Overcoming therapeutic resistance is one of the major hurdles in cancer care. One mechanism contributing to therapeutic resistance is a process in which epithelial cells switch to a mesenchymal state (epithelial-to-mesenchymal transition or EMT). The precise mechanisms driving EMT-mediated therapeutic resistance have, however, not been elucidated.

Here, we study ten cell line pairs, for which parental cell lines were made resistant to either a targeted or chemotherapy-based treatment. First, we show by miRNA-200 overexpression that treatment resistance is driven by EMT. Next, we demonstrate that DNA methylation changes occur within each cell line pair and show that exposure to 5-azacytidine or knock down of DNA methyltransferases (DNMTs), both of which globally demethylate cells, result in EMT reversal and increased therapeutic sensitivity. This suggests DNA methylation to causally underlie EMT and treatment resistance. We also observe significant overlap in methylation profiles between resistant lines, suggesting a common epigenetic mechanism to cause resistance to therapy. In line with this hypothesis, cross-resistance to other targeted and chemotherapies is observed, while importantly, this is lost upon demethylation of the cells. Finally, we clinically validate that DNA methylation changes drive EMT-mediated resistance to sorafenib in patients with advanced hepatocellular carcinoma (HCC). Specifically, we develop a capture-based protocol to interrogate DNA methylation in low amounts of circulating tumor DNA (ctDNA). By interrogating the methylation status in liquid biopsies, longitudinally collected during sorafenib treatment, we assess whether DNA methylation changes also drive EMT and therapy resistance in a clinical setting. Particularly, by monitoring methylation changes in EMT genes, we are able to predict tumor response and acquired resistance to sorafenib.

We propose methylation changes underlying EMT to constitute a common resistance mechanism to cancer therapies. This process can be reversed pharmacologically and monitored non-invasively in ctDNA to predict resistance to treatment.