ZEB1-mediated fibroblast polarization controls inflammation and sensitivity to immunotherapy in colorectal cancer

Abstract

The EMT-transcription factor ZEB1 is heterogeneously expressed in tumor cells and in cancer-associated fibroblasts (CAFs) in colorectal cancer (CRC). While ZEB1 in tumor cells regulates metastasis and therapy resistance, its role in CAFs is largely unknown. Combining fibroblast-specific Zeb1 deletion with immunocompetent mouse models of CRC, we observe that inflammation-driven tumorigenesis is accelerated, whereas invasion and metastasis in sporadic cancers are reduced. Single-cell transcriptomics, histological characterization, and in vitro modeling reveal a crucial role of ZEB1 in CAF polarization, promoting myofibroblastic features by restricting inflammatory activation. Zeb1 deficiency impairs collagen deposition and CAF barrier function but increases NFκB-mediated cytokine production, jointly promoting lymphocyte recruitment and immune checkpoint activation. Strikingly, the Zeb1-deficient CAF repertoire sensitizes to immune checkpoint inhibition, offering a therapeutic opportunity of targeting ZEB1 in CAFs and its usage as a prognostic biomarker. Collectively, we demonstrate that ZEB1-dependent plasticity of CAFs suppresses anti-tumor immunity and promotes metastasis.

Keywords: Cancer-Associated Fibroblast Plasticity; Colorectal Cancer; Immune Checkpoint Blockade; Tumor Microenvironment.

Figure 1. Context-dependent role of stromal ZEB1 during colorectal carcinogenesis.

(A, B) ZEB1 IHC on human (A) and ZEB1/E-cadherin IF stainings on mouse (B) CRC samples. Red and blue arrowheads depict cells with high and low/absent ZEB1 detection, respectively. (C–H) AOM/DSS model (C) showing representative endoscopic images (D) (dotted lines indicate unobstructed area) and quantification of colon obstruction in Fib^Ctrl and Fib^ΔZeb1 mice (n = 24/17 for Fib^Ctrl/Fib^ΔZeb1, Genotype: P = 0.0282, two-way ANOVA, day 75: P = 0.0382, Šídák’s multiple comparisons test), macroscopic images (E, colons opened longitudinally, arrowheads point to individual tumors) and quantification of tumor volume and number (F) (n = 23/16 for Fib^Ctrl/Fib^ΔZeb1, number: P = 0.0714, volume: P = 0.0180, Student’s t test). Quantitative analysis of ZEB1 depletion, proliferation (KI67) and apoptosis (cl. Caspase 3) (G) with representative KI67 IHC is given (H). For ZEB1, the fraction of positive stromal cells was quantified. For KI67 and cl. Caspase (CASP) 3, the fraction of all tumor cells was quantified (n = 17/11 for Fib^Ctrl/Fib^ΔZeb1, Zeb1: P = 0.0054, KI67: P = 0.0020, cl. CASP3: P = 0.1126, Mann–Whitney test). (I–L) Schematic overview of orthotopic transplantation of tumor organoids into the cecum of Fib^Ctrl and Fib^ΔZeb1 mice (I). AKP^re organoids were generated after re-culturing cells retrieved from orthotopic AKP tumors. Analysis of tumor onset (detected by palpation) after orthotopic AKP organoid transplantation (J) (n = 12/10 for Fib^Ctrl/Fib^ΔZeb1, P = 0.6401, Mantel–Cox test) as well as of liver metastasis incidence (K) (AKP: P = 0.0427, AKP^re: P = 0.0425, Fisher’s exact test) and numbers (L) after orthotopic transplantation of AKP or AKP^re tumor organoids (n = 12/10 for Fib^Ctrl/Fib^ΔZeb1 with AKP and 12/14 for Fib^Ctrl/Fib^ΔZeb1 with AKP^re, AKP: P = 0.8469, AKP^re: P = 0.0007, two-way ANOVA). All mice with AKP^re transplantation were treated with control IgG and are shown also in Fig. 5. Mice with AKP transplantation were treatment-naïve. Data information: Data are presented as mean ± SEM (D) or mean ± SD (F, G, L). Scale bars represent 50 µm (A, B), 5 mm (E) or 100 µm (H). Source data are available online for this figure.

Figure 2. scRNA sequencing of CAFs reveals a key role of ZEB1 for fibroblast plasticity.

(A) Experimental scheme for isolation of CAFs and other cell types. AOM/DSS adenomas (left) and primary tumors from the orthotopic AKP model (right) of Fib^Ctrl and Fib^ΔZeb1 mice were enzymatically dissociated, and fibroblasts were enriched by depletion of CD31+, CD45+, and EPCAM+ cells by flow cytometry. Numbers of analyzed mice and sorted cells are shown. (B–G) Fibroblast analysis from the AOM/DSS model. (B) Integrated UMAP-(Leiden) clusters of CAFs from n = 3 mice per genotype were subjected to differential abundance (‘DA’) analysis. 3 significant DA regions between Fib^Ctrl and Fib^ΔZeb1 were found (Wilcoxon P values for regions 1, 2, and 3: 5.37 × 10⁻¹⁴; 1.35 × 10⁻¹² and 5.62 × 10⁻⁰⁶). Note that region 2 derived exclusively from one mouse and was therefore neglected. (C) Projections of iCAF and myCAF gene signature scores (Elyada et al, 2019). Note the highlighted DA regions from (B). (D) Log₂ fold changes of representative DAseq marker genes in DA regions 1 and 3, designated as “iCAF” and “myCAF” regions, respectively (P < 0.01, as determined by STG (stochastic gates) within DAseq). (E) Independent t-SNE clustering in Fib^Ctrl (left) and Fib^ΔZeb1 mice (right). (F) Cluster similarities defined by cluster annotations based on “SingleR” scores (see “Methods” for details) (Aran et al, 2019). Grayscale shows the log₂-transformed numbers of Fib^ΔZeb1 cells assigned to the different Fib^Ctrl clusters. (G) Heatmap showing the similarity (annotation scores) of gene expression in CAF clusters with published gene sets. (H–J) scRNA-seq of fibroblasts, immune cells and tumor cells after orthotopic transplantation of AKP organoids. (H) Data from n = 4 mice per genotype of the orthotopic model were subjected to scRNA sequencing and t-SNE clustering in Fib^Ctrl (left) or Fib^ΔZeb1 tumors (right). (I, J) Down- and upregulated gene sets in Fib^ΔZeb1 CAFs in comparison to the two Fib^Ctrl CAF clusters, as determined using Enrichr. Differential genes between Fib^ΔZeb1 CAFs and each of the two Fib^Ctrl CAF clusters were individually determined and pooled before enrichment analysis (FDR ≤ 0.1; P ≤ 0.05, as determined by default Welch t tests within the findMarkers function). (K, L) Tumor sections from mice after transplantation with AKP organoids were subjected to multiplexed immunostaining (n = 5/6 for Fib^Ctrl/Fib^ΔZeb1). (K) Representative images of fibroblast and CAF markers (VIM, αSMA, C3, MHCII). (L) Individual cells were UMAP embedded based on their CAF marker intensities and the density distribution of cells from Fib^Ctrl or Fib^ΔZeb1 mice is shown. Data information: Scale bars represent 100 µm (K). Source data are available online for this figure.

Figure 3. ZEB1 is critically involved in myofibroblast differentiation and functionality.

(A) Representative IF staining of ZEB1 and αSMA in fibroblasts after in vitro recombination. (B) qRT-PCR analysis of myofibroblast markers Acta2 and Pdgfrb in fibroblasts after in vitro recombination, (n = 4/6; 2 independent fibroblast lines per genotype in 2/3 biological replicates for Acta2/Pdgfrb, Acta2: P = 0.0014, Pdgfrb: P = 0.0190, Student’s t test). (C) Gene set enrichment analysis using myCAF and extracellular matrix (GO: 0031012) signatures. (D) Representative images and quantification of collagen contraction assay (n = 5/4 independent Fib^Ctrl/Fib^ΔZeb1 lines; P < 0.0001, Student’s t test). (E) Quantification of relative wound area in Fib^Ctrl and Fib^ΔZeb1 mice during a skin wound healing model (n = 16/22), day 1: P = 0.0539, day 2: P = 0.0326, two-way ANOVA). (F, G) Representative images (F) and quantification (G) of Picrosirius red staining of tumor sections from Fib^Ctrl and Fib^ΔZeb1 mice after AOM/DSS and orthotopic transplantation (45/50 image sets derived from n = 13/9 Fib^Ctrl/Fib^ΔZeb1 mice for AOM/DSS and 84/76 image sets derived from n = 8/9 Fib^Ctrl/Fib^ΔZeb1 mice for the orthotopic tumor model, AOM/DSS: P = 0.0023, orthotopic: P = 0.0015, Mann–Whitney test). Areas from representative images are marked in green. (H) Quantification of T-cell migration through a transwell insert alone or with a layer of fibroblasts after in vitro recombination of Zeb1 (n = 7 independent lines, no Fib vs Fib^Ctrl: P < 0.0001, no Fib vs Fib^ΔZeb1: P = 0.0002, Fib^Ctrl vs Fib^ΔZeb1: P = 0.0012, Tukey’s multiple comparisons test). Data information: Data are presented as mean ± SD (B, D, G, H) or mean ± SEM (E). Scale bars represent 200 µm (A) or 80 µm (F). Source data are available online for this figure.

Figure 4. ZEB1 attenuates inflammatory signaling in fibroblasts and limits immune cell infiltration in multiple CRC models.

(A, B) IHC-based quantification of immune cell infiltration and PD-L1 expression in tumors from Fib^Ctrl and Fib^ΔZeb1 mice in the AOM/DSS model (A) and after orthotopic transplantation of AKP tumor organoids (B). Number of experimental mice per genotype are indicated (CD4, CD8, FOXP3, F4/80, B220, PD-L1: P = 0.0414, 0.2745, 0.0005, 0.3751, 0.0134, 0.0082 (AOM/DSS), P = 0.3383, 0.0068, 0.5852, 0.0401, 0.0122, 0.0545 (orthotopic), Student’s t test). (C) qRT-PCR analysis of Cxcl1 and Ccl2 mRNA expression in fibroblasts after in vitro recombination of Zeb1 and stimulation with IL1α (n = 4, two independent fibroblast lines per genotype in two biological replicates, 0, 5, 24 h: P = 0.9248, 0.0070, 0.1816 (Ccl2), P = 0.9998, 0.0337, 0.0890 (Cxcl1), Šídák’s multiple comparisons test). (D) Western blot of NFκB pathway activity in fibroblasts after 15 min of IL1α stimulation. β-ACTIN detection was used as a loading control. (E) IHC-based quantification of phospho-NFκB p65 (Ser536) in mice after AOM/DSS tumorigenesis or transplantation of AKP^re organoids. Numbers of experimental mice per genotype are indicated (AOM/DSS: P = 0.0435, orthotopic: P = 0.0238, Student’s t test). (F) Quantification of T-cell attraction to fibroblasts after in vitro recombination of Zeb1 (n = 7 independent lines, no Fib vs Fib^Ctrl: P = 0.0014, no Fib vs Fib^ΔZeb1: P < 0.0001, Fib^Ctrl vs Fib^ΔZeb1: P = 0.0043, Tukey’s multiple comparisons test). (G–I) Fib^Ctrl and Fib^ΔZeb1 fibroblasts were seeded in 3D ECM and expression of fibroblast subtype markers was quantified by qRT-PCR. Baseline marker expression of Fib^Ctrl and Fib^ΔZeb1 fibroblasts without additional treatment (G) and after treatment with IL1α (H) or TGFβ (I) relative to the respective untreated condition (n = 9/6 for Fib^Ctrl/Fib^ΔZeb1 fibroblast lines, Ccl2, Cxcl1, Acta2, Tagln: P = 0.6830, 0.8861, 0.0184, 0.0084 (baseline), P = 0.9318, 0.5463, 0.3182, 0.7161 ( + IL1α), P = 0.0670, 0.0077, 0.1469, 0.0346 ( + TGFβ), Student’s t test). Data information: Data are presented as mean ± SD (A–C, E–I). Source data are available online for this figure.

Figure 5. Loss of Zeb1 in fibroblasts enables response to ICB therapy.

(A) Kaplan–Meier analysis showing tumor-free survival of Fib^Ctrl and Fib^ΔZeb1 mice after orthotopic transplantation of AKP^re organoids and intraperitoneal injection of anti-(a)-PD-L1 antibodies or control IgGs, as indicated by red arrows. Recombination of Zeb1 was induced by tamoxifen food starting from day (d) 3. Mice were considered tumor-free if no tumor was detected by palpation. Numbers of experimental mice per condition are indicated (Fib^Ctrl-ICB vs Fib^ΔZeb1-ICB: P = 0.0189, Mantel–Cox test). (B) Representative H&E images and quantification of tumor volumes after orthotopic transplantation of AKP^re tumor organoids and ICB. Only tumors collected after d28 were included in this analysis. Number of experimental mice per condition are indicated. IgG-treated mice contributed to the initial AKP^re analysis (Fib^Ctrl-ICB vs Fib^ΔZeb1-ICB: P = 0.0397, Fib^ΔZeb1-IgG vs Fib^ΔZeb1-ICB: P = 0.0792, Šídák’s multiple comparisons test). (C, D) ICB in the AOM/DSS model, applied by intraperitoneal injection of a-PD-L1 and a-CTLA-4 antibodies or control IgGs. Starting at d70 of AOM/DSS tumorigenesis antibodies were administered two times/week to all mice with at least 25–30% colon obstruction. This time point corresponds to d0 of ICB. Representative endoscopic images at d0 and d35 of ICB in the AOM/DSS model. Dotted line indicates unobstructed areas (C). Quantification of colon obstruction relative to d0 (D). Numbers of experimental mice per condition are indicated (Fib^Ctrl-ICB vs Fib^ΔZeb1-ICB: two-way ANOVA, day 35: P = 0.0211. Data information: Data are presented as mean ± SD (B) or mean ± SEM (D). Scale bars represent 1 mm (B). Source data are available online for this figure.

Figure EV1. Loss of Zeb1 in fibroblasts does not affect morphology of primary tumors in the orthotopic transplantation model.

(A) IF images and quantification of ZEB1 expression in tumor stroma after orthotopic transplantation of AKP tumor organoids (n = 10/9 independent mice for Fib^Ctrl/Fib^ΔZeb1, P < 0.0001, Student’s t test). (B) Quantification of primary tumor engraftment in treatment-naïve Fib^Ctrl and Fib^ΔZeb1 mice after orthotopic transplantation of AKP tumor organoids. Numbers of experimental mice are indicated. (C) Tumor volume after orthotopic transplantation of AKP tumor organoids (n = 8/5 independent mice for Fib^Ctrl/Fib^ΔZeb1, P = 0.7584, Student’s t test). (D, E) Representative H&E stainings of AKP (D) and AKP^re (E) tumor sections. Top left corners show higher magnification of the indicated regions. (F–H) Analysis of tumors after orthotopic transplantation of AKP^re tumor organoids in treatment-naïve Fib^Ctrl and Fib^ΔZeb1 mice. (F) Tumor onset (n = 4/5 for Fib^Ctrl/Fib^ΔZeb1; P = 0.6401, Mantel–Cox test). (G) Quantification of tumor engraftment. Numbers of experimental mice are indicated. (H) Tumor volume (n = 4/4 independent mice for Fib^Ctrl/Fib^ΔZeb1). (I–K) Analysis of tumors after orthotopic transplantation of AKP^re tumor organoids in control IgG-treated Fib^Ctrl and Fib^ΔZeb1 mice. These mice are shown again as controls in Fig. 5. (I, J) Tumor onset (I) and quantification (J) after orthotopic transplantation of AKP^re tumor organoids (n = 12/14 for Fib^Ctrl/Fib^ΔZeb1, P = 0.4564, Mantel–Cox test). (K) Tumor volumes (n = 8/10 independent mice for Fib^Ctrl/Fib^ΔZeb1). Only tumors collected after day 28 were included. Data information: Data are represented as mean ± SD (A, C, H, K). Scale bars represent 50 µm (A) or 1 mm (D, E). Source data are available online for this figure.

Figure EV2. scRNA-seq in AOM/DSS and orthotopic models show reduced CAF diversity and impaired subtype-specific gene expression in Fib^ΔZeb1 tumors.

(A) Phenotypic annotation of AOM/DSS CAFs (n = 3 mice per genotype; corresponding to Fig. 2) upon integrated clustering by scoring of iCAF/myCAF/apCAF/mCAF/vCAF gene signatures (Bartoschek et al, ; Elyada et al, 2019) and displayed on UMAP Leiden clusters, pooled according to the determined scores. Note that ‘i/myCAFs’ share features of iCAFs and myCAFs and that the identities of 3 cell clusters could not be determined (ND) using these gene sets. (B) Genotype distribution of cells in AOM/DSS CAFs in DA regions (please refer to Fig. 2). The fraction of Fib^Ctrl and Fib^ΔZeb1 CAFs in each DA region or among all CAFs is shown. (C, D) Gene set enrichment analysis using Enrichr of marker genes from DAseq region 1 (C) and 3 (D) as compared to all other CAFs (Benjamini-Hochburg corrected Fisher’s exact test). (E) t-SNE sub-clustering of fibroblasts separately in Fib^Ctrl (left) and Fib^ΔZeb1 mice (right) of the orthotopic model showing less clusters in Fib^ΔZeb1. (F) Cluster similarities defined by cluster annotations based on ‘SingleR’ scores (see "Methods" for details) (Aran et al, 2019). Grayscale shows the log₂-transformed number of cells across clusters. Note the low similarity of Fib^ΔZeb1 cells with Fib^Ctrl clusters 2 and 5. (G) Heatmap showing the similarity (annotation scores) of gene expression in CAF clusters with published gene sets. Note, the high scores of Fib^Ctrl clusters 2 and 5 when compared with ‘iCAF’ and ‘myCAF’ signatures, respectively, and absence in Fib^ΔZeb1. Source data are available online for this figure.

Figure EV3. Multiplexed IF staining of tumor sections reveals an enrichment of iCAF-like cells in Fib^ΔZeb1 tumors.

(A, B) Single marker images (VIM, αSMA, C3, MHCII, EPCAM, CD45, DAPI) and merge of all channels for the representative images of Fib^Ctrl (A) and Fib^ΔZeb1 (B) orthotopic tumors shown in Fig. 2K. (C) Normalized staining intensities of CAF markers (αSMA, C3, MHCII) in UMAP embedding of CAFs from Fib^Ctrl and Fib^ΔZeb1 tumors. (D) Quantification of the distribution of CAF subtypes based on thresholds for αSMA (myCAF-like), C3 (iCAF-like) or MHCII (apCAF) staining intensity (n = 5/6 independent mice for Fib^Ctrl/Fib^ΔZeb1, myCAF: P = 0.0469, iCAF: P = 0.0499, Student’s t test). Data information: Data are presented as mean ± SD (D). Scale bars represent 100 µm (A, B). Source data are available online for this figure.

Figure EV4. Loss of Zeb1 in fibroblasts impairs tumor progression and increases T-cell infiltration in the invasive non-inflammation-driven AOM/p53 model.

(A) Schematic representation of the AOM/p53 model. (B) Quantification of numbers and volumes of tumors in the colons of Fib^Ctrl and Fib^ΔZeb1 mice at the endpoint (n = 15/22 for Fib^Ctrl/ Fib^ΔZeb1, number: P = 0.0433, volume: P = 0.0202, Mann–Whitney test). (C) Macroscopic evaluation of the most advanced/progressed tumor per mouse categorizing ‘T4’ as fully invasive (penetrating the muscle) or not (‘T1-T3’) (fraction of mice is given, n = 15/22 for Fib^Ctrl/Fib^ΔZeb1, P = 0.0252, Fisher’s exact test). (D) Representative H&E stainings with a higher magnification of the indicated region to the right. (E) IHC-based quantification of immune cell infiltration and stromal PD-L1 expression of tumors from Fib^Ctrl and Fib^ΔZeb1 mice. Numbers of experimental mice per genotype are indicated (CD4, CD8, FOXP3, B220, PD-L1: P < 0.0001, P = 0.0010, 0.3618, 0.3748, 0.0745, Student’s t test). (F–H) AOM/DSS model until day 50 in Fib^Ctrl and Fib^ΔZeb1 mice. Macroscopic analysis of early adenomas (number: P = 0.8339, volume: P = 0.2088, Student’s t test) (F) and IHC quantification of epithelial proliferation and cell death (KI67: P = 0.5403, cl. CASP3: P = 0.2773, Student’s t test) (G), as well as immune cell infiltration (H). Numbers of experimental mice per genotype are indicated (CD4, CD8, FOXP3, B220: P = 0.0296, 0.9515, 0.0524, 0.3996, Student’s t test). Data information: Data are presented as mean ± SD (B, E–H). Scale bars represent 1.5 mm (D, left) or 100 µm (D, right). Source data are available online for this figure.

Figure EV5. Monitoring of immune cell infiltration after ICB and effect of late-stage deletion of Zeb1 on ICB.

(A) IHC-based quantification of immune cell infiltration and PD-L1 expression in tumors from Fib^Ctrl and Fib^ΔZeb1 mice after orthotopic transplantation of AKP^re organoids and intraperitoneal injection of a-PD-L1 antibodies or control IgGs. Numbers of experimental mice per condition are indicated (CD4, CD8, FOXP3, B220, PD-L1: P = 0.0032, 0.0076, 0.6442, 0.0027, >0.9999 (Fib^Ctrl-ICB vs Fib^ΔZeb1-ICB), p = 0.1568, 0.0525, 0.0304, 0.2150, 0.0004 (Fib^ΔZeb1-IgG vs Fib^ΔZeb1-ICB), Šídák’s multiple comparisons test). IgG-treated mice contributed to initial AKP^re analysis (Fig. 1). (B) Kaplan–Meier analysis showing tumor-free survival of Fib^Ctrl and Fib^ΔZeb1 mice after orthotopic transplantation of AKP^re organoids and intraperitoneal injection of a-PD-L1 antibodies or control IgGs, as indicated by red arrows. Recombination of Zeb1 was induced by tamoxifen food starting from day (d) 14. Mice were considered tumor-free if no tumor was detected by palpation. Numbers of experimental mice per condition are indicated (P = 0.7583, Mantel–Cox test). (C) Quantification of tumor volumes after orthotopic transplantation of AKP^re tumor organoids and ICB with late recombination of Zeb1. Only tumors collected after d28 were included in this analysis (n = 8 independent mice, P = 0.6276, Student’s t test). (D) IF-based quantification of stromal cells expressing ZEB1 (n = 8/8 independent mice for Fib^Ctrl/Fib^ΔZeb1, P = 0.0002, Student’s t test). (E) IHC-based quantification of immune cell infiltration and PD-L1 expression in AOM/DSS tumors from Fib^Ctrl and Fib^ΔZeb1 mice receiving ICB (2 intraperitoneal injections of a-PD-L1 and a-CTLA-4 antibodies or control IgGs per week starting at d70 of AOM/DSS tumorigenesis). Numbers of experimental mice per condition are indicated (CD4, CD8, FOXP3, B220, PD-L1: P = 0.9902, 0.0802, <0.0001, 0.0070, 0.0003 (Fib^ΔZeb1-IgG vs Fib^ΔZeb1-ICB), P = 0.6241, 0.7615, 0.0070, 0.3163, 0.0144 (Fib^Ctrl-IgG vs Fib^ΔZeb1-IgG), Šídák’s multiple comparisons test). Data information: Data are presented as mean ± SD (A, C–E). Source data are available online for this figure.