Understanding Phenotypic Plasticity Induced by Environmental Factors Using Computational Markov Modeling

Cancer plasticity contributes to tumorigenesis, tumor progression, and drug resistance. Plasticity has traditionally been attributed to stem‐like cells, but recent studies used markers of stemness and differentiation to purify specific phenotypes from a mixed population of cancer cells by flow cytometry, and found that both differentiated and dedifferentiated phenotypes could revert over time to a heterogeneous equilibrium mixture, through stochastic transitions. A given instance of heterogeneity can in theory arise either from multiple static phenotypes or from transitions between non‐static phenotypes, and therefore, understanding plasticity requires measuring transition rates. We developed a Markov‐modeling computational approach to analyze time‐series intensity data from flow cytometry experiments, and to quantify phenotypic change in terms of cell state transitions. The software employs optimization and novel statistical bootstrap methods to compute best‐fit transition probabilities (and associated dispersion measures or confidence intervals) for stochastic transitions between multiple cell states. Applying this method to the MCF10A model of breast cancer provided quantification of stochastic transitions as follows. Cells were grown in different culture conditions: basal medium supplemented with the components of MCF10a complete growth medium (insulin, EGF, hydrocortisone, cholera toxin), given individually or together. For each condition, cells were purified into populations of epithelial‐like CD44 + CD24 + cells and stem‐like CD44 + CD24 − cells. The purified populations were cultured for 12 days, and the population fraction of each state was measured using flow cytometry on days 4, 8 and 12 post‐sorting. Analysis using Markov models showed that Hydrocortisone (HC), and Cholera toxin (CTX) cause the largest changes in transition rates, compared with control (basal media). CTX drove the equilibrium toward a more differentiated CD24 + phenotype. In contrast, HC supplementation drove toward the CD24 − phenotype, which showed increased mammosphere formation and an aggressive mRNA profile. The bootstrapping method showed these transition rate changes to be statistically significant. In conclusion, we developed a novel algorithm to estimate transition probabilities from FACS data, and to make statistical comparisons between transition probabilities. Applying our method to a cell culture model of triple‐negative breast cancer showed that exposure to hydrocortisone increased the rate of cell transitions toward a more malignant phenotype, suggesting that chronic exposure to the stress hormone cortisol might be particularly deleterious. This software will be released open‐source and is currently available by request. Support or Funding Information Singapore Ministry of Health's National Medical Research Council (NMRC) under its Open Fund Individual Research Grant scheme (OFIRG15nov062). National Research Foundation (NRF), Prime Minister's Office, Singapore, under its CREATE programme, Singapore‐MIT Alliance for Research and Technology (SMART) BioSystems and Micromechanics (BioSyM) IRG This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal .