The Paradigm for Human Breast Cancer Development

Most human breast cancers appear to evolve from normal epithelial cells in terminal duct lobular units (TDLUs) through a series of increasingly abnormal stages over long periods of time, probably decades in most cases (Fig. 1) [1–3]. The key stages in this progression, which is non-obligatory, are generically referred to as hyperplasias, atypical hyperplasias, in situ carcinomas, invasive carcinomas and, finally, metastatic disease. There are important general characteristics that distinguish one stage from the next that accumulate and increase with progression. The transition from normal cells to hyperplasias is characterized by increased growth due to epithelial hyperplasia. Hyperplasias are very common and may be due to delayed differentiation rather than genetic damage per se, creating fertile soil for the accumulation of random genetic mutations leading to progression to more advanced clonal precursors. Alterations of cell adhesion and polarity distinguish atypical hyperplasias from hyperplasias as the epithelium begins to pile up and distend acini. In situ carcinomas are characterized by further growth and the appearance of enormously increased histological and biological diversity compared to earlier precursors, which may be accelerated by acquiring mutations resulting in genetic instability. Invasion into surrounding stroma defines the transition from in situ to invasive carcinomas and genetic alterations in tumor epithelium appear to induce the participation of stromal cells in this process. Metastasis is characterized by spread of disease to other sites in the body. Identifying the biological alterations associated with early precursors before the development of substantial diversity may reveal effective strategies for the prevention of the majority of breast cancers, and some progress has already been made. For example, ERα is highly elevated in nearly all early precursors, and drugs targeting this receptor (e.g. tamoxifen) reduce breast cancer by 50% in recent chemoprevention trials of high risk women [4, 5]. Unfortunately, most defects responsible for the development and progression of premalignant breast disease remain unknown and, hopefully, future studies will shed light on these important issues. In this regard, mouse models of breast cancer represent one of the most important tools available to study breast cancer evolution. This chapter briefly reviews the general types of mouse models that have been developed over the years.

Figure 1
figure 1

The Wellings–Jensen model of human breast cancer evolution. The Wellings–Jensen model proposes that the cellular origin of human breast cancers occurs in the normal terminal duct lobular unit [TDLU] and that the putative precursors represent a non-obligatory series of increasingly abnormal stages which progress to cancer over long periods of time, probably decades in most cases. The key stages in the so-called ductal lineage [representing about 80% of all breast cancers] are referred to as columnar cell hyperplasia [CCH] or hyperplastic enlarged lobular units [HELUs], atypical ductal hyperplasia [ADH], ductal carcinoma in situ [DCIS], invasive ductal carcinoma [IDC], and metastatic [MET] carcinoma. There are several major types of histologically-defined invasive breast cancers [IBCs], including ductal, tubular, mucinous, medullary, and invasive lobular carcinoma [ILC]. All subtypes except ILC can be considered as subtypes of IDCs in the sense that they appear to evolve from the same ductal precursors. The evolution of the so-called lobular lineage [representing the remaining 20% of carcinomas] is histologically relatively distinct, and the key stages in this setting are referred to as atypical lobular hyperplasia [ALH], lobular carcinoma in situ [LCIS], and ILC. However, the practice of referring to breast cancers as ductal or lobular implies that they originate and reside in ducts and lobules, respectively, which is a historical misconception in the sense that both lineages are thought to arise from progenitor cells in normal TDLUs and both can occupy ducts and lobules. Furthermore, many IBCs show complex combinations of the various types of cancers, emphasizing that these histology-based classifications, while very useful, oversimplify enormous diversity.

Full size image

Mouse Models of Breast Cancer

Traditional Models

The mouse has played an important role in providing models to develop a basic understanding of the development of breast cancer. Over the past 75 years, there has been a progressive development of new models to study premalignant and malignant breast disease. The first era, roughly from 1920 to 1960, represented a time when inbred mice were developed and recognized as unique models to study the susceptibility to specific cancers, including breast cancers. Studies on a variety of inbred strains led to the hypothesis that certain genetic, hormonal, and environmental factors (e.g. MMTV, chemical carcinogens) played interactive roles determining susceptibility to breast cancer [6]. Of the many basic tenets learned from using these models, there were two important basic understandings that are pertinent for this discussion. First, both MMTV and chemical carcinogens induced premalignant lesions which were demonstrated by transplantation studies to progress to malignant breast cancers [6–8]. Second, the studies identified specific genes that were activated by insertional mutagenesis in the case of MMTV, or new mutations in the case of chemical carcinogens. The genes activated by insertional mutagenesis included wnts, fgfs and notch, [8], each of which are thought to play a role in the development of human breast cancer. New genes are still being identified by this approach. Genes mutated by chemical carcinogens included the ras gene family [8]. Although mutated ras is seen at only a very low frequency in human breast cancer, deregulation of ras activity is seen in a high percentage of human breast cancers [9].

Transgenic Models

With the exploration of the molecular basis for cancer development, the second era of model building, roughly from 1980–1995, generated models based on over-expression of specific oncogenes targeted to the mammary gland or deletion of specific tumor suppressor genes in a germ line specific manner. Over 100 such models have been constructed and described with genes that include c-myc, p53, v-ras, TGFa, wnt-1, cyclin D1, polyoma mT, c-neu, SV40LT, and IGF-1, among others [10]. These models have been used to evaluate not only the pathobiology, signal transduction pathways, and gene expression signatures, but also as a means to test the efficacy of specific chemoprevention strategies [11]. The value of such models lies in their ability to examine the effects of single genes and genes in combination on the development of mammary tumors. However, such models are not without limitations. In particular, the extent of gene expression is difficult to control as multiple gene copies can be inserted into the genome. Similarly, deletion of tumor suppressor genes was originally not targeted to the mammary gland thus there were often lethal effects on the fetus or neonate. However, the information gained on the importance of specific genes that play a role either in early or late stages of breast cancer development was enormous and led to an appreciation of the numerous pathways that can be independently disrupted to facilitate tumorigenesis.

Conditional Models

The ability to activate (e.g. c-myc, c-neu) or delete (e.g. brca-1, p53) specific genes in the mammary gland at a specific time in development was a great advance in modeling which represents the third era of model building and started around 1997 [12–14]. The advantage of such models lay in the ability to activate genes at any stage of development [e.g. post pubertal, mature, pregnancy, parous] and deactivate the gene of interest at a later stage. Thus, such models provided information on the stage specificity of oncogene action, the oncogene addiction hypothesis, and the identification of secondary events that enabled the altered cells to become independent of the original initiation stimulus [15]. One limitation of conditional mouse models is that the penetrance of the gene alteration is usually very high so that the vast majority of the mammary epithelial cells are affected. This is in marked contrast to what is thought to happen in the human situation where only a few cells in the breast epithelial population are affected by the initiating gene alteration. It would be ideal to specifically target a low percentage of epithelial cells with an activated gene and examine the consequences. More importantly, if one could target subsets of the mammary epithelial cell population (i.e., the stem cell, the differentiated cell, the ductal progenitor, etc], then one could examine the relationship between cell phenotype and tumor phenotype.

A New Model

In 2007, a new approach to targeting specific genes to the mammary gland was published [16]. This approach uses the avian leukosis virus receptor A (TVA) targeted to the mammary gland as a transgene. The replication competent ALV LTR serves as a cloning vector for the virus (RCAS) and exogenous genes that are expressed behind the viral env are expressed upon productive infection of cells carrying the TVA. RCAS vectors are introduced efficiently by intraductal injection into post-pubertal mice. Although the frequency of infected cells is low (<1%), the mammary tumor incidence is high and with relatively rapid onset. Oncogenes such as Polyoma mT, neu, c-myc and wnt-1 have been used in this model. One of the major advantages of the model is that it mimics the condition of an initiated cell emerging in a field of normal cells. Additionally, if the vector can be modified to infect specific subpopulations, then the phenotypes emerging from specific cell subtypes can be accurately traced.

What have we learned from mouse models in the past decade?

Since the last review of mouse models in the Journal of Mammary Gland Biology and Neoplasia in 2000 [17], multiple studies have made important contributions to our understanding of breast cancer development, and many were based on new models that mimic certain important aspects of human breast cancer that were missing in earlier models.

Hormone-Dependent Tumors

The vast majority of premalignant hyperplasia and mammary tumors arising in most mouse models are either estrogen-receptor negative or ovarian hormone-independent [8]. This is in contrast to similar lesions appearing in human breast where at least 70–80% of breast cancers are ER-positive and at least 50% are responsive to hormonal therapies. There are several mouse models where hyperplastic precursor lesions and their resulting tumors are ER-positive and hormone-responsive. The earliest model is the hormone dependent tumor arising in mice that carry the GR variant of the MMTV. These lesions are pregnancy dependent tumors that often regress after pregnancy [18, 19]. Since these lesions are rare in human breast literature, research on these lesions has not progressed over the past 30 years. The second model is based on the treatment of conventional BALB/c mice with the progestin methyoxyprogesterone acetate (MPA). The premalignant lesions are ER-positive ductal hyperplasias which progress to ER-positive breast cancers[20, 21]. This intriguing model has been pursued in a limited fashion but deserves more recognition. The importance of MPA as a factor which increases risk in human breast cancer make it especially relevant to examine the biological and molecular characteristics of the lesions that arise in this setting. Considering the large number of postmenopausal women who have used MPA as part of hormone replacement therapy, such a model could provide critical and immediately translational information with respect to prevention modalities. Stable cell lines have been developed from these hormone-dependent tumors [22]. The third and newest model is the p53 tumor suppressor gene model which exists in several forms, including a germ line null [23], as a conditional knockout [14], and as specific mutant transgenic [24]. The model is characterized by cellular aneuploidy and ER-positivity in the hyperplastic stages [23], a significant incidence of ER-positive tumors [14, 23], and a strong response to either anti-progesterone or tamoxifen-mediated chemoprevention [25, 26]. In addition, the mammary tumors that arise show a remarkable diversity of gene expression profiles that closely mimics that seen in primary human breast cancers [27]. In many ways, this model mimics significant subsets of human breast cancers and is a valuable model to test the effects of chemopreventive agents on the development and progression of premalignant breast disease. Several new mouse models are based on the over-expression of ER alpha. Over-expression of ER alpha alone results in ER-positive hyperplasia that resemble ductal hyperplasia in the human [28]. If ER alpha over-expression is imposed upon a background of BRCA1 loss and p53 heterozygosity, then a significant percentage of the mammary preneoplasias and cancers are ER-positive [29]. If ER alpha over-expression is combined with Tag over-expression, then all of the mammary cancers are ER-positive [30].

Genomic Instability

A high percentage of human breast cancers are aneuploid or show other evidence of genetic instability [31–33]. This instability can manifest as loss or gain of entire chromosomes, loss or gain of portions of chromosomes, translocations, and mutations. Most mouse models do not mimic faithfully these chromosomal changes. Exceptions are few but one of the models that does exhibit chromosome instability and is well characterized is the p53 null model [34, 35]. One of the mechanisms driving this chromosome instability is the p53 regulation of certain genes (e.g. MAD2, Separase) involved in sister chromatid exchange [35]. Interestingly, Separase RNA is over-expressed in human breast cancers compared to normal breast cells (Zhang, et al. submitted). A second model is the BRCA-1 heterozygote which is highly tumorigenic and exhibits strong evidence of genetic instability [36]. The mechanisms driving genetic instability are believed to involve altered DNA repair due to low BRCA-1 activity. A third model of interest, albeit not a mouse model, is that of the estradiol-driven carcinogenesis in the ACI rat [37]. Chemical carcinogen-induced mammary tumors in the rat are ER-positive but generally euploid and rarely metastatic [37]. The estradiol-treated ACI rat overcomes these limitations and exhibits a histopathology that includes ductal carcinoma in situ (DCIS), tumors that are metastatic and aneuploid. The mechanisms underlying the development of aneuploidy in this model are not completely understood. Other mouse models where genetic instability has been described include the Polyoma mT transgenic mammary gland [38], overexpression of luteinizing hormone resulting in ovarian hyperstimulation [39], overexpression of the prolyl isomerase Pin1 [40], and mutation of the Bloom Syndrome gene (BLM) [41].

Histopathology

The histopathology of breast cancer in different models is discussed in detail in chapter [8] of this issue [8]. Alveolar hyperplasias generated in traditional mouse models are histologically similar to so-called columnar cell hyperplasias (CCHs), also referred to as hyperplastic enlarge lobular units (HELUs) in humans, which represent very early potential precursors of breast cancer. In addition, some of these progress to lesions resembling atypical ductal hyperplasia (ADH) in humans. The most important contribution of the new transgenic models is that several exhibit a histopathology of premalignant lesions that shows close similarities to that seen in human breast disease. DCIS is a predominant lesion in the p53 null [42], the c-neu [43], polyoma mT [44], the SV40LT [45] and the IGF-1 models [46]. Also, unique hyperplasias that resemble ADH are present in the ER-transgenic model [28]. The DCIS lesions that arise in the p53 null and polyoma mT models have been shown by transplantation methods to be tumorigenic, give rise to metastatic lesions and are responsive to anti-hormone therapy [42, 44].

Oncogene Addiction

Until the existence of conditional transgenic models, it was impossible to examine the importance of the continued presence of the initiation stimulus. With the ability to turn off the expression of an oncogene, one could test the importance of the oncogenic stimulus for continued existence of the neoplastic phenotype. As expected, the answer is complex. In the models of c-myc, c-neu and wnt-1, the expression of these genes results in rapid tumor development. However, if one turns off gene expression by withdrawing doxcycyline from the drinking water, the tumors exhibit different fates. In many cases, the tumors will regress demonstrating that the maintenance of the tumor phenotype is dependent on continued function of the oncogene. However, a significant proportion of the tumors do not regress. For instance, in c-myc mediated tumorigenesis, removal of c-myc expression results in only 50% of the tumors regressing. The remaining 50% continue to grow and are characterized by the presence of activated Kras2 mutation [12, 47]. In contrast, decreasing wnt-1 expression in tumors results in virtually all of the tumors regressing. However, tumors do reappear in the absence of wnt-1 signaling if they acquire loss of a p53 allele or p19ink4a allele [48, 49]. Of interest, the tumors that reappear exhibit genetic instability as well as an increase in epithelial-mesenchymal transition (EMT). In c-neu induced mammary tumors, 94% of the primary and metastatic tumors regressed following down-regulation of c-neu [50]. However, if the animals were maintained for a further 5–10 months, a significant number of mice (7/11) demonstrated tumor reoccurrence. The recurrent tumors did not exhibit c-neu signaling, indicating an independent secondary event was responsible for the tumor recurrence. The significance of these experiments for clinical breast cancer is that secondary events are often dominant over the initial oncogenic events and represent new escape pathways from targeted therapeutics.

Epithelial-Mesenchymal Transition

EMT has gained substantial attention recently because of the apparent importance of this process for tumor invasion, motility, and metastasis [51, 52]. Classically, EMT refers to the process in embryological development where epithelial cells lose some of their distinctive properties and assume the appearance and behavior of mesenchymal cells. This process is essential for critical developmental stages such as gastrulation, neural crest formation, and heart valve formation, as well as other tissue forming events [53]. The molecular mechanisms are multiple and involve extracellular signals, soluble growth factors, specific transcription factors, and intracellular effector molecules. The ultimate result is the disassembly of junctional complexes, alteration of cell shape, and active motility of the epithelial cell. These events have been linked to the acquisition of invasive potential in DCIS and subsequent spreading outside the mammary duct into the underlying stroma. The conditional c-neu mouse model described above is a classic example of activation of EMT. Tumors that recurred after c-neu was down-regulated by withdrawal of doxycycline had a mesenchymal phenotype, down-regulation of E-cadherin and cytokeratin 8 and up-regulation of S100A4 (a fibroblast specific protein). Snail, a transcription factor that is a primary inducer of EMT, was highly up-regulated in the recurrent tumors. Subsequent experiments demonstrated that over-expression of Snail in primary c-neu-induced tumors generated the same EMT phenotype. The importance of Snail expression in human breast cancers is revealed by examination of four different data sets that show high Snail expression levels predict decreased relapse-free survival [54].

The role of EMT in human breast cancer is somewhat controversial because pathologists rarely observe histological evidence of EMT; although it is common in many different models of mouse mammary cancer [51]. One possible explanation for this paradox is that EMT may be transitory in human breast cancer and briefly exerts its most important role in invasion from the duct into the stroma, but thereafter re-assumes an epithelial phenotype. Supporting this idea, the elegant studies of Condeelis demonstrate a transitory EMT phenotype as tumor cells are wandering in the stroma to a site of intravasation [55–57]. It is very difficult to detect structurally different cells at the border of a growing tumor or isolated wandering cells on a histological section. The studies by Chodosh and coworkers using his elegant conditional mouse models have shed light on this phenomenon and with other in vitro experimental approaches have emphasized the importance of EMT for understanding invasion and subsequent metastasis.

Chemoprevention

The new generation of mouse models that are based on specific molecular alterations to drive the development of mammary tumors have also provided powerful tools to test chemopreventive agents. Although there are always going to be species dependent pharmacological differences between mouse and human for any chemical, the targeting of a chemical to a specific biological pathway in tumor models that mimic subsets of human breast cancer represents a powerful and clinically translatable assay. Several recent experiments illustrate the potential usefulness of these models. The c-neu, SV40LT, polyoma mT and p53 null models have been used to test the chemopreventive efficacy against the development of both premalignant hyperplasias as well as invasive tumors. Chemicals used have been tamoxifen [25, 58–60], retinoids [43, 61, 62], celecoxib [63, 64], and tyrosine kinase inhibitors [65, 66]. Different models exhibit different levels of susceptibilities to the different agents. For instance, tamoxifen is superior to the other agents for prevention of tumorigenesis in p53 null mammary cells whereas, in the c-neu model, rexinoids are the most effective. Although these agents can prevent the development of both premalignant hyperplasia and invasive tumors, established premalignant cells are less responsive to chemopreventive agents than the analogous normal mammary epithelial cell. Interestingly, this same conclusion on the stage specificity to chemopreventive agents was also drawn on experiments examining premalignant hyperplasias derived from traditional models of mammary cancer [67].

Gene Expression Profiling

Gene expression profiling has provided new insights into the subtypes of human breast cancer [68–74]. This same approach has been applied to a limited extent on the mammary tumors arising in different transgenic mouse models [75]. Depending on the dominant oncogene, the different mammary tumors fall into similar categories of luminal, basal, her-2, etc. Microarray studies of tumors arising in the p53 null model identified a group of uniformly altered genes which are also commonly altered in human breast cancers, including genes involved in transcriptional regulation, proliferation, apoptosis, cytokine signaling, cytoskeletal elements, and DNA repair, among others [27]. Similarly, a gene expression signature identified from tumors in the SV40LT antigen mouse model also identified a group of human breast cancers with particularly poor clinical outcome [76]. Thus, the value of these types of studies is that the gene signatures can identify potential novel targets for new therapeutics.

Promising New Models

A recent review has summarized the salient features of the majority of transgenic mouse models [10]. It is of interest to briefly discuss new models that are just gaining attention and that should provide important new insights into human breast cancer or. One such model is similar to human invasive lobular carcinoma [ILC) [77]. The incidence of ILC is about 10–15% of human breast cancer, and they are associated with certain unique molecular alterations such as an inactivation of E-cadherin. The mouse model of ILC was developed by inactivating E-cadherin upon a background of mammary epithelium-specific loss of p53 function. The result was a high incidence of multifocal mammary tumors histologically resembling human ILC, which are highly metastatic. Such a model provides an opportunity to not only decipher the mechanistic role of E-cadherin in ILC development but also provides identification of new therapeutic targets specific to ILC. IGF-1, its receptor, its binding partners, and downstream regulated molecules represent another pathway where recent models offers new insight into a gene family that is associated with a high risk for breast cancer. Overexpression of the IGF-1R results in DCIS lesions and invasive mammary cancers with a short latency [46]. Interestingly, in human breast cells, activated IGF-1R causes an EMT which was associated with induction of Snail, downregulation of E-cadherin and increased motility [78]. Similarly, overexpression of the insulin receptor substrates (IRS-1 and IRS-2) results in ductal hyperplasias, invasive cancers and metastasis [79]. One pathway activated by the IRS was beta-catenin, a component of the wnt-1 signaling pathway [79].

What is lacking in mouse models?

There are many “types” of human breast cancers, depending on how they are being evaluated. For example, based on traditional histology, pathologists recognize five major subtypes of invasive carcinomas [ductal, lobular, tubular, mucinous, and medullary), and a larger number of rare subtypes (e.g. adenocystic carcinomas, etc.). Based on certain gene expression profiles assessed by microarrays, there are five so-called intrinsic subtypes (normal-like, luminal A, luminal B, basal, erbB2+), and so on. Any given mouse model mimics at best only a small subset of the types of breast cancer occurring in humans. Although not all of the newer mouse models have been thoroughly evaluated, collectively the studies based on them have provided substantial new information about the development and progression of human breast cancer overall, as well as potential novel therapeutic approaches. However, there are still several aspects of human disease which are not particularly well represented in models. One such area is metastasis. Many mouse models of breast cancer show metastases, although the majority spread to the lungs through blood vessels. In contrast, human breast cancers commonly spread to bone and brain, in addition to lung, primarily through lymphatics. Are these differences intrinsic to species differences or is there a fundamental pathway that is not represented in mouse mammary tumors? There are several mouse mammary tumors that have been selected for metastasis to bone, suggesting that the molecular changes that dictate metastasis to sites other than lung are intrinsic to mouse mammary tumors [80–83]. Interestingly, these models were all derived from traditional carcinogen-induced mouse tumor models. The enigma of primarily lung metastases may be due to the predominant vascular route used by most mouse mammary tumors. If tumors could be engineered to escape via a lymphatic route, it is possible that bone and brain metastases would be more common in mouse models. However, the presence of three well defined models of metastasis to bone and elsewhere does provide the basis for unraveling the mechanistic basis for metastasis and the opportunity to develop specific therapies. A focused effort on developing bone and brain specific metastases from tumors arising in specific genetically engineered models, such as the BRCA-1 or triple negative or ER-positive would provide new and valuable tools.

A second area needing an innovative approach is the area of modeling human DCIS. DCIS is a common premalignant lesion of the human breast which requires surgery followed by radiation and sometimes hormonal therapy. It is highly diverse histologically and genetically, similar to invasive cancer [1]. Although DCIS is recognized in some in animal models, they are relatively limited in diversity. Our understanding of the progression of DCIS to invasive breast cancer would be greatly facilitated if human DCIS, in all of its diversity, could be grown as xenografts. Current approaches to developing such xenografts have been relatively unsuccessful, perhaps because the progression of the DCIS involves not only factors intrinsic to the transformed breast epithelium, but also changes in adjacent myoepithelial cells, basement membrane, and other stromal cells such as fibroblasts. This is one area of modeling that needs a radical re-invention of approaches.

Finally, there has been a revolution in our interest and understanding of the role of stromal-epithelial interactions in the evolution of mammary cancer. This recognition is based on many types of experiments. One of the initial and most important influences has been the pioneering experiments of Bissell and coworkers [84, 85]. These investigators have systematically examined and delineated the importance of the extracellular matrix factors for growth and differentiated function of normal mammary cells as well as the growth of cancer cells. Recently, they demonstrated that the growth of established human breast cancers was strongly influenced by defined stroma components not only under in vitro conditions but also as xenografts. The effect of the stromal microenvironment on the progression of premalignant mammary cells has been demonstrated using transplantation of premalignant cells into the mammary fat pad that had been treated prior to transplantation by irradiation [86]. This effect was correlated with decreased TGFB signaling in the stroma. The experiments of Kuperwasser and coworkers have focused on developing a xenograft model to grow human breast cells in the mammary fat pad of an immuno-compromised mouse [87]. One of the important elements to make this a viable approach was the recognition that the stroma of the mouse and the human are fundamentally different in the extent of fibroblasts and collagenous connective tissue. The co-injection of human fibroblasts and epithelial cells into the mammary fat pad resulted in enhanced takes and growth of the human breast epithelial cells. The elegant studies of Pollard and coworkers have highlighted the importance of the wandering mononuclear cells in the stroma [i.e., macrophages, mast cells, leukocytes) for the growth of normal mammary cells in development as well as for the growth of tumor cells [88–90]. The targeting of stromal components by modern genetic engineering approaches would provide a powerful approach to testing the functional role of specific stromal factors on the development and progression of mammary cancer [91]. The combination of genetically modified fibroblasts (either deleting or over-expressing specific genes of interest) in combination with transplantation of mammary epithelial cells (either of mouse or human origin] would provide new approaches to functionally testing the importance of factors in a tissue microenvironment that more accurately mimics what occur sin the human breast.

Summary

Mouse modeling of human breast cancer has developed tremendously over the past ten years such that current models replicate many essential features of the variety of disease found in the human breast. The new models provide for not only a deeper understanding of the fundamental events that mediate initiation, development and progression of the disease, but new opportunities to develop and test therapeutic strategies that are specific for a subtype of human breast cancer. There are additional questions to be solved and many more experiments to be done but to paraphrase the poet T.S. Eliot, our “exploring will bring us to the place where we started but we will know it for the first time.”