Reprogramming during epithelial to mesenchymal transition under the control of TGFβ

Abstract

Epithelial-mesenchymal transition (EMT) refers to plastic changes in epithelial tissue architecture. Breast cancer stromal cells provide secreted molecules, such as transforming growth factor β (TGFβ), that promote EMT on tumor cells to facilitate breast cancer cell invasion, stemness and metastasis. TGFβ signaling is considered to be abnormal in the context of cancer development; however, TGFβ acting on breast cancer EMT resembles physiological signaling during embryonic development, when EMT generates or patterns new tissues. Interestingly, while EMT promotes metastatic fate, successful metastatic colonization seems to require the inverse process of mesenchymal-epithelial transition (MET). EMT and MET are interconnected in a time-dependent and tissue context-dependent manner and are coordinated by TGFβ, other extracellular proteins, intracellular signaling cascades, non-coding RNAs and chromatin-based molecular alterations. Research on breast cancer EMT/MET aims at delivering biomolecules that can be used diagnostically in cancer pathology and possibly provide ideas for how to improve breast cancer therapy.

Keywords:

• epithelial-mesenchymal transition
• signal transduction
• transforming growth factor β
• tumor invasiveness

Abbreviations

bHLH	=	basic helix-loop-helix
BMP	=	bone morphogenetic protein
CSC	=	cancer stem cell
DNMT	=	DNA methyltransferase
EMT	=	epithelial-mesenchymal transition
FGF	=	fibroblast growth factor
HDAC	=	histone deacetylase
lncRNA	=	long non-coding RNA
mTORC	=	mammalian target of rapamycin complex
MET	=	mesenchymal-epithelial transition
miRNA	=	micro-RNA
MAPK	=	mitogen activated protein kinase
PDGF	=	platelet derived growth factor
PRC	=	polycomb repressive complex
TF	=	transcription factor
TGFβ	=	transforming growth factor β

Introduction

Cancer is a complex disease with a long evolutionary trajectory stemming from the introduction of genetic mutations in normal cells to the biochemical alterations of chromatin, signaling pathways and cell biological processes. The final manifestation of this disease involves malignant alterations in tumor cell differentiation, migration and miscommunication with the immune and other systems that survey the homeostasis of the organism. The epithelial-mesenchymal transition (EMT) is a biological process that takes place during organismal development, with prominent examples being the formation of the primitive streak, the neural crest delamination, heart valve differentiation and lung organogenesis.¹ EMT has also been described in the pathology of fibrotic disorders and cancer, as a mechanism that links the effects of chronic inflammation to the wounding of tissues, resulting in epithelial cell dedifferentiation, deposition of excessive extracellular matrix and acquisition of motile properties by cells that otherwise are destined to be relatively immotile, as a consequence of gradual loss of proper tissue architecture.^2,3 Epithelial cells alter their apico-basal polarity, remodel their intermediate filaments and microfilaments and intercellular adhesions, secrete new matrix proteins and glycosaminoglycans, and acquire a mesenchymal state that correlates to enhanced migratory capacity.¹ In addition, in the context of cancer, EMT reprograms epithelial cells and generates cancer stem cells (CSCs) that exhibit resistance to chemical or radiological therapy or senescence induced by oncogenic stimuli, and provide necessary attributes that promote tumor cell metastasis.^4-6 An important feature of cells undergoing EMT during embryonic development or cancer progression is the reversibility of the process (mesenchymal-epithelial transition, MET), which generates new epithelial cells that build new embryonic tissues or metastatic colonies of secondary tumor growth. Because of the prominence of cell biological and molecular studies on the process of EMT in breast cancer, the present article focuses on this type of malignancy; however, many of the principles that govern EMT appear to be universal and independent from tissue type.

Growth and Transcriptional Factors Governing the EMT Program

One of the reasons behind the universal principles that guide EMT processes is the dependency of these processes on a cohort of secreted growth factors, cytokines and intracellular transcription factors that execute the signals provided by the growth factors. The list of the extracellular guiding cues that initiate and maintain the EMT process over time is large, and includes environmental factors such as smoking and ultraviolet radiation, tissue-specific pathological conditions such as hypoxia and associated cytokines including the transforming growth factor β (TGFβ), fibroblast growth factor (FGF), Notch, insulin-like growth factor, Wnt, interleukin-like EMT inducer and more.¹ This example nicely illustrates the principle of inter-dependent cytokine requirement for EMT to get established in certain cases of breast cancer that exhibit estrogen receptor deregulation: the reduced expression of estrogen receptors in breast cancer and the development of hypoxia lead to induced secretion and extracellular activation of TGFβ, which starts a cascade of signaling events that include the pathways of Notch, interleukin-like EMT inducer and insulin-like growth factor, which cause secretion and activation of the vascular endothelial growth factor, catalyzing neo-angiogenesis that fosters sustained EMT and subsequent tumor cell extravasation and metastasis.⁷

The EMT program can be triggered by a specific group of transcription factors in response to the above signaling pathways (e.g., TGFβ, FGF, Notch, etc.) during development or cancer progression. These EMT transcription factors (EMT-TFs) coordinately regulate the repression of epithelial genes and upregulation of mesenchymal genes, thus promoting migratory and invasive cellular behavior. The well-studied classical EMT-TFs are the Snail, ZEB and Twist family members, which converge to repress E-cadherin – a key feature of EMT.^8,9 Snail1 (Snail) and Snail2 (Slug) are zinc finger proteins^10-12; ZEB1 (δEF1) and ZEB2 (SIP1) are zinc finger and homeobox domain proteins;^13,14 and Twist1 and Twist2 are basic helix-loop-helix proteins (bHLH).^15,16 Additional bHLH family members, E47 and E2-2, are also known to induce EMT.¹⁷

The embryonic chromatin protein high mobility group A2 (HMGA2) integrates EMT signals downstream of TGFβ by induction of Snail1, Snail2, Twist1 and repression of inhibitor of differentiation 2 (ID2).^18-20 Another transcriptional regulator induced by TGFβ is Sox4, a transcription factor of the SRY-related HMG-box family and is also involved in embryonic development.^21,22 Sox4 regulates the expression of the classical EMT-TFs in mammary epithelial cells during EMT. The EMT-TFs also regulate the expression of each other in a feed-forward loop,^18,23,24 establishing the EMT nuclear reprogramming that is reminiscent of the embryonic stem cell fate regulatory circuitry controlled by master transcription factors, Oct4, Nanog and Sox2.²⁵

The complexity of signaling crosstalk during EMT can be presented by studying the TGFβ and Wnt pathways. To present deeper analysis of the crosstalk in this section, we diverge slightly and cite EMT cases from organs other than the breast. Components of the Wnt pathway, namely, the adaptor protein axin, the protein kinase glycogen synthase kinase 3β (GSK3β) and β-catenin interact and modulate the activity of Smad proteins of the TGFβ pathway in the cytoplasm or in the nucleus. Cytoplasmic axin associates with Smad3 and facilitates Smad3 phosphorylation and overall efficiency of downstream transcriptional responses to TGFβ.²⁶ On the other hand, cytoplasmic axin brings GSK3β to Smad3, which phosphorylates Smad3 and induces its proteasomal degradation thus keeping the steady state pool of Smad3 under control.²⁷ This mechanism leads to general control of Smad3 signaling by axin/GSK3β. In the mammary EpH4 cell model of EMT, oncogenic activation (e.g., by an inducible c-Fos-estrogen receptor transgene) caused activation of the nuclear β-catenin/T-cell-specific transcription (TCF) 4 complex together with the autocrine activation of TGFβ and downstream Smad nuclear complexes that mediate mesenchymal gene expression.²⁸ During endothelial-mesenchymal transition (EndoMT) that forms the valves in the heart in response to TGFβ2, EMT induced by TGFβ3 in the palate, or EMT developed during lung fibrosis, β-catenin/TCF complexes are critical, together with Smads for the induction of mesenchymal genes in the transforming endothelial cells or for the repression of E-cadherin in the regressing epithelial cells of the palate.^29-31 These examples of EndoMT will be discussed further later. In the case of lung EMT after inflammation, a specific protein complex between Smad2 and phosphorylated β-catenin on tyrosine 654 that is mediated by activation of integrin-α3β1 is a crucial determinant that controls E-cadherin repression.²⁹ In a similar manner, Smad3 can facilitate nuclear accumulation of β-catenin during proliferation and maintenance of adult mesenchymal stem cells.³² The TGFβ-Wnt signaling network does not only contribute to the generation of transcriptional complexes between Smads and β-catenin, but also contributes to the regulation of Snail stability and transcriptional activity since Snail is a direct target of phosphorylation by the GSK3β kinase.³³ The result of such molecular interactions between TGFβ and Wnt signaling components is evident in numerous cases of epithelial tumorigenesis. In mouse models of breast and intestinal cancer that are driven by Wnt ligand overactivation, a high degree of synergism between Wnt and TGFβ signaling was shown to contribute to the progression of tumorigenesis.³⁴ In breast cancer models, the non-canonical Wnt5a, among all Wnt family ligands, appears to play critical roles in cooperation with or acting upstream of TGFβ signaling. TGFβ signaling controls Wnt5a expression in the mammary gland and Wnt5a signaling depends on TGFβ receptor function.³⁵ According to this mechanism, TGFβ and non-canonical Wnt5a act as tumor suppressors and when these 2 factors are lost, breast cancers gain stable canonical Wnt/β-catenin nuclear activity that promotes tumor growth.³⁶ However, it should be mentioned that Wnt5a is required for tumor cell invasion and CSC properties in basal-like breast cancer.³⁷ In the prostate, upon androgen reduction, TGFβ gets elevated and the resulting hyperproliferation of prostate epithelial cells often leads to tumorigenesis, a mechanism that depends on expression and signaling by multiple Wnt family ligands.³⁸ In vitro EMT caused by the cooperation between Ras and TGFβ signaling leads to the switch from cadherin-dependent adhesion to integrin receptor-dependent adhesion and this process coordinates with the secretion of Wnt5a and the activity of the non-canonical Wnt pathway of planar cell polarity. Finally, the link between EMT and cancer cell stemness in breast cancer cell models has also been shown to be mediated by the coordinate activity of TGFβ, secreted non-canonical Wnt members and inhibitors of bone morphogenetic protein (BMP) signaling that act in an autocrine manner to promote cancer stemness.³⁹ These and numerous other experimental examples enlighten our understanding of the molecular networks that control the EMT process in the breast and many other organs.⁹

General Aspects of TGFβ-Induced EMT

The links between TGFβ signaling and EMT were developed since the early days of the molecular dissection of this signal transduction pathway and have been firmly established in the field through a large set of solid publications.^40,41 Early studies of TGFβ signaling in mouse mammary epithelial cells, a) Namru murine mammary gland, NMuMG cells, which are non-tumorigenic, showed time-dependent and reversible EMT in response to TGFβ^42-45; b) in EpRas cells, which are mouse mammary EpH4 cells transformed by the H-Ras oncogene, showed a more stable EMT that is dependent on autocrine TGFβ and also developed tumors in mouse xenografts.⁴⁶ TGFβ could elicit all the hallmarks of EMT such as loss of cell-cell contacts, including adherens and tight junction disassembly, cytoskeletal adaptations including focal adhesion-bound actin stress fibers, intense perinuclear vimentin filaments accompanied by gradual loss of cytokeratin filaments, spindle-like morphology and enhanced motility.⁴³ The complex cell biology of the EMT process can be relatively well understood and classified when viewed as a program of cell differentiation emanating under the guidance of a single growth factor like TGFβ. According to this notion, the EMT program can be divided to specific subprograms (Fig. 1) that include the above-mentioned junctional complex and cytoskeletal programs, but also 2 additional subprograms, the secreted factor and extracellular matrix programs.⁴¹ In brief, TGFβ induces the synthesis and secretion of several other growth factors, cytokines and chemokines and many of these collectively contribute to the establishment of EMT, tumor cell invasiveness and metastasis. Some of these growth factors optimally operate in a remodeled extracellular matrix that is composed by newly synthesized and deposited proteins or glycosaminoglycans, that mediate new cell-cell interactions, paracrine cell activation and enhanced cell migration.⁴⁷

Figure 1. Mechanisms in TGFβ-induced EMT and BMP-induced MET. A mammary epithelial cell is shown on top with the 2 sister signaling pathways of TGFβ and BMP. These 2 growth factors are color coded to match the process of EMT and MET respectively. Signaling by the dimeric extracellular proteins is mediated by plasma membrane receptors (type II (RII) and type I (RI), which phosphorylate Smad proteins or activate other non-Smad signaling proteins (mainly protein kinases shown in gray boxes). Phosphorylation events on the type I receptor or on the Smads are shown with a black circled white P. Smads form complexes with Smad4 and together with non-Smad kinases transmit signals into the nucleus where different target genes are regulated positively (black arrows) or negatively (red arrows). Target genes of the pathways are symbolized as circles on the chromatin thread and are divided for simplicity and based on the text in 4 groups: a) miRNA and lncRNA genes (pink); b) co-transcription factor genes (orange), e.g., HMGA2, C/EBP or IDs in the case of BMP signaling; c) EMT-TF genes (purple); d) effector genes including secreted polypeptides, junctional, extracellular matrix and cytoskeletal components (yellow). MiRNAs are shown to negatively regulate various other target mRNAs. LncRNAs are shown to participate with chromatin regulators (black box on the left). These chromatin regulators participate in the chromatin-based epigenetic control of every step in the signaling and gene regulatory cascade (not shown). Make note of the sequential regulatory cascades whereby for example, the effector (yellow) gene is regulated by incoming Smad and non-Smad signaling, co-TFs and EMT-TFs, together with the conserted action of the lncRNA/chromatin factors (not shown). Such signaling mechanisms lead to EMT or MET leading to changes of cell architecture and differentiation as depicted at the bottom of the figure. Platelets and other cell types assist to the process of EMT, whereas the contribution of accessory cell types that promote MET is not well investigated (question mark). The generation of cancer stem cells is indicated with dark pink color and the metastatic breast carcinoma epithelial cells are colored as dark pink to indicate their development via MET from the CSCs. The latter is a speculative hypothesis and not yet demonstrated experimentally. Note the lack of normal basement membrane in the metastatic epithelial colony, which is also hypothetical and aims at emphasizing that the metastatic carcinoma cells and the epithelial cells in the primary cancer may not generate identically organized tissue architecture.

In this section we will diverge again from discussing strictly breast-related EMT. While the majority of experimental studies of EMT focus on the prototype isoform TGFβ1, the first demonstration of a link between TGFβ and EMT came from studies of EndoMT that generates the cushions in the heart valves, a process operating under the guidance of the TGFβ2 isoform.⁴⁸ Thus, TGFβ2 initiates the dissociation of adherent endothelial cells, and induces expression of TGFβ3, which promotes myoepithelial cell invasion into the developing tissue in chicken embryos. Species-specific differences are known to exist with respect to the necessity and the developmental timing during which these different TGFβ isoforms become activated and functional.⁴⁸ Accordingly, mouse TGFβ2 appears to be self-sufficient and does not require TGFβ3. Careful analysis of all 3 TGFβ isoforms using the respective knockout mice, showed that normal heart valve EndoMT can occur in the complete absence of TGFβ1 or TGFβ3, whereas knockout of TGFβ2 led to a defective EMT in assays of explanted embryonic heart tissue, which corroborates the in vivo phenotype of the TGFβ2 knockout mice that exhibit enlarged heart cushions.⁴⁹ Downstream of TGFβ2 lies Snail1 at least in mouse endothelial cells, which contributes to the upregulation of many mesenchymal genes, including α-smooth muscle actin (α-SMA), SM22α and calponin.⁵⁰ In mice, the TGFβ2 signal also cooperates with the Wnt, β-catenin/TCF4 transcriptional complex to mediate the EndoMT and in vitro experiments using endothelial cells showed that α-SMA expression is regulated by crosstalk between TGFβ2 and β-catenin/TCF4,³⁰ as mentioned above. While TGFβ2 seems to initiate the EndoMT in the heart valves, invasiveness depends on BMP2 signaling derived from myocardial cells and Snail1 activity that is stabilized after BMP2 exposure, both events lying under the control of Notch1 signaling, which operates in the adjacent endocardial tissue.⁵¹ These 2 cytokines, TGFβ2 and BMP2, seem to contribute to EndoMT and invasiveness by binding to the type III co-receptor β-glycan, which coordinates signaling by both TGFβ-specific type I receptors, activin receptor-like 5 (ALK5), and BMP-specific type I receptors, ALK2 and ALK3.⁵² Upstream of the action of TGFβ2 lies the MAP kinase kinase MEKK3, whose signaling in the developing heart of mice regulates the transcriptional induction of TGFβ2.⁵³ Thus, heart valve morphogenesis is a good example of TGFβ2-dependent EMT.

Another tissue that shows isoform-specific dependence for its EMT is the palatal epithelium that undergoes EMT in order for the 2 sides of the epithelium to fuse and generate the intermediary seam. TGFβ3 is the key player in this process, which causes the EMT by repressing E-cadherin via a nuclear Smad-TCF4 transcriptional complex on the E-cadherin promoter region as explained above.³¹ Not only does TGFβ3 promote the formation of the Smad-β-catenin-TCF4 complexes, but also induces the levels of TCF4 at the transcriptional level.⁵⁴ In addition to Smads, non-Smad signaling contributes to the stabilization and activation of Snail1 and ZEB2 during palatal EMT, and additionally, the small GTPase Rho and its downstream Rho kinase contribute to mesenchymal phenotypic changes in the palate, including actin cytoskeleton remodeling.^55,56

Some additional examples of EMT can be listed based on their sensitivity to different TGFβ isoforms. Similar to the heart, pulmonary valve morphogenesis requires EndoMT, which is driven by TGFβ2 and is antagonized by the pro-angiogenic factor VEGF-A.⁵⁷ Retinal epithelial cells undergo EMT at least in vitro and in vivo during fibrosis that leads to retinopathies, and this EMT is driven by TGFβ2, which then induces expression of the Notch ligand Jagged1, that synergistically with TGFβ2-Smad and non-Smad signaling promotes the expression of EMT-TFs that mediate the mesenchymal transition.⁵⁸ In the context of cancer-related EMT, all 3 TGFβ isoforms can induce hepatocarcinoma EMT that is mediated by downstream induction and signaling of the platelet derived growth factor (PDGF) A.⁵⁹ In EMT and metastatic spreading of breast cancer with oncogenic phosphatidylinositol 3’ kinase (PI3K)/Akt pathway activation, TGFβ2 activity is required; Akt directly phosphorylates Twist1, causing its stabilization and transcriptional activation, and leading to the enhanced expression of the TGFβ2 gene.⁶⁰ Finally, EMT and MET cycles are controlled by the reciprocal loop between TGFβ and members of the miR-200 family.^61,62 In zebrafish, the tyrosine phosphatase PEZ regulates the transcriptional expression of TGFβ3, which then drives several cases of developmental EMT in the heart, pharynx and in somites.⁶¹ In several tumor cells that exhibit EMT, ZEB1 directly represses the expression of miR-200 family members, and these microRNAs (miRNAs) negatively regulate expression of TGFβ2, among many other targets.⁶² Thus, the EMT-TF network facilitates the further expression and activity of specific TGFβ isoforms during cancer progression toward metastasis,⁶² or during tissue fibrosis such as in the kidney.⁶³

It is worth pointing out 2 aspects of TGFβ cell physiology that relate to the process of EMT and often generate conflicting views on the topic: a) TGFβ is known to arrest the cell cycle of normal mammary epithelial cells at the early G1 phase.⁴⁵ Thus, at least in vitro, mammary epithelial cells undergo EMT while their cell cycle is arrested. This is in agreement with the general view that EMT correlates with cell cycle arrest and enhanced survival.⁷ b) However, the pro-EMT action of TGFβ represents a pro-tumorigenic and pro-metastatic aspect of the biology of this growth factor in the context of cancer.⁶⁴ In the sequence of multi-stage events of cancer progression, the pro-tumorigenic TGFβ actions are usually thought to follow the tumor suppressive effects that TGFβ exhibits as a factor that maintains epithelial tissue homeostasis.⁶⁴ In other words, the pro-tumorigenic, pro-EMT effects of TGFβ, at the cellular level correlate to one of the hallmarks of the tumor suppressive effects TGFβ has. This complicated concept is explained by the observation that long-term exposure to TGFβ bypasses the initial growth arrest response of mammary epithelial cells in culture, and proceeds with sustained cell proliferation in the presence of a mesenchymal phenotype.⁴² This adaptive response of breast cells is dependent on multiple signaling inputs in addition to TGFβ.^7,42 In other words, while TGFβ arrests the mammary epithelial cell cycle and initiates EMT, concomitant exposure of cells to other cytokines and growth factors in the context of the tumor microenvironment ensure the progression of carcinogenesis and eventually help so that breast cancer cells maintain mesenchymal features while having bypassed the proliferation arrest that is imposed on their epithelial progenitors by TGFβ.

It is therefore appropriate to briefly discuss the 2 faces of TGFβ biology during cancer progression. The current state of the art proposes that tumor suppressive mechanisms induced by TGFβ in the breast and many other epithelial tissues involve the transcriptional and Smad-mediated upregulation of cell cycle inhibitor genes such as p15 and p21, which block the cell cycle at the early G1 phase.^65,66 In addition, a critical prerequisite of cell cycle inhibitor gene upregulation is the transcriptional repression of c-myc, a condition that is often bypassed in breast cancers with oncogenic myc activation or overexpression.⁶⁵ It is worth emphasizing that epithelial cells where c-myc is overexpressed do not respond anymore to TGFβ in terms of stopping their cell cycle, however, they maintain the EMT response, as c-myc redirects Smad complexes to the Snail1 gene causing the induction of EMT.⁶⁷ Many tissue-specific knockout mouse models have also been generated that clearly demonstrate the tumor suppressor activity of TGFβ in a variety of organs, including the breast.⁶⁴ Some of these models have made clear that the more sensitive cell type to the tumor-derived TGFβ are the resident fibroblasts of the tumor stroma or immune cells, such as T lymphocytes that contribute to enhanced tumor growth and cancer progression when TGFβ receptor or Smad signaling is lost in these cells.⁶⁸

The other face of TGFβ in tumor progression is a pro-tumorigenic that seems to be more universal and usually is linked to every tumor type studied so far. Primary determinants of this activity of TGFβ are: a) the potent immunosuppressive action of TGFβ, which inhibits proliferation and differentiation of B and T lymphocytes, and thus, tumor-derived TGFβ generates a locally immunocompromized microenvironment that is beneficial to the advancement of tumorigenesis⁶⁸; b) the EMT that can be mediated either in an autocrine or paracrine manner, and associated with EMT, the enhanced abundance of cancer stem cells and metastatic stem cells that disseminate more effectively to distant sites of tumor spread⁴¹; c) the pro-angiogenic effects of TGFβ, which are usually indirect and mediated by the transcriptional induction of VEGFs and Notch family ligands that ensure a more robust tumor vasculature.⁶⁴ Thus, such pro-tumorigenic and pro-metastatic actions of TGFβ seem to overtake and last for the lengthier period of tumor progression.

Smad and Non-Smad Signaling in EMT

TGFβ makes a remarkable inducer of EMT because it involves both Smad and non-Smad signaling (Fig. 1) to regulate genes controlling cell motility and invasion, by remodeling the actin cytoskeleton and extracellular matrix.^41,47,69 Smad proteins play a vital role in TGFβ-induced EMT as inhibition of Smad2, Smad3 and Smad4 functions, and overexpression of the negative regulator Smad7, blocked EMT in NMuMG cells.^45,70 TIF1γ (TRIM33), a histone binding protein, antagonizes EMT by competing with Smad4 for binding of active Smad2/3 complexes.⁷¹ In addition, Smad proteins form complexes with members of the AP1 family to induce genes related to invasiveness, e.g., matrix metalloproteinase genes MMP2 and MMP9, during EMT.⁷²

The robust EMT response to TGFβ in NMuMG cells could be attributed to the genotype of these cells, which harbor one wild-type and one mutant p53 R277C allele. Silencing of the wild-type p53 allele hastened the onset of TGFβ-induced EMT, whereas p53 stabilization by nutlin counteracted the EMT effects.⁷³ The p53 protein blocks EMT by upregulating miR-34 and miR-200, which target Snail1 and ZEBs, respectively.^74,75 In agreement with this, molecular analysis of p53-null breast cancers classified them as having the claudin-low tumor gene signature, that is, cancers with dominant EMT and cancer stem cell characteristics.⁷⁶

In addition, TGFβ directly regulates the majority of the EMT-TFs.⁴¹ We found that TGFβ, via Smad signaling, upregulates HMGA2 to drive EMT,¹⁹ and this has been complemented independently in vivo as HMGA2 overexpression causes metastasis,⁷⁷ further reinstating the link between EMT and metastasis. A related family member HMGA1, overexpressed in basal-like breast cancers, is required for EMT and self-renewal to promote metastasis.⁷⁸ Snail1 is another direct target of TGFβ/Smad signaling and of HMGA2.^20,79 Furthermore, Smads and Snail1 interact and are recruited to repress E-cadherin and coxsackie and adenovirus receptor (CAR), a tight junction protein.⁸⁰ Likewise, ZEB1 and ZEB2 are upregulated by TGFβ and also able to interact with Smad proteins.^81,82

EMT can be induced via the non-canonical TGFβ signaling pathway that signals in parallel to Smad activation downstream of the 2 TGFβ receptors, the type II (TβRII) and type I (TβRI)/ALK5 receptor serine/threonine kinases. TGFβ activates the mitogen activated protein kinase (MAPK) and PI3K pathways (Fig. 1) to regulate cytoskeleton organization, migration and invasion, which are characteristics of the EMT response.⁸³ TGFβ receptors signal through the ShcA adaptor protein to activate the ERK MAPK and p38 MAPK pathways to promote breast cancer cell proliferation and invasion.^84,85 TβRI recruits the TRAF6 ubiquitin ligase and becomes polyubiquitinated, leading to the cleavage of the TβRI intracellular domain, which enters the nucleus to activate Snail1 and MMP2 genes.⁸⁶ TGFβ upregulates Sox4, in a Smad-independent manner, to regulate a group of mesenchymal genes and the histone methyltransferase EZH2 during EMT.^21,22 Furthermore, Sox4 creates a feedback loop by activating canonical TGFβ signaling to maintain EMT.²² Overexpression of Sox4 is often found associated with the aggressive, triple-negative breast cancer subtype.²² TGFβ disrupts cell polarity during EMT when it activates TβRII, which phosphorylates the polarity complex protein Par6 to recruit the Smurf1 ubiquitin ligase. Smurf1 ubiquitinates and targets RhoA for degradation, destabilizing the actin filament network and thereby dissolving the tight junctions.⁸⁷

Downstream of the PI3K lies the Akt signaling pathway that plays an important role in cell survival, motility and metastasis. Inactivating mutations in PTEN, the phosphatase responsible for counteracting PI3K activity, is commonly found in human malignancies. Although several features, like the mechanism of activation, are shared between the 3 isoforms of Akt (Akt1-3), distinct signaling effects downstream of the specific isoforms are beginning to emerge.⁸⁸ In a study by Arboleda and colleagues Akt1-3 were overexpressed in breast and ovarian cancer cells. Only Akt2 overexpressing cells displayed increased invasive behavior in an in vitro invasion assay.⁸⁹ When injected into the mammary fat pad of immunocompromized mice, Akt2 transfected MDA-MB-435 breast cancer cells showed a higher number of lung metastases compared to the parental cell line. This effect was abrogated when a kinase-dead Akt2 was expressed. These findings indicate that Akt2 needs to be activated for metastasis to occur. In agreement, overexpression of active Akt2 in 2 different transgenic mammary carcinoma models was found to markedly increase the incidence of pulmonary metastases, in contrast to overexpression of active Akt1.⁹⁰ The pro-metastatic properties of Akt2 may reside in its capacity to regulate the activity of Twist1 and Snail1, 2 classical EMT-TFs.^91,92 Expression of both Twist1 and Akt2 was found to be elevated in highly invasive breast cancer cells,⁹¹ and silencing of Akt2 decreased Twist1-induced migration and invasion. Moreover, Twist1 was found to bind E-box elements in the Akt2 promoter and thereby enhance its expression.⁹¹ Snail1 and Akt2 have been shown to directly interact in the cell nucleus, an interaction that requires the C-terminal half of Snail1 and the plekstrin homology-domain in Akt2.⁹² Snail1 and Akt2 bind together to the E-cadherin promoter and cooperate to repress expression of E-cadherin. Silencing of Akt2 prevents Snail-induced downregulation of E-cadherin.⁹² Finally, Iliopoulos and colleagues showed that the miR-200a family regulates E-cadherin repression and EMT.⁹³ The amount of miR-200a present in a cell was shown to be dependent on the balance between Akt1 and Akt2 expression. A low ratio of Akt1 to Akt2 promotes EMT by reducing the level of miR-200a, again highlighting Akt2 as a driver of EMT.

Activation of the PI3K pathway by TGFβ additionally leads to signaling via the mammalian target of rapamycin complexes -1 and -2 (mTORC1 and mTORC2). The activity of mTORC2, not mTORC1, is essential during TGFβ-induced EMT.^94,95 However, mTORC1 activation by TGFβ, in addition to increasing protein synthesis and cell size, promotes migratory and invasive behavior of mammary epithelial cells.⁹⁴ TGFβ also induces integrin-linked kinase (ILK) phosphorylation of Rictor (a core component of mTORC2 that sets it apart from mTORC1) and thereby promotes the interaction of Rictor with ILK to regulate EMT.⁹⁶ Knockdown experiments showed that Rictor is required during TGFβ-induced EMT but any functional difference between Rictor/mTORC and Rictor/ILK complexes remains unknown.

Epigenetic Rewiring and Noncoding RNAs in EMT

DNA methylation and post-translational modifications of histones, both of which are intrinsically linked, have an impact on gene expression and chromatin function.⁹⁷ Gene repression is associated with nucleosomal histone 3 di-/tri-methylation on lysine 9 (H3K9me2/3) or tri-methylation on lysine 27 (H3K27me3) and DNA methylation; whereas active genes are associated with histone acetylation and histone H3 tri-methylation on lysine 4 (H3K4me3).⁹⁸

EMT entails dynamic changes in expression of the epithelial and mesenchymal gene programs. In recent years, there has been a keen interest in understanding the epigenetic regulation during EMT (reviewed in Ref.^99,100). Several studies have shown that genome-wide epigenetic changes occur during EMT in response to TGFβ (Fig. 1) or during overexpression of EMT-TFs.^21,101,102 Histone modification marks, such as H3K9me2 and H3K4me3, were reduced globally by the action of the dual histone demethylase LSD1 that demethylates H3K4 and H3K9 in mouse hepatocytes undergoing TGFβ-induced EMT.¹⁰² Similarly, differences in repressive H3K27me3 marks mediated by EZH2, which is part of the polycomb repressive complex 2 (PRC2), were observed globally in mammary epithelial cells that overexpressed Twist1, Snail1 or were stimulated with TGFβ.^21,101 Mammary epithelial cells exposed to TGFβ over time gained repressive histone H3K9me2 and lost acetylated H3K9 active marks, an epigenotype also identified in triple-negative breast tumor tissues.¹⁰³ The gain or loss of specific histone marks are often correlated with the regulation of EMT-related genes.^21,101,102 For example, the Snail1 promoter contains high levels of the repressive mark H3K27me3 in epithelial cells. TGFβ upregulates the histone demethylase KDM6B, which removes the methyl groups from H3K27 tails on the Snail1 promoter and thereby allowing Snail1 to be expressed and cause EMT. Furthermore, KDM6B promotes cell invasion and invasive breast carcinomas show high expression levels of KDM6B compared to normal breast tissue.¹⁰⁴

Epigenetic modifications may not be determined by specific DNA sequences, but DNA and histone modifying factors often interact and are recruited to the chromatin by, for example, EMT-TFs which recognize E-box DNA elements with the consensus 5’-CANNTG-3’, and therefore exhibit gene or sequence specific functional effects.⁸ The mechanism of E-cadherin repression by Snail1 is mediated through the recruitment of histone deacetylases (HDACs) and the corepressor SIN3A.¹⁰⁵ Several recent studies have shown that Snail1 also interacts with additional histone modifying enzymes (e.g., G9a, Suv39H1, EZH2, LSD1 and LOXL2)^103,106-109 and DNA methyltransferases (DNMTs)¹⁰³ during EMT. An active E-cadherin promoter is marked by H3K4me3 and has to be first demethylated by the demethylase LSD1,^108,110 before H3K9 could be di-methylated by G9a,¹⁰³ or tri-methylated by Suv39H1,¹⁰⁶ to mediate repression of E-cadherin. Knockdown of G9a, Suv39H1 or LSD1 in breast cancer cells reduced migration and metastasis in xenograft models.^103,106,110

Recently, Snail1 got implicated in the reorganization of heterochromatin by LOXL2, which removes methyl groups from H3K4 by deamination.¹¹¹ Snail1 pairs with LOXL2 to downregulate non-coding pericentromeric transcripts (major satellites) that are required for heterochromatin formation and this led to a temporal release of HP1α from the heterochromatin at the initial stage of EMT. Forced expression of major satellites inhibits the release of HP1α and thus, blocks TGFβ-induced EMT. In an inducible EMT model, Javaid et. al analyzed the chromatin dynamics upon Snail1 activation in human mammary epithelial cells and found that Snail1 represses epithelial genes at the early phase of EMT and sustained Snail1 expression activates mesenchymal genes at the late stage of EMT.¹¹² The short time frame during which HP1α is displaced from chromatin¹¹¹ coincides with the early regulation of epithelial genes by Snail1,¹¹² suggesting that heterochromatin rearrangement by Snail1 is not sequence-specific.

Twist1 recruits SET8 to mono-methylate H4K20 at the promoters of both E-cadherin and N-cadherin genes, mediating the cadherin switch during EMT and breast cancer progression.¹¹³ However, it is still not clear how the H4K20me1 mark contributes to repression of E-cadherin or activation of N-cadherin gene expression.¹¹³ Furthermore, Twist1 could lead to a global change in genes having bivalent marks H3K4me3 and H3K27me3, which could explain the cell plasticity of cells undergoing EMT.¹⁰¹ Twist1 forms a complex with another chromatin regulator, BRD4, to regulate WNT signaling and promote invasion and tumorigenicity in basal-like breast cancer cells.³⁷

PAD4 is a known histone modifier, which citrullinates arginine residues on histones, and has been recently shown to maintain epithelial integrity in mammary epithelial cells by citrullinating and stabilizing GSK3β. Depletion of PAD4 leads to GSK3β destabilization, increased TGFβ signaling and onset of EMT. Low expression levels of PAD4 correlated with invasiveness in breast tumor samples.¹¹⁴ However, the global pattern and functional effects of histone citrullination by PAD4 on chromatin during EMT remain unclear.

Long-term silencing of genes is achieved through DNA methylation, in which DNMTs methylate the C5 position of cytosine in CpG dinucleotides.⁹⁷ De novo DNA methylation of several promoters, most notably of the E-cadherin gene, occurs in human mammary epithelial and breast cancer cells undergoing EMT.^103,115,116 Thus, TGFβ signaling leads to increased DNA methylation at the E-cadherin promoter.¹⁰³ Sustained DNA methylation of epithelial genes is maintained by DNMT1 in mammary cells with EMT features, and this involves active TGFβ/Smad signaling as knocking down Smad2, but not Smad4, could prohibit DNMT1 from binding to its target promoters and therefore caused re-establishment of the epithelial phenotype.¹¹⁷

Noncoding RNAs (ncRNAs) are transcribed in many cell types and these transcripts do not encode for any proteins but yet are important in biological processes. The most studied members of ncRNA are the miRNAs that are first synthesized as primary miRNA, cleaved by Drosha into precursor-miRNA and then processed into mature miRNA (averaging 22 nucleotides) by Dicer. MiRNAs inhibit gene expression by targeting mRNA for degradation or by hindering ribosomal translation.¹¹⁸ The miR-200 family (miR-200a/b/c, miR-141 and miR-429) and miR-205 are the most prolific miRNAs in controlling the EMT program, as they negatively regulate ZEB1 and ZEB2 expression.^119,120 In a feedback mechanism, ZEB1 binds to the miR-200 promoters to reinforce the mesenchymal phenotype.¹²¹ TGFβ signaling decreases the levels of miR-200 family and miR-205 during EMT, through EMT-TFs,¹²¹ or through de novo methylation of their promoters,¹²² and allows the stabilization of ZEB1 and ZEB2 levels. Overexpression of miR-200 in mesenchymal cells led to MET.^119,120 Furthermore, BMPs induce miR200 family and miR-205 and promote MET during reprogramming.¹²³ The tumor suppressor p53 maintains MET and decreases invasion of cancer cells by inducing miR-34 to target and degrade Snail1. Similar to the miR-200/ZEB feedback loop, Snail1 can bind the E-boxes in the miR-34 promoter and repress its expression.¹²⁴ TGFβ upregulates expression of miR-155, whose target is RhoA, leading to the dissolution of tight junctions in mammary epithelial cells.¹²⁵ High expression of miR-155 correlates with invasive breast cancers. A second target of miR-155 is C/EBPβ, a differentiation transcription factor that activates promoters of epithelial genes like E-cadherin and CAR. C/EBPβ is downregulated by miR-155 during TGFβ-induced EMT and its expression is often lost in triple-negative breast cancers, which are often negative for E-cadherin expression.¹²⁶

There has been a growing interest in another group of ncRNAs, known as long noncoding RNA (lncRNA), in relation to cancer progression and metastasis. LncRNAs have lengths of up to 100 kb and are implicated in epigenetic regulation of genes.¹²⁷ An example is the Hotair lncRNA, which binds and guides PRC2 to its gene targets, where PRC2 mediates H3K27 methylation and thereby silencing of genes. Recently, Hotair has been shown to be transcriptionally induced by TGFβ and is required for the EMT process and the gain of cancer stem cell properties. Inhibition of Hotair in breast cancer cells prevented downregulation of E-cadherin.¹²⁸ Hotair is highly expressed in metastatic breast cancers and correlates with poor prognosis.¹²⁹ The H19 lncRNA is another target of TGFβ signaling and is also regulated in a feed-forward mechanism by TGFβ-induced Snail2 during EMT. H19 and Snail2, together with miR-675, repress E-cadherin expression and promote metastasis in vivo.¹³⁰ Upregulation of Malat1 lncRNA by TGFβ has been described in bladder cancer. Malat1 associates with Suz12, another component of PRC2, and regulates EMT genes, and knockdown of either Malat1 or Suz12 inhibits tumor metastasis.¹³¹

Thus, the EMT program involves chromatin remodeling, generating a plastic epigenetic landscape characterized by bivalent genes, i.e., genes with both active H3K4me3 and repressive H3K27me3 marks.¹⁰¹ Such “poised” chromatin is then available to undergo the inverse changes during the MET process that is important for somatic reprogramming to induced pluripotent stem cells.¹³² This could explain the case why EMT has to precede MET during somatic cell reprogramming.¹³³

EMT and Breast Cancer Stem Cells

Data accumulating during recent years support a link between EMT and generation of cancer stem cells. The concept of breast CSCs was first established by the Clarke laboratory and defines a population of cancer cells that has the ability to form new tumors (tumor initiating cells).¹³⁴ Breast CSCs constitute a minority of the tumor cells and show high expression of the hyaluronan receptor CD44 and low levels of the surface sialoglycoprotein CD24. A few years later it was demonstrated that breast CSCs could be enriched in populations that underwent EMT (Fig. 1) due to the high expression of EMT-TFs, like Snail1 and Twist1.^5,6 The connection between EMT and stem cell properties may reside in the process of reprogramming that epithelial cells undergo during EMT,^99,102 as will be exemplified in this section.

By transfecting differentiated somatic cells with specific transcription factors that promote stemness, Oct4, Klf4, Myc and Sox2, pluripotent stem cells can be induced (iPSCs).¹³⁵ It was recently shown that optimal reprogramming to stem cells was obtained by step-wise introduction of the factors listed above and involved a sequential EMT-MET process.¹³³ Data from other research groups also confirm the reprogramming to iPSCs as a multistep process, where pluripotency during the early EMT phase is promoted by TGFβ. In the later MET phase, BMP signaling plays a key role by upregulating miR-200 family members that enforce the epithelial phenotype.¹²³ These studies support the notion that reprogramming during EMT is not only relevant to cancer but has direct links to the more physiological process of induced stemness and the generation of pluripotent stem cells. Cycles of EMT and MET seem to play an important role during reprogramming of normal stem cells and cancer cells, suggesting that these processes may provide an adaptive force when cells change their differentiation potential. This is compatible with the findings that TGFβ, BMP and Wnt signaling pathways mediate these transitions in differentiation and stemness in the context of both normal and breast cancer cells.³⁹

Several examples of signaling pathways and transcription factors regulating both EMT and stemness exist. The mesenchymal state of mammary cells is established by the EMT-TF FOXC2, which also contributes to CSC potential at least in part by inducing transcription of the PDGF β-receptor.¹³⁶ Signaling via PDGFRβ is also required for survival of cancer cells undergoing EMT.¹³⁷ In agreement, sunitinib, a clinically approved PDGF receptor inhibitor, reduced both the CSC population and the metastatic potential of breast cancer cells with high FOXC2 expression. A second example is the Hippo pathway and its downstream effector protein TAZ, which can inactivate the epithelial polarity protein Scribble, thus causing EMT.¹³⁸ In addition, TAZ has been shown to promote stemness in breast cancer cells when overexpessed and activated by Hippo pathway kinases. A hallmark feature of cancer cells is their escape from oncogene-induced senescence and apoptosis. Interestingly, EMT-TFs like Twist1 and ZEB1/2 interfere with key pathways regulating cell survival, such as the p53, pRb and PP2A phosphatase, resulting in the development of breast cancers that have characteristic enrichment in stem-like features.¹³⁹ Resistance to cell death is indeed an important feature not only during the process of EMT, but also for the generation of CSCs.¹⁴⁰ Based on a variety of findings, it could be argued that EMT and changes in differentiation induce stem like properties in cancer cells, as opposed to the hypothesis of CSCs being derived from malignant transformation of stem-like progenitor cell types in the mammary gland.

Normal and malignant stem cells are dependent on a tissue microenvironment often called the stem cell niche. Metabolic conditions in the niche are commonly found to regulate cancer stemness and one prominent example in the breast cancer niche is hypoxia. Hypoxia affects multiple molecular pathways, one being the Notch ligand Jagged2 that causes EMT and promotes breast CSC growth; consistent with this observation, Jagged2 expression and active Notch signaling have been found to correlate with poor prognosis and metastasis of human breast cancers.¹⁴¹ The significance of hypoxia for the development of stem cell properties is illustrated by an experiment where breast cancer cells were exposed to cycles of hypoxia and normoxia. This change in microenvironment resulted in enrichment of breast cancer cells with enhanced tumor-initiating potential in mouse transplantation experiments.¹⁴² The hypoxia-inducible protein carbonic anhydrase IX (CAIX) contributes to changes in local pH induced by hypoxia.¹⁴³ CAIX can promote breast CSC proliferation by activation of mTORC1 signaling leading to EMT and high expression of Jagged1 and Notch1. Treatment of mice with a CAIX inhibitor, suppressed CSCs and combination therapy with paclitaxel and the CAIX inhibitor potently blocked metastasis in these mice.¹⁴³

The concept that EMT links to cancer stemness could potentially offer novel possibilities for improved prognostics and therapeutic intervention for human breast cancer. A recent study addressing the phenotype of circulating tumor cells in the blood of breast cancer patients revealed a diverse cohort of cells exhibiting both epithelial and mesenchymal features.¹⁴⁴ Interestingly, a correlation to poor prognosis was suggested for circulating tumor cells harboring mesenchymal features, including TGFβ signaling intermediates and EMT-TFs.

Platelet-Derived TGFβ and NF-κB Signaling as Inducers of EMT and Metastasis

As described above, TGFβ is a central mediator of EMT and subsequent metastasis. In the tumor microenvironment, TGFβ is expressed by stromal cells such as macrophages but also by the tumor cells themselves. Another major source of TGFβ are megakaryocytes in the bone marrow and their circulating derivatives, platelets.¹⁴⁵ TGFβ, along with a plethora of other growth factors and cytokines, are stored in platelet α-granules and released upon activation of the platelets. Tumors generate a pro-thrombotic environment and promote enhanced coagulation and platelet activation primarily due to deregulated expression of tissue factor. As a consequence, cancer patients commonly have an elevated turnover of platelets in their circulation and increased risk of thrombotic occlusion of vessels.¹⁴⁶ In a recent study from the group of R.O. Hynes it was demonstrated that direct contact between tumor cells and platelets induces EMT (Fig. 1) and primes the tumor cells for metastases.¹⁴⁷ Co-culturing of platelets with either breast or colon carcinoma cells induced a more invasive mesenchymal-like phenotype of the tumor cells, which resulted in enhanced seeding of tumor cells in the lungs of mice when injected through the tail vein. By creating a megakaryocyte/platelet-specific knock-out of TGFβ (Pf4-cre;TGFβ1^fl/fl), it was elegantly shown that platelet-derived TGFβ is crucial to promote EMT and increased invasiveness. In addition to providing TGFβ, the direct contact with platelets triggered NF-κB signaling in the tumor cells, which was also required for platelet-induced EMT to take place. Platelets have for long been recognized to promote metastasis and potential mechanisms include protection of the tumor cells from immune recognition and shear stress in the circulation, as well as easier tumor cell adherence and subsequent extravasation through the vessel wall when platelets bind to the tumor cells.¹⁴⁸ The study by Labelle et al adds another mechanism to the list and provides the first molecular explanation for platelet-induced metastasis. These findings highlight the potential benefit of keeping platelet activity as low as possible in cancer patients, without increasing the risk of bleedings. Indeed, recent clinical data show that patients that have been on low dose aspirin for cardiovascular disease for several years have a dramatically decreased risk to develop metastatic cancer.^149,150 This connection has therapeutic relevance since aspirin efficiently prevents platelet activation.

MET and Effective Metastatic Colonization

Although the process of EMT generates mesenchymal and stem-like cells with enhanced invasive and metastatic capacities, it appears that what matters for the effectiveness of metastatic dissemination and secondary tumor growth is the ability of metastatic stem cells to generate new epithelial cancer cells after undergoing MET (Fig. 1). Early studies focusing on signaling by the BMP pathway and its impact on EMT, established that BMPs can counteract TGFβ-induced EMT and promote MET in mammary epithelial cells.^151,152 BMP signaling is mediated by distinct receptor serine/threonine kinases and distinct Smads, Smad1, Smad5 and Smad8 (Fig. 1), however, Smad4 also serves as a cofactor of the other Smads in both BMP and TGFβ pathways.^41,47,69 Using this canonical Smad signaling pathway, BMP induces the transcription of genes of the ID family, which can negatively regulate the function of pro-EMT TFs with bHLH domains, thus counteracting EMT and possibly stimulating the MET. Accordingly, TGFβ represses the expression of ID family genes (Fig. 1) and this is required for the establishment of EMT.^151,152 This principle has been nicely exploited therapeutically in mouse models where breast cancer metastasis was effectively blocked by the delivery of BMP ligands to the host animals.¹⁵³ Recent work has also revealed a more complete and unavoidably more complex picture of the relationship of ID gene regulation and MET in breast cancer metastasis.¹⁵⁴ Upon induction of EMT by TGFβ, mesenchymal breast cancer cells also upregulate ID1, which binds to Twist1, inactivates its transcriptional activity and promotes MET at metastatic colonies.¹⁵⁴ This is an interesting model that emphasizes the temporal aspect of regulation of key EMT-TFs during the EMT-MET cycle. In the same line of evidence, transgenic mice expressing Twist1 in a conditional manner in the skin and upon a chemical carcinogenesis protocol, demonstrated that transient Twist1 induction is critical for EMT induction and invasive tumor growth, however, when the squamous carcinomas would metastasize, Twist1 had to be silenced so that MET would prevail in order for metastatic epithelial colonies to be established after dissemination through the vasculature.¹⁵⁵ However, one should consider that breast cancer metastasis in the absence of MET may well be possible and this clearly applies in certain mouse models such as the EpRas model described above.⁴⁶ The evidence for a requirement of a succession of EMT-MET cycles suits the current understanding of the multifunctional roles TGFβ and BMP play in breast cancer progression. This evidence also cautions experimentalists about the relative difficulty of applying this basic knowledge of tumor biology to the effective use of novel therapeutics that may perturb the EMT-MET cycle or the corresponding TGFβ-BMP signaling networks.^7,156 Careful consideration of the possibilities these alternate signaling processes may offer to the development of therapeutic approaches is warranted.

Perspectives

Here we have highlighted some key examples and processes that explain how cytokines of the TGFβ family promote the EMT and how they may also have an impact on the inverse process of MET. A major need for current investigation in this field is to understand deeper signaling crosstalk between multiple pathways that are initiated by diverse growth factors and cytokines and their relative contribution to EMT and breast cancer metastasis. We need to explore deeper the links between chemical modifications of chromatin, which currently are referred to as epigenetic mechanisms of gene regulation, to the action of specific EMT-TFs and the process of breast cancer stem cell generation. The specific involvement of non-coding RNAs will continue to provide novelty to this area of cancer biology. Of equal importance are studies that aim at clarifying the spatiotemporal control and the molecular analysis of EMT-MET cycles in breast cancer progression, linking the cyclical or reversible processes to the plasticity of chromatin modifications during the metastatic trajectory. The possibility that MET is also driven by the EMT-generated CSCs is an interesting open hypothesis (Fig. 1) that promises continues excitement in this booming area of tumor biology.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgments

We thank all past and present members of our groups for their contributions to the scientific work emanating from our laboratories. This article summarizes work from a large number of published papers; however, due to space limitations, we have been unable to include all relevant publications in our discussion. We apologize to those authors whose relevant work has not been included in this review article.

Funding

AM acknowledges funding by the Ludwig Institute for Cancer Research, the Swedish Cancer Society, the Swedish Research Council and the Marie Curie Initial Training Network “IT-Liver” under the European Union FP7 program. A-KO acknowledges funding by the Swedish Cancer Society.