How studies in developmental epithelial-mesenchymal transition and mesenchymal-epithelial transition inspired new research paradigms in biomedicine Free

Epithelial cells usually form an organized continuous sheet or layer known as an epithelium, which forms a protective barrier within the body. Epithelial cell characteristics include apico-basal polarity, extensive cell-cell interactions (including adherens and tight junctions), as well as distinctive cell morphologies (e.g. squamous, cuboidal or columnar). Epithelial-mesenchymal transition (EMT) is the process by which epithelial cells adopt a mesenchymal state, when cells are loosely organized, have unstable polarization and can be motile. Some distinguishing features of EMT include the progressive loss of epithelial markers and the gain of mesenchyme genes. In addition to transcriptome changes, there is extensive remodeling of the cytoskeleton, including the development of cell protrusions, breakdown of cell junctions and loss of polarity (Huang et al., 2012; Nieto et al., 2016; Yang et al., 2020) (Fig. 1). Conversely, mesenchymal-epithelial transition (MET) is the reverse process by which mesenchymal cells organize – or reorganize – into an epithelium (Pei et al., 2019).

EMT and MET are observed frequently during embryonic development (reviewed by Lim and Thiery, 2012; Sheng, 2021), with the first example of EMT often occurring during gastrulation, such as following invagination of the ventral furrow in Drosophila, a transient structure that appears in the ventral region of the blastoderm of insects and allows mesodermal progenitor cells to invaginate and subsequently separate from the blastoderm epithelium to migrate along the inner surface of the blastoderm cavity (Leptin, 1999). Gastrulation in amniotes may involve EMT at the primitive streak, which appears as a groove in the epiblast where mesodermal and endodermal progenitor cells ingress (Najera and Weijer, 2023). Another striking example of EMT operates in neural crest cells (NCCs), which dissociate from the lateral borders of the neural plate and provide mesenchyme that can differentiate into a distinct variety of cell types primarily related to the nervous system, skin pigmentation and craniofacial structures (reviewed by Leathers and Rogers, 2022). These mesenchymal cells can migrate within the embryo to form specific tissues and produce localized extracellular matrix (ECM) (Perris and Perissinotto, 2000).

Mesenchymal cells can subsequently form epithelial structures by MET. MET is crucial to forming the mammalian epiblast and, subsequently, organ rudiments from post-migratory mesenchymal cells. In some tissues, cells can undergo secondary or tertiary EMT to become motile again; e.g. in the somite (Christ and Ordahl, 1995), valve formation from endocardium (Markwald et al., 1977; Mikawa and Gourdie, 1996), dissolution of the Mullerian ducts (Trelstad et al., 1982) and fusion of the palate from opposing palatal shelves (Fitchett and Hay, 1989).

EMT is also a key feature in the promotion of carcinoma progression. Carcinoma cells acquire different degrees of EMT to engage in local invasion of the para-tumoral tissue at the primary site. EMT can also favor the distant dissemination of carcinoma cells, which establish as micrometastases in distant organs. Foci of carcinoma cells eventually expand as clinically detectable metastases through the activation of MET. EMT is also involved in forming carcinoma stem cells, allowing clonal expansion (Lambert and Weinberg, 2021; Nieto et al., 2016; Thiery, 2002; Tsai et al., 2012). Therefore, an understanding of EMT biology has important implications for developing treatments interfering with the progression of carcinoma.

The contributions of developmental biology to biomedicine have been significant in elucidating genetic and molecular mechanisms underlying human diseases, including developmental differences, fibrosis, cancer and aging. Here, we illustrate how studies of EMT and MET in developmental biology have impacted the wider study of biology, including evolution, disease and biomedicine. We discuss the history of EMT research that led to the identification of key molecular players that regulate the process (Fig. 2). We then focus on examples of EMT and MET that occur during cardiac development to illustrate events where these processes can be distinguished from concurrent differentiation into the multiple cell types originating during gastrulation or in neural crest development. We then discuss how these studies in developmental biology have impacted the broader field in the context of evolution and in vitro cell differentiation. We conclude by discussing how studies of EMT in development have informed our understanding of developmental differences, diseases including cancer and therapeutic opportunities. This Primer largely focuses on EMT because less is known about the molecular mechanisms driving MET but we do mention examples where relevant.

EMT was probably first described in vivo as an ‘epithelial-mesenchymal transformation’ by Elizabeth Hay after studies of primitive streak formation in chick (Greenburg and Hay, 1982) (Fig. 2). In parallel, in vitro analyses of collagen gel cultures, showed that cell migration could be seen within an ECM (Bernanke and Markwald, 1982; Elsdale and Bard, 1972; Schor, 1980). Subsequent use of these gels demonstrated that EMT could be induced and visualized using embryonic epithelial and endothelial cells (Greenburg and Hay, 1982; Runyan and Markwald, 1983). These experiments pointed to ECM components as potential inducers of EMT. Indeed, a link between ECM and cell morphology was previously noted by Overton because frog epithelial-derived cells covered basal lamina surfaces, whereas mesenchymal-derived cells could invade them (Overton, 1979).

Molecular insights into EMT-related processes arose from concurrent studies of tumor cells in vitro. Tumor cells secrete transforming growth factor α (TGFα) and TGFβ into conditioned media, which can induce normal cells to acquire a transformed phenotype in soft agar culture (de Larco and Todaro, 1978). Subsequent studies found that TGFβ regulates cell proliferation (Roberts et al., 1980; Sporn and Todaro, 1980), ECM secretion (Ignotz and Massague, 1986) and the expression of cell adhesion receptors (Heino et al., 1989). The use of ‘transforming’ when initially describing TGFβ was somewhat of a misnomer because the expression of ECM proteins and receptors by normal cells in soft agar enabled them to anchor and survive in this milieu, without actual transformation into a tumor phenotype (Ignotz and Massague, 1986). However, further studies with collagen tissue culture and embryonic heart tissue identified the ability of TGFβ isoforms to induce EMT in development (Potts et al., 1991; Potts and Runyan, 1989) and subsequent studies linked EMT and metastasis with TGFβ (Miettinen et al., 1994). Furthermore, EMT is now associated with a variety of extracellular signals, including fibroblast growth factors (FGFs), Notch and Wnts, that either induce or regulate the inductive response to EMT (Lamouille et al., 2014; Savagner et al., 1994; Thiery et al., 2009; Valles et al., 1990). However, TGFβ remains the most commonly used or investigated inducer of EMT both in vivo and in vitro (Derynck and Akhurst, 2007; Katsuno and Derynck, 2021) (Fig. 3).

The genetic regulation of EMT was revealed by the large-scale mutagenesis screen in Drosophila by Nusslein-Volhard and colleagues (Nusslein-Volhard and Wieschaus, 1980; Nusslein-Volhard et al., 1984), which uncovered mutations that abolished the formation of the ventral furrow, which marks the initiation of gastrulation in Drosophila. Through the combined efforts of several groups, two genes, snail (sna) and twist (twi), were shown to encode zinc-finger and basic helix-loop-helix transcription factors, respectively (Boulay et al., 1987; Thisse et al., 1988). sna and twi control the formation of mesoderm in Drosophila (although with different penetrance) via a gene-regulatory network in which Dorsal induces both twi and sna expression and twi can maintain its own expression (Alberga et al., 1991; Leptin, 1991). In addition to their role in mesoderm specification, Twist and Snail promote invagination during gastrulation by orchestrating the remodeling of ventral epithelial cell adherens junctions in the blastoderm, thereby constricting their apex and causing tissue bending (Leptin, 1999). After invagination, EMT is initiated by Twist through the activation of FGF receptor signaling (Leptin, 1999; Leptin and Affolter, 2004).

Based on the initial findings in Drosophila embryos, the role of sna was addressed in vertebrates after identifying two orthologs: Snai1 in mouse (Nieto et al., 1992) and SNAI2 in chick and Xenopus (Nieto et al., 1994). In the chick embryo, SNAI2 (previously known as SLUG) is crucial for the ingression of mesodermal cells at the primitive streak and NCC individualization from the neural epithelium (Nieto et al., 1994). The inactivation of Snai1 in mouse causes gastrulation defects in which the mesoderm forms but does not acquire a mesenchymal phenotype (Carver et al., 2001). These striking findings suggest that EMT is transcriptionally regulated by an identifiable set of genes, such as SNAI1 and SNAI2 in vertebrates. The use of the embryonic heart and collagen assay models mentioned previously also linked TGFβ signaling to the regulation of Snai2 expression (Romano and Runyan, 2000). In addition to vertebrates, a similar scenario (albeit somewhat different regarding the signaling cascades) was uncovered in sea urchin gastrulation (Saunders and McClay, 2014; Wu and McClay, 2007; Wu et al., 2008), in which Snail and Twist also initiate EMT (Byrum and Martindale, 2004; Technau, 2020). Thus, these two controlling genes are conserved essential regulators of gastrulation in metazoans. Nonetheless, Snail family members are also implicated in cell survival and motility (Barrallo-Gimeno and Nieto, 2005).

Several other transcription factors were also identified as EMT inducers (Yang et al., 2020), most notably Zeb1 and Zeb2, the vertebrate orthologs of the Drosophila ZFH-1 gene (Postigo and Dean, 2000). Zeb1 and Zeb2 contain three and four zinc fingers, respectively, which recognize bipartite E boxes in gene promoters, such as E-cadherin (Peinado et al., 2007). Inactivation of ZEB2 impacts the migration of vagal NCCs, leading to Hirschsprung phenotype and the persistence of E-cadherin expression in the neural tube (Van de Putte et al., 2003). The evidence that Zeb2 is a bona fide transcription factor that induces EMT was initially obtained in vitro in control and carcinoma cell lines (Comijn et al., 2001). Zeb1 also induces EMT in carcinoma (Eger et al., 2005). Both Zeb1 and Zeb2 are negatively regulated by a feedback loop involving microRNA (miR)200 and mir205 family members (Gregory et al., 2008). There is currently a consensus that EMT is mediated by one or more of these core transcription factors (Snai1, Snai2, Zeb1, Zeb2 and Twist2), although at least 20 additional transcription factors are associated with EMT and may replace or combine with the core factors during EMT (Yang et al., 2020). We next discuss the extent to which other features of EMT are conserved within the animal kingdom.

Genetically programmed obligate multicellularity is an essential feature of animal development. Epithelial structures generate sub-compartments with coordinated apico-basal polarity, allowing functional specialization in a multicellular organism. Ectoderm- and endoderm-derived tissues exhibit epithelial characteristics as their initial and final morphology, undergoing EMT and MET as in the mammary placodes and liver bud. Mesoderm cells exhibit a mesenchymal morphology upon differentiation. As they are located between the two epithelial structures of ectoderm and endoderm, their functionality is defined by motility and morphological flexibility. Nevertheless, most mesoderm lineages undergo multiple rounds of EMT and MET in development to arrive at their final homeostatic states. These may be of an epithelial nature for internal compartmentalization (e.g. kidney, vessels, mesothelial cells, etc.) (Sheng, 2021). The evolution of epithelial and mesenchymal characteristics, however, predates that of three-germ layered animals. Non-bilaterian metazoan clades (Porifera, Placozoa, Ctenophora and Cnidaria) all have epithelial tissues of specialized physiological functions and a small number of mesenchymal cells sandwiched between layers of epithelial cells (Eitel et al., 2018; Leys et al., 2009; Nielsen, 2019; Schierwater et al., 2021; Srivastava et al., 2008). Interestingly, some of the molecular features of EMT are even present in choanoflagellatea, the closest unicellular relative of Metazoa (Box 1).

EMT and MET are reported to be involved during wound healing and regeneration among those clades (Costa et al., 2020; Kraus et al., 2020; Technau, 2020), suggesting that the last common ancestor of animals was already equipped with cellular biological features, which are now known as ‘epithelial’ and ‘mesenchymal’. Supporting this, phylogenomic analyses reveal the presence of almost the entire molecular repertoire of epithelial markers in those stem metazoans, including cadherin-mediated homophilic cell-cell interaction, collagen/laminin/fibronectin-based matrix proteins, integrin-mediated cell-matrix interaction, apico-basal polarization, and WNT and TGFβ signaling pathways (Belahbib et al., 2018; Jacques et al., 2022; Nielsen, 2019; Srivastava et al., 2008) (Fig. 4). Core EMT transcription factors, including members of the SNAI and TWIST gene families, appeared before Bilateria (Barrallo-Gimeno and Nieto, 2009; DuBuc et al., 2019), and the ZEB gene family appeared at the origin of Bilateria (Seetharam et al., 2010). These lines of evidence support the concept that epithelium, mesenchyme and their interconversions are key, ancient metazoan traits, possibly being used to regulate functional specialization and morphogenetic plasticity in the multicellular organization, especially that of epithelia, and in their response to environmental stress (e.g. wound healing and regeneration). The molecular regulatory network controlling EMT is therefore predicted to pre-date networks that control other conserved bilaterian features, such as gastrulation and dorso-ventral and antero-posterior tissue polarization.

The observation that EMT and MET during embryogenesis are often accompanied with lineage specifications indicates that they might have roles in addition to stimulating cell migration and morphogenesis. With the elucidation of core EMT transcription factors and signaling pathways, potential functions of EMT and MET in regulating cell fate conversions can now be directly investigated in in vitro cell cultures. In immortalized, non-tumorigenic human mammary epithelial cells (HMLEs), ectopic expression of EMT transcription factors (Snail or Twist) or TGFβ treatment induces EMT and generates stem cell-like cells (Mani et al., 2008). Furthermore, combined ectopic expression of Snai2 and Sox9 in primary mammary epithelial cells (MECs) induces the formation of mammary stem cells where Snai2 functions to stimulate EMT and promote the conversion of luminal progenitors to mammary stem cells (Guo et al., 2012). In vivo studies in mice have confirmed that Snai2 regulates gland-reconstituting mammary stem cells and mammary gland morphogenesis (Nassour et al., 2012; Ye et al., 2015). Together, these studies suggest that EMT transcription factors induce a cellular status favoring the acquisition of stemness. Based on this idea, a protocol that induces the differentiation of mesenchymal tumor-initiating cells (TICs) in breast cancers has been established by PKA-stimulated MET (Pattabiraman et al., 2016).

Somatic reprogramming provides an ideal in vitro model to investigate the role of EMT and MET in cell fate conversions. Dynamic changes between epithelial and mesenchymal statuses have been observed in the classic Yamanaka factors (Oct4, Sox2, Klf4 and Myc) induced somatic cell reprogramming. It is reported that a BMP-driven MET facilitates the initiation of somatic reprogramming of mouse embryonic fibroblasts (Samavarchi-Tehrani et al., 2010), Sox2, Oct4 and Myc act to suppress the mesenchymal TGFβ/Snail signaling, and Klf4 promotes the epithelial program during reprogramming (Li et al., 2010). Conversely, genes that promote actin stress fiber formation in fibroblasts, such as TESK1 and LIMK2, are barriers to MET and somatic reprogramming (Sakurai et al., 2014). Interestingly, the sequential introduction of Yamanaka factors first activates a transient EMT program to transduce mouse embryonic fibroblasts into a ‘primed status’ that can efficiently undergo MET and reprogramming (Liu et al., 2013). Further investigation of the early EMT in a chemically defined reprogramming medium revealed that the metabolic switch from oxidative phosphorylation to glycolysis (OGS) cooperates with EMT to facilitate reprogramming. EMT transcription factors stimulate the expression of glycolytic genes to facilitate OGS while OGS stimulates the expression of mesenchymal genes through epigenetic modifications; thus, they form a positive-feedback loop during early reprogramming (Sun et al., 2020). A similar metabolic switch to glycolysis occurs in the neural crest at the onset of migration and it functions to promote EMT and migration (Bhattacharya et al., 2020).

Embryonic stem cells (ESCs) are typical epithelial cells, and multiple EMT and MET cycles are observed during targeted differentiation of ESCs in vitro, reminiscent of EMT and MET associated with lineage differentiation in vivo. Typical EMT changes (such as downregulation of E-cadherin, upregulation of N-cadherin, SNAI1 and SNAI2, and increased motility) occur during feed-free induced differentiation of human ESCs (hESCs) (Eastham et al., 2007). Inhibition of E-cadherin in hESCs leads to a partial EMT, which is not sufficient to induce differentiation (Aban et al., 2021); however, E-cadherin inhibition switches hESCs from self-renewal to a differentiation program upon Wnt stimulation by releasing cytoplasmic β-catenin to stimulate the expression of SNAI2 (Huang et al., 2015).

EMT is also observed in activin A-induced definitive endoderm (DE) differentiation of hESCs and is functionally required because inhibition of EMT by TGFβ inhibitor or SNAI1 knockout blocks DE differentiation (D'Amour et al., 2005; Li et al., 2017). Similarly, Snai1 knockout mouse ESCs fail to initiate EMT and these cells cannot exit the epi-ESC status and are thus defective in lineage commitment (Lin et al., 2014).

EMT and MET cycles have also been reported in in vitro transdifferentiation. For example, human gastrointestinal epithelial cells can be induced into endodermal progenitor cells by defined factors that can further differentiate into hepatocytes and other endodermal lineages (Wang et al., 2016). During this process, a round of EMT and MET occurs, which mirrors a similar EMT and MET cycle during hepatic differentiation of hESCs (Li et al., 2017). An EMT and MET cycle has also been reported during transdifferentiation of mouse embryonic fibroblasts into neuronal cells, and both the EMT and the MET are required for efficient neuronal conversion (He et al., 2017). Similarly, transient TGFβ treatment improves the transdifferentiation of mouse astrocytes into induced dopaminergic neurons (Rivetti di Val Cervo et al., 2017).

Together, these data support the idea that a transient mesenchymal state (partial EMT) generally facilitates cell fate conversions in somatic reprogramming, differentiation and transdifferentiation. EMT-related events, such as TGFβ signaling, EMT transcription factors, E-cadherin or metabolic switches have been proposed to regulate cell fate conversions in these in vitro models. Testing whether some of these mechanisms also operate in EMT-accompanied lineage differentiation in development will be interesting.

Developmental EMTs continue to have much to offer in the elucidation of the fundamental aspects of the process. As pathological EMTs reflect the recapitulation of embryological events, mechanisms found in development will continue to provide context for EMT pathologies. A significant advantage to developmental systems is timing. Whereas EMT in pathology is often a slow and asynchronous process extending over weeks to years, developmental EMT is relatively rapid and the sequence of events in a tissue undergoing EMT is often completed quickly, as seen in the temporal analysis of neural crest or cardiac cushion formation (Basch et al., 2000; Runyan and Markwald, 1983). This temporal advantage can be coupled with a collection of large numbers of samples to produce sufficient materials for detailed molecular analysis.

In addition to advancing our understanding of embryonic development, the ability to generate patient-specific induced pluripotent stem cells (iPSCs) provides opportunities to investigate the pathogenesis of these diseases and to develop potential therapeutic strategies in vitro. As MET is an early obstacle for somatic reprogramming of fibroblasts, cells with epithelial properties might be better source cells for reprogramming. To this end, urine-derived cells, which can be collected non-invasively and have epithelial-like properties have been established as the preferred source for the generation of human iPSCs, either by retroviral or episomal (virus-free) delivery of Yamanaka factors (Xue et al., 2013; Zhou et al., 2011). Furthermore, tumorigenic Myc can be replaced by the SV40T+miR-302-367 cluster, which promotes MET to facilitate reprogramming (Liao et al., 2011) to improve the safety profiles further (Xue et al., 2013). Thus far, dozens of patient-derived iPSCs have been successfully established by a highly accessible, virus-free and chemically defined reprogramming protocol (Xue et al., 2013) that overcomes some of the safety limitations for deriving iPSC lines and paves the way for their future clinical applications in regenerative medicine.

The complete failure of EMT or MET is incompatible with life in metazoans, as there are no disorders in newborns linked to a complete loss of EMT- and/or MET-associated mechanisms. This is consistent with mouse models of EMT regulators, such as Snai1, which are embryonic lethal due to loss of mesenchyme formation at gastrulation (Carver et al., 2001). It does appear that partial or local defects in components of EMT or MET produce developmental differences.

Two of the most common developmental differences involve cardiac and craniofacial (palate) development (www.cdc.gov/ncbddd/birthdefects/data.html). Valvular and septal defects are commonly observed heart defects that appear to reflect a partial loss of EMT (Briggs et al., 2012). Pathogenic classification of congenital cardiac malformations prominently focuses on defects produced by EMT-derived neural crest and cardiac cushion cells (Clark, 1996).

As TGFβ is a central regulator of EMT and MET, examining mouse models with TGFβ signal impairment provides further evidence that these processes are involved in developmental differences. Although the three TGFβ isoforms provide some redundancy of function (Roberts and Sporn, 1992), the loss of TGFβ3 produces neonatal mice with a palatal defect and neonatal death (Brunet et al., 1995). Mice with a deletion of TGFβ2 have cardiac valve defects (Azhar et al., 2011; Sanford et al., 1997). Double TGFβ2 and TGFβ3 mutant mice have more severe cardiac defects than TGFβ2 mutant mice (Dunker and Krieglstein, 2002). Furthermore, defects in EMT-derived tissues are associated with loss of TGFβ receptors, including Alk2, Alk3, Alk4, TBRII, TBRIII and endoglin, as well as downstream signal transduction molecules including Smad2 and Smad4 (Goumans and Mummery, 2000; Mercado-Pimentel and Runyan, 2007).

Two distinct events in the heart are instructive of EMT mechanisms: first, the formation of heart valve primordia by the endothelium lining the heart (Markwald et al., 1977); and second, the formation of the epicardium (Dettman et al., 1998; Mikawa and Gourdie, 1996) (Fig. 5).

The heart is initially derived from a subset of primary mesenchyme produced at gastrulation. These mesenchymal progenitors briefly form a splanchnopleural cell layer (MET) in the early embryo and then undergo a secondary EMT to produce the mesenchymal progenitors of the myocardium and the endocardium. Formation of the myocardial and endocardial cell layers is achieved by a secondary MET process (von Gise and Pu, 2012). Thus, the early heart is a tube of endothelium surrounded by a tube of developing myocardium, each produced by two rounds of EMT and MET. ECM secreted by the myocardium separates these definitive myocardial and endocardial cell layers called the cardiac jelly (Person et al., 2005). In the region where the atrioventricular valves (mitral and tricuspid) develop, there is a further expansion of the cardiac jelly into two opposing swellings (cardiac cushions) secreted from the adjacent myocardium. EMT in the atrioventricular canal is sometimes called endothelial-mesenchymal transition (EndoMT). EndoMT is defined as the EMT of a specific single-layered epithelium common to vascular tissues; however, there appears to be little difference in the fundamental mechanisms of signaling and transition by which EMT is accomplished.

The formation of cardiac valves from endothelial cells of the cardiac cushions as a tertiary EMT (EndoMT) was first detailed in electron micrographs (Markwald et al., 1979, 1977). The endothelial cells of the valve-forming region become hypertrophied, dissociate from each other and extend filopodia into the underlying ECM. A subset of these cells delaminate from the endothelial layer and invade the cardiac jelly (Markwald et al., 1977). As mentioned previously, the invasion process seen in vivo is closely mimicked in vitro by the development of a collagen gel assay (Runyan and Markwald, 1983).

Chick atrioventricular canal heart segments, collected before EndoMT, produce an endothelial monolayer surrounding an aggregate of myocardial cells on the surface of collagen gels. When the subjacent myocardium is retained in the cultures, the endothelium develops a fibroblastic morphology accompanied by cell-cell separation. A subset of the separated cells continues to invade the underlying gel matrix as mesenchymal cells. Removal of the myocardium from the explant before the endothelial cells display separation, disrupts this EndoMT (Runyan and Markwald, 1983). These data suggest EndoMT is induced by a signal produced by the adjacent myocardium and has been shown to occur in vitro with the same developmental timing observed in the intact embryo. Furthermore, the in vitro assay faithfully recapitulates differences between the AV canal and ventricular tissues (where EndoMT does not occur). The collagen assay was then used to show that cardiac EndoMT could be induced by the addition of 5-10 ng/ml of any exogenous TGFβ isoform or inhibited by the addition of a blocking pan-TGFβ antibody (Nakajima et al., 1997; Potts and Runyan, 1989). Although TGFβ was known to regulate cell proliferation (Massague et al., 1991), this was the first evidence that it was an inducer of EMT (Potts and Runyan, 1989). Subsequent work showed that the TGFβ2 isoform is the most sensitive inducer of the EMT transcription factors Snai2 and Runx2-I and the most likely isoform to produce this EndoMT (Romano and Runyan, 2000; Tavares et al., 2018). The transcription factors Zeb1 and Zeb2 appear to be regulated by additional inductive signals, such as olfactomedin 1 (OLFM1) or TGFβ3 (Boyer et al., 1999; Doyle et al., 2006; Lencinas et al., 2013).

Although all endothelial cells of the atrioventricular canal appear to be activated by the initial inductive signal (Markwald et al., 1979; Runyan et al., 1990), only 7-10% of cells complete cell invasion into the ECM; remaining cells revert to endothelia (Lencinas et al., 2013). Differential regulation of cell separation and invasion, and changes in transcription factor expression during EndoMT suggests that the valve-forming endothelia display epithelial-mesenchymal plasticity and that additional signals are involved in the complete transition to mesenchyme (Lencinas et al., 2013; Lotto et al., 2023; Zhang et al., 2018).

Finally, experimental analysis based on trichloroethylene exposure, as an example of a potent cardiac teratogen, shows that both direct inhibition of EMT in vitro and the perturbation of contraction and blood flow are important signaling components of valvular EndoMT in vivo (Boyer et al., 2000; Makwana et al., 2010). Trichloroethylene exposure in humans is linked to cardiac defects involving valves and septa (Bove et al., 2002; Goldberg et al., 1990).

The epicardium is a cell layer that arises from the pro-epicardial organ to cover the outer surface of the heart with a superficial layer of cells during the heart looping stage of development (Ho and Shimada, 1978). After enveloping the heart, the epicardium undergoes an EMT to produce a population of mesenchymal cells visible in the sub-epicardial matrix between the epicardium and the myocardium, and within the cardiac muscle (Braitsch and Yutzey, 2013; Mikawa and Gourdie, 1996). Unlike the mesenchyme derived from the endothelium that produces little invasion into the myocardium, epicardial-derived mesenchymal cells penetrate the myocardium and eventually transit across the myocardium to contribute to the heart valves (Lockhart et al., 2014). Lineage tracing of cells of epicardial origin has shown that they differentiate into cardiac fibroblasts and several cell types of the coronary blood vessels, including smooth muscle cells and hematopoietic cells, as well as a limited number of endothelial cells (Mikawa and Gourdie, 1996). Studies in the chick and the mouse have linked retinoic acid, FGF2 and PDGFα to epicardial EMT (Braitsch and Yutzey, 2013; Dettman et al., 1998; Smith et al., 2011). Transcription factor regulation of epicardial EMT is complex with many factors identified, including roles for Wt1 and Hand2, but the core EMT transcription factors, Snai1, Snai2 and Twist1, are also involved in this EMT (Braitsch and Yutzey, 2013; Takeichi et al., 2013).

Wound healing in epithelial tissues is a fundamental process in tissue repair. Any defect in the orchestrated healing process may lead to the formation of chronic wounds or scars and keloids in humans (Vu et al., 2022). In mouse skin models of wound healing (Vu et al., 2022), basal and suprabasal epithelial cells at the leading edge of the wound undergo a partial EMT characterized by a decrease in junctional components of the adherens junctions and hemidesmosomes, transition to calcium-sensitive desmosomes (Garrod et al., 2005; Wang et al., 2018), and a decrease of transcription factors associated with the maintenance of the epithelial state, such as Ovol2 (Haensel et al., 2020). These migratory cells also acquire a repertoire of integrins interacting with the ECM and increased metalloproteolytic activities. Notably, these pioneer cells also express higher levels of SNAI2 than epithelial cells of healthy skin (Haensel et al., 2020; Savagner et al., 2005). Leading cells progressively cover the granulation tissue, helped by cells localized more peripherally to the wound that contribute to the regeneration of the epidermis by proliferation and collective cell migration. The leading cells must acquire transiently an intermediate EMT phenotype to migrate and cover the wound. Any alteration in the acquisition of the intermediate state will impair healing. Deletion of Ovol2 accelerates the migration of individual cells rather than collective cell migration. A defined balance of activity of Ovol2 and Zeb1, is one mechanism allowing these cells to be momentarily positioned in an intermediate EMT stage (Haensel et al., 2019). The intermediate EMT migratory leader cells transit from oxidative phosphorylation to glycolysis and hypoxia (Haensel et al., 2020). Topical application of MiR210, a hypoxia-induced gene, restores effective wound healing in diabetic mice by triggering glycolysis (Narayanan et al., 2020). A similar deregulation of the EMT phenotype is observed in hypertrophic scars and keloids, which express higher levels of Fsp1 (Aifm2) and vimentin (Hahn et al., 2016). Hyperbaric oxygen therapy, applied to combat the adverse effect of excessive EMT, similarly attempts to reduce the number of activated fibroblasts in keloids. This approach may take advantage of bone marrow stem cells, which can enhance wound healing by activating MET in the migratory keratinocytes (Chen et al., 2012). EndoMT, and its reverse mechanism MendoT, are crucial mechanisms operating to transiently establish a de novo capillary network to provide an adequate blood supply during wound healing. However, a sustained EndoMT contributes to an accelerated fibrotic process, such as keloid formation (Gurevich et al., 2021; Zhao et al., 2021). It has been suggested that inhibiting the Notch and Sox9 pathways significantly reduces scarring. Another strategy to control the transient intermediate EMT phenotype would be to activate IL17 secreted by skin-resident lymphocytes to alter downstream HIF1a and the glycolytic pathways (Konieczny et al., 2022).

Fibrosis, caused by an excessive number of stromal cells, including activated fibroblasts, considerably alters the functions of essential organs such as the lung, liver and kidney, resulting in an extremely severe and potentially deadly disease. The origin of activated fibroblasts was believed to be, in part, derived from the epithelial or endothelial components of these tissues undergoing extensive EMT (Kalluri and Neilson, 2003; Zeisberg et al., 2007a,b). However, these findings were challenged by subsequent studies in the liver, showing that hepatocytes do not engage an EMT program and thus are not transdifferentiated into myofibroblasts (Taura et al., 2010). In more recent studies, the proportion of myofibroblasts derived from an EMT of the kidney epithelium did not exceed 5%, or 10% if derived from the endoMT, which are significantly below the original estimates (LeBleu et al., 2013). To further investigate this crucial issue, additional studies in murine models took advantage of fibrosis induction by either unilateral ureteral obstruction, folic acid or nephrotoxic serum-induced nephritis in the kidney epithelium (Grande et al., 2015; Lovisa et al., 2015). In this model, the deletion of Snail or Twist in the kidney epithelium attenuated fibrosis. The kidney epithelium underwent a partial EMT but did not transform into myofibroblasts. The expansion of this population resulted from the massive production of TGFβ, whereas the production of TNFα created an adverse inflammatory reaction. In addition, TGFβ arrested kidney epithelial cells in G2, thus preventing their regeneration in the altered epithelium. These new findings have clarified the indirect role of EMT in fibrosis, leading to modified therapeutic interventions to interfere with TGFβ receptor signaling that preferably target downstream signaling by STAT3 or NR4A1 (Nieto et al., 2016). BMP7 signaling through Smad1, Smad5 and Smad8 can also be used to compete with the canonical Smad2/3 and Smad4 complex (Kim et al., 2020). Molecular pathways associated with fibrosis in diabetic nephropathy may also guide the development of new therapeutic strategies (Cao et al., 2022).

Acute heart injury induces massive fibrosis. These fibroblasts derive in part from endoMT. However, the reverse phenomenon (i.e. MendoT) can contribute to neo-vascularization. This finding can be leveraged to attenuate the impact of fibrosis and to restore proper vascularization (Ubil et al., 2014). Understanding the mechanisms generating fibroblasts directly or indirectly from EMT or EndoMT will contribute to the refinement of therapeutic strategies targeting fibrosis in different organs, as there are unmet needs for these diseases that have become prevalent worldwide (Liu et al., 2022).

Mesenchymal status confers refractoriness to chemotherapy and targeted therapeutics (Nieto et al., 2016). Most recent studies demonstrate the importance of EMT intermediate states. For example, the analyses of carcinoma cell populations derived from skin tumors of a mouse transgenic model have revealed that different intermediate states exhibit either proliferative, invasive, stem-like or metastatic properties. These different populations in the primary tumors are likely to cooperate to execute all the steps initiating metastasis (Pastushenko et al., 2018). A similar finding was found in a pancreatic carcinoma model through the lineage tracing of carcinoma cells during tumor progression (Simeonov et al., 2021) and highlighted that only some EMT intermediate states can lead to dissemination and detectable metastases. The documentation of different EMT statuses is present in multiple cancers, including primary breast cancers and circulating tumor cells (Sarrio et al., 2008; Tan et al., 2014; Yu et al., 2013). The concept of EMT has now been taken into consideration in more than forty clinical trials (clinicaltrials.gov) and a tentative list of drugs potentially interfering with EMT in carcinoma has been recently established (Castaneda et al., 2022; Jonckheere et al., 2022).

Extensive developmental studies have revealed that EMT and MET-regulated cell migration and morphogenesis, many of which are accompanied by lineage differentiations, are involved in developing various organs. Early models of cardiac and palatal EMT first identified essential elements of EMT, including tissue interaction, TGFβ-mediated regulation of EMT transcription factors, and the alternative roles of EMT and apoptosis. Additional developmental, oncogenic and fibrotic model systems have greatly expanded the diversity of EMT regulators and their roles in development and pathologies. EMT may contribute to lineage differentiation indirectly by transporting a cell to its destiny in vivo where a progenitor cell can receive the proper signal required for differentiation. Recently, EMT and MET processes have been observed in in vitro cell fate conversion events such as somatic reprogramming, pluripotent stem cell (PSC) differentiation and transdifferentiation. These in vitro models provide an opportunity to investigate additional roles of EMT and MET during cell fate conversions. One common theme from these studies is that a cell state with partial mesenchymal character (partial EMT) usually facilitates cell fate conversions. Elevated glycolysis is essential for this cell state by providing ATP, substrates for multiple biosynthesis pathways and metabolites that regulate global epigenetic modifications. Epigenetic changes, EMT and metabolic reprogramming form an intermingled regulatory network to regulate cell fate changes. In addition, classic EMT regulators, such as TGFβ and EMT transcription factors, also have EMT-independent functions in cell fate regulation. These mechanistic insights from in vitro models might provide opportunities to understand further the multiple roles of EMT/MET in development or related diseases

We apologize for not being able to include all relevant work that has contributed to our current understanding of EMT in this Primer. The authors thank the editorial team and the anonymous referees for their input on the manuscript during the peer-review process.