Varying susceptibility of subpopulations along the epithelial‐mesenchymal spectrum to undergo EMT

Abstract

To decode the response of a population to an external stimulus, information about the composition of the population and how the constituent subpopulations respond to the stimulus is essential. In their study, Sekeroglu et al. highlight how seemingly identical clones, based on cell‐surface E‐cadherin levels, derived from MCF10A cells exhibit varying levels of susceptibility to undergo epithelial‐mesenchymal transition (EMT) upon treatment with an EMT inducer. Highlighting the transcriptional and epigenetic differences among the clones driving these distinct outcomes, the study emphasizes the need to characterize tumor phenotypic heterogeneity across multiple regulatory layers to assess these functional differences.

Cells in an isogenic cancer population can show distinct extents of induction of epithelial‐mesenchymal transition (EMT) to the same stimulus. Such nongenetic cell‐to‐cell variability originates from differences in many cellular‐level processes being modulated to varying degrees among individual cells.

Abbreviations

EMT

Epithelial‐Mesenchymal Transition

OE

over expression

SCC

Single Cell Clone

Introduction

Advanced molecular profiling of patient tumor samples has led to characterizing phenotypic heterogeneity within tumors [1]. The presence of intratumor phenotypic heterogeneity is often associated with increased invasiveness, higher immune evasion, and drug resistance [2]. Such phenotypic heterogeneity is shaped by several micro‐environmental factors such as spatial variation in oxygen levels and/or extracellular matrix (ECM) density [3, 4]. However, recent in vitro studies suggest that even in a constant environment, there exist distinct subpopulations within the drug‐naïve population that survive drug exposure to different extents and contribute to drug resistance [5]. Such differences among subpopulations are shown to have non‐genetic origins, with subtle molecular differences, as cells in the population reversibly enter a rare ‘pre‐resistant’ state that expresses genes that are markers of the resistant state. Further, pre‐treatment of drug‐naïve cells with chromatin‐level modifiers alters their adaptive abilities [6, 7]. Thus, differences in molecular and epigenetic states among cell subpopulations can determine their initial and long‐term responses to drug exposure.

Recent efforts have identified different cancer cell subpopulations along the epithelial‐mesenchymal (E‐M) phenotypic axis and mapped their functional implications. In breast cancer, the extent of E‐M heterogeneity is usually higher in basal and triple negative breast cancer cell lines as compared to luminal ones [8, 9]. Further, generating single‐cell clones from heterogeneous SUM149T mouse mammary cancer cells can lead to distinct clonal populations that span the E‐M heterogeneity spectrum but can be significantly different from the parental population and have distinct growth and invasion characteristics in vivo [1]. Further, E‐M subpopulations exhibit distinct temporal responses in AKT and ERK signaling upon stimulation with epidermal growth factor (EGF) [10]. Thus, distinct E‐M subpopulations may exist within a parental population, and such heterogeneity contributes to the overall population‐level response to external stimuli.

EMT‐responsiveness based on intra‐ and extracellular factors

The work presented by Sekeroglu et al. [11] in The FEBS Journal further characterizes the E‐M heterogeneity and their molecular and functional differences using MCF10A breast cancer cells as the model system. MCF10A cells were found to be distributed along the E‐M spectra, as measured by E‐cadherin intensity and single‐cell RNA‐seq data analysis. Moreover, upon treatment with TGFβ3 (an EMT inducer), not all cells underwent a canonical EMT even after 8 days, with some cells retaining their E‐cad levels, albeit now in cytoplasmic regions. To investigate how any pre‐existing differences among the E‐M states can shape distinct responses to TGFβ3 treatment, the authors generated several clonal cell lines by bulk sorting E‐cad‐positive and E‐cad‐negative cells from the parental population and by generating their single‐cell clones. Interestingly, regardless of the starting E‐cad level(s), the clonal populations exhibited an E‐cad distribution similar to that of the parental cell line. However, these seemingly identical clonal populations showed distinct responses to TGFβ3 exposure when evaluated for a panel of epithelial and mesenchymal markers at the transcriptional level. While all clonal populations showed upregulation of the mesenchymal program with concomitant downregulation of the epithelial program, a single‐cell clone (SCC) originating from an E‐cad‐positive cell (Pos‐D9) showed unexpected upregulation of CDH1 (gene encoding E‐cad). This E‐cad upregulation was accompanied by GRHL3 upregulation, which plays a potent role in stabilizing the epithelial state; additional experiments with GRHL3 overexpression in parental MCF10A cells further prevented downregulation of CDH1 despite TGFβ3‐induced upregulation of mesenchymal genes. This potential stabilizing role of GRHL3 on E‐cad was also evident when tracking simultaneous changes in EpCAM and Ecad levels in parental MCF10A cells upon TGFβ3 exposure. A dominant population with the EpCAM(neg)/Ecad (pos) state emerged upon treatment, with transcriptional states spread across the E‐M spectrum, showing varying degrees of resistance to full EMT. Further, Pos‐D9 SCC, with its reduced responsiveness to EMT inducer than parental MCF10A cells, showed higher metastatic potential upon injection in the yolk sac of 2‐day post‐fertilization (dpf) embryos. Thus, heterogeneity in cell‐intrinsic factors, which is often not apparent with individual cell‐surface markers, leads to distinct cellular and functional outcomes.

Similar differences in E‐M subpopulation responsiveness to EMT induction have been observed in other studies using MCF10A cells [12] and with human mammary epithelial (HMLE) cells [13]. EMT induction in HMLE cells for 21 days led to a loss of EpCAM expression in most cells in the population; however, a small fraction remained positive for EpCAM. Upon sorting these EpCAM (pos) and EpCAM (neg) cells and providing a 10‐day holiday from EMT‐inducing signals, it was observed that EpCAM‐negative cells retained their mesenchymal phenotype, as determined by morphology and levels of canonical genes. In contrast, EpCAM‐positive cells preserved their epithelial profiles. Further, while both EpCAM (pos) and EpCAM (neg) cells continued to grow in two‐dimensional culture conditions, only EpCAM (pos) cells formed branching organoid structures. Moreover, isolation of single‐cell clones (SCCs) from HMLE cells yielded two clonal types: one responsive to EMT induction (M‐SCC) and the other nonresponsive to EMT (E‐SCC). Analysis of gene expression and global chromatin changes among the two clones during EMT induction revealed the potential role of transcription factors ZEB1 and GRHL2 in EMT responsiveness and reversibility. While ZEB1 overexpression (OE) in E‐SCC led to loss of EpCAM and downregulation of epithelial genes, only knockdown of GRHL2 in E‐SCC stabilized the ZEB1 OE‐induced mesenchymal state. Similar existence of EMT responsive and non‐responsive subpopulations, identified based on EpCAM status, was observed across several other breast cancer cell lines and metastatic effusions of stage IV patients, with EpCAM (pos) (with their higher EMT non‐responsiveness) leading to the growth of tumors upon orthotopic injections in mice and higher metastasis to the lungs.

Apart from the role of cell‐intrinsic GRHL3 levels in regulating and differentiating responses across subpopulations, Sekeroglu et al. also highlighted the role of an extracellular factor, forskolin, in regulating cAMP signaling. Forskolin is a cAMP‐activating supplement used in the culture media to grow MCF10A cells. The authors found that while the absence of forskolin did not affect cell growth or apoptosis, it significantly reduced cellular response to TGFβ3 treatment, with about 50% of cells maintaining their EpCAM(pos)/Ecad(pos) status, compared to 16% in the presence of forskolin. Further, forskolin's absence also reduced the transcriptional suppression of CDH1 and the activation of VIM and ZEB1 upon TGFβ3 treatment. Counterintuitively, the absence of forskolin led to lower levels of E‐cad protein in the population compared with its presence upon TGFβ3 treatment, suggesting a potential role for forskolin in post‐translational stabilization for E‐cad. Overall, forskolin's absence reduced the strength of coordinated response to TGFβ3 treatment, suppressing epithelial genes and activating mesenchymal genes, thereby reducing susceptibility to undergo EMT.

Conclusion

Phenotypic heterogeneity among cells can exist across multiple levels—cell surface marker, mRNA and protein levels, and chromatin states, such that two subpopulations that seem similar at one level might be significantly different at other levels (Fig. 1A). Such differences at multiple levels can lead to distinct, functionally significant responses among cells when exposed to the same stimuli (Fig. 1B). Current efforts have helped us understand how differences at one regulatory level contribute to different functional outcomes. However, how different levels of cell‐state regulation interoperate and manifest in observed heterogeneity patterns remains an open area of research. Future endeavors are to understand this emergent crosstalk, with the aim of predicting plausible population responses to external stimuli.

Fig. 1.

Manifestation of cellular heterogeneity at multiple levels and resulting differences in cellular response to external stimulus. (A) The binary grouping of cells in a population based on the presence (positive) and absence (negative) of a cell‐surface marker could be further divided into multiple subpopulations based on the high‐dimensional internal state of the cells defined by the mRNA and protein levels and chromatin structure. (B) The multi‐dimensional origins of heterogeneity within the population lead to distinct cellular responses upon treatment with an external stimulus.

Conflict of interest

The authors declare no conflict of interest.

Author contributions

PJ and MKJ wrote the manuscript.