A six-step scheme illustrating the generation of a cancer-specific EMT signature. Note that tumours and cell lines have their own cancer-specific EMT signatures. (Top right panel) Red and green bars on sample enrichment score (ES) bar chart indicate, respectively, mesenchymal-like (Mes) and epithelial-like (Epi) samples selected for building the BinReg EMT signature. (Middle right panel) Heatmap of the EMT signature from Significance Analysis of Microarray (SAM)/Receiver Operating Characteristics (ROC) analysis. The colour bar shows the EMT phenotype probability of cell line or tumour samples, sorted from most Epi to most Mes. Red and green bars indicate Mes and Epi samples selected for SAM/ROC analysis. (Bottom right panel) Plots of empirical cumulative distribution function of Mes (red) and Epi (green) gene sets.Dot plot of EMT score (mean ± SEM) for breast cancer cell lines (n = 34) with spindle- and non-spindle-like morphologies. Mann–WhitneyU-testP-value is shown.Immunohistochemistry staining heatmap of Oestrogen Receptor (ER), Progesterone Receptor (PR), and Epi (CDH1, ERBB2, CK19) as well as Mes (CK5, VIM, CDH2) markers (black = low, red = high, white = no data). Breast cancer cell lines (n = 39) are aligned from the most Epi to most Mes based on the EMT score, as shown by the bar chart. Dot plot is the-log₁₀P-value of two-sample Kolmogorov–Smirnov test. Arbitrary threshold ofP < 0.001 was used to define Epi, intermediate and Mes cell lines. Breast cancer cell line microarrays and subtype are from GSE16795 (Hollestelleet al, ). Subtype colour code: blue, Luminal; maroon, Basal.

Circos plot illustrating the generic EMT signature: the overlap of ovarian (blue), breast (purple), lung (green), colorectal (yellow), bladder (red) and gastric (orange) cancer-specific EMT signatures is shown. Links indicate overlapping genes (red = mesenchymal, green = epithelial). Heatmap on the inner ring indicates weight computed based on Significance Analysis of Microarray (SAM) fold-change, false discovery rate, Receiver Operating Characteristics (ROC) and number of samples of a gene in each cancer-specific EMT signature (red = high, blue = low weight). On the outermost ring, genes are represented by ticks and aligned from the highest SAM fold-change to the lowest for each cancer type. Selected genes are labelled.EMT score (mean ± SEM;y-axis) of breast cancer molecular subtypes as predicted using ssGSEA and signature from Pratet al () in non-laser-capture micro-dissected (non-LCM) cohort (n = 3,992; upper panel) and LCM cohort (n = 417; lower panel). The Mann–WhitneyU-testP-value of binary comparison for each subtype is given. Colour code: maroon, Basal; yellow, Claudin-low; light blue, Luminal-A; dark blue, Luminal-B; orange, ERBB2+; green, Normal-like. N.A, not applicable. Note that noP-value is available for Claudin-low and Normal-like subtypes in lower panel becausen < 3.

Scatter plot of EMT scores (mean ± SEM;y-axis) of various cancers in clinical samples (upper panel) and cell lines (lower panel) sorted by cancer type and mean EMT score. EMT score nearer to +1.0 is more mesenchymal-like (Mes), whereas EMT score nearer to −1.0 is more epithelial-like (Epi). EMT scores of overlapping cell lines in Cancer Cell Line Encyclopedia (CCLE) (Barretinaet al, ) and SANGER/COSMIC (Garnettet al, ) collections were averaged.

A, B Plot of log₂ hazard ratio (HR; mean ± 95% confidence interval) comparing (A) overall survival (OS) and (B) disease-free survival (DFS) of Epi and Mes tumours in different cancers and cohorts. DFS includes progression-free, recurrence-free and distant metastasis-free survival (cohorts inclusive of distant metastasis-free survival were indicated with *). CorrespondingP-values from the log-rank test are given next to each cohort, and those with significant differences (P < 0.05) are marked red. Log₂ HR < 0.0 indicates Epi tumours with survival benefit, whereas log₂ HR > 0.0 indicates Mes tumours with survival benefit. Meta-analysisP-value for effect or heterogeneity was computed using DerSimonian–Laird binary random effect (for overall) or Peto fixed effect method (for individual cancer).

Bar plots of breast (n = 270; left panel) and ovarian (n = 328; right panel) cancers stratified by EMT status and clinical response based on response evaluation criteria in solid tumours (RECIST). Regimen was neoadjuvant doxorubicin and cyclophosphamide for breast cancer, and platinum-based adjuvant/progression/recurrence chemotherapy for ovarian cancer. Percentage distribution of EMT status is given in each clinical response group. Abbreviation: CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease. Green, epithelial-like (Epi); orange, intermediate; red, mesenchymal-like (Mes).Volcano plot of EMT correlation with drug sensitivity regardless of cancer type.Rho ∈ [−1.0, +1.0] (x-axis) and-log₁₀P-value (y-axis) were computed by Spearman's correlation coefficient test. Dashed line ofP-value = 0.1 is plotted. Red and green indicate higher drug resistance in Mes tumours (Rho ∈ [0, +1.0]) and Epi tumours (Rho ∈ [−1.0, 0]), respectively.Kaplan–Meier analysis comparing overall survival (left panel) and disease-free survival (right panel) of Epi (green) and Mes (red) ovarian cancer patients who underwent a treatment regimen with (dark colour) or without (light colour) paclitaxel.P-value reported was computed by log-rank test. Abbreviation: HR = hazard ratio.