PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Clin Cancer Res. Author manuscript; available in PMC 2010 November 18.
Published in final edited form as:
PMCID: PMC2987696
NIHMSID: NIHMS103499

The Macrophage Colony Stimulating Factor-1 Response Signature in Breast Carcinoma

Abstract

Purpose

Macrophages play an important role in breast carcinogenesis. The pathways that mediate the macrophage contribution to breast cancer and the heterogeneity that exists within macrophages are incompletely understood. Macrophage colony stimulating factor-1 (CSF1) is the primary regulator of tissue macrophages. The purpose of this study was to define a novel CSF1 response signature and to evaluate its clinical and biological significance in breast cancer.

Experimental Design

We defined the CSF1 response signature by identifying genes overexpressed in tenosynovial giant cell tumor and pigmented villonodular synovitis (tumors composed predominantly of macrophages recruited in response to the overexpression of CSF1) as compared with desmoid type fibromatosis and solitary fibrous tumor. To characterize the CSF1 response signature in breast cancer, we analyzed the expression of CSF1 response signature genes in eight published breast cancer gene expression datasets (n=982) and performed immunohistochemistry and in situ hybridization for CSF1 response genes on a breast cancer tissue microarray (n=283).

Results

In both the gene microarray and tissue microarray analyses, a consistent subset (17–25%) of breast cancers shows the CSF1 response signature. The signature is associated with higher tumor grade, decreased expression of estrogen receptor, decreased expression of progesterone receptor, and increased p53 mutations (p <0.001).

Conclusions

Our data show that the CSF1 response signature is consistently seen in a subset of breast carcinomas and correlates with biological features of the tumor. Our findings provide insight into macrophage biology and may facilitate the development of personalized therapy for patients most likely to benefit from CSF1-targeted treatments.

Keywords: tumor microenvironment, stromal signature, macrophage, breast cancer, CSF1

Statement of Clinical Relevance

Breast cancer is the most common cancer among American women. Although it is known that the tumor microenvironment plays an important role in the pathogenesis of breast cancer, there are currently few available therapeutic agents to target the breast cancer microenvironment. Clinical and experimental studies suggest that macrophages play an important role in breast carcinogenesis. Macrophage colony stimulating factor-1 (CSF1) is the primary regulator of tissue macrophages. In the current study, we have defined a novel CSF1 response signature. Using multiple breast cancer gene expression datasets (n=982) and a tissue microarray (n=283), we show that a reproducible subset of breast cancers (17–25%) are enriched with the CSF1 response signature. Breast cancers with the CSF1 response signature are significantly more likely to be higher grade, estrogen receptor negative, progesterone receptor negative, and contain p53 mutations (all p<0.001). Therapies targeted at CSF1 and other mediators of macrophage behavior are currently being developed. We believe that the CSF1 response signature described in this paper will facilitate the development of drugs to target previously unrecognized mediators of the CSF1 response in breast cancer and may provide a novel technique for identifying patients most likely to respond to CSF1- and macrophage-targeted therapies.

Introduction

Clinical and experimental studies suggest that macrophages play an important role in breast carcinogenesis (13). Macrophage colony stimulating factor-1 (CSF1) is the primary regulator of tissue macrophages. Studies on breast cancer have demonstrated that the protein expression of CSF1 and CSF1 receptor (CSF1R) correlate with increased inflammation and poorer prognosis (4, 5). CSF1 has been shown to promote progression to malignancy in mouse mammary tumors (6) and blockade of CSF1 suppressed growth of human mammary tumor xenografts (7). Previous studies in breast cancer have focused primarily on expression of CSF1 or CSF1R. Here we show that an expanded list of CSF1 responsive genes can be used to identify subsets of breast carcinomas that may be uniquely sensitive to CSF1- and macrophage-targeted therapies.

Previously we reported that the related soft tissue tumors tenosynovial giant cell tumor (TGCT) and pigmented villonodular synovitis (PVNS) are made up of a heterogeneous population of cells, in which a small subset of tumor cells contain a translocation involving CSF1 resulting in the recruitment of CSF1R expressing macrophages that constitute the majority of the tumor mass (8). We hypothesize that the distinct gene expression pattern of TGCT and PVNS can be seen as a surrogate for the macrophage response to CSF1.

In the current study, we first define a CSF1 response gene expression signature by identifying a set of genes specifically and highly expressed in TGCT and PVNS. We then evaluate the expression of these CSF1 response genes in multiple breast cancer gene expression datasets and a breast cancer TMA and find that a consistent subset of breast cancers expresses the CSF1 response signature. The characterization of this new pathway in breast cancer provides insight into the regulation of the breast cancer tumor microenvironment by CSF1. This finding may facilitate the development of personalized therapy for patients most likely to benefit from CSF1- and macrophage-targeted treatments.

Methods

Determination of the CSF1 Response Gene Signature

We have previously shown that in TGCT/PVNS, a translocation occurring in a small subset of tumor cells results in the recruitment of a macrophage rich inflammatory infiltrate that composes the majority of the tumor mass (8). Therefore, we hypothesized that the expression profile of TGCT/PVNS primarily reflects the biology of the non-neoplastic macrophage-rich inflammatory infiltrate responding to secreted CSF1 by the minority of tumor cells with the CSF1 translocation. To define the CSF1 response signature, significance analysis of microarrays was performed to identify genes that show significantly increased expression in PVNS (n=8) and TGCT (n=7) as compared with desmoid type fibromatosis (n=7) and solitary fibrous tumor (n=6), with a minimum fold change of 2.5 and false discovery rate of 0.02%, based on gene expression profiling previously performed (8). This analysis resulted in the identification of 603 genes that constitute the CSF1 response signature (Supplemental Workbook).

Breast Cancer Datasets

We utilized five publically available whole tumor breast cancer data sets (NKI (9), Perreard (10), GSE1379 (11), GSE1456 (12), GSE3494 (13)) that contain gene expression data on a total of 856 cases with clinical follow-up, and we utilized three laser capture microdissection (LCM) breast cancer datasets (GSE5847 (14), GSE9014(15), GSE10797(16)), that contain gene expression data obtained separately from the stroma and epithelium (GSE5847 (14) and GSE10797 (16)) or solely from the stroma (GSE9014 (15)) from 126 cases. Additional information on the datasets is provided in the Supplemental Methods.

Gene Expression Data Analysis

For all data sets, the expression data were downloaded and imported into the dChip 2006 software (http://biosun1.harvard.edu/complab/dchip/). Expression values were standardized gene-wise, by subtracting the mean and dividing by the standard deviation of the expression values for each gene. Unsupervised hierarchical clustering was performed in each dataset with the Cluster 3.0 software (http://bonsai.ims.u-tokyo.ac.jp/~mdehoon/software/cluster/software.htm#ctv), using the uncentered Pearson correlation as the distance metric and average linkage clustering. The resulting heatmap and dendrogram were visualized on Java Treeview (http://jtreeview.sourceforge.net/).

Determination of CSF1 Response Core Gene and Case Clusters

We defined the CSF1 response core gene cluster as the largest cluster of genes in each whole tumor dataset that showed greater than 25% correlated expression. We defined the CSF1 response case cluster as the largest cluster of cases that showed high levels of expression of the CSF1 response core gene cluster and exhibited greater than 10% correlated expression. After applying these rules, we defined the CSF1 response core gene set as being composed of genes that are present on the microarray platform in at least 2 whole tumor data sets and present in the CSF1 response core gene cluster in either all whole tumor data sets (if gene was present on platform in only 2–3 of data sets) or absent in a single data set (if gene was present on platform in 4–5 data sets). Using these criteria, the original set of 603 CSF1 response genes indentified through gene expression profiling on TGCT and PVNS was reduced to a “CSF1 response core gene set” of 112 genes (Supplemental Workbook) that were consistently and coordinately expressed in the 5 whole tumor breast cancer datasets.

Analysis of Clinicopathological Variables

In the whole tumor data sets examined, the measured survival outcomes include: disease-free survival (Perreard, GSE1379, GSE1456, combined n=294), disease-specific survival (GSE1456, GSE3494, combined n=395), and overall survival (NKI, Perreard, GSE1456, combined n=529). The Kaplan-Meier estimate was used to compute survival curves, and the log rank p value was computed to assess statistical significance. For association tests, the Pearson Chi Square test was used. To compare ordinal or ratio variables in two independent groups, either the Mann-Whitney U test or Student T test were performed. Statistical computing was performed using SPSS 15.0 for Windows.

Evaluation of CSF1 Response Protein Localization in the Breast Cancer Microenvironment

To determine the patterns of CSF1-response protein expression in the breast cancer microenvironment, we performed immunohistochemistry and in situ hybridization on a breast cancer TMA (TA221), which contains samples from 283 breast carcinomas obtained from Stanford University Medical Center. We measured CSF1 and CSF1R RNA expression by in situ hybridization and the expression of four additional CSF1 response proteins (FCGR3a, FCGR2a, CTSL1, CD163) by immunohistochemistry. To select CSF1 response markers for evaluation by in situ hybridization or immunohistochemistry, we identified the 112 genes that were consistently present in the CSF1 response core cluster (the “CSF1 response core gene set”; Supplemental Workbook). Out of this list of 112 candidates, we identified FCGR3a, FCGR2a, CTSL1, and CD163 as five markers for which commercial available antibodies that performed well on formalin fixed paraffin embedded tissue were available. The primary antibodies used were FCGR3a (CD16) (MCA1816, mouse monoclonal, AbD Serotec, CA), CTSL1 (MCA2374, mouse monoclonal, AbD Serotec, CA), FCGR2a (CD32) (AB45143, rabbit monoclonal, Abcam, UK), CD163 (NCL-CD163, mouse monoclonal, Novocastra, CA). The immunohistochemical reactions were visualized using mouse and rabbit versions of the EnVision + system (DAKO, Carpinteria, CA) using diaminobenzidine. CD163 staining was performed with the Ventana Benchmark Autostainer. In situ hybridization of TMA sections for CSF1 and CSF1R were performed based on a protocol published previously (8, 1719). The immunohistochemical and in situ hybridization studies were interpreted by histopathological evaluation by a surgical pathologist (IE). A case was determined to show the CSF1 response signature if it showed coordinate expression (score ≥ 1) of at least 4 of the 5 markers. The digital images, collected using computerized microscopes (BLISS, Bacuslabs; and Ariol, Applied Imaging Inc), are available for all stained cores through the accompanying website (http://tma.stanford.edu/tma_portal/CSF1_breast). Additional detailed information on the antibodies used, staining procedure, and scoring technique can be found in the Supplemental Methods. Institutional review board approval was obtained for these studies.

Functional Gene Set Analysis

To determine the functional significance of gene sets, we utilized the DAVID: Database for Annotation, Visualization, and Integrated Discovery (20). To generate protein-protein interaction (PPI) networks, gene sets were uploaded into STRING 7.0 (21), and the following active prediction methods were employed (neighborhood, co-expression, gene fusion, experiments, co-occurrence, database, and text mining) with a medium confidence score (0.400). The Cytoscape software platform (22) was used to visualize the PPI networks, and the Cytoscape plug-in Network Analyzer 2.52 was used to evaluate the topological characteristics of the networks (23).

Results

A Reproducible Core Subset of CSF1 Response Genes Show Coordinate Expression in Breast Carcinoma

We performed two-way unsupervised hierarchical clustering on the five whole tumor breast cancer gene expression datasets with the 603 CSF1 response genes. In each dataset, we observed a core cluster of CSF1 response genes that showed high levels of coordinated expression among a cluster of breast cancer cases. We defined the CSF1 response core gene set as being composed of the genes which were present in the CSF1 response core gene cluster most consistently across the five whole tumor datasets (see Methods). The resulting CSF1 response core gene set consists of 112 CSF1 response genes (Supplemental Workbook).

Functional gene set analysis demonstrates that the core CSF1 response gene set is most highly associated with annotation terms relating to cellular defense and immune response (response to biotic stimulus, defense response, immune response, response to stimulus; all Fischer exact p < 1.5E-15 computed using the DAVID: Database for Annotation, Visualization, and Integrated Discovery (20); See Supplemental Workbook for full annotation results). The core CSF1 response genes associated with these annotation terms include CSF1R, FCGR3a, FCGR2a, and CD163. The core gene set contains several members of the cathepsin class of lysosomal cystein proteases (CTSL1, CTSS, CTSC), which are known to be expressed by macrophages and to be important for regulating antigen presentation (24).

We performed protein-protein interaction (PPI) network analysis on both the core CSF1 response genes and the non-core CSF1 response genes. This analysis shows that the PPI network created by the core CSF1 response genes has a higher average clustering coefficient, average number of connections to other proteins, average neighborhood connectivity, and average closeness centrality (all p ≤ 0.002) as compared with the network created with the non-core CSF1 response genes (Supplemental Fig. 1 and Supplemental Table 1). These findings suggest that the PPI network created by the CSF1 response core gene set is more centralized and tightly connected than the PPI network created by the CSF1 response non-core proteins.

These functional gene set and PPI network analyses suggest that by filtering the 603 CSF1 response gene set to the 112 core CSF1 response genes, we have selected for the core genes that are likely to operate in a common CSF1 induced immune response module in the breast carcinoma microenvironment.

A Reproducible Subset of Breast Carcinomas Demonstrate the CSF1 Response Signature

In each of the five whole tumor breast carcinoma datasets, a subset of similar size (17–25%) of breast carcinomas demonstrates the CSF1 response gene signature (Fig. 1). Cases of breast cancer with the CSF1 response signature were more likely to be ER negative, PR negative, higher grade, and larger (Supplemental Table 2).

Figure 1
Unsupervised Hierarchical Clustering of Breast Carcinomas with CSF1 Response Genes in Five Datasets. A, Perreard (n=91). B, GSE1379 (n=60). C, GSE1456 (n=159). D, GSE3494 (n=251), E, NKI (n=295). Within the heatmap, yellow represents high expression, ...

The GSE3494 data set contains information pertaining to the p53 mutation status of 251 cases of breast carcinoma. From this data set, Miller et al. defined a p53 expression signature and used diagonal linear discriminant analysis to classify breast cancer cases according to the signature (13). In our analysis of this dataset, we find that breast cancers with the CSF1 response signature are significantly more likely to harbor a p53 mutation and to be enriched with the p53 mutation gene expression signature (Supplemental Table 2).

The NKI and GSE1456 datasets have previously been stratified by others into molecular subcategories based on initial gene expression studies by Perou and Sorlie and colleagues (basal, ERBB2+, luminal A, luminal B, normal-like) (25, 26). Breast cancers with the CSF1 response signature were significantly more likely to be basal or ERBB2 and less likely to be normal-like, with no significant association with the luminal A and luminal B molecular subtypes (Supplemental Table 2).

The CSF1 Response Signature Shows a Variable Association with Survival

We evaluated the relationship of the CSF1 signature with patient survival in each of the five whole tumor datasets. In the NKI dataset, which is limited to patients younger than 53 years old with stage I or II disease, the CSF1 response signature showed an association with decreased overall survival (10 year survival = 64% in tumors with CSF1 response signature vs. 73%; Log rank p = 0.044; Fig. 2A). In all other datasets, we were unable to identify a statistically significant association with survival (all p > 0.15, Fig. 2A). When survival data is pooled from the five datasets, we find a trend for an association of the CSF1 response signature with decreased survival but are unable to identify a statistically significant association (p > 0.099, Fig. 2B, Supplemental Table 2).

Figure 2
Kaplan Meier Survival Curves for Breast Cancer Cases Stratified by CSF1 Response Signature. A, The survival curves are displayed individually for each of the five datasets. B, The survival curves display data pooled from the five datasets for disease ...

The lack of a consistent association of the CSF1 response signature with poor survival was unanticipated, given the fact that the signature is highly and consistently correlated with features known to predict poor outcome (higher grade, ER negativity, PR negativity, p53 mutations, basal and ERBB2 molecular subtypes). To further evaluate the complex relationship of the CSF1 response signature with survival, we performed several subset analyses. When the survival analysis was limited to ER negative tumors, the CSF1 response signature showed a trend for an association with improved overall survival (10 year survival of 54% in tumors with CSF1 response signature vs. 36%; Log rank p = 0.103, Fig. 3A) and improved disease specific survival (10 year survival 90% in tumors with CSF1 response signature vs. 65%; Log rank p = 0.099). Among cases that are grade 1 or 2, the CSF1 response signature is associated with decreased survival (10 year survival 75% in tumors with CSF1 response signature vs. 84%; Log rank p = 0.006, Fig. 3B). Among cases enriched with the p53 mutation signature, the CSF1 response signature showed an association with improved disease specific survival (10 year survival 72% in tumors with CSF1 response signature vs. 49%; Log rank p =0.047, Fig. 3C). These subset analyses suggest that the CSF1 response signature demonstrates a complex relationship with survival, in which the signature is associated with poor prognosis among low grade tumors and shows a trend for an association with improved prognosis among ER-negative tumors and among tumors with a p53 mutation gene expression signature. When the data is not sub-stratified by these factors, the CSF1 response signature does not show a statistically significant relationship with prognosis.

Figure 3
Kaplan Meier Survival Curves for CSF1 Response Signature Subset Analyses. Kaplan Meier survival curves are displayed for breast cancer cases stratified by CSF1 response signature for subsets determined by: A, estrogen receptor status; B, grade; and C, ...

The CSF1 Response Signature is Present in LCM Breast Cancer Datasets

To further assess the pattern of CSF1 response gene expression in the epithelium and stroma of breast carcinomas, we evaluated the expression of CSF1 response genes in three publically available LCM breast carcinoma datasets. Two of the datasets (GSE5847 and GSE9014) contain samples from both breast cancer epithelium and stroma. Unsupervised hierarchical clustering of both of these datasets shows that a subset of cases (18/74, 24%) demonstrates the CSF1 response signature (Supplemental Fig. 2A–B). Most cases (13/18) with the CSF1 response signature express the signature coordinately in both the epithelium and stroma, while 4/18 cases express the signature in only the epithelial sample and 1/18 in only the stromal sample. It should be noted that inflammatory cells may infiltrate the breast cancer epithelium, and consequently their gene expression may be included in either the “epithelial” or the “stromal” compartment of LCM datasets. The third LCM dataset (GSE10797) contains gene expression data from stromal samples of 52 breast cancers. Unsupervised hierarchical clustering performed on this dataset shows a cluster of 13/52 (25%) stromal samples with the CSF1 response signature (Supplemental Fig. 2C). The GSE10797 dataset contains information on tumor grade and recurrence free survival for all samples. Analysis of these data show that the CSF1 response signature identified in stromal samples is associated with higher tumor grade (10/13 CSF1 response stromal samples from grade 3 carcinomas vs. 16/39; p=0.025), which agrees with the association seen in the whole tumor gene expression profiling.

The Expression of CSF1 is Correlated with Stromal Expression of CSF1 Response Proteins in Breast Carcinoma

To confirm the existence of a CSF1 response signature in breast cancer by a different technique and to determine the cellular localization and patterns of expression of CSF1 and CSF1 response proteins in breast carcinoma, we examined the expression of CSF1 and 5 CSF1 response markers by in situ hybridization and immunohistochemistry on a breast cancer TMA containing samples from a total of 283 patients.

Of the 206 cases with evaluable data for CSF1 and at least 4 of the 5 CSF1 response proteins, CSF1 was expressed in the stroma in 72 cases (35%) and in the malignant epithelium in 107 cases (52%). The expression of CSF1 in either the epithelium or stroma was associated with stromal expression of CSF1 response proteins: 41/107 (38%) of cases with epithelial CSF1 expression showed coordinate stromal expression of at least 4 CSF1 response proteins vs. 17/99 (17%) for cases with no epithelial CSF1 (p = 0.001); and 37/72 (51%) of cases with stromal CSF1 expression showed coordinate stromal expression of at least 4 CSF1 response proteins vs. 21/134 (16%) of cases with no stromal CSF1 expression (p = 5.49e-008). These data demonstrate that either the epithelial or the stromal expression of CSF1 is significantly correlated with the stromal expression of CSF1 response genes and further suggest that the expression of the CSF1 gene plays a major role in coordinating the CSF1 response signature.

Unsupervised hierarchical clustering of stromal and epithelial CSF1 with the CSF1 response proteins demonstrates that CSF1R and stromal CSF1 expression show the strongest pairwise correlation (Fig. 4). Of the CSF1 response proteins, CD163 (macrophage-associated antigen) stains the highest percentage of cases (149/226, 66%) and shows the least coordinated expression with the other markers (Fig. 4). CD163 is a member of the scavenger receptor cysteine-rich superfamily and is expressed on most subpopulations of mature tissue macrophages (27). These findings suggest that CD163 labels the majority of macrophages, and the CSF1 response signature captures a distinct and novel subset of CD163 expressing macrophages.

Figure 4
Coordinate Expression of CSF1 and CSF1 Response Proteins in Breast Cancer. A, Unsupervised hierarchical clustering of 272 breast carcinomas based on tissue microarray staining for CSF1 and five CSF1 response markers (CSF1R, CD163, FCGR3a, FCGR2a, CTSL1). ...

The Coordinate Expression of CSF1 Response Proteins in the Breast Cancer Microenvironment Correlates with the Clinicopathologic Features Observed in the Gene Expression Profiling Analysis

There were at least four of five evaluable CSF1 response markers for 252 cases on the breast cancer TMA. Of these 252 cases, 64 (25%) showed coordinate expression of at least 4 of the CSF1 response proteins. These cases were significantly more likely to be ER negative, PR negative, higher grade, EGFR positive, ductal differentiation, and show significantly higher numbers of proliferating malignant epithelial cells as measured by Ki67% (Supplemental Table 3). The expression of CSF1 response proteins showed no association with the presence of lymph node metastasis at time of diagnosis, tumor size, or expression of Her2 (all p>0.48) (Supplemental Table 3). The proportion of cases coordinately expressing CSF1 response proteins in the TMA data (25%) is similar to the proportion of cases with the CSF1 response signature seen in the whole tumor gene expression data (17–25%) and the LCM gene expression data (25%). The clinicopathological associations seen in the TMA data correlate with the findings of the gene expression data.

Discussion

Clinical and experimental studies have demonstrated that tumor associated macrophages play an important role in breast carcinogenesis (13, 6, 7). Macrophages represent a heterogeneous cell type, but the clinical and biological significance of this heterogeneity is incompletely understood (28). CSF1 is a growth factor that acts through the cell surface receptor CSF1R. Activation of CSF1R by CSF1 stimulates the proliferation, differentiation, and survival of macrophages and influences macrophage chemotaxis, phagocytosis, and synthesis and secretion of proteolytic enzymes and cytokines (3). Lin and colleagues have used a mouse breast cancer model to demonstrate that CSF1 promotes malignant progression in mammary tumors (6). The specific cellular pathways that mediate CSF1 behavior in breast cancer are largely undefined.

In the current study, we define a CSF1 response signature obtained by gene expression profiling of TGCT and PVNS, two related soft tissue tumors composed predominantly of macrophages that express CSF1R and are recruited in response to the expression of CSF1 by neoplastic tumor cells (8). The profiling of these tumors captures a CSF1 response, since the CSF1R expressing macrophages far outnumber the neoplastic cells, which typically compose ~10% of the tumor cells; thus the RNA measured in the profiling experiments is predominately derived from macrophages responding to CSF1. We hypothesized that this CSF1 response occurs in a subset of breast carcinomas, in which carcinoma cells or stromal cells secrete CSF1 resulting in the recruitment of a macrophage rich inflammatory infiltrate, and we posit that evidence of this process will be observable in gene expression profiling datasets.

To evaluate this hypothesis, we first examined the expression of CSF1 response genes in five breast cancer datasets. We defined the CSF1 response gene set by identifying the 603 genes that showed the highest levels of expression in TGCT/PVNS as compared to desmoid type fibromatosis and solitary fibrous tumor. We then evaluated the expression of these genes in five breast carcinoma datasets and filtered the original list of 603 genes down to the 112 core CSF1 response genes that showed the highest level of coordinated expression across the five breast cancer datasets.

This core group of CSF1 response genes is highly enriched for annotation terms relating to immune defense (Supplemental Workbook). The core gene set includes multiple genes known to be expressed by macrophages and involved in macrophage function, including CSF1R, FCGRI, FCGR3a, FCGR2a, CD163, and CCL5, which is a chemoattractant for macrophages and has recently been shown to be secreted by mesenchymal stem cells to enhance the motility, invasion, and metastasis of breast cancer cells (29). The core gene set includes several cathepsins (CTSL1, CTSS, CTSC), which are lysosomal proteases expressed by macrophages and important for antigen presentation (24). The macrophage lineage encompasses a heterogeneous group of macrophage sub-populations, and the biological and functional significance of these sub-populations is only beginning to be elucidated (28). A study by Grage-Griebenow and colleagues identified a novel dendritic cell-like subtype of monocyte that expresses FCGRI (CD64) and FCGR3a (CD16) (30), and is characterized by high accessory capacity for activated lymphocytes and high expression of HLA-DR, and CD86. All four of these markers (FCGRI, FCGR3a, HLA-DR, CD86) were identified in our study as members of the core CSF1 response gene set. In addition to these selected genes, we have identified numerous other core CSF1 response genes whose behavior in macrophages or in relation to CSF1 have not previously been characterized (Supplemental Workbook).

In both our gene microarray and TMA analyses, we find that the CSF1 response signature is associated with a characteristic clinicopathological phenotype in breast cancer (higher histologic grade, increased frequency of p53 mutations, decreased expression of ER and PR, increased expression of EGFR, and increased Ki67 proliferation index). The association of tumor expression of EGFR with expression of CSF1 response proteins in the tumor microenvironment is particularly interesting, as the expression of CSF1 by breast cancer cells has been shown to promote the expression of EGF by macrophages, which in turn promotes the expression of CSF1 by breast carcinoma cells leading to the adoption of a more invasive phenotype in a positive feedback loop (31, 32). It is known that EGFR is expressed by basal subtype breast cancers (33), and in the current study we find an association of basal molecular subtype with the CSF1 response signature (Table 2).

Despite the fact that breast cancers enriched with the CSF1 response signature harbor several molecular (p53 mutations, basal and ERBB2 molecular subtypes) and histopathological (larger tumor size and higher histologic grade) features conferring poor prognosis, we find an association with poor prognosis in only one of the five breast cancer datasets and no significant association is identified when pooling data from all five datasets (Fig. 2, Supplemental Table 2).

To better define the relationship of the CSF1 response signature with prognosis, we performed several subset analyses. Interestingly, we find that among grade 1 and 2 tumors, the CSF1 response signature is associated with decreased survival (Fig 3B). In contrast, among tumors harboring a p53 mutation signature the CSF1 response signature shows an association with improved prognosis (Fig. 3C). This finding is compatible with work by Lee and colleagues demonstrating that CSF1 activates p53-independent pathways to induce growth arrest of human breast cancer cells (34). This finding suggests that in cases with a p53 mutation, the CSF1 response pathway could potentially stimulate p53-independent growth arrest. In an additional subset analysis, we find that among ER negative cases there is a trend for increased survival in cases enriched with the CSF1 response signature (Fig. 3A). It has been shown that tumor associated macrophages may promote malignant progression by secreting estrogens (35, 36), and therefore this mechanism of tumor pathogenesis might be prevented in ER negative breast cancers. It has recently been shown by Teschendorff and colleagues that increased expression of immune response genes correlates with improved prognosis in ER negative breast cancer (37). These findings suggest that several immune response pathways (including the CSF1 response pathway) may play a protective role in ER negative breast cancers.

These data show that the influence of CSF1 and the CSF1 response in breast cancer depends not only on the behavior of stromal and inflammatory cells but also on the particular genotypic and phenotypic characteristics of the carcinoma cells. A recent study by Tamimi and colleagues underscores the complex relationship between CSF1 expression and host characteristics and shows that increased serum levels of CSF1 is associated with decreased risk of breast cancer in premenopausal women and increased risk of breast cancer in postmenopausal women (38).

In the current study, we have defined a novel CSF1 response signature seen in breast carcinoma. We believe that our findings will not only afford a more comprehensive understanding of the mediators and pathways involved in the macrophage CSF1 response in breast cancer, but will also provide a valuable resource for the development of additional therapeutic agents to target heretofore unrecognized mediators in the CSF1 response in breast cancer. Furthermore, the measurement of genes from the CSF1 response core gene set in clinical samples from breast cancer patients may provide a novel technique for identifying patients most likely to respond to CSF1- and macrophage-targeted therapies. It is well documented that unrecognized molecular heterogeneity within a clinical cancer trial may lead to underestimation of therapeutic benefits if potential responders and non-responders are not identified prior to treatment (39). Potent therapies directed against gene targets, such as ER and HER2 in breast cancer, only work well in tumors that highly express those gene products. Recent studies have demonstrated the ability to use genomic signatures to guide the use of chemotherapeutics (40, 41), and the technique has been validated in clinical trials in ovarian (42) and breast cancer (43). Therapies targeted at CSF1 and other mediators of macrophage behavior are currently being developed (7, 44, 45). We believe that the CSF1 response signature could potentially serve as a resource for drug development and as a clinically useful tool for identifying patients most likely to respond to CSF1- and macrophage-targeted therapies.

Supplementary Material

Supp Methods

Supplemental Figure 1. A, CSF1 Response Signature Core Protein-Protein Interaction Network. B, CSF1 Response Signature Non-Core Protein-Protein Interaction Network. The figures depict the PPI networks created from the CSF1 response signature core proteins (A) and the CSF1 response signature non-core proteins (B). The PPI networks are created and visualized using the STRING database. The nodes represent proteins and the links between nodes represent predicted protein-protein interactions. The thickness of the links corresponds to the confidence of the predicted interaction. The network created using the 80 core proteins (A) is a more centralized and tightly connected network than the network created using the 285 non-core proteins (B) (See Supplemental Table 1).

Supplemental Figure 2. Unsupervised Hierarchical Clustering of Breast Carcinomas with CSF1 Response Genes in Three Laser Capture Microdissection Breast Cancer Datasets. A, GSE5847 (n=46 cases with 1 stromal and 1 epithelial sample per case). B, GSE9014 (n=28 cases with 1 stromal and 1 epithelial sample per case). C, GSE10797 (n=52 cases, with 1 stromal sample per case). Within the heatmap, red represents high expression, black represents median expression, and green represents low expression. The red highlighted region of the dendrograms above the heatmaps indicates the cluster of samples with the CSF1 response signature in each dataset. The colorbar on the left of each heatmap indicates whether the CSF1 response gene belongs to the core CSF1 response gene set. The core CSF1 response genes are indicated in red and the non-core CSF1 response genes are green.

Supp Table

Supp workbook

Footnotes

Conflict of Interest: There are no conflicts of interest.

References

1. Condeelis J, Pollard JW. Macrophages: obligate partners for tumor cell migration, invasion, and metastasis. Cell. 2006;124:263–6. [PubMed]
2. Pollard JW. Tumour-educated macrophages promote tumour progression and metastasis. Nat Rev Cancer. 2004;4:71–8. [PubMed]
3. Sapi E. The role of CSF-1 in normal physiology of mammary gland and breast cancer: an update. Exp Biol Med (Maywood) 2004;229:1–11. [PubMed]
4. Scholl SM, Pallud C, Beuvon F, et al. Anti-colony-stimulating factor-1 antibody staining in primary breast adenocarcinomas correlates with marked inflammatory cell infiltrates and prognosis. J Natl Cancer Inst. 1994;86:120–6. [PubMed]
5. Maher MG, Sapi E, Turner B, et al. Prognostic significance of colony-stimulating factor receptor expression in ipsilateral breast cancer recurrence. Clin Cancer Res. 1998;4:1851–6. [PubMed]
6. Lin EY, Nguyen AV, Russell RG, Pollard JW. Colony-stimulating factor 1 promotes progression of mammary tumors to malignancy. J Exp Med. 2001;193:727–40. [PMC free article] [PubMed]
7. Aharinejad S, Paulus P, Sioud M, et al. Colony-stimulating factor-1 blockade by antisense oligonucleotides and small interfering RNAs suppresses growth of human mammary tumor xenografts in mice. Cancer Res. 2004;64:5378–84. [PubMed]
8. West RB, Rubin BP, Miller MA, et al. A landscape effect in tenosynovial giant-cell tumor from activation of CSF1 expression by a translocation in a minority of tumor cells. Proc Natl Acad Sci U S A. 2006;103:690–5. [PubMed]
9. van de Vijver MJ, He YD, van’t Veer LJ, et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med. 2002;347:1999–2009. [PubMed]
10. Perreard L, Fan C, Quackenbush JF, et al. Classification and risk stratification of invasive breast carcinomas using a real-time quantitative RT-PCR assay. Breast Cancer Res. 2006;8:R23. [PMC free article] [PubMed]
11. Ma XJ, Wang Z, Ryan PD, et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell. 2004;5:607–16. [PubMed]
12. Pawitan Y, Bjohle J, Amler L, et al. Gene expression profiling spares early breast cancer patients from adjuvant therapy: derived and validated in two population-based cohorts. Breast Cancer Res. 2005;7:R953–64. [PMC free article] [PubMed]
13. Miller LD, Smeds J, George J, et al. An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival. Proc Natl Acad Sci U S A. 2005;102:13550–5. [PubMed]
14. Boersma BJ, Reimers M, Yi M, et al. A stromal gene signature associated with inflammatory breast cancer. Int J Cancer. 2008;122:1324–32. [PubMed]
15. Finak G, Bertos N, Pepin F, et al. Stromal gene expression predicts clinical outcome in breast cancer. Nat Med. 2008;14:518–27. [PubMed]
16. Casey T, Bond J, Tighe S, et al. Molecular signatures suggest a major role for stromal cells in development of invasive breast cancer. Breast Cancer Res Treat. 2008 [PubMed]
17. St Croix B, Rago C, Velculescu V, et al. Genes expressed in human tumor endothelium. Science. 2000;289:1197–202. [PubMed]
18. Iacobuzio-Donahue CA, Ryu B, Hruban RH, Kern SE. Exploring the host desmoplastic response to pancreatic carcinoma: gene expression of stromal and neoplastic cells at the site of primary invasion. The American journal of pathology. 2002;160:91–9. [PubMed]
19. West RB, Corless CL, Chen X, et al. The novel marker, DOG1, is expressed ubiquitously in gastrointestinal stromal tumors irrespective of KIT or PDGFRA mutation status. The American journal of pathology. 2004;165:107–13. [PubMed]
20. Dennis G, Jr, Sherman BT, Hosack DA, et al. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 2003;4:3. [PubMed]
21. von Mering C, Jensen LJ, Kuhn M, et al. STRING 7--recent developments in the integration and prediction of protein interactions. Nucleic Acids Res. 2007;35:D358–62. [PubMed]
22. Shannon P, Markiel A, Ozier O, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13:2498–504. [PubMed]
23. Assenov Y, Ramirez F, Schelhorn SE, Lengauer T, Albrecht M. Computing topological parameters of biological networks. Bioinformatics. 2008;24:282–4. [PubMed]
24. Honey K, Rudensky AY. Lysosomal cysteine proteases regulate antigen presentation. Nat Rev Immunol. 2003;3:472–82. [PubMed]
25. Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours. Nature. 2000;406:747–52. [PubMed]
26. Sorlie T, Perou CM, Tibshirani R, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A. 2001;98:10869–74. [PubMed]
27. Fabriek BO, Dijkstra CD, van den Berg TK. The macrophage scavenger receptor CD163. Immunobiology. 2005;210:153–60. [PubMed]
28. Gordon S, Taylor PR. Monocyte and macrophage heterogeneity. Nat Rev Immunol. 2005;5:953–64. [PubMed]
29. Karnoub AE, Dash AB, Vo AP, et al. Mesenchymal stem cells within tumour stroma promote breast cancer metastasis. Nature. 2007;449:557–63. [PubMed]
30. Grage-Griebenow E, Zawatzky R, Kahlert H, Brade L, Flad H, Ernst M. Identification of a novel dendritic cell-like subset of CD64(+)/CD16(+) blood monocytes. Eur J Immunol. 2001;31:48–56. [PubMed]
31. Goswami S, Sahai E, Wyckoff JB, et al. Macrophages promote the invasion of breast carcinoma cells via a colony-stimulating factor-1/epidermal growth factor paracrine loop. Cancer Res. 2005;65:5278–83. [PubMed]
32. O’Sullivan C, Lewis CE, Harris AL, McGee JO. Secretion of epidermal growth factor by macrophages associated with breast carcinoma. Lancet. 1993;342:148–9. [PubMed]
33. Nielsen TO, Hsu FD, Jensen K, et al. Immunohistochemical and clinical characterization of the basal-like subtype of invasive breast carcinoma. Clin Cancer Res. 2004;10:5367–74. [PubMed]
34. Lee AW, Nambirajan S, Moffat JG. CSF-1 activates MAPK-dependent and p53-independent pathways to induce growth arrest of hormone-dependent human breast cancer cells. Oncogene. 1999;18:7477–94. [PubMed]
35. Mor G, Yue W, Santen RJ, et al. Macrophages, estrogen and the microenvironment of breast cancer. J Steroid Biochem Mol Biol. 1998;67:403–11. [PubMed]
36. Mor G, Sapi E, Abrahams VM, et al. Interaction of the estrogen receptors with the Fas ligand promoter in human monocytes. J Immunol. 2003;170:114–22. [PubMed]
37. Teschendorff AE, Miremadi A, Pinder SE, Ellis IO, Caldas C. An immune response gene expression module identifies a good prognosis subtype in estrogen receptor negative breast cancer. Genome Biol. 2007;8:R157. [PMC free article] [PubMed]
38. Tamimi RM, Brugge JS, Freedman ML, et al. Circulating colony stimulating factor-1 and breast cancer risk. Cancer Res. 2008;68:18–21. [PMC free article] [PubMed]
39. Betensky RA, Louis DN, Cairncross JG. Influence of unrecognized molecular heterogeneity on randomized clinical trials. J Clin Oncol. 2002;20:2495–9. [PubMed]
40. Bild AH, Potti A, Nevins JR. Linking oncogenic pathways with therapeutic opportunities. Nat Rev Cancer. 2006;6:735–41. [PubMed]
41. Potti A, Dressman HK, Bild A, et al. Genomic signatures to guide the use of chemotherapeutics. Nat Med. 2006;12:1294–300. [PubMed]
42. Dressman HK, Berchuck A, Chan G, et al. An integrated genomic-based approach to individualized treatment of patients with advanced-stage ovarian cancer. J Clin Oncol. 2007;25:517–25. [PubMed]
43. Bonnefoi H, Potti A, Delorenzi M, et al. Validation of gene signatures that predict the response of breast cancer to neoadjuvant chemotherapy: a substudy of the EORTC 10994/BIG 00-01 clinical trial. Lancet Oncol. 2007;8:1071–8. [PubMed]
44. Allavena P, Signorelli M, Chieppa M, et al. Anti-inflammatory properties of the novel antitumor agent yondelis (trabectedin): inhibition of macrophage differentiation and cytokine production. Cancer Res. 2005;65:2964–71. [PubMed]
45. Paulus P, Stanley ER, Schafer R, Abraham D, Aharinejad S. Colony-stimulating factor-1 antibody reverses chemoresistance in human MCF-7 breast cancer xenografts. Cancer Res. 2006;66:4349–56. [PubMed]