A Wnt-Induced Phenotypic Switch in Cancer-Associated Fibroblasts Inhibits EMT in Colorectal Cancer

Colorectal cancer is one of the most frequent types of tumors and the fourth leading cause of cancer-related deaths worldwide (1). The therapeutic options for patients with advanced colorectal cancer are still limited. One main challenge is a high genetic heterogeneity both at the interindividual (2, 3) and intratumoral level (4). In addition, prognosis and therapy responses are influenced by the tumor microenvironment (TME) that has been recognized as key regulator for the hallmarks of cancer (5). In a bidirectional cross-talk, tumor cells educate the surrounding stroma, which in turn regulates tumor metabolism, drug resistance, immune defense, and capacity to metastasize. Cancer cell dissemination has been closely linked to epithelial–mesenchymal transition (EMT), a program of cell plasticity that is dependent on stromal signals (6). How oncogenic changes in tumor cells affect their capacity to shape and respond to the TME has not been fully understood.

One of the most prominent cell types in the TME are fibroblasts. In colorectal cancer, a mesenchymal-rich stroma has been associated with poor prognosis (7–9) and an immunosuppressive environment (10). Cancer-associated fibroblasts (CAF) are characterized by induction of markers including alpha smooth muscle actin (αSMA) and platelet-derived growth factor receptor α (PDGFRα) that are also found in activated fibroblasts during wound healing (11, 12). CAFs can be recruited from various sources including resident tissue fibroblasts and mesenchymal stem cells (13) and recent studies have identified diverse CAF populations in solid cancers of breast (14), colon (15), lung (16), and pancreas (17, 18). However, we currently lack mechanistic insights of the specification and function of CAF subtypes.

In the normal gut, precise control of the Wnt signaling pathway in the epithelium controls stem cell maintenance, proliferation, and differentiation. Inactivating mutations of Adenoma polyposis coli (APC) act as the main initiating driver of colorectal cancer that confers growth independence from Wnt ligands that are normally restricted to the stem cell niche (19, 20). While the tumor-intrinsic role of Wnt signaling pathway is well established (21), we here set out to study a potential role in the TME. We combined protocols for genetic modification of gastrointestinal tumor organoids with transplantation and primary cell coculture models to investigate the consequences of stromal-specific Wnt perturbation. We report that Wnt signaling in CAFs induces subtypes with distinct roles in colorectal cancer growth and progression.

Wnt3^HA (22), Apc^min (23), Rag1^tm1Mom (Rag1; ref. 24), NOD-scid IL2Rgamma^null (NSG; ref. 25), Col1a2-creER^T2 (26), Ctnnb1^fl (27), and Rosa-CAG-LSL-tdTomato-WPRE (28) mice have been described. Rag1, Col1a2-creER^T2, and Ctnnb1^fl/fl mice were maintained on a C57BL/6 background. Experiments were performed using cohorts of sex-matched littermates that were randomly allocated. No blinding was performed except for the Col1a2-creER^T2; Ctnnb1^fl/fl experiments. Wnt3^HA/HA; Apc^min/+ mice together with their control littermates (Wnt3^HA/HA; Apc^wt) were kept under normal chow diet for 20 weeks before they were sacrificed. For transplantation, organoids were collected in cold medium, mechanically dissociated, centrifuged for 5 minutes at 1,200 RPM and the pellet was resuspended in the respective organoid medium (see below) supplemented with 25% Matrigel (Corning) and Rho kinase inhibitor Y-27632 (10 μmol/L; Biotrend). Per mouse, 100 μL organoid suspension representing 2 confluent wells (12-well plate) were injected subcutaneously into the right flank. Tumor volumes were measured with a caliper and calculated using the formula: V = π × (d² × D)/6, where d is the minor tumor axis and D is the major tumor axis. Once tumors reached an average size of 200 mm³, Col1a2-creER^T2; Ctnnb1^fl/fl and Col1a2-creER^T2; Ctnnb1^+/fl mice were kept under tamoxifen diet (400 mg/kg) for a duration of 14 days.

For organoid overexpression of Dkk1 and Sfrp1, the open reading frames were PCR amplified from mouse cDNA of E5-embryos and whole colon, respectively, using Phusion polymerase (Thermo Fisher Scientific) and the followings primers Dkk1_fwd: GAACTAAACCGTCGACCATGATGGTTGTGTGTGCAGCG, Dkk1_rev: GAATTCCCGGCTCGAGTTAGTGTCTCTGGCAGGTGTGGAG and Sfrp1_fwd: GAACTAAACCGTCGACCATGGGCGTCGGGCGCAGCGCG, Sfrp1_rev: GAATTCCCGGCTCGAGAGTATCACTTAAAAACAGACTGGAAGGTGGGACACTCGTG. Gel-purified PCR products were inserted into the SalI and XhoI restriction sites of pLV-Pgk::EGFP-IRES-DsRed-P2A-puro vector using In-Fusion HD Cloning Plus Kit (Takara Bio) according to manufacturer's instruction and plasmids were then confirmed by Sanger sequencing. The NICD was amplified from mouse cDNA of whole colon using the following primers: N1icd_fwd: CAGGACCGGTTCTAGAGCGCTGCCACCATGAAGCGCCGGCGGCAGCATG, N1icd_rev: TTGTTGCGCCGGATCCTGCCGGTGCACCCACGGTG and introduced into the XbaI and BamHI restriction sites of lentiCas9-Blast (Addgene plasmid #52962, kind gift of Feng Zhang) using the in-Fusion HD Cloning Plus kit.

All cells used in this study were regularly tested for Mycoplasma and replaced from low passage number cryo-stocks. Mouse organoids (29) and human colorectal cancer organoids (30) were established and cultured as reported previously. Wnt3a, R-spondin-1 and Noggin conditioned media (CM) were prepared as described previously (29). Generation of mouse tumor organoids (APN and APTK) and Dkk1- and Sfrp1-expressing organoids was performed by liposomal transfection and lentiviral transduction as described in the Supplementary Methods. Isolation of primary mouse colon fibroblasts was described in ref. 29. Fibroblasts generated from C57BL/6 mice were maintained in DMEM (Thermo Fisher Scientific), 10% FBS (Sigma-Aldrich), and 1× penicillin/streptomycin (Thermo Fisher Scientific). Upon confluency, cells were split in a 1:3 ratio. Organoid/fibroblast cocultures were performed as described previously (17) using 12-well tissue culture inserts (Transwell Corning; pore size 8 μm). In the lower compartment, 50,000 fibroblasts were seeded per well either directly on tissue culture treated plates or embedded in 75%–80% BME (Amsbio). In the top compartment, tumor organoids were seeded either in reduced medium (advanced DMEM/F12 supplemented with 10 mmol/L HEPES, 1× Glutamax, 1× penicillin/streptomycin) or supplemented with 20% Wnt3a CM or 5 μmol/L IWP-2 (Sigma-Aldrich). For normalization, equal volumes of control CM from parental L-cells were added. For gel contraction assay, 50,000 mouse fibroblasts were resuspended in 300 μL reduced medium supplemented with 160 μL of 3 mg/mL rat tail collagen I (Thermo Fisher Scientific), 16.8 μL 10× PBS before addition of 8 μL 1 mol/L sodium hydroxide. The suspension was then transferred to a 24-well plate and allowed to solidify at 37°C, 5% CO₂ for 45 minutes. Collagen disks were released before addition of reduced medium that was supplemented with 20% Wnt3a CM, 10% R-spondin1 CM, 5 μmol/L IWP-2 or CM from tumor organoids. For normalization, equal volumes of control CM from parental L-cells (for Wnt3a), HEK293Tcell (for R-spondin1) were added. For cocultures, 50,000 fibroblasts and organoids (1:12 of confluent 12 wells) were embedded as above and incubated in reduced medium. Gel contraction was documented using a Canoscan 8800F scanner (Canon) and the disk circumference was measured by ImageJ. Contraction index was determined using the following formula: 100 − (circumference at time point ×/initial circumference).

Two-dimensional (2D) and three-dimensional (3D) seeded fibroblasts were cocultured with APN tumor organoids as indicated above and CM was collected after 4 days, filtered through 0.45 μm filter and stored at −20°C, before measurement of Il6 (DY406), Il1ra (DY480), and Tnfα (DY410; all from R&D systems) according to the manufacturer's protocol.

Organoid CM was collected after 4–5 days in regular culture medium. Control CM was collected in parallel from organoids transduced with a GFP lentivirus. Activity was tested on normal mouse small intestinal organoids and cell viability was assessed after 4–5 days using CellTiter-Glo (Promega) according to manufacturer's instructions using a Lumistar Galaxy plate reader (BMG Labtech).

For fixation and paraffin embedding of organoids, cells were gently collected in cold medium, fixed overnight in 2% paraformaldehyde at 4°C, before one drop of hematoxylin was added to the organoid pellet, mixed with 100 μL of HistoGEL (Thermo Fisher Scientific), and incubated for 30 minutes on ice. Tissues were fixed in 4% paraformaledehyde overnight at 4°C. Fixed samples were dehydrated, embedded in paraffin, and sectioned. Four μm sections were deparaffinized and automatically processed using a Bond-Max (Leica) and the Bond Polymer Refine Detection System (Leica) as described by the manufacturer's instructions. Stained sections were scanned with a ScanScopeCS2 using a 20× objective. For immunofluorescence, sections were deparaffinized and antigen retrieval was performed in a pressure cooker for 20 minutes. Sections were blocked in PBS supplemented with 0.2% Tween 20 and 10% goat serum (Sigma-Aldrich) for 45 minutes and then incubated with primary antibody overnight at 4°C. Sections were washed and incubated with secondary antibodies for 1 hour at room temperature together with 2 drops/mL of NucBlue (Thermo Fisher Scientific). Sections were mounted using Prolong Gold Antifade (Thermo Fisher Scientific) and documented using an Evos microscope (Thermo Fisher Scientific; 20× objective). Whole mount staining and detection of Wnt3-HA, were performed as reported before (22) and signals were analyzed with a SP5 scanning confocal microscope (Leica) using a 20× objective. Skin tissues of Col1a2-CreER^T2; control or Col1a2-CreER^T2; Rosa-CAG-LSL-tdTomato-WPRE mice were collected and 10 μm cryosections were mounted with Prolong Gold Antifade (Thermo Fisher Scientific) and documented using an Evos microscope (Thermo Fisher Scientific; 20× objective). A full list of antibodies and staining conditions is shown in Supplementary Table S1A.

Tumors were collected and dissociated using 0.1% Collagenase type I (Thermo Fisher Scientific) and 0.2% Dispase type II (Sigma-Aldrich), filtered using a 70 μm cell strainer and centrifuged for 5 minutes at 1,500 RPM. Red blood cell lysis was performed by incubating the cell pellet for 3 minutes in 1 mL of Hybri-Max buffer (Sigma-Aldrich). Counted cells were resuspended in PBS containing 0.5% BSA and 2 mmol/L EDTA (FACS buffer) in a concentration of 10⁶ cells/100 μL and incubated for 15 minutes with 1 μL of mouse BD Fc Block (BD Biosciences) per 1 million cells. Cells were stained with anti-Epcam-VioBlue (Miltenyi Biotec), anti-Pdgfrα-APC (Thermo Fisher Scientific), anti-Cd31-PE-Cy7 (BD Biosciences) and Fixable viability dye eFluor780 (Thermo Fisher Scientific) for 20 minutes on ice, washed and resuspended in FACS buffer. Epcam⁺ tumor cells or Pdgfrα⁺ CAFs were collected using a FACS Fusion (BD Biosciences). Data were analyzed using FlowJo v10 software (FlowJo LLC).

Total RNA was extracted from organoids, fibroblasts, sorted tumor cells, and sorted CAFs using NucleoSpin-RNA Kit (Macherey-Nagel) according to the manufacturer's instructions. In transwell cocultures, the RNA from top and bottom compartments were collected separately. For qRT-PCR analysis, 0.1–0.5 μg RNA was reverse transcribed using random hexamers (Thermo Fisher Scientific) and M-MLV reverse transcriptase enzyme (Promega) according to the manufacturer's instructions. Relative gene expression was measured using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific) on a StepOnePlus instrument (Applied Biosystems). The relative gene expression was calculated using 2^(ΔΔC_T) method by normalization to Hprt expression. The list of the primers used is shown in Supplementary Table S1B. RNA sequencing was performed as described previously (see Supplementary Methods; ref. 30).

All error bars represent the mean ± SEM. Statistical significance was calculated by two-tailed unpaired t test on two experimental conditions or two-way ANOVA when repeated measurements were compared, with P < 0.05 considered statistically significant: ns, nonsignificant. All the experiments were repeated at least two times independently unless mentioned in the legend. Statistical analysis was performed using Prism 7.0 (GraphPad Inc.).

All expression datasets generated in this study have been deposited in Gene Expression Omnibus (NCBI) under accession number GSE155364.

All animal housing and experimental procedures were approved by the institutional animal care and the regional animal welfare office, Darmstadt (F123/1004, F123/1031). Human colorectal cancer organoids were generated after written informed consent as part the interdisciplinary Biomaterial and Databank Frankfurt (iBDF) of the University Cancer Center Frankfurt (Frankfurt, Germany) that was approved by the ethics committee at the University Hospital Frankfurt (274/18).

Recently, we studied the distribution of Wnt3 protein in the intestinal stem cell niche using an epitope-tagged mouse allele (Wnt3^HA; ref. 22). Because Wnt3-expressing Paneth cells are induced by canonical Wnt signals (29), constitutive pathway activation upon APC loss of function may further promote Wnt3 expression in tumors. This is consistent with the Paneth cell metaplasia that is frequently found in human adenomas (31). To study the endogenous Wnt3 protein level in adenomas, we generated Wnt3^HA/HA Apc^min/+ mice. IHC analysis indeed revealed strong induction of Wnt3 expression (Fig. 1A) and we hypothesized that rather than Apc-deficient tumor cells, the adenoma-surrounding stroma may respond to the signal and thereby influence colorectal carcinogenesis (Fig. 1B).

To dissect the role(s) of the Wnt-responsive stroma, we generated transplantable tumor organoids that secrete the Wnt antagonists Secreted frizzled-related protein 1 (Sfrp1) or Dickkopf-related protein 1 (Dkk1) as a tool to globally block Wnt signaling (Fig. 1B): First, we tested functional secretion of overexpressed Dkk1 or Sfrp1 in mouse tumor organoids. Sfrp1 and Dkk1 CM were collected after 3–4 days and then transferred to normal mouse intestinal organoids (Supplementary Fig. S1A). We observed a concentration-dependent decrease of growth, indicating functional secretion of both antagonists (Supplementary Fig. S1B). In the following experiments, we focused on Sfrp1 that showed a more potent inhibition compared with Dkk1. Tumorigenic organoids were generated from p53^fl/fl intestinal organoids, where Apc loss-of-function mutation was induced by CRISPR/Cas9 together with p53 ablation by cotransfection of Cre plasmid. The resulting Apc/p53-deficient (AP) organoids were clonally expanded and mutations were confirmed (Supplementary Fig. S1C–S1E). AP organoids were transduced with a Notch1 intracellular domain (NICD) lentivirus to activate Notch signaling. The resulting APN organoids, represent an adenocarcinoma-like colorectal cancer genotype (32), and all three drivers were required for successful growth upon subcutaneous transplantation (Supplementary Fig. S1F and S1G). Cotransduction with Lenti-GFP-DsRed (control) or Lenti-Sfrp1-DsRed (Supplementary Fig. S1H) was performed to obtain Apc^KO/p53^KO/NICD^OE/GFP^OE and Apc^KO/p53^KO/NICD^OE/Sfrp1^OE that were named “control” and “Sfrp1” in the following experiments, respectively.

In vitro, Sfrp1 expression did not affect organoid expansion, size, or proliferation (Supplementary Fig. S2A–S2F). In addition, qRT-PCR analysis showed unchanged expression of Wnt target genes, as well as p53 targets (Supplementary Fig. S2G and S2H). However, upon subcutaneous transplantation into Bl/6 mice, strong reduction of tumor growth was observed (Fig. 1C). Similar results were obtained in immunocompromised (Rag1) and immunodeficient (NSG) mice (Fig. 1D and E), suggesting that the defect in tumor growth depends on a nonimmune, stromal compartment. Importantly, histologic and mRNA expression analysis of transplanted tumor cells again revealed no general reduction in proliferation (Fig. 1F and G) and Wnt signaling (Fig. 1H–J). We observed no difference in apoptosis as studied by cleaved caspase-3 staining (Fig. 1K and L) and slight dysregulation of the p53 target Pmaip1 (Fig. 1M). Similar results were obtained using Apc^KO/p53^KO/Tgfbr2^KO/Kras^G12D-OE mouse colon tumor organoids (APTK; Supplementary Fig. S2I–S2P; ref. 33), suggesting that interference with Wnt in the microenvironment reduces tumor growth independently of the tumor genotype.

To address the molecular consequences of Wnt blockage, we performed RNA sequencing and compared the transcriptional response in transplanted tumor organoids (after flow cytometric sorting of Epcam-positive cells) with organoids cultured in vitro (Fig. 2A). While pronounced changes were observed between transplanted Sfrp1 and control cells, minor changes were induced in vitro (Fig. 2B; Supplementary Tables S2 and S3). Except for the expression of the Sfrp1 transgene, no global correlation was found. Interestingly, gene set enrichment analysis (GSEA) analysis showed a strong enrichment of signatures associated with EMT, extracellular matrix, TGFβ signaling and inflammation in Sfrp1 tumor cells only in vivo (Fig. 2C). qRT-PCR analysis confirmed that Zeb1 and Vim were strongly induced upon Sfrp1 expression in vivo but not in vitro (Fig. 2D), while Cdh1 expression remained unchanged. Consistently, Wnt/β-catenin signaling was unaffected, collectively arguing for a noncell autonomous effect. On sections, we could confirm Zeb1 protein induction in scattered cells in the tumor bed of Sfrp1 tumors, as well as a reduced E-cadherin protein level, indicative of EMT (Fig. 2E–G). To further characterize this result, control and Sfrp1 organoids from subcutaneous grafts expressing DsRed were traced by flow cytometry. Indeed, we found that 15.8% ± 0.6% of the Epcam-negative population was of tumor cell origin in Sfrp1 tumors compared to 4.1% ± 1.8% in the control (Fig. 2H and I). Furthermore, Sfrp1-induced EMT was also observed using APTK organoids (Supplementary Fig. S3), suggesting a general role of stromal Wnt signals in regulation of tumor malignancy.

To distinguish whether canonical or noncanonical Wnt signals that are both blocked by Sfrp1 are involved, we generated tumor organoids overexpressing Dkk1 that targets the canonical pathway (34). Again, we observed unaffected growth of APN organoids after Dkk1 expression in vitro (Supplementary Fig. S4A–S4H) but strong inhibition of tumor growth (Supplementary Fig. S4I and SIJ) without major effect on tumor cell proliferation, apoptosis or Wnt signaling (Supplementary Fig. S4K–S4R). Furthermore, Dkk1 strongly induced EMT in tumor cells and increased the number of Epcam-negative cells of tumor origin (Supplementary Fig. S4S–S4U), demonstrating that blockage of the canonical Wnt pathway is sufficient to induce the observed phenotypes.

Next, we investigated which stromal cell population is the primary target of Wnt inhibition. Given the preserved reduction of tumor growth (Fig. 1C–E; Supplementary Fig. S2IJ) and EMT induction (Fig. 2E–I; Supplementary Figs. S3 and S5A–S5C) in immunocompetent and immunocompromised mice, we excluded a contribution of the immune-compartment. CAFs have been implicated in regulation of tumor growth and progression (11, 12) and their abundance characterizes a malignant colorectal cancer subtype (7–9). To study whether CAFs are affected by Wnt blockade, we performed co-immunofluorescence (co-IF) analysis for vimentin, a general fibroblast marker and αSMA, a well-established activation marker (11, 12). Interestingly, block of Wnt signaling using both Sfrp1 and Dkk1 dramatically reduced the activation of CAFs (Fig. 3A–F). The number of tumor-adjacent fibroblasts appeared comparable, but a significantly increased number of nonspindle-shaped vimentin-positive cells was found in the central regions of Sfrp1 and Dkk1 tumors, presumably representing tumor cells after EMT. Similar reduction of CAF activation was observed in APN tumors transplanted in NSG mice (Supplementary Fig. S5D–S5F), as well as upon Sfrp1 expression in APTK tumors in Bl/6 mice (Supplementary Fig. S5G–S5I), indicating that this phenotype is not dependent on host immune cells or a specific tumor genotype.

To investigate whether Wnt signaling directly regulates fibroblast activation, we performed collagen gel contraction assays. Indeed, addition of Wnt and R-spondin to mouse colonic fibroblasts increased their contractility, indicative of fibroblast activation (Fig. 3G and H). Of note, we observed that Wnt-induced contractility does not require R-spondin (Fig. 3I and J). Block of Wnt production (by IWP-2 treatment) did not affect baseline contractility, arguing against presence of endogenously expressed Wnt ligands in monoculture (Fig. 3G and H). In contrast, transfer of conditioned medium from APN-Sfrp1 or APN-Dkk1 organoids reduced fibroblast contractility arguing for Wnt production in tumor cells (Fig. 3K and L). Together, our data suggest a direct role Wnt signaling for fibroblast activation.

To characterize the transcriptional response of CAFs, we performed RNA sequencing of sorted Pdgfrα-positive CAFs from control and Sfrp1 tumors. As a control, mouse colonic fibroblasts were stimulated with Wnt in vitro (Fig. 4A; Supplementary Tables S4 and S5). GSEA analysis showed that Wnt/β-catenin signaling was significantly downregulated in fibroblasts isolated from Sfrp1 tumors but enriched upon stimulation in vitro, validating our experimental strategy (Fig. 4B; Supplementary Table S6). Furthermore, Wnt-induced cell-cycle signatures were observed in vitro and in vivo. In contrast, signatures related to invasiveness and inflammation were only enriched in CAFs from Sfrp1 tumors. Together, Wnt signaling appears to regulate diverse phenotypes in fibroblasts including contractility, proliferation, and inflammatory signaling.

Because not all CAFs are Pdgfrα-positive (11, 12), we sought for a more global strategy to record stromal effects after Wnt blockage. To this end, we performed bulk RNA sequencing of xenografted colorectal cancer patient-derived organoids (PDO), where bioinformatic analysis allows to distinguish tumor transcripts (human reads) from stromal transcripts (mouse reads; ref. 35). Sfrp1 was stably introduced in two independent PDO lines that were transplanted into Rag1 mice. Similar to mouse organoids, we found reduced tumor growth (Supplementary Fig. S6A–S6G) and number of activated CAFs (Supplementary Fig. S6H–S6J), indicative of a conserved mechanism. Interestingly, we could not recapitulate induction of vimentin- or Zeb1-positive cells and reduction of E-cadherin in the tumor bed (Supplementary Fig. S6H–S6M), suggesting that EMT-inducing signals are incompatible between the species as reported for several cytokine–receptor interactions (36, 37). Following bulk RNA sequencing and species deconvolution (Supplementary Fig. S7A), we confirmed a reduction of Wnt/β-catenin signaling and induction of inflammatory responses in the mouse stroma (Supplementary Fig. S7B).

Recently, two subtypes of CAFs were identified in pancreatic cancer: myCAFs, characterized by high expression of αSMA (Acta2) and inflammatory-like CAFs (iCAF), characterized by the expression of Il6 (17). We tested whether similar subtypes are regulated by Wnt signaling in colorectal cancer, and found that the iCAF signature was indeed enriched in sorted fibroblasts from Sfrp1 tumors (Fig. 4C; Supplementary Table S7), as well as in stroma from Sfrp1 PDO xenografts (Supplementary Fig. S7C; Supplementary Table S8), whereas the myCAF signature was more abundant in the controls, suggesting that Wnt signaling regulates this phenotypic switch.

To obtain an in vitro model for fibroblast plasticity in colorectal cancer, we adapted the coculture protocols that were recently reported for pancreatic cancer (Fig. 4D; ref. 17). Indeed, we could confirm induction of the iCAFs and myCAFs in colonic fibroblasts (Supplementary Fig. S8AB): The myCAF marker, Acta2 was induced in fibroblasts cultured on a 2D support. Transwell coculture with tumor organoids (APN) further increased the expression. In addition, Axin2 expression was strongly induced in 2D cocultures arguing for a Wnt supportive effect (Supplementary Fig. S8A). iCAF markers (Il1a and Il6) were strongly induced by coculture when fibroblasts were seeded in a 3D matrix (Supplementary Fig. S8B), as reported before (17). We then tested the reciprocal effect on cocultured APN tumor organoids. qRT-PCR analysis revealed a pronounced induction of markers of invasiveness (Zeb1 and Vim) with iCAFs but not myCAFs (Supplementary Fig. S8C). This result was confirmed in APTK tumor organoids (Supplementary Fig. S8D), indicating that iCAFs induce a general promalignant phenotype in colorectal cancer cells. We then tested the effect of Wnt modulation on the CAF and tumor cell status by qRT-PCR. In monocultures, high Wnt-independent base-line expression of the myCAF marker Acta2 was observed in 2D cultured fibroblasts (Fig. 4E). In contrast, addition of Wnt3a-CM was sufficient to induce weak Acta2 expression also in 3D fibroblasts. The Wnt inhibitor IWP-2 had no pronounced effect in monocultures consistent with our observation that fibroblast contractility does not depend on endogenous Wnt. Irrespective of Wnt modulation, only weak expression of iCAF markers was observed in fibroblast monocultures. Upon coculture with APN tumor organoids, however, we observed that Wnt significantly increased the expression of Acta2 and Axin2 in 2D and 3D fibroblasts and downregulated iCAF markers (Il6, Tnfa, and Il1a) in 3D fibroblasts (Fig. 4E and F). IWP-2 had an opposite effect and strongly increased iCAF marker expression. ELISA showed that 3D cocultured fibroblasts also secrete significantly higher levels of proinflammatory cytokines (Tnfα, Il1ra, and Il6), which was further enhanced or reduced by addition of IWP-2 and Wnt3a-CM, respectively (Supplementary Fig. S8E). In the organoids, iCAF-induced Zeb1 and Vim expression was blunted by addition of Wnt3a-CM, whereas IWP-2 treatment further increased the Vim expression (Fig. 4G). Cdh1 mRNA expression was not strongly affected as observed in vivo (Fig. 2D). Together, our results indicate that endogenous Wnts control fibroblast subtype identity thereby inducing a partial EMT in colorectal cancer cells.

Recently, we have shown that Wnt signals exhibit a short range in the intestinal stem cell niche (22). In a tumor context, the induction of myCAFs and tumor invasiveness may also depend on the distance between the cell compartments. Therefore, we tested in a transwell setting, if coseeding of tumor organoids together with fibroblasts in direct contact or in separate 3D matrices affects the invasiveness of APN organoids (seeded in the upper compartment) (Fig. 5A). We observed that expression of Zeb1 and Vim in tumor organoids was significantly decreased when fibroblasts were seeded in close contact, suggesting that juxtracrine signals induce the myCAF phenotype (Fig. 5B). Furthermore, collagen gel-contraction assays showed that direct coculture with tumor organoids increased the contractility of fibroblasts, which was modulated by Wnt and IWP-2 treatment (Fig. 5C–F) or by coculture with Sfrp1- and Dkk1-expressing tumor organoids (Fig. 5G and H), indicating that direct tumor cell interaction regulates fibroblast subtype differentiation. To identify functional mediators of the stromal cross-talk, we used a recently described computational method (NicheNet) to infer ligand-induced signaling based on gene expression data (38). Differentially expressed genes in fibroblasts (Supplementary Table S4) and tumor cells (Supplementary Table S3) of Sfrp1 tumors were analyzed to identify iCAF mediators involved in EMT. Among predicted iCAF ligands, we found the inflammatory mediators Il1a, Cxcl5, Il1rn, Tnfsf13b (BAFF), and Cxcl10 (Supplementary Fig. S9A), that could be linked to EMT gene expression in tumor cells (Supplementary Fig. S9B). Further inspection also confirmed expression of the cognate receptors for Il1 and BAFF ligands in tumor cells (Supplementary Fig. S9C) supporting a functional involvement.

To genetically validate our findings, we generated a conditional Col1a2-creER^T2; Ctnnb1^fl/fl mouse model to ablate Wnt/β-catenin signaling specifically in fibroblasts (Bcat^Δ/Δ; Fig. 6A). After tamoxifen administration, Bcat^Δ/Δ mice showed a rapid loss of body weight and had to be sacrificed after 1–2 weeks (Supplementary Fig. S10A and S10B). Inspection of internal organs showed reduced size of colon, liver and spleen compared with littermate controls (Bcat^+/Δ; Supplementary Fig. S10C–S10G). Histologic analysis of the small intestine and colon of Bcat^Δ/Δ mice did not reveal obvious morphologic differences or changes in epithelial cell proliferation (Supplementary Fig. S10H and S10I). To delete Ctnnb1 in a tumor context, APN tumor organoids were subcutaneously injected, followed by tumor growth and tamoxifen administration (in Bcat^+/Δ and Bcat^Δ/Δ mice). By crossing the Col1a2-creER^T2 to Rosa-CAG-LSL-tdTomato-WPRE mice, we could confirm widespread Cre activity in fibroblasts in the skin (Supplementary Fig. S10J). Consistent with Sfrp1/Dkk1 overexpression, Ctnnb1 deletion in fibroblasts decreased tumor size and weight without pronounced effects on proliferation and β-catenin staining in tumor cells (Fig. 6B–F). αSMA expression was significantly reduced in CAFs of Bcat^Δ/Δ mice, confirming the requirement for fibroblast activation (Fig. 6G and H). Furthermore, vimentin expression was prominently increased in tumor cells in Bcat^Δ/Δ mice, suggesting widespread EMT induction (Fig. 6G and I). To study the histopathologic consequences, we analyzed the presence disseminating tumor cell clusters (1–10 cells) at the invasive front of primary tumors. Indeed, we found a significant increase of tumor cell clusters in Sfrp1, Dkk1, and Bcat^Δ/Δ tumors compared with their littermate controls (Supplementary Fig. S11A–S11C). Together, our results emphasize a key role of Wnt/β-catenin signaling in fibroblasts for tumor growth and invasiveness.

Consequently, we tested the association of Wnt activity with patient outcome in a clinical colorectal cancer cohort (GSE39582; ref. 39). By scoring combined Wnt target gene expression, we found that patients with low activity displayed significantly worse relapse-free survival (RFS; Fig. 7A), consistent with previous reports (9, 32). We furthermore confirmed increased Wnt activity in the canonical colorectal cancer subtype (CMS2) while the aggressive mesenchymal subtype (CMS4) displayed reduced activity (Fig. 7B). Interestingly, the Wnt level was able to stratify colorectal cancer subtypes and a low level was linked with reduced RFS in CMS1, CMS2, and CMS4 patients (Fig. 7C–F). These results indicate that low Wnt levels can be considered as independent prognostic factor in colorectal cancer. In line, we found that the combined expression of Wnt antagonists (DKK1–4/SFRP1,2,4,5) was associated with reduced RFS (Fig. 7G). Here, CMS4 patients showed highest expression of Wnt antagonists (Fig. 7H), arguing for an inhibitory role of the mesenchymal stroma. In summary, these results support a prognostic relevance of stromal Wnt signaling in colorectal cancer.

Here, we report a mechanism that controls plasticity between CAF subtypes with opposing phenotypes in colorectal cancer. In our experimental models, canonical Wnt signals were necessary and sufficient to induce myofibroblasts and to block inflammatory CAFs. The resulting tissue cross-talk inversely links tumor growth and invasiveness, revealing a conserved mechanism for generation of cellular heterogeneity in the TME. We discuss how dynamic changes of Wnt ligands and antagonists from diverse sources may organize the tumor architecture and influence tumor progression.

Transcriptomic stratification of colorectal cancer has identified a fibroblast-rich stroma as indicator for poor prognosis (7–9). This mesenchymal subtype (CMS4) has been associated with a more advanced stage at diagnosis and with generally reduced levels of Wnt signaling (9). We have recently shown that while tumor-specific Wnt signals downstream of APC-loss are reduced, Wnt signals that are characteristic for nontumor cells are enriched in CMS4 tumors, suggestive of a response in stromal cells (30). Here we describe that Wnt activity can further stratify prognosis in CMS1, CMS2, and CMS4 tumors, indicating that a low stromal level is more generally associated with poor disease outcome. In future, single cell analysis will be informative to record Wnt activity in different stromal populations. We observed a striking differential regulation of tumor growth and invasiveness. This is reminiscent of the described inverse correlation between the level of tumor growth and fibroblast infiltration in colon cancer (40). EMT was induced in transplanted Wnt-independent tumor organoids after fibroblast-specific perturbation. Thus, reduced Wnt signaling in fibroblasts may contribute to tumor progression in a similar manner as described for reduction of tumor-cell intrinsic Wnt activity (41). Despite the important immunomodulatory function of fibroblasts in other contexts, we excluded an immune involvement in our experiments. This could be result of a low immunogenicity of the genetically engineered organoid models or a more immunosuppressive microenvironment in subcutaneous tumors. Previous work, showed a promalignant role of autocrine Wnt2 in fibroblasts in xenografts of human colorectal cancer cell lines (42) and in mouse breast cancer cell lines expressing Wnt7 (43). This discrepancy may be explained by tumor cell–intrinsic signaling, dependence on specific tumor (geno-)types and/or by species incompatibilities in xenografts. Future work will be required to dissect the underlying signals and genetic factors.

Within the TME a considerable heterogeneity and dynamic changes have been described for CAFs in diverse tumor types (14–18, 44). However, the mechanisms that control the distinct populations and their exact functions still remain largely unknown. Herein, we report that Wnt regulates two conserved fibroblast subtypes that were recently identified in pancreatic cancer (17). Unlike in the pancreas where αSMA⁺ myCAFs restrict tumor growth (45), we observed a trophic function. iCAFs induced tumor cell EMT, and our analysis suggested that the Il1 and BAFF pathways could be involved in this phenotype in support with the promalignant role of inflammatory signals in colorectal cancer (46). TGFβ-induced IL11 secretion from CAFs could represent a parallel pathway to induce colorectal cancer progression (47). By adapting described in vitro models (17), we could show that coculture with tumor organoids strongly increased myCAF differentiation together with Axin2 expression. This induction of endogenous Wnt signaling was dependent on culture in 2D and was blocked in a 3D matrix, indicating that high matrix stiffness represents a permissive state for Wnt-induced myCAF differentiation. Interestingly, a close link between Wnt signaling and fibrosis of skin and kidney has been described previously (48, 49), arguing for similarities between fibroblast subtypes in distinct (patho)physiologic conditions. Recent data have identified Il1/NFκB and Lif/Stat3 as positive and TGFβ/Smad as negative regulators for iCAFs (50). However, the preferential location of myCAFs in close proximity to tumor cells could not be explained yet. Our finding that tumor invasiveness can be partially reversed by juxtaposition of fibroblasts and tumor cells indicates that direct cell interactions induce myCAF formation. This is consistent with the short range of Wnt ligands in the gut (22), providing a plausible model how tissue heterogeneity and distribution of fibroblasts is coordinated.

In the normal gut, myofibroblasts are a main source Wnt signals (19, 20) together with epithelial Paneth cells that express Wnt3 in the small intestine (29). Paneth cell metaplasia is frequent in colon adenomas (31) and our data suggest that a Wnt-rich microenvironment in early adenomas may further support tumor growth by induction of myCAFs. Perturbation of this positive signaling loop, for example, by reduced secretion Wnt ligands or unbalanced ratio of tumor and stroma cells could locally induce formation of iCAFs and invasiveness. In addition, endogenous induction of Wnt antagonists has been linked to progression of several tumor types including colorectal cancer, pancreatic ductal adenocarcinoma, and melanoma (51–53) and we found that Wnt antagonist expression is enriched in the CMS4 subtype. In colorectal cancer, where the majority of tumor cells are Wnt independent, these stromal effects may play a particularly important role, which argues that therapeutic targeting of the Wnt pathway may adversely induce an invasive microenvironment.

H.F. Farin and F.R. Greten report grants from DFG and Hessen State Ministry for Higher Education, Research, and the Arts during the conduct of the study. No disclosures were reported by the other authors.

M.H. Mosa: Formal analysis, investigation, writing-original draft, writing-review and editing. B.E. Michels: Formal analysis, writing-review and editing. C. Menche: Methodology, writing-review and editing. A.M. Nicolas: Methodology, writing-review and editing. T. Darvishi: Methodology, writing-review and editing. F.R. Greten: Resources, supervision, writing-review and editing. H.F. Farin: Conceptualization, supervision, writing-original draft, writing-review and editing.

This work was supported by the LOEWE Center “Frankfurt Cancer Institute” funded by the Hessen State Ministry for Higher Education, Research and the Arts [III L 5 - 519/03/03.001 - (0015)] and the DFG research group “Cell plasticity in CRC” (FOR2438).

We would like to thank Marina Pešić for p53^fl/fl/Tgfbr2^fl\fl organoids, Petra Dinse for help with histology, and Stefan Stein for support with flow cytometry. We thank the High Throughput Sequencing Unit of the Genomics & Proteomics Core Facility, DKFZ, for providing excellent services and Moritz Greif for informatics support. The UCT biobank is acknowledged for support with collection of patient tissues. Cell lines for preparation of conditioned media were a kind gift from Hans Clevers and Calvin Kuo. Johan van Es and Hans Clevers are kindly acknowledged for support with the APC^min experiments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.