Cell-state plasticity is the underlying mechanism through which cancer cells undergo dynamic, non-genetic changes contributing to their proliferation and metastasis. Cancer cells can be thought to exist within a spectrum of phenotypic states connected by trajectories defining dynamic transitions within a cancer cell-state landscape. At one end of this spectrum, are the metastasis-seeding cancer stem cells (CSCs) with invasive mesenchymal properties critical to tumour initiation and therapy resistance. Non-CSCs generated through differentiation of CSCs are progeny cells with epithelial properties sitting on the other end of this spectrum. The dynamic cell-state transitions between these extremes are regulated by intrinsic programs such as the epithelial-to-mesenchymal transition (EMT) and the reverse mesenchymal-to-epithelial transition (MET). This study aims to define the dynamic transitions that drive CSCs into a chemotherapy-sensitive non-CSC-like state through the MET program in the cancer cell-state landscape.
We developed a neural ODE network called MIOFlow that learns continuous dynamics from static transcriptomic data and applied it to time-lapsed single-cell expression measurements from an in-vitro triple-negative breast cancer MET differentiation system. This led to the identification of core transcriptional factors associated with the emergence of the epithelial trajectory (non-CSC state). We constructed a gene regulatory network to sketch the transcriptional circuitry underlying the MET. The regulatory effect of one gene, the estrogen-related receptor alpha (ESRRA) was validated using orthogonal approaches including RNAi, western blotting, and immunofluorescence. Indeed, the knockdown of ESRRA increased epithelial cell state marker CDH1 (E-cadherin) expression. The downregulation of ESRRA within tumorsphere was validated temporally, together with the concomitant downregulation of mesenchymal cell-state marker ZEB1 and upregulation of CDH1. Furthermore, inhibition of ESRRA in triple-negative breast cancer cells prevented metastatic seeding in lungs via in vivo tail vein injections.
MIOFlow, thus presents an innovative approach to characterizing the dynamic molecular programs derived from time-point-based transcriptomic datasets. Using this approach, we have defined the key transcription factors and validated ESRRA as a key driver of MET in the cancer which have broad implications for translational research aimed at driving CSCs out of their aggressive state to inhibit metastasis and chemotherapy-resistant disease.