Tumor-extrinsic discoidin domain receptor 1 promotes mammary tumor growth by regulating adipose stromal interleukin 6 production in mice

Discoidin domain receptor 1 (DDR1) is a collagen receptor that mediates cell communication with the extracellular matrix (ECM). Aberrant expression and activity of DDR1 in tumor cells are known to promote tumor growth. Although elevated DDR1 levels in the stroma of breast tumors are associated with poor patient outcome, a causal role for tumor-extrinsic DDR1 in cancer promotion remains unclear. Here we report that murine mammary tumor cells transplanted to syngeneic recipient mice in which Ddr1 has been knocked out (KO) grow less robustly than in WT mice. We also found that the tumor-associated stroma in Ddr1-KO mice exhibits reduced collagen deposition compared with the WT controls, supporting a role for stromal DDR1 in ECM remodeling of the tumor microenvironment. Furthermore, the stromal–vascular fraction (SVF) of Ddr1 knockout adipose tissue, which contains committed adipose stem/progenitor cells and preadipocytes, was impaired in its ability to stimulate tumor cell migration and invasion. Cytokine array–based screening identified interleukin 6 (IL-6) as a cytokine secreted by the SVF in a DDR1-dependent manner. SVF-produced IL-6 is important for SVF-stimulated tumor cell invasion in vitro, and, using antibody-based neutralization, we show that tumor promotion by IL-6 in vivo requires DDR1. In conclusion, our work demonstrates a previously unrecognized function of DDR1 in promoting tumor growth.

that tumor promotion by IL-6 in vivo requires DDR1. In conclusion, our work demonstrates a previously unrecognized function of DDR1 in promoting tumor growth.
Excessive adiposity has been linked to increased breast cancer recurrence and mortality in both pre-and postmenopausal women (1)(2)(3)(4)(5)(6). The underlying mechanisms of adiposity-associated cancer burden are likely multifactorial, including elevated production of hormones, cytokines, reactive oxygen species, and extracellular matrix (ECM). 6 Obesity is also associated with altered adipose tissue homeostasis and metabolic reprogramming. These changes in adipose tissue could collectively impact tumor progression through both systemic and paracrine mechanisms. Given the abundance of tumor-surrounding adipocytes in breast tissue, communication between tumor and mature adipocytes has naturally been the primary focus of mechanistic studies of obesity-related cancer burden (7)(8)(9). However, it is increasingly evident that, in addition to mature adipocytes, altered abundance and properties of human adipose stem/stromal cells (ASCs) and elevated fibrosis contribute to tissue remodeling associated with tumor-associated adipose tissue (2, 10 -12). For example, we and others have shown previously that human ASCs are a significant source of local estrogens that stimulate ER␣ϩ breast tumor growth (11,(13)(14)(15).
Discoidin domain receptor 1 (DDR1) is a cell surface tyrosine kinase receptor that binds to and is activated by collagens (16 -19). DDR1 is predominantly expressed in normal epithelial cells, and its aberrant expression is associated with multiple solid cancer types. For these reasons, the current literature on DDR1 function in cancer biology has exclusively been focused on its activity in tumor cells (20 -22). However, comparative gene expression profiling shows that stromal DDR1 expression in invasive breast cancer is significantly elevated versus normal breast stroma (23) (6.4-fold, p ϭ 1 ϫ 10 Ϫ15 ), suggesting a possible DDR1 function in stromal cells during cancer progression.
In support, we previously reported a DDR1-dependent signaling pathway that regulates adipose production of estrogens in human ASC cultured in vitro (15). Furthermore, we found that the DDR1 function in human ASCs is not shared by other collagen receptors, including integrins or DDR2 (15), indicating a uniquely important role of DDR1 in regulating endocrine/paracrine ASC functions. Despite these lines of emerging evidence, there is a lack of in vivo evidence that definitively establishes a causal relationship between stromal DDR1 and cancer progression.
In this work, we utilized a Ddr1 knockout (KO) mouse model and syngeneic mouse mammary tumor cells to examine the role of host DDR1 in mammary tumor progression. To complement in vivo tumor studies, we assessed in vitro the tumor cellpromoting ability of the stromal-vascular fraction (SVF) of mouse adipose tissue, which is enriched with multipotent stem/ progenitor cells and functionally similar to human ASCs (24). We conducted comprehensive cytokine profiling to identify the adipose stroma-secreted cytokine IL-6 as an important mediator of stromal DDR1 function in tumor pathogenesis. For the first time, our data provide compelling mechanistic insight into the role of stromal DDR1 in breast tumor growth in vivo.

Genetic ablation of host Ddr1 blunts mammary tumor growth
To interrogate the role of host DDR1 in cancer progression, we used a previously established Ddr1 whole-body KO mouse model on the C57BL/6 genetic background (25). We first confirmed DDR1 protein expression in WT mouse SVF and its depletion in the counterpart from homozygous KO mice (lanes 2 and 3, Fig. 1a). The two Ddr1-dependent protein bands in WT mouse stroma are likely DDR1 isoforms because of alternative splicing (18). As a positive control for DDR1, we used primary ASCs isolated from human breast tissue (lane 1, Fig. 1a). For the in vivo tumor study, we orthotopically injected two syngeneic murine mammary tumor cell lines, AT-3 and M-Wnt, into 8-to 10-week-old female Ddr1 WT or homozygous KO recipient mice. To avoid potential animal cage-based variation, pairs of WT and KO mice from the same litter were used in tumor and cell culture experiments throughout our study. No significant body weight difference was observed between the WT and KO cohorts at the time of tumor study (data not shown).
The AT-3 cell line was derived from an murine mammary tumor virus-PyMT (polyoma middle T) transgenic mouse mammary tumor (26, whereas M-Wnt was established from an murine mammary tumor virus-Wnt-1 transgenic mouse mammary tumor (27). We chose these two tumor cell lines because both are syngeneic with C57BL/6 mice and have been used as models for triple-negative breast cancer (28 -31). Tumor sizes were assessed by caliper measurement over a period of 4 -7 weeks, and tumors were weighed upon harvest. In both AT-3 ( Fig. 1, b-d) and M-Wnt (Fig. 1, e-g) syngeneic tumor models, tumors grew more robustly in WT mice versus Ddr1 KO counterparts. Because DDR1 in tumor cells is also known to promote tumor progression, we examined DDR1 expression in tumors from WT and Ddr1 KO hosts. DDR1 protein levels in KO hosts were not lower than those in WT counterparts ( Fig. S1a), further supporting tumor-extrinsic activity of host DDR1 in promoting tumor growth in vivo. Based on Ki67 and phospho-histone H3 staining, we did not find any significant difference in tumor cell proliferation between the WT and KO cohorts (Fig. S1, b and c). However, tumors from Ddr1 KO hosts displayed elevated apoptosis, as measured by TUNEL (Fig. S2a). The same tumors from KO hosts also expressed less Ctnnb1 (␤-catenin) and Cdh2 (N-cadherin) but more Cdh1 (E-cadherin) versus those tumors in WT hosts (Fig. S2b), suggesting reduced epithelial-mesenchymal transition for tumors in Ddr1 KO hosts.

Host DDR1-dependent ECM remodeling in the tumor microenvironment
Under various physiopathological conditions, such as hypertensive nephropathy, collagen-triggered DDR1 activation is known to induce an inflammatory response, which, in turn, leads to excessive collagen synthesis and exaggerated fibrosis (19). To determine whether a similar DDR1-dependent positive feedback loop occurred in the mammary tumor microenvironment, we first conducted intratumoral collagen histochemistry with picrosirius red (PSR) on M-Wnt tumors harvested from Ddr1 WT and KO mice. Intratumoral PSR staining intensity was significantly reduced in tumors from Ddr1 KO mice versus their WT counterparts (Fig. 2, a and b), consistent with reduced collagen in Ddr1 KO mice. In further support, Col1a1 mRNA levels were markedly dampened in M-Wnt tumors in Ddr1 KO Host DDR1 promotes tumor growth through IL-6 mice (Fig. 2c). Expression of Col3a1 and ␣-smooth muscle actin (␣-SMA), another hallmark for accumulation of tumor-associated stromal cells (32), followed the same trend, but it was not statistically significant (Fig. 2c). Taken together, our findings support the notion that host DDR1 contributes to matrix remodeling in the tumor microenvironment.

Mouse SVF stimulates tumor cell migration and invasion in a stromal DDR1-dependent manner
Because we previously showed that DDR1 is important for the paracrine action of primary human ASCs cultured in vitro (15), we asked whether host mouse DDR1 played a similar role in the current animal models. Upon isolation of SVF from WT and Ddr1 KO mice, we first verified DDR1 depletion in KO-SVF by immunoblotting (Fig. 3a). An MTT assay indicated no appreciable difference in cell proliferation between WT-and KO-SVF populations (Fig. 3b). Using a Boyden chamber-based co-culture system, we found that medium conditioned by WT-SVF significantly stimulated both migration (Fig. S3) and invasion (Fig. 3, c and d) of both M-Wnt and AT-3 murine mammary tumor cells. In stark contrast, medium conditioned with Ddr1 KO-SVF had a substantially impaired ability to stimulate tumor cell migration (Fig. S1) and invasion (Fig. 3, c and d). Consistent with the in vivo finding of elevated collagen deposition in tumors from the WT host versus the Ddr1 KO host (Fig. 2), medium conditioned with WT-SVF stimulated Col1a1 mRNA expression in tumor cells to a greater extent than that conditioned with KO-SVF (Fig. 3e). Thus, the in vitro system with SVF-conditioned medium recapitulates the observed DDR1 effect on the tumor microenvironment in the syngeneic mouse tumor models.
To determine the durability of the SVF effect on tumor cells, M-Wnt and AT-3 tumor cells were retrieved from the exposure to SVF-conditioned medium and assessed alone for their invasive behaviors in the absence of the conditioned medium. As shown in Fig. S4, a and b, tumor cells that had been exposed to medium conditioned with WT-SVF retained the more aggressive invasive behavior versus their counterparts exposed to either medium alone or medium conditioned with KO-SVF. Of note, neither WT nor KO-SVF-conditioned medium had any effect on the proliferation of M-Wnt or AT-3 tumor cells (Fig. S4, c and d). Thus, our findings clearly indicate that DDR1 in mouse SVF confers to tumor cells a prolonged invasive phenotype.

Stromal DDR1 promotes tumor cell invasion by regulating SVF-secreted IL-6
To identify the DDR1-dependent, SVF-secreted factor(s) that promote tumor cell invasion, we conducted a cytokine screen using a commercial array consisting of 111 cytokines and chemokines. SVF-conditioned medium from two indepen-  Host DDR1 promotes tumor growth through IL-6 dent pairs of WT and Ddr1 KO mice was used for probing the cytokine/chemokine array. The levels of several cancer-related, SVF-secreted factors were reduced in conditioned medium from KO-SVF versus WT-SVF, including IL-6, IL-11, CCL17, and vascular endothelial growth factor ( Fig. 4a and data not shown). Because our previously published work showed that human ASC-secreted IL-6 contributes to the tumor-promoting action of ASC (33), we chose to focus on the functionality of the differential levels of IL-6 between mouse WT-and KO-SVF. We first used RT-qPCR (Fig. 4b) and ELISA (Fig. 4c) to confirm reduced IL-6 mRNA and protein levels, respectively, in KO-SVF versus WT-SVF in multiple independent pairs of WT and KO samples. Consistent with our published results for human ASCs (33), mRNA and protein expression of Cofilin, an upstream regulator of IL-6 production in the human stromal compartment, was also substantially reduced in mouse KO-SVF as compared with WT-SVF (Fig. 4, d-f).
To determine whether DDR1-dependent IL-6 secretion contributes to SVF-promoted tumor cell invasion, we used an IL-6 -neutralizing antibody to reduce IL-6 activity in WT conditioned medium. Pretreatment with the anti-IL-6 antibody obliterated the difference in invasion stimulation between WT and KO-SVF (Fig. 5, a and b), clearly indicating that IL-6 is an important mediator of the DDR1-dependent effect on tumor cell invasion. In a reciprocal experiment, addition of recombinant IL-6 reduced the difference in stimulation of tumor invasion between WT-and KO-SVF from 1.7-to 1.4-fold (Fig.  5, c and d). Because exogenous IL-6 did not completely eliminate the difference between WT-and KO-SVF, we infer from this result that additional DDR1-regulated, SVF-secreted factors besides IL-6 likely play roles in mediating host DDR1 signaling.

IL-6 stimulation of mammary tumor growth in vivo is host DDR1-dependent
To interrogate the in vivo functional relationship between DDR1 and IL-6 in cancer progression, we systemically administered the anti-IL-6 -neutralizing antibody in M-Wnt-

Host DDR1 promotes tumor growth through IL-6
bearing WT and Ddr1 KO mice (34). Tumor sizes were monitored by caliper for a 5-week period following tumor transplantation (Fig. 6a), and tumors were weighed upon harvest at the end point (Fig. 6, b and c). Consistent with the in vitro co-culture findings (Fig. 5, a and b), IL-6 neutralization in WT mice significantly mitigated tumor growth (compare the first and third columns in Fig. 6c). Notably, the same antibody treatment in Ddr1 KO mice did not lead to any further reduction in tumor growth (compare the second and fourth columns in Fig.  6c). Taken together with the findings from the above in vitro cytokine experiments, these in vivo results strongly suggest that host DDR1 is an important upstream regulator of IL-6 production and its tumor-promoting function.

Discussion
To date, published studies of DDR1 in cancer have been limited to its action in tumor cells (20,21). Using genetically engineered mice and syngeneic tumor models, we demonstrate the importance of host DDR1 in tumor growth, significantly extending the current understanding of the DDR1 tumor-promoting function. We further show that DDR1 in adipose tissuederived SVF is a previously unappreciated regulator of stromal IL-6 secretion that affects tumor cell migration and invasion. Notably, we provide compelling in vivo evidence for a functional link between host DDR1 and the tumor-promoting activity of IL-6. Given the cell surface localization of DDR1 and its innate tyrosine kinase activity, pharmacologic abrogation of host DDR1 function in tumor progression could mitigate cancer burden.
Emerging evidence indicates that adipose tissue-derived stem/progenitor cells are a significant source of stroma-secreted factors that have a profound impact on tissue regeneration as well as pathogenesis, including in cancer (2, 35, 36). Obesity is a well-known factor associated with poor prognosis for multiple cancer types, including breast cancer. Furthermore, obesity-associated fibrosis is an increasingly recognized hallmark of adipose dysfunction that is tightly associated with other adiposity-related changes such as inflammation (37). As a known collagen receptor, DDR1 is a key player in a collageninitiated positive feedback loop that ultimately results in excessive ECM accumulation in various non-cancer disease models (19). Our in vitro and in vivo data suggest that the same positive feedback loop most likely also manifests in the mammary tumor microenvironment. In this regard, DDR1-dependent secretion of various inflammatory factors, including those detected in our cytokine assay, could serve dual functions: they promote tumor growth and, at the same time, exacerbate obesity-associated ECM remodeling and inflammation. We propose that stromal DDR1 is part of a signaling network that links the ECM, stromal cells, and tumor cells in the same tumor microenvironment (see the model in Fig. 6d). Although our current work used mice receiving a normal (not a high-fat) diet for studying primary tumor growth, it will be important to investigate the role of host DDR1 in other aspects of cancer biology, including obesity-associated tumor progression and metastasis.
The tumor-promoting functions of IL-6 in both tumor and host cells have been well documented as important in inflammation and breast cancer (33, 34, 38 -45). In further support, high circulating IL-6 levels are associated with poor prognosis in breast cancer patients (46). Although our in vitro co-culture experiments focused on IL-6 secretion by adipose SVF, multiple cell types, including tumor and other stromal cells, most likely contribute to IL-6 levels in the tumor microenvironment and in the circulation. However, it is also worth noting that genetic ablation of host Ddr1 in our syngeneic mouse models completely eliminates the effect of IL-6 -neutralizing antibody on tumor growth. Furthermore, neither AT-3 nor M-Wnt mammary tumor cells express appreciable amounts of IL-6 ( Fig. S5). We therefore favor the possibility that stromal IL-6 predominantly contributes to the tumor-promoting activity of IL-6 observed in our study and that this adipose stromal pool of IL-6 is under the tight control of SVF DDR1. In support, it has been reported previously that mouse preadipocytes express significantly higher levels of IL-6 than mature adipocytes (47).
Our work does not exclude the possible involvement of other DDR1-dependent, SVF-secreted factors besides IL-6 that could also contribute to tumor promotion. Consistent with this possibility, Ddr1 KO-conditioned medium still exhibited lower invasion-promoting activity than WT-condition medium, even in the presence of an excessive amount of recombinant IL-6 (Fig. 5, c and d; 10 ng/ml recombinant IL-6 versus 0.7 and 0.2 ng/ml for endogenous IL-6 in WT-and KO-SVF, respectively). Further, with IL-6 -neutralizing antibody, tumor growth in Ddr1 KO mice was still slower versus WT counterparts. Although this difference could be due to incomplete IL-6 neutralization in vivo, an alternative explanation is that host DDR1 regulates expression of additional tumor-promoting factors besides IL-6. Potential candidates for future investigation include cytokines and chemokines that displayed differential levels between WT and Ddr1 KO-derived SVF. Although the

Host DDR1 promotes tumor growth through IL-6
current whole-body Ddr1 KO animal model allows us to distinguish DDR1 in tumor versus host cells, DDR1 from multiple host cell and tissue types besides adipose tissue-derived stromal cells could also influence tumor growth. There is currently no suitable Cre-lox system in vivo to create adipose SVFspecific gene KO in mice. More sophisticated tissue-specific mouse models could shed light on this question.
This study of syngeneic mouse tumor models significantly extends our previously published work using cultured primary human ASCs (15,33). Although human samples obviously bear more clinical relevance, the use of genetically engineered mouse models circumvents individual-based variation associated with clinical cohorts. Furthermore, compared with xenograft tumor models, the syngeneic tumor models used in this work ensure an immune-competent host environment that more faithfully recapitulates the tumor microenvironment in humans. It is satisfying that both lines of complementary investigation in mouse and human systems clearly point to an unequivocal role of a conserved DDR1-dependent signaling pathway that dictates production of a tumor-promoting secretome in breast cancer.

Primary cells isolation and culture
Primary SVF was isolated from mouse inguinal fat pads using standard procedures (48). Briefly, harvested fat tissue was digested for 15-18 h on a rotating shaker at 37°C using Dulbecco's modified Eagle's medium/F12 (Stem Cell Technologies, catalog no. 36254) supplemented with 10% gentle collagenase/ hyaluronidase (Stem Cell Technologies, catalog no. 07912) and 1% penicillin/streptomycin plus 2% fetal bovine serum (FBS). Debris was removed by a 70-m cell strainer (BD Biosciences, catalog no. 352350), followed by centrifugation at 100 ϫ g for 5 min at 4°C. Red blood cells were removed by lysis using ammonium chloride solution (Stem Cell Technologies, catalog no. 07800) at room temperature for 5 min, followed by centrifugation at 600 ϫ g for 5 min at 4°C. Cell pellets were washed and resuspended in SVF culture medium (Dulbecco's modified

In vivo tumor study and IL-6 antibody neutralization
Ddr1 ϩ/Ϫ mice on the pure C57BL/6 background were bred to generate Ddr1 ϩ/ϩ (WT) and Ddr1 Ϫ/Ϫ (KO) littermates, which were used in all in vitro and in vivo experiments. Murine mammary tumor cells (M-Wnt and AT-3) were propagated up to 70% confluency, harvested by trypsinization (0.05% trypsin-EDTA), washed, and resuspended in PBS. Cells were injected into the mouse mammary gland fat pad using 1 ϫ 10 5 cells/100 l for M-Wnt and 2 ϫ 10 5 cells/100 l for AT-3 per injection. Tumor volume was measured by caliper at the indicated time points, and tumor growth was measured for 4 to 7 weeks. Tumor weight was measured at the time of termination. In vivo IL-6 neutralization was performed by intraperitoneal injection of IL-6 -neutralizing antibody (BD Biosciences, clone MP5-20F3, catalog no. 554398) or isotype control anti-IgG (BD Biosciences, clone R3-34, catalog no. 554682) at 100 g per mouse every 3 days starting 2 days before tumor inoculation. All animal experiments were performed after obtaining approval from the University of Texas Health San Antonio Institutional Animal Care and Use Committee. All methods were carried out in accordance with the Institutional Animal Care and Use Committee-approved guidelines.

Picrosirius red staining
Tumors harvested from mice were fixed in 10% neutral-buffered formalin, dehydrated, embedded in paraffin, and sectioned at 3-m thickness. The picrosirius red staining was performed as described by the manufacturer using 0.2% phosphomolybdic acid (Electron Microscopy Sciences, catalog no. RT-26357-01), Sirius red 0.1% in saturated picric acid (Electron Microscopy Sciences, catalog no. RT-26357-02), and 0.01 N hydrochloric acid (Electron Microscopy Sciences, catalog no. RT-26357-03). The intensity of positive staining was estimated by measuring the optical density ϭ log [I B /I O ], where I B is the average intensity in the background, and I O is the average intensity of the stained area. The percent positive staining was calculated by taking the ratio of area of pixels stained to the total area of pixels in the background. A minimum of 10 fields per tumor section were measured and averaged by ImageJ.

Cell migration and invasion assay
Tumor cells grown to 80% confluency were harvested by trypsinization, washed with PBS twice, resuspended in culture medium without FBS, and co-cultured with but physically separated from either Ddr1 WT-SVF or KO-SVF in a Transwell system. Briefly, 5 ϫ 10 4 tumor cells were seeded in the top chamber. The bottom chamber was filled with medium with 1% FBS alone or medium with 1% FBS plus 3 ϫ 10 4 WT-SVF or KO-SVF. After 12 h of co-culture at 37°C, unmigrated cells on the upper side of the insert were gently removed with a cotton swab. Migrated cells on the undersurface of the inserts were stained by crystal violet. Six fields per insert were counted under an optical microscope.
For the cell invasion assay, the inserts were overlaid with ice-cold Matrigel Basement Membrane Matrix Growth Factor Reduced (Corning, catalog no. 354483) at 5 mg/ml with 50 l Matrigel/well. The Matrigel was allowed to settle for 30 min at 37°C. 5 ϫ 10 4 tumor cells were seeded on the top chamber, and either medium alone or SVF was seeded in the bottom chamber. After 20 h, invaded cells were stained by crystal violet and counted by optical microscopy as described above for the migration assay. The invasion assay in Fig. 5 was conducted using 10 ng/ml recombinant mouse IL-6 (R&D Systems, catalog no. 406-ML-005) supplementation or 1 g/ml anti-IL6neutralizing antibody (R&D Systems, catalog no. MAB406) in the bottom chamber with either WT-or KO-SVF. For the experiment shown in Fig. S2, a and b, tumor cells were removed from the top chamber of the inserts (0.4-m pore, Millipore, catalog no. MCHT12H48) after 3-day incubation with either medium alone or SVF seeded in the bottom chamber. Tumor cells were then examined in the Boyden chamber assay for cell invasion with 10% FBS-containing medium in the bottom chamber without SVF-conditioned medium.

RT-qPCR
RNA samples were reverse-transcribed using the ImProm-II reverse transcription system (Promega, catalog no. A3800). Real-time PCR was set up using Luminaris Color High Green High ROX qPCR Master Mix (Thermo Fisher Scientific, catalog no. K0364) and run in an Applied Biosystems 7900HT work station equipped with SDS 2.4 software. All primers used for RT-PCR were designed using Primer-3 software (Sigma-Aldrich). The primer sequences were as follows:

Western blotting
Protein lysates were prepared in Laemmli sample buffer, and the protein amount was estimated using a BCA protein assay kit (Pierce, catalog no. 23225). For DDR1 Western blotting, samples were run on SDS-PAGE and transferred to an H-bond nitrocellulose membrane using standard procedures. The membrane was blocked using 5% BSA and immunoblotted with anti-DDR1 (D1G6) XP rabbit mAb (Cell Signaling Technology, catalog no. 5583), GAPDH ( 14 C10) rabbit mAb (Cell Signaling Technology, catalog no. 2118), Cofilin mouse mAb (Santa Cruz Biotechnology, catalog no. SC-53934), and ␣-tubulin mouse mAb (Calbiochem, catalog no. CP06). Protein detection was done using ECL Plus Western blotting substrate (Pierce, catalog no. 32132).

Host DDR1 promotes tumor growth through IL-6 Cytokine array/ELISA
A membrane-based antibody array was used to detect the secretome of Ddr1 WT-and KO-SVF. To reduce the baseline levels of growth factors and cytokines in FBS, SVF was cultured in medium containing 1% FBS (heat-inactivated) for 2 days. Cell-free supernatant was collected and used in the Proteome Profiler Mouse XL cytokine array kit (R&D Systems, catalog no. ARY028) following the instructions of the manufacturer. For the ELISA, SVF was incubated in 1% FBS (heat-inactivated) medium for 2 days. Concentrations of IL-6 in the WT-and KO-SVF cell supernatant or AT-3/M-Wnt tumor cell supernatant were detected utilizing the mouse IL-6 ELISA kit (Invitrogen, catalog no. KMC0061) according to the kit instructions.
MTT proliferation assay 1 ϫ 10 3 cells were seeded into a 96-well plate and incubated for 1-4 days. 50 l of MTT solution (3 mg/ml) was added to the culture medium at the indicated time point and incubated at 37°C for 1 h until the purple precipitate became visible. The medium was removed without disturbing the purple precipitate. 100 l of DMSO was added to each well and mixed homogeneously, and the absorbance was recorded at 570 nm. For the experiment shown in Fig. S2, c and d, tumor cells were trypsinized from the top chamber of the inserts (0.4-m pore; Millipore, catalog no. MCHT12H48) after 3 days of co-culture with either medium alone or SVF seeded in the bottom chamber. Tumor cells were seeded into a 96-well plate and incubated for 2 days to monitor the proliferation rate as described above.
For the TUNEL assay, slides were deparaffinized by graded ethanol (100%, 90%, 70%, and 60%) and PBS wash and subsequently permeabilized by 0.2% Triton X-100 wash for 5 min. Nick-end DNA fragmentations were labeled with the DeadEnd Fluorometic TUNEL system kit (Promega, catalog no. G3250) according to the instructions of the manufacturer. Slides were washed with PBS and mounted with Vectashield mounting medium with DAPI (Vector Laboratories, catalog no. H-1200). TUNEL-positive cells were visualized by fluorescence microscopy and quantified by ImageJ software.

Statistical methods
Mean comparison of two groups was assessed by two-tailed Student's t test. Mean differences of multiple groups were examined by one-way analysis of variance followed by multiple comparison tests. Two-way analysis of variance was used for tumor growth curve analysis. In all assays, p Ͻ 0.05 was considered statistically significant.