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

ZEB1 is expressed heterogeneously in the TME of both human and murine tumors (Fig. 1A,B), and high expression of ZEB1 in stromal cells of pancreatic cancer is a prognostic factor for poor survival in patients (Bronsert et al, 2014). To investigate a functional involvement, we generated fibroblast-specific Zeb1-deleted mice (Fib^ΔZeb1), by combining the Zeb1^fl/fl genotype (Brabletz et al, 2017) with the constitutive Col6a1-Cre transgene (Koliaraki et al, 2015) or with the tamoxifen-inducible Col1a2-CreERT2 knock-in allele (Zheng et al, 2002). Cre recombinase activity in both lines was restricted to fibroblasts as identified by Cre reporter alleles (Appendix Fig. S1A,B), and deletion of Zeb1 was confirmed at the genomic (Appendix Fig. S1C) and protein level in the colon of both models (Appendix Fig. S1D,E). Of note, loss of Zeb1 in fibroblasts did not compromise colon morphogenesis or homeostasis (Appendix Fig. S1F,G).

Fib^ΔZeb1 mice were subjected to the inflammation-driven AOM/DSS model (Neufert et al, 2007) (Fig. 1C), where deletion in fibroblasts did not affect overall or tumor-free survival of mice (Appendix Fig. S2A,B). Intriguingly, endoscopic scoring revealed increased colonic obstruction (Fig. 1D), which was confirmed by significantly increased individual and total adenoma volumes and a trend to increased tumor numbers in Fib^ΔZeb1 mice at endpoints (Fig. 1E,F; Appendix Fig. S2C). Histologic differentiation was not affected by loss of stromal Zeb1 (Appendix Fig. S2D), but endoscopic analysis showed slightly increased colonic inflammation (Appendix Fig. S2E–G). By immunohistochemistry (IHC) fewer ZEB1-positive stromal cells and increased epithelial proliferation accompanied by slightly reduced cell death was observed in Fib^ΔZeb1 mice (Fig. 1G,H). These data demonstrate that loss of Zeb1 in fibroblast promotes inflammation-driven adenoma growth.

Sporadic CRC progression was modeled by orthotopic transplantation of tumor organoids (Fumagalli et al, 2018) into Fib^ΔZeb1 and Fib^Ctrl mice (Fig. 1I). Tumor organoids were genetically engineered from normal colonic organoids to harbor mutations in Apc, Tp53 and Kras loci (Apc^Δ/Δ, Kras^G12D, Tp53^Δ/Δ; AKP) and either transplanted directly into syngeneic mice, or re-cultured upon one round of orthotopic growth to promote further tumor progression in vivo (AKP^re). AKP^re tumors indeed displayed more aggressive tumor growth with earlier onset and less differentiated morphology. However, deletion of Zeb1 in fibroblasts did not affect the engraftment of organoids, overall survival, primary tumor size, and tumor morphology upon transplantation of AKP (Figs. 1J and  EV1A–D) or AKP^re organoids (Fig. EV1E–K). Strikingly, spontaneous metastasis to the liver was decreased in Fib^ΔZeb1 mice, regardless of AKP or AKP^re transplantation, as reflected in the fraction of metastasis-bearing mice and the number of metastases per mouse (Fig. 1K,L), indicating a pro-metastatic role of ZEB1 in CAFs of sporadic and progressed CRC. Overall, these findings suggest that ZEB1 in fibroblasts regulates colon cancer initiation and progression in a tumor context- and stage-dependent manner.

We applied scRNA-seq to gain insights into the cellular heterogeneity of CAFs and the transcriptional changes upon Zeb1 loss. AOM/DSS tumors were enzymatically dissociated, and CAFs were enriched by fluorescence-activated cell sorting (FACS) of CD326 (EPCAM)−, CD45−, and CD31− cells before sequencing (SORT-seq, Fig. 2A, left). Integrated clustering of cells from both genotypes was performed, followed by annotation using previously reported CAF signatures (Bartoschek et al, 2018; Elyada et al, 2019) (Fig. EV2A). Differential abundance analysis (Zhao et al, 2021) was performed for unbiased identification of genotype-specific differences. Prominent abundance differences were observed in two regions that were enriched for myCAF and iCAF-specific genes (Fig. 2B–D). Fib^Ctrl cells were overrepresented in the myCAF region and Fib^ΔZeb1 cells in the iCAF region (Fig. EV2B–D). Genotypes were studied independently to avoid that Fib^ΔZeb1 cells are influenced by Fib^Ctrl cells. Here, walktrap clustering resulted in eight and six CAF clusters in Fib^Ctrl and Fib^ΔZeb1 tumors, respectively (Fig. 2E). Cross-correlation of transcriptomes indicated ambiguity in Fib^ΔZeb1 clusters number 1 and 4 (Fig. 2F). Furthermore, no Zeb1-deficient CAFs were matched to Fib^Ctrl cluster 8, altogether suggesting impaired diversification and subtype specification in Fib^ΔZeb1 CAFs. The transcriptomes of Fib^Ctrl CAFs correlated well with “iCAF” and “myCAF” archetypes described in pancreatic cancer (Elyada et al, 2019; Ohlund et al, 2017) (Fig. 2G). In contrast, Fib^ΔZeb1 CAFs displayed less clusters with myCAF-like identity.

We also examined non-inflammation-driven orthotopic tumors by scRNA-seq. Following FACS, CAFs were analyzed in parallel to isolated CD45+ immune and EPCAM+ tumor cells (Fig. 2A, right). A separate analysis of genotypes resulted in two CAF clusters for Fib^Ctrl but only one for Fib^ΔZeb1 tumors (Fig. 2H). Individual analysis of CAFs confirmed that loss of Zeb1 limits the CAF repertoire (Fig. EV2E) and cross-annotation confirmed increased ambiguity in Fib^ΔZeb1 (Fig. EV2F). We observed a higher correlation of CAF clusters with “iCAF” and “myCAF” archetypes in Fib^Ctrl compared to Fib^ΔZeb1 (Fig. EV2G). Consistent with AOM/DSS CAFs, differential gene expression analysis between Fib^Ctrl and Fib^ΔZeb1 CAFs showed reduced signatures for ECM organization and increased inflammatory terms (Fig. 2I,J). In summary, our data show that independent of the CRC model, CAF diversification and specification is strongly impaired when ZEB1 is absent.

To study the differentiation defects in situ, we performed multiplex immunofluorescence staining. In orthotopic tumors, fibroblasts were defined as VIM+, CD45−, EPCAM− cells, and CAFs were further characterized by αSMA, C3, and MHCII staining (Figs. 2K and  EV3A,B). Quantitative image analysis and dimensionality reduction allowed classification into myCAF, iCAF, and apCAF-enriched populations (Fig. EV3C). While the majority of cells in Fib^Ctrl displayed a myofibroblast-like phenotype, CAFs in Fib^ΔZeb1 tumors were predominated by the inflammatory subtype (Fig. 2L). Here, Zeb1 loss did not affect apCAFs or result in a mixed αSMA+ /C3+ iCAF/myCAF identity (Fig. EV3A–D).

We next studied the functional consequences of Zeb1 loss and CAF polarization. We established adherent colon fibroblast cultures from Zeb1^fl/fl mice, which showed a typical myofibroblast-like morphology. Induction of Cre recombinase activity resulted in efficient Zeb1 deletion in vitro, which was confirmed by immunofluorescence staining (Fig. 3A). Of note, Zeb1 deficiency-induced loss of fibrillary αSMA staining, demonstrating reduced myofibroblast differentiation and CAF activation (Sahai et al, 2020). Consistently, Zeb1-deleted fibroblasts showed lower expression of myofibroblast markers by qRT-PCR and bulk RNA sequencing analyses (Fig. 3B,C) and strongly reduced contraction in collagen (Fig. 3D). In contrast, we observed no major changes in proliferation or senescence of Zeb1-deficient fibroblasts (Appendix Fig. S3A,B). To study myofibroblast-specific functions in vivo, skin wound healing assays revealed delayed wound closure in Fib^ΔZeb1 compared to Fib^Ctrl mice (Fig. 3E; Appendix Fig. S3C). Furthermore, the impaired myCAF polarization in AOM/DSS adenomas and orthotopic tumors was associated with reduced deposition of collagen (Fig. 3F,G). Since collagenous ECM can constitute a physical barrier for immune cell infiltration, we conducted an in vitro migration assay to test the ability of fibroblasts to block the passage of mouse splenocytes (Fig. 3H). Strikingly, Fib^ΔZeb1 fibroblasts showed a pronounced barrier defect. Collectively, these data demonstrate the fundamental role of ZEB1 in myofibroblast specification and function.

To clarify whether the inflammatory polarization in combination with the reduced collagen deposition of Zeb1-deleted CAFs affects tumor immune cell infiltration, we performed IHC analyses. Strikingly, in inflammation-induced AOM/DSS tumors, we found increased infiltration of CD4+ T cells and FOXP3+ Tregs, but not of CD8+ cells or F4/80+ macrophages into Fib^ΔZeb1 tumors (Fig. 4A), indicating modulation of adaptive anti-tumor immunity. Concomitantly, increased B-cell infiltration and an overall upregulation of the immune checkpoint molecule PD-L1 were observed, suggesting tolerance induction upon T-cell activation. Similar to AOM/DSS adenomas, increased T- and B-cell infiltration and expression of PD-L1 was observed in Fib^ΔZeb1 orthotopic tumors, reflecting sporadic CRC (Fig. 4B). Yet, in contrast to AOM/DSS tumors, CD8+ T cells and macrophages were enriched. Since the primary tumor size was comparable in Fib^Ctrl and Fib^ΔZeb1 orthotopic tumors (Fig. EV1C,H,K), the observed T-cell infiltration apparently did not result in an effective anti-tumor response, presumably due to T-cell inactivity.

To investigate if the differences between both models can be explained by the inflammatory environment in the AOM/DSS model, we employed another model of sporadic CRC, in which AOM-induced mutagenesis in combination with epithelial cell-specific Tp53 deletion results in invasive CRC (Neufert et al, 2021; Schwitalla et al, 2013) (Fig. EV4A). Here, deletion of Zeb1 in fibroblasts resulted in fewer, smaller and less invasive tumors, indicating delayed tumor progression in Fib^ΔZeb1 mice in the AOM/p53 model (Fig. EV4B–D). We observed enhanced infiltration of CD4+ and CD8+ cells but no and only trending increase in FOXP3+ Tregs and stromal PD-L1 levels, respectively (Fig. EV4E). Collectively, our data show a context-dependent immunomodulatory role of ZEB1 expression in fibroblasts. This notion was supported by analysis of AOM/DSS tumors at day 50 during the acute phase of inflammation. Even before overt tumor outgrowth, significantly increased CD8+ and FOXP3+ cell infiltration was observed (Fig. EV4F–H), suggesting that the acute inflammation triggers a compensatory regulation of the immune microenvironment upon Zeb1 loss in fibroblasts.

To explore the mechanism for the altered immune cell infiltration in Zeb1-deficient CAFs, we tested the response of cultured fibroblasts to the chemokine IL1α, a crucial stimulator of inflammatory polarization in CRC CAFs (Nicolas et al, 2022). We monitored basal and IL1α-induced NFκB activation in cultured Fib^Ctrl and Fib^ΔZeb1 fibroblasts by qRT-PCR and Western blot analyses. Upon stimulation, Fib^ΔZeb1 fibroblasts showed increased expression of NFκB targets including Ccl2 and Cxcl1 that both have been described as iCAF markers (Biffi et al, 2019; Elyada et al, 2019; Nicolas et al, 2022) (Fig. 4C). Consistently, strongly increased levels of phosphorylated NFκB and IκBα were observed in lysates of Fib^ΔZeb1 fibroblasts that were stimulated with IL1α for 15 min (Fig. 4D), when Fib^Ctrl cells already partially resolved the signal, suggesting an overshooting pathway activation. Indeed, immunostaining of phospho-NFκB showed increased pathway activity in both Fib^ΔZeb1 AOM/DSS and orthotopic tumors (Fig. 4E). Furthermore, secretome analysis in cultured CAFs from AOM/DSS tumors identified increased basal levels of CCL2 upon Zeb1 loss (Appendix Fig. S4A,B), which was confirmed by IHC (Appendix Fig. S4C). Notably, increased phospho-NFκB and CCL2 staining were also observed during the acute inflammation in Fib^ΔZeb1 mice at day 50 of AOM/DSS treatment (Appendix Fig. S4D,E). Our data suggest that elevated cytokine production from Zeb1-deficient CAFs may act as a potent lymphocyte chemoattractant. Thus, the influence of Fib^Ctrl or Fib^ΔZeb1 fibroblasts on T-cell recruitment was tested in a transwell migration assay using stimulated T cells in co-culture with fibroblasts, where migration was significantly increased towards a Fib^ΔZeb1 fibroblast monolayer (Fig. 4F).

As a plastic cell population, the observed changes could result from an intrinsic differentiation bias, or a failure to respond to external signals. To tackle this question, we seeded CAFs in a 3D ECM, representing a quiescent status, or applied IL1α or TGFβ treatment, to direct them into the myCAF and iCAF lineages, respectively (Ohlund et al, 2017). qRT-PCR showed that under baseline conditions Fib^ΔZeb1 expressed unaffected levels of iCAF markers (Ccl2 and Cxcl1) but significantly reduced levels of myCAF markers (Acta2 and Tagln) (Fig. 4G). While the induction of inflammatory genes by IL1α was comparable in this setting (Fig. 4H), after TGFβ addition Fib^ΔZeb1 CAFs showed a striking defect to downregulate inflammatory genes (Fig. 4I). In contrast, the TGFβ-mediated induction of myCAF markers was less prominently affected. Collectively, our data indicate that ZEB1 limits the sensitivity of fibroblasts to NFκB-activating signals, which causes an iCAF bias affecting tumor immune infiltration.

Given the refractoriness of non-hypermutated CRC to ICB therapy, the increased immune cell infiltration and induction of checkpoint molecules observed in Fib^ΔZeb1 mice may point to a strategy to render MSS tumors more sensitive. We investigated this hypothesis in orthotopic and AOM/DSS tumors. Following transplantation of AKP^re organoids, tumor-bearing mice were injected subcutaneously with anti-PD-L1 antibodies for 3 weeks. Strikingly, anti-PD-L1 therapy but not control IgG significantly delayed tumor growth of Fib^ΔZeb1 in comparison with Fib^Ctrl mice, eventually resulting in smaller Fib^ΔZeb1 tumors at the endpoint (Fig. 5A,B). In addition, infiltration of CD4+ , CD8+ T cells, and B cells was further increased by ICB in tumors from treated Fib^ΔZeb1 mice, whereas FOXP3+ cells were not affected. PD-L1 expression was reduced in Fib^ΔZeb1 mice upon ICB, indicating successful reactivation of the adaptive immune response (Fig. EV5A). In a more clinically relevant setting, we explored whether ICB sensitivity is also induced by later Zeb1 deletion, when tumors have already formed. For this purpose, we inactivated Zeb1 in orthotopic tumors simultaneously with ICB administration (Fig. EV5B–D). Although not significant, a similar trend towards smaller tumors was observed in Fib^ΔZeb1 mice, when Zeb1 was depleted from a preformatted immune TME. In autochthonous AOM/DSS tumors, we employed dual ICB targeting PD-L1 and CTLA-4, as we found a substantial increase in the number of Tregs in tumors of Fib^ΔZeb1 mice in this model (Fig. 4A). Tumor growth was monitored longitudinally via endoscopy, and adenoma-bearing mice were subjected to ICB after recovery from cyclic DSS-induced inflammation at day 70. Strikingly, dual ICB abrogated tumor growth in Fib^ΔZeb1 mice combined with a more robust increase in CD8+ T cells, ablation of FOXP3+ cells, and reduced PD-L1 expression (Figs. 5C,D and  EV5E). Together, our analyses demonstrate that loss of Zeb1 in fibroblasts induces sensitivity to ICB in both colitis-induced and sporadic tumor models.

Animal husbandry and all experiments were performed according to the European Animal Welfare laws and guidelines. The protocols were approved by the committee on ethics of animal experiments of Bavaria (Regierung von Unterfranken, Würzburg; TS-18/14, 55.2-DMS-2532-2-270, -2-952, and -2-1133) and of Hessen, Germany (Regierungspräsidium Darmstadt F123/1031, F123/1040, and F123/2001). Power analysis was used to calculate the sample size required for animal experiments. Animals were kept on a 12:12 h light-dark cycle and provided with food and water ad libitum in the animal facilities of the Friedrich-Alexander University of Erlangen-Nürnberg and the Georg-Speyer-Haus Frankfurt. Col1a2-CreERT2^Tg/+;Zeb1^fl/fl and Col6a1-Cre^Tg/+;Zeb1^fl/fl mice were kept on FvB and C57BL/6 backgrounds, respectively. Col6a1-Cre^Tg/+;Zeb1^fl/fl;Vil-Flp^Tg/+;Tp53^FRT/FRT Fib^ΔZeb1 mice were kept on a mixed FvB/C57BL/6 background.

Rosa26-tdTomato (Madisen et al, 2010); RRID:IMSR_JAX:007914) and Rosa26-mTmG (Muzumdar et al, 2007); RRID:IMSR_JAX:007576) mice have been described previously. Conditional Zeb1 knockout mice (Brabletz et al, 2017); MGI:5901939) were crossed with mice expressing Col6a1-Cre (Armaka et al, 2008); GI:3775430) or Col1a2-CreERT2 (Zheng et al, 2002); RRID:IMSR_JAX:029567) to generate Cre-positive Fib^ΔZeb1 mice or Cre-negative Fib^Ctrl mice. Age-matched littermates of both sexes were used for experiments and randomly assigned to groups within genotypes. Genotyping primer sequences are listed in Appendix Table S1.

AOM/DSS driven adenoma were induced as described (Neufert et al, 2007). Briefly, mice were injected intraperitoneally (i.p.) with 10 mg/kg of azoxymethane (AOM, Sigma, A5486) in 0.9% NaCl), prior to administration of three cycles of 1.75% dextran sulfate sodium (DSS, MP Biomedicals, SKU-0216011080) in the drinking water ad libitum, each interrupted by 2 weeks of regular water and finally sacrificed at day 84 or at ethical endpoints. A separate cohort of mice was sacrificed at day 50 at the peak of inflammation. Colonic inflammation and tumor growth was monitored by endoscopy under isoflurane anesthesia employing the Colorview endoscopic system (Karl Storz), as described before (Becker et al, 2006), expressed as numbers of adenoma and/or the maximal colon obstruction by the adenoma in percent of the full colonic diameter at the most blocked site. Colonic inflammation was visually scored from ‘0’ (zero) to ‘4’ (four), with ‘0’ representing normal, ‘1’ signs of thickening, increased vascularity, and/or fibrin visible, moderate granularity and/or stool being still shaped, ‘2’ stronger thickening, vascularization, granularity and unshaped stool (no blood), ‘3’ loss of transparency, extreme granularity, rectal blood (clots) and/or traces of blood in the mucosa and/or in the unshaped or spread stool, and ‘4’ (partially) liquid stool, several sites of bloody mucosal damage or non-rectal liquid blood in the colon. Stool consistency was scored individually from ‘1’ to ‘4’ as well, with ‘0’ being normal and solid, ‘1’ softer than usual, ‘2’ unshaped, ‘3’ almost liquid but containing solid pieces of stool and/or containing traces of (clotted) blood, ‘4’ being liquid and/or mostly bloody. For immune checkpoint inhibition, mice were injected i.p. with anti-PD-L1 (BXC-BE0101) and anti-CTLA-4 (BXC-BE0131) antibodies or isotype controls (BXC-BE0094 and BXC-BP0087) twice per week (10 mg/kg each, Bio X cell). Therapies started at day 70 for all mice, unless colon obstruction by adenomas did not reach 25–30%, as determined by endoscopy. Adenoma growth was then monitored weekly via endoscopy, and mice were sacrificed at day 105.

The AOM/p53 model was based on the previously reported Tp53ΔIEC model (Neufert et al, 2021; Schwitalla et al, 2013), using Villin-Flp mediated recombination and intestinal epithelium-specific inactivation of biallelic FRT-flanked Tp53 (intron 1 and 10). Experimental animals were generated by sequential crossings to generate experimental Col6a1-Cre^Tg/+;Zeb1^fl/fl;Vil-Flp^Tg/+;Tp53^FRT/FRT Fib^ΔZeb1;Δp53 and Col6a1-Cre^+/+;Zeb1^fl/fl;Vil-Flp^Tg/+;Tp53^FRT/FRT Fib^Ctrl;Δp53 littermates. AOM/p53 tumors were induced by six weekly doses of AOM (10 mg/kg, i.p.) in 0.9% NaCl, as described previously (Neufert et al, 2021; Schwitalla et al, 2013). Tumor growth was monitored by endoscopy as in the AOM/DSS model and animals were sacrificed after 161 days or at any other ethical endpoint.

Orthotopic transplantation of tumor organoids was performed as described (Fumagalli et al, 2018). In brief, AKP or AKP^re organoids were mechanically dissociated, seeded in high-concentration collagen pads (6.5 mg/mL) and allowed to recover for 48 h. Collagen pads were transplanted below the muscularis externa of the mouse cecum, and serosal wounds were covered by an anti-adhesion barrier. Recombination of Zeb1 was induced at day 14 for AKP organoids and at day 4 for AKP^re organoids by 400 mg/kg tamoxifen diet (custom order based on Altromin 1824P). Tumor growth was monitored three times a week by palpation, and mice were sacrificed when tumors exceeded 1 cm in diameter. For immune checkpoint inhibition, mice were injected i.p. with anti-PD-L1 (Bio X cell, BE0101) or matched control antibodies (Bio X cell, BE0090) at days 14, 21, and 28. To ensure equal treatment of mice, only tumors collected after day 28 were considered for size comparison. For AKP^re transplantation, only a small batch of untreated mice was used, and most analyses were performed in mice treated with control antibodies as mentioned in the figure legends for the reduction of mouse numbers.

Wound closure of a skin excision wound in mice was determined for 10 days (Lin et al, 2011). Briefly, circular wounds were applied under sterile conditions in anaesthetized mice into the dorsum after depilation. The dorsal skin was lifted at the midline and punched through two layers of skin using a disposable biopsy punch (6 mm in diameter, Kai medical, Solingen #BP-60F). Wound size was determined at indicated time points with a digital calliper and by transferring wound outline to a transparent film for area calculation using ImageJ.

Primary mouse intestinal organoid cultures were established as reported (Fan et al, 2019). Briefly, mouse colons were cut into small fragments, washed with PBS, and epithelial crypts were dissociated by incubation with 10 mM EDTA, passed through a 100 µm mesh and collected by centrifugation. Colon crypts were seeded at high density in BME (R&D Systems, 3533-010-02) and cultured in Advanced DMEM/F12 (Thermo Fisher Scientific, 12634028) with 20% Wnt3a conditioned medium, 10% Noggin conditioned medium, 10% R-spondin1 conditioned medium, B27 supplement (Thermo Fisher Scientific, 17504-044), 500 mM N-Acetylcysteine (Merck, A9165), 500 µg/mL human EGF (Peprotech, AF-100-15) and 500 µM A83-01 (Tocris, 2939/10). Established organoids were passaged twice a week by mechanical dissociation and reseeding.

For generation of tumor organoids, colons from Lox-Stop-Lox-Kras^G12D mice were used (Jackson et al, 2001). Both Apc and Tp53 alleles were mutated by co-transfection of Cas9 with the respective sgRNA plasmids. The oncogenic Kras^LSL.G12D allele was recombined by pAC1-Cre (ATCC, 39532) transfection. Modified lines were clonally expanded, and successful modification confirmed by Sanger sequencing. The hCas9 and gRNA_GFP-T2 plasmids were a kind gift from George Church (Addgene plasmids #30205 and #41820) and the Apc sgRNA plasmid kindly provided by Hans Clevers (Schwank et al, 2013). sgRNA oligonucleotides targeting Tp53 (sense: 5’-GTTTTAGAGCTAGAAATAGCAAG-3’ antisense: 5’-AGTGAAGCCCTCCGAGTGTCGGTGTTTCGTCCTTTCCACAAGAT-3’) were inserted into gRNA_GFP-T2 plasmid by inverse cloning as published (Schwank et al, 2013).

Tumor organoids from established tumors were obtained by enzymatic digestion of primary tumors with 0.1% Collagenase I (Thermo Fisher Scientific, 17018029), 0.2% Dispase II (Merck, D4693) and 2 U/mL DNase I (New England Biolabs, M0303L). Cells were passed through a 40 µm mesh and seeded at high density in BME.

For fibroblast cultures, normal colon fragments after epithelial crypt isolation or minced AOM/DSS tumors were washed in PBS and further incubated with 10 mM EDTA. After 1 h, tissue fragments were seeded into 10-cm culture plates coated with 0.1% gelatine (Sigma, G2500) in DMEM (Thermo Fisher Scientific, 31966-047) with 10% FBS (Sigma, F7524) and Pen/Strep (Thermo Fisher Scientific, 15140-122). After several days, fibroblasts expanded from tissue fragments attached to the plates, which were collected by trypsinization and propagated as regular two-dimensional cell cultures on gelatine-coated plates. Rates of proliferation were analyzed by plating 200 fibroblasts in hexaplicates in 96-wells grown for 10 days, respectively, and cell confluence was determined by live-cell microscopy using an IncuCyte S3 instrument (Sartorius). Cell senescence was determined by the Senescence β-Galactosidase Staining Kit (Cell Signaling, 9860) 3 days after plating 1 × 10⁴ fibroblasts in six-well plates according to the manufacturer’s instructions. SA-β-Gal positive and total cell numbers were counted from 15 random images each (>700 cells per line).

For the collagen contraction assay, 1 × 10⁵ fibroblasts were seeded in 500 µL Collagen I solution (Thermo Fisher Scientific, A1048301) adjusted to 1 mg/mL according to the manufacturer’s instructions into a 24-well suspension plate. After polymerization of collagen, the collagen disks were detached from the plate and contraction was determined by monitoring the area of the discs over time.

For the barrier function and T-cell attraction assays, splenocytes were isolated from OT1 mice (Hogquist et al, 1994). T cells were stimulated by the addition of SIINFEKL peptide (AnaSpec, AS-60193-1) for 4 h and expanded for 3 days prior to an experiment. For the barrier function assay, 12-well transwells with 8-µm pore size (Greiner, 665638) were coated with 0.1% gelatine, and 2 × 10⁵ fibroblasts were seeded on top of the transwell. After 48 h, 3 × 10⁴ T cells were added on top of the confluent fibroblast layer and cells in the lower compartment were monitored microscopically. For the T-cell attraction assay, 1 × 10⁵ fibroblasts were seeded into a 12-well adhesion culture plate. After 24 h, a transwell with 3-µm pore size (Corning, CLS3462-48EA) containing 3 × 10⁴ T cells was added to the well, and attracted cells in the lower compartment were monitored microscopically.

For inflammatory activation, colon fibroblasts were plated in a six-well plate and after 24 h treated with 1 ng/mL murine recombinant IL1α (Biolegend, 575002) in PBS for the indicated periods. The medium was replaced by 1 mL of fresh medium 2 h before treatment and supplemented with 1 mL 2 ng/mL IL1α containing medium at the starting point. At indicated time points, cells were harvested and processed for RNA and protein isolation.

For analysis in 3D, 2 × 10⁵ fibroblasts were seeded in 8 × 25 µL of BME and cultured in Advanced DMEM/F12. Cells were analyzed either untreated or treated with 10 ng/mL recombinant hTGFβ (Peprotech, 100-21) or 1 ng/mL recombinant mIL1α. Medium was replaced by fresh medium every 24 h and cells were harvested and processed for RNA isolation after 96 h.

Tumors of orthotopic transplantation were dissected and fixed in 4% PFA/PBS overnight at 4 °C. For histological and IHC analyses, paraffin-embedded tumors were sectioned at 3 µm and stained using the Bond-Max device (Leica) and the Bond Polymer Refine Detection system (Leica, DS9800). Sections were imaged using an Aperio CS2 digital pathology slide scanner (Leica) and marker expression was quantified with macros in Aperio ImageScope (v12.4).

AOM/DSS and AOM/p53 colons were collected, flushed with PBS and longitudinally opened for imaging and determining tumor sizes and numbers using a caliper. Colons were mounted as Swiss Rolls, fixed in 4% PFA/PBS overnight at 4 °C, paraffin-embedded, sectioned at 3–4 µm and subjected to hematoxylin/eosin (H&E), IHC or IF staining as described previously (Krebs et al, 2017). After antibody incubations, washing steps, and DAB reactions, the slides were counterstained with Mayer’s hematoxylin before dehydration and mounting (Roti®-Histokitt, 6638.2). Analysis and image acquisition was performed using a Leica DM5500B microscope. Scoring of infiltration into AOM/DSS tumors was estimated from the cellularities of infiltrated cell-type marker-positive cells on IHC slides, with none (0), rare (1), few (2), several (3), and many/abundant (4) classification. For anti-ZEB1 IF of Fib^Ctrl/Fib^ΔZeb1 (Col6a1-Cre) mice, cryosections from fresh frozen colon specimens were fixed in 4% PFA for 10 min, permeabilized for 10 min in 0.25% Triton X-100/PBS, blocked in 3% BSA/PBS and incubated with anti-ZEB1 antibodies. After washing and incubation with Alexa594-conjugated secondary antibodies, DAPI-stained sections were mounted (Antifadent AF1, Citifluor). Images were acquired using a Leica DM5500B microscope.

For immunofluorescence staining of tumors from orthotopic transplantation, citrate-based antigen retrieval was applied with deparaffinized sections. Slides were blocked with 20% goat serum (Merck, G9023) in PBST and stained for 1 h with primary antibodies and for 30 min with secondary Alexa488/647-conjugated antibodies. Sections were imaged using an Evos FL microscope and marker expression was quantified using CellProfiler (McQuin et al, 2018). Antibodies and dilutions are listed in Appendix Table S2.

FFPE sections (3 μm) from orthotopic tumors (AKP) were stained with Opal 7‐Color Automation IHC Kits (Akoya Bioscience) in the BOND‐RX Multiplex IHC Stainer (Leica) following established protocols (Strack et al, 2020). Each section was put through 6 sequential rounds of staining, which included blocking in 5% BSA followed by incubation with primary antibodies; for staining conditions see Appendix Table S2. Nuclei were counterstained with 4′,6‐diamidino‐2‐phenylindole (DAPI) contained in the Opal 7‐Color Automation IHC Kits and slides were mounted with Fluoromount‐G (SouthernBiotech). Imaging was performed with the PhenoImager HT imaging system (Akoya Bioscience). Tumor regions of interest were defined manually, and out-of-focus images or high background signals were excluded from further analysis. Images were analyzed using the phenotyping application of the ‘inForm’ software V2.54.10 (Akoya Bioscience) and the fluorescence intensities for individual cells were exported. Epithelial cells (33 ± 6%) and immune cells (12 ± 5%) were excluded by EPCAM or CD45 intensity thresholds, respectively. CAFs (7 ± 2%) were separated from other cells by thresholds of αSMA, C3 or MHCII. Assigned phenotypes were myCAFs (αSMA+ , C3−, MHCII−), iCAFs (C3+ , αSMA−, MHCII−), apCAFs (MHCII+ , αSMA−, C3−) and mixed myCAF/iCAF identity (αSMA+ , C3+ , MHCII−). Cell numbers were then exported and statistically analyzed by Student’s t test. Normalized intensity values of αSMA, C3 and MHCII were used to plot CAFs by UMAP (umap V0.2.10.0) and to display the staining intensities and the cell density distribution (ndensity, ggplot2 V3.4.2).

Slides were deparaffinized, rehydrated, and then incubated in Picrosirius Red solution (abcam, ab246832) for 1 h at RT before washing in 0.5% acidified water, dehydration, and mounting (Roti®-Histokitt, 6638.2). Polarized light imaging was done by using a Leica DM5500B microscope equipped with a polarization filter to monitor green thinner collagen fibers and yellow-orange bundled fibers by refraction and birefringence. Percentages of total stained areas were analyzed on polarized light images using CellProfiler by RGB conversion, global background-based thresholding for pixel identification, followed by binary image generation of each channel for measuring area occupied by pixels.

Protein extraction and western blotting were carried out as described (Krebs et al, 2017; Schuhwerk et al, 2022). Briefly, cells grown in 6-well plates were lysed in 150 mM NaCl, 50 mM Tris-HCl pH 8.0, 0.5% Na-Desoxycholate (w/v), 0.1% SDS (w/v), 1% NP40 (v/v), 1× complete protease inhibitor (Roche, 4693132001), 1 mM PMSF, 1× PhosSTOP (Roche, 4906837001) for 20 min at 4 °C. Protein concentrations were determined by using the BCA Protein Assay (Thermo Fisher Scientific, 23225) according to manufacturer’s instructions. Up to 30 µg of protein lysate was separated by SDS-PAGE and transferred to nitrocellulose membranes before antibody incubation. Detection was carried out using Western Lightning Plus-ECL (Perkin Elmer, NEL103001EA) and a ChemiDoc^TM Imaging System (BioRad). Antibodies and dilutions are listed in Appendix Table S2.

For secretome analysis, a commercially available secretome array kit was used (R&D systems, ARY028) according to the manufacturer’s instructions. Samples from AOM/DSS CAF supernatants were prepared by surgically detaching adenomas and mincing using scalpels and incubating the tissue in digestion buffer containing 0.05% Collagenase D (w/v), 0.3% Dispase II (w/v), 0.05% DNase I (w/v), 4% FBS (v/v), in DMEM/F12 medium (Thermo Fisher Scientific, 31331028) for 30-45 min at 37 °C with constant agitation. In total, 10 mL of cold washing buffer (sterile PBS containing 2% FBS) was added, and the suspension passed through a 70 µm strainer. After centrifugation and erythrocyte lysis in ACK buffer (150 mM NH₄Cl, 10 mM KHCO₃, 0.1 mM EDTA; pH 7.2-7.4) for 2 min at RT, washed and collected cells were resuspended in DMEM/F12/10% FBS and plated in 12-well plates. Adherent cells were passaged in 1:1 ratios when reaching confluence. After 2–3 passages, the supernatants of individual confluent 6-well vessels were collected and kept on ice. After centrifugation at 1000 × g for 5 min at 4 °C, supernatants were aliquoted and frozen at −80 °C. For the secretome array, samples were added to the equilibrated and blocked membranes for incubation overnight at 4 °C. After incubation with the detection antibody cocktail for 1 h, with the Streptavidin-HRP mix for 30 min and the detection mix for 1 min at RT, secretomes were detected by imaging like for western blot. Intensities were normalized to the background of each respective membrane.

Total RNA of cultured cells was isolated using the RNeasy Plus Mini Kit (Qiagen, 74136) or the NucleoSpin RNA kit (Macherey-Nagel, 740955.250) and 200–500 ng of total RNA was used to synthesize cDNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1622) or M-MLV Reverse Transcriptase, RNase H Minus, Point Mutant (Promega, M3682) according to manufacturers’ instructions. Subsequent qRT-PCR was performed in triplicates in 384-well plates using primers and Roche universal probe library (UPL) with TaqMan^TM Universal MasterMix II (Thermo Fisher, 4440044) and LightCycler® 480 II (Roche) or in 96-well plates using primers with PowerUp^TM SYBR^TM Green MasterMix (Thermo Fisher, A25778) and CFX Opus 96 Real-Time PCR System (Biorad). Primer sequences are listed in Appendix Table S3.

Single-cell transcriptomes were generated using a commercially available 384-well plate approach (SORT-Seq2, Single Cell Discoveries, SCD (Muraro et al, 2016)). To this end, primary tumors were dissected and dissociated into single cells as for establishment of organoid lines from tumors. After erythrocyte lysis in ACK buffer, cells were incubated with Fc Block and antibodies (concentrations listed in Appendix Table S2) in PBS/2% FCS/2 mM EDTA. Single viable cells (eFluor-) were gated for EPCAM+ or CD45+ or double negative cells. EPCAM−, CD45− cells were further gated for CD31 expression and triple-negative cells were considered fibroblasts for sequencing. Cells were index-sorted into 384-well capture plates containing 50 µL lysis buffer and barcoded primers covered by 10 µL of mineral oil by flow cytometry using a BD FACSAria Fusion sorter (BD biosciences) and capture plates were sent for paired-end sequencing at SCD (Illumina Nextseq™ 500). Sequences from read 1 were used for assigning reads to cells and libraries, whereas read 2 was aligned to the ensemble transcriptome (genome assembly GRCm38) using bwa version 0.7.10 (Li and Durbin, 2010). The transcript count table was generated by SCD using a custom-written script (https://github.com/anna-alemany/transcriptomics/tree/master/mapandgo). Transcript counts and metadata for all samples were imported and stored as a single-cell experiment object (SingleCell-Experiment (Amezquita et al, 2020) version 1.12.0). During quality control, cells with high mitochondrial content (isOutlier, scater (McCarthy et al, 2017) version 1.18.5) were discarded to remove low-quality cells that may have been damaged during processing or may not have been fully captured by the sequencing protocol. For analysis of CAFs, cells with high expression of Ptprc and Epcam genes were excluded to avoid contamination by residual immune/epithelial cells. The samples were normalized by computing the log₂-transformed normalized expression values across all genes for each cell (logNormCounts, scater). Next, all batches were subset to the common features across all samples to enable downstream analysis. To account for the differences in samples due to plates (“batch effect”), we used the fastMNN algorithm (correctExperiments, batchelor (Haghverdi et al, 2018) version 1.6.2). Subsequently, clusters were identified using a graph-based approach (buildSNNGraph with k set to 15, scran (Lun et al, 2016) version 1.18.5) and the walktrap algorithm (igraph version 1.2.6). Cluster-specific marker genes were identified using the findMarkers function (scran) that uniquely define one cluster against the rest. Functional enrichment analysis in Metascape (https://metascape.org/) and Enrichr (https://maayanlab.cloud/Enrichr/) was applied to identify pathways and processes that were enriched in each cluster based on differentially expressed genes (FDR ≤ 0.1; P ≤ 0.05). Furthermore, expression profiles of identified clusters were compared with those previously published for CAF subtypes using the AddModuleScore function (Tirosh et al, 2016) from Seurat (Hao et al, 2021) version 4.0.0. To this end, each cell was assigned a score using the module of genes associated with the “published clusters”. A positive score suggested that this module of genes is expressed in a particular cell more than would be expected, given the average expression of this module across the population. A mean score for cluster-specific cells was calculated to obtain scores per cluster for each module of genes associated with “published clusters”. Using a reference dataset with known labels, the SingleR (Aran et al, 2019) approach (version 1.4.1) labels new cells from a test dataset based on similarity to the reference. Thereby, we assessed the similarity between clusters from Fib^ΔZeb1 and Fib^Ctrl, where Fib^Ctrl was set as a reference. For the scRNA-seq dataset from non-inflammation-driven orthotopic tumors, lower quality clusters with indistinct or ambiguous cell-type identities were excluded by further sub-clustering.

The DAseq tool (Zhao et al, 2021) was used to study the differential abundance of cells from AOM/DSS scRNA Fib^Ctrl and Fib^ΔZeb1, using read counts after basic quality control filtering implemented in the standard Seurat (Hao et al, 2021) workflow. Cells with 100–6000 detected genes and less than 10% mitochondrial counts were included. The entire dataset which includes Fib^Ctrl and Fib^ΔZeb1 samples, was used to normalize the data, identify the highly variable genes, scale the data, compute the PCA, use nearest neighbor embedding, integrate the batches using harmony and recluster the data. Cell phenotypes were assigned by using Seurat’s AddModuleScore and published gene sets (Bartoschek et al, 2018; Elyada et al, 2019). DAseq was applied to identify differential abundance (Zhao et al, 2021) and three differentially abundant cell populations were identified. One contained almost exclusively cells from one SORT-Seq plate and was therefore neglected. Statistical significance is reported in the respective figures or figure legends. Characteristic genes for the two other DA populations were identified using the marker gene function included in the DAseq package, which uses stochastic gates.

The integrity of total RNA samples was assessed by Bioanalyzer2100 before sequencing and processed at Novogene UK, Cambridge. RNA-Seq raw reads files were analyzed using nfcore RNA-seq pipeline version 3.11.0 (Ewels et al, 2020). Briefly, raw sequencing reads were aligned to the mouse genome GRCm38 using STAR (Dobin et al, 2013), the reads aligning to each annotated gene were quantified by the Salmon software (Patro et al, 2017), and differential gene expression was performed using DESeq2 R package (Love et al, 2014). Gene set enrichment analysis (GSEA) was performed on all (filtered) genes across the different experimental conditions using the R packages “clusterProfiler” (v4.10.0) and “enrichPlot” (v1.22.0) with MSigDB and custom gene sets (Elyada et al, 2019; Ohlund et al, 2017; Wu et al, 2021). To account for multiple comparisons, P values were adjusted using the Benjamini–Hochberg method.

Information on the number of biologically independent samples analyzed and the number of times experiments were performed is included in the figure legends. All the representative experiments were repeated at least three times unless otherwise stated. Statistical analyses were performed using GraphPad Prism, within the provided R packages or online tools for gene set enrichment analyses, and details on statistical methods are included in the figure legends. All error bars represent SD or SEM as indicated. Where possible animals were assigned to experimental groups using simple randomization and the investigators were blinded for initial animal data collection on tumor/metastasis parameters. No blinding was applied for other experiments since investigators needed information about the groups to correctly perform and analyze the experiments.