Intra- and inter-cellular drivers of Epithelial-Mesenchymal heterogeneity in cancer cells

Abstract

Phenotypic heterogeneity is a fundamental feature reported across prokaryotic and eukaryotic cells. This heterogeneity enables an isogenic cell population to be composed of distinct subpopulations with varying functional traits, such as migration, proliferation, and stemness, and varied susceptibility of cells to a drug. Consequently, the cell population can exercise bet-hedging, leading to higher fitness in the face of varying environments. In cancer, phenotypic heterogeneity also manifests as a higher propensity for cells to metastasize, thus aggravating the clinical progression of the disease. However, the cellular processes responsible for initiating and maintaining such cellular heterogeneity have been poorly investigated in cancer.

In carcinomas, a well-studied axis of phenotypic heterogeneity is along the Epithelial (E)-Mesenchymal (M) spectrum. Cancer cells can undergo E-to-M and M-to-E transitions either spontaneously or in response to microenvironmental changes such as hypoxia. These bidirectional transitions – collectively referred to as Epithelial-Mesenchymal-Plasticity – lead to dynamic phenotypic heterogeneity on the timescales of days to weeks, as observed across cancer types in vitro and in vivo. Despite such longitudinal experimental data, the cellular processes that can explain these dynamic heterogeneity patterns remain elusive. This thesis focuses on identifying cellular processes that drive E-M phenotypic heterogeneity seen in cancer cells.

First, PMC42-LA cells were found to contain 80% EpCAM-high (epithelial) cells and 20% EpCAM-low (mesenchymal) cells. This E: M ratio of 80:20 was recapitulated over eight weeks when individual E and M populations were isolated and cultured independently. Such slow convergence to the parental phenotypic distribution led us to hypothesize the role of asymmetric cell division in enabling E-M plasticity. Through a mechanism-based mathematical model tracking the dynamics of a population over multiple cell divisions, we demonstrated that an asymmetric distribution of key molecular regulators of E and M states during cell division, i.e. from a parent to two daughter cells, can result in E-M plasticity and explains the dynamic heterogeneity patterns observed for PMC42-LA cells.

Second, experimental data in MCF10A cells report that the duration of induction of E-to-M transition through TGFβ governs the rate at which cells recover to an epithelial state post-TGFβ withdrawal. While TGFβ treatment had been associated with chromatin-based epigenetic changes leading to a seemingly irreversible transition, how time-varying chromatin-based epigenetic changes over multiple cell generations can impact M-to-E transitions remained elusive. We developed a phenomenological model of chromatin-based epigenetic regulation of an E gene (MIR200) by an M transcription factor (ZEB) and showed that the turnover rates of chromatin changes caused during TGFβ withdrawal determined the timescales of recovery of an Estate.

Lastly, in HCC38 cells, the E and M subpopulations were shown to promote M-to-E and E-to-M transitions, respectively, in the co-culture experiments, indicating the role of cell-cell interactions in shaping the dynamics of heterogeneity. We developed a set of mathematical models, each capturing different mechanisms such as variable growth rates, phenotypic transitions, and the impact of cell-cell interaction through changes in growth/transition rates. Our comparative analysis based on simulation and fitting of the different models to the experimental data elucidated that phenotypic transitions were necessary to capture the observed E and M cell fractions dynamics. Further, the role of E and M subpopulations in impacting each other’s growth rate led to better fits with experimental time-course data.

Overall, this thesis unravels different mechanisms that can qualitatively or quantitatively explain the evolution of E-M heterogeneity reported experimentally in breast cancer cells. It also lays the ground for future experiments to be designed to test the model predictions.