more than a quar ter century ago described EMT as being induced by a diverse array of contextual signals. For example, cul tured amnion cells were 'transformed' from an epithe lial phenotype into fibroblastlike cells in response to a leukocyte medium 17 . Endocardial cells underwent EMT in response to signals from adjacent cardiac muscle 18 . Hepatocyte growth factor (HGF) was found to induce transformation of epithelial cells into migratory fibro blasts 19 . Fibroblast growth factor 1 (FGF1) induced an 'epithelial plasticity' response in bladder carcinoma 10 , connecting EMT to cancer. TGFβ, overexpressed in cancers and required for cardiac EMT 20 , was found to be a potent inducer of EMT in cultured cells 11 . These observations provided the first indications that diverse extracellular signals, including soluble factors and components of the extracellular matrix (ECM), could act together to evoke EMT programmes in responding epithelial cells.
These early descriptive studies shed little light on the mechanisms operating within individual cells that enable the induction of EMT. Discoveries in the field of Drosophila melanogaster developmental genetics led to the identification of master regulators of EMT -for example, the transcription factors Snail and Twist, which act pleiotropically to orchestrate mesoderm formation during gastrulation 21 . The identification of related tran scription factors in chordates revealed the high degree of conservation of these factors during metazoan evo lution 22 and thus highlighted the importance and rele vance of studying various developmental animal model systems in order to understand EMT regulation.
Research aimed at identifying molecular regulators of EMT began on a large scale in the 1990s. For example, identification of the Snailrelated transcription factor Slug (also known as Snai2) as an inducer of EMT dur ing chick gastrulation and neural crest cell formation illustrated that specific transcription factors can act as key upstream regulators of EMT 23 . The finding that Slug expression can convert epithelial carcinoma cells into mesenchymal derivatives made a strong case for a connection between embryonic EMT and cancer pro gression 24 . This notion was reinforced by the observation that the Snail family of transcription factors are capable of inducing EMT and invasiveness (the capacity to leave the epithelial tissue and migrate into the underlying tis sue) in normal epithelial cells, in part through transcrip tional repression of the gene encoding Ecadherin [25][26][27][28] . Additional EMTTFs, notably E47, Twist1, Zeb1 and Zeb2, were identified by means of their ability to evoke morphological and molecular changes associated with EMT [29][30][31][32] . It is important to note that these EMTTFs usually cooperate with one another to orchestrate EMT. A large number of studies have also revealed that EMTinducing signals can regulate the expres sion and activity of these EMTTFs, doing so via both transcriptional and posttranscriptional mechanisms.
As our understanding of the inductive signals and transcriptional control of EMT evolved, it became apparent that the activation and execution of EMT does not require changes in DNA sequence and can be revers ible. This made it clear, in turn, that EMT occurs as a result of complex epigenetic regulatory programmes, much like those operating at different stages of devel opment. During development, some cell popu lations may undergo multiple rounds of EMT and MET, indi cating substantial phenotypic plasticity. For example, during renal morphogenesis, the epithelial cells lining renal vesicles are derived from renal mesenchymal cells via MET, while these mesenchymal cells in turn are descendants of epithelial cells in the epiblast via EMT 33 . During somite formation, paraxial mesenchyme cells undergo MET to form epithelial somites, which then undergo EMT to give rise to the sclerotome 34 . Likewise, during the pathogenesis of cancers and fibrosis, EMT is activated to various degrees (from partial to fully) and is often reversible, revealing a plasticity that can yield cells residing in a spectrum of states, between a fully epi thelial phenotype and a fully mesenchymal phenotype 3 (considered as the end points of EMT). Thus, EMT does not result in a single mesenchymal state, but rather in a variety of intermediate states with various degrees of epi thelial and mesenchymal features. This finding presents a major challenge to the EMT research community: how best to capture the diversity and plasticity of the EMT programmes operating in various biological contexts.

EMT in development, cancer and fibrosis
A driving paradigm for the growth of this research field has been that EMT operates in normal tissues during development and wound healing, but is also a driver in the pathogenesis of cancer and fibrosis. The common starting point of diverse EMTs is the downregulation of certain features of the epithelial phenotype. Importantly, however, although it is now recognized that the EMT programmes do not operate as binary switches that shunt cells from fully epithelial to fully mesenchymal extremes, it remains unclear whether discrete phenotypic states are arrayed along the epithelialtomesenchymal (EtoM) phenotypic spectrum or, alternatively, a continuum of such states exist that lack distinct, definable boundaries.

Neural crest
A multipotent cell population formed at the interface between the neuroepithelium and the epidermis. Neural crest cells give rise to multiple cell types, including the neurons, glial cells, pigment cells, fibroblasts, smooth muscle cells, odontoblasts and adipocytes.

Müllerian duct
A coelomic epithelium-derived tubular structure present in early reproductive tract development. Müllerian duct gives rise to the oviduct, uterus, cervix and part of the vagina in the female embryo. it degenerates in the male embryo through the action of anti-Müllerian hormone.

Renal vesicles
Epithelialized vesicles derived from the metanephric mesenchyme during kidney development. Each renal vesicle gives rise to one nephron and, together with vascular-derived glomerulus and ureteric bud-derived collecting duct, forms the basic unit of renal filtration.

Epiblast
A population of pluripotent cells during early mammalian, reptilian and avian development. The epiblast undergoes epithelialization, initiates gastrulation morphogenesis and gives rise to the three principal germ layers (ectoderm, mesoderm and endoderm).

Somite
An epithelial vesicle formed through periodic budding from the mesenchymal pre-somitic mesoderm. The somite is located between the neural tube and the intermediate mesoderm and gives rise to the sclerotome (axial bones) and dermomyotome (dermis and skeletal muscles).

Sclerotome
A part of the epithelial somite that undergoes EMT first and gives rise to the vertebrae and ribs. The extent to which such intermediates represent sta ble states in specific biological contexts is also unclear. A continuum of EMT intermediate states might enable rapid interconversion between cells possessing various combinations of these traits, a process viewed as hav ing high phenotypic plasticity. Moreover, it is possible that the phenotypic states between the fully epithelial and fully mesenchymal end points might not be arrayed along a linear spectrum, and that multiple alternative paths can operate to enable an epithelial cell to advance towards a mesenchymal state. Finally, the cells activating EMT programmes in adult tissues under pathological conditions commonly express combinations of epithe lial and mesenchymal markers and rarely complete the entire EMT programme, suggesting that 'partial EMTs' represent the norm rather than the exception.

Development.
During animal development, cells of epi thelial origin often migrate long distances from their original position to their final destination. The exit and detachment from the epithelial sheet and the following migration to distant locations depend on cells acquiring a mesenchymal state through EMT. Moreover, in many cases, the cells later undergo a permanent or temporary reversal to an epithelial state through MET (for example, endoderm cells) or switch to a different state (for example, neural crest cells). A great degree of morphological vari ability is associated with epithelial cells that participate in developmental EMTs, ranging from cells possessing fully formed epithelial cell-cell junctions (including tight junctions, adherens junctions and/or desmosomes) and an underlying basement membrane (to which they adhere through hemidesmosomes), such as the pluri potent epiblast cells of amniotes 35 , to the primitive epithe lial cells giving rise to mesendoderm in Xenopus laevis and zebrafish, which exhibit only apical-basal polarity and incompletely assembled cell-cell junctions 36 .
In many cases, the quasimesenchymal state is not reached through a complete loss of cell-cell junctions, but instead by changes in the nature and dynamics of junction formation and dissolution, which may explain how cells with mesenchymal characteristics can exhibit collec tive cell migration 37 -that is, the migra tion of cohorts of cells that seem to be held together by various types of cell-cell junctions. Such plastic, quasimesenchymal phenotypes are observed in cells that migrate collectively and are held together partially by cadherinbased cell-cell contacts, in endoderm and meso derm cells of D. melanogaster, zebrafish, X. laevis and mouse [38][39][40][41][42][43] , and in neural crest cells of zebrafish, chick and X. laevis 44,45 . It is important to note that not every migratory process employed by epithelial cells involves EMT, as is the case in chicken epiblast morphogenesis before the formation of the primitive streak 46 .
Cancer. During the multistep progression of carcino mas that are initially benign, epithelial cells acquire a few distinctly mesenchymal traits that confer to them the ability to invade adjacent tissues, locally, and then to disseminate to distant tissues. Much of this phenotypic progression towards increased invasiveness depends on the activation of EMT 3,47-51 . Carcinoma cells might be able to perform collective migration locally without activating EMT, possibly using collective migration mechanisms similar to those used during development. However, it is unclear whether primary carcinoma cells can complete the entire process of metastatic dissemina tion without activating, at least transiently, components of the EMT programme. The behaviour of carcinoma cells that transition to intermediate epithelial/mesenchy mal states (E/M states) (that is, partial EMTs) echoes the behaviour of epithelial cells during normal development. Cancer cells proceed through a gradation of phenotypic states, each associated with combinations of epithelial and mesenchymal markers 3,51,52 .
The activation of alternative EMT programmes and the progression of individual cells to different states along the EtoM spectrum can generate extensive pheno typic heterogeneity within tumours. Supporting this notion, multiple E/M cell subpopulations with dis tinct chromatin landscapes and gene expression signa tures have been reported in skin and mammary primary tumours, and these subpopulations are often spatially localized within specific areas of a tumour 53 . Moreover, hybrid E/M states are enriched in circulating tumour cells (CTCs) that are released by primary breast and lung can cers and their metastases 54,55 , ostensibly reflecting the cel lular heterogeneity seen within the originating primary tumours. Such phenotypic plasticity and heterogeneity may provide cancer cells with increased adaptability and resistance, enabling them to respond to a variety of external cues and physiological stresses 3,49,51,52,56,57 . Thus, because tumour cells encounter diverse microenviron ments as they navigate the multiple steps of the metastatic cascade (Box 1) and migrate through and reach different tissues, various hybrid E/M phenotypes may provide a survival advantage in these distinct environments, such as blood and lymphatic vessels and primary and second ary tumour sites. The tissue of origin of the tumour cell, specific combinations of expressed EMTTFs, and chro matin modifications may also determine the phenotypic heterogeneity of these various hybrid E/M states.
The diversity of EMTassociated cancer cell pheno types is reflected in the discrepancies in experimental and histopathological observations of human tumours that have fueled a longstanding debate regarding the roles of EMT in cancer progression 4,5,58-60 . Such discrep ancies can often be attributed to the use of different EMT markers and the analysis of particular EMTTFs as markers of this programme; in addition, the EMT programmes operating in different tissues might differ from one another. For instance, Snai1 and Twist1 have both been shown to be important for metastasis in the PyMTdriven breast cancer model 61,62 , but dispensable for metastasis in a pancreatic cancer model 7 , which instead depends on Zeb1 (rEF. 63 ). Further confusion has come from the observation that carcinoma cells undergoing a partial EMT can reduce their epithelial phenotype through posttranslational mechanisms 64

Adherens junctions
A type of cell-cell junction in which the plasma membranes of apposing cells adhere to each other through cadherinmediated homophilic interactions. Adherens junctions associate with condensed actin filaments intracellularly.

PyMT
A breast cancer metastasis model in which the polyomavirus middle T-antigen is overexpressed in mammary epithelial cells as a driver of breast tumour development.
seen during the course of embryonic development, cancerassociated EMT is only activated partially and transiently, making endstage markers of a fully mesen chymal state uninformative in cancer studies. Another complicating factor is that EMT programmes have been linked to additional traits that are not associated with canonical EMT regulation, such as stemness, cell sur vival rate, metabolic changes and, in the case of cancer cells, resistance to anticancer therapeutic drugs 65,66 .
Fibrosis. EMT has also been observed to occur and play a role in diverse types of fibrosis (including in the lung, liver and kidney), with EMTTF expression shown to be a prerequisite for fibrosis development in mouse models 3 . As in cancer progression, the role of EMT in organ fibrosis has been the subject of active debate. A central issue in this debate is the origin of the myofibroblasts that accumulate in fibrotic tissues. These cells represent a specialized fibroblast population involved in collagen secretion and thus in the develop ment and progression of interstitial fibrosis, which is the major cause of the disease in different tissues. Early lineagetracing studies supported the hypothesis that myofibroblasts arise from EMTdriven conversion 67 , but subsequent lineagetracing analyses have not provided compelling evidence of epithelial cells as precursors of fibrosisassociated myofibroblasts 68 . More recent studies have shown that renal epithelial cells undergo a partial EMT that is crucial for disease progression, but that they do not directly contribute to the formation of the myofibroblast population 68,69 . Instead, they lose their normal tubular function, and these damaged cells release paracrine signals to the renal interstitium, reshap ing the microenvironment. The release of TGFβ converts existing fibroblasts into myofibroblasts, and the secre tion of additional cytokines and chemokines probably recruits macrophages to the stroma. Hence, damaged renal epithelial cells promote both fibrogenesis and inflammation, which are hallmarks of renal fibrosis 69,70 . While the debate concerning the contribution of EMT to different types of fibrosis continues, the demonstrated requirement of EMTTF expression strongly suggests that the activation of EMT is indeed required for the development of several types of fibrosis.

Definitions of EMT and its associated terms
To facilitate the investigation of multifaceted EMT pro cesses and discussion among diverse groups of research ers studying EMT, we propose the following definitions of EMT and its associated terms to stand as a reference. We encourage researchers to adhere to this recommended nomenclature.

Box 1 | The metastasis cascade
The metastatic process is thought to consist of the following sequential steps (see the figure). Initial escape from the primary site (invasion) requires that the epithelial tumour cells loosen their cell-cell junctions to become motile (step 1) and that they degrade the basement membrane and extracellular matrix (eCm); breakdown of these physical barriers allows cells to migrate and invade into nearby normal tissues (step 2). The next step of metastasis is termed 'intravasation', during which tumour cells invade across the endothelial lamina, penetrate into the vascular or lymphatic vessels and thereby enter the systemic circulation (step 3). only a small number of the released neoplastic cells appear to be capable of surviving attack by the shear forces encountered in the circulation and by anoikis (a form of programmed cell death that occurs when cells detach from the surrounding eCm) provoked by the loss of anchorage to solid eCm. eventually, some of the survivors may arrest in and extravasate through the capillary endothelium of distant organs into the parenchyma of these organs (extravasation) (step 4). In the new stromal environment that they encounter, an even smaller subset succeed in establishing themselves (step 5) and in proliferating from minute growths (micrometastases) into fully malignant secondary tumours that are clinically detectable and eventually life-threatening (secondary growth) (step 6). The activation of epithelial-mesenchymal transition (emT) can provide tumour cells with the ability to migrate, invade, intravasate and extravasate. once they reach distant organs, these mesenchymal cells revert to a more epithelial identity via mesenchymal-epithelial transition (meT) in order to regain proliferating ability, to form secondary growth in distant organs. Figure adapted

Interstitial fibrosis
Excessive deposition of extracellular matrix by fibroblasts in interstitial space, leading to tissue hardening and scarring.

Renal interstitium
The renal parenchymal space that is outside the tubular, glomerular and vascular structures.

Fibrogenesis
The phenomenon of excessive production of extracellular matrix by fibroblasts. volume 21 | June 2020 | 347 nATuRe RevIeWS | MolECulAr CEll BioloGy EMT. A multifaceted and often reversible change in cellular phenotypes during which epithelial cells lose their apical-basal polarity, modulate their cytoskeleton and exhibit reduced cell-cell adhesive properties. Cells may individually or collectively acquire mesenchy mal features and increase motility and invasive ability. Typically, a switch in intermediate filament usage from cytokeratins to vimentin is observed after a complete EMT. Cortical actin filament in epithelial cells also undergoes marked rearrangement during EMT. While the characteristics of fully epithelial cells are relatively clearly defined, our current knowledge does not allow us to define the mesenchymal state with specific cellular characteristic or molecular markers that are universal endproducts of all EMT programmes.

MET.
Reciprocal changes in cellular phenotype that reverse EMTinduced phenotypes, during which mesenchymallike cells may acquire apical-basal polar ity, reorganize their cytoskeleton, and exhibit increased cell-cell adhesion, resulting in an organized epithelium. MET occurs during embryonic development (for exam ple, cardiac development, kidney morphogenesis and somite formation) and cancer.

Endothelial-mesenchymal transition.
As with epithelial cells, endothelial integrity depends on cell-cell junctions, apical-basal polarity and interactions with an underly ing basement membrane. Endothelial-mesenchymal transition (EndoMT) more accurately indicates the phe nomenon in such cell populations and resembles EMT in most aspects, except for the replacement of Ecadherin by VEcadherin. EndoMT thereby enables endothelial cells to attenuate or deconstruct their functional integrity and apical-basal polarity, to acquire motile and invasive behaviour and to activate changes in gene expression that are driven by certain EMTTFs. Similar to epithelial cells in EMT, endothelial cells that have activated EndoMT pro grammes exhibit a variety of intermediate or partial pheno types, as discussed above for EMT. EndoMT was described initially during embryonic heart development 71 and subsequently in the context of cardiac fibrosis 72 .
Epithelial-mesenchymal plasticity. We favour and rec ommend use of the term 'epithelial-mesenchymal plas ticity' (EMP) to describe the ability of cells to adopt mixed E/M features and to interconvert between intermediate E/M phenotypic states arrayed along the epithelialmesenchymal spectrum that cannot be easily distinguished on the basis of our current understanding. This plasticity has been variably referred to as partial EMT, hybrid E/M status, a metastable EMT state, EMT continuum and EMT spectrum; in all cases, the cells express a mixture of epithelial features (such as cytokeratins) and mesenchy mal features (such as cell migration) and markers. EMP indicates an ability to move readily between these various states, although the stability of the various states varies in different biological contexts. EMP is widely observed in development, wound healing and cancer. In addition to a mesenchymal type of migration, as observed during mesoderm formation, EMP can also participate in collec tive migration -for example, during tubulogenesis and wound healing. EMP also accounts for the reversibility of the EMT programme. Epithelial cells going through EMT give rise to cell populations that may enter reversi bly into states with various proportions of epithelial and mesenchymal features. EMP is thought to provide cells with the fitness and flexibility to fulfil the diverse require ments during the course of either developmental or pathological processes.

EMT-TFs.
In many if not most settings, both in cell culture and in vivo, EMP involves some degree of tran scriptional regulation. Several transcription factors belonging to the Snail, Twist and Zeb families have been found to control cell-cell adhesion, cell migra tion and ECM degradation, and to play evolutionarily conserved central roles in the execution of EMT in various biological settings and organisms (TABlE 1). Of note, all the developmental EMT processes described to date involve at least one member of these fami lies of core EMTTFs. Other transcription factors have been shown to impact EMT in certain contexts (Supplementary Table 1). However, these transcrip tion factors are also involved in other cellular processes (for example, proliferation, apoptosis or stemness). In addition, many of the EMTTFs are also expressed in nonepithelial cells, ranging from fibroblasts to haema topoietic precursors, and in cancer types involving nonepithelial derivatives (melanoma, glioblastoma and leukaemia), where they play important roles during tumour progression, often beyond classic EMT phases. Although we use the term EMTTFs to describe all tran scription factors associated with EMT, it is important to keep in mind that their expression alone is not sufficient to indicate that EMT is occurring.

Recommendations on the criteria to define EMT
In the current EMT literature, both cellular and molec ular descriptors have been used to define EMT in var ious biological systems. Below we provide several key recommendations on how to better use this informa tion to cover the enormous complexity and plasticity of the EMT programme in diverse developmental and pathological settings.

EMT status cannot be assessed on the basis of one or a small number of molecular markers.
EMT constitutes changes of cell behaviour that involve the loss of certain epithelial characteristics and the gain of certain mes enchymal traits. The complex series of cellular changes occurring during EMT require the cooperation of a large number of molecular factors. On the basis of their involvement in the process, these factors can be divided into three groups: EMTinducing signals, EMTTFs and EMT markers that define and constitute various epithelial and mesenchymal cell characteristics. In the literature, diverse cellular and molecular descriptors have been used to define EMT in different biological sys tems, which has been a major source of confusion. For example, some studies define partial loss of Ecadherin as an indication of EMT, while others argue that the maintenance of certain levels of expression of epithelial markers such as cytokeratins is indicative of cells not having undergone EMT. Given the complex manifesta tions of the EMT programme, it has become clear that inferring the involvement of EMT in any process cannot rely solely on a few salient molecular markers, such as Ecadherin and vimentin 73 . More importantly, the use of various EMT molec ular markers to characterize the phenotypic state of individual tumour cells has revealed that such cells, as described earlier, can simultaneously express both epi thelial and mesenchymal genes. The core EMTTFs are often coexpressed in various combinations in order to orchestrate complex EMT programmes, and they involve various members of EMTTF families, such as Snai (Snai1 and Snai2) and Zeb (Zeb1 and Zeb2), depend ing on the specific biological context 74 . Importantly, posttranscriptional regulation of EMT regulators at both the mRNA and protein levels is crucial in con trolling EMT. Such regulation is often neglected in studies that use RNA expression exclusively to survey EMT molecular markers. A focus on defining EMT programmes exclusively on the basis of the expression of specific molecular markers such as these under represents the enormous complexity and plasticity of the EMT programmes in diverse developmental and pathological settings.
The primary criteria for defining EMT status should be changes in cellular properties together with a set of molecular markers, rather than relying solely on molecular markers. One major feature that unites all the variant EMT programmes is the initial attenuation or deconstruction, to varying degrees and with diverse manifestations, of the epithelial phenotype. Epithelial cells harbour complexes that mediate cell-cell interac tions, most notably adherens junctions, tight junctions and desmosomes (Fig. 1). Apical-basal polarity guides the proper organization of tight junctions, adherens junctions and desmosomes in epithelial cells. Polarity complexes, including the Par, Crumbs and Scribble complexes [75][76][77] , define the apicallateral and basallateral domains of epi thelial cells (Fig. 1). During the early phase of EMT, loss of apical-basal polarity is often the first event to be observed and can lead to the destabilization of adhesion complexes, such as the tight junctions and adherens junctions at the lateral membrane 78,79 , as well as to the activation of EMTTFs 80 . The decrease or loss of epithelial adherens junctions and desmo somes occurs via transcriptional repression by the core EMTTFs of the genes encod ing junctional proteins. The cytoplasmic relocalization of adherens junction proteins, such as Ecadherin, via posttranscriptional regulation is also an early feature of EMT initiation in various EMT models 81 .
Another key function of the EMT programmes is to provide stationary epithelial cells with the ability to migrate by invading through extracellular matri ces secreted by both epithelial and mesenchymal cells. Thus, during EMT, epithelial cells often need to breach the basement membrane in order to migrate away from their epithelium of origin 35,82 . Migration of cells that have undergone EMT does not necessarily require the cells to lose all epithelial features, and a switch of intermediate filaments from cytokeratin to vimentin can facilitate cell migration. Depending on the extent of cell-cell adhe sion loss, epithelial cells can migrate as single cells, in a mesenchymal manner, or collectively, while remain ing attached with one another via weakened but still operative cell-cell interactions. This indicates that the breakdown of these tightly regulated epithelial struc tures, the gain of motility and the ability to degrade ECM during EMT are inaccurately represented by the simple expression, or lack of expression, of selected markers. Furthermore, complex posttranslational modifica tions of key proteins play critical roles in governing the complex cellular processes that occur during EMT. For these reasons, researchers should describe EMT and MET as functional changes in the biological proper ties of cells rather than focusing largely on changes in a few readily monitored molecular markers. It is through this perspective that our understanding of EMT could faithfully reflect the function of EMT during animal development and pathological events (Fig. 1). Therefore, whenever it is experimentally feasible, EMT should be assessed through the combination of cellular properties and multiple molecular markers.

EMT-TFs and other molecular markers are valuable indicators of EMT, but they should be assessed in conjunction with changes in cellular characteristics to define EMT.
The morphological and functional changes that can be observed in cells during EMT often result from changes in gene expression. Many, but not all, EMTassociated changes in gene expression result directly or indirectly from the actions of EMTTFs, which play key roles in driving EMT. Indeed, most EMT programmes are volume 21 | June 2020 | 349 nATuRe RevIeWS | MolECulAr CEll BioloGy associated with the activation of expression of one or several core EMTTFs. Although core EMTTFs often initiate EMTassociated changes in gene expres sion (TABlE 1), a large number of other EMTTFs and numerous microRNAs and long noncoding RNAs (lncRNAs) have also been shown to contribute to or play critical roles in diverse EMTs (Supplementary  Table 1). Decreased association between βcatenin or p120catenin and Ecadherin, achieved by posttranslational modifications, can also greatly weaken the adhesive functions of adherens junctions. Reduced expression of junctional and polarity proteins is often visible during EMT. Depending on the cell type and the extent to which cells advance through an EMT programme, cells undergoing an EMT may begin to express vimentin, to suppress cytokeratin, to shift expression of key integrins and so forth. These changes in gene expression are often seen as being indicative of EMT or as markers of EMT, although, considering the extensive variations in EMT, their overall value in the diagnosis of EMT needs to be considered with caution. Beyond this small set of commonalities, it is difficult to define other changes as contributing universally to all the diverse manifestations of EMT programmes that have been described in the rapidly expanding lite rature. Furthermore, our current knowledge does not allow us to know whether there is a linear succession of cellbiological changes as cells advance progressively through an EMT programme, or whether a diverse series of routes radiates in multiple directions from the starting point of attenuation or loss of epithelial junctions.

Finding reliable EMT markers requires a combinatorial approach, as well as distinguishing between EMT-associated and non-EMT-associated functions.
Analysing the state of a cell that is engaged in EMT often requires the use of markers that are specific to a specific biological context. To be able to assess in which position on the EMT spectrum a cell resides, it might be necessary to use a set of criteria that are standard ized for the specific biological context. Obtaining quan titative EMT marker measurements should always be coupled with cellular and functional analyses of EMT status, as described above. Importantly, recent studies have linked EMT to various other cellular programmes and functions, including cancer cell stemness, resistance to apoptosis, genome instability, cancer drug resistance and metabolic adaptation. Many components of EMT regulatory pathways, including EMTTFs, also affect other important cellular functions and phenotypes and are themselves regulated through diverse signals that may or may not involve canonical EMT. For example, EMTTFs such as Snai1/2 and Twist1 also regulate can cer cell survival and cancer drug resistance 66,83 , and it is currently unknown whether cell survival and drug resistance are regulated independently of EMT. Likewise, many ECM remodeling proteins that are important for breaching the basement membrane can be regu lated in both EMTdependent and EMTindependent manners. Furthermore, although cells may switch to a different cell fate upon EMT, the EMTassociated cel lular changes from epithelial to more mesenchymal phenotypes are independent of cell differentiation or cell dedifferentiation. Therefore, it is important to note that the cellbiological definition of EMT strictly refers to cellular features describing the EtoM phenotypes, while we should keep these associated phenotypes in mind when examining the role of EMT in various biological settings.
Implications for future EMT research Fifty years and over 38,000 publications after Betty Hay's pioneering observations, the concept of EMT has now been widely applied in biomedical research. It provides a unifying framework for developmental and cancer studies, which is evidenced by the exponentially grow ing number of EMTrelated publications. Such a frame work holds the promise of farreaching breakthroughs in cancer diagnosis and treatment, for cancer biologists, and of bridging the gap in understanding normal and pathological epithelial organization and morphogene sis, for developmental and cell biologists. To realize this promise, it is desirable that the EMT community reach a consensus on the definition of EMTrelated terms and on the conceptual framework for approaching EMT as a biological process with quantifiable molecular descriptors and cellular readouts. All the authors of this Consensus Statement article have agreed to adhere to the recommendations on nomenclature presented here in their future research publications and recommend that other researchers in the EMT and the larger biological research communities also follow these guidelines. Only by minimizing semantic misinterpretation and data mis communication can we begin to appreciate the diversity of individual EMTs and uncover conserved themes in EMT regulation between development and disease.
Studies using cell lines, developmental systems and cancer models have revealed a diversity of EMTinduced phenotypes and have highlighted remarkable complex ity in the execution and regulation of EMT. Looking forward, to decipher the complexity and plasticity of the EMT programme, we propose that EMT research, while remaining anchored in traditional developmen tal, cell and cancer biology, should be explored within a broader conceptual context. The EMT field has in recent years attracted the interest of a diverse group of researchers with expertise in systems biology, biophysics, stem cell biology, pathology and mathematical model ling. This remarkable strength of interest will enable crossdisciplinary collaborations and push this field of research forward. We expect that future EMT stud ies will apply multidisciplinary approaches in order to gain increased mechanistic understanding of EMT. One open question in the cancer EMT field is the extent to which the stabilization of specific hybrid E/M states, or the dynamic switch between E/M states in response to distinct cues from the microenvironment, favours the metastatic process. With many important aspects of EMT remaining unexplored, advancing EMT research will require technological innovations to enable the study of both developmental and cancerassociated EMT at the singlecell level. These innovations will include singlecell live imaging, lineage tracing, gene expression analyses and studies of genetic and epigenetic

Cancer cell stemness
The presence of subpopulations of cancer cells that exhibit stem cell-like behaviour.

Dedifferentiation
The conversion of cells from a more differentiated state to a less differentiated one. Dedifferentiation is often associated with the acquisition of progenitor or stem cell-like properties.
modifications. Finally, a combination of mathematical modelling with carefully constructed experimental analyses will be important to gaining a mechanistic understanding of EMT plasticity.
Another major challenge is the translation of the cur rent knowledge of EMT heterogeneity and plasticity into clinical practice. While we are far from understanding the functional implications of EMT heterogeneity, sev eral clinical trials have already incorporated the notion of EMT plasticity 3 , thereby opening the way for novel therapies that exploit EMT heterogeneity. Singlecell sequencing of normal tissues, primary tumours, circu lating tumour cells and metastases, combined with cellu lar analyses and functional validations, will capture the diversity and plasticity of EMT and have the potential to reveal the molecular alterations underlying tumour progression and the diverse responses to therapy.
An increased understanding of the EMT mechanisms associated with these behaviours offers the poten tial for targeted therapy to prevent cancer metastasis. For example, while the inhibition of EMTassociated changes might reduce cancer cell dissemination in earlystage carcinoma, preventing MET in dissemi nated tumour cells might inhibit metastatic outgrowth in distant organs. Experimental and clinical studies have shown that the development of resistance to vari ous therapies, including chemotherapies and immuno therapies, is tightly associated with EMT phenotypes 83 . These studies suggest that targeting EMT, or the cells capable of executing it, holds promise in overcoming therapy resistance, which is a major challenge in cancer treatment.