Wnt1 Expression Induces Short-Range and Long-Range Cell Recruitments That Modify Mammary Tumor Development and Are Not Induced by a Cell-Autonomous β-Catenin Effector

Xenograft model studies have shown that tumor-associated, or genetically modified, activated stromal cells can promote tumor cell growth. Here, we examined mammary tumors arising in response to two different transgene-mediated Wnt signaling effectors: Wnt1 (a ligand with cell-nonautonomous effects) and ΔNβ-catenin (a constitutively active form of the intracellular effector). Although the route of tumor development has been shown to be similar for these two models, histologic analysis shows that Wnt1-induced tumors are associated with tracts of activated stroma, whereas most ΔNβ-catenin–induced tumors are solid adenocarcinomas. Furthermore, quantification of the “reactive stroma index” indicates that abundant activated stroma correlates with accelerated tumor progression. Wnt1-expressing mammary epithelial cells induce Wnt-specific target gene expression in local stromal cells (Wnt1-induced secreted protein 1/CCN4) but also induce long-range effects. Thus, mice with rapid tumor progression have 2-fold more circulating endothelial progenitor cells in peripheral blood than control or ΔNβ-catenin transgenic mice. Using tagged bone marrow (BM) transplants, we show that BM-derived cells are massively recruited to infiltrate the stroma of Wnt1-induced tumors where they differentiate into multiple cell types. Thus, localized ectopic expression of the proto-oncogene Wnt1 in mammary glands induces systemic responses, and we propose that this response modifies the tumorigenic outcome. [Cancer Res 2008;68(24):10145–53]

It has become clear that the interactions that occur between epithelial and stromal cells to regulate organ development and function are reactivated and exploited during tumor initiation and progression to produce unregulated growth and vascularization (1). Coculture and in vivo grafting analyses show that the altered stromal cells (often called activated or reactive stroma) promote tumorigenesis of nontumorigenic epithelial cells (2–4). Recent studies also suggest that stromal cells undergo genetic or epigenetic alteration as they respond to tumor cells (5–7) and that genetically modified stromal cells can initiate tumor development of epithelial cells (8). Stromal changes are known to occur in breast tumors. For instance, infiltrating ductal carcinomas, the most common type of breast cancer, show remarkable changes of stromal cell gene expression profile, extracellular matrix composition, and stromal cell morphology (9, 10). Specific changes have recently been implicated as a functional component of tumor progression by their consistent association with tumors with poor prognosis (11, 12). Because stroma comprises many organ-specific cell types, the elucidation of functional interactions requires the observation of the whole tumor process in situ, followed by deconstruction in vitro.

Wnt family members mediate epithelial-mesenchymal interactions during a variety of organogenic processes, including tooth, kidney, and mammary gland (13–15). Although Wnt ligands can induce several different signaling cascades, only the stabilization of β-catenin and binding to TCF/LEF transcription factor family members (the canonical pathway) has been shown to be tumorigenic, and aberrant activation of canonical Wnt signaling is implicated in a variety of human tumors (16, 17). Wnt transactivational responses are highly cell type–specific and have been characterized for mammary gland (18). However, the genes that are key to the mammary oncogenic program are not yet known.

Transgenic mice ectopically expressing Wnt1 ligand or a constitutively active form of β-catenin (ΔNβ-catenin) in mammary glands under the control of an MMTV-LTR promoter develop mammary tumors (19, 20). The origin of both Wnt1-induced and ΔNβ-catenin–induced adenocarcinomas seems to be similar. Both effectors induce the accumulation of stem/progenitor cells, and both induce highly differentiated (and bilineal) adenocarcinomas in a hyperplastic background (21, 22). Here, we show that only Wnt1-induced preneoplastic hyperplasias and tumors are associated with activated stroma. This resembles the fibrotic stroma induced by ectopic overexpression of matrix metalloprotease-3 (23, 24). We propose that Wnt1 expression in mammary epithelial cells activates neighboring stromal cells by a paracrine mechanism whereas ΔNβ-catenin is a cell-autonomous effector. By comparing the mechanism of tumor development in these two transgenic strains, the physiologic consequences of stromal activation can be evaluated in vivo. We show that abundant activated stroma correlates with more aggressive tumor development, and it is accompanied by the infiltration of bone marrow (BM)–derived cells that contribute to angiogenesis and stromalization.

Mice. All mice used in this study were FVB/NJ background. MMTV-Wnt1 mice were purchased from the Jackson Laboratory, and MMTV-ΔNβ-catenin mice were made and described by Dr. Pam Cowin (20). Rosa26-hPAP mice were kindly provided by Dr. Eric Sandgren (University of Wisconsin). The transgenic lines were bred through male hemizygotes. To determine the onset of mammary tumor formation, mice were palpated weekly (sensitive to 1-mm tumor masses).

Reactive stroma index. A representative H&E-stained section per tumor was examined by five blinded observers and scored for the reactive stroma index based on the percentage contribution of a stroma area in tumor mass (i.e., 0–10% stroma area = 0, 11–20% stroma area = 1, and so on, >50% stroma area = 5). The average of five points was used as the reactive stroma index for each sample.

Antibody biotinylation. Mouse monoclonal anti-hPAP antibody (clone 8B6, Sigma) and a polyclonal anti–Wnt1-induced secreted protein 1 (WISP1) antibody (sheep polyclonal, R&D Systems) were biotinylated with a protein biotinylation kit (Dojindo Molecular Tech.), as instructed by the manufacturer.

Immunohistochemistry. We used standard methods to assay protein expression by immunohistochemistry. Antibodies used in this study are as follows: anti-K5 (1:300, rabbit polyclonal, Covance), anti-FSP1 (1:100, rabbit polyclonal, Dako), anti-CD31 (clone MEC7.46, 1:50, Abcam), anti–von Willebrand factor (1:200, rabbit polyclonal, Dako), anti–FITC-conjugated α-smooth muscle actin (α-SMA;clone 1A4, 1:250, Sigma), anti-F4/80 (clone CI:A3-1, 1:100, AbD Serotec), Cy3-conjugated donkey anti-rat IgG (1:100, Jackson ImmunoResearch), Alexa Fluor 546 goat anti-rabbit IgG (1:250, Molecular Probes), Alexa Fluor 488 goat anti-rabbit IgG (1:200, Molecular Probes).

Preparation of normal mammary fibroblast cells/hyperplasia-associated fibroblast cells and growth curve determination. To generate immortalized normal mammary fibroblast cells (NMF) or Wnt1-induced hyperplasia-associated fibroblast cells (HAF), mammary fat pad from adult FVB control or FVB–MMTV-wnt1 mice were chopped with scissors into ∼1 mm³ chunks and then digested with collagenase (3 mg/mL; Worthington) and trypsin (1.5 mg/mL; Worthington) for 90 min at 37°C with agitation. Enzyme-digested tissues were washed with medium and then filtered through a 40-μm cell strainer. The filter-through fraction was mostly single cells and small organoids, and these were plated on tissue culture plates. After 1 h of incubation at 37°C, unattached cells (mostly organoids and epithelial cells) were washed off using PBS. Only fibroblast cells grew out in DMEM + 10% fetal bovine serum (FBS), deduced from their expression of the fibroblast markers (FSP1 and SMA) and their lack of epithelia marker expression (K8 and K5; data not shown).

To determine growth rate, 2 × 10⁴ cells per well were plated in 24-well plates in 10% FBS + DMEM. After overnight incubation, medium was changed to either fresh 10% or 2% FBS-containing medium. Total cell number was determined by trypsinization, followed by counting using hematocytometer. Growth curves were plotted using average number of triplicates and SD.

Quantification of microvessel density and tumor-associated macrophage density. CD31 or F4/80 stained tissue images were processed with ImageJ software to determine the total area of marker-positive cells (pixel²). For comparison of samples, the area determination was divided by the total number of nuclei in the field, using ImageJ plug-in software, ITCN (Center for Bio-Image Informatics at University of California-Santa Barbara). The procedure for image processing is described in detail in the supplementary files.

Fluorescence-activated cell sorting. To quantify circulating endothelial progenitor cell number, peripheral blood (PB) cells were collected in PBS with 50 mmol/L EDTA after heart puncture and incubated in ALK buffer [150 mmol/L NH₄Cl, 10 mmol/L KHCO₃, 0.1 mmol/L EDTA (pH 7.3)] on ice for 10 min to lyse RBC. Cells were washed twice with Hank's balanced saline + 2% FBS (HF), and cell number was counted using a hemocytometer to calculate the total PB cell number per 1 mL of blood. Cells were resuspended in incubation buffer (2% FBS, 20 mmol/L HEPES, 2 mmol/L EDTA in DMEM), and then labeled with anti–CD45-FITC (clone 30-F11, eBioscience), anti–CD34-APC (clone RAM34, eBioscience), and anti–Flk1-PE (clone Avas 12α1, BD Biosciences) antibodies at 4°C for 30 min. After washing twice with cold HF, cells were resuspended in HF containing propidium iodide (PI; 5 μg/mL) and sorted using FACS Caliber (BD Biosciences). PI-negative live cells were gated and analyzed for their expression of CD45^lowCD34⁺Flk1⁺ [characterizing the endothelial progenitor cell (EPC) population] using Flowjo (Treestar, Inc.) software. The absolute number of EPCs was calculated by multiplying the percentage of EPCs (typically <0.015%) by the total number of PB cells per mL of blood, as described by Asahara and colleagues (25).

BM transplant. Female transgenic mice (5–7 wk) were irradiated with 4.5 Gray γ-irradiation twice, separated by 3 h. Whole BM cells (10⁶) from FVB-Rosa26-hPAP mice were injected retroorbitally. Reconstitution of the hematopoietic lineage was confirmed by assaying the expression of heat-resistant alkaline phosphatase on the cell surface of PB cells 4 wk after transplantation. The percentage of reconstitution was >99%. Mice were sacrificed when tumors were palpable, and tumors were removed and either fixed in 4% PFA in PBS for overnight or embedded in ornithine carbamyl transferase solution for immunohistochemical analysis.

Tumor cell dissociation and isograft. Tumor cells were isolated as described by Orimo and colleagues (25) with some modifications. Briefly, chopped tumors were incubated in the dissociation buffer [1 g tissue/4 mL buffer; 1 mg/mL collagenase type I (Boehringer Mannheim), 125 units/mL hyaluronidase (Sigma), 20 mmol/L HEPES buffer (pH 7.5), 10% FBS in DMEM] for 3 h at 37°C with agitation. Fresh cold medium (10% FBS + DMEM) was added to enzyme-digested cells and then centrifuged at 400 × g for 10 min. Cells were resuspended with medium, and single cells were prepared by filtering through a 40-μm cell strainer. RBCs were lysed by incubating them in ALK buffer [150 mmol/L NH₄Cl, 10 mmol/L KHCO₃, 0.1 mmol/L EDTA (pH 7.3)].

For isografts of tumor cells, isogenic, 3-wk-old FVB female mice were anesthetized using avertin and endogenous epithelia of inguinal mammary fat pads were surgically removed. Tumor cells (5 × 10³) were counted and suspended in DMEM + 2% FBS with 5 μg/mL Matrigel (BD Biosciences) and injected into a small pocket that was made using a surgical forceps in the fad pad, as described previously (22).

Statistical analysis. All data sets were compared by two-sided Wilcoxon rank sum test using Mstat software (Dr. Norman Drinkwater, University of Wisconsin). All data were presented as means with SD error bars.

Real-time quantitative PCR. Inguinal #4 mammary glands were dissected out, lymph nodes were surgically removed, and then tissues were stored in RNAlater solution (Ambion). Total RNA was isolated using RNeasy Lipid Tissue Mini kit (Qiagen), as instructed by the manufacturer. Total RNA (750 μg) was reverse transcribed in a 60-μL reaction volume using QuantiTect Rev. Trasnscription kit (Qiagen). One microliter of cDNA was mixed with 5 μL of Platinum Super Mix (Platinum SYBER Green qPCR SuperMix-UDG with Rox, Invitrogen) and 4 μL of forward and reverse primer (0.5 μmol/L), and the PCR was performed using ABI7900-HT (50°C 2 min, 95°C 2 min, 45 cycles of 95°C 15 s, 53°C 30 s, 72°C 30 s, and followed by a dissociation curve). Further details of the normalization procedure and primer sequences are provided in supplementary data.

Wnt1-induced hyperplasias and tumors are associated with activated stroma. Wnt signaling is required for embryonic mammary placode formation and is at least partly mediated by periepithelial stroma (15). Several Wnt ligands are expressed during embryonic and juvenile mammary development, but not Wnt1 (26).

To determine whether activated stroma is present in Wnt1-expressing mammary tissues, we compared the histologic features of tissues from MMTV-Wnt1 mice with tissues from MMTV-ΔNβ-catenin mice and nontransgenic control mice. Because activated stroma is often associated with collagen deposition (9), we analyzed mammary tissue sections from 3-month-old animals that were stained with Masson's Trichrome (Fig. 1A). Both Wnt1-expressing and ΔNβ-catenin–expressing mammary glands developed preneoplastic hyperplasia at this age but only Wnt1-expressing hyperplastic glands were extensively surrounded by collagen-rich (blue stain) matrix with embedded stromal cells.

Differences with respect to stromal activation were even more evident for Wnt1-induced tumors. Histologic examination of H&E-stained tumor sections showed that most Wnt1-induced tumors were associated with tracts of activated stroma, whereas most ΔNβ-catenin–induced tumors were solid adenocarcinomas, which were encapsulated with a stromal sheath but showed little infiltration (Fig. 1B,, top). To assess whether tumor-associated stromal cells display reactive stroma markers, tumor sections were stained with antibody to the myofibroblast cell marker α-SMA (27). Because SMA is also expressed in myoepithelial cells, tumor sections were costained with a myoepithelial cell marker, K5, to distinguish the two different cell types. Thus, immunohistochemical analyses showed that the stromal tracts were indeed enriched in SMA-positive fibroblasts (Fig. 1B , bottom), and in many cases, those activated stromal cells also expressed another activated fibroblast cell marker, FSP1 (data not shown).

These histologic features were also found in MMTV-Wnt1 mice crossed to a C57/Bl6 background (data not shown) and were evident in tumors from independently generated strains of transgenic Wnt1 and mutant β-catenin mice (28, 29). Furthermore, Wnt10b-induced tumors displayed a fibrotic stroma that was histologically similar to Wnt1-induced tumors (30). Therefore, this is a robust phenotype that associates with the expression of epithelial Wnt ligands.

Abundant stroma correlates with accelerated tumor progression. Both Wnt1-induced and ΔNβ-catenin–induced tumor formation with quite high penetrance (77% and 96%, respectively) in a year time period but with quite different kinetics (Fig. 1C). Tumor onset (defined as the time for 10% of mice to develop tumors) was much more rapid in MMTV-Wnt1 mice (6.3 weeks, n = 48) compared with 18 weeks in MMTV-ΔNβ-catenin mice (n = 25; Supplementary Table S1). Indeed, 40% of MMTV-Wnt1 mice developed tumors before 120 days of age compared with <10% MMTV-ΔNβ-catenin mice. In contrast, the rest of MMTV-Wnt1 mice that did not develop tumors rapidly seemed to be relatively tumor-resistant, so tumors appeared slowly or not at all (reflected in the penetrance). Accordingly, MMTV-Wnt1 mice fall into (at least) two groups: one with aggressive tumor progression and the other with delayed or no tumor progression.

To quantify the grade of stromal activation for early and late developing Wnt1-induced tumors, we classified Wnt1-induced tumors into two groups [group 1, tumors that developed before 120 days of age (n = 19); group 2, tumors (n = 15) that developed later than 120 days]. The stromal indices were measured in a blinded fashion, as described in Materials and Methods. Group 1 Wnt1-induced tumors have 30% higher reactive stroma indices than group 2 tumors (3.9 ± 0.8 compared with 3.0 ± 1.1, respectively; P = 0.026), correlating abundant stroma with accelerated tumor progression. Interestingly, 60% of group 2 Wnt1-induced tumors displayed hemorrhagic “blood lakes” (Supplementary Fig. S1), whereas this was uncommon in group 1 Wnt1-induced tumors (16%). This may indicate more leaky tumor vascular structures or higher rates of necrosis in group 2 Wnt1-induced tumors.

ΔNβ-catenin–induced tumors were also (arbitrarily) divided into two approximately equal-sized groups for comparison with the Wnt1 transgenic cohort (group 1, n = 7, <190 days; group 2, n = 8, 190-360 days). None of these tumors were significantly associated with stromal cells (group 1 = 1.6 ± 1.0; group 2 = 1.2 ± 1.2), and there was no significant difference between group 1 and group 2 tumors.

Previous studies suggest that the MMTV-LTR is also active in several hematopoietic cell types (31). Therefore, we assessed whether the recruitment of BM-derived cells in Wnt1-induced tumors was an artifact of ectopic MMTV-driven Wnt1 expression in hematopoietic cells, by isografting tumor cells into nontransgenic hosts. Wnt1 and ΔNβ-catenin tumor cells were prepared, and 5 × 10³ tumor cells were transplanted into cleared fat pads of 3-week-old control female mice. Both Wnt1 and ΔNβ-catenin isografts formed palpable tumors in 45 days. Importantly, they formed histologically identical tumors to the typical Wnt1-induced or ΔNβ-catenin–induced tumors (Fig. 1D). Thus, tumors from Wnt1 tumor cell isografts are associated with activated stromal tracts and blood lakes, whereas tumors from ΔNβ-catenin tumor cell isografts form solid adenocarcinomas. Accordingly, we propose that expression of Wnt1 in hematopoietic cells is not required for inducing short-range and long-range recruitment. Rather, Wnt1-expressing tumor cells are sufficient to activate surrounding stromal cells and recruit BM-derived cells.

Wnt1 activates surrounding stromal cells. Wnt proteins are lipid modified and often considered to be short-range morphogens or even to act as autocrine factors, depending on the extracellular milieu (32, 33). For example, the short cellular range of Wnt signaling has been specifically visualized during limb development, using an Axin2 reporter gene, and found to be a few cells deep.³

Because MMTV-Wnt1 is expressed in luminal cells (31) and is therefore separated from stromal cells by the myoepithelium and basement membrane, it is important to determine whether Wnt1 directly activates surrounding stromal cells, or recruits them by an indirect mechanism. Previous studies from other groups have shown that WISP1 (CCN4) is a Wnt1/β-catenin target gene expressed specifically in the fibroblasts surrounding Wnt1-expressing tumor cells (34, 35). Accordingly, we assessed the induction of WISP1 mRNA expression in transgenic mammary glands by real-time quantitative PCR (qPCR) assay. Interestingly, WISP1 expression was induced ∼5-fold in Wnt1-expressing mammary glands but not in ΔNβ-catenin–expressing glands (or controls; Fig. 2A). To determine whether ectopic ΔNβ-catenin expression induces canonical Wnt signaling, the expression level of Wnt/β-catenin target gene Axin2 (36) was also evaluated using the same total RNA samples (Fig. 2B). Axin2 has been shown to be a uniquely robust and universal Wnt/β-catenin target gene (18, 37, 38). Wnt1-expressing glands have 1.7-fold more Axin2 mRNA than control mammary glands (P < 0.001). On the other hand, ΔNβ-catenin–expressing glands have 4-fold more Axin2 mRNA levels compared with control and 2.3-fold more than Wnt1-expressing glands (P < 0.01), indicating that ΔNβ-catenin expression activates canonical Wnt signaling in mammary gland, and it is a more potent canonical signaling agonist than Wnt1. Furthermore, we also confirmed that WISP1 protein was associated with the stroma of Wnt1-induced tumors (Fig. 2C), which is consistent with previously reported results of in situ hybridization of mRNA (34). Therefore, we propose that Wnt1 directly activates signaling in nearby stromal cells by a paracrine mechanism.

To determine whether ectopic Wnt1 expression affects the growth potential of surrounding stromal cells, we prepared stromal cells from control mammary glands (called NMFs) or Wnt1-induced hyperplastic mammary glands (called HAFs) and compared their growth rates (Fig. 2D). In a high serum (10% FBS) medium, HAFs grew faster than NMFs. Whereas NMFs arrested in low serum (2% FBS), HAFs survived and divided, indicating that HAFs have persistent changes of their phenotype, which are maintained in tissue culture condition. To test the effect of Wnt treatment on fibroblasts, NMFs were incubated in Wnt3a conditioned media or control media (contains <2% FBS) and were found to survive over 10 days of culture with added Wnt3a, but not without (Supplementary Fig. S2). Taken together, these data suggest that Wnt ligands are survival/growth factors for mammary fibroblast cells and that the stromal fibroblast components of Wnt1-induced hyperplasia may have acquired genetic or epigenetic alterations that result in increased growth/survival potential.

Angiogenesis in Wnt1-induced tumors. Because angiogenesis is a rate-limiting step in tumor progression and has been proposed to be recruited by reactive stroma (39), we compared tumor vascularization for group 1 Wnt1-induced and ΔNβ-catenin–induced tumors. Tumor sections were stained with anti-CD31 antibody to quantify microvessel density [determined by CD31-positive area (pixel²) per cell; (Fig. 3A and Supplementary Fig. S3). Analysis of 72 fields (five to eight different fields per tumor for each of six Wnt1-induced tumors and five ΔNβ-catenin–induced tumors) showed no significant difference between Wnt1-induced and ΔNβ-catenin–induced tumors in microvessel density (Fig. 3A,, bottom left). However, we found that ∼55% of microvessels in Wnt1-induced tumors were associated with activated stromal tracts (Fig. 3A,, bottom right), whereas vessels in ΔNβ-catenin–induced tumors were mostly mixed with tumor cells, some of which had a dilated structure (*; Fig. 3A). Further high-resolution analysis showed that vessels embedded in the activated stroma (SMA⁺ myofibroblast cells rich area) had a relatively thin and serpentine structure (arrow; Fig. 3B), whereas tumor cell-associated vessels were most often aggregates of CD31⁺ endothelial cells (arrow head; Fig. 3B) that resemble the dilated vessels in ΔNβ-catenin–induced tumors (Fig. 3B). Taken together, these data suggested that we should test whether there were differences between the angiogenic mechanisms that supported these two tumor types.

Abundant stroma correlates with higher numbers of circulating EPCs. Accumulating evidence suggests that BM-derived EPCs play a role in the angiogenesis that accompanies pathogenic changes, including tumorigenesis (40). Additionally, studies have shown that carcinoma-associated fibroblasts promote tumor angiogenesis by mobilizing and recruiting EPCs (25). Therefore, we assessed whether Wnt1-activated stroma induces the mobilization of EPCs. To quantify EPC mobilization, the EPC-enriched population (CD45^lowCD34⁺Flk1 VEGFR2⁺ cells) was quantified in PB cells from control (age-matched nontransgenic littermates), Wnt1-induced, and ΔNβ-catenin–induced tumor-bearing mice by flow cytometry (ref. 41; Fig. 4). Group 1 Wnt1-induced tumor-bearing mice had over 2-fold more circulating EPCs compared with control and group 2 Wnt1-induced tumor-bearing mice (P < 0.02). Although the difference was not statistically significant (P = 0.062), group 1 Wnt1-induced tumor-bearing mice had ∼1.8-fold more circulating EPCs than ΔNβ-catenin–induced tumor-bearing mice. Therefore, these data suggest that abundant activated stroma and accelerated tumor development correlate with higher numbers of circulating EPCs.

BM-derived cells are recruited to Wnt1-induced tumors. To examine the contribution of BM-derived cells to tumor vascularization, we transplanted BM cells from Rosa26-hPAP (human placenta alkaline phosphatase) mice into lethally irradiated MMTV-Wnt1 and MMTV-ΔNβ-catenin mice (Fig. 5). These irradiated transgenic mice developed tumors with similar kinetics and pathologic phenotypes to the nonirradiated transgenic mice (data not shown). Tumors from chimeric BM transgenic mice were removed and examined by immunohistochemistry using biotin-conjugated hPAP antibody. BM-derived cells (hPAP⁺) were extensively recruited to Wnt1-induced tumors, whereas only a few hPAP⁺ cells were present in ΔNβ-catenin–induced tumors (Fig. 5A).

To determine whether BM-derived cells incorporated into newly formed blood vessels, tumor sections were costained with biotin-hPAP and either CD31 or von Willebrand factor antibodies (Fig. 5B). Interestingly, most hPAP⁺ cells were perivascular, and only a subpopulation of endothelial cells was hPAP and CD31 or von Willebrand factor double positive (arrow). The quantitative analysis (total 64 different fields; 8 fields/tumor, described as % double-positive cells/total endothelial cells) showed that there were ∼2-fold more vessel-incorporated BM-derived cells in Wnt1-induced tumors than ΔNβ-catenin–induced tumors (P = 0.02), which parallels the number of circulating EPCs (Fig. 5C). BM-derived cells were enriched in stroma-associated vessels compared with epithelial tumor cells.

We assessed the tumor-associated fibroblast cells that were derived from BM (42). Double-labeling with biotin-hPAP and FSP1 antibodies showed that a high proportion of FSP1⁺ cells in the infiltrated stromal area of Wnt1-induced tumors were hPAP⁺ BM-derived cells, whereas only a few FSP1+ cells were present in ΔNβ-catenin–induced tumors (arrow; Fig. 5D). Thus, Wnt1-expressing tumor cells induce infiltration of BM-derived cells that contribute to tumor vascularization and stromalization.

More tumor-associated macrophages are recruited to Wnt1-induced tumors. Tumor-associated macrophages (TAM) have been shown to modify breast tumor development and metastasis (43). To evaluate whether stromal recruitment was associated with the recuitment of TAMs, we quantified the number of F4/80 (pan-macrophage marker)–positive cells in Wnt1-induced and ΔNβ-catenin–induced tumors (Fig. 6). Wnt1-induced tumors showed a ∼2.5-fold increase F4/80-positive area per cell compared with ΔNβ-catenin–induced tumors (Fig. 6A). The majority of TAMs in Wnt1-induced tumors were associated with stromal tracts, whereas TAM in ΔNβ-catenin–induced tumors mostly resided in the lumen of blood vessels (Fig. 6B).

Our data show that localized production of Wnt1 ligand in the mammary gland is associated with a systemic response, including localized stromal activation, long-range (humoral) BM activation, endothelial precursor cell mobilization, and the infiltration of macrophages. Although the Wnt signaling pathway is necessary for normal morphogenesis (including mammary development; ref. 44), dysregulation is highly oncogenic for (many) epithelia. Although Wnt ligands stimulate several signaling pathways, only the β-catenin/TCF canonical route has been shown to be oncogenic. We compared the process of breast tumor development in response to a secreted Wnt1 ligand with that observed in response to the cell-autonomous ΔNβ-catenin Wnt effector and found that the disease process was much more complex when Wnt ligands were ectopically expressed. Thus, some tumors seemed much faster, whereas other mice were delayed or prevented from developing tumors. ΔNβ-catenin–induced tumors are less morphologically heterogeneous, contain minimal stroma, and are more uniform in their time of appearance (median latency). Because the tumor precursor cell and the oncogenic signaling pathway are presumed to be similar for both models, we propose that the variability of the final tumor load in the Wnt1 strain is derived from altered tumor progression mediated by the stroma and subsequent body response.

Activation of stromal cells by Wnt signaling pathways. We have shown that stromal activation (measured as enlarged stromal compartments, expression of desmoplastic markers such as SMA, and activation of Wnt target genes such as WISP1) is a feature of preneoplastic Wnt1-expressing glands, but not β-catenin–induced glands. Therefore, stromal activation must depend directly upon Wnt1 ligand signaling and not upon a molecule induced by epithelial cell β-catenin/TCF transactivation.

We propose that a noncanonical Wnt response is involved in stromal activation. Thus, although canonical Wnt signaling (via a Frizzled-LRP pair) clearly induces the accumulation of nuclear ΔNβ-catenin in cultured fibroblasts, we could not observe a similar phenomenon in vivo in these glands (data not shown). However, we have shown that Wnt1 expression in mammary epithelial cells induces the Wnt target gene WISP1 in the subjacent stroma (Fig. 2A and C). Data from another group suggests that although there are TCF-binding sites in the WISP1 promoter, it is the cAMP-responsive element binding protein (CREB) binding site that directs transactivation in response to Wnt1 or ΔNβ-catenin (35). This is confirmed by a recent report that ascribes the physiologic effects of noncanonical Wnt signaling to ATF/CREB-mediated transactivation (45). These effects are likely mediated by Frizzled alone or by alternate Wnt receptors (such as Ror). During development, these noncanonical pathways are key to morphogenetic movements of whole embryos and cell motility.

Wnt1-expressing mammary epithelial cells can recruit BM-derived cells during tumor development. Our present study shows that Wnt1-expressing tumor cells can mobilize EPCs and recruit BM-derived cells. Thus, we find that the pattern of stromal colonization in isografts is characteristic of the primary tumor. When aggressive tumors, with high stromal colonization, are isografted to virgin mammary fat pads, their phenotype is similar (Fig. 1D). It is, therefore, unlikely that there is a stochastic host-derived switch that determines the tumor-stromal phenotype. These BM-derived cells then differentiate into various cell types, including endothelial cells and fibroblast cells. Furthermore, Wnt1-induced tumors are associated with ∼2.5-fold more macrophages than ΔNβ-catenin–induced tumors. These observations indicate that Wnt1-expressing tumor cells are capable of communicating with circulating or resting BM cells and recruiting them to the tumor site. Wnt proteins are lipid modified and interact with heparan sulfates and are usually considered short-range morphogens (32, 33). However, reports suggest there may be lipid carriers that could turn Wnt ligands into humoral agents (46). Our preliminary experiments could not find blood-borne Wnt signaling activity in serum from Wnt transgenic mice (using a very sensitive reporter assay in vitro; data not shown).

We propose instead that mammary stroma expresses a bioactive recruiting molecule in response to Wnt1 interaction. Two obvious candidates are stromal cell-derived factor 1 (SDF1) and vascular endothelial growth factor (VEGF). These have been shown to be major chemokines that mediate mobilization and recruitment of BM-derived cells. VEGF administration increases circulating EPCs in vivo (47), and ectopic expression of VEGF induces homing of circulating mononuclear myeloid cells (48). SDF1 expression is induced in hypoxic lesions, and it increases the migration and homing of circulating CXCR4⁺ progenitor cells (49). In addition, tumor-associated fibroblast cells express higher level of SDF1, which correlates with more circulating EPCs (25). However, we could not find evidence in support of a role for either of these mediators. Instead, using qPCR analyses, we found that Wnt1-induced hyperplastic mammary glands and tumors have ∼2-fold less SDF1α and VEGFα mRNA level compared with control mammary glands (data not shown). We have broadened the search for the paracrine effector that communicates to BM.

Whereas there is little direct data to support a major role of canonical Wnt signaling during the initiation of human breast tumors, there is data that suggests that the loss of the extracellular Wnt pathway inhibitor, secreted Frizzled-related protein 1, is important to breast cancer (50). Furthermore, a recent study showed that FZDB, a negative Wnt signaling regulator, is down-regulated in breast cancer–associated stroma compartment and that it correlates with poor prognostic outcome (11), suggesting that activation of Wnt signaling in stroma could have an important role in breast cancer development. Given the widespread expression of noncanonical Wnt ligands in mammary gland, our data suggest that up-regulated extracellular Wnt signaling could represent a mechanism for inducing localized stromal activation and systemic recognition of the premalignant lesion by the body.

No potential conflicts of interest were disclosed.

Grant support: National Cancer Institute RO1 CA90877 and McArdle Lab training award T32 CA09135 (Y.C. Kim).

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.

We thank Andreas Friedl and Bob Liu for their help during the preparation of this manuscript.