Epigenetic feedback and stochastic partitioning during cell division can drive resistance to EMT

Abstract

Epithelial-mesenchymal transition (EMT) and its reverse process mesenchymal-epithelial transition (MET) are central to metastatic aggressiveness and therapy resistance in solid tumors. While molecular determinants of both processes have been extensively characterized, the heterogeneity in the response of tumor cells to EMT and MET inducers has come into focus recently, and has been implicated in the failure of anti-cancer therapies. Recent experimental studies have shown that some cells can undergo an irreversible EMT depending on the duration of exposure to EMT-inducing signals. While the irreversibility of MET, or equivalently, resistance to EMT, has not been studied in as much detail, evidence supporting such behavior is slowly emerging. Here, we identify two possible mechanisms that can underlie resistance of cells to undergo EMT: epigenetic feedback in ZEB1/GRHL2 feedback loop and stochastic partitioning of biomolecules during cell division. Identifying the ZEB1/GRHL2 axis as a key determinant of epithelial-mesenchymal plasticity across many cancer types, we use mechanistic mathematical models to show how GRHL2 can be involved in both the abovementioned processes, thus driving an irreversible MET. Our study highlights how an isogenic population may contain subpopulation with varying degrees of susceptibility or resistance to EMT, and proposes a next set of questions for detailed experimental studies characterizing the irreversibility of MET/resistance to EMT.

Keywords: GRHL2; asymmetric cell division; epigenetics; epithelial-mesenchymal transition; mesenchymal-epithelial transition.

Figure 1. GRHL2 correlates with an epithelial phenotype.

Scatter plots showing correlation of GRHL2 with EMT-TF ZEB1 and CDH1 (E-cadherin) in TCGA datasets and CCLE: (A) breast cancer, (B) colon adenocarcinoma, (C) colorectal adenocarcinoma, (D) ovarian carcinoma, (E) CCLE. R, p denote Pearson’s correlation coefficient and corresponding p-value for corresponding plot.

Figure 2. GRHL2/ZEB1 axis correlates with EMT/MET across cancer types.

(A) Pairwise Pearson’s correlation between different EMT and MET regulatory genes in the CCLE dataset. Pearson’s correlation value (cor) of each gene pair is represented as the size of the circle and filled with corresponding color from the color palette represented below the ranging from −1(red) to +1(blue). Boxes highlighted by the black squares represent insignificant (p > 0.01) correlation. (B) Scatter plot of genes correlated using Pearson correlation method with GRHL2 and ZEB1 in CCLE dataset. Each dot represents one gene and coordinates are Pearson cor values with ZEB1 and with GRHL2. Color of the dots is based on the p-value obtained from correlation test, Blue dots for genes having p < 0.05 with GRHL2 and p > 0.05 with ZEB1. Red dots for genes having p < 0.05 with ZEB1 and p > 0.05 with GRHL2. Green dots for genes having p < 0.05 with GRHL2 and ZEB1. Black dots for genes having p > 0.05 with GRHL2 and ZEB1. Numbers in each quadrant represent the number of genes in that quadrant.

Figure 3. EMT decision-making network.

(A) A core network regulating EMT via two mutually inhibition loops between miR-34 (miR-200) and SNAIL (ZEB). Signal I represents external EMT-inducing signals such as HGF, TGF-β, NF-κB and HIF1α, among others. GRHL2 is a transcription factor which forms a mutually inhibitory loop with ZEB. (B) Bifurcation diagram (shown for the levels of ZEB mRNA) with I as the bifurcation parameter. Solid lines represent stable states, i.e., epithelial, hybrid or mesenchymal state, and dashed lines represent unstable states. Shaded rectangle represents the values of I for which all three phenotypes can co-exist. (C) Starting from epithelial state (miR-200 = 17,000, mZEB = 50, ZEB = 10,000 molecules) with different pre-fixed threshold value of inhibition of ZEB by GRHL2, 100 cells are treated by fixed high EMT-inducing signal (I = 75,000 molecules). (D) Population distribution changes as a function of signal I.

Figure 4. Epigenetic feedback on GRHL2 self-activation.

(A) The bifurcation diagrams for core EMT circuit with/without epigenetic feedback on self-activation of GRHL2. Black curve denotes the case without any epigenetic feedback; blue curve represents the epigenetic feedback case. (B) Starting from 100% cells in an epithelial state (miR-200 = 17,000, mZEB = 50, ZEB = 10,000 molecules), simulation results showing how the population changes as a function of simulation time, on addition of noise. Dashed lines represent no epigenetic feedback case, solid lines represent case with strong epigenetic feedback (a = 0.22) on GRHL2’s self-activation (Signal I₀ = 75,000 molecules). (C) A representative dynamical trajectory for no epigenetic feedback case. (D) A representative dynamical trajectory for strong epigenetic feedback case.

Figure 5. Epigenetic feedback on inhibition of ZEB by GRHL2.

(A) The bifurcation diagrams for core EMT circuit with/without epigenetic feedback on the inhibition of ZEB by GRHL2. Black curve represents the case without any epigenetic feedback; blue curve represents the epigenetic case. (B) Starting from all cells in an epithelial state (miR-200 = 17,000, mZEB = 50, ZEB = 10,000 molecules), simulation results showing how the population changes as a function of simulation time. Dashed lines represent no epigenetic feedback case, and solid lines represent case with strong epigenetic feedback (α = 0.14) on the inhibition of ZEB by GRHL2 (Signal I₀ = 75,000 molecules). (C) The percentage of population which exhibit M phenotype, for varying values of α. (D) A sample dynamical diagram for strong feedback case.

Figure 6. Reversibility of EMT starting from epithelial state (miR-200 = 17,000, mZEB = 50, ZEB = 10,000 molecules), a cell is treated by different time duration (5, 10, 20 arbitrary units [a.u], as marked by arrow) of high EMT-inducing signal (I = 125,000 molecules), corresponding to the {H, M} bistable region.

Then, this signal is reduced to a lower level (I = 71,000 molecules) corresponding to the {E, M} bistable region. (A–C) Represents the case without epigenetic feedback, and (D–F) represents the case with strong epigenetic feedback on inhibition of ZEB by GRHL2.

Figure 7. A seemingly irreversible MET at the population level.

Starting with a population of mesenchymal cells on day 0 (all cells with very high value of the EMT inducer), the EMT inducer was withdrawn in fixed dosages each day for the first 10 days. As a result, a large fraction of cells in the population underwent MET. Day 11 onwards, fixed dosages of the EMT inducer were added each day for the next 10 days. In the absence of GRHL2 (left panel), the fraction of mesenchymal cells went back to nearly 100%, same as the value on day 0. However, in the presence of GRHL2 (right panel), ~15% of the cells in the population were epithelial on day 20. These cells thus represented a subpopulation that had undergone a MET that is irreversible at least on the time scale investigated here. The mean over 16 independent simulation runs is shown here. The error bars indicate the standard deviation over the independent runs.