Until recently, the acceptance of a microenvironmental contribution to tumor development was limited mostly to that of a role for angiogenesis in solid tumors beyond a certain size [4]. The more recent, albeit broad, focus on the importance of the microenvironment, in particular inflammation, in essence represents a renaissance of the historic view of cancer as a never-healing wound [5]; in chickens, wounding itself can be a co-carcinogen that creates a conducive tissue environment for Rous sarcoma virus (RSV) tumorigenesis (reviewed in [6]), and bacterial or viral infections in humans induce chronic inflammation, leading to a significantly increased risk of cancer (reviewed in [7]). Thus, although sporadic or inherited mutations in crucial genes might represent initiating events in tumorigenesis, chronic inflammation favors the selection of additional features in initiated cells that might promote their full malignant transition (reviewed in [8]).

Our understanding of the importance of the microenvironment has now shifted: whereas previously it was regarded as merely providing passive support for epithelial cells during carcinoma, the microenvironment is now considered an active contributor to tumor progression. The recent discoveries that the capability of the microenvironment, in this case the bone marrow, is the sole cause of hematopoietic disorders [9,10] bear witness to the possibility that the tumor microenvironment essentially could control the course of breast tumor progression.

Upon transplantation of a breast cancer cell line to mice transplanted with bone marrow from double-mutant recombinase-activating gene-1 (RAG-1)^-/- β-gal transgenic and green fluorescent protein (GFP) transgenic mice, 20% of the myofibroblasts in the resultant tumors were bone-marrow-derived [15]. In another study, SV40 large T antigen-, human telomerase reverse transcriptase (hTERT)- and RasV12-transfected human breast cells recruited Rag-1^null and GFP transgenic bone-marrow-derived cells to comprise ~90% of the tumor-associated cells, of which the majority expressed α-smooth muscle (α-sm) actin [16]. Given that cancer is a tissue-specific disease and that human breast stroma differs appreciably from the stroma of mouse mammary gland [3], we believe that additional markers of human fibrocytes are needed. This is particularly pertinent to breast tissue. Currently, the most widely used marker of fibrocytes, CD34, is expressed by all fibroblasts in normal breast but is lost in α-sm-actin-positive myofibroblasts [17].

We have shown previously that resident, normal fibroblasts readily undergo conversion to myofibroblasts in response to tumor cells in culture (reviewed in [3,18]). Multiple studies have documented that the converted stroma in turn supports cancer cell growth and metastasis [3,18–22]. That the reactive stroma itself, once converted by any means, might also signal to epithelial cells and eventually convert them to cancer cells is beginning to be appreciated.

A comparison of stromal cells isolated from tumors and normal tissues, respectively, suggests that stromal cells might indeed provide cues for malignancy. In recombination studies of human prostate-derived cells, normal fibroblasts support growth arrest and normal histology of the epithelium, whereas so-called cancer-associated fibroblasts (CAFs) direct immortalized prostatic epithelial cells towards adenocarcinomas [23]. The myoepithelial cells isolated from ductal carcinoma in situ (DCIS) were shown to be grossly abnormal and to secrete many cytokines and matrix metalloproteinases (MMPs) that normal myoepithelial cells did not [24]. Thus, stromal as well as myoepithelial cells might be a switch for overt malignancy in luminal epithelial cells. Conversely, an experimental model of DCIS has demonstrated that only myoepithelial cells isolated from normal glands can inhibit progression of DCIS to invasive carcinoma [24].

Evidence that not only certain cell populations but also changes in the microenvironment as a whole can promote tumorigenesis also comes from studies of irradiated mouse mammary gland stroma. Ionizing radiation is a known albeit weak carcinogen in the mammary epithelium. In addition to its widely accepted DNA-damaging (i.e. mutagenic) effect, irradiation elicits persistent microenvironmental alterations in ECM composition and cytokine activities. Transplantation of a p53-null mouse mammary epithelial cell line to cleared mammary fat pads showed that tumor incidence was 81% in irradiated animals but only 19% in sham-irradiated animals. Experiments in animals with one-sided irradiation confirmed that tumors formed only on the irradiated side [25].

It is intriguing to note the dramatic epigenetic changes that were clearly transmitted to progenies after combined irradiation and transforming growth factor-β (TGF-β) treatment of single human breast epithelial cells [26]. There was a heritable disruption of epithelial cell polarity and multicellular organization in three-dimensional (3D) assays in the entire surviving population, indicating non-mutational but heritable mechanisms [26]. Of note, the stroma itself also has been suggested as a target of the chemical carcinogen N-nitrosomethylurea, which was previously thought to exert its effect only through direct mutation induced in the epithelium [27]. However, confirmation of these latter findings via the use of marked stromal cells would be essential for interpretation of the data.

The stromal compartment also includes mast cells, macrophages and T-cell subtypes, all of which are emerging as essential modulators of tumor growth, and more research clearly needs to be done in this area. The influence of immune cell infiltrates on tumor development might be favorable or adverse, depending on the cell types involved. Mast cells are attracted to tumors by tumor-derived chemoattractants and affect tumor development in accordance with local tumor conditions. In a mouse tumor model, recruitment of mast cells was required for angiogenesis and expansion of Myc-induced pancreatic islet tumors [28], whereas in human breast cancer, the presence of mast cells might be an independent, favorable prognostic factor [29]. Macrophages, as a major population of infiltrating cells in the tumor stroma, are considered to have an overall tumor-supporting role. The attraction of tumor-associated macrophages is thought to be caused by hypoxia, which stimulates macrophages to perform pro-angiogenic functions [30]. Pollard and coworkers, who were the first to describe the importance of macrophages in breast tumors, developed a transgenic mouse model that displays a progression series from benign tumor to malignancy. The results show a marked increase in macrophage infiltration prior to the transition to malignancy, which is associated with angiogenesis, thus showing that the microenvironment is directly involved in the development of the high-density tumor vasculature that is required for tumor progression (reviewed in [31]). These findings highlight the crucial role that the host stromal reaction, including the inflammatory cell infiltrate, has in modulating cancer progression. Immune cells, which seem to be very widely involved in tumor development, are easily accessible, making them a potential and attractive tool for tumor therapy, as discussed below.

Transgenic mouse models have also pointed to the importance of ECM and TGF-β signaling in tumor development. Disruption of the ECM in the mammary gland of normal mice ([32], discussed in more detail below) led to the formation of reactive stroma [33] long before tumors were formed [34]. Importantly, knockout of TGF-β receptor II revealed prostatic hyperplasia in the stroma, as well as in adjacent epithelium, leading to prostatic intraepithelial neoplasia, a presumed forerunner of carcinoma [35]. In spite of the fact that TGF-β signaling has so far not been found to be inactivated during the normal course of tumor pathogenesis, these latter studies provide a small but growing number of counter-examples to the assumption that mutations in epithelial cells are a required initiating event for neoplastic change.

It has long been postulated that the observed lifetime protective effect of an early full-term pregnancy against estrogen-receptor-positive breast cancers among postmenopausal women is due to the complete cellular differentiation that the gland undergoes during the process of pregnancy ([37] and references therein). This belief has recently been questioned experimentally: it seems that it is not differentiation per se but exposure to estrogen at a critical stage of the gland’s development that causes the protective effect [38].

However, the long-term protective effect of pregnancy is not constant and varies with age at the time of the first pregnancy; moreover, pregnancy is associated with its own paradoxical but relatively transient increase in risk of breast cancer, which often leads to metastasis with poor prognosis. The increase of risk at first pregnancy peaks 6.5 years post-partum and persists for up to 15 years after parturition [39]. That such a window of time can be defined suggests that an event subsequent to pregnancy is involved in breast cancer progression. One obvious post-partum change in mammary gland physiology after pregnancy is gland involution, during which a tissue-remodeling program similar to those of inflammation and wound healing is activated [40]. Mammary gland involution has been investigated widely in rodent models. Thus, the stroma from mammary glands undergoing weaning-induced involution contains elevated levels of MMPs (including MMP-3), urokinase plasminogen activator, fibrillar collagen and proteolytic fragments of laminin and fibronectin [41–43]. The possible effect of ECM components has been tested directly: MDA-MB-231 cells were xenografted to mammary fat pads with extracts of endogenous mammary ECM from quiescent, nulliparous and involuting rat glands, respectively. Although the volumes of the resultant tumors were the same, the ability to metastasize was highly dependent on the source of the matrix: ECM from involuting glands significantly promoted tumor cell metastasis [43]. This suggests that the physiological status of the microenvironment might determine epithelial cell potential and fate, even in the form of tissue extract. Although the underlying molecular mechanisms still await elucidation, the prominent proteolytic modulators of the breast tissue microenvironment, MMPs, undoubtedly have an important role in the process. In the experiment above, the stromal ECM of the mammary glands from involuting rats was collected between days four and six after weaning, coinciding with the MMP-3-dependent phase of involution [44].

We have shown a role for MMP-3 in tumor development in physiological settings by creating MMP-3 transgenic mice in which the BM was disrupted in the middle of pregnancy and mammary glands underwent ‘involution’ in mid-pregnancy instead of during weaning [32]. As these animals aged, they developed mammary tumors [34]. Although MMP-3 does not seem to influence breast cancer susceptibility in humans, the MMP-3 5A/6A promoter polymorphism seems to be linked to a higher risk for metastasis among breast cancer patients [45]. The molecular pathways by which MMP-3 might induce epithelial—mesenchymal transition (EMT) as well as genomic instability, two crucial events in the process of malignant progression, are now being dissected [46]. Exposure of mouse mammary epithelial cells to MMP-3 cleaves E-cadherin, alters the cytoskeleton and cell shape and provokes the expression of Rac1b. This in turn causes a rise in cellular reactive oxygen species (ROS) and an increase in the transcription factor Snail, leading to EMT. The above scenario can be achieved by elevating ROS, thus bypassing MMP-3, or by expression of Rac1b [46].

MMP-3 has also been implicated in overall age-associated risk of cancer development. Although most age-related cancers arise from epithelial cells, accumulation of senescent stromal cells can favor tumor development by both compromising tissue renewal capacity and secreting cytokines and proteolytic enzymes, which alter tissue homeostasis. Thus, senescent fibroblasts have been shown to support epithelial proliferation and to alter epithelial morphology and functional differentiation in part through high-level secretion of MMP-3 [47].

Once breast tumors are formed, they are often characterized by the presence of an extremely dense collagenous stroma, the so-called desmoplastic response, which is produced by myofibroblasts in the interstitium. Desmoplasia is a prominent feature of the most frequent breast carcinoma lesion, ‘infiltrating ductal carcinoma not otherwise specified’, which constitutes 75–80% of all breast tumors [3]. The desmoplastic response forms stromal tissue that can vary from being predominantly cellular, with accumulated fibroblasts and myofibroblasts and little collagenous tissue, to dense and collagenous, with few stromal cells present [50]. How the nature of the desmoplastic reaction relates to the composition of the ECM and breast density before tumor development is not known. Nevertheless, direct histopathological analyses have shown increased expression of molecular markers of desmoplasia, such as collagen, lumican, dEcoRIn and syndecan-1, in the stroma of women with high density breasts [51,52].

The relevance of the varieties of desmoplasia for tumor progression might be difficult to discern, but the presence of fibrotic foci, defined as radiating fibrosclerotic scars with varying cellularity, in the desmoplastic stroma has been shown to be a significant histopathological parameter of worse prognosis [53]. Recent discoveries further imply that the physical properties of a tissue might determine specification, morphology, differentiation and function. It has been shown recently that matrix elasticity per se directs lineage differentiation of naïve mesenchymal stem cells, thus implicating physical properties of the microenvironment in fate determination [54], and it is likely that similar mechanisms might apply to tumor development. Another elegant study addressed whether desmoplasia and tissue stiffening might actively promote malignant behavior [55]. Compression analyses of normal and malignant mammary tissues from transgenic mice showed that normal tissue was very soft, whereas tumors and adjacent stroma were relatively stiff. Human breast epithelial cell lines were then cultured in matrices with stiffness ranging between that of normal tissue to that of tumor tissue. Increased collagen content of the substrate compromised tissue organization, inhibited lumen formation and destabilized cell—cell junctions in non-malignant structures. In addition, increase in substrata rigidity enhanced growth by inducing Rho-generated cytoskeletal tension to promote focal adhesion assembly and increased extracellular-signal-regulated kinase activation [55]. Thus, normal tissue function (i.e. homeostasis) is favored by a ‘soft’ microenvironment and low tension, whereas tumorigenic behavior is initiated when the ECM becomes chronically stiffer and/or when mechanotransduced tension rises [55].

The stromal profiles are also being explored at the genome level. In germ-line BRCA1/2 (see Glossary)-related breast cancers, the average frequencies of loss of heterozygosity or allelic imbalance (LOH/AI) in stroma and epithelium were reported to be similar, whereas in sporadic breast cancer, the average epithelial LOH/AI frequency far exceeds the average stromal LOH/AI frequency [61]. Further scrutiny has revealed that stroma-specific LOH/AI is associated with mutations in the somatic tumor suppressor gene TP53 and regional lymph-node metastases in sporadic but not hereditary breast cancers, suggesting that genomic alterations in stromal cells contribute to the clinical outcome [62]. However, because genetic alterations in the stroma of sporadic breast cancer can be distinct from, and hence independent of, those of adjacent malignant epithelium, the epithelial and stromal mutations are not derived from a common progenitor but rather seem to both undergo similar selection pressures in the tumor microenvironment [63]. In some instances, stromal changes might not even be restricted to the microenvironment: recently, frequent genetic alterations have been revealed in normal skin in breast cancer patients [64]. In spite of the limited sample number (twelve), these findings provide evidence of a genetic basis for the aberrant behavior of explanted normal skin fibroblasts from patients with breast cancer, a phenomenon that was originally described more than two decades ago [65]. It is tempting to speculate that a broader spectrum of LOH/AI affecting multiple cell populations in the organ might ultimately be manifested in a more ‘localized’ carcinoma [64].

Despite the dual role of TGF-β in normal function and cancer, breast tumor cells in which the TGF-β signaling pathway has gone awry have been shown to undergo apoptosis in response to rapamycin [73]. In the future, the simultaneous use of agents such as rapamycin incombination with TGF-β inhibitors might be even more effective in targeting tumor cells and at the same time might abrogate the cancer-promoting effect of the microenvironment.

Further resolution of the stromal component itself, in terms of cellular and ECM compositions and gene expression profiles combined with histological analysis, could improve risk assessment and add useful parameters of prognosis. One result of this approach is the recent recognition of the scar-like centre, the fibrotic focus, of invasive breast carcinomas as a surrogate marker for hypoxia and lymphangiogenesis, in addition to being a practical and reproducible histological prognostic parameter [74]. Another important finding is that ECM signatures identify breast cancer subgroups with different clinical outcome [75].