During evolution from primitive species, a mechanism, termed epithelial-mesenchymal transition (EMT), was established to create mesenchymal cells from a pre-existing epithelial cell layer. However, the direct and full conversion from an epithelial to a mesenchymal state is not observed in all species (Yin et al., 2009). Intermediate-staged cells harbor the greatest plasticity because they are able to either reverse to an epithelial state or progress to a mesenchymal phenotype. Epithelial cells are characterized by well-developed, intercellular contacts, whereas mesenchymal cells seldom form intercellular contacts. Therefore, to better characterize the hallmarks of intermediate-stage cells, it is important to understand how intercellular contacts change during early EMT. This Cell Science at a Glance article focuses on adhesion and polarity, which are restructured during EMT in vitro. The molecular anatomy of junctional complexes and the dynamics of some of their components during EMT will be described in intermediate stages that lead to a fully converted mesenchymal state.
Molecular anatomy of junctional complexes
A prototypic monolayered epithelium is characterized by apical, lateral and basal plasma membrane domains. The lateral domain contains morphologically distinct, well-demarcated intercellular adhesive structures, including tight junctions (TJs), adherens junctions (AJs) and desmosomes, that contribute to the establishment of apical-basal polarity (Nelson, 2009). Below, we describe the dynamics involved in the formation of these junctional complexes during the establishment of apical-basal polarity.
Nectins comprise a family of cell adhesion proteins that contain three immunoglobulin (Ig) domains. Nectins are involved in nascent contact between epithelial cells (Takai and Nakanishi, 2003) through Ca2+-independent heterophilic binding (e.g. nectin-1 binds to nectin-3), which forms the scaffold upon which E-cadherin-based adhesive structures assemble and recruit other adherens components. Through their PDZ-binding domain, nectins associate with afadin, a large cytoplasmic protein that binds directly to actin microfilaments (Takai and Nakanishi, 2003). E-cadherin, a prototypic, Ca2+-dependent, adhesion molecule that contains five Ig domains, subsequently accumulates at these nectin scaffolds and forms trans-homophilic interactions through its N-terminus, and cis-interactions between the first and second Ig domains (Harrison et al., 2011). The armadillo repeat-containing proteins β-catenin and p120-catenin interact with the cytoplasmic domain of E-cadherin through their C-terminal and juxtamembrane segments, respectively. β-catenin binds to α-catenin, which, in turn, binds to other proteins ensuring the connection to actin microfilaments (Nishimura and Takeichi, 2009). In conjunction with afadin, proteins that connect to actin microfilaments enforce nascent junctional stability. However, such a stabilizing effect is still debated, because actin polymerization is inhibited by monomeric α-catenin (the form that binds cytoplasmic β-catenin), but not dimeric α-catenin (Drees et al., 2005; Yamada et al., 2005).
These nascent cadherin-based adhesions, often called ‘punctates’, progressively mature to form AJs (Adams et al., 1996). This first requires the active assembly of actin filaments within cadherin-rich punctates, which involves nucleation of actin polymerization by the actin-like protein complex ARP2/3 and actinin-4 (Tang and Brieher, 2012). Mature AJs provide a scaffold for the assembly of a circumferential acto-myosin contractile belt at the apex of epithelial cells to ensure stable interactions with neighboring cells that reside as epithelial sheets (Shapiro and Weis, 2009). EPLIN (also known as LIMA1), an actin-filament-bundling protein, establishes such connections in fully developed adherens junctions, forming the zonula adherens at the apex of the lateral domain of polarized epithelial cells (Taguchi et al., 2011). Therefore, actin dynamics have a profound effect on AJ stability.
AJ formation and stability are controlled by several protein complexes, in particular GTPases and their corresponding G-protein exchange factors (GEFs). For example, TARA (also known as TRIOBP), an actin-binding adaptor protein, activates the GEF triple functional domain (TRIO), which in turn promotes actin polymerization through Rac activation (Yano et al., 2011). The Rac, CDC42 GTPases and ARP2/3 also induce localized actin polymerization to initiate nascent junctions in the vicinity of lamellipodia by forming complexes with the Abi-WAVE complex [the Abelson interactor proteins 1 and 2 in complex with a member of the Wislott-Aldrich syndrome protein (WASP) family]. Abi also binds Diaphanous, a member of the formin family, which competes for the same binding site as WAVE. This Abi-Diaphanous complex polymerizes linear actin filaments and stabilizes junctions. Both the Abi-Diaphanous and Abi-WAVE complexes form in the vicinity of E-cadherin–β-catenin complexes (Ryu et al., 2009). SHROOM3, a PDZ domain-containing adaptor protein, binds and activates Rho-associated protein kinase (ROCK), which stimulates apical constriction of the actin belt through phosphorylation of the myosin light chain (Nishimura and Takeichi, 2008). AJ assembly also requires actomyosin contractility (Miyake et al., 2006), with Myosin IIA having a role in punctate E-cadherin clustering into AJs, and Myosin IIB providing stability for the actin belt (Smutny et al., 2011). During epithelial assembly, epithelial cells must sense and transduce mechanical forces. These tensile forces that are exerted on cadherin complexes result in the unfolding of α-catenin to reveal cryptic vinculin-binding sites, which nucleate polymerization of new actin microfilaments (Gomez et al., 2011; Yonemura et al., 2010).
Desmosomes are well-described adhesive junctions that are characteristic to epithelial cells. The cytoplasmic domains of desmocolins (DC1 and DC2) and desmogleins (DG1, DG2 and DG3) bind to plakoglobin, plakophilins (PKP1, PKP2 and PKP3) and desmoplakins (DP1 and DP2) (Franke, 2009; Thomason et al., 2010). Desmoplakins form electron-dense plaques that firmly connect to the thick network of intermediate cytokeratin filaments (Thomason et al., 2010). Many desmosomal proteins (plakoglobin, PKP2 and desmogleins) associate with AJ proteins, such as E-cadherin (Hinck et al., 1994; Lewis et al., 1997), β-catenin (Chen et al., 2002) and p120-catenin (Kanno et al., 2008) to mutually strengthen cadherin- and desmosome-based cell adhesions. Like AJs, desmosomes initially form as immature Ca2+-dependent structures (Thomason et al., 2010), and mature into hyper-adhesive desmosomes. Evidence suggests that AJs and desmosomes cooperate during junctional assembly; this strengthening and maturation ultimately contributes to epithelial cell integrity (Huen et al., 2002; Vasioukhin et al., 2001).
TJ assembly commences after the formation of AJs and desmosomes. TJs prevent the passage of fluid between the luminal compartment and the underlying stroma. Despite the identification of more than 40 proteins involved in TJs, their precise molecular composition remains unclear (Anderson and Van Itallie, 2009). Notably, the paracellular transport of ions or small metabolites occurs through TJs via anastomosing fibrils, designated ‘TJ strands’, which initiate from homo- and heterotypic interactions between the tetraspan proteins claudins (Furuse, 2010); for instance, claudin-3 binds to claudin-1 and claudin-2 (Furuse and Moriwaki, 2009). The cytoplasmic domains of claudins and occludins associate with the zonula occludens proteins [ZO1, ZO2 and ZO3 (also known as TJP1, TJP2 and TJP3, respectively)], which contain three PDZ domains. Occludins and other tetraspan proteins, such as tricecullin, are recruited for TJ maturation following claudin activity. ZO proteins interact with other PDZ-containing proteins, such as multi-PDZ-domain protein 1 (MUPP1, also known as MPDZ) and PALS1-associated to tight junction protein (PATJ, also known as INADL), and assemble into scaffolds that are connected to actin microfilaments (Furuse, 2010; Furuse and Tsukita, 2006). Also closely associated with TJs are the transmembrane junctional adhesion molecules (JAMs; JAM1, JAM2 and JAM3) that contain two Ig domains (Fukuhara et al., 2002), and the coxsackie virus and adenovirus receptor (CXADR). These proteins interact with ZO1 and afadin. Nectin adhesion complexes also contribute to the recruitment of claudins, occludins and JAMs.
The highly polarized epithelial cells reside above a layer of basement membrane, comprising a complex network of extracellular matrix (ECM). Epithelial cells connect to the basement membrane through integrins that are located on the basal epithelial surface, and bind to peptide motifs found on collagens, laminins and other ECM components (Hay, 1993; LeBleu et al., 2007; Timpl and Brown, 1996). Hemi-desmosomes are specialized adhesive structures that are assembled by clustering of the α6β4 integrin (Margadant et al., 2008). The cytoplasmic domain of the β4 integrin subunit subsequently recruits an array of proteins, including plectin, which is connected to cytokeratin filaments (Margadant et al., 2008). Other integrin heterodimers cluster into focal complexes, focal adhesions or fibrillar adhesions that are linked to the actin microfilament network. This so-called ‘integrin adhesome’ also includes numerous kinases, phosphatases and adaptor proteins (Geiger and Yamada, 2011) and senses mechanical forces through Talin and p130CAS (also known as BCAR1) (del Rio et al., 2009; Sawada et al., 2006).
Apical-basal polarity is progressively established during junctional maturation. In vitro studies indicate that the basal side is first determined when cells attach and spread to the substratum. As the adhesion junctional complexes mature, the lateral cell domain is increasingly segregated from the apical domain, with mature TJs defining the boundary between the apical and lateral domains. Apical-basal cell polarity is controlled by three evolutionarily conserved protein complexes in Drosophila: (1) the Crumbs complex, comprising Crumbs3, PALS1 (known as Stardust) and Patj; (2) the PAR complex, comprising PAR6, PAR3 and atypical protein kinase C (aPKC); and (3) the Scribble complex, consisting of Scribble, Lethal giant larvae (Lgl) and Discs large (Dlg) (St Johnston and Ahringer, 2010). In immature cell-cell contacts, PAR6 is bound to aPKC, independent of PAR3, which is found in nascent AJs associated with afadin (Martin-Belmonte and Perez-Moreno, 2012). During maturation, PAR3 relocates into the Crumbs complex and is recruited to TJs (Nelson, 2009; St Johnston and Ahringer, 2010).
The dynamic changes of junctional complexes during EMT
EMT triggers the sequential destabilization of cell-cell adhesive junctions and the regulatory machinery that controls cell polarity. This transforms epithelial cells into a plastic and motile state, with a mesenchymal phenotype (Thiery and Sleeman, 2006). The cartoon in the accompanying poster depicts two bona fide epithelial cells that become progressively separated mesenchymal-like cells through the gradual loosening of cell-cell contacts, which is comparable to ‘unzipping’. Separation starts with the sequential loss of TJ, AJ and desmosome integrity, transitioning cells from a polarized epithelium to a post-EMT mesenchymal state. This transition involves intermediate stages, which offers the greatest potential for epithelial plasticity and highlights one of the important hallmarks of EMT, reversibility.
EMT reversibility is closely tied to the dynamic and efficient remodeling of cell adhesion complexes (Yap et al., 2007). Adhesion proteins undergo constant recycling into and out of junctions, which is counter-intuitive, as these junctions are assumed to be stable. For instance, a fraction of the E-cadherin molecules at AJs in highly polarized cells have a half-residence time of only several minutes in the junction (de Beco et al., 2009). As such, E-cadherin is in constant traffic between the cell surface and the recycling endosomes (Bryant and Stow, 2004; de Beco et al., 2009; Hong et al., 2010; Le et al., 1999; Troyanovsky et al., 2006). However, in cells with an intermediate phenotype, such as A431 squamous carcinoma cells and E-cadherin-transfected Chinese hamster ovary (CHO) cells, E-cadherin turnover is not only faster, but also appears to be governed by non-clathrin-mediated endocytosis (Hong et al., 2010). TJs are also constantly remodeled whilst maintaining a barrier between epithelial cells (Shen et al., 2008; Steed et al., 2010). Thus, EMT is likely to be dependent on these constitutively established dynamic changes at cell junctions.
Early intermediate state of EMT
The absence of well-developed TJs in in vitro cell lines might be responsible for the paucity of research emphasizing the role of TJs in EMT (Thiery and Huang, 2005). The central role of TJ stability in EMT regulation was discovered serendipitously in a luminescence-based interactome mapping for transforming growth factor-beta (TGFβ) signaling (Barrios-Rodiles et al., 2005; Ozdamar et al., 2005). Occludin was found to bind directly to TGFβ type I receptor (TGFBR1). TGFBR1 forms a complex with PAR6 and occludin at TJs independently from its ligand. Upon stimulation with TGFβ, TGFBR2 is recruited to the TJ complex and phosphorylates PAR6, which binds to the E3-ubiquitin ligase, SMURF1. Smurf1 then ubiquitinates RhoA leading to its degradation, which interferes with the cortical actin cytoskeleton and ultimately contributes to TJ disassembly (Ozdamar et al., 2005). In mammary epithelial cells, activated EGFR-related protein 2 (ERBB2) directly interacts with the PAR6-aPKC complex and disrupts the apical polarity of multi-acinar structures. This ERBB2-PAR6-aPKC association promotes PAR3 dissociation from the PAR6-aPKC complex (Aranda et al., 2006). Therefore, the breakdown of TJs and disruption of the PAR complex owing to the dissociation of PAR3 are the key events in the early intermediate state of EMT.
Late intermediate state of EMT
The late intermediate state of EMT is hallmarked by AJ destabilization. First, the E-cadherin complex is destabilized by post-translational modifications, such as increased phosphorylation, its internalization and degradation of E-cadherin and β-catenin (Bryant and Stow, 2004; D'Souza-Schorey, 2005). Evidence for this comes from growth factor or receptor tyrosine kinase (RTK)-mediated induction of EMT; examples include hepatocyte growth factor (HGF) and its receptor Met, epidermal growth factor (EGF) and EGFR, and fibroblast growth factor (FGF) and FGFR (Thiery, 2002). In HGF-induced EMT, Met co-localizes with E-cadherin and they are both co-endocytosed (Kamei et al., 1999) through the upregulation of endogenous ADP-ribosylation factor 6 (ARF6) GTPase (Palacios et al., 2001). This leads to the phosphorylation of E-cadherin on two tyrosine residues and its subsequent ubiquitination by the E3-ubiquitin ligase HAKAI (also known as CBLL1) (Fujita et al., 2002; Mukherjee et al., 2012). EGF also induces EMT in non-transformed cells, such as normal ovarian surface epithelium (Salamanca et al., 2004) and in various human carcinoma cells that overexpress EGFR (Barr et al., 2008). During the early stages of EGF-induced EMT, a rapid increase in tyrosine phosphorylation of the E-cadherin–β-catenin complex disrupts cell-cell adhesion and causes cell dissociation (Fujii et al., 1996; Shibamoto et al., 1994), possibly through caveolin-1-dependent endocytosis of E-cadherin (Lu et al., 2003). Aside from RTKs, Src tyrosine kinase also has a crucial role during this stage of EMT (Rodier et al., 1995). Tyrosine phosphorylation of its substrates β-catenin and p120-catenin decreases their binding affinity to E-cadherin. In lysophosphatidic acid (LPA)-induced ovarian cancer cell dispersal, the Src-related tyrosine kinase, Fyn, phosphorylates β-catenin and p120-catenin, thus, destabilizing the β-catenin–p120-catenin–α-catenin complex and resulting in the breakdown of cell adhesions (Huang et al., 2008). However, Src can also induce E-cadherin endocytosis-mediated AJ disassembly by tyrosine phosphorylation-independent mechanisms. In MDCK cells, Src induces the depletion of T-cell lymphoma and invasion metastasis 1 (TIAM1), a Rac-specific GEF, through TIAM1 phosphorylation and proteasomal degradation (Palovuori et al., 2003; Woodcock et al., 2009). Interestingly, E-cadherin-mediated cell adhesion might also inhibit ligand-dependent RTK activation, such as that of EGFR (Perrais et al., 2007; Qian et al., 2004), which raises the question as to how cells that express E-cadherin at AJs trigger EMT through growth factor-mediated stimulation. One plausible explanation is that cells that respond to RTK signals might have entered an early intermediate state through a β-catenin-dependent mechanism.
Unlike in AJs, desmosome stability is affected by serine phosphorylation by members of the protein kinase C (PKC) family (Volberg et al., 1992). Plakophilin 2, which regulates intercellular junction assembly, is a crucial scaffold for PKCα (Bass-Zubek et al., 2008). PKCα signaling also regulates the dynamics of desmoplakin through the Ca2+ ATPase SERCA2 (also known as ATP2A2), which stabilizes desmosomes (Hobbs et al., 2011). However, plakoglobin, which is structurally related to β-catenin, can be phosphorylated at both tyrosines and serines (Gaudry et al., 2001; Shibamoto et al., 1994). Desmosome disassembly often accompanies AJ disruption during late intermediate EMT, when TJs are also weakened. This suggests an interdependence between desmosomes and AJs, which is not surprising given the fact that E-cadherin, plakophilin and desmoplakin genes are direct targets of transcriptional repressors, such as Zinc finger E-box-binding homeobox 1 (ZEB1), which is a key driver of EMT (Vandewalle et al., 2005). Initial transcriptional changes, such as upregulation of snail homologue 1 and 2 (SNAI1 and SNAI2), frequently occur at this late intermediate state, and their activation is often transient and results in a partial EMT (Carrozzino et al., 2005; Savagner et al., 1997). In an inducible system, the incomplete downregulation of E-cadherin by SNAI1 in MDCK cells results in a partial EMT effect on TJs (Carrozzino et al., 2005). However, when expression of SNAI1 and SNAI2 is sustained, a full EMT phenotype can be initiated. SNAI1-overexpressing mouse EpH4 epithelial cells fully execute EMT by downregulating occludins and claudins-3, -4 and -7 (Ikenouchi et al., 2003), followed by the post-transcriptional loss of ZO1 (Ohkubo and Ozawa, 2004).
Transcriptional regulation of SNAI1 and SNAI2 at this stage also contributes to the effect that EMT has on other polarity complex proteins, such as Crumbs3. In MDCK cells, expression of TGFβ upregulates SNAI1 transcripts at around 48 hours, which is followed by downregulation of Crumbs3 transcription after 96 hours, concomitant with the loss of E-cadherin (Whiteman et al., 2008).
Full mesenchymal state of EMT
The full mesenchymal state of EMT is achieved through the transcriptional repression of desmosomal, AJ and TJ proteins (Moreno-Bueno et al., 2008). Interestingly, β-catenin, p120-catenin and plakophilin 2 have nuclear localization signals (NLS) and can shuttle between the nucleus, cytoplasm and cell membrane (Fagotto et al., 1998; Mertens et al., 2001; Roczniak-Ferguson and Reynolds, 2003). Once displaced from junctional complexes, a fraction of these proteins is targeted to the nucleus where they interact with transcription factors to activate their downstream target genes, including EMT drivers (Gottardi and Gumbiner, 2001). Nuclear β-catenin has been identified in single migratory cells at the colon carcinoma invasive front, which has undergone full EMT (Brabletz et al., 2005a; Brabletz et al., 2005b). It has been suggested that nuclear β-catenin coactivates Wnt signaling with TCF4 and lymphoid enhancer-binding factor 1 (LEF1), and that formation of the β-catenin–TCF4 complex directly executes full EMT through induction of the E-cadherin transcriptional repressor ZEB1 (Sánchez-Tilló et al., 2011). However, thus far, there is no direct evidence for the involvement of nuclear p120-catenin or plakophilin 2 in the execution of EMT.
Conclusions and future perspectives
The conversion of polarized epithelia into mesenchymal-like cells follows an ordered sequence. Initially, EMT targets the PAR polarity complex and the closely associated TJs. During the intermediate states, AJs and desmosomes weaken and fragment through post-translational modifications and transcriptional repression. Concomitant with these ‘loss-of-function’ events, many ‘gain-of-function’ events occur, where vimentin-containing filaments replace intermediate cytokeratin filaments in mesenchymal-like cells to induce motile and invasive properties that potentially favor local invasion and metastasis (Thiery, 2002). This plasticity is closely linked to the junctional dynamics with which cell-cell adhesions are formed and remodeled.
Epithelial and mesenchymal phenotypes govern the sensitivity of targeted therapies or chemotherapies in human cancers. Thus, it might be possible to exploit reversibility of EMT for therapeutic purposes (Arumugam et al., 2009; Buck et al., 2007; Thomson et al., 2005; Yauch et al., 2005). Drug resistance tends to either result from or induce carcinoma clones that have undergone EMT (Kajiyama et al., 2007; Yang et al., 2006). Therefore, reversing EMT could be a promising strategy to increase carcinoma sensitivity to drugs. For example, tyrosine kinase inhibitors have been implicated in EMT reversal in cancers (Choi et al., 2010; Lorch et al., 2004; Nagai et al., 2011). New genes that mediate EMT are expected to be uncovered in the near future. Therefore, elucidation of their translational modifications and integration into specific gene regulatory networks will contribute towards the development of EMT-based therapeutic interventions (Chua et al., 2011). The EMT reversal process is reminiscent of nascent junctional maturation; consequently, further exploration of the factors involved is essential to shed light on the molecular basis of EMT. It will be very exciting to see whether these factors constitute valuable targets for therapeutic EMT reversal.
This work was supported by the Agency for Science, Technology and Research (A*STAR) Biomedical Research Council, Singapore; and the Health Research Council of New Zealand International Investment Opportunities Fund (IIOF) [grant number 1010124640].
A high-resolution version of the poster is available for downloading in the online version of this article at jcs.biologists.org.
- © 2012. Published by The Company of Biologists Ltd