Investigation of the heterogeneity of cancer cells using single cell Ca²⁺ profiling

Abstract

Calcium (Ca²⁺) is an essential and ubiquitous second messenger that controls numerous cellular functions. The regulation of intracellular Ca²⁺ oscillations defines a Ca²⁺ signature that is representative of the cellular identity and the phenotype of a cell. In cancers, aberrant Ca²⁺ fluxes have been associated with characteristics such as migration, proliferation or resistance to chemotherapy. However, it remains unclear whether cancer cells have specific Ca²⁺ signatures that are representative of their oncogenic properties. To reveal the existence of oncogenic Ca²⁺ signatures, we develop a method combining single cell calcium imaging and unsupervised and supervised machine learning approaches. Briefly, we generated a database of 27,439 single-cell agonist-induced Ca²⁺ responses obtained from 16 prostate and colon cancer cell lines. Graph-based unsupervised clustering was used to define and compare Ca²⁺ signatures of cancer cell lines. A supervised neural network model trained on the collection of agonist-induced Ca²⁺ responses successfully identify individual cancer cells and predicted some of their characteristics. We extend these methods to observe Ca²⁺ signatures of docetaxel-resistant cancer cells and in a co-culture model mimicking the interaction of cancer cells with cancer-associated fibroblasts. Our models successfully highlight subtle changes in those Ca²⁺ signatures and are able to distinguish individual cancer cells from fibroblasts and predict their sensitivity to Docetaxel. Taken together, this study provides a proof of concept that cancer cells can be defined by their Ca²⁺ signature and conversely that Ca2+ signaling can serve as a predictor of cancer cell characteristics.