An “elite hacker”

Abstract

The year 2011 marked the 40 year anniversary of Richard Nixon signing the National Cancer Act, thus declaring the beginning of the “War on Cancer” in the United States. Whereas we have made tremendous progress toward understanding the genetics of tumors in the past four decades, and in developing enabling technology to dissect the molecular underpinnings of cancer at unprecedented resolution, it is only recently that the important role of the stromal microenvironment has been studied in detail. Cancer is a tissue-specific disease, and it is becoming clear that much of what we know about breast cancer progression parallels the biology of the normal breast differentiation, of which there is still much to learn. In particular, the normal breast and breast tumors share molecular, cellular, systemic and microenvironmental influences necessary for their progression. It is therefore enticing to consider a tumor to be a “rogue hacker”—one who exploits the weaknesses of a normal program for personal benefit. Understanding normal mammary gland biology and its “security vulnerabilities” may thus leave us better equipped to target breast cancer. In this review, we will provide a brief overview of the heterotypic cellular and molecular interactions within the microenvironment of the developing mammary gland that are necessary for functional differentiation, provide evidence suggesting that similar biology—albeit imbalanced and exaggerated—is observed in breast cancer progression particularly during the transition from carcinoma in situ to invasive disease. Lastly we will present evidence suggesting that the multigene signatures currently used to model cancer heterogeneity and clinical outcome largely reflect signaling from a heterogeneous microenvironment—a recurring theme that could potentially be exploited therapeutically.

The Breast Developmental “Program”: Functional Differentiation Requires Intricate and Dynamic Interactions between Epithelial and Stromal Cells and Their Underlying Extracellular Matrix

The mammary gland is comprised of a series of polarized, bilayered epithelial ducts residing in a complex microenvironment comprised of basement membrane and other extracellular matrix (ECM) molecules and a number of stromal cell types, including adipocytes, fibroblasts, endothelial cells, mesenchymal stem cells, mast cells and macrophages among others.^1-4 The exact ratio and localization of these cell types varies between species—for instance, adipose tissue comprises the majority of the mouse mammary gland mesenchyme and immediately juxtaposes the ducts, but is a minor component that is spatially segregated from the epithelium of the human breast stroma, which comparatively is more ECM-enriched.⁵ Furthermore, the stromal microenvironment in the breast varies dramatically in fat and ECM composition even between adjacent lobules.⁶ As such, whereas we may use the mouse mammary gland development as a proxy for human breast development, the absolute importance of each stromal component to species-intrinsic organogenesis can only be estimated.

The mammary gland is unique among organs in that the majority of its development and functional differentiation occur post-natally during puberty and pregnancy. In response to estrogen, progesterone and other hormones expressed during puberty in females, the primordial ducts begin to infiltrate into the surrounding stromal tissue. This developmental program requires extensive remodeling of the underlying ECM surrounding the cells, regulated epithelial cell proliferation and collective migration until the ducts span the entire fat pad, a process termed branching morphogenesis.^1,7,8 In many ways, the coordinated proliferation and migration during branching is similar to that of an invading tumor (Fig. 1). Branching morphogenesis is not unique to the mammary gland and rather seems to be a developmental program conserved in the differentiation of multiple organs, such as the salivary gland and lung, though the patterns of bifurcation are unique to individual glands and are strongly influenced by the stroma.⁷

Figure 1. Dynamic microenvironments regulate both normal mammary gland development and malignant progression. Branching morphogenesis during normal mammary gland development requires an extensive dialog between epithelial and stromal cells and their underlying extracellular matrix (ECM). These events resurface or are amplified during malignancy, and often are required for cancer progression. As such, the microenvironment represents a “security vulnerability” through which tumors can “hack” the normal developmental “program” to manifest themselves.

In a classical series of experiments, it was shown that grafting the embryonic salivary gland mesenchyme into the adult mammary gland, or co-grafting salivary stroma with mammary gland epithelium in the subrenal capsule, changes the epithelial branching architecture to resemble that of the salivary gland.^9,10 Furthermore, co-culturing mammary tumors with embryonic mammary gland stroma in a trans-filter causes tumor differentiation and “reversion” into a more benign state not observed in controls; this phenotypic reversion is reversible if the tissues are transplanted back into mice.^11,12 More recently, in a series of dramatic experiments, it was shown conversely that the mammary gland microenvironment can cause trans-differentiation of cells originating from other organs, including the testes, neural progenitor cells, and embryonic cancer cells, into mammary epithelial cells that are integrated into functional ducts and are able to produce milk,^13-15 providing compelling evidence that cell differentiation and branching architecture is instructed by the tissue microenvironment.

The stromal microenvironment contributes many biochemical and mechanical signals, the sum of which direct proper mammary gland branching morphogenesis. Several paracrine-acting growth factors originating from the stromal compartment have been shown to promote ductal outgrowth and lumen formation, such as HGF, FGF, IGF, EGF, neuregulin, TGFα, epimorphin and Wnts, whereas TGFβ inhibits branching to refine branching architecture and spacing.^16-23 Whereas our understanding of the exact mechanisms by which the microenvironment directs mammary gland branching is fragmentary, from what we know, it is likely a combination of several cells that are responsible for enriching the tissue microenvironment with these instructional signals under stringent regulation in space and time.

Of the cell types participating in these heterotypic interactions, adipose cells are among the most abundant present within the mammary stroma, and are indispensible to mammary gland development and function.^24-26 The stromal fat provides not only a scaffold upon which the branching epithelia and its associated vasculature is supported and constrained, but also serves as an essential reservoir of metabolites and signaling molecules to direct gland development. Using 2D gel mass-spectrometry analysis, 359 proteins were identified in the adipose of normal human breast tissue including many diverse growth factors and chemokines, suggesting adipocytes are central participants in the heterotypic signaling interactions guiding mammary gland vascularization, epithelial cell proliferation and migration, recruitment of leukocytes, and ultimately its functional differentiation.²⁷ In fact, there is now an entire literature on the adipose as a tissue that secretes a suite of “adipokines” having endocrine, paracrine and autocrine roles.²⁸

The adipose can secrete and respond to both estrogen and prolactin, hormones necessary for proper mammary gland branching and milk production, through the expression of prolactin, p450 aromatase (a prolactin-regulated gene), and the estrogen and prolactin receptors.^24,29-33 These findings raise the intriguing possibility that, in response to endocrine prolactin (and possibly estrogen), the adipose initiates a paracrine-acting positive feedback loop of prolactin and estrogen secretion to increase the local concentration of lactogenic hormones in the mammary gland microenvironment, thus potentiating alveolar development. Indeed, expression of the estrogen receptor in the fat pad is necessary for proper mammary gland branching morphogenesis,^30,31 suggesting that the ability of the adipose to both respond to and secrete estrogen is necessary for epithelial glandular function. As such, the adipose tissue, long regarded as a passive lipid storage scaffold, is anything but static.

In addition to adipocytes, it is clear that cells of the immune system have pleiotropic roles in mammary gland development. Through an elegant combination of mouse genetics, sub-lethal irradiation and bone marrow transplantation experiments, it was shown that macrophages are recruited to the mammary stroma near branching ducts in response to the CSF-1 chemoattractant, secrete EGF to promote epithelial cell proliferation and chemotaxis, and are necessary for proper ductal outgrowth, terminal end bud geometry and side branching.^34,35 Macrophages are thought to clear apoptosed epithelial cells from the ducts during lumen formation, but additionally have been shown to regulate the alignment of fibrillar type I collagen near the terminal end buds.³⁶

Other cells comprising the innate immune system have been implicated in mammary gland branching morphogenesis—for example, eosinophils are recruited to the developing mammary gland end buds in response to eotaxin, and similar to macrophages, are thought to regulate ductal side branching.³⁴ Mast cells coordinate epithelial cell proliferation and mammary gland branching morphogenesis independent of macrophage recruitment and collagen deposition via a mechanism requiring mast cell degranulation and release of serine proteases, though the specific targets of mast cell proteases are not yet described.³⁷ Thus, several lineages of leukocytes appear to collaborate in defining the pattern of mammary gland branching architecture, and as will be discussed in subsequent sections, malignant progression.

Fibroblasts are another cell type present in the developing mammary gland stroma that, as the name implies, are thought to primarily function in the deposition of the collagen-rich ECM sheathing the branching ducts.^6,38 Similar to other stromal cells, fibroblasts secrete many factors acting on epithelial cells to promote branching and ductal outgrowth, and fibroblast conditioned media is sufficient to cause mammary epithelial cell branching in collagen-1 gels.³⁹ Among the best characterized of factors secreted by fibroblasts is HGF, a morphogen having profound influence on epithelial cell proliferation, migration, tubule formation, and ductal branching by activating PI3K signaling downstream of its receptor c-Met.^40-45 Using a combination of genetics and the classical DeOme transplantation of mammary epithelial tissue into the cleared fat pad,⁴⁶ it was shown that FGF10, a growth factor known to be secreted by fibroblasts, is necessary for mammary gland placode formation during embryogenesis, suggesting that fibroblast and epithelial communication is important in very early (prenatal) mammary gland development.⁴⁷ Fibroblast-derived FGF2 (also called basic-FGF) promotes mammary epithelial cell branching in collagen gels, but is also known to promote angiogenesis, a process important for establishing nutrient and oxygen supply not only for the normal mammary gland, but additionally within the tumor microenvironment.^39,48,49 The metastasis-linked protein S100A4 (FSP1) is secreted from fibroblasts during ductal elongation and pregnancy, and induces MMP3 activity necessary for the branching of mammary epithelial organoids ex vivo,⁵⁰ providing another example of how events governing tumor progression often have correlative biology in normal development. While in vivo evidence is currently lacking owing to the infrequency of fibroblasts in the stroma relative to adipocytes and the lack of genetic studies, it would be no surprise if other stromal derived growth factors and morphogens are additionally provided by fibroblasts during development—indeed, many of these factors are known to be secreted by fibroblasts within the breast tumor microenvironment.⁵¹

Perhaps the most imperative role for fibroblasts in postnatal mammary gland development is in their shaping of global mammary gland architecture through deposition and remodeling of the collagen-enriched ECM comprising the majority of the human breast stroma; indeed, collagen is the most abundant protein in humans and thought to comprise approximately 30% of our proteome.⁵² The intra- and inter-lobular ECM is comprised mostly of fibrillar collagen and fibronectin, but additionally contains several glycoproteins such as tenascins, osteonectin and decorin, all of which are important regulators of mammary gland architecture.⁶ Collagen and fibronectin are responsible for maintaining tissue architecture by controlling the mechanical strength and organization of tissues, and by regulating cellular adhesion, respectively. In seminal studies, it was shown that tissue mechanics have a profound influence on the developmental fate of mesenchymal stem cells⁵³ and on maintaining breast epithelial cell polarity,⁵⁴ underscoring the importance of the fibrillar collagen ECM in normal breast biology. Collagen fibril formation is an elaborate process involving several post-translational modifications, and upon deposition, collagen fibers can be further crosslinked, aligned or cleaved by lysyl oxidases, transglutaminases or MMPs, respectively, to fine-tune the tissue-specific mechanical properties.⁵⁵

The collagen-rich ECM provides the architecture supporting gland development, yet it also represents a physical barrier that must be overcome by the branching ducts during stromal invasion—this is accomplished through coordinated expression of several MMPs and TIMPs in response to microenvironmental cues, and these events are necessary for proper gland architecture.^20,39,56-60 It is clear that the spatiotemporal regulation of tissue mechanics through collagen remodeling and crosslinking enzymes is necessary to maintain gland homeostasis and can go awry in cancer.^60-62 However, it is unclear whether the activity of some of the modifiers, such as members of the lysyl oxidase family, is necessary for normal mammary gland branching or involution—we eagerly await such a study.

There are several lines of evidence that the stromal ECM through which the branching epithelium invades is not sufficient for normal mammary gland functional differentiation. Purified luminal epithelial cells cultured in fibrillar (type I) collagen gels alone do not properly polarize and instead have “inside-out” morphology upon staining with polarity markers—co-culturing luminal epithelial cells with myoepithelial cells in collagen gels restores their correct apicobasal polarity.⁶³ Additionally, tenascins present within the stromal ECM interfere with cellular adhesion to fibronectin through competitive binding to fibronectin receptors,⁶⁴ and it was further shown that tenascin C prevents mammary epithelial cell differentiation and milk protein synthesis.⁶⁵ So how does one reconcile the prevalence of the stromal ECM with its demonstrated inability to encode mammary epithelial cell functional differentiation—if anything, the stromal ECM inhibits functional differentiation!

It turns out that epithelial ducts are not in direct contact with the stromal ECM in vivo, and rather are in contact with the basement membrane—a specialized form of ECM that segregates the epithelial ducts from the stromal compartment and provides a myriad of biochemical and biophysical cues to allow mammary gland functional differentiation.^66,67 The basement membrane contains proteins and glycoproteins that are very different from the stromal ECM, and is comprised mostly of laminins, network-forming (non-fibrillar) type IV collagen, entactin/nidogen, proteoglycans and other glycoproteins.^6,67 In contrast to the stromal-associated ECM deposited primarily by fibroblasts, the basement membrane is deposited primarily by luminal and myoepithelial cells.^63,68,69 Expression of laminin-322 (previously known as laminin-5), collagen IV and their receptor, integrin α2β1, peaks during branching morphogenesis and their signaling is necessary for proper branching architecture; genetic knockout of either the α2 or the β1 integrin subunits impairs branching morphogenesis and alveologenesis during pregnancy.^70-73 Similarly, inhibiting deposition of collagen IV, a basement membrane component central to supporting alveolar architecture, leads to involution-like collapsed ductal architecture and decreased side branching.⁷⁴

Of the basement membrane proteins, laminins are thought to be most critical for encoding tissue-specific differentiation. Laminins are high molecular weight heterotrimeric proteins that are covalently assembled from single α, β and γ subunits (of which there are 5, 3 and 3 genes, respectively) in tissue-specific complexes. Culturing primary mammary epithelial cells or nonmalignant human breast cells in laminin-rich ECM (lrECM), commercially available as Matrigel or Cultrex, yields acinar-like colonies with apicobasal polarity, a central lumen, and geometry similar to acini in lobules in vivo—if supplied with lactogenic hormones, primary mouse mammary epithelial cells can even synthesize milk proteins when cultured in this microenvironment, a result not observed if cells are cultured on “attached” collagen gels or on plastic.^69,75-79 The pleiotropic mechanisms by which laminins, particular laminin-111, encode tissue-specific functional differentiation in the mammary gland are reasonably well-elucidated, and involve targeted gene expression via ECM-responsive transcription factors, cytoskeletal and nuclear scaffold reorganization, suppression of proliferative signaling, establishment of polarity and other events that are reviewed in further detail elsewhere.⁶⁷

In summary, multiple cell types within the epithelial and stromal compartment of the mammary gland participate in extensive heterotypic communication with neighboring cells and their underlying ECM during development. The sum of these complex microenvironmental interactions—the developmental “program”—drives the coordinated epithelial cell proliferation and collective migration that shapes the mammary architecture, and later, polarizes cells into functionally differentiated units. Whereas we have limited this discussion to the mammary gland development, it is most likely that similar biology is found in the development of other organs. The mammary gland provides a robust and convenient avenue to allow such studies because of its unique post-natal development and maturation. We contend that many of the processes necessary for normal mammary gland biology are deregulated or exploited during breast cancer progression (Fig. 1 and Table 1), thus having a full understanding of normal breast biology may be necessary to having a better understanding of the complexity of breast cancer. It is clear that dynamic interactions between breast tumors and their associated stroma are required for cancer progression, and that in many ways a tumor “hacks” into the normal developmental program for personal gain—a subject we discuss next.

Table 1. Breast cancer invasion and normal mammary gland branching share many similarities.

	Branching morphogenesis		Breast tumor invasion
Similarities	Requires heterotypic, paracrine interactions between epithelial and stromal cells
	Stromal cells secrete growth factors guiding epithelial cell proliferation and directional migration
	Proliferative epithelial cells secrete growth factors to enrich their adjacent stroma
	Conserved growth factors and receptors used, albeit at different amplitudes of signaling
	Epithelial cells must invade and traverse the stroma
	ECM destruction, alignment, and remodeling as epithelium invades the stroma
	Liberation of sequestered growth factors and cryptic fragments drive angiogenesis and chemotaxis
Differences	Regulated proliferation and migration	Aberrant mitogenic signaling and network substructure
	Transient stromal-epithelial interactions	Exaggerated stromal-epithelial interactions
	Migration confined by fat pad	Migration infiltrates adjacent tissues and beyond
	Epithelial cell apoptosis yields a lumen	Epithelial cell necrosis yields inflammation
	Deposition of basement membrane	Destruction of basement membrane
	Differentiation into polarized, bilayered ducts	Dedifferentiation; loss of polarity and myoepithelium
	End product is a functional tissue	End product is a pathological tissue

The Breast Tumor as a “Hacker”: Interactions between the Tumor Epithelium and the Microenvironment Drive Tumor Progression

Given the complexity of all organs including the normal mammary gland, it is not surprising that the biology of breast cancer and the influence of the breast tumor microenvironment on progression from ductal carcinoma in situ (DCIS) through metastatic disease is also elaborate and a subject of intense research effort. Although it is clear that mutational aberrations linked with breast cancer etiology likely originate in luminal epithelial cells or their progenitors, genomics approaches have revealed that all cell types present in the breast acquire altered patterns of gene expression during breast cancer progression.^80-82 Furthermore, using laser-capture microdissection and microarray analysis, it was shown recently that whereas approximately 5,900 genes are differentially expressed in the epithelium during the transition from normal to DCIS, only three epithelial genes had differential expression from DCIS to invasive carcinoma, suggesting the profound importance of the stroma in the development of both the normal mammary gland and of breast cancer.⁸³ Recent reviews have discussed the complexity of the breast cancer microenvironment and its influences on cancer progression.^38,84-86 Therefore, here we emphasize molecular underpinnings of tumor progression that mirror the biology of the normal mammary gland and further present evidence to suggest that much of the heterogeneity observed in breast cancer can be explained by intra- and inter-tumoral heterogeneity within the tumor and its elaborate microenvironment. In particular, many of the biological underpinnings necessary for the transition from DCIS to invasive carcinoma have parallel roles during normal mammary gland branching morphogenesis.³

Changes in the breast tumor microenvironment are often observed as early as the DCIS stage or even earlier (Fig. 1), where hyperactive mitogenic signaling in epithelial cells results in secretion of many chemokines causing accumulation of leukocytes, mesenchymal stem cells, endothelial cells, fibroblasts and myofibroblasts within the tumor microenvironment either through chemotaxis or through differentiation.^80,83,87-90 Once present, the stromal and epithelial cells participate in reciprocal and paracrine-acting signaling loops, stabilizing the elevated localization of macrophages, myofibroblasts and fibroblasts at the DCIS, which then remodel and condition the ECM and promote tumor cell proliferation.^83-85,91-93 The altered microenvironment can, depending on the individual tumor location, genetics and systematic influences, increase ECM stiffening, vascularity, breast tissue density and/or result in calcifications potentially to the level of detection by mammographic screening or MRI.^94-97 Indeed, the introduction of routine mammographic screening artificially elevated the “incidence” of DCIS, and arguably has led to its overtreatment.⁹⁸ Eventually, the cells within the DCIS breach the myoepithelial layer and the basement membrane, defining the moment when an invasive breast carcinoma is formed. As such, myoepithelial cells are thought to be “natural tumor suppressors” central to maintaining tissue polarity, a role that begins to deteriorate during DCIS and that is entirely lost in invasive breast carcinomas.^63,80,99-101

In addition to classical secretion of growth factors through the Golgi complex, tumor cells can further recruit and signal to stromal cells via non-classical secretion pathways such as through shedding of membrane vesicles (including exosomes) or though necrosis. Exosomes are cup-shaped nanovesicles that contain hundreds of proteins, including MMPs, inflammatory cytokines and activated growth factor receptors, and are thought to play an important role in both mammary gland function and in breast cancer progression to metastatic disease.^102,103 Necrosis is frequently observed in highly proliferative breast cancers and high grade DCIS (such as comedo-type) owing to the lack of sufficient oxygen and nutrients to support the population, and represents another mechanism of bulk protein transportation. In particular, necrosis is known to promote inflammation and angiogenesis within the tumor microenvironment, similar to “wound healing.”^85,104-107 As will be discussed in more detail later, inflammation and “wound healing” are observed in the normal mammary gland biology, particularly during involution.

Reciprocal, paracrine-acting, heterotypic cellular communication underlies both tumor progression and mammary gland branching morphogenesis. In response to signals from the tumor epithelium, mesenchymal stem cells and/or fibroblasts differentiate into myofibroblasts that begin to secrete SDF-1, VEGF, EGF, HGF, TGFβ and other growth factors to promote angiogenesis and tumor epithelial cell proliferation, motility, and invasion.^{51,91,108-110} Likewise, tumor-associated macrophages recruited to the tumor by CSF-1 secrete multiple growth factors that promote angiogenesis and lure cancer cells into the stroma to facilitate their eventual metastasis.^92,111 Through the expression of proteases including MMPs and uPA family members among others, tumor epithelial cells eventually breach the basement membrane and enter the stroma.^112,113 Recruited immune cells and fibroblasts additionally express MMPs, cathepsins and other proteases that aid in this process, similar to the role of mast cells in normal branching morphogenesis.^{92,109,113,114} In fact, it must be again emphasized that all of these processes are observable during normal branching morphogenesis and are deregulated and exploited as tumors develop.

The skewed presence of immune cells and cancer-associated fibroblasts (CAFs) within the tumor stroma, and the stromal infiltration by renegade epithelial cells that are normally spatially segregated from the stroma by an intact basement membrane, results in tremendous remodeling of the stromal ECM analogous to perturbations observed during branching. The normal and tumor-associated ECM contains glycoproteins and proteoglycans—sticky proteins that can bind and sequester many growth factors, chemokines and cytokines that become liberated by the elevation of proteolytic activity within the tumor microenvironment, leading to signaling that promotes further tumor epithelial cell proliferation and migration, angiogenesis, lymphangiogenesis and inflammation.^{6,113,115-117} Similar storage of growth factors via sequestration is important for normal mammary gland development. In addition to the signaling illustrated above (albeit with regulated amplitude), HGF is thought to modulate mammary gland branching through a mechanism requiring interaction with heparan sulfate proteoglycans,⁴³ and TGFβ activation from a latent to active form depends on sequestration in the ECM and subsequent proteolysis.^117,118

ECM molecules themselves can contain “cryptic fragments” that are liberated upon proteolysis in both normal mammary gland and tumors.^116,119 A fragment in the gamma chain of laminin-322 (formerly called laminin-5), once cleaved by one of several MMPs predicted to target the region, can function as a epithelial cell chemoattractant by binding and activating signaling through EGFR—a surprising function that appears to be important during normal mammary gland pregnancy and involution.^120,121 Several unique cryptic fragments are documented in collagen IV, the liberation of which, depending on their balance, can positively or negatively enforce angiogenesis.¹¹⁶ Perlecan proteolysis additionally yields peptides that regulate endothelial cell adhesion and angiogenesis.¹²² All of these cryptic fragments ostensibly would be liberated during the progression from DCIS to invasive breast carcinoma, though it remains untested whether their presence would provide additional mitogenic signals above and beyond the potentially saturating levels of growth factors that accumulate in the tumor microenvironment during this time.

In addition to degrading normal tissue ECM, stromal cells of the tumor microenvironment, most notably CAFs, extensively deposit new ECM and align the existing stromal ECM into a tumor-associated ECM during cancer progression, which can have profound influence on the tissue architecture, mechanics, and biochemistry.^{83,109,123,124} A dramatic example of this is the extensive desmoplasia that often accompanies invasive carcinomas, which is comprised mainly of ECM, myofibroblasts and fibroblasts, and which is normally restricted to tissue regeneration during wound healing; in some cases the desmoplasia accounts for the majority of the tumor mass.^107,125,126 Collagen within the tumor microenvironment becomes aligned and mechanically stiffer than the normal breast stromal ECM during cancer progression through cross-linking enzymes such as lysyl oxidase, and in addition to altering signal transduction within cells by clustering integrins, is thought to provide “tracks” upon which stromal and epithelial cells can migrate to promote eventual metastasis.^61,127-129 Lysyl oxidase expression is regulated by HIF1-α, suggesting that hypoxia within tumors can signal to crosslink the ECM to drive metastasis; inhibiting lysyl oxidase eliminated hypoxia-induced metastasis in vivo.¹³⁰ Additional molecules present in the tumor-associated ECM, such as fibronectin and tenascins, additionally support tumor cell motility through the stroma and correlate with poor clinical outcome.¹³¹ In contrast to collagen, which undergoes structural rearrangements post-secretion into the tumor microenvironment, it seems that the tumor-associated fibronectin and tenascins have different splicing isoforms than their normal breast counterparts that instruct alternative biology in the tumor microenvironment.^132-135 The cause and consequence of these alternative splicing events, and their context-specificity in breast cancer subtypes, remains to be fully characterized.

The extensive desmoplasia and inflammatory microenvironment within breast tumors has led to the notion that tumors are “wounds that do not heal.”¹⁰⁷ Indeed, classical experiments have shown that the wound response drives tumor formation in adult chickens previously infected with the Rous sarcoma virus in ovo, that wounding is both necessary and sufficient to induce tumors, and that wounding leads to tumorigenesis through TGFβ-mediated inflammation at the wound site.^136-139 In addition to demonstrating the importance of the normal tissue microenvironment in suppression of malignancy, these studies were the first to identify a pro-tumorigenic role for TGFβ, which was previously thought to exclusively function as an inhibitor of proliferation and branching morphogenesis. Induction of a wound-like microenvironmental response is observed during involution in the normal mammary gland, additionally involves signaling through the TGFβ pathway, and is linked with postpartum gland repopulation and ECM deposition.^140-142 In recent work, postpartum-associated cancer was mechanistically linked to inflammation via induction of COX-2 and collagen crosslinking; mice administered non-steroidal anti-inflammatory drugs (NSAIDs) during involution had a decrease in postpartum tumor formation, decreased collagen fibrillogenesis and decreased subsequent metastasis.¹⁴³ This is particularly groundbreaking research, as breast cancers diagnosed shortly following pregnancy are associated with poor clinical outcome and could potentially be combated through prophylactic NSAIDs.¹⁴² Furthermore, it has been previously shown that inflammatory breast cancer, the most aggressive of breast cancer subtypes,¹⁴⁴ occurs more frequently in pregnancy-associated breast cancers¹⁴⁵—it is enticing to speculate that inflammatory breast cancer and pregnancy-associated breast cancer share common origins, both of which could be intervened or even prevented by anti-inflammatory agents.

In summary, there is clear evidence that the tumor microenvironment elaborates (or even “hacks”) many biological programs of the developing mammary gland microenvironment, and like its normal counterpart, the breast tumor microenvironment remains incompletely understood. We advocate that understanding the “code” and “scripts” responsible for the normal developmental “program” is necessary before one can understand how a tumor can hack its vulnerabilities, as a tumor ultimately depends on normal biological processes to evolve and sustain itself. In addition to the influences of the microenvironment on cancer progression illustrated above, there is mounting evidence that the microenvironment can further direct the response of cancer cells to systemic therapy.^146-151 Indeed, several gene expression profiling studies have yielded biological “signatures” that are able to forecast clinical outcome and/or predict who may benefit most from chemotherapy, and a recurrent theme emerging in these studies is the importance of the tumor microenvironment. This will be briefly discussed in the final section of this review.

The “Security Vulnerability”: The Tumor Microenvironment Creates Heterogeneity in Breast Cancer Biology and Clinical Outcome as Evidenced by Molecular Profiling

Pathologists have known for decades that breast cancer cannot be described as a single disease, and heterogeneity based on tumor morphology, location, grade, lymph node metastasis, expression of hormone and growth factor receptors is well documented, and for several biomarkers, is routinely assessed in the clinic.^152-157 However, in addition to heterogeneity between tumors, it is becoming increasingly appreciated that a high degree of molecular and morphological heterogeneity exists even within tumors,^156,158,159 further complicating the development of therapeutic strategies and our understanding of disease progression. Through the use of gene expression profiling and other genomics approaches, the complexity and heterogeneity of breast cancer has been the subject of much research in the past decade, and emerging evidence has indicated that the tumor microenvironment plays a pivotal role in driving tumor heterogeneity.

The advent of whole genome expression profiling by microarray analysis has revealed that, in addition to clinical subtypes, there are additional “intrinsic subtypes” of breast cancer, including at least two estrogen receptor (ER)-positive “luminal” subtypes, and three ER-negative subtypes termed “basal-like,” “ERBB2-positive” and “normal-like.”^160-162 These intrinsic subtypes behave as distinct clinical entities with characteristic prognoses and responses to treatment.^162-164 Other tumor subtypes with gene expression profiles and clinical outcome suggestive of specific biology have been characterized in ER-negative breast tumors; these include the “claudin-low” subtype that is enriched with mesenchymal-like cancer stem cells deficient in cell-cell adhesion proteins^165,166 and the “molecular apocrine” subtype that has elevation of androgen receptor signaling,^167-169 suggesting the full extent of tumor heterogeneity remains to be seen. Furthermore, the intrinsic subtypes were originally identified using invasive breast cancers that were morphologically “not otherwise specified”—yet up to 10% of invasive breast cancers fall into one of 30 rare morphological subtypes (known as histological “special types”) having distinct molecular features.¹⁷⁰

There is emerging evidence suggesting that the biology of specific breast cancer intrinsic subtypes reflect significant contributions from the tissue microenvironment. For instance, the “ERBB2-positive” intrinsic subtype contains several genes that appear to be co-expressed with the ERBB2 gene—yet there is low correlation between patients having “ERBB2-positive” intrinsic subtype and the clinically-defined ERBB2-amplified breast cancer subtype (who are eligible to receive targeted therapy against ERBB2/HER2), suggesting that other microenvironmental signals (most likely through other EGFR family members) influence a set of genes associated with “ERBB2-ness” in the absence of gene amplification.^171,172 In fact, tumor samples separated from the breast stroma using laser-capture microdissection are markedly deficient in “ERBB2-positive” and “normal-like” intrinsic subtypes, and these subtypes are not observed in established breast cancer cell lines, indicating that these subtypes of ER-negative breast cancer may be molecular represented due to heterogeneity in stromal cells within the breast tumor microenvironment.^173,174 Another subtype with obvious ties to the microenvironment is the so-called “B-cell high/IL-8 low” subset of “triple-negative” (ER-, PR- and HER2-negative) breast tumors; this subset is characterized by enrichment of B-cell markers, sparse inflammation and a very good prognosis atypical of “triple-negative” tumors.¹⁷⁵ Thus, the tumor microenvironment can influence the heterogeneity, biology and outcome of breast cancer even among subtypes canonically thought to have the worst clinical outcome, which can be exquisitely demonstrated by molecular profiling. The continuing emergence of “subtypes within subtypes” of breast cancer does not lessen the validity of the tumor intrinsic subtypes—rather, it finetunes it and emphasizes that tumors are complex, heterogeneous and continuously evolving creatures that requires us to consider the biology not only of the tumor epithelial cells, but additionally their modified and elaborate stroma.

Other seminal microarray-based tumor profiling approaches have led to the identification of prognostic gene expression signatures; that is, sets of genes whose expression can forecast patient outcome independent of standard clinical parameters. These prognostic signatures include the 70-gene (MammaPrint) signature, the 76-gene signature and the Genomic Grade Index among others.^176-180 In an alternative strategy, changes in gene expression that accompany growth arrest and polarization during acinar morphogenesis were measured using microarrays; these genes were found to be prognostic of patient outcome in independent cohorts.^181,182 Although unique genes are represented in each signature, these prognostic signatures are highly concordant in their patient risk stratification performance due to tracking of similar biological pathways and processes, such as cell proliferation, mitogenic signaling, ECM remodeling, cytoskeletal dynamics and ECM deposition genes¹⁸³—all of which are regulated by stromal signaling within the tumor microenvironment (see earlier sections). Indeed, several of these cancer-associated biological pathways are also enriched in the mammary gland terminal end buds during normal branching morphogenesis, further highlighting the similarities between normal branching and tumor invasion as measured by gene expression profiling.¹⁶

Of these “first-generation” microarray-derived prognostic signatures, the 70-gene signature¹⁷⁶ is by far the most clinically successful, and contains multiple microenvironment-related genes such as FGF, TGFβ, MMP and collagen family members.¹⁸⁴ The prognostic performance of this set of genes has been validated in several independent cohorts worldwide,^185-190 and the genes were adapted into MammaPrint,¹⁹¹ the first multi-gene clinical diagnostic test to receive FDA approval. The MammaPrint assay is currently being evaluated in a 6,600 patient prospective, multicenter, randomized clinical trial called MINDACT (Microarray in node-negative and 1–3 positive lymph-node disease may avoid chemotherapy),^192,193 which will directly assess whether using MammaPrint in routine clinical practice will lead to more accurate predictions of who could be safely spared chemotherapy vs. standard clinical parameters.

Other molecular profiling studies have directly interrogated the influence of the tumor microenvironment on breast cancer outcome, and have provided additional insight into the underlying mechanisms governing epithelial-stromal interactions.^147,194-204 In one of the first gene expression profiling studies with relevance to the breast microenvironment, a molecular signature of serum-activated vs. quiescent fibroblasts was characterized and found to contain several genes involved in wound response.¹⁹⁴ This and independent “wound-healing” signatures were shown to be prognostic of breast cancer clinical outcome, ostensibly through molecular events originating from fibroblasts within the breast tumor microenvironment.^194,195,203 Recent work using gene expression of stromal cells microdissected from primary breast tumors has led to additional stromal-derived predictors of clinical outcome²⁰¹ and response to neoadjuvant (presurgical) chemotherapy,¹⁴⁷ further underscoring the importance of the microenvironment in tumor biology. Using a similar microdissection approach, it was shown that tumor epithelial cells acquire enrichment in motility and proliferation-related genes whereas stromal cells acquire a reactive phenotype during cancer progression,¹⁹⁷ recapitulating a theme central in the majority of gene expression prognostic signatures and one that is manifested as well during normal branching morphogenesis and involution (see earlier sections). Lastly, others have focused on characterizing the influence of ECM protein expression and organization on patient prognosis, and were able to identify ECM-related gene expression modules that correlate with tumor intrinsic subtypes and with clinical outcome.²⁰² Whereas these signatures are promising and clearly demonstrate a molecular basis by which the stromal microenvironment influences tumor behavior, they will require much further development and validation before incorporation into the clinic. Nevertheless, these studies provide evidence suggesting the microenvironment, beyond individual genes, is a master regulator of breast cancer progression, heterogeneity, and prognosis, and represents a “security vulnerability” that can be exploited by a tumor during malignant progression.

Conclusions

Through the development and astronomical improvements in DNA sequencing, comparative genome hybridization, gene expression microarrays and proteomic technologies, we now have the ability to measure the sequence, copy number and expression of every gene in a tissue at single unit (base, copy number, mRNA and codon/amino acid) resolution. However, despite these advances in technology, there is still much to learn about the nature of heterogeneity and development even in the normal breast. In addition to the comparisons we have presented in detail, there are further parallels between the biology of the normal and the pathological breast that we haven't discussed or where the details are still emerging, such as the epithelial stem/progenitor cell hierarchy—the differentiation of which is most likely encoded by cues from the tissue microenvironment.^205-208 From what we know, it is clear that the development of a functional breast or a (dysfunctional) breast tumor—the program and its “hacked” counterpart—requires extensive dialog between stromal and epithelial cells and relaying of biochemical and biophysical signals from the ECM. Many of the same pathways and interactions observed in the normal mammary gland during development are exploited by tumors during progression to metastatic disease. As is clear from the above discussion, we advocate that understanding normal breast biology is necessary to understand the pathological breast and there is still much work to do. After all, didn't this malicious program originate from something that used to be normal?

Glossary

Abbreviations:

CSF

colony-stimulating factor

CAF

cancer-associated fibroblast

DCIS

ductal carcinoma in-situ

(lr)ECM

(laminin-rich) extracellular matrix

EGF(R)

epidermal growth factor (receptor)

ER

estrogen receptor

FGF

fibroblast growth factor

HGF

hepatocyte growth factor

IGF

insulin-like growth factor

MINDACT

microarray in node-negative and 1-3 positive lymph-node disease may avoid chemotherapy

MMP

matrix metalloproteinase

MRI

magnetic resonance imaging

NSAIDs

non-steroidal anti-inflammatory drugs

PR

progesterone receptor

SDF

stromal cell-derived factor

TGF

transforming growth factor

TIMP

tissue inhibitor of metalloproteinases

TNF

tumor necrosis factor

VEGF

vascular endothelial growth factor

Disclosure of Potential Conflicts of Interest

L.J.van'tV. declares an employment/leadership role and has stock or other ownership interests at Agendia Inc. (Chief Research Officer).