The microenvironment in breast cancer progression: biology and implications for treatment

Breast cancer comprises a heterogeneous group of malignancies derived from the ductal epithelium. The microenvironment of these cancers is now recognized as a critical participant in tumor progression and therapeutic responses. Recent data demonstrate significant gene expression and epigenetic alterations in cells composing the microenvironment during disease progression, which can be explored as biomarkers and targets for therapy. Indeed, gene expression signatures derived from tumor stroma have been linked to clinical outcomes. There is increasing interest in translating our current understanding of the tumor microenvironment to the development of novel therapies.


Introduction
Breast cancer is the most common malignancy and the leading cause of cancer-related death in women worldwide [1]. Whereas localized disease is largely curable, metastatic or recurrent disease carries a dismal prognosis. Th e tumor microenvironment is now recognized as an important participant of tumor progression and response to treatment. As a result, there is increasing interest in developing novel therapies targeting the microenvironment, particularly as it relates to invasive and metastatic progression.
Th e normal breast duct consists of a luminal epithelial cell layer surrounded by myoepithelial cells, which produce and attach to the basement membrane. Th e breast microenvironment is composed of extracellular matrix (ECM) and numerous stromal cell types, including endo thelial and immune cells, fi broblasts, and adipocytes ( Figure 1). Early work investigating epithelial-mesenchymal interactions in tissue diff erentiation demonstrated that embryonic mesenchyme strongly infl uences the terminal diff erentiation of both embryonic and adult epithelia [2]. Th e infl uence of ECM is also observed in cell culture whereby normal mammary epithelial c ells in laminin-rich three-dimensional matrix form acini with a central lumen, become responsive to lactogenic hormones, and are capable of producing milk proteins [3,4]. Components of the microenvironment, including macrophages, myoepithelial and endothelial cells, and several ECM molecules, have been shown to play critical roles in mammary duct morphogenesis [5]. Similarly, the tumor microenvironment is increasingly recognized as a major regulator of carcinogenesis [6]. For decades, pathologists have appreciated the wound-like appearance of desmoplastic tumors, including some breast carci nomas. Th e now-famous assessment by Dvorak that 'tumors are wounds that do not heal' is being redefi ned at the molecular level as the role of the tumor micro environment in cancer progression is elucidated [7].
Breast tumors evolve via sequential progression through defi ned stages, starting with epithelial hyperproliferation and progressing to in situ, invasive, and metastatic carcino mas [8]. Both clinical and experimental data suggest that ductal carcinoma in situ (DCIS) is a precursor of invasive ductal carcinoma (IDC; Figure 2A,B) [9,10]. DCIS lesions contain proliferating neoplastic cells confi ned to the duct ( Figures 1B and 2). A critical, but poorly understood, step in breast cancer progression is the transition from in situ to invasive ductal carcinoma, which is defi ned by the loss of myoepithelial cell layer and basement membrane ( Figure 2). Th e subsequent spread of tumor cells to distant sites results in metastatic disease. Importantly, the tumor microenvironment has been implicated in each of these steps of cancer progression.

Comprehensive molecular profi ling of the microenvironment during tumor progression
In a pioneering study, Allinen and colleagues [11] isolated multiple cell types from normal breast, DCIS and IDC lesions and analyzed their gene expression profi les using Abstract Breast cancer comprises a heterogeneous group of malignancies derived from the ductal epithelium. The microenvironment of these cancers is now recognized as a critical participant in tumor progression and therapeutic responses. Recent data demonstrate signifi cant gene expression and epigenetic alterations in cells composing the microenvironment during disease progression, which can be explored as biomarkers and targets for therapy. Indeed, gene expression signatures derived from tumor stroma have been linked to clinical outcomes. There is increasing interest in translating our current understanding of the tumor microenvironment to the development of novel therapies.
serial analysis of gene expression (SAGE). In addition, genetic changes were detected by cDNA array comprehensive genomic hybridization and single nucleotide poly morphism arrays. Th e results of this study demon strated altered gene expression patterns in each cell type analyzed during breast cancer progression. Myoepithelial cells from normal breast tissue and DCIS had the highest number of diff erentially expressed genes. In ductal carcinoma in situ (DCIS), epigenetically and phenotypically altered myoepithelial cells (shown as brown cells) are surrounded by a still largely continuous basement membrane. Altered myoepithelial cells in DCIS are unable to aid polarization and organize the structure of the normal duct. At the same time in the stroma, the numbers of fi broblasts and infi ltrated leukocytes are increased and angiogenesis is enhanced. Cancer-associated fi broblasts (shown as yellow-green fi broblasts) and infi ltrated leukocytes elevate secretion of growth factors, cytokines, chemokines, and matrix metalloproteinases (MMPs) to promote tumor progression. Potential crosstalk between cell-cell and cell-matrix interactions are aberrantly regulated by both autocrine and paracrine networks of proteolytic enzymes, cytokines, and chemokines (red arrows; not all possible interactions are indicated). Interactions between stromal and cancer cells may interact with each other via paracrine signaling rather than direct cell-cell contact. A signifi cant fraction of these encode secreted proteins, suggest ing the activation of aberrant autocrine and paracrine regulatory loops. Many of the genes involved in normal myo epithelial cell diff erentiation and function were down regulated in DCIS-associated myoepithelial cells, including those encoding laminin and oxytocin receptor, whereas genes that promote tumorigenesis were increased, including CXCL12 and CXCL14. Importantly, clonal genetic aberrations were only identifi ed within the malignant epithelium. Although some contro versy remains regarding the existence and relevance of genetically abnormal stromal cells in human breast cancer, the majority of data support the hypothesis that gene expression and functional changes observed in the tumor microenvironment are not due to genetic alterations.

Myoepithelial cells
In a related study, Ma and colleagues [12] utilized laser capture microdissection and cDNA microarrays to analyze the gene expression profi les of patient-matched samples of normal-and tumor-associated epithelium (DCIS and IDC) and stroma. Again, the most dramatic gene expression changes both in the stromal and epithelial compartments were observed in the normal to DCIS transition. However, several ECM-degrading proteases, including matrix metalloproteinase (MMP)2 and MMP14, showed elevated expression during in situ to invasive carcinoma transition, which may play a role in the destruction of the basement membrane. One limitation of this study is that the diff erent stromal cells were not individually isolated prior to analysis. Additionally, laser capture microdissection is unable to separate luminal and myoepithelial cells and to fully isolate stroma from epithelium. In fact, the authors noted that the majority of genes with increased expression in IDC compared to DCIS epithelium were likely the result of stromal cell contamination.
Tumor-associated stromal cells maintain their altered phenotype in cell culture during prolonged passage [13], indicating hereditary changes such as epigenetic modifica tions, as genetic alterations are very rarely detected [14]. Hu and colleagues [15] tested this hypothesis by analyzing the comprehensive DNA methylation patterns of multiple cell types from normal breast tissue, DCIS, and IDC using methylation-specifi c digital karyotyping. Signifi cant methylation changes were identifi ed in each cell type during tumor progression. Th ese data imply that epigenetic modifi cations are at least in part responsible for the altered phenotype of cells composing the microenvironment in breast cancer.

The importance of myoepithelial cells
Th e key characteristic of invasive progression is the loss of the myoepithelial cell layer and basement membrane ( Figure 2B). Studies utilizing both in vitro co-culture and xenograft models have demonstrated that normal myoepithelial cells inhibit tumor growth. To better characterize the role of myoepithelial and stromal cells in the transition from DCIS to IDC, Hu and colleagues [16] utilized the MCF10DCIS.com xenograft model, which forms DCIS-like lesions that spontaneously progress to IDC with histological and molecular characteristics resembling human lesions. Co-injection of normal myoepithelial cells effi ciently suppressed the growth of MCFDCIS xenografts and the transition to IDC whereas fi broblasts had tumor growth and progression-promoting eff ects. Gene expression profi ling and immuno histochemical analysis of luminal epithelial and myoepithelial cells from MCFDCIS xenografts and human breast tissues have identifi ed transforming growth factor β (TGFβ) and Hedgehog pathways as specifi cally expressed in myo epithelial cells, implying an important role in these cells. Correlating with this, downregulation of TGFBR2, SMAD4, or GLI1 in MCFDCIS cells resulted in a decrease in myoepithelial cells and enhanced progression to invasion. Th ese studies suggest that the loss of myoepithelial cells promotes DCIS to IDC transition. Th e mechanism by which this loss occurs in human tumors is unclear. One hypothesis is that the diff erentiation of myo epithelial progenitors to fully diff erentiated myoepithelial cells is compromised due to signals emitted by tumor epithelial and stromal cells. Th e identifi cation and characterization of these paracrine factors may lead to the development of novel therapeutic approaches for the prevention and treatment of invasive breast cancer.

Cancer-associated fi broblasts
Normal fi broblasts maintain the extracellular environment through the production and remodeling of the ECM. Carcinoma-associated fi broblasts (CAFs) have distinct characteristics and substantial data support a role for CAFs in promoting tumor progression. CAFs are themselves heterogeneous, with a subset of them identifi ed as myofi broblasts expressing alpha smooth muscle actin (αSMA), others expressing fi broblast activation protein (FAP), desmin, S100A4 protein, and Th y-1 [17].
Orimo and colleagues [13] demonstrated that CAFs promote tumor growth and increase tumor angiogenesis by secretion of stromal derived factor (SDF)-1/CXCL12, which acts in a paracrine fashion to increase tumor cell proliferation via CXCR4. Hepatocyte growth factor (HGF), acting through the c-Met receptor tyrosine kinase, is another CAF-derived factor that has been impli cated in promoting tumor progression and metastasis. Th e paracrine activation of c-Met on tumor cells by HGF increases invasion of experimental DCIS lesions in xenografts [18]. Interestingly, co-culture of normal mammary fi broblasts with breast cancer cells can 'educate' the fi broblasts to secrete HGF and increase their tumor-promoting activities [19].
Th e origin of carcinoma-associated fi broblasts has been actively investigated and multiple hypotheses have been proposed. One possibility is that they are derived from native interstitial fi broblasts whose phenotype has been modifi ed by persistent aberrant signaling from neighboring tumor epithelial cells. Alternatively, they can be diff erentiated from bone marrow-derived mesenchymal stem cells that are recruited to the tumor site via endocrine stimulation by tumor-derived factors. Th is hypothesis is supported by the identifi cation of bone marrow-derived cells within tumors from patients who have previously received gender mis-matched allogenic bone marrow transplantation [20], although the recruitment of these cells by itself does not indicate functional relevance. However, a recent study demonstrated that certain xenografts 'instigate' the growth and metastasis of weakly tumorigenic cell lines via the activation and recruit ment of bone marrow-derived cells [21]. In particu lar, granulin-expressing bone marrow-derived cells stimulate both tumor progression and the desmoplastic response of resident fi broblasts. Th ese data as well as data from other labs [22][23][24][25] support the hypothesis that tumor-derived signals stimulate the bone marrow to produce and emit cells that promote tumor progression. Th us, targeting bone marrow-derived cells may infl uence the treatment of both localized and metastatic disease.

Matrix remodeling in tumor progression
Th e ECM of breast cancers is markedly abnormal and believed to promote tumor progression. MMPs are a large family of endopeptidases, which are synthesized predominantly by fi broblasts and normally participate in tissue remodeling and wound healing. Besides degrading ECM components, MMPs can also activate chemokines, cytokines, adhesion molecules, and growth factors, which contribute to tumor progression by increasing tumor cell proliferation (such as the release of insulin-like growth factor from ECM by MMP3 and -7) or by promoting angiogenesis (for example, activation of angio genic factors by MMP1, -2, -9, and -14) [26].
Abnormal physical characteristics of breast tumors, such as abnormal collagen cross-linking resulting in ECM stiff ening, also contribute to progression. Th e forces generated by this stiff ening lead to enhanced integrin and growth factor signaling that promotes invasion. Lysyl oxidase, an amine oxidase commonly expressed in breast tumors, promotes collagen cross-linking and enhances ECM stiff ening and its inhibition increases tumor latency and decreases tumor burden in the MMTV-Neu model of breast cancer [27]. In addition, elevated expression of lysyl oxidase-like 2 is associated with worse prognosis in early stage estrogen receptor (ER)-negative breast cancers [28].

Leukocytes
Th e link between infl ammation and cancer and the impor tance of infi ltrating leukocytes in tumor development are widely accepted, but the mechanisms mediating immune and tumor cell cross-talk are poorly understood. Immune cells are one of the most dynamic cell populations present within tumors and healing wounds and during the remodeling of breast tissue in pregnancy and involution [29,30]. During physiologic wound healing and breast tissue involution, immune responses are activated, but balanced towards the suppression of overt infl ammation, facilitating re-epithelialization and tissue healing [29,31].
High numbers of infi ltrating leukocytes are present in DCIS with focal myoepithelial cell layer disruptions [32], suggesting that they might play a role in invasive progression. Indeed, several groups have shown that tumor-associated macrophages (TAMs) facilitate angiogenesis, ECM degradation, and tumor invasion through activation of epidermal growth factor receptor signaling, secretion of proteases and paracrine signaling between tumor cells [33,34]. Loss of macrophages in colony stimulating factor (CSF)-1 defi cient mice (Csf1 op/op ) had no eff ect on tumor initiation but dramati cally reduced malignant progression [35]. To determine whether human breast cancer cells have similar response to macrophages, xenografts derived from human MCF-7 cells in immunodefi cient mice were treated with either mouse CSF-1 antisense oligonucleotide or CSF-1 small interfering RNAs. Th ese treatments suppressed mammary tumor growth by decreasing macrophage infi ltration, the production of MMPs and vascular endothelial growth factor (VEGF)-A, and endothelial cell proliferation [36]. Cumulatively, these functional studies in mouse models of breast cancer emphasize a prominent role for macrophages during breast cancer progression, and provide a plausible explanation for the association between TAMs and poor clinical prognosis.
Besides macrophages, other immune cells have also been implicated in breast cancer development. In a spontaneous mouse model of breast cancer the number of CD4+ regulatory T lymphocytes increased and systemic depletion of T cells using interleukin-2 immuno toxin fusion protein inhibited tumor growth and maintained a strong and persistent anti-tumor immune response [37]. A recent study analyzed tumor-infi ltrating CD8+ lympho cytes in breast tumors and found that a higher frequency of these cells in stroma surrounding the tumor was associated with better patient survival [38]. Because these cells are required for cell-mediated immunity, these results may indicate an active anti-tumor immune response against breast tumors, the intensity of which may infl uence the risk of distant metastatic progression.

The microenvironment of metastases
Although detailed cellular and molecular characterization of the metastatic tumor microenvironment has not been performed, numerous functional studies support a role for tumor epithelial-stromal cell interactions. During metastatic progression, tumor cells encounter and must survive in a number of diff erent microenvironments, such as blood, lymphatics, lymph nodes, and distant organs. Th e specifi c destination where the metastatic cell forms metastases may be mediated by the production of chemoattractants, by the local organ and/or the secretion of various products by the primary tumor that can create a favorable environment (Figure 3). Weinberg and colleagues [39] have shown that MDA-MB-231 xenograft tumors growing on one side of the mouse, termed instigator, mobilized bone marrow pre cursors via secreting osteopontin to home to secondary metastatic sites, where they promoted the growth of a less malignant cell line. Despite the fact that the mechanism by which osteopontin supported metastasis is not clear, this study not only demonstrates the recruitment of bone marrowderived cells to metastatic sites but also highlights the systemic eff ects of primary tumor growth.
In animal models of breast cancer, F4/80 + CSF -1R + CD11b + Gr1 -CX3CR1 high CCR2 high and VEGFR1 high host macrophages, distinct from classical macrophages, are recruited to metastatic breast cancer cells extravasating in the lungs [40]. In another study, primary tumor-stimulated macrophages increased the metastatic ability of tail-vein injected tumor cells via secretion of MMP9 and VEGF [41], although lung metastases only developed in mice bearing primary tumors, again indicating sys temic eff ects of tumor growth.
A recent study provided another example of how paracrine signals emitted by stromal cells may play important roles in promoting breast cancer metastasis. Receptor activator of nuclear factor-κB (RANK) is highly expressed in human breast carcinoma cells, but the source of RANK ligand (RANKL) and its role in breast cancer metastasis was mostly unknown. In a mouse model of ERBB2-driven mammary tumors, Tan and colleagues [42] identifi ed a role for RANKL in the formation of lung metastases. Only CD4 + regulatory T cells in the stroma of mammary tumors produced RANKL, which stimulated RANK-expressing ERBB2-driven mammary tumors to metastasize. In addition, most RANKL + T cells were located adjacent to myofi broblasts expressing T-cell-attracting chemokine CCL5. Th ese studies indicate the possibility that recruitment of distinct populations of leukocytes and stromal cells is required in the process of metastasis.

The prognostic relevance of microenvironmental changes
Interest in targeting the microenvironment comes not only from the identifi cation of 'druggable' targets (enzymes and receptors) but also from clinical data demonstrating that stroma-derived gene expression patterns predict clinical outcome. One of the fi rst such studies, by Wang and colleagues [43], fi rst defi ned an 'activated fi broblast' gene expression signature that was then analyzed in genome-wide expression data of bulk breast tumor samples from patients with lymph nodenegative disease to determine if it has prognostic rele vance. A 76-gene signature was identifi ed that predicted shorter distant metastasis-free survival independent of patient age, tumor size, grade, or ER status. Th e authors proposed that patients with smaller (<2 cm) tumors and good prognosis based on their signature may not benefi t from adjuvant chemotherapy and could be spared associated morbidities [43]. A related fi broblast signature was developed by Chang and colleagues [44], defi ned as 'core serum response' genes activated in fi broblasts exposed to serum. Many of the functions of the identifi ed genes were related to wound healing, such as matrix remodeling, myofi broblastic activation, cell proliferation and motility. Th is gene signature was then validated as a predictor of clinical outcome when applied to expression profi ling of whole tumor samples [44,45].
More recently, investigators have focused on breast tumor stroma-derived gene expression changes. Finak and colleagues [46] analyzed gene expression changes within tumor stroma and identifi ed 'outcome-linked' clusters that were independent of tumor grade, size, hormone receptor, and lymph node status. Th e poor outcome cluster was associated with increased expression of hypoxia-and angiogenesis-related genes, whereas the good outcome cluster was enriched for T cell immune responses and natural killer cell markers. Th e authors derived a 26-gene signature that predicted clinical outcome independent of tumor ER or human epidermal growth factor receptor (HER)2 status, implying distinct stromal subtypes distinct of breast tumor subtype. In a related study, Farmer and colleagues [47] analyzed the gene expression profi les of reactive tumor stroma from biopsies obtained prior to treatment of ER-negative tumors and derived a signature that predicted clinical response to neoadjuvant chemotherapy. Similarly, Bergamaschi and colleagues [48] defi ned several ECM signatures based on gene expression profi ling of whole tumor samples that pre dicted clinical response. In addition to these global profi ling studies, individual prognostic markers have also been identifi ed. In DCIS, patients with tumors expressing low levels of CD10 (a myoepithelial cell surface marker) had a higher risk of local relapse [49]. While potentially interesting, small sample size and mixing outcomes from patients treated with mastectomy or lumpectomy limit the validity of the fi ndings. Th e expression of several CAF-derived proteins is also associated with clinical outcome. For example, high levels of platelet-derived growth factor-β receptor or SDF-1/CXCL12 and decreased caveolin-1 are associated with worse clinical outcome [50][51][52].

Therapeutic targeting of the microenvironment
New insights into the tumor microenvironment, both focused and global, are identifying novel therapeutic Metastasis is a complicated multistep process that requires cancer cells to escape from the primary tumor, survive in the circulation, seed at distant sites and grow. Each step involves stromal cell and paracrine interactions of the microenvironment. Aberrantly secreted chemokines and cytokines from the primary tumor circulate into the blood stream, creating a premetastatic niche even before tumor cell mobilization. Secreted factors functionally activate bone marrow-derived cells, which are then released into the circulation to subsequently incorporate these cells into distant organs, such as lung and liver, to create a favorable microenvironment for the cancer cell to be seeded. For the cancer cells to invade into the blood circulation, proteases are produced by bone marrow-derived cells, including macrophages and fi broblasts. Following tumor cell intravasation, a series of steps is required for the establishment of secondary tumors in the metastatic sites. Disseminated cancer cells preferentially form metastases at sites where activated bone marrow-derived cells are localized and the primary tumor has created a favorable environment at the local organ. After seeding, persistent growth of the metastatic tumor requires the establishment of a vasculature that can be possibly achieved through the production of angiogenic growth factors. targets. Currently, three types of tumor microenvironment-targeting therapies are in clinical practice: aromatase inhibitors (which target the aromatase enzyme predominantly expressed by stromal components), angiogenesis-modulating agents (including anti-VEGF receptor antagonists), and inhibitors of HER family receptors (such as trastuzumab, which inhibits receptor signaling on epithelial cells triggered by stroma-produced growth factors).

Metastatic sites
Whereas aromatase inhibitors and trastuzumab have become standard therapy, the clinical eff ectiveness of angiogenesis inhibitors is less clear [53]. In addition, there is a concern that inhibition of angiogenesis may enhance disease progression based on data in animal models where treatment with anti-angiogenic agents increased invasiveness and metastatic spread [54,55]. Potential selection for hypoxia-tolerant clones or establishment of leaky, metastasis-promoting vessels could explain these results [56]. In addition to these concerns, bevacizumab, a clinically approved VEGF inhibitor, has been associated with signifi cant adverse reactions, including hemorrhage, neutropenia, gastrointestinal perfora tion, and thromboembolic events. A recently published meta-analysis of 16 randomized controlled clinical trials administering bevacizumab demonstrated that this agent, when used in combination with chemotherapy, was associated with increased risk of fatal treatment-related adverse events compared to the use of chemotherapy alone [57]. Whereas targeting the tumor microenvironment is an exciting possibility, side eff ects resulting from disruption of homeostatic functions in normal tissues are very likely, as was demonstrated by the poor tolerability of MMP inhibitors. In the past years, numerous targets have been investigated in early clinical trials, including antibodies targeting FAP, c-Met antagonists and multi-targeted receptor tyrosine kinase inhibitors such as sunitinib [58,59]. Some of these have been plagued by poor side eff ect profi les whereas others have been well tolerated, but ineff ective.
In addition to drugs being developed against novel targets, the anti-tumor eff ects of several older agents seem to be mediated through microenvironmental actions. For example, bisphosphonates (e.g., zoledronic acid), which are used for the treatment of osteoporosis and the management of bone metastasis, are now recog nized to have activity outside of the skeleton, including direct anti-tumor eff ects on the malignant epithelium, and modulating angiogenesis and immune cell infi ltration [60].
Osteoclasts are an important component of the normal bone microenvironment as well as bone metastases. Meta static tumor cells secrete growth factors that activate osteoclasts, which degrade bone and release additional growth factors, triggering a paracrine cascade that promotes tumor growth and bone destruction. Denosumab is a monoclonal antibody that binds RANKL and inhibits osteoclast function. Recently, denosumab was compared to zoledronic acid in a phase III randomized clinical trial in breast cancer patients with bone metastases [61]. Th e results of this trial demonstrated that denosumab was well tolerated, and superior to zoledronic acid in delaying time to complications of bone metastases (that is, pathological fractures) but did not improve survival.
An important new hypothesis in targeting the tumor microenvironment is the induction of microenvironmental 'reprogramming' . Rolny and colleagues [62] recently published an intriguing discovery that overexpression of histidine-rich glycoprotein (HRG) in murine syngeneic tumor models induced 'normalization' of TAMs and blood vessel structure. Importantly, this was associated with decreased breast tumor growth and pulmonary metastasis, and increased sensitivity to chemotherapy. Th e authors demonstrated that the eff ects of HRG were dependent on the presence of TAMs and in particular TAM conversion from the 'M2' pro-tumor/ pro-angiogenic phenotype to the 'M1' anti-tumor/proinfl ammatory phenotype. In addition, HRG expression was associated with vessel normalization, which was also dependent on TAM activity. Th is study has linked the phenotypic switching of TAMs by HRG with orchestration of vascular normalization. Th is eff ect of HRG on TAMs seems to be mediated through the down-regulation of placental growth factor, though the precise mecha nism is unclear.
Coussens and colleagues have recently des cribed that the ratio of macrophages to T cells predicts clinical outcome, with increased macrophage recruitment associated with worse outcome [63]. Interestingly, cytotoxic chemotherapy induces the recruitment of TAMs into invasive carcinomas by increasing the expression of CSF-1, a macrophage-recruiting cytokine. Inhibition of TAM recruitment by several approaches increased the effi cacy of chemotherapy by decreasing tumor development and metastasis in a CD8+ T cell-dependent manner. Th e authors postulate that chemotherapy increases TAM recruitment that subsequently modulates T cells, favoring the CD4+ T-cell phenotype, which leads to inhibition of anti-tumor immunity. Inhibition of TAMs promotes CD8+ T-cell recruitment and is associated with increased anti-tumor immunity. Th ese data support the development of novel compounds that target TAMs and, in concert with cytotoxic chemotherapy, can encourage anti-tumor immunity [63].
Response to chemotherapy can be assessed by changes in tumor size and imaging characteristics as well as histopathological assessment. Tumor growth can progress, stabilize or regress in response to chemotherapy. In the case of a good tumor response characterized by tumor shrinkage, it is possible that the tumor microenvironment actively participates in the tissue remodeling. A simplistic model would be that classic cytotoxic therapies kill tumor cells, which then gives stromal components the opportunity to 'mop up' the necrotic debris. An alternative hypothesis is that the microenvironment, either as a direct eff ect of chemotherapy or in response to signals derived from the assaulted epithelium, acquires an altered phenotype that independently inhibits tumor growth. Identifi cation of these microenvironmental changes that take place during tumor regression have not been intensively studied. Such studies may identify 'reprogramming' events that can be pharmacologically mimicked with novel, non-cytotoxic agents. Such manipulation of the microenvironment to promote an anti-tumor phenotype in stromal components represents a novel treatment strategy.
Metronomic therapy refers to the frequent or continuous administration of low doses of chemotherapy with the goal of eliciting an anti-tumor response while minimizing side eff ects. Interestingly, metronomic therapies have been implicated in inhibiting angiogenesis, promoting a benefi cial immune response and tumor dormancy [64]. Th e mechanisms by which metronomic therapies infl uence these changes are largely unknown. One possible explanation is that these chronic therapies are re-modeling the epigenetic landscape of the tumor microenvironment. Just as the epigenetic changes identifi ed in tumors possibly arise from chronic exposure to pro-tumorigenic signals derived from malignant epithe lium, one could postulate a similar aff ect from chronic exposure to anti-neoplastic agents.
Epigenetic therapies, such as the histone deacetylase inhibitor suberoylanilide hydroxamic acid (also called vorinostat), are currently under clinical investi ga tion for the treatment of breast cancer. While developed to target the malignant epithelium, their eff ect on the microenvironment may induce alterations that help orchestrate an anti-tumor response. Currently, there are no reports of the gene expression or epigenetic profi les of tumor samples obtained from patients treated with metronomic therapy or histone deacetylase inhibitors. Th ese data will be valuable to our understanding of the microenvironmental changes induced by these therapies.
Besides identifying new therapeutic targets, the microenvironment has also been implicated in chemotherapy resistance. Weaver and colleagues [65], working with three-dimensional cultures, demonstrated that sensitivity to chemotherapy could be infl uenced by cellular polarity, which is mediated in part by integrin expression and exposure to basement membrane. Hiscox and colleagues [66] demonstrated that resistance to fulvestrant, an antiestrogen, promotes an invasive phenotype secondary to increased epithelial expression of c-MET, which is then activated by fi broblast-produced HGF. Loeff er and colleagues [67] generated an oral vaccine against FAP and studied its eff ect on the growth of multidrug-resistant breast cancer in murine xenografts. Th e vaccine decreased tumor collagen I, an ECM component previously implicated in chemotherapy resistance, and tumors from these animals had a signifi cant improvement in chemotherapy uptake as well as decreased tumor growth resulting in increased survival. Th ese data demonstrate that, in addition to promoting progression, the microenvironment can modulate sensitivity or resistance to chemotherapy.

Conclusions and future directions and challenges
Breast cancer remains a major clinical challenge with considerable mortality as well as treatment-associated morbidity. Novel treatment strategies are urgently needed, especially in the setting of metastatic disease where outcomes are still dismal. Th e breast cancer microenviron ment is a complex mixture of cells, the proteins they secrete, and the ECM in which they reside. Alterations within the microenvironment are now recognized during key steps of tumor progression, making them attractive candidates for therapeutic modulation. Th e relative genomic stability of stromal cells makes the develop ment of chemoresistance to stromal-target therapy less likely. Furthermore, the epigenetic modifi cations that contribute to phenotypic alterations, while inheritable, are reversible, and there is mounting interest in 'normalizing' the altered stroma, thereby abrogating its tumor-supporting role.
One major obstacle facing stromal-targeted therapy is avoiding disruption of homeostatic function in normal tissues. Despite these challenges, our improved understanding of key aspects of tumor progression should lead to treatment strategies that can discriminate normal tissue from neoplasm.
How the tumor microenvironment changes during chemotherapy-induced tumor regression is still poorly understood. Insights into these changes may identify important pathways, which can be activated using noncytotoxic therapies. As the mainstay of aggressive forms of breast cancer will continue to rely heavily on cytotoxic therapies for the foreseeable future, agents without these characteristics will be particularly valuable in combi nation trials.
Translating our burgeoning knowledge of microenviron mental infl uences on tumor progression into clinical practice is challenging. For example, targeting bone marrow-derived mesenchymal cells, which infl uence both primary tumor growth and the metastatic niche, prior to clinically evident metastatic disease makes intuitive sense. However, testing these potentially impor tant agents in early clinical trials of recurrent or refractory disease may not yield signifi cant improvements in such advanced disease. Th oughtful clinical trial design, including neoadjuvant therapy during which pre-and post-treatment tumor samples can be analyzed, will be vitally important in developing stromal-targeted therapy. Despite these challenges, taken together, the majority of data support the rationale for targeting the tumor microenvironment in the treatment of breast cancer.

Competing interests
KP receives research support and is a consultant to Novartis Pharmaceuticals, Inc.; KP is also a member of the scientifi c advisory boards of Metamark Genetics Inc., and Theracrine Inc., and owns stocks of Aveo Pharmaceuticals Inc.