Fig 2: Phenotypic plasticity maintains EpCAMlo and EpCAMhi cells in a stochastic equilibrium.(A) Analysis of plasticity of EpCAMhi and EpCAMlo cells from HCT116 (upper panel) and SW480 (lower panel). EpCAMhi and EpCAMlo cell fractions were sorted and plated in culture. At different time points, as indicated, cells were reanalyzed by flow cytometry for their levels of CD44 (y-axis) and EpCAM (x-axis) expression. (B) Flow cytometric analysis of single CD44hiEpCAMhi and CD44hiEpCAMlo HCT116 cells sorted by FACS and cultured for 72 days. Three representative individual single-cell clones per cell fraction are shown. (C) Dynamics of the EpCAMhi and EpCAMlo subpopulations from the HCT and SW480 cell lines as measured by FACS (% of total) over time. Under each graph, a schematic shows the estimated transition probabilities from the fitted two-state Markov model. Figure 2—source data 1.Phenotypic plasticity maintains EpCAMlo and EpCAMhi cells in a stochastic equilibrium: clonal analysis.The table encompasses a selection of individual single-cell sorted EpCAMhi and EpCAMlo HCT116 and SW480 clones. Single cells were sorted, grown up, and analyzed at two indicated times points by flow cytometry for their CD44 and EpCAM expression. Percentage of cells present in the set EpCAMhi and EpCAMlo gates is presented in the bar graph. Total success rate of clones grown out of single cells: HCT116 EpCAMhi 15 clones out of 24 sorted cells, HCT116 EpCAMlo: 31/48; SW480 EpCAMhi: 5/36; SW480 EpCAMlo: 11/36.

Fig 3: RNAseq analysis reveals enhanced Wnt signaling in EpCAMlo cells.(A) Multidimensional scaling analysis of RNAseq profiles of EpCAMhi and EpCAMlo cells from the HCT116 and SW480 lines. Red: HCT116, black: SW480, circle: EpCAMhi, triangle: EpCAMlo. (B) Ingenuity Pathway Analysis (IPA) of the HCT116 and SW480 expression profiles from the multicell line analysis (p adjusted value <0.01; log2 fold change <–1.5 and >1.5). Red marked pathways highlight the enhanced involvement of pathways involved in epithelial to mesenchymal transition, Wnt signaling, and the formation of colon cancer metastasis in the EpCAMlo subpopulation compared to EpCAMhi cells. (C) TOP-Flash luciferase reporter analysis of Wnt signaling activity in colon cancer cell lines HCT116 and SW480 upon treatment with 4 µM Chiron for 3 days. Each bar represents the mean ± SD of two independent experiments. (D) Flow cytometric analysis using antibodies directed against CD44 and EpCAM of control and 4 µM Chiron—figure supplemented HCT116 (A) and SW480 (B) cultures. Graphs show percentage of cells within the CD44hiEpCAMhi and CD44hiEpCAMlo gates relative to the control. Each bar represents the mean ± SD of two independent experiments. (E) Immunofluorescence analysis of control and Chiron-treated HCT116 (left panel) and SW480 (right panel) cells. After 3 days of treatment, cells were fixed with 4% paraformaldehyde and stained with antibodies against EpCAM (green) and ZEB1 (red). Nuclei were visualized by DAPI staining of DNA (blue). Scale bar: 100 µm. (F) Flow cytometric analysis of three HCT116 cell lines with differential ß-catenin mutation status, a parental HCT116 (HCT116-P) cell line harboring one WT and one mutant allele (Ser45 del), and two HCT116-WT and HCT116-MT cell lines harboring one WT or one mutant allele, respectively, generated by disruption of the other allele in the parental cell line (Kim et al., 2019). Figure 3—source data 1.List of differentially expressed genes in EpCAMlo vs. EpCAMhi cells in HCT116 and SW480.A total of 152 and 353 differentially regulated genes were identified between the CD44hiEpCAMhi and CD44hiEpCAMlo cells in HCT116 and SW480, respectively (p adjusted<0.01, when applying a log2 fold change of <–1.5 and > 1.5).

Fig 4: ZEB1 and ESRP1 differential expression in quasi-mesenchymal and highly metastatic EpCAMlo colon cancer cells.(A) Gene rank plot showing differentially expressed genes between EpCAMhi and EpCAMlo with combined analysis of HCT116 and SW480. (B) RT-qPCR ESRP1 and ZEB1 expression analysis of HCT116 and SW480 EpCAMhi, EpCAMlo, and bulk subpopulations. GAPDH expression was used as control (means ± SEM, n=3). **=p < 0.01. (C) ESRP1 and ZEB1 western analysis in HCT116 and SW480 EpCAMhi, EpCAMlo, and bulk fractions. β-Actin was used as loading control. (D) RT-qPCR and western analysis of ZEB1 and ESRP1 expression in ZEB1-OE and -KD HCT116 and SW480 cells. Expression values were normalized in each sample with those from the parental HCT116 and SW480 cell lines. HCT116 and SW480 cells transduced with the shZEB1 lentivirus were induced by 1 µg/mL doxycycline for 72 hr. Expression values were normalized with those from non-induced cells; GAPDH expression was employed as control (means ± SEM, n=3). *=p < 0.05, **=p < 0.01. β-Actin was used as loading control. (E) RT-qPCR ZEB1 and ESRP1 expression analysis in ESRP1-OE and -KD HCT116 and SW480 cells. Two independent ESRP1-OE clones were selected for each cell line. Expression values were normalized in each sample with those from the parental HCT116 and SW480 cell lines. HCT116 and SW480 cells transduced with the shESRP1 lentivirus were induced by 1 µg/mL doxycycline for 72 hr. Two independent clones were selected for each cell line. Expression values were normalized with those from non-induced cells; GAPDH expression was employed as control (means ± SEM, n=3). *=p < 0.05, **=p < 0.01. (F) CD44/EpCAM FACS analysis of HCT116 and SW480 EpCAMlo and EpCAMhi subpopulations in ESRP1-OE cells. Two independent clones are shown for each cell lines. Figure 1—source data 1.Original files and labelled bands of western blots in Figure 1C–D.

Fig 5: Identification and characterization of EpCAMlo cells in colon cancer cell lines.(A) Flow cytometric analysis of the colon cancer cell lines HCT116 (left panel) and SW480 (right panel) with antibodies directed against CD44 and EpCAM. EpCAM/CD44-positive and -negative regions (gray quadrants) were defined as in Figure supplement 1 using multiple isotype controls and are shown by the quadrants in the plots. Notably, both HCT116 and SW480 revealed a continuum of different EpCAM and CD44 expression levels with a large CD44highEpCAMhigh (EpCAMhi) cluster followed by a tail of gradually decreasing EpCAM and increasing CD44 levels. By applying specific gates, cells were divided in a large EpCAMhi cluster, together with a considerably smaller CD44highEpCAMlow (EpCAMlo) subpopulation. To ensure good separation from the large EpCAMhi cluster and maximal sorting purity, EpCAMlo cells were gated as CD44hi events = 60% of the EpCAM fluorescence intensity of the left border of the EpCAMhi gate and sorted from =50% of that value. Variable percentages of EpCAMlo cells were found to feature the HCT116 (5.0% ± 2.5%) and SW480 (16.7% ± 13%) cell lines, respectively. For the sake of simplicity, gates are shown in the figure only if they encompass sizeable percentages of cells. Graphs show representative analysis of one experiment. (B) Phase-contrast microscopy images of sorted EpCAMhi and EpCAMlo cells from HCT116 (upper images) and SW480 (lower images) cells. While EpCAMhi cells formed compact colonies with characteristic epithelial morphology, EpCAMlo cells showed a more spindle- and mesenchymal-like appearance. Scale bar: 100 µm. (C) Intrasplenic injection of bulk, EpCAMhi, and EpCAMlo cells from HCT116 (left panel) and SW480 (right panel). For each transplantation experiment, 2 × 104 cells were injected in the spleen of a recipient NSG mouse. 4 (HCT116) and 8 (SW480) weeks after injection, mice were sacrificed and individual tumors counted. Single and double asterisks indicate significant differences (p<0.05 and p<0.01, respectively). HCT116: bulk (n = 8), EpCAMhi (n = 9), and EpCAMlo (n = 7). SW480: bulk (n = 4), EpCAMhi (n = 4), and EpCAMlo (n = 4). (D) Images of mouse livers 4 (HCT116) and 8 (SW480) weeks after orthotopic injection with 104 cells. Scale bar: 5 mm. Figure 1—source data 1.EpCAMlo cells among colon cancer cell lines.The percentage of CD44hi/EpCAMlo subpopulation was determined in a panel of commonly employed colon cancer cell lines by flow cytometric analysis. Please note that, within the same cell line, these percentages can vary depending on the passage number and culture conditions. If indicated by Lindeman et al., consensus molecular subtype classification of the cell lines is shown. Figure 1—source data 2.Cell cycle analysis of EpCAMhi and EpCAMlo cells in HCT116 and SW480.Cells fractions were sorted and plated in culture. After 72 hr, cells were fixed and stained with propidium iodide. Cell cycle distribution was assayed by flow cytometry. Tables demonstrate average and standard deviation of three independent experiments. Figure 1—source data 3.Quantification of EpCAMhi/lo percentages of all liver metastases as determined by FACS.