The Continuing Search for Causality between Epithelial-to-Mesenchymal Transition and the Metastatic Fitness of Carcinoma Cells

Epithelial-to-mesenchymal transition (EMT) in carcinoma has seen its definition refined, from basic phenotypic characterization that assumed invasive potential, to heterogeneous, complex, and dynamic transcriptomic profiles with presumed functional consequences on the metastatic potential of cancer cells (1). The proposed invasive potential of carcinoma cells exhibiting EMT features is ascribed to the dissolution of intercellular junctions and subsequent loss in basal-apical polarity, remodeling of cytoskeleton promoting motility, and transcriptomic changes driven by core transcription factors (TF) and regulators. Owing in part to single-cell RNA (scRNA) sequencing studies, the program is now better understood as a collection of EMT transitional states or E/M hybrids with a spectrum of partial EMT programs. While more encompassing a definition, the newly appreciated heterogeneity of EMT is, in part, used to explain the challenges that remain in capturing these cells in the metastatic cascade and in identifying EMT rate-limiting functions. EMT in carcinoma cells may be a driver, or alternatively an adaptative response to tumor microenvironmental changes, impacting features of cancer stemness as well as chemoresistance and immune resistance. EMT can be associated with various aspects of cancer progression; however, a definite role for this complex collection of cellular transition states in metastasis remains challenging to ascertain. Here we offer a perspective on the causal relationship between EMT and cancer metastasis and discuss the evolution of the concept of EMT to better understand metastatic fitness.

EMT, alongside its reversal process, MET (mesenchymal-to-epithelial transition), has been schematized in numerous biological illustrations. In developing embryos, EMT-MET has been functionally established. In experimental models of carcinoma, however, defining the rate-limiting function of EMT-MET remains a challenge, and in need of more relevant models to discern which carcinoma cells contribute to circulating tumor cells, and/or disseminating tumor cells. Despite many efforts to favor this plastic cellular transformation for metastasis, the experimental capture of the execution of EMT-MET process in metastatic disease has not yet been realized. In many ways, the cartoon depiction of carcinoma cells undergoing EMT, and subsequently MET for metastatic nodules to become appreciable (clinically or in experimental system), has replaced the burden of proof when affirming a causal relationship between EMT and metastasis. Experimental “windows” into the process of EMT in vitro and in primary tumors have been amply documented, together with reports of circulating cancer cells with mesenchymal and epithelial features, giving support for acquisition of EMT for invasion into systemic circulation. However, confirmation that such cells seed the metastatic site, and possibly grow to form secondary tumors via MET, is lacking. The limitation for such capture rests primarily in the dynamism of EMT-MET, wherein MET effectively erases the evidence of EMT in metastasis. Lineage tracing studies using complex genetically engineered mice (GEM) in both pancreatic and breast cancer models offered a novel outlook into fate-mapping EMT cells in metastasis (2, 3). These studies however showed that metastatic nodules were deprived of cancer cells that would have been labeled to show acquisition of an EMT program in their lifespan. In GEM with spontaneous and metastatic pancreatic ductal adenocarcinoma (PDAC), EMT was readily captured in primary tumor when expression of mesenchymal genes αSMA and FSP1 were lineage traced, and such cells were seen in distant sites (lungs), yet the bulk of metastatic disease (macrometastases) did not demonstrate expression of either of the mesenchymal markers tested (2). A similar approach was used in two distinct GEM of breast cancer metastasis, using Vimentin and FSP1 as mesenchymal markers to lineage trace EMT, and again metastases were deprived of such EMT-traced cells (3). Similarly in the context of breast cancer GEM, vimentin lineage tracing of EMT failed to capture metastatic nodules, whereas N-cadherin did (4). Notwithstanding the elegance of the N-cadherin lineage tracing strategy (4), this approach may have captured collective migration of epithelial cells. Further studies are needed to ascertain how these mesenchymal markers fit in the evolving EMT transcriptomic signature. Recent observations using bar coding of aggressive murine PDAC cells subsequently seeded into the pancreas mice showed clonal bottlenecking for metastases, supporting metastatic fitness may result from a continuous selective process (5). The study also informed that metastases from two mice at a single timepoint displayed scRNA sequencing profiles enriched in a late-hybrid EMT state, though it is unknown whether this is rate limiting, and whether this reflects an initiating event or feature of established metastatic nodules. It should also be noted that in all lineage tracing approaches, one may not be able to differentiate whether EMT occurring in the primary tumor give rise to colonizing cells, or whether secondary nodules, formed from the seeding of primary tumor epithelial cells, subsequently gain dynamically mesenchymal or stem cell–like gene expression profile to survive their new environment. This chicken and egg situation could be vexing to experimentalists. To date, the data from attempts to trace the EMT-MET process in metastasis are not supportive of its absolute requirement. Our evolving understanding of cancer metastasis has brought into focus the complexity of cellular behaviors in the primary tumor, with cancer cell genetic drivers in a balancing act with the tumor microenvironment, resulting in numerous survival and therapeutic resistance pathways. It is also noted that dedifferentiated tumors that are intrinsically mesenchymal-like (e.g., basal and squamous carcinoma) may efficiently metastasize by mechanisms that differ from the prototypical EMT program. This may also challenge our visualization of EMT in the metastatic cascade. It would be prudent to postulate that metastasis likely emerges via numerous pathways, including non-EMT pathways such as collective migration of epithelial cells.

EMT has been observed by many in both animal models of cancer as well as in human tissues, and more often than not, these observations have been correlated with poor outcome though this may differ in distinct cancer types, poor response to therapy, disease recurrence, and increase metastatic burden in mouse models (as reviewed previously; ref. 1). Numerous reports have shown correlative association between the observation of EMT, metastasis, and chemoresistance; and EMT has been reported to precede metastasis in GEM of pancreatic cancer. Overexpression or downregulation of expression of TFs described as central to the transcriptomic changes noted in EMT cells have shown, largely in orthotopic settings, to impact metastatic disease in the expected direction. However, it is also widely appreciated that the genomic landscape of cancer cells likely plays a role in their metastatic potential. A simplistic view of this phenomenon is shown in isogenic mouse cancer cell lines with various metastatic potential, despite noncorrelated expression of EMT TF. While loss-of-function and gain-of-function studies suppressing the core EMT TFs (Twist1, Snail1, Zeb1) in cells prior to introduction in mice is informative, such experiments are limited in enabling the evaluation of the rate-limiting function of spontaneously arising EMT in carcinoma. Any ex vivo manipulation is unlikely to mimic the now known heterogeneity of EMT in spontaneously arising tumors. This notion is particularly critical given primary tumors evaluated for spontaneously arising EMT cells show that these cells are a minority population. Intravenous introduction of cancer cells previously enriched for mesenchymal-like versus epithelial-like features hardly supports a functional role for EMT in metastasis, but rather corroborate notions that certain phenotypes may be more resilient and adaptive or stem cell like.

To refine the rate-limiting role of EMT in GEM with spontaneously arising cancer, genetic tour de force experiments were pursued. A clever approach using inducible Twist1 expression showed squamous cell carcinomas had increased metastatic disease when Twist1 was turned “on” in the primary tumor but turned “off” at the secondary site (6). However, continuous Twist1 expression (thereby limiting MET) was not sufficient to abrogate metastasis, and metastasis occurred despite any manipulation of Twist1 expression. Furthermore, primary tumor burden was increased when Twist1 expression was induced, challenging interpretations of increased metastatic disease. Conditional deletion of Twist1 or Snail1 in PDAC GEM showed significant suppression of EMT in primary tumors and a significant reduction in chemoresistance, yet no changes in metastatic burden was noted (7). Interestingly, while compensation by other TFs was suspected because suppression of EMT was not sufficient to limit metastasis, such putative compensation was not needed to impede chemoresistance. Notably, conditional deletion of both Twist1 and Snail1 failed to suppress metastatic burden (8). Zeb1 conditional deletion in PDAC GEM did suppress metastasis by 50%; however, this was in the context of a 50% reduction in primary tumor burden (9), questioning whether reduction of metastasis merely reflected a slower emergence and progression of primary tumors. Indeed, the conditional deletion of Zeb1 failed to suppress metastasis in PDAC GEM (8). In breast cancer GEM with EMT lineage tracing capability, overexpression of miR-200 showed impact on chemoresistance, but failed to suppress metastasis (3). In contrast, conditional deletion of CDH2 (N-cadherin) reduced metastatic burden (4), though it was not reported what was the impact on primary tumor growth kinetic. The findings negating a rate-limiting role for EMT in metastasis generated discussion, and, perhaps for the better, renewed the interest in deciphering the distinct functions of EMT for chemoresistance versus metastasis. Such discussion should not merely continue to shift the goal post for characterization of EMT every time data do not fit the often used cartoon version of EMT in metastasis, but rather evolve toward refining what cancer cell plasticity means in the context of metastasis.

Given EMT is so dynamic and heterogeneous, can it remain a therapeutic target? Should specific TFs at the core of EMT be shown as dispensable for metastasis, it does not necessarily follow that EMT is not important for cancer progression and response to therapy. Recent focus on resolving the hybrid EMT transcriptome and phenotypic plasticity may offer novel understanding of EMT and therapeutic avenues. Though therapeutic opportunities for EMT may be challenging to reach given the heterogeneity of the program and shared pathway activation in other stromal cells in the tumors, they may be synergistic with targets of cancer genetic driver. A successful “EMT targeting” therapy may see its benefit by limiting chemoresistance. EMT signature has been shown to provide valuable insight into cancer progression, survival, and treatment response. While the lack of metastasis-promoting mutations does not favor genetic selection for metastasis, recent studies on metastases of PDAC in mice showed a significant role for non-genetic drivers of metastasis, supporting transcriptomic adaptation for metastasis fitness (5). These exciting results energizes research that aim to decipher transcriptomic changes necessary for growth of nodules at secondary site, a pursuit that may be more fruitful than debating whether EMT-MET program partake in every step of the metastatic cascade. In this regard, epigenetic mechanisms underlying transcriptomic adaptation of metastatic cells may become rate-limiting targets for metastasis. Keeping in mind the enormous heterogeneity of EMT transcriptomes and their adaptation to EMT TF suppression, metastatic fitness may simply reflect mesenchymal and mesenchymal-like cells transcriptomic variability (10). Metastatic fitness and EMT may thus converge and reconcile through the lens of transcriptomic variability giving rise to adaptative phenotypes under a dynamic and continuous selection process.

J.P. Thiery reports other support from Biocheetah Pte Ltd Singapore and Lion TCR Pte Ltd Singapore and personal fees and other support from Biosyngen Pte Ltd Singapore outside the submitted work. V.S. LeBleu reports personal fees and other support from Stellanova Therapeutics outside the submitted work.