Update on Epithelial-Mesenchymal Plasticity in Cancer Progression

Abstract

Epithelial-mesenchymal transition (EMT) is a cellular process by which epithelial cells lose their characteristics and acquire mesenchymal traits to promote cell movement. This program is aberrantly activated in human cancers and endows tumor cells with increased abilities in tumor initiation, cell migration, invasion, metastasis, and therapy resistance. The EMT program in tumors is rarely binary and often leads to a series of gradual or intermediate epithelial-mesenchymal states. Functionally, EMT plasticity (EMP) improves the fitness of cancer cells during tumor progression and in response to therapies. Here, we discuss most recent advances in our understanding of the diverse roles of EMP in tumor initiation, progression, metastasis and therapy resistance, and address major clinical challenges due to EMP-driven phenotypic heterogeneity in cancer. Uncovering novel molecular markers and key regulators of EMP in cancer will aid the development of new therapeutic strategies to prevent cancer recurrence and overcome therapy resistance.

INTRODUCTION

Epithelial to Mesenchymal Transition (EMT) is a cellular process during which epithelial cells lose some of their epithelial characteristics and gain mesenchymal phenotypes to allow cell movement. In physiological contexts, activation of EMT plays critical roles during many embryonic processes, including gastrulation, neural crest formation, heart valve formation, and wound healing; while in pathological contexts, EMT has been considered a key driver of organ fibrosis and cancer progression and metastasis (1).

In tumor cells, activation of EMT can be triggered by various signals from the tumor microenvironment, such as Wnt, Notch, growth factor receptor signals, inflammatory cytokines, hypoxia (2) and changes in extracellular matrix (ECM) density and rigidity (3–5). Induction of EMT is orchestrated by a number of core EMT transcription factors (TFs), including SNAI1/2, TWIST1/2 and ZEB1/2. Together, they play pleiotropic roles to directly or indirectly repress expression of epithelial genes such as CDH1 (E-cadherin) and TJP1 (ZO-1), and promote expression of mesenchymal genes, such as VIM (Vimentin), FN1 (Fibronectin), and CDH2 (N-cadherin), FSP1 (6, 7). These molecular changes allow carcinoma cells to lose apical–basal polarity and epithelial junctions and to acquire mesenchymal characteristics such as mesenchymal morphologies as well as increased migratory and invasive capacities to breach the basement membrane.

Although EMT has been traditionally viewed as a static transition between two distinct cell populations, epithelial (E) vs. mesenchymal (M) cells during embryonic development, recent studies revealed that individual carcinoma cells activate EMT to different extents, thus entering hybrid or partial E/M states. Such tumor cells often express different levels of E and M markers simultaneously and exhibit phenotypic, transcriptional and epigenetic features of both epithelial and mesenchymal cells, and such properties are best described as Epithelial-Mesenchymal Plasticity (EMP) (8). Research in the past decade revealed a number of critical roles of EMP in cancer progression (Figure 1). Here we review publications from the last 6 years focusing on the impact of EMP on acquisition of cancer stem-like properties, metastasis, tumor dormancy, and cancer therapy resistance.

Figure 1. Epithelial-Mesenchymal Plasticity in tumor progression and metastasis.

During early stages of tumor progression, tumor cells undergo partial (Hybrid E/M) or total EMT (Mesenchymal), acquiring the ability to (1) invade into adjacent connective tissues as single mesenchymal cells or collectively as clusters of hybrid E/M cells, and (2) intravasate into the blood vasculature. (3) Once in circulation, maintenance of EMT programs endows CTCs with enhanced survival and (4) extravasation ability. (5) After initial seeding to distant organs, DTCs may remain (6) dormant for years until they resurge into clinically detectable lesions. Activation of MET is required to reinitiate the growth of dormant DTCs into (7) macrometastases.

EMP IN CANCER STEMNESS

During tumor development, acquisition of malignant traits is associated with the presence of a small pool of cancer cells with tumor initiating properties, named cancer stem cells (CSCs) (9). CSCs were functionally identified by their ability to initiate tumor growth when implanted in mice, thereby they are also termed tumor initiating cells (TICs). Importantly, given their capacity to self-renew and to give rise to phenotypically diverse subpopulations of tumor cells, CSCs have been implicated in tumor initiation, tumor heterogeneity, metastasis, and disease recurrence after therapies. Since most CSCs are quiescent and slow cycling, they are also tightly associated with chemoresistance (10, 11).

Several studies over the past decade revealed that EMT promotes the acquisition of cancer stem cell properties. The connection between EMT and cancer stemness was first suggested by Brabletz et al. (12) in 2005 when the authors proposed the “migrating cancer stem (MCS)-cell” concept to explain that although EMT and CSCs cover distinct aspects of carcinogenesis, they should be considered as parts of the same process in order to explain the diverse cellular changes occurring during the evolution of metastatic lesions. Later, Mani and colleagues (13, 14) reported for the first time that induction of EMT through TWIST1 or SNAI1 expression in normal or neoplastic mammary epithelial cells induced the expression of cancer stem cell markers, enabling mammosphere or tumorsphere formation in vitro respectively and, more importantly, increasing tumor initiation capacity in vivo (13, 15). Consistent with these initial reports, several other groups reported that EMT is important for the acquisition of stem-like traits, which contribute to tumor formation, metastasis, tumor dormancy and therapy relapse in different tumor types including breast (13–16), ovarian (17), pancreatic (18), squamous cell carcinoma (SCC) (19), colon (20), prostate (21) and gastric cancers (22).

Although TICs isolated either from mouse or human mammary glands or mammary carcinomas express numerous EMT markers and exhibit a mesenchymal phenotype (13), tumor cells with an epithelial phenotype are also reported to show stem-like features both in vitro (23) and in vivo (24–26). Using skin tumor models in which individual stages of the tumor development including tumor initiation, maintenance/propagation and malignant progression are well defined, Beck and colleagues (25) demonstrated that TWIST1 expression was critical for skin tumor initiation and progression in a dose-dependent manner. The authors showed that high levels of TWIST1 were necessary for metastasis progression, while low levels of Twist1 were required for skin tumor initiation and maintenance by promoting cell proliferation and conferring stem-cell-like properties through an EMT-independent mechanism. Accordingly, using an inducible Twist1 construct in human mammary epithelial cells, studies by Schmidt et al. (26) showed that long term expression of TWIST1 induced EMT without conferring any mammosphere-forming ability, but subsequently, turning off TWIST1 could revert tumor cells into an epithelial phenotype and conferred strong mammosphere-forming abilities. Furthermore, transient expression of TWIST1, which was not sufficient to induce EMT morphological changes, primed a subset of mammary epithelial cells for stem-cell-like properties, which only emerged and persisted following Twist1 deactivation. These studies suggest that the ability of both skin and breast cells to acquire stem-cell-like properties does not depend on activation of the EMT program per se; instead, the tumor initiating ability is associated with an epithelial phenotype and is regulated by oncogenic functions of EMT-TFs independent of EMT.

These seemingly conflicting results complicate the relationship between EMT and stemness, but recent data in ovarian (27) and breast cancers (28, 29) showed that tumor cells residing in a hybrid E/M state formed tumors when injected in mice and displayed features generally associated with cancer stemness, whereas fully epithelial or fully mesenchymal tumor cells lost their tumor initiation abilities. Accordingly, Pastushenko and colleagues (30) reported that acquisition of a hybrid E/M state through the deletion of Fat1 in both human and mouse SCC accelerated tumor initiation upon subcutaneous transplantation in mice and increased clonogenicity of human SCC in vitro. In a previous study from the same group, the authors show that while tumor cells in different hybrid E/M states presented similar TIC frequency, the ones in the earlier EMT intermediate state already exhibited increased TIC frequency, suggesting that tumor stemness does not increase further in more advanced EMT transition states in SCCs (31). Consistent with these results, the hybrid E/M clone P4B6 obtained from OPCT-1 human prostate cancer cells expressed high levels of stem cell markers with faster and larger tumor formation in vivo (32). Moreover, in non-small cell lung cancer (NSCLC), Andriani and colleagues (33) found that in response to TGF-β treatment, the E/M hybrid A549 (hybrid E > M) and LT73 cells (hybrid M > E) expressed a higher level of CD133, a classical marker of lung cancer CSCs (34) associated with higher tumorigenic potential in vivo. Interestingly, in NSCLC tumors, the hybrid E/M phenotype – identified by co-expression of E-cadherin and SNAI2 – is correlated with significantly poorer survival in comparison to tumors displaying largely epithelial features, or even epithelial and mesenchymal tumors combined. Together these studies suggest that tumor cells in a hybrid E/M state have increased stem-like properties and are more aggressive (33).

Consistent with these results, in triple-negative breast cancer (TNBC) cells, expression of integrin β4 (ITGB4) is reported to identify a specific E/M hybrid subpopulation that presents more stem-like characteristics, tumor initiating abilities and exhibited worse prognosis in patients (35). Mechanistically, expression of the EMT-TF ZEB1 reduced tumor initiation abilities of the highly mesenchymal SUM159 TNBC cells by blocking the of ITGB4 expression (35). Accordingly, another study by Kröger et al. (29) showed that forcing breast cancer cells into a complete mesenchymal state through the expression of Zeb1 and activation of the noncanonical Wnt signaling resulted in a poorly tumorigenic cell population. In contrast, high level of the EMT-TF SNAI1 and activation of the canonical Wnt signaling drove breast cancer cells to a highly tumorigenic hybrid E/M state. These studies argue that acquisition of the tumor initiating ability in breast cancer is associated with an E/M hybrid state characterized by expression of SNAI1, whereas expression of ZEB1 induces the cells to acquire a mesenchymal phenotype, which is associated with loss of stem-like properties.

Interestingly, while expression of ZEB1 inhibits stem-like properties by locking tumor cells in a mesenchymal state in breast cancer, Zeb1 depletion in pancreatic ductal adenocarcinoma (PDAC) cells reduced their tumorigenic capacity in vivo and destroyed their sphere-forming abilities in vitro, suggesting that ZEB1 is crucial for cellular plasticity and stemness/tumorigenic properties in PDAC (36). Zeb1 expression also correlates with more aggressive precursor lesions and poor outcomes in human pancreatic cancer (18, 37). Consistent with the role of ZEB1 in PDAC tumor initiation, single-cell RNA sequencing (scRNA-seq) experiments identified a subpopulation of colon cancer cells in a hybrid E/M state, which is characterized by Zeb1 expression, acquisition of stem-like features and the ability to locally invade and metastasize to distant organs (38). These studies lead to the conclusion that different EMT-TFs can induce or inhibit stem-like properties depending on the tumor context.

Although all these studies demonstrate that stem-like properties and tumor initiating abilities are mainly associated with the acquisition of an E/M hybrid state instead of a fully epithelial or fully mesenchymal phenotype, Ruscetti and colleagues (39) showed that in CPKV mouse model of prostate cancer, both hybrid E/M and mesenchymal tumor cells presented enhanced tumor initiating capacity in vivo and stemness characteristics in vitro compared to the epithelial tumor cells. Interestingly, while hybrid E/M tumor cells displayed a proliferative, stem/progenitor cell phenotype, the mesenchymal ones exhibited a quiescent stem cell phenotype. Accordingly, inhibition of epithelial-mesenchymal plasticity by knocking down the chromatin remodeling gene HMGA2 in prostate cancer cells inhibited stemness in vitro and strongly reduced tumor growth and metastasis in vivo (39). Therefore, distinctive EMT intermediate states could confer different tumor initiating properties.

Several studies also attempted to dissect the spatial organization of CSC subpopulations at distinct EMT states in tumor tissues. In breast cancer, Liu and colleagues (40) show that the mesenchymal-like CSCs expressing CD24⁻/CD44⁺ are typically located towards the invasive edge of the tumor at the tumor-stroma interface, whilst the hybrid ALDH1 (aldehyde dehydrogenase 1)⁺ CSCs (originally considered as epithelial cells in Liu et al. (40) but later discovered to be hybrid E/M stem-like cells based on RNA seq experiment (41)) are found in the more interior tumor region. This heterogeneity in spatial localization of CSCs subsets is also observed in prostate cancer (39, 42). A gradient of EMT-inducing signals, such as TGF-β, in the tumor environment, likely contributes to this spatial pattern formation in pancreatic cancer (43, 44). Moreover, although in vitro experiments showed that CSCs could significantly promote both single and collective cell migration in breast cancer, in vivo analyses show that the majority of CSCs located at the invasive front of the tumors are in a hybrid E/M state and exhibit a collective migration pattern (45).

Stem-like properties are not only enriched in tumor cells capable of initiating primary tumors, but are also observed in circulating tumor cells (CTCs) isolated from peripheral blood of cancer patients. scRNA-seq of CTCs isolated from breast cancer patients showed that tumor cells in hybrid E/M states also express stem-related genes (46). Moreover, staining for cytokeratin, ALDH1, and Twist1 in peripheral blood mononuclear cells (PBMC) from 130 breast cancer patients showed that the incidence of CTCs with stem-like characteristics and partial EMT phenotypes increased after first-line chemotherapies, suggesting that cells undergoing EMT might represent a chemoresistant subpopulation associated with poor prognosis and recurrence (47). 80% CTCs in men with castration-resistant prostate cancer and over 75% of CTCs in women with metastatic breast cancer co-express epithelial, mesenchymal and stem cell markers (48). Furthermore, CTCs isolated from colon cancer patients were found to exhibit strong E/M hybrid phenotypes and stem-like properties (49, 50).

Taken together, these studies suggest that although tumor cells in a fully epithelial, fully mesenchymal or hybrid E/M states can all have the potential to initiate tumors in vivo, the ones in hybrid E/M states have the highest tumor initiation capability (Table1). Moreover, EMT-TFs could insert different roles in conferring stem-like properties depending on the tumor context. It is currently unknown whether the stem-like properties conferred by EMP are universal across tumor types. Single-cell analysis approaches to uncover specific stem-like markers for individual E/M hybrid states through the EMP spectrum could help to identify the entire spectrum of E/M hybrid states for in-depth functional analyses of their corresponding tumor initiation properties. Moreover, since hybrid E/M states and stem-like properties are widely observed in both primary tumor cells and CTCs, a deeper knowledge about the connection between EMP and cancer stemness could also reveal novel prognostic markers for metastasis, recurrence and therapy resistance.

Table 1.

Association between stemness/tumor-initiation potential and EMP in different cancer types

Cancer Type	Experiments performed	EMP/stemness association	Reference
Breast	In vivo	Yes	Al-Haji et al., 2003
	In vivo/in vitro	Yes	Mani et al., 2008
	In vitro	Yes	Morel et al., 2008
	In vivo/in vitro	Yes	Morel et al., 2012
	In vitro	No	Ocana et al., 2012
	In vitro	No	Schmidt et al., 2015
	In vivo/in vitro	Yes	Goldman et al., 2015
	In vivo	Yes	Bierie et al., 2017
	In vivo/in vitro	Yes	Kröger et al., 2019
	CTC from patients- In vitro test	Yes	Armstrong et al., 2011
			Cheng et al., 2019
			Papadaki et al., 2019
Ovarian	In vivo	Yes	Strauss et al.,2011
	In vivo/in vitro	Yes	Hojo et al., 2018
Squamous Cell Carcinoma	In vivo	No	Beck et al., 2015
	In vivo	Yes	Pastushenko et al., 2018
	In vivo/in vitro	Yes	Pastushenko et al., 2021
Prostate	In vivo	No	Celia-Terrassa et al., 2012
	In vivo/in vitro	Yes	Ruscetti et al., 2015 Ruscetti et al., 2016
	In vivo/in vitro	Yes	Harner-Foreman et al., 2017
Lung	In vivo/in vitro	Yes	Andriani et al., 2016
Pancreatic	In vivo/in vitro	Yes	Krebs et al., 2017
Colon	CTC from patients- In vitro test	Yes	Cayrefourcq et al., 2015
			Grillet et al., 2017
	In vitro	Yes	Sacchetti et al., 2021

EMP IN TUMOR METASTASIS

Activation of EMT is considered a key mechanism by which cancer cells disseminate from the primary site to distant organs in various epithelial cancer types (6, 36, 51, 52). Interestingly, histological analyses of human carcinoma metastases showed that most distant metastases present epithelial phenotypes that largely resemble their primary tumors, instead of mesenchymal phenotypes (12). These observations, together with the fact that mesenchymal tumor cells cannot be easily distinguished from neighboring stromal cells in tumor tissues, make it challenging to identify tumor cells undergoing EMT in human tumor samples. Therefore, the involvement of EMT in tumor metastasis was in doubt for a long time, especially in the clinical pathology field.

Using various mouse and human tumor models, several studies demonstrate that activation of the EMT program induces tumor cell dissemination; upon arriving at distant organs, disseminated tumor cells (DTCs) undergo a reversion process called mesenchymal-epithelial transition (MET) to allow metastatic outgrow at distant organs (23, 53–57). Using an inducible Twist1 mouse model of SCC, our study demonstrates that expression of TWIST1 is sufficient to activate EMT and allow tumor cell dissemination into the blood circulation; while at distant organs, downregulation of Twist1 to turn off EMT is required for DTCs to proliferate and form macrometastases (55). Similar results were obtained by Tran and colleagues (55, 58) using an inducible Snail1 mouse model of breast cancer. Together, these studies show that activation of EMT by TWIST1 or SNAI1 can potently induce tumor invasion, intravasation and extravasation while MET is crucial for metastasis outgrowth in distant organs. Moreover, Reichert et al. show that p120catenin (ctn) loss in a PDAC mouse model is sufficient to induce EMT and increase metastasis formation (59). However, while p120ctn/E-cadherin loss specifically directs metastasis formation to the lung, its expression is required to induce MET and allow metastasis formation specifically at the liver, suggesting that EMT may also play a role in metastasis organotropism. These studies reveal the dynamic changes between EMT and MET happening during cancer progression and metastasis. The impacts of MET in metastatic dormancy will be discussed in depth in the following section.

Although EMT and MET programs had traditionally been viewed as a binary switch between two alternative static states, mesenchymal or epithelial, it is now widely accepted that EMT is not an “all-or-none” process. Instead EMT is often incomplete, leading to tumor cells residing at intermediate hybrid E/M states between fully epithelial and fully mesenchymal phenotypes (6, 8, 60). To dissect the complexity and plasticity of EMT in metastasis, several groups developed elegant lineage-tracing systems to mark tumor cells undergoing EMT in breast tumor mouse models and follow their fates during metastasis progression (61). Using EMT lineage tracing strategies based on the expression of two mesenchymal reporters, Vimentin and Fsp1 in MMTV–PyMT and MMTV–Neu models of breast cancer, Fischer et al. (62) show that lung metastases are mainly formed by epithelial cells that had never switched on the Fsp1⁺ or Vim⁺ mesenchymal state. Moreover, overexpression of miR-200, which was previously reported to inhibit EMT by suppressing EMT-TFs Snail1/2, Twist1, Zeb1/2 (63, 64), did not inhibit the formation of lung metastasis, but instead, it made tumor cells more susceptible to the chemotherapeutics. Based on these findings, the authors concluded that EMT is not required for lung metastasis in breast cancer, but it contributes to chemoresistance (62). Interestingly, using the same Fsp1-based lineage-tracing reporter by Fischer et al. (62) in combination with a real-time E-cadherin-based epithelial-mesenchymal state reporter, Bornes and colleagues (61) show that Fsp1 is not expressed in the majority of carcinoma cells that have undergone EMT in primary tumors or in the majority of DTCs in hybrid E/M or mesenchymal states that show strong metastatic potentials. More recently, using a novel dual recombinases-mediated genetic lineage tracing strategy based on the expression of Vimentin or N-cadherin in MMTV-PyMT mouse model, Li et al. (65) show that while the primary tumor cells express both activated Vimentin and N-cadherin, only the early mesenchymal marker N-cadherin is activated during EMT and is functionally required for tumor metastasis. Accordingly, immunofluorescence staining of MMTV–PyMT tumor sections confirmed that end-stage mesenchymal markers, such as Fsp1 and Vimentin, are not expressed in most of the mesenchymal and hybrid E/M carcinoma cells (66). Together, these studies show that during metastasis, acquisition of hybrid E/M states, which cannot be identified through the expression of late-stage mesenchymal markers (such as Vimentin and Fsp1) indeed contributes to distant metastasis formation in breast cancer, whereas fully mesenchymal cells fail to colonize the lung (Figure 2) (67).

Figure 2. Summary of lineage-tracing experiments performed to study the role of EMT in cancer metastasis.

In lineage-tracing experiments performed in breast cancer mouse models, cancer cells undergoing EMT are marked based on the expression of mesenchymal markers such as N-cadherin, Vimentin and Fsp1 while epithelial cells were marked based on E-cadherin⁺ expression. Only E-cadherin⁺ and hybrid E/M tumor cells expressing N-cadherin are able to efficiently intravasate and colonize the lung. In lineage tracing experiments done in PDAC mouse models, cancer cells in an E/M hybrid state characterized by E-cadherin cytoplasmic localization efficiently intravasate and are detected as tumor cell clusters in the blood circulation.

In line with these findings, the acquisition of hybrid E/M states has also been identified in PDAC models and associated with high metastatic potential using an innovative inducible CRISPR-Cas9-based lineage recorder capable of simultaneously capturing both transcriptional and phylogenetic information at the single cell level (68). Lineage tracing experiments based on the expression of the epithelial marker E-cadherin showed that the majority of PDAC tumors are in an E/M hybrid state, which is characterized not only by low expression of EMT TFs and co-expression of epithelial and mesenchymal genes, but also by re-localization of the epithelial proteins from the plasma membrane to the cytoplasm (69). These studies, together with the data previously described in breast cancer, highlight that partial EMT programs are required for cancer metastasis formation and the hybrid E/M states are defined not only by the expression but also by the subcellular localization and post-transcriptional/post-translational modifications of EMT TFs and many other EMT markers (Figure 2) (60).

Another major approach to determine the role of EMT in tumor metastasis is to genetically delete core EMT-TFs in various mouse tumor models. Zheng and colleagues (70) showed that in PDAC mouse models, EMT suppression by genetic deletion of either Snai1 or Twist1 did not reduce pancreatic tumorigenesis, local invasion, systemic dissemination or metastasis to distant organs, but instead it increased cancer cell proliferation, thus leading to enhanced sensitivity to gemcitabine treatment and increased overall survival. Interestingly, in the same mouse tumor model used by Zheng and colleagues (69), expression of other EMT-TFs, such as Zeb1, Sox4 and Snail2, was found to persist upon deletion of Snail1 or Twist1. More importantly, Krebs et al. (36) showed that depletion of Zeb1 led to loss of cancer cell plasticity by locking tumor cells in an epithelial state, which strongly reduced tumor grade, invasion and notably metastasis during PDAC progression. These results lead to the conclusion that deletion of Twist1 or Snai1 alone is not sufficient to suppress EMT in the pancreatic tumor mouse model used by Zheng and colleagues (71) and that Zeb1 deletion has a much greater impact on pancreatic tumor metastasis. While TWIST1 seems to be dispensable for PDAC metastasis, it is required for the expression of other EMT TFs (Snai1/2, Zeb2) in a small subset of breast primary tumors and CTCs displaying hybrid E/M states. Deletion of Twist1 inhibited basal-like tumor progression, intravasation and metastasis, suggesting that depending on the tumor context, different EMT-TFs can activate partial EMT programs to promote metastasis.

Although the hybrid E/M states have been mainly identified and studied in breast and pancreatic cancer mouse models, activation of EMP is also reported during cancer cell dissemination and metastasis in other tumor types. In an autochthonous murine model of prostate cancer, while both hybrid E/M and fully mesenchymal carcinoma cells exhibit enhanced invasive capacity compared with epithelial tumor cells, the mesenchymal tumor cells are able to persist in circulation and survive in the lung following intravenous injection and the hybrid E/M carcinoma cells are capable of generating macrometastases (39). Both skin SCC and breast cancer mouse models showed that tumor cells in hybrid E/M states had increased ability to undergo vascular extravasation, lung colonization and metastasis as compared to fully mesenchymal cells when injected intravenously. Moreover, following orthotopic injections, the majority of SCC CTCs detected in the blood circulation were in hybrid EMT states, demonstrating that activation of partial EMT programs in SCCs is not only important for metastasis formation but also allows cancer cells to intravasate more efficiently (30, 31). Hybrid E/M states were also identified in colon cancer by sc-RNA seq experiments and tumor cells in a quasi-mesenchymal state were shown to present the phenotypic plasticity underlying local invasion and distant metastasis (38).

While fully mesenchymal cells migrate and invade as single cells, hybrid E/M cancer cells present both mesenchymal and collective cell migration and invasion patterns (67, 69, 72–75). Hybrid E/M PDAC cells migrate and invade mainly as collective groups in vitro, and more than 50% of CTCs were found to exist as tumor cell clusters in the blood circulation following orthotopic implantation. A minority of PDAC tumors undergoing full EMT exhibit primarily single-cell invasion phenotypes coupled with single CTCs in the blood circulation (69). Accordingly, intravital tumor imaging, live tumor slice culture imaging and 3D culture systems confirmed that hybrid E/M breast cancer cells could migrate and invade as clusters (67). RNA-sequencing analyses of CTCs collected from breast cancer patients showed that the CTCs population was formed by both hybrid E/M and fully mesenchymal cells (76) and, interestingly, although rare in the circulation compared with single CTCs, CTC clusters presented 23- to 50-fold higher metastatic potential in mouse models of breast cancer (77). Accordingly, patient-derived tumor xenograft models of colorectal cancer (CRC) showed that tumor cell clusters in hybrid E/M states seeded spontaneous metastases more often than single tumor cells (78). Furthermore, CTC clusters expressing both epithelial and mesenchymal markers can be readily detected in patients with advanced stages and correlated with poorer patient outcomes in various cancer types (8).

In conclusion, these studies demonstrate that tumor cells retaining both epithelial and mesenchymal characteristics can more readily respond to environmental signals and adapt to various stresses in the microenvironment during the metastasis journey. Studies discussed in this section highlight the importance of carefully defining the EMT status in individual tumors by evaluating the expression, subcellular localization and post-transcriptional/translational modifications of EMT regulators in combination with changes in cellular and functional properties. Moreover, analyzing the EMT states often requires the use of markers that are specific to each biological context (36, 60, 71). One open question in the cancer EMT field is to understand how stabilization of specific hybrid E/M states or dynamic and continuous switches between diverse E/M states in response to distinct microenvironmental signals function during individual steps of metastasis. Studying the EMT process at the single-cell level by using novel technologies, such as lineage tracing experiments, intravital live imaging, gene expression analyses and genetic and epigenetic manipulations, could help to address this question. Finally, a combination of interdisciplinary approaches from bioengineering to mathematical modeling will be important to gain a deeper mechanistic understanding of EMP in metastasis.

EMP IN METASTASIS DORMANCY AND RECURRENCE

Metastasis is the leading cause of cancer-related deaths (79). Notably, metastatic lesions may develop decades after primary cancer diagnosis and treatment, especially in breast and prostate cancer patients (80). This protracted period of latency is caused by the long-term survival of dormant malignant cells following their dissemination into distant organs, which persist as single cells or as small clusters in a clinically undetectable quiescent state (80). Metastatic dormancy is mediated by complex networks of tumor cell-intrinsic pathways and signals from the metastatic niche, which confer DTCs with selective advantage to endure several stresses including therapy resistance, host anti-tumor immunity and metabolic stress (81). Dormant DTCs may eventually adapt to the new tissue microenvironments and escape from their cell cycle arrest, leading to the outgrowth of macrometastases and disease relapse (81). Despite current efforts, the molecular mechanisms underlying metastasis dormancy and recurrence are poorly understood.

Metastasis is considered a highly inefficient process, as only a small fraction of DTCs survive the initial colonization step and manifest into overt metastases at distant sites (82). Importantly, the metastatic competency of DTCs has been linked to EMP, which is key for the adaptation of DTCs to new microenvironmental constraints, to overcome immune surveillance, and to exit metastatic dormancy and ultimately develop clinically detectable lesions (31). While the impacts of EMT on tumor cell dissemination have been extensively studied, a growing body of literature suggests that its reversal through MET is required to reinitiate the growth of dormant DTCs during metastatic colonization of distant organs (Figure 1). Histopathological evaluation of paired primary and secondary tumors revealed that the majority of metastatic lesions have epithelial morphology and high levels of E-cadherin expression resembling their primary tumor counterparts (12). In prostate and bladder cancer cell lines, constitutive EMT programs suppressed stem cell properties and reduced the metastatic potential of DTCs if not reverted (24). Bonnomet et al.(83) detected a downregulation of the mesenchymal marker vimentin in lung metastases opposed to its high expression in CTCs. Accordingly, loss of the epithelial marker E-cadherin promoted tumor cell invasion but increased the susceptibility to ROS-mediated cell death, thus reducing CTC survival, cell proliferation and metastatic seeding in human and murine models of invasive ductal carcinoma (IDC) (84).

Why do DTCs need to undergo MET to establish distant metastases? A number of previous studies revealed the functional consequences of the EMT program on cell cycle regulation. In 2004, Vega et al. (85) demonstrated that SNAI1 expression caused cell cycle arrest in MDCK cells and during embryonic development by inhibiting RB phosphorylation, repressing the transcription of cyclins D1 and D2, and maintaining an elevated expression of p21. In SCC cells, repression of cyclin D1 was modulated by direct binding of ZEB2 to its promoter region, causing cell cycle arrest at G1 phase (86). Later reports assessed the relevance of MET in dormant DTC reactivation and cancer recurrence by modelling the gene expression dynamics of EMT-TFs during multiple stages of tumor progression. Tsai et al. (55) demonstrated that inducing Twist1 expression in primary SCC tumors promoted EMT and tumor cell dissemination, but its reversal following metastatic seeding favored tumor cell proliferation at the distant sites and was essential to establish macro metastases. Similarly, continuous Snail1 overexpression promoted tumor cell dissemination but decreased metastatic competency in MMTV-PyMT mice (58). Overexpression of miR-200 induced Zeb1/2 silencing and Sec23a-mediated cytokine secretion, which in turn favored metastatic colonization by promoting re-expression of E-cadherin (87). In addition, loss of the paired-related homeobox transcription factor (PRRX1) in mesenchymal cancer cells induced MET and was associated with metastatic recurrence in mouse models and high-grade IDC patients (23). The role of PRRX1 in cell dormancy was further characterized using in vitro and in vivo models of head and neck SCC. Specifically, PRRX1 cooperated with TGF-β signaling to promote EMT and invasion, and sustained metastatic dormancy mediated by p38 and TGF- β2 through downregulation of miR-642b-3 (88). Takano et al. (89) described opposing roles for Prrx1 isoforms A and B in tumor progression. Prrx1 isoform B (Prrx1b) contributed to PDAC cell dissemination through Prrx1b-HGF axis, which is reported to promote EMT in several types of cancer (90). In liver metastases, expression of Prrx1b was only detected in small lesions and significantly decreased in larger proliferative metastases, whereas Prrx1a played opposite functions, thus stimulating MET and metastatic recurrence (89). The association between EMT/MET plasticity, metastatic dormancy and recurrence was also confirmed by single cell analyses of metastatic breast cancer PDX models demonstrating an enrichment in stemness, survival, EMT, and dormancy/quiescence signatures in DTCs from animals with low vs. high metastatic burdens (91). Taken together, these studies demonstrate that although EMT is relevant for tumor cell dissemination, it may also limit the expansion of dormant DTCs and recurrence by promoting growth arrest. In distal organs, reversal of EMT in dormant micrometastases endow DTCs with enhanced proliferation and is required to form overt metastases.

Recent studies added more complexity to the field by demonstrating that DTCs with intermediate E/M phenotypes display higher tumor initiating and metastatic potentials than fully mesenchymal cancer cells (30, 31, 69). The reason behind these observations is that hybrid E/M states have the most plasticity to switch between a continuum of epithelial-mesenchymal phenotypes, thus facilitating their ability to overcome environmental constraints and grow in secondary sites (8). Harper. et. al. (92) identified a hybrid population of Her2+ breast cancer cells with high Twist1 and CK8/18+, but low E-cadherin expression that were capable of disseminating at early stages of tumor development. Notably, early disseminated hybrid cells remained predominantly dormant in bone marrow and lungs and eventually initiated metastasis, which correlated with Twist1 downregulation and higher expression of E-cadherin. Using both mathematical and experimental models, Celia-Terrassa et al. showed that hysteretic EMT increased cellular plasticity and reduced E/M extreme phenotypes, enabling faster cellular response and enhancing metastasis (93). In line with these observations, tumor cells could reside in distinct populations of stable hybrid E/M states and fully reprogrammed mesenchymal cells in breast cancer mouse models. However, cells that completed EMT displayed decreased metastatic potentials, suggesting that a complete EMT renders DTCs at disadvantage for proliferation and colonization in distant organs (29). EMP signatures were also detected upon tumor cell dissemination in ER+ breast cancer models, which are clinically characterized by their late recurrence (94). In ER+ breast cancer, hybrid E/M tumor cells with high expression of mesenchymal genes and downmodulation of E-cadherin are associated with metastatic dormancy and low proliferation rates (94). Interestingly, single cell analysis of early disseminated breast cancer cells in MMTV-Her2 and MMTV-PyMT mice revealed that the transcription factor ZFP281 induced heterogeneous mesenchymal and pluripotency-like transcriptional programs that maintained early DTCs in a dormant state by preventing the acquisition of epithelial traits and cell cycle programs (95). Overall, these data support the notion that MET is required for the resurgence of dormant tumor cells to form macrometastases in distant organs.

Metastatic dormancy is a key step in tumor progression, often associated with stemness, chemoresistance and incurable recurrence (80). Interestingly, stabilizing the expression of mesenchymal programs to maintain DTCs in a dormant state has been proposed as a strategy to prevent metastatic resurgence in cancer patients (96). Despite its clinical relevance, our comprehension of the molecular mechanisms controlling metastatic dormancy has been limited by the lack of specific cellular and surface markers for dormant DTCs and the challenge to detect dormant micrometastases in either cancer patients or mouse models. Recent advances demonstrated that the awakening of dormant DTCs is modulated by a plethora of signaling molecules and cell types residing in the metastatic niche, which could contribute to MET (81, 97, 98). Moreover, MET could be consequence of the absence of EMT-inducing signals derived from primary tumor microenvironments in distant organs (7). Indeed, tumor cells expressing different levels of epithelial and mesenchymal markers are associated with specific tumor microenvironmental conditions (31). In this study, EpCAM+ hybrid E/M tumor cells maintained intermediate E/M phenotypes following subcutaneous transplantation but completely reverted to the epithelial phenotype after lung colonization, suggesting that different microenvironments can drive changes of EMT states. Metastatic microenvironments and dormancy could also be impacted by environmental factor such as smoking. For instance, chronic exposure to nicotine promoted breast cancer metastasis by creating a favorable environment for resurgence of dormant metastases in the lung. Specifically, nicotine induced the polarization of lung neutrophils towards pro-tumorigenic N2 functions, which promoted MET in DTCs through the secretion of a glycoprotein termed lipocalin 2 (LCN2) (99).

Conversely, some microenvironmental cues promote the maintenance of EMT and dormancy programs in DTCs as mechanisms of survival to microenvironmental stress. In CRC models, the COP9 signalosome 8 (CSN8) regulated EMT and dormancy through NF-κB/HIF-1α signaling to facilitate the adaptation of DTCs to hypoxic environments (100). During bone colonization, the osteogenic niche contributed to tumor cell plasticity and endocrine resistance by triggering adaptative epigenetic changes in disseminated ER+ breast cancer cells through activation of STAT3 signaling, EMT and stemness programs (101). Direct contributions of TGF-β, a potent inductor of EMT, to cell dormancy have been reported in several cancer types (98). In a head and neck SCC model, bone marrow-derived TGF-β2 promotes downmodulation of CDK4 and metastatic dormancy of DTCs via MAPK/p38 signaling (97). Similarly, NG2+/Nestin+ mesenchymal stem cells from the bone marrow perivascular niche secreted bone morphogenic protein (BMP7) and TGF-β2 to activate metastatic dormancy via SMAD, p38 and p27 signaling (95). Other mechanisms of communication between cancer DTCs and their metastatic niche include the release of exosomes containing microRNAs. Adipose mesenchymal stem cells (MSCs) induced dormancy and EMT in different human breast cancer cell lines via secretion of several miRNA (102). Bone marrow-derived exosomes containing miR-222 promoted EMT via Notch3 targeting, quiescence and therapy resistance in a subset of breast cancer DTCs (103).

While our understanding of metastatic dormancy remains in its infancy, research in the past decade clearly demonstrate a critical role of MET in promoting macrometastasis growth. Future studies could shed light on the exact MET-inducing signals and molecular mechanisms by which multiple factors, including MET, induce reactivation of dormant DTCs in distant organs. The goal is to unveil novel targets for therapeutic intervention of cancer recurrence, yet therapeutic strategies blocking EMT should be carefully considered due to the role of MET in macrometastasis growth. Targeting EMT could be beneficial to tackle tumor cell dissemination, but it could also reactivate the growth of dormant DTCs via MET. Instead, keeping disseminated tumor cells in the mesenchymal state to lock them into dormancy could be an alternative approach for patients with invasive carcinomas. Furthermore, uncovering novel molecular markers that predict recurrence risks for cancer patients with dormant DTCs could help to identify those patients that likely benefit from such therapies.

CLINICAL IMPLICATIONS OF EMP

Many works of exceptional significance have underpinned the essential role of EMP in tumor progression and metastasis. It is therefore not surprising that EMP is considered a potential target for developing effective strategies to prevent cancer recurrence. Although major challenges may arise from EMP-driven phenotypic heterogeneity, the clinical implications of EMP in cancer prognosis, treatment and therapy resistance have been profoundly revised in the past ten years.

EMP as a marker of cancer progression

As discussed in previous sections, activation of EMT in the primary tumor endow cancer cells with enhanced stem cell-like properties, acquired mobility, invasion and metastatic competency, which are critical for colonization and subsequent formation of overt metastases in distant organs. In numerous studies, acquisition of mesenchymal traits by malignant cancer cells correlates with increased tumor cell invasion, metastasis, therapy resistance, and poor patient survival. For instance, progression from non-invasive to muscle-invasive bladder cancer presented a strong association with the acquisition of EMT-related gene signatures (104). In addition, transcriptomic analyses revealed an increase of EMT gene expression in late stages of bladder cancer, which was associated with decreased survival and therapy resistance (105). EMT-related signatures had similar predictive values in many types of human carcinomas, including breast, hepatic, colorectal and gastric cancers (106–109).

The clinical significance of EMP and hybrid E/M states on survival outcomes are studied in various cancer cell lines, mouse models and patient cohorts. Huang et al. (110) classified an ovarian cancer library comprised of 43 different cell lines in four groups based on their E/M status, namely epithelial, intermediate-epithelial, intermediate-mesenchymal or mesenchymal signatures. Tumor cells at the intermediate-mesenchymal state presented enhanced resistance to anoikis, colony formation capacity and plasticity in vitro. Accordingly, ovarian and breast cancer patients with intermediate E/M signatures have poor prognosis and increased metastatic recurrence (111, 112). Interestingly, cutting-edge transcriptomic technologies were used to evaluate both tumor-intrinsic and microenvironmental pathways shaping EMP during tumor progression. Using a combination of high-throughput genomic approaches, Puram et al. (113) analyzed primary head and neck cancer samples and their matched lymph node metastases and uncovered a tight association between a partial EMT gene signature and metastasis. Similarly, Tagliazucchi et al. (114) interrogated the distribution and microenvironmental interactions of different EMT states (epithelial, hybrid E/M, and mesenchymal) in multiple cancer tissues and identified genomic and environmental hallmarks shaping the mesenchymal transformation of primary tumors. All these findings highlight the involvement of EMP and tumor cell heterogenicity in cancer progression. In the long run, understanding the clinical implications of specific E/M hybrid states and their control mechanisms in response to tumor microenvironmental signals could aid the identification of new biomarkers of recurrence for high-risk patients.

The presence of CTCs in the blood of cancer patients is often associated with poor prognosis, as it is indicative of the ability of malignant cells to disseminate into distant organs (115). Interestingly, molecular markers of EMT are enriched in CTCs, which correlates with increased recurrence in several human cancer types (116–118). Yu et al. (76) identified dynamic changes of epithelial and mesenchymal markers in CTC from breast cancer patients: whereas expression of mesenchymal genes was higher in CTCs from advanced cancer patients and reverse shifts between E/M states were detected after treatment, suggesting a close association between EMP and therapy responses. Cohen et al. (119) performed a longitudinal study of CTCs from breast cancer patients undergoing different types of therapies and demonstrated that CTCs from patients with disease progression shifted toward a mesenchymal stem-like phenotype after initiation of therapy. In line with previous observations, CTCs from patients with positive therapy responses had higher expression of epithelial markers. Additional interesting studies characterized nine CTC cell lines derived from one patient with metastatic colorectal cancer following cancer treatment and progression (120). Analyses of EMP markers in these cell lines showed expression of both epithelial and mesenchymal signatures, revealing an increased plasticity that is associated with metastasis (120). Overall, assessing the EMT status in both primary tumors and CTCs following treatment and progression could yield better treatment choices and improve survival of patients with cancer recurrence.

EMP and therapy resistance

Multiple lines of evidence suggest that dynamic changes in E/M states contribute to the generation of heterogeneous tumor cell populations with varied drug sensitivities (121). Importantly, acquisition of mesenchymal traits by tumor cells is often detected in clinical samples after standard cancer treatments, including different combinations of chemotherapeutic agents, radiotherapy, and targeted therapies. In breast, prostate and rectal tumors, EMT-related gene expression signatures were enriched after treatment with neoadjuvant therapies (122–124). Associations between EMT and resistance to radiotherapy, hormone therapy, and immunotherapy were also observed (125–127). For instance, SCC lines that become resistant to chemotherapeutic agents, such as paclitaxel and cisplatin, present high cellular plasticity and stem-like characteristics. These hybrid E/M subpopulations, defined by high levels of stemness markers (CD44/CD24) and low levels of epithelial markers (EpCAM), exhibited enhanced resistance to therapy when compared to fully epithelial or mesenchymal cells (128). Tissue samples from castration-resistant prostate cancer patients showed increased hybrid E/M phenotypes with co-expression of SNAI1 and Cytokeratin (129). Moreover, multiple signaling pathways regulating EMP-induced therapy resistance were described in breast cancer models. Bierie et al. (35) identified a population of ITGB4-positive cancer stem cells with intermediate E/M phenotypes linked to a decreased relapse-free survival in patients receiving chemotherapies. In addition, Wnt/β-catenin signaling induced a partial EMT, marked by expression of both mesenchymal and epithelial markers, and contributed to multidrug resistance in different cancer cell lines (130, 131). Taken together, these data suggest that hybrid E/M tumor cells present a strong association with therapy resistance, possibly by their increased ability to survive various stresses from therapies.

Uncovering the molecular basis underlying therapy resistance is critical for developing targeted therapies against resistant cancer cells. Therefore, it is of utmost importance to understand the mechanisms by which EMP endows tumor cells with the ability to overcome diverse types of cancer therapies. Several mechanisms underlying EMP-induced therapy resistance have been described (121) and are summarized in Figure 3.

Figure 3. Mechanisms underlying EMP-driven therapy resistance.

Activation of EMT programs confers therapy resistance through several mechanisms. EMT confers resistance to cell death by decreasing apoptosis mediated via NF-κB and PI3K/AKT signaling pathways, and anoikis. In addition, EMT mediates the acquisition of stem-like characteristics and reduced proliferation through cell cycle arrest. EMT decreases drug intracellular levels by enhancing the expression of ABC transporters. Notably, EMT promotes resistance to targeted therapies through aberrant signaling activation, such as SNAI1-mediated overexpression of Axl, which activates PI3K/Akt signaling and overrides EGFR inhibition by tyrosine kinase inhibitors (TKIs). EMT contributes to immunotherapy resistance via several potential mechanisms, including enhanced immune checkpoint protein expression, increased autophagy to escape CTL-mediated lysis, loss of IFN-γ signaling, and defective antigen presentation.

First, EMT confers resistance to cell death by downmodulation of pro-apoptotic signaling pathways. For example, SNAI1 increased the survival of tumor cells to different types of therapies through activation of multiple signaling pathways, including NF-κB and PI3K/AKT, which antagonize p53-mediated apoptosis (85, 132, 133). Similarly, SNAI2 suppressed the expression of PUMA, a pro-apoptotic gene regulated by p53, promoting cisplatin resistance in NSCLC cells (134). In lung cancer cell lines, CDH1 depletion decreased apoptosis signaling via TRAIL and the death receptors DR4 and DR5 (135). Interestingly, EMT plasticity also increased tumor cell resistance to detachment-induced cell death or anoikis. Ovarian carcinoma cell lines with hybrid E/M properties presented increased anoikis resistance and spheroid-forming ability in vitro (110). In the same study, restoring E-cadherin expression and epithelial phenotypes via Src kinase inhibition abrogated spheroidogenesis, thus decreasing the metastatic potential of tumor cells.

Second, activation of EMT promotes cell cycle arrest and reduces tumor cell proliferation (85, 86). In turn, decreased proliferation protects primary tumor cells and DTCs against chemotherapeutic agents that target cell division. Through the reversal of EMT, dormant disseminated tumor cells may reactivate their proliferation into overt metastases years after initial successful treatment.

Third, EMT promotes therapy resistance through the acquisition of stem cell-like properties. CSCs have increased tumor-initiation capacity and are responsible for generating intratumor heterogeneity and driving cancer recurrence (136). Moreover, CSCs display enhanced resistance to chemotherapy when compared to more differentiated tumor cells and are enriched after conventional cancer treatments (137). The link between stemness, EMP and chemoresistance was addressed in both preclinical and clinical studies. Both CSC-like properties and EMT were observed in tongue SCC cells with acquired cisplatin resistance, mediated by the FZD7/Wnt/β-catenin pathway (137). In vitro, loss of the E3-ubiquitin ligase FBXW7 promoted EMT via Zeb2 stabilization, which conferred CRC cells with enhanced migration, stemness and chemoresistance (138). The transmembrane protein Metadherin (MTDH) promoted TWIST1-mediated EMT, which was linked to the acquisition of CSC traits and drug resistance in breast cancer cell lines and patient specimens (139).

Last, EMT modulates expression of drug transporters and cellular receptors to gain therapy resistance through unique mechanisms, including decreasing drug intracellular levels and acquiring protection against targeted agents. Notably, EMT promoted expression of multiple ATP-Binding-Cassette (ABC) transporters responsible for multidrug resistance, which increased drug efflux and resistance to doxorubicin in breast cancer cells (140). In addition, EMT increased tumor cell resistance to targeted therapies by aberrant activation of alternative signaling molecules. Using in vitro and in vivo models of NSCLC, Zang et al. (141) demonstrated that protection against EGFR tyrosine kinase inhibitors (TKIs) was mediated by EMT through activation of the receptor tyrosine kinase Axl. In a similar study, Byers et al. (142) developed an EMT signature to predict resistance to EGFR and PI3K/AKT inhibitors in NSCLC patients, which showed a higher resistance to targeted therapies in mesenchymal tumor cells. In mouse models, cells with a mesenchymal phenotype presented increased levels of Axl and were more sensitive to Axl inhibition. Moreover, combined treatment with Axl and EGFR inhibitors improved therapy response in mesenchymal tumors that are resistant to EGFR inhibition alone.

Notably, recent studies also addressed the impacts of EMT in acquired resistance to immunotherapy, a topic that has been extensively reviewed by experts in the field (143). Briefly, EMP modulates tumor cell response to immunotherapy through multiple tumor-intrinsic mechanisms. For instance, EMP signatures were linked to increased expression of immune checkpoint proteins, including PD-1, PD-L1, PD-L2, CTLA and TIM3, in multiple types of cancer (144). Mesenchymal cancer cells could also become insensitive to IFN-γ signaling via IRF1 downmodulation, thus overcoming IFN-γ-mediated anti-tumor effects induced by immune checkpoint blockade therapy (145). Other rising mechanisms of EMP-mediated immunotherapy resistance include increased tumor cell autophagy to escape CTL-mediated lysis (146), and defective antigen presentation, marked by SNAI1-dependent downmodulation of MHC-I and HLA-I downmodulation (147, 148). In addition, EMP contributes to the recruitment of immunosuppressive microenvironments through increased secretion of specific signaling molecules (143). Whether some of these mechanisms, such as decreased antigen presentation, are directly regulated through EMT or are induced by EMT-driven stemness needs further investigation.

EMP has emerged as a key driver of cancer progression and therapy resistance in many tumor types. Therefore, therapeutic targeting of EMT has been considered a promising approach to prevent recurrence and overcome therapy resistance. The potential strategies to target EMT are divided in four categories: 1) blocking signal transduction pathways promoting EMT, 2) downmodulating the expression or inhibiting the function of EMT-TFs, 3) targeting invasive mesenchymal cells, and 4) locking cells in dormancy by blocking MET (149). Despite all the progress made over the past decade, especially in the development of pharmacological approaches to inhibit paracrine factors that induce EMT, concerns are raised for the use of anti-EMT therapies in the clinical setting. As discussed above, inhibition of EMT could reactivate proliferation of dormant DTCs, thus promoting the growth of macrometastases in patients carrying clinically undetectable micrometastases. Moreover, EMT could provide the plasticity to give rise to heterogeneous tumor cell populations with diverse drug sensitivities, thus promoting therapy resistance. Combining compounds to target EMP with standard anti-proliferative cancer therapies could reduce tumor heterogeneity, dissemination, and improve therapy response in cancer patients. In addition, future research efforts may identify new EMT markers in primary tumor cells and CTCs to improve patient stratification and prediction of response to various therapeutics, ultimately improving the survival of patients at risk of cancer recurrence.