Ets2 in Tumor Fibroblasts Promotes Angiogenesis in Breast Cancer

Tumor fibroblasts are active partners in tumor progression, but the genes and pathways that mediate this collaboration are ill-defined. Previous work demonstrates that Ets2 function in stromal cells significantly contributes to breast tumor progression. Conditional mouse models were used to study the function of Ets2 in both mammary stromal fibroblasts and epithelial cells. Conditional inactivation of Ets2 in stromal fibroblasts in PyMT and ErbB2 driven tumors significantly reduced tumor growth, however deletion of Ets2 in epithelial cells in the PyMT model had no significant effect. Analysis of gene expression in fibroblasts revealed a tumor- and Ets2-dependent gene signature that was enriched in genes important for ECM remodeling, cell migration, and angiogenesis in both PyMT and ErbB2 driven-tumors. Consistent with these results, PyMT and ErbB2 tumors lacking Ets2 in fibroblasts had fewer functional blood vessels, and Ets2 in fibroblasts elicited changes in gene expression in tumor endothelial cells consistent with this phenotype. An in vivo angiogenesis assay revealed the ability of Ets2 in fibroblasts to promote blood vessel formation in the absence of tumor cells. Importantly, the Ets2-dependent gene expression signatures from both mouse models were able to distinguish human breast tumor stroma from normal stroma, and correlated with patient outcomes in two whole tumor breast cancer data sets. The data reveals a key function for Ets2 in tumor fibroblasts in signaling to endothelial cells to promote tumor angiogenesis. The results highlight the collaborative networks that orchestrate communication between stromal cells and tumor cells, and suggest that targeting tumor fibroblasts may be an effective strategy for developing novel anti-angiogenic therapies.


Introduction
Although the microenvironment of normal tissue can inhibit tumor growth, changes induced by tumor cells actually influence this microenvironment to promote tumor progression and metastasis [1,2,3]. For example, disruption of Tgfbr2 in fibroblasts leads to growth and increased oncogenic potential of adjacent epithelial cells in stomach and prostate [4]. Cancer associated fibroblasts (CAFs) were also shown to mediate immune cell recruitment in an NF-kB dependent manner in a mouse model of squamous cell carcinoma [5]. In addition, recent evidence demonstrates that a human stromal gene signature can be correlated with tumor prognosis and evolution of breast cancer [6]. Stromal fibroblasts are one of the major components within the tumor microenvironment that drive a number of characteristics associated with tumor malignancy, including remodeling of the extracellular matrix, immune cell recruitment and modulation, angiogenesis and tumor cell proliferation and invasion [7,8]. Identification of specific molecules and signaling pathways within this stroma which promote tumor initiation and progression is essential for the potential development of novel therapeutics to be used independently or in combination with other tumor cellspecific treatments [9].
Ets2 (v-ets erythroblastosis virus E26 oncogene homologue 2) is an evolutionarily conserved proto-oncogene known to be overexpressed in a variety of human cancers including breast and prostate [10,11]. As a downstream effector of the Ras/Raf/ MAPK pathway, Ets2 regulates the expression of a number of genes with potentially important functions in the tumor microenvironment, including growth factors, adhesion molecules, extracellular proteases and anti-apoptotic genes. In mouse mammary tumor models, Ets2 was shown to support tumor growth from an undefined cell population in the tumor stroma [12,13].
Previous work from our group has shown Ets2 functions to promote tumor progression and metastasis in at least two compartments of the mammary tumor microenvironment, including tumor macrophages and breast stromal fibroblasts in which Pten was deleted [14,15]. However, questions concerning both the fibroblast-specific and oncogene-dependent functions of Ets2, as well as the molecular and cellular mechanisms of action remain unanswered. To address these issues, we used a Cre/loxP genetic approach to investigate the role of Ets2 in mammary epithelial cells and mammary fibroblasts in the context of PyMT and ErbB2 driven tumors. This approach revealed that fibroblast Ets2 promotes tumor growth and progression, whereas its function in epithelial cells is dispensable for tumor progression. Transcriptome profiling of both normal and tumor fibroblasts revealed tumor specific targets of Ets2 including genes involved in matrix remodeling and angiogenesis in the context of both PyMT and ErbB2. Consistent with the gene expression results, loss of Ets2 in tumor fibroblasts decreased tumor associated blood vessels in both models. Additionally, Ets2 deficiency in tumor fibroblasts significantly impaired angiogenesis in vivo in matrigel plug assays. The fibroblast Ets2 transcriptomes were represented in human breast cancer stroma and correlated with patient outcome in independent whole-tumor breast cancer gene expression data sets. These results indicate that Ets2 functions in a tumor-dependent but tumor cell non-autonomous manner to affect tumor growth and angiogenesis.

Ethics Statement
The use of animals was in compliance with federal and University Laboratory Animal Resources at The Ohio State University regulations and was conducted under under protocol number 2007A0120-R1 which was reviewed and approved by the Ohio State University Institutional Animal Care and Use Committee.

Tumor Studies
All mice used in these studies were inbred 10 generations to the FVB/N (F10) background. The generation of Fsp-Cre and Ets2 loxP mice has been previously described [15,16]. Ets2 db mice were a gift from Dr. Robert Oshima (The Burnham Institute, La Jolla, California). PyMT tumor bearing mice were sacrificed 4 weeks after palpable tumors were identified. Excised tumors were weighed and processed for histological analysis. X-gal staining was done as described elsewhere [16]. ErbB2 mice were sacrificed at 16 weeks of age, and all mammary glands were harvested and processed for histological analysis. The total area of tumor lesions was calculated using Zeiss AxioVision 3.1 software.

Lectin Injections and Matrigel Plug Assay
Fluorescein or DyLight 594 Lycopersicon esculentum lectin (Vector Laboratories) was used to visualize blood vessels as described [17]. For the in vivo matrigel angiogenesis assays, 250,000 primary fibroblasts were mixed with ice cold growth factor reduced matrigel (BD Bioscience) and subcutaneously injected into the flanks of two month old FVB/N mice. Matrigel plugs were retrieved 5 days post injection for subsequent analysis.
X-gal Staining, Immunostaining, Gelatinase in situ Zymography, and Microscopy Imaging and Quantification X-gal staining was performed as previously described [15]. The following antibodies were used in this study: CD31 (BD Bioscience), Ki67 (Dako) and cleaved caspase-3 (Cell Signaling). Appropriate secondary antibodies were used for IF or IHC as described elsewhere [15]. DQ-gelatin (Molecular Probes) was used for gelatinase in situ zymography as described by Mook et al. [18]. For CD31 and lectin quantification, images were made from multiple tumors in at least three different mice; a minimum of five images were taken per mouse from random tumor fields. Microscopic images for CD31, lectin and zymography quantification were acquired with a Zeiss Axiocam HRc camera and quantified using ImageJ (http://rsbweb.nih.gov/ij/), Fiji [19] or MetaMorph softwares, whereby the percentage of total positive area was calculated. For quantification of Ki67 and cleaved caspase 3 staining, images from random tumor fields were taken, and the total number of positive cells was represented as a percentage of the total number of tumor cells.
Whole mount staining. Inguinal mammary glands were isolated from 6 week or 14 days pregnant female mice and mounted on glass slides using a Q-tip to spread the tissue. Tissue was fixed using Carnoy's fixative (60% ethanol, 30% chloroform and 10% acetic acid) at 4uC overnight, and then stained with carmine alum stain (Sigma) for 12 hours. After thorough rinsing in water, mammary glands were dehydrated with alcohol and preserved in xylene.

Primary Mammary Fibroblast Culture, Endothelial Cell Isolation and Microarray Analysis
Mammary fibroblasts were isolated from female mice using gravity separation [20]. Briefly, mammary tissue was digested with collagenase/hyaluronidase overnight in a CO 2 incubator. After neutralization, cells were separated by gravity five times. Supernatant containing stromal cells was plated with 10% FBS DMEM. Second passage cells were used for subsequent analysis. For endothelial cell isolation, mammary glands were digested with collagenase until a single cell suspension was achieved. Cells were then stained with fluorescently conjugated CD31 antibody (BD Pharmingen) and subjected to cell sorting. RNA was extracted with Trizol (Invitrogen) and purified with RNAeasy kit (Qiagen) following manufacturers' instructions. Microarray profiling was performed with Affymetrix Mouse Genome 430A 2.0 or 430 2.0 GeneChips at the OSUCCC microarray facility. All microarray data has been deposited and can be accessed through the following links: http://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?token = dberdqwuosoegdi&acc = GSE16989 http://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?token = flmxjguyowqmojq&acc = GSE44166 http://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?token = txmnnecmegeaopw&acc = GSE44118 Differentially expressed genes were identified by the WEDGE ++ method [21] or Robust Multichip Average method (RMA) [22]. Quantitative RT procedures are described elsewhere [23]. Primers and probes used for real time PCR are available in Table S6. categories in ToppGene Suite [25]. Statistical significance was determined using 1,000 random permutations of each gene set to obtain a nominal P value.
Generating the human stroma heat map using the Ets2 gene lists. Of the 107 and 69 genes regulated by Ets2 in PyMT and ErbB2 derived tumor fibroblasts, 88 and 61 had human orthologs identified using Ensembl and MGI databases. When queried against the McGill Cancer Center's Breast Stroma Microarray data (GSE9014 and GSE4823), 85 and 57 genes were also represented on the Agilent custom array used by the McGill group. To identify the genes most differentially expressed in human tumor stroma versus normal stroma, a variance cutoff of.0.5 was applied across the multiple human samples. The 38 PyMT derived and 36 ErbB2 derived genes which met this cutoff criterion were subsequently used to generate a heat map for the human stroma dataset (52 normal stroma and 49 tumor stroma samples). The separation of normal stroma and tumor stroma was significant using gene signatures from both models (P = 0.00085 for PyMT and 0.00004 for ErbB2), as determined by 10,000 random permutations using the one sided Wilcoxon rank sum test. Similar analysis was used for our PyMT derived Ets2 dependent endothelial cell data, with 22 of 33 human orthologs represented in the McGill breast stroma data. In this case, the separation of normal stroma and tumor stroma was not found to be significant (p-value = 0.1014).
Kaplan-meier survival analysis. The Ets2 tumor fibroblast 38 and 36 gene signatures as well as the endothelial 22 gene signature were used to query gene expression data from three independent breast cancer microarrays -Rosetta (NKI) dataset, Stockholm (GSE1456) and the McGill University (GSE9014) dataset. A Cox proportional hazards model associated to a L2 norm constraint [26] was applied to estimate the coefficients b associated to each gene from the human orthologs corresponding to the Ets2tumor fibroblast 38 or 36 gene signatures. A prognostic index for each patient is defined as the sum of the gene expression levels pondered by the estimated coefficients b. A 10-fold cross validation was used to predict the outcome for each patient: the patients are randomly divided into 10 groups; 9 groups are chosen to form the training set while the last group represents the test set. The coefficients b are estimated from the training set and the prognostic index for each patient in the test set is computed. This step is repeated 10 times to consider each group of patients as the test set once. The 10-fold cross validation is performed 10 times in order to be independent of the random partitioning. A prognostic index threshold is then defined as the 50 percentile of all prognostic indexes. All the patients that have a prognostic index superior (respectively inferior) to this threshold are considered as high risk (respectively low risk) patients. A logrank test is finally applied to evaluate the statistical significance between the two groups.

Statistical Analysis
Number of animals and experiments are as indicated in the figures. Analysis of variance (ANOVA) model assuming unequal variance with Holm's methods was used to study the differences among groups in the PyMT tumor study (Fig. 1A). A nonparametric Mann-Whitney test was used to compare lesion sizes in ErbB2 tumors ( Fig. 2A, left panel). A Chi square test was used to compare the number of ErbB2 lesions larger than 1mm 2 ( Fig. 2A, right panel). A Welch's t-test assuming unequal variance was used to calculate p-values for Ki67 staining and all other immunofluorescence staining.

Ets2 Deletion Specifically Effects Tumor Progression from the Fibroblast Compartment in Multiple Breast Cancer Models
The ablation of a conditional Ets2 loxP allele [27] was studied both in mammary stromal fibroblasts and in mammary tumor epithelial cells in the MMTV-PyMT (PyMT) mammary cancer model [28]. For this purpose, Fibroblast Specific Protein 1 promoter-Cre (Fsp-Cre) was used for specific deletion in tumor fibroblasts, while MMTV-Cre was used to target epithelial cells [15,29]. Initial experiments indicated that deletion of both alleles in homozygous Ets2 loxP/loxP mice was inefficient with several different Cre-driver lines [14,15], including Fsp-Cre (data not shown). In order to facilitate the complete ablation of Ets2 within the fibroblast and epithelial compartments, mice were generated carrying the Cre transgene along with one conditional Ets2 loxP allele and a conventional Ets2 knockout allele targeting the DNAbinding domain, Ets2 db [30]. We evaluated tumor development in both PyMT;Fsp-Cre;Ets2 db/loxP and PyMT;MMTV-Cre;Ets2 db/loxP experimental groups and in control PyMT;Ets2 db/loxP mice. No difference was observed in tumor initiation; however tumor progression was significantly impeded by the ablation of Ets2 specifically in mammary fibroblasts. The average weight of tumors 30 days after tumor initiation in PyMT;Fsp-Cre;Ets2 db/loxP mice was 1.5-fold lower than in either control PyMT;Ets2 db/loxP or PyMT;MMTV-Cre;Ets2 db/loxP mice (P,0.01, Fig. 1A). Additionally, H&E staining of inguinal mammary glands from control PyMT;Ets2 db/loxP mice showed solid tumors throughout the mammary gland, whereas tumors had not invaded the entire fat pad of PyMT;Fsp-Cre;Ets2 db/loxP mammary glands (Fig. 1B, yellow outline indicates area with less tumor invasion).
The specificities of both Fsp-Cre and MMTV-Cre were confirmed by introducing the Rosa26 LoxP reporter gene into a subset of the experimental mice [31]. X-gal staining of mammary tumor sections confirmed the stromal-or epithelial-specific expression of the respective Cre transgenes (Fig. S1A). PCR-based analysis of genomic DNA from epithelial cells isolated from PyMT;Ets2 db/loxP or PyMT;MMTV-Cre;Ets2 db/loxP mice demonstrated efficient recombination of the single Ets2 loxP allele (Fig. S1B). Western blot assays confirmed the loss of ETS2 [32] protein in mammary stromal fibroblasts isolated from PyMT;Fsp-Cre;Ets2 db/loxP mice (Fig. S1C). Cultured mammary fibroblasts stained with antiphospho-Ets2 pT72 antibody further demonstrated the loss of ETS2 activation in fibroblasts from PyMT;Fsp-Cre;Ets2 db/loxP mice [15] (Fig. S1D). Interestingly, whole mount analysis of normal developing and pregnant mammary glands showed similar branching of epithelial ducts in both Ets2 db/loxP and Fsp-Cre;Ets2 db/loxP animals (Fig. S1F, S1G), suggesting Ets2 expression in fibroblasts was not essential for proper mammary gland development. Additionally, whole mount analysis of mammary glands isolated form 6-week old PyMT;Ets2 db/loxP animals revealed extensive neoplastic progression, however tumors in mammary glands from PyMT;Fsp-Cre;Ets2 db/loxP mice remained restricted to the nipple area, with little or no invasion into and beyond the mammary lymph node (Fig. S1H).
Careful examination of tumor histology revealed complete disruption of the basement membrane and invasive carcinoma in tumors from PyMT;Ets2 db/loxP mice, while many tumor lesions in PyMT;Fsp-Cre;Ets2 db/loxP mammary glands were restricted to the confines of the basement membrane (Fig. 1C). As assessed by Ki67 staining, there was an approximate 2.5 fold reduction in epithelial cell proliferation in tumors with Ets2 deleted in the fibroblast compartment (P,0.01, Fig. 1D). Staining with cleavedcaspase-3 failed to detect any difference in apoptosis between genetic groups (Fig. S1E), indicating that stromal Ets2 impacts tumor growth through modulation of epithelial proliferation.
To determine if fibroblast Ets2 function was oncogene dependent, we combined the MMTV-ErbB2 oncogene with Ets2 db/loxP and Fsp-Cre;Ets2 db/loxP mice. Since an expected delay  in tumor progression was predicted based on our results in the PyMT model, mice harboring the ErbB2 oncogene were harvested at 16 weeks of age, just before palpable tumors were detected in this model. Using consecutive histological sections, we calculated the number and sizes of carcinoma lesions. The ErbB2;Fsp-Cre;Ets2 db/loxP mice had nearly half as many overall lesions compared to those found in ErbB2;Ets2 db/loxP mice (P,0.01, Fig. 2A, left panel, Fig. 2B). The lesions were also significantly smaller in ErbB2;Fsp-Cre;Ets2 db/loxP compared to controls ( Fig. 2A). Closer examination of the data revealed a three-fold decrease in the number of lesions.1mm 2 in size in mammary glands from ErbB2;Fsp-Cre;Ets2 db/loxP mice (P,0.01, Fig. 2A, right panel). As illustrated by higher magnification H&E staining, deletion of Ets2 in tumor fibroblasts significantly reduced the expansion of carcinoma lesions into the surrounding fat pad tissue (Fig. 2C). As in the PyMT model, Ki67 staining revealed a 2-fold decrease in proliferation of epithelial cells surrounded by Ets2 null fibroblasts, implicating a similar mechanism of controlling tumor growth in this model as well (Fig. 2D). Collectively, these data support a stromal fibroblast-specific function for Ets2 in facilitating the progression and invasion of mammary tumors induced by both PyMT and ErbB2 epithelial expression.

Gene Expression Changes Reveal the Tumor Specific Effects of Ets2
To uncover the molecular mechanisms underlying Ets2 function in stromal fibroblasts in both PyMT and ErbB2 induced tumors, we profiled RNA isolated from primary mammary fibroblasts for changes in global gene expression (Affymetrix Mouse Genome 430A 2.0 GeneChip platform or 430 2.0 GeneChip platform). Fibroblasts were isolated when tumors were not yet invasive, from 9-week old mice in the PyMT model and from 16-week old mice in the ErbB2 model. Fibroblasts from Ets2 db/loxP and Fsp-Cre;Ets2 db/loxP control mice were also isolated at the same time points and RNA was subjected to gene expression profiling. Assessment of cultured fibroblast purity has been previously described [15]; staining of cultured mammary fibroblasts with both vimentin and antiphospho-Ets2 pT72 antibodies confirmed both the enrichment for fibroblasts (.95% fibroblasts) and the efficiency of Ets2 conditional knockout (Fig. S1D). To identify Ets2 target genes specifc to tumor fibroblasts, we compared expression changes between normal fibroblasts at both ages (Fsp-Cre;Ets2 db/loxP vs. Ets2 db/loxP ) and tumor associated fibroblasts from both models (TAFs; PyMT;Fsp-Cre;Ets2 db/loxP vs. PyMT;Ets2 db/loxP and ErbB2;Fsp-Cre;Ets2 db/loxP vs. ErbB2;Ets2 db/loxP ). Comparison of normal fibroblasts from 9-10 week old mice identified only 22 genes to be differentially expressed upon Ets2 inactivation (fold change.4 and negative log p-value (NLP).4.5, Fig. S2A and Table S1). In contrast, when the PyMT oncongene was present in epithelial cells, 107 genes were differentially expressed with Ets2 ablation in fibroblasts (fold change.4 and NLP.4.5, Fig. 3A, Fig. S2A and Table S2). Consistent with a tumor-specific role for Ets2, only 4 genes were in common between normal and tumor fibroblasts (Fig. S2A). Our analysis using samples from the ErbB2 model yielded a similar tumor specific Ets2dependent gene expression pattern, with 15 genes identified as differentially expressed in normal Ets2 null fibroblasts isolated from 16-week old mice compared to controls, whereas 69 genes changed when Ets2 was inactivated in ErbB2 tumor associated fibroblasts (Fig. 3C, Fig. S2C, Table S3 and Table S4). In the ErbB2 model 1 gene was commonly mis-regulated in between normal and tumor fibroblasts. (Fig. S2C). Quantitative real-time PCR (qRT-PCR) was used to confirm the microarray data, and.85% of the expression changes were verified in at least two independent sets of RNA from tumor fibroblasts isolated from either genetic model (Fig. S2E, S2F).
Gene set enrichment analysis (GSEA) was used to analyze the data [24]. ECM remodeling, cell migration and cell growth-related gene sets were found to be significantly differentially expressed in PyMT tumor fibroblasts compared to cells lacking Ets2 (Fig. 3B). Angiogenesis was another biological process affected by Ets2 deletion, but the trend did not reach statistical significance (Fig. 3B). GSEA analysis of the ErbB2 model identified ECM remodeling, cell migration, cell growth-related genes and angio- PyMT;Ets2 db/loxP and PyMT;Fsp-Cre;Ets2 db/loxP mice. Scale bar, 200 mm. Slides were counterstained with DAPI (blue). Right: graph represents percent CD31 positive area quantified using ImageJ, Fiji (n = 3, bars represent means 6 SD, **P,0.01, Welch's t-test assuming unequal variance). C. Left: in vivo tumor vasculature visualized by intrafemoral injection of DyLight 594 lectin (green). Scale bar, 200 mm. Right: graph represents percent lectin positive area quantified using ImageJ, Fiji (n = 3, bars represent means 6 SD, ** P,0.01, Welch's t-test assuming unequal variance). D. Left: immunofluorescence staining for CD31 (red) in mammary gland tumors from 16 week old ErbB2;Ets2 db/loxP and ErbB2;Fsp-Cre;Ets2 db/loxP mice. Scale bar, 200 mm. Slides were counterstained with DAPI (blue). Right: graph represents percent CD31 positive area quantified using ImageJ, Fiji (n = 3, bars represent means 6 SD, **P,0.01, Welch's t-test assuming unequal variance). E. Heatmap depicting the genes significantly differentially regulated in endothelial cells isolated from PyMT;Ets2 db/loxP vs. PyMT;Fsp-Cre;Ets2 db/loxP mammary glands harvested at 9 weeks (n = 3, fold change.2, *P#0.05). F. GSEA plots depicting ECM remodeling, cell adhesion and cell chemotaxis to be enriched in endothelial cells isolated from PyMT;Ets2 db/loxP mammary glands as compared to PyMT;Fsp-Cre;Ets2 db/loxP mammary glands. NES: normalized enrichment score. doi:10.1371/journal.pone.0071533.g004 Ets2 in Tumor Fibroblasts Promotes Breast Cancer PLOS ONE | www.plosone.org genesis to be significantly differentially regulated in tumor associated fibroblasts compared to Ets2 null fibroblasts (Fig. 3D). To determine whether the same genes were driving the enrichment of these biological processes in the two different breast cancer models, the number of overlapping genes present in the leading edge subset for each gene ontology (GO) category was determined (Fig. 3E). Interestingly, between 15-20% of genes present in the leading edge subset of GO categories in the PyMT samples were also present in the leading edge in the ErbB2 samples. This indicates that although Ets2 may have unique targets that are specific to a particular oncogene, a significant portion of targets are regulated similarly by Ets2 independent of the oncogene. Manual annotation of differentially expressed genes in fibroblasts from both PyMT and ErbB2 models confirmed that genes known to be involved in angiogenesis and ECM remodeling were targets of Ets2 (Fig. S2B and S2D).

Ets2 in Fibroblasts Promotes Angiogenesis in a Cell Autonomous Manner
To verify whether Ets2 deletion in fibroblasts compromised angiogenesis as predicted by the microarray results, tumor vasculature was quantified by intracardiac or femoral vein injection of tomato lectin [33]. This analysis revealed a marked decrease in the vascular network in Ets2-deleted versus nondeleted control tumors in both the PyMT and ErbB2 models (P,0.01, Fig. 4A, C). Additionally, staining using the endothelial cell specific marker CD31 revealed a decrease in the recruitment of new vasculature in PyMT or ErbB2 tumors lacking Ets2 in fibroblasts compared to controls (P,0.01, Fig. 4B, D, see S3A and S3B for additional representative images).
To better understand how Ets2 function in tumor associated fibroblasts promotes angiogenesis, we isolated endothelial cells from the mammary glands of 9-week old PyMT;Ets2 db/loxP and PyMT;Fsp-Cre;Ets2 db/loxP mice and performed gene expression profiling on isolated endothelial cell RNA (Affymetrix Mouse Genome 430 2.0 GeneChip platform). Our comparison of endothelial cells from PyMT;Ets2 db/loxP and PyMT;Fsp-Cre;Ets2 db/loxP mammary glands (n = 3 biological replicates) identified 33 differentially regulated genes (fold change.2, P,0.05, Fig. 4E, Table S5). Using GSEA we observed differential regulation of genes involved in ECM remodeling, cell adhesion and cell chemotaxis in tumor endothelial cells when Ets2 was deleted in fibroblasts (Fig. 4F). Although further analysis is required to determine the exact mechanism by which Ets2 signaling in fibroblasts affects endothelial cells, our results highlight crosstalk among these different cells in the tumor microenvironment.
To experimentally establish whether these effects on angiogenesis were dependent on the cell autonomous function of Ets2 in fibroblasts, we isolated primary PyMT tumor fibroblasts with or without Ets2 and used these cells for an in vivo matrigel plug angiogenesis assay (Fig. 5A) [34,35]. Matrigel containing PyM- T;Ets2 db/loxP and PyMT;Fsp-Cre;Ets2 db/loxP tumor fibroblasts was injected subcutaneously into opposite flanks of syngeneic mice, and plugs were harvested 5 days later for analysis by CD31 staining. These experiments revealed that fibroblasts containing Ets2 promoted recruitment of endothelial cells more efficiently than fibroblasts lacking Ets2 (P,0.01, Fig. 5B). The results directly implicate Ets2 as a driving factor in tumor fibroblasts in de novo blood vessel formation. Mmp9, a gene directly regulated by Ets2 [15], is known to play a critical role in angiogenesis, including in the release of active VEGF-A in the ECM [36]. To determine if MMP9 activity was also dependent on Ets2 function in fibroblasts irrespective of tumor cells, matrigel plugs were analyzed by in situ Figure 6. Ets2 gene signatures are represented in human breast tumor stroma. A. Heat map displaying the expression of the human orthologs of the PyMT derived 38-gene Ets2 signature in normal-and tumor-stroma from human breast cancer patients. 16 genes were upregulated (denoted by red bar on the y-axis) and 22 genes were downregulated (denoted by the blue bar on the y-axis) in Ets2 null tumor fibroblasts. Red and blue regions inside the heat map indicate relative gene expression levels (red, increased and blue, decreased) between the normal and tumor stroma (P = 0.00085, one-sided Wilcoxon rank sum test, based on 10,000 permutations). B. Heat map displaying the expression of the human orthologs of the ErbB2 derived 36-gene Ets2 signature in normal-and tumor-stroma from human breast cancer patients. 3 genes were upregulated (denoted by red bar on the y-axis) and 33 genes were downregulated (denoted by the blue bar on the y-axis) in Ets2 null tumor fibroblasts. Red and blue regions inside the heat map indicate relative gene expression levels (red, increased and blue, decreased) between the normal and tumor stroma (P = 0.00004, one-sided Wilcoxon rank sum test, based on 10,000 permutations). C. Expression of the 29 PyMT-driven Ets2-dependent genes present in the NKI whole tumor data set correlates with patient outcome. Kaplan-Meier curves of high risk and low risk groups based on expression of the 29 Ets2 tumor specific genes (**P,0.0001). D. Expression of the 30 ErbB2-driven Ets2-dependent genes present in the NKI whole tumor data set correlates with patient outcome. Kaplan-Meier curves of high risk and low risk groups based on expression of the 30 Ets2 tumor specific genes (**P,0.0001). E. Expression of the 7 endothelial cell genes dependent on Ets2 fibroblast signaling present in the NKI whole tumor data set correlates with patient outcome. Kaplan-Meier curves of high risk and low risk groups based on expression of the 7 endothelial cell genes (**P,0.0001). doi:10.1371/journal.pone.0071533.g006 Ets2 in Tumor Fibroblasts Promotes Breast Cancer PLOS ONE | www.plosone.org zymography, revealing a 2-fold decrease in MMP9 activity when Ets2 was deleted in tumor fibroblasts (P,0.01, Fig. 5C). The results of these in vivo experiments supports the cell autonomous role of fibroblast Ets2 in ECM remodeling and angiogenesis.

Ets2 Fibroblast Gene Signatures are Represented in Human Breast Cancer Stroma and Correlate with Worsened Patient Outcomes
To evaluate the relevance of our findings to human breast cancer, we compared the PyMT driven Ets2-dependent fibroblast transcriptome to gene expression data derived from laser-captured human breast tumor stroma and normal stroma [6]. The expression of the human orthologs of 38 genes regulated by Ets2 in the PyMT model (see Materials and Methods) could significantly distinguish human breast tumor stroma from normal stroma compared to 10,000 random gene lists (P = 0.00085, Fig. 6A). Similar results were observed when Principal Component Analysis (PCA) was used (Fig. S4A). Importantly, the Ets2 gene signature enriched in human tumor stroma correlated with poor patient outcome in two independent and well-annotated whole tumor breast cancer gene expression datasets, the NKI and Stockholm datasets [37,38], (Fig. 6C and Fig. S4C); the 38-gene Ets2 expression signature also correlated with patient outcome in the McGill dataset (Fig. S4C).
The ErbB2 derived Ets2 tumor fibroblast gene signature was also compared to human breast cancer stroma (see Materials and Methods). The 36 gene signature was found to be significantly enriched in human breast tumor stroma with only 2 patients miscategorizing (P = 0.00004, Fig. 6B). PCA confirmed the result (Fig.  S4B). The expression of these 36 genes was significantly correlated with patient outcomes in the NKI dataset, as well as in the Stockholm and McGill datasets ( Fig. 6D and Fig. S4D, respectively).
In contrast to the results we obtained using our fibroblast gene signatures, the endothelial cell gene signature was not able to distinguish human breast tumor stroma from normal stroma (Fig.  S5A, S5B). However, 9 variably expressed genes (variance cutoff.0.5, see Materials and Methods) used for the analysis were able to significantly predict patient outcome in the NKI data set, as well as in the Stockholm and McGill datasets ( Fig. 6E and Fig.  S5C, respectively).

Conclusions
The tumor fibroblast is critical to breast tumor progression and metastasis, and may provide novel therapeutic opportunities for cancer treatment. In this context, the results presented here provide three significant and inter-related conceptual advances for the breast tumor microenvironment field.
First, the genetic studies demonstrate that Ets2 regulates an oncogenic gene expression program in tumor stromal fibroblasts that promotes tumor growth. This role of Ets2 is specific to the stroma since Ets2 is dispensable for tumor progression in malignant epithelial cells. Although the exact genes regulated by Ets2 in mammary fibroblasts vary depending on the oncogenic signals coming from the tumor cell, there is a significant overlap of target genes and the overall outcome of increased tumor growth and tumor angiogenesis remain consistent with both oncogenes tested. The activation of Ets2 stromal fibroblast-specific expression programs represents one mechanism by which the tumor co-opts the microenvironment to serve in its progression to malignancy.
The lack of Ets2 function in the tumor cells shown here is consistent with a previous report from the Oshima lab [13]. Careful examination of this earlier study indicated that the activity of MMTV-Cre utilized was highly mosaic and thus inefficient recombination of the DNA binding region of this Ets2 allele could account for the apparent lack of Ets2 function in epithelial cells. In the current work, we used a different Ets2 conditional allele targeting exons 3-5 configured so that only one floxed allele needs to undergo recombination mediated by MMTV-Cre. Our study, ensuring more efficient recombination of the conditional Ets2 allele, unequivocally demonstrates that Ets2 is not required in mammary tumor cells in the PyMT model employed in both studies. Functional redundancy of Ets2 with other Ets-family members and/or collaborating transcription factors could account for the lack of a significant effect on tumor growth in epithelial cells. Studies in colon cancer mouse models and more recently in human prostate cancer samples provide evidence that Ets2 can also function as a tumor suppressor gene [39,40,41]. In either scenario, cell context is critical in determining whether Ets2 promotes tumor progression.
A second significant finding presented here is that the consequence of inactivating this Ets2-driven stromal program is tumor specific, sparing normal development of the mammary gland. Comparisons with human stromal gene expression demonstrate the mouse Ets2 pathway functions in human breast cancer progression and correlates with worsened patient outcomes. The results of the matrigel plug assay results imply that fibroblasts acquire a tumor associated ''memory'' likely dependent on epigenetic changes in these cells, and responsible for the constitutive activation of fibroblasts. Targeting Ets2-dependent DNA and histone modifications in tumor associated fibroblasts may lead to more effective treatments in patients with breast tumors without toxicity to normal tissues [42].
Finally, the Ets2 mechanism involves the activation of a gene expression program that leads to increased angiogenesis. Targeting the tumor vasculature remains an attractive therapeutic target. However, current strategies in which single molecules like VEGF are targeted have demonstrated limited success clinically. Our results indicate that the changes in tumor angiogenesis observed in vivo are likely regulated by the cumulative action of many Ets2 target genes acting in concert. Defining and targeting the pathway hubs controlled by Ets2 provides an alternative strategy that should be more effective at altering multiple cell-cell and cellmatrix interactions required for tumor angiogenesis.
In conclusion, we have identified Ets2 as a fibroblast-specific effector of tumor growth through a mechanism that involves increased tumor angiogenesis. The results emphasize the potential utility of identifying tumor-dependent, stromal-specific targets for therapeutic intervention in human breast cancer. Figure S1 A. X-gal staining of 10 week old mammary glands from