Fig 2: NOTCH1 activation and ZEB1 expression in invasive ESCC predicts poor prognosis. a, b Representative IHC images for ICN1 and ZEB1 in ESCC cells in a deep invasive tumor nest and ESCC cells invading into lymphatic vessels. a ESCC #55; b ESCC #62 in Supplementary Data 1. c Representative IF image for ICN1 and CD44 in invasive ESCC cells. Scale bars, 50 µm in a–c. d Survival curve for 185 post-surgical ESCC patients with or without ICN1 expression at the tumor invasive front. e Model of the role of Notch1 in EMT in ESCC. Notch1 activation promotes tumorigenicity and heterogeneity in SCC via EMT. Notch1 drives squamous-cell differentiation by inducing Notch3 in normal squamous epithelia as well as SCC cells. Notch3 limits EMT. In response to TGFß from the tumor microenvironment (e.g., cancer-associated fibroblasts and inflammatory cells), however, transcriptional repression of Notch3 by ZEB1 permits Notch1-mediated induction of CD44H cells via EMT. Notch activation may result in the generation and maintenance of CD44H cells possessing mesenchymal properties and enhanced malignant potential. EMT allows neoplastic cells to cope with stress during carcinogenesis and disease progression (e.g., genotoxic stress induced by 4NQO). CD44H cells produce pro-tumorigenic cytokines (e.g., IL-6) and tissue remodeling factors (e.g., MMP13, LOX, and POSTN) (Supplementary Fig. 3e)

Fig 3: Notch1 facilitates expansion of CD44H cells with mesenchymal properties. a qRT-PCR analysis for indicated genes comparing EN60 intratumoral CD44L and CD44H cells. *P < 0.05 vs. CD44L; # P < 0.0001 vs. CD44L; n = 3 per group. b Representative flow cytometry scatter plot determining CD44H cells in EN60 tumors grown for 4 weeks with or without DOX-induced ICN1 (EN60-ICN1 TetOn). c Experimental design of serial transplantation experiments and tumor formation rates by CD44L and CD44H cells purified from parental xenograft tumors carrying DOX-inducible ICN1 (EN60-ICN1 TetOn and TE11-ICN1 TetOn). Parental tumors were grown in mice without DOX treatment and dissociated for FACS-purification of CD44L and CD44H cells. Purified cells were serially transplanted (103 cells per injection site) into recipient mice to monitor tumor formation. Recipient mice were treated with or without DOX. TE11 carried DNMAML1 or zeo (empty vector control). *P < 0.05 for CD44L and DOX (+) vs. CD44L and DOX (-), n = 10–12 per group; ns not significant for CD44L and DOX (+) vs. CD44H (with or without DOX treatment), n = 10 per group; # P < 0.05 vs. TE11-ICN1 TetOn -zeo CD44L and DOX (+), n = 10 per group. d Flow cytometry analysis for EpCAM-negative cells in TE11 with NOTCH1 deletion (N1?) or wild-type control (N1WT). Cells were treated with or without 5 ng/ml TGFß for 72 h. *P < 0.05 vs. TGFß (-); # P < 0.0001 vs. N1WT and TGFß (+), n = 3. Data are presented as mean ± s.e.m. e Multicolor IF for E-cadherin and Zeb1 in representative single-cell-derived organoids from 4NQO-induced ESCC tumors of Notch1 loxP/loxP mice with or without ex vivo Cre-mediated recombination. Zeb1 expression diminished in organoids upon Cre-mediated Notch1 deletion (arrowheads). Box denotes area that is magnified in panel below. KC keratinized core of organoids. Scale bars, 20 µm. Bar diagrams represent mean ± s.d. in a, d. At least two independent replicates were performed for all experiments. Student’s t-test was used for paired data comparisons in a, d. Chi square with Fisher’s exact test was used for percentage comparisons in c

Fig 4: Notch1 promotes ESCC tumorigenesis. a Schematic of 8×CSL-GFP reporter. Notch activation permits GFP reporter expression via concatemeric CSL-binding sites. Constitutively active UBC promoter drives tRFP expression concurrently. Cells without tRFP represent host-derived stromal cells. b Representative flow cytometry scatter plots determining 8×CSL-mediated GFP reporter activation, indicating Notch active population of tRFP-labeled ESCC cells in TE11 and EN60 xenograft tumors. On average, 43.5 ± 0.5% s.d. of EN60 and 11.2 ± 6.5% s.d. of TE11 cells comprised the live GFPPos/tRFPPos fraction across three independent tumors. c, d Tumor growth curves in immunodeficient mice carrying TE11 tumors of indicated genotypes. Upon xenograft transplantation, mice were treated with DOX to induce ICN1 or NOTCH1-directed shRNA. Tumor growth was monitored for indicated time periods. In c, *P < 0.01 vs. zeo and DOX (-), # P < 0.05 vs. zeo and DOX (+), n = 6–8 per group. In d, *P < 0.05 vs. DOX (-); n = 6 per group. e TE11 cells with CRISPR/Cas9-mediated NOTCH1 deletion. Immunoblotting confirms ICN1 loss in NOTCH1-deleted TE11 (N1?), but not non-targeted control TE11 (N1WT) cells. Bar graph shows tumor formation rate in immunodeficient mice 7 weeks after xenograft transplantation. *P < 0.05 vs. N1WT, n = 8 per group. f Esophageal keratinocytes isolated from 4NQO-induced IEN (dysplasia) lesions or ESCC tumors of Notch1 loxP/loxP mice were utilized to generate 3D esophageal organoids in the presence or absence of ex vivo Cre-mediated recombination. qRT-PCR analysis confirms inhibition of Notch1 gene expression upon ex vivo Cre-mediated recombination in ESCC organoids. *P < 0.000 vs. Cre (-), n = 3. Organoid formation rate was evaluated at day 14 post-plating and is represented as relative to Cre (-) for IEN and ESCC. *P < 0.000 vs. Cre (-) IEN; # P < 0.001 vs. Cre (-) ESCC. Line graphs represent mean ± s.e.m. in c, d. Bar diagrams represent mean ± s.d. in f. At least two independent replicates were performed for all experiments. Repeated measures ANOVA with Tukey’s post hoc test were used for multiple comparisons in c, d. Chi square with Fisher’s exact test was used for percentage comparisons in e. Student’s t-test was used for paired data comparisons in f

Fig 5: TGFß signaling facilitates ESCC tumor growth, Notch1 signaling, and EMT. a Growth curve for mice bearing TE11 or HNSCC PDX tumors receiving either ID11 anti-TGFß blocking antibody or control IgG via intraperitoneal injection at indicated time points (arrows). ID11 neutralizes all TGFß isoforms. *P < 0.05 vs. IgG, n = 5–8 per group. b, c p53 +/+ mice were treated with 4NQO for 16 weeks. Six weeks following 4NQO withdrawal, mice were treated with ID11 anti-TGFß blocking antibody or control IgG via intraperitoneal injection three times weekly for a period of 2 weeks. In b, quantification of ICN1 IHC scoring in esophageal epithelium with representative IHC images for IgG-treated and anti-TGFß blocking antibody-treated animals. *P < 0.05; n = 4–5 per group. Scale bar, 50 µm. In c, multicolor IF for E-cadherin and Zeb1 in representative 4NQO-induced IEN lesions from IgG-treated and anti-TGFß blocking antibody-treated animals. Data indicate mean ± s.e.m. in a, b. At least two independent replicates were performed for all experiments. Repeated measures ANOVA with Tukey’s post hoc test were used for multiple comparisons in a. Student’s t-test was used for paired data comparisons in b