IKKβ acts as a tumor suppressor in cancer-associated fibroblasts during intestinal tumorigenesis

To target fibroblasts and CAFs throughout intestinal tumorigenesis, we employed tamoxifen-inducible Col1a2Cre-ER^T2 mice that restrict Cre-mediated recombination to all fibroblastic cells (Zheng et al., 2002). Lineage tracing in Col1a2Cre-ER^T2;Rosa26R^CAG-tdTomato (Fig. 1 A) mice confirmed recombination in subepithelial stromal cells in both untransformed intestine and colonic tumors using the azoxymethane (AOM)/dextran sulfate sodium (DSS)–induced model of colitis-associated tumorigenesis (CAC; Greten et al., 2004). Immunofluorescent staining confirmed widespread expression of collagen I and IV in Col1a2Cre-ER^T2-targeted cells in both unchallenged intestine and in AOM/DSS-induced tumors, but only partial overlap with vimentin, Fsp1, and α-SMA expression (Fig. 1 B). Recombination was not limited to the intestine and, in agreement with the original characterization of the mice (Zheng et al., 2002), we observed widespread RFP expression in other organs, including lung, liver, pancreas, kidney, and spleen (Fig. 1 C). Flow cytometric analysis confirmed that Col1a2Cre-ER^T2 mice target ∼80% of all PDGFRα⁺ cells in the intestine, which express mesenchymal markers CD29 and CD44, but not PDGFRβ (Fig. 1, D–F). Importantly, we could not detect recombination in other stromal cells, such as CD45⁺ leukocytes, Ter119⁺ erythroid cells, CD31⁺ endothelial cells, and EpCAM⁺ epithelial cells (Fig. 1, D–F).

To examine the effect of IKKβ-dependent NF-κB activation in CAFs during intestinal tumorigenesis, we crossed conditional Ikkβ^F/F mice with Col1a2Cre-ER^T2 mice, and Ikkβ deletion was confirmed in intestinal vimentin⁺ fibroblasts isolated from Col1a2Cre-ER^T2;Ikkβ^F/F animals (termed Ikkβ^ΔFib; Fig. 1, G and H), but not in respective splenocytes (Fig. 1 I) or sorted lamina propria CD45⁺ cells (Fig. 1 J) upon tamoxifen administration. Loss of Ikkβ in fibroblasts did not cause any overt phenotype, and Ikkβ^ΔFib mice remained healthy when kept on a tamoxifen-containing diet without further challenge. However, when we performed the CAC model, Ikkβ^ΔFib animals were surprisingly not protected from tumor development and developed similar number of tumors as Ikkβ^F/F littermate controls (Fig. 2 A). Moreover, absence of IKKβ in fibroblasts even strongly promoted tumor growth (Fig. 2 B), and Ikkβ^ΔFib mice developed a higher number of larger tumors (Fig. 2, C–E). Increased tumor size was accompanied by an increased proliferation rate and decreased apoptotic activity in tumor cells from Ikkβ^ΔFib mice (Fig. 2, F–I). Comparable results were also obtained in a model of sporadic tumorigenesis when we repetitively injected AOM without inducing DSS-mediated colitis (Fig. 2, J–L). Furthermore, when we assessed tumor vascularization before sacrifice on day 84 of the colitis-associated tumor model by in vivo confocal laser scanning microscopy (CLSM) blood vessel area and length in Ikkβ^ΔFib tumors was significantly increased when quantified by functional capillary density (FCD; Fig. 2, M and N). Flow cytometric analysis of stromal cells confirmed a significant increase in CD45⁻CD31⁺ endothelial cells (Fig. 2 O). To quantify the number of fibroblasts in the colon, we used the receptor tyrosine kinase PDGFRα as a marker due to its expression in >90% of fibroblast-like cells (Erez et al., 2010). Indeed, in Ikkβ^ΔFib tumors, the number of EpCAM⁻CD45⁻PDGFRα⁺ fibroblasts was markedly increased (Fig. 2 P), whereas the number of CD45⁺ leucocytes remained unchanged (Fig. 2 Q). Immunohistochemical analysis of Ikkβ mutant tumors revealed significantly increased phosphorylation of Akt^S473 and Stat3^Y705 and increased nuclear β-catenin in tumor cells (Fig. 2, R–Z). Collectively, these data suggested that lack of IKKβ in fibroblasts promotes tumorigenesis either because of an increase in the number of activated fibroblasts and/or elevated secretion of cytokines that were capable of activating Akt, Stat3, and Wnt signaling pathways that presumably control tumor cell proliferation and cell death (Cirri and Chiarugi, 2012).

To identify possible IKKβ-regulated factors that could be responsible for the paracrine stimulation of tumor cell proliferation and angiogenesis, we performed qPCR on purified EpCAM⁻CD45⁻PDGFRα⁺ CAFs sorted by flow cytometry from tumors of Ikkβ^ΔFib and Ikkβ^F/F mice. In line with a more activated phenotype of Ikkβ-deficient fibroblasts, αSma, Fap, and Col1a gene expression was markedly elevated, yet Fgf1 or bFgf expression was comparable (Fig. 3 A). In contrast, transcription of genes encoding classical NF-κB–dependent proinflammatory cytokines, such as Il6, Cox2, Cxcl1, and Cxcl12 was reduced, whereas Il-1β, Cxcl2, and Cxcl5 expression levels remained unchanged (Fig. 3 A). CAFs are the main drivers of stromal TGF-β–driven programs associated with poor clinical outcome in CRC (Calon et al., 2014). We therefore examined gene expression levels of Tgfb1, Tgfb2, and Tgfb3, but could not detect any relevant changes in their expression levels between the different genotypes. However, Smad7 and Smurf1 coding for negative regulators of TGFβ signaling, and the former known to be transcriptionally regulated by NF-κB (Bitzer et al., 2000; Freudlsperger et al., 2013), were markedly down-regulated in the absence of IKKβ, causing up-regulation of several TGFβ-controlled targets, including Decorin, Ctgf, Cald1, Postn, Il11, and Igfbp7 (Fig. 3 A). Importantly, however, Hgf mRNA coding for one of the most prominent promitogenic factors secreted by CAFs was markedly up-regulated (Fig. 3 A). HGF is a pleiotropic cytokine produced mainly by fibroblasts, and it acts on adjacent epithelial and endothelial cells by binding to cell surface c-Met receptor, promoting cell survival, proliferation, and migration via Akt, Stat3, and Wnt activation (Hoot et al., 2010; Nakamura et al., 2011; Organ and Tsao, 2011). HGF is also known to be a potent angiogenic factor, stimulating endothelial cell recruitment, motility, and growth (Bussolino et al., 1992; Trusolino et al., 2010). Overexpression of HGF or its receptor c-Met is seen in many tumors, including CRC, and is associated with poor prognosis (Stein et al., 2009; Liu et al., 2012). Accumulating evidence suggests a role of CAFs, and particularly HGF, in maintaining the cancer–stem cell niche (De Wever et al., 2008; Vermeulen et al., 2010; Quante et al., 2011). We therefore analyzed sorted EpCAM⁺ tumor cells and detected consistent up-regulation of various cancer stem cell markers, including Ascl2, Tnfrsf19, Sox2, Sox9, EphB2, EphB3, Dclk1, and Lgr5, as well as EMT master regulators Snail and Slug in tumor cells derived from Ikkβ^ΔFib mice, indicating increased stemness (Fig. 3 B). This strongly supported the possibility that the observed phenotype in Ikkβ^ΔFib mice regarding increased IEC proliferation, increased angiogenesis, and stemness could be explained by elevated HGF expression in Ikkβ-deficient fibroblasts.

To examine whether the decreased expression of proinflammatory cytokines in Ikkβ-deficient fibroblasts would affect the outcome of DSS-induced colitis, we examined animals of both genotypes on day 15 of the AOM/DSS model during the acute phase of inflammation. However, loss of Ikkβ in fibroblasts did not affect the initial DSS-induced epithelial cell death (not depicted), and therefore the extent of inflammation determined by weight loss (Fig. 4 A), histological damage, and number of ulcerations (Fig. 4, B and C) was indifferent between both genotypes. Whereas Tnfa and Il1b mRNA levels were not altered in Ikkβ^ΔFib whole mucosa, Il6 and Cxcl1 expression was decreased (Fig. 4 D). Importantly, despite no differences in the mRNA levels of Tgfb1, Tgfb2, and Tgfb3, already at this early time point Hgf and Fgf1 and bFgf gene expression was markedly elevated in mucosa of DSS-challenged Ikkβ^ΔFib mice (Fig. 4 D). This was paralleled by increased epithelial Met phosphorylation in Ikkβ^ΔFib mice (Fig. 4, E–H), which presumably was responsible for promoting IEC proliferation (Fig. 4, I–K). Myofibroblasts play an active role in regulation of the type and duration of leukocyte infiltration after acute inflammation (Pilling et al., 1999; Servais and Erez, 2013), and deregulation of this process can lead to inappropriate immune responses promoting tumorigenesis (Östman and Augsten, 2009). Fibroblasts can also suppress inflammation by inducing FoxP3⁺ regulatory T cell proliferation directly by secreting TGF-β into the microenvironment (Yang et al., 2010) or in an IL-15–dependent manner (Clark and Kupper, 2007). We therefore isolated and analyzed the lamina propria–infiltrating immune cell populations during DSS-induced colitis by flow cytometry. Although we found no differences in the relative numbers of CD45⁺, EpCAM⁺, CD31⁺, and PDGFRα⁺ cells (Fig. 5 A), or CD11b⁺F4/80⁺, CD11b⁺Gr-1⁺, and CD11b⁺CD11c⁺ myeloid cells (Fig. 5 B), we found a significant increase in the relative numbers of CD3⁺CD4⁺ T cells, as well as CD4⁺Foxp3⁺ regulatory T cells in the Ikkβ^ΔFib group. There was also a slight increase in CD4⁺IFNγ⁺ T cells, but it was not significant, the levels of CD4⁺IL17a⁺ T cells remained unchanged between the two groups (Fig. 5 C).

To examine whether elevated Hgf transcription was a direct cell autonomous effect of Ikkβ deletion in fibroblasts or an indirect consequence by an altered microenvironment, we examined HGF production in purified ex vivo cultured intestinal fibroblasts. To this end, we purified intestinal fibroblasts from tumor-bearing Ikkβ^ΔFib and Ikkβ^F/F mice, as well as on day 15 of the CAC regimen. Fibroblasts were seeded in 24-well plates and cultured to 80% confluence. After 24 h of serum starvation, they were either left untreated or stimulated with TGFβ (10 ng/ml) or EGF (10 ng/ml) separately or simultaneously over a period of up to 72 h. Independent of the time point, whether cells had been isolated or the stimulus that had been applied, after DSS-induced colitis or AOM/DSS-induced tumorigenesis Ikkβ-deficient fibroblasts secreted up to fourfold more HGF compared with cells from littermate controls (Fig. 6, A and B). Similar effects were observed when cells were treated with PMA (unpublished data). These results clearly indicated that IKKβ directly suppresses HGF production in a cell autonomous manner. So far, NF-κB has not been shown to directly bind to the Hgf promoter, yet considering the clear induction by TGFβ in Ikkβ-deficient fibroblasts and the decreased Smad7 expression in Ikkβ-deficient fibroblasts (Fig. 3 A), we speculated that IKKβ/NF-κB may affect HGF secretion via Smad7-dependent regulation of TGFβ signaling. Thus, using a retroviral approach, we overexpressed Smad7 in Ikkβ-deficient intestinal fibroblasts (Fig. 6 C), which indeed completely prevented TGFβ- and EGF-dependent HGF secretion in Ikkβ-deficient fibroblasts compared with cells that had been transduced with a GFP-control retrovirus (Fig. 6 D).

To finally confirm that enhanced HGF secretion by Ikkβ-deficient fibroblasts was responsible for the observed increase in tumor growth, we inhibited c-Met activation by feeding mice of both genotypes a diet containing a selective c-Met inhibitor INCB28060 (capmatinib; Liu et al., 2011) during the recovery phase after each DSS cycle (Fig. 7 A). Capmatinib did not affect tumor incidence, but blocked tumor growth in control animals (Fig. 7 B). However, in Ikkβ^ΔFib mice, c-Met inhibition was even more apparent and capmatinib suppressed both tumor incidence, as well as tumor size. Two out of the eight treated Ikkβ^ΔFib mice (25%) did not develop any adenomas, and capmatinib prevented the enhanced tumor growth in untreated Ikkβ^ΔFib mice. Consequently, tumor size was now comparable to those in capmatinib-treated control animals. Moreover, capmatinib blocked angiogenesis in Ikkβ^ΔFib mice to levels observed in Ikkβ^F/F tumors (Fig. 7 D) and normalized the number of regulatory T cells in the mucosa during the acute inflammatory response (Fig. 7 H). In line with decreased tumor growth, activation of Stat3 and Akt and expression of nuclear β-catenin were down-regulated in the presence of capamatinib in Ikkβ^ΔFib mice (Fig. 7, E–G). Collectively, these data confirmed that, indeed, enhanced HGF secretion by Ikkβ-deficient fibroblasts accounted for the enhanced tumor promotion in Ikkβ^ΔFib mice by regulating tumor cell proliferation, cell death, angiogenesis, and T cell polarization.

Canonical IKKβ-dependent NF-κB activation is considered a key signaling pathway linking inflammation and tumorigenesis (Greten et al., 2004). In tumor cells, as well as inflammatory immune cells, NF-κΒ controls a variety of functions that ultimately favor all stages of tumorigenesis (Karin, 2009). In line with this notion, CAFs in in pancreatic, skin, and mammary cancer express a proinflammatory NF-κB signature, and expression of dominant-negative IKKβ in cotransplanted CAFs inhibited tumorigenesis (Erez et al., 2010). Here, we report the unexpected observation that Ikkβ in Col1a2-expressing fibroblasts suppressed tumor growth in a model of CAC. This was strictly dependent on elevated HGF secretion, which stimulated cell proliferation, increased angiogenesis, and induced T reg cell polarization during colitis. Interestingly, pronounced HGF production was not limited to inflammation-associated tumorigenesis. We obtained similar results when sporadic colon tumors were initiated by AOM without additional DSS-triggered inflammation. The MAPK Tpl2 has been suggested to be involved in the upstream activation of NF-κB (Vougioukalaki et al., 2011). Interestingly, mice that lack Tpl2 in ColVI expressing intestinal myofibroblasts show a very similar HGF-dependent phenotype in the CAC model, but not in an inflammation-independent model (Koliaraki et al., 2012). Surprisingly, however, IKKβ deletion in ColVI-expressing cells causes tumor inhibition in the CAC model, as discussed by Koliaraki and Kollias in this issue. Considering that Col1a2 targets nearly 80% of PDGFRα⁺ cells, thus recombining in a much larger fraction of fibroblasts, raises the possibility that IKKβ/NF-κB may have distinct functions in different fibroblast subpopulations, possibly explaining the discrepancy between these results, as well as the previously reported findings when CAFs had been co-injected in xenografts of pancreatic and mammary cancer (Erez et al., 2010). Furthermore, the diametric opposite results in these two CAC studies here underscores the notion that CAFs represent much more than just one single cell type and provide strong evidence for the importance of different subpopulations within this cell lineage. Considering this high degree of plasticity in fibroblasts, NF-κB signaling may be differently affected or confer distinct functions depending on the respective activation status of CAF subpopulations. Nevertheless, evidence emerging form other mesenchymal cell types, including adipocytes, macrophages, and muscle-derived stem cells NF-κB negatively regulates Hgf transcription in response to proinflammatory stimuli (Yin et al., 2014; Proto et al., 2015).

Our data further highlights the great importance of HGF for colorectal carcinogenesis. HGF is a potent and pleiotropic growth factor with important roles in regeneration and homeostasis in many adult tissues (Trusolino et al., 2010). It is secreted by mesenchymal cells, especially fibroblasts in a preform, and requires cleavage by extracellular proteases to produce the biologically active heterodimer, which can then bind to its receptor c-Met and activate downstream signaling. The C-terminal motif of c-Met contains docking sites for numerous downstream signaling pathways, including PI3K, GAB1, phospho-lipase Cγ1, and STAT3 (Ponzetto et al., 1994; Trusolino et al., 2010), leading to cell proliferation, reduced apoptosis, and enhanced angiogenesis. HGF has also been identified as the stromal factor responsible for maintenance of the cancer stem cell niche (Vermeulen et al., 2010) and aberrant activation of this signaling pathway correlates with a poor prognosis in CRC (Arlt and Stein, 2009). Indeed, sorted tumor cells from Ikkβ^ΔFib mice expressed higher levels of several stem cell and EMT markers. Interestingly, HGF seemed to be also responsible for T reg cell polarization during acute colitis considering the normalization upon capmatinib administration. This is consistent with findings in preclinical models of autoimmune diseases such as osteoarthritis, autoimmune myocarditis, and in CNS demyelinating autoimmune diseases when HGF promoted T reg cell accumulation by suppressing dendritic cell function (Okunishi et al., 2007; Benkhoucha et al., 2010).

One important upstream regulator of HGF secretion in CAFs is TGF-β. TGF-βRII–deficient fibroblasts secrete elevated levels of HGF, and fibroblast-specific deletion of TGF-βRII triggers development of spontaneous prostate and forestomach cancer (Bhowmick et al., 2004). In squamous cell carcinoma (SCC), it has been shown that HGF signaling is the central mediator of myofibroblast-derived paracrine signaling promoting tumorigenesis after TGF-β-dependent fibroblast to myofibroblast conversion. SCC cells produce TGF-β, thus activating adjacent fibroblasts, and the resulting myofibroblasts produced more HGF than their precursors, which significantly promoted tumor invasion (Lewis et al., 2004). We found TGFβ1 induces HGF secretion at significantly higher levels after genetic ablation of NF-κB signaling, and we could demonstrate that SMAD7 plays a key role in the control of HGF secretion in murine CAFs. SMAD7 negatively regulates TGFβ signaling, whereas it is transcriptionally regulated by NF-κB (Baugé et al., 2008; Yan et al., 2009; Freudlsperger et al., 2013). Cross talk between NF-κB and TGFβ signaling via SMAD7 has already been described in head and neck cancer (Freudlsperger et al., 2013), osteosarcoma (Eliseev et al., 2006), and breast and gastric cancers (Hong et al., 2007). Importantly, the enhanced TGFβ signaling in IKKβ-deficient CAFs did not just promote HGF release, but also led to the up-regulation of several other TGFβ-dependent target genes that have recently been identified to predict poor prognosis in CRC patients (Calon et al., 2015). Thus, in addition to the previously identified negative regulation of IL-1β secretion causing increased susceptibility to sepsis when inhibiting IKKβ (Greten et al., 2007), our data here raise additional concern regarding the use of specific IKKβ inhibitors for the treatment of CRC. However, considering that these effects seem to be restricted to CAFs, and that in intestinal tumor cells and myeloid cells NF-κB clearly confers tumor-promoting functions (Greten et al., 2004; Schwitalla et al., 2013a,b), it remains to be determined what net effect pharmacological IKKβ inhibition may have.

Col1a2-creER^T2 mice and Rosa26R-tdTomato reporter mice were purchased form The Jackson Laboratory. Ikkβ^F/F have been described previously (Greten et al., 2004). All mice including cre-negative littermate controls were crossed on a FVB background for at least four generations and in all experiments littermate controls were used. To ensure sustained recombination mice were kept on a diet containing tamoxifen (400 mg/kg; LASvendi) throughout the duration of the CAC model. AOM/DSS-induced tumorigenesis was performed essentially as previously described (Bollrath et al., 2009). 10 mg/kg AOM (Sigma-Aldrich) was injected i.p. at day 0, 5 d before mice received 2% DSS (MP Biosystems) in the drinking water for 5 d, followed by 16 d of regular water, which was repeated twice. Severity of colitis was assessed histologically as described before (Eckmann et al., 2008). The c-Met inhibitor INCB28060 (capmatinib) was purchased from Selleckchem and incorporated (18 mg/kg) along with tamoxifen (400 mg/kg) into rodent diet AIN-76A (Research Diets, Inc.). Control groups received tamoxifen containing AIN-76A diet only. All experiments were approved by the Regierungspräsidium Darmstadt.

Colons were removed, opened longitudinally, rolled as swiss rolls, and fixed in 4% PFA overnight at 4°C. Tissues were subsequently dehydrated and paraffin embedded. 4 µm sections were used for histological analysis. For immunohistochemistry: after deparaffinization and rehydration of the tissue, antigen retrieval was done by incubating the tissue section in 10 mM sodium citrate buffer, pH 7.4 in sub boiling temperature in the microwave for 20 min. After cooling and washing the tissue, the endogenous peroxidase activity was quenched with 3% hydrogen peroxidase for 10 min at room temperature. Sections were blocked with 5% normal donkey serum in PBS containing 1% BSA and 0.1% Tween-20 for 1 h at room temperature. Endogenous avidin and biotin were blocked using the avidin-biotin kit (Vector Labs), following the manufacturer’s instructions. The following primary antibodies were used: α-BrdU (MCA2060; AbDSerotec), α-cleaved caspase 3 (9661S; Cell Signaling Technology), α-p-Stat3^Y705 (9145; Cell Signaling Technology), α-pAkt^S473 (3787S; Cell Signaling Technology), α-β-catenin (06–734; EMD Millipore), α-p-Met (ab5662; Abcam), α-E-cadherin (610182; BD). To determine cell proliferation, mice were injected i.p. with 75 mg/kg BrdU (Sigma-Aldrich) 90 min before sacrifice. Quantification of positive nuclei was done using eSlide manager version 12.0.1.5027 after image acquisition by the Aperio Scanscope XT (Leica).

Colons were harvested and thoroughly washed with PBS containing 100 U/ml penicillin/streptomycin and 50 µg/ml gentamicin sulfate. For isolation of fibroblasts on day 15 of the AOM/DSS regimen, the entire distal one-third of the colon was harvested. For isolation of CAFs only tumor tissue was harvested and used for cell separation. After repeated washing, tissue was minced and digested in 1 mg/ml collagenase I (Sigma-Aldrich), 1 mg/ml Dispase II (Roche), 50 µg/ml DNase (Sigma-Aldrich) in 10 ml RPMI containing 2% FBS for 60 min. Supernatant was filtered through a 40-µM cell strainer, and the single-cell suspension was plated on T75 flasks. Cells were cultured in complete DMEM containing 10% FCS, 1% glutamax, 1% nonessential amino acids, 1% penicillin/streptomycin. Cells between the fourth and eighth passage were used for experiments. Cell stimulation was done after a 24-h serum starvation period and using DMEM containing only 1% PenStrep (Gibco), 1% glutamax (Gibco), and 1% nonessential Amino Acids (Gibco). For RNA analysis, cells were sorted by flow cytometry and directly collected in RNA lysis buffer.

RNA from tissue/cells was harvested using RNeasy mini kit (QIAGEN). SuperScript II Reverse transcription (Life Technologies) was used for cDNA synthesis. qRT-PCR was then performed on a StepOne plus Real-Time PCR detection system (Applied Biosystems) using the SYBR Green PCR Master Mix (Roche) according to the manufacturer’s instructions. Forward and reverse primers were added at a concentration of 0.2 pmol/ml in a final volume of 20 µl. The primers used are listed in Table S1.

Immunoblotting was done using standard procedures. Protein lysates from cell pellets/tissue was separated using SDS-gel electrophoresis and transferred onto PVDF membranes. They were blocked with 5% milk in PBS containing 0.1% Tween 20 and incubated with specific antibodies. The following primary antibodies were used: α-p-Met (ab5662; Abcam) and α-SMAD-7 (sc-11392; Santa Cruz Biotechnology, Inc.). Secreted HGF in the supernatants was measured using an ELISA kit (R&D Systems) according to manufacturer’s instructions.

Flow cytometry was performed using BD FACSCanto II flow cytometer, and analysis was done using Flow-Jo software, v8.8.6 (Tree Star). A schematic representation of the preparation and gating strategy is provided in. Fig. S1. Colon was harvested, washed thoroughly, minced using sterile scalpels, and digested in 1 mg/ml collagenase I (Sigma-Aldrich), 1 mg/ml Dispase II (Roche 04942078001), 50 µg/ml DNase (Sigma-Aldrich) in 10 ml of RPMI containing 2% FBS for 60 min. The supernatant was passed through a 40 µm cell strainer and the single cell suspension was stained using following antibodies: α-CD3 APC-eFluor 780(47–0032), α-CD4 eFluor 450(48–0041), α-CD8a PerCP-Cyanine5.5 (45–0081), α-IFN-κ Alexa Fluor 488(53–7311), α-IL-4 PE-Cyanine7 (25–7041), α-IL-17A APC (17–7177), α-Foxp3 PE (12–4771), α-CD11b APC-eFluor 780 (47–0112), α-CD11c FITC (11–0114), α-F4/80 APC (17–4801), α-Ly-6G (Gr-1) PE-Cyanine7 (25–5931), α-PDGFRα PE (12–1401-81), Fixable Viability Dye eFluor 506 (65–0866; all antibodies from eBioscience). For in vitro stimulation and intracellular staining, cells were plated on 6-well plates, stimulated with PMA (20 ng/ml) and ionomycin (1 µg/ml) in RPMI medium with 10% FCS and 1% PS and 1% Brefeldin A (eBioscience) for 5 h. Cells were then harvested and stained with specific antibodies for the surface markers and a fixable viability dye. The cells were then fixed using IC fixation buffer (eBioscience) and stained with antibodies for the intracellular stains in 1× wash/perm buffer.

Endoscopic imaging and quantification of angiogenesis was done using probe-based confocal laser endomicroscopy (CLE; CellVizio; Mauna Kea Technologies), enabling real-time in-vivo microscopy of mucosal surfaces. Anesthetized mice were injected with 10 µl of 1% fluorescein i.v. and, immediately after, examined with a miniprobe (Mauna Kea Technologies). CLE images were recorded as grayscale video sequence files at 12 frames per second for ten minutes 10 min after fluorescein injection. At least 10 images per tumor were used to calculate the FCD of the tumor areas ulsing the CellVizio software.

To prepare retroviral supernatants, pBabe Flag Smad7 plasmid (a gift from R. Derynck, University of California, San Francisco, CA; plasmid #14836; Addgene) was transfected into Phoenix-Eco packaging cells using the CaPO₄ transfection protocol. The virus-containing supernatant was collected 48 h after transfection. For retroviral transduction, one degree fibroblasts were plated in 6-well dishes at 2 × 10⁵ cells (40–60%) confluence. After filtering and centrifugation, the virus supernatant containing 4 µg/ml polybrene was added and the cells were cultured for 48 h at 37°C. Selection of infected clones was done with puromycin. Control fibroblasts were transduced with eGFP retrovirus.

Data are expressed as mean ± SEM. Statistical analysis methods used were standard two-tailed Student’s t test for two datasets and ANOVA, followed by Bonferroni post-hoc test for multiple datasets using Prism4 (GraphPad). P-values ≤ 0.05 were considered significant.

Fig. S1 shows scheme of cell fractionation and confirmation of cell purity. Table S1, available as an Excel file, shows sequences of oligonucleotides used in real-time PCR.

We thank Natalia Delis, Kathleen Mohs, Christin Danneil, Eva Rudolf, and Petra Dinse for expert technical assistance.

This work was supported in part by the LOEWE Center for Cell and Gene Therapy Frankfurt (III L 4-518/17.004) and institutional funds from the Georg-Speyer-Haus, as well as grants from the Deutsche Forschungsgemeinschaft (Gr1916/11-1) and the ERC (ROSCAN-281967) to F.R. Greten.

The authors declare no competing financial interests.