REVIEWS

BREAST CANCER METASTASIS: MARKERS AND MODELS

Britta Weigelt*, Johannes L. Peterse and Laura J. van t Veer*
Abstract | Breast cancer starts as a local disease, but it can metastasize to the lymph nodes and distant organs. At primary diagnosis, prognostic markers are used to assess whether the transition to systemic disease is likely to have occurred. The prevailing model of metastasis reflects this view it suggests that metastatic capacity is a late, acquired event in tumorigenesis. Others have proposed the idea that breast cancer is intrinsically a systemic disease. New molecular technologies, such as DNA microarrays, support the idea that metastatic capacity might be an inherent feature of breast tumours. These data have important implications for prognosis predicition and our understanding of metastasis.
Breast cancer is the most common malignant disease in Western women. In these patients, it is not the primary tumour, but its metastases at distant sites that are the main cause of death. Recently, the rates of metastasis and mortality in breast cancer patients have decreased as a result of early diagnosis by mammographic screening and the implementation of systemic ADJUVANT THERAPY. Adjuvant therapy can help eradicate breast tumour cells that might have already spread to distant sites by the time of diagnosis. In women with breast cancer who are younger than 50 years of age, chemotherapy increases their 15-year survival rate by 10%; in older women the increase is 3%1. However, chemotherapy has a wide range of acute and long-term side effects that substantially affect the patients quality of life2. As it is not possible to accurately predict the risk of metastasis development in individual patients, nowadays more than 80% of them receive adjuvant chemotherapy, although only approximately 40% of the patients relapse and ultimately die of metastatic breast cancer. Therefore, many women who would be cured by local treatment alone, which includes surgery and radiotherapy, will be over-treated and suffer the toxic side effects of chemotherapy needlessly. New PROGNOSTIC MARKERS are urgently needed to identify patients who are at the highest risk for developing metastases, which might enable oncologists to begin tailoring treatment strategies to individual patients. Gene-expression signatures of primary breast tumours might be one way to identify the patients who are most likely to develop metastatic cancer, and would therefore benefit from adjuvant therapy. In addition, gene-expression profiling of breast tumours might also help to identify new therapeutic targets. Improving our understanding of the molecular mechanisms of the metastatic process might also improve clinical management of the disease. According to the widely held model of metastasis, rare subpopulations of cells within the primary tumour acquire advantageous genetic alterations over time, which enable these cells to metastasize and form new solid tumours at distant sites3. Many studies have challenged this genetic-selection model of metastasis in the past46, but only the recent data obtained by gene-expression profiling of human breast carcinomas79 received broader attention. The DNA-microarray studies reported that primary breast tumours that developed metastases could be distinguished by their gene-expression profile from those that remained localized. The data imply that the metastatic capacity of poor-prognosis breast tumours might be acquired by mutations at much earlier stages of tumorigenesis than was previously assumed10. This review will describe the current state of the clinicopathological and molecular prognostic markers

ADJUVANT THERAPY

Cytotoxic chemotherapy and/or endocrine therapy after surgical removal and/or radiotherapy of the primary tumour. Adjuvant therapy is used to ensure that all microscopic disseminated cancer cells are destroyed.
PROGNOSTIC MARKER

A characteristic of a patient or tumour at the time of diagnosis that can be used to estimate the chance of the disease recurring in the absence of therapy.

*Division of Experimental Therapy and Division of Diagnostic Oncology, The Netherlands Cancer Institute, Amsterdam, The Netherlands. Correspondence to L.J.v.V. e-mail: l.vt.veer@nki.nl.
doi:10.1038/nrc1670

NATURE REVIEWS | C ANCER

VOLUME 5 | AUGUST 2005 | 591

2005 Nature Publishing Group

REVIEWS
have an aggressive disease and develop distant metastases within 3 years after the initial detection of the primary tumour. However, the manifestation of metastases at distant sites 10 years or more after the initial diagnosis is also not unusual11. Patients with breast cancer are therefore at risk of experiencing metastasis for their entire lifetime. The heterogeneous nature of breast cancer metastasis makes it difficult not only to define cure for this disease, but also to assess risk factors for metastasis. A wide range of histopathological subtypes of invasive breast cancer have been identified, of which the invasive ductal carcinomas, defined as a type of cancer not classified into any of the other categories of invasive mammary carcinoma, represent the largest group12 (TABLE 1; online supplementary information S1 (figure)). Although some of the morphologically distinct, special types of breast tumour, which represent 510% of all breast cancers, have certain favourable prognostic features, histological typing in general is only a weak prognostic marker of metastasis risk12. Once disseminated, metastases from carcinoma of the breast are formed in various organs. The common sites for metastatic spread are bone, lung and liver (reviewed in REF. 13) (FIG. 1).
Established prognostic markers

Summary Current prognostic criteria only poorly predict the metastasis risk for an individual breast cancer patient. Therefore, many women receive cytotoxic chemotherapy unnecessarily. Gene-expression signatures of human primary breast tumours predict more accurately than current prognostic criteria which patients are destined to relapse and ultimately die of metastatic breast cancer, and should therefore receive adjuvant therapy. New molecular insights challenge the traditional model of metastasis, and indicate that the metastatic capacity of breast tumours is an inherent feature, and not necessarily a late, acquired phenotype. Local breast cancer might have a non-metastatic, good-prognosis stem cell of origin; metastasizing systemic breast cancer might have a metastatic, poorprognosis stem cell of origin.

as well as the current thinking on the metastatic process, which is based on microarray profiling studies and model systems. Both for the clinical and model systems, two general concepts have been proposed: metastasis is either, first, an intrinsic capacity, or, second, an acquired feature, of primary tumour cells. In the clinic this can be described as breast cancer that develops either as local or as systemic disease. We propose that an integrative metastasis model, in which metastasis is an intrinsic feature of breast cancer, might best explain the clinical and experimental observations.
Clinical features of breast cancer metastasis

Breast cancer is a clinically heterogeneous disease. Approximately 1015% of patients with breast cancer
Table 1 | Histopathological types of invasive breast carcinoma
Histopathological type of invasive breast carcinoma Invasive ductal carcinoma, not otherwise specified Invasive lobular carcinoma Mixed type, lobular and ductal features Tubular/invasive cribriform carcinoma Mucinous carcinoma Medullary carcinoma Invasive papillary carcinoma Invasive micropapillary carcinoma Metaplastic carcinoma Adenoid cystic carcinoma Invasive aprocrine carcinoma Neuroendocrine carcinoma Secretory carcinoma Lipid-rich carcinoma Acinic-cell carcinoma Glycogen-rich, clear-cell carcinoma Sebaceaous carcinoma
Data from REFS 11,12.

Frequency 5080% 515% 45% 16% <5% 17% <12% <3% <5% 0.1% 0.34% 25% 0.010.15% <16% 7 cases 13% 4 cases

10-year survival rate 3550% 3550% 3550% 90100% 80100% 5090% Unknown Unknown Unknown Unknown Unknown Unknown Unknown Unknown Unknown Unknown Unknown

The risk of metastasis development increases with the presence of lymph-node metastasis, a largersized primary tumour and loss of histopathological differentiation (grade) 1418 , which are the established breast cancer prognostic markers TABLE 2. In patients with tumour-negative axillary lymph nodes, vessel invasion is an additional predictor for distant recurrence 19,20 TABLE 2 . Despite this, approximately one-third of women with breast tumours that have not spread to the lymph nodes develop distant metastases, and about one-third of patients with breast tumours that have spread to the lymph nodes remain free of distant metastases 10 years after local therapy16,21. Markers that can predict the site of metastasis are also scarce. It has been shown that oestrogen-receptor-positive breast tumours have a predilection to metastasize to bone22, whereas invasive lobular carcinomas recur with increased frequency in the gastrointestinal tract and ovaries23,24. Today, the traditional prognostic markers are able to confidently identify the group of approximately 30% of patients, who are most likely to have either a very favourable or a very poor outcome. For the remaining 70% of patients, of whom approximately 30% will still develop metastases25, new prognostic markers are needed to help identify low-risk and high-risk groups, to pinpoint those patients who are most likely to benefit from systemic adjuvant treatment.
Recent prognostic markers

Substantial efforts have been made to identify additional prognostic markers that characterize patients with breast cancer who are at the highest risk of

592 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer

REVIEWS
has gained clinical relevance due to the introduction of trastuzumab, a therapeutic monoclonal antibody that is directed against the receptor, and which helps prolong survival in patients with metastatic breast cancer29. Moreover, increasing evidence also indicated that ERBB2 might be a PREDICTIVE MARKER for response to adjuvant chemotherapy and endocrine therapy (reviewed in REFS 31,32). This might explain why the testing of newly diagnosed breast cancer specimens for ERBB2 status has achieved standard of practice status for the management of breast cancer. This is despite the prognostic value of ERBB2 for disease-free and overall survival in patients with lymph-node-positive breast cancer being specified as weak-to-moderate by the World Health Organization Classification of Tumours 12 . Clearly, additional well-controlled and well-designed studies with sufficient follow-up time must be conducted to adequately validate the prognostic significance of ERBB2. Detection of disseminated tumour cells. To establish a metastasis, tumour cells have to invade their surrounding host tissue, enter the circulatory blood stream, arrest in capillary beds of distant organs, invade the host tissue and proliferate. As small tumours of less than 2 mm in diameter already receive a vascular blood supply33, it is likely that cancer cells have spread throughout the body years before they are first detected. The development of an assay to detect these cells before the manifestation of distant metastases might therefore be useful for patient prognosis. The search for circulating tumour cells started in the late 1980s, and today both immunohistochemical staining and PCR-based approaches are available to detect disseminated tumour cells. These methods are based on the presence of breast epithelial markers, in peripheral blood, bone marrow and lymph nodes (reviewed in REF. 34). Owing to technical issues regarding the rarity of the disseminated cells and the background expression levels of these markers, only a few studies have been published that examine the relationship between the presence of circulating tumour cells in peripheral blood and patient outcome. Two clinical studies showed that the disseminated-tumour-cell load in peripheral blood is associated with shortened disease-free intervals and reduced overall survival in patients with early breast cancer35,36. Stathopoulou et al. tested the peripheral blood of 148 patients for cytokeratin-19 mRNA 35 , whereas Zach and colleagues tested 310 patients for mammaglobin mRNA36, both using a nested reverse transcriptase (RT)-PCR approach. Although both studies selected patients with operable breast cancer, the incidence of patients with positive tests varied. Stathopoulou et al. found 44 patients (30%) to be positive for cytokeratin-19 mRNA, of whom 19 developed a metastasis, whereas Zach et al. detected mammaglobin mRNA in only 5 patients (2%), who all developed distant metastases. Remarkably, the detection of putative circulating tumour cells predicts early recurrence at distant sites in patients with breast cancer; in the

70

Percentage of 2,050 cases

60 50 40 30 20 10 0

um

al Br ain

ar t

ey dn Ki

ra

in

nt es tin

Sp

nc

rit o

Pa

ric a

G as tro i

Pe

Organ

Figure 1 | Most common metastasis sites of breast cancer at autopsy. Primary breast cancer cells metastasize through the blood vessels to various distant organs, preferentially, to the lung, liver and bones. Patients frequently develop metastases at multiple sites. Data adapted from REF. 13.

metastasis development. To meet the requirements of a prognostic marker, the potential marker should be tested retrospectively in large patient cohorts with a long follow-up period. MULTIVARIATE ANALYSIS needs to be done in conjunction with established markers to assess its independent value. Subsequently, the findings should be validated by an independent group of researchers, and, ideally, a PROSPECTIVE STUDY should confirm the prognostic significance of the tested marker26,27. A large number of putative molecular prognostic markers have been reported in the literature, but only a few of these have so far fulfilled the above described requirements TABLE 2. Preliminary data indicate that nearly all these markers are only of prognostic benefit in certain subgroups, or are not independent of the established markers25. Many other markers are still under clinical investigation. So, what are the most promising recently developed prognostic markers, and what is their potential to accurately predict metastatic potential and patient outcome?
MULTIVARIATE ANALYSIS

A statistical test that examines more than two variables at the same time.
PROSPECTIVE STUDY

A study in which a selected group of patients is followed over time to determine differences in the rate at which disease develops in relation to the investigated factor. These factors might include drugs, procedures or diets.
PREDICTIVE MARKER

A characteristic of a patient that is associated with the response or lack of response to a particular therapy.

ERBB2. Among many biomarkers epidermal gorwth factor receptor 2 (ERBB2; also known as HER2/neu) has raised much attention as a possible prognostic marker. The human ERBB2 proto-oncogene encodes a transmembrane receptor with constitutive tyrosinekinase activity. ERBB2 is overexpressed due to gene amplification in 1530% of human breast cancers28,29. The prognostic value of ERBB2 was first claimed in 1987 REF. 28, and from then on it has been extensively studied. Ross and colleagues recently reviewed the published literature concerning ERBB2, including 81 studies and 27,161 patients30. Most studies reported that ERBB2 amplification or overexpression is associated with poor outcome in patients with axillarylymph-node metastases, but it is not associated with poor outcome in patients with tumour-negative lymph nodes30. However, the ERBB2 status of breast tumours

NATURE REVIEWS | C ANCER

Ad

re n

Pe

al

Th

gl

ry Ut er us

ng

ve r

ne

ds

re a

lee

yr oi

Sk

eu

rd iu

Bo

Lu

ne

He

an

Li

Pl

va

VOLUME 5 | AUGUST 2005 | 593

2005 Nature Publishing Group

REVIEWS

Table 2 | Breast cancer metastasis prognostic markers

Marker Tumour size Use in clinic Established Metastatic determinants Tumours under 2 cm in diameter have a low risk of metastasis; tumours of 25 cm have a high risk of metastasis; tumours over 5 cm have a very high risk of metastasis If there are no lymph-node metastases, the risk of metastasis is low; if lymph-node metastases are present, the risk of metastasis is high; the presence of over 4 lymph-node metastases is associated with very high metastasis risk Grade 1 tumours have a low risk of metastasis; grade 2 tumours have an intermediate risk of metastasis; grade 3 tumours have a high risk of metastasis The presence of tumour emboli in over 3 blood vessels is associated with metastasis High protein levels of uPA and PAI1 are associated with high metastasis risk Low steroid-receptor levels are associated with metastasis Details Independent prognosis marker References 1417

Axillary lymphnode status

Established

Related to tumour size

14,16,17

Histological grade

Established

Related to tumour size

14,16,18

Angioinvasion

Established in patients with lymph-nodenegative tumours Newly established marker Established for adjuvant therapy decision

In patients with lymph-nodenegative tumours Independent prognosis marker Short-term predictor of metastasis risk (5 years); related to histological grade In patients with lymph-nodepositive tumours Tested in patients with lymph-nodenegative tumours

19,20

uPA/PAI1 protein level Steroid-receptor expression

5560

14

ERBB2 gene amplification and protein expression Gene-expression profiling

Established for adjuvant therapy decision Currently being tested

ERBB2 amplification/overexpression is associated with metastasis A good signature of 70 genes is associated with low metastasis risk; a poor signature of 70 genes is associated with high metastasis risk

28,30,31

7,8

PAI1, plasminogen activator inhibitor 1; uPA, urokinase-type plasminogen activator.

MICROMETASTASIS

Micrometastases were originally defined as small occult metastases of less than 0.2 cm in diameter. Nowadays, the term also includes disseminated tumour cells that are present in peripheral blood, bone marrow or lymph nodes.

study by Stathopoulou and colleagues, it even does so independently of traditional prognostic indicators. Further clinical studies with standard techniques in clearly defined patient populations will be needed to establish the clinical significance of circulating breast tumour cells in peripheral blood. However, not all studies have supported the prognostic value of detecting circulating epithelial cells in blood37, which might be attributed to the low sensitivity of the immunohistochemical methods that have been used. As opposed to the analysis of peripheral blood, there are a significant number of studies that aim to define the prognostic value of breast cancer cells that are detected in bone-marrow aspirates. Bone marrow represents a relevant site of breast cancer metastasis, and epithelial cells are normally not present in this location, which enables their immunohistochemical discrimination by antibodies against epithelial proteins, mainly against cytokeratins. Many studies have demonstrated a correlation between the presence of epithelial cells in bonemarrow aspirates and reduced disease-free intervals and overall survival3841. By contrast, several studies did not confirm the bone-marrow status to be an independent

prognostic indicator4244. A meta-analysis of 20 published studies that encompassed 2,494 patients reported that the prognostic impact of detected epithelial cells in bone marrow remains to be substantiated45. Large multicentre trials are therefore required to determine the validity of this approach, involving standardized detection and quantification procedures. The contradictory findings regarding the prognostic potential of disseminated tumour cells might be due to the fact that <0.1% of the cancer cells that have entered the blood circulation are able to establish a metastatic lesion46. This implies that most of the MICROMETASTATIC cytokeratin-positive cells detected in the bone marrow are not capable of forming metastases at distant sites, as supported by the large number of these cells found to harbour only a poor proliferative potential47. These data are in line with the idea that only a few cancer cells actually harbour tumour-initiating capacity and could be considered as breast cancer stem cells48. Interestingly, cytokeratin-19, which is frequently used as a breast epithelial marker in immunocytochemical and molecular assays (reviewed in REF. 34), was found to be a putative marker of stem cells in the breast 49,50.

594 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer

REVIEWS
Plasminogen activator uPA and its inhibitor PAI1. The early steps of the metastatic cascade involve the degradation of the extracellular matrix (ECM) and subsequent invasion of the surrounding host tissue by cancer cells. This degradation is accomplished by several enzyme systems, including, among others, the matrix metalloproteinases (MMPs) and the urokinase-type plasminogen activator (uPA) system51. This system consists of the serine protease uPA, its glycolipid-anchored receptor uPAR, and its two serpin inhibitors, plasminogen activator inhibitor 1 (PAI1) and plasminogen activator inhibitor 2 (PAI2). uPA converts plasminogen to plasmin, which degrades matrix components and activates latent metalloproteinases and latent growth factors51. Interestingly, PAI1 is not produced by the epithelial cancer cell but by the stromal cells in the tumours52. This indicates that stroma and tumour cells have coordinated effects on the processes controlling proteolysis in cancer. Duffy et al. were the first to show that patients with primary breast carcinomas with high levels of uPA activity had a significantly shorter disease-free interval than patients with low levels of activity. In addition to uPA, increased levels of its inhibitor PAI1 were paradoxically also reported to be a prognostic marker in both patients with node-positive and nodenegative breast tumours53. However, independent of its protease-inhibitory capacity, PAI1 also has a role in cell migration51 and promotes tumour invasion and angiogenesis54. A large number of studies confirmed PAI1 as well as uPA to be independent prognostic markers of disease-free survival and overall survival in breast cancer patients5558. The combined assessment of both markers, uPA and PAI1, more accurately predicts metastasis risk and patient survival time than either marker alone, or the established prognostic markers58. Remarkably, and in contrast to the investigations regarding disseminated tumour cells, there have been no studies published that discredit the association between uPA or PAI1 levels with breast cancer metastasis. These markers therefore appear to be reliable prognostic indicators. The levels of uPA and PAI1 in protein extracts of primary breast tumour tissue are determined by an ENZYMELINKED IMMUNOSORBENT ASSAY. A pooled analysis by the European Organization for Research and Treatment of Cancer (EORTC) that involved 8,377 breast cancer patients and had a median follow-up of 79 months showed that, apart from lymph-node status, high levels of uPA and PAI1 were the strongest prognostic markers for diseasefree survival and overall survival59. In patients with lymph-node-negative tumours, the levels of uPA and PAI1 were the strongest predictor of metastasis59. In addition, the two markers might also be used to predict response to therapy, in particular the response to adjuvant chemotherapy 60. So far, uPA/PAI1 are the only recently developed markers that have true prognostic use for patients with breast cancer, according to the Tumour Marker Utility Grading System26, a system that evaluates the clinical value of tumour markers and establishes an investigational agenda for evaluation of new tumour markers. However, these markers have not yet seen wide application in clinical practice. Gene-expression profiling of breast cancer. In search of new prognostic markers that predict metastasis risk in patients with breast cancer, most studies have examined the correlation between only one or a few markers and clinical outcome TABLE 3. Considering the heterogeneity of the disease, prediction of the metastatic potential of a tumour might require the analysis of many different markers at once. This is made possible by the introduction of DNA-microarray technology, which can analyse gene expression in a genome-wide fashion BOX 1. The first key finding in breast cancer using the DNA-microarray technology and an unsupervised analysis was the gene-expression-pattern-based classification of breast tumours into four previously unrecognized subtypes61. Three biologically distinct subgroups of oestrogen-receptor-negative breast carcinomas have been identified: the basal-like group, which expresses cytokeratin-5 and cytokeratin-17; the ERBB2+ group, which expresses several genes in the ERBB2 amplicon including ERBB2 and the gene encoding growth-factor-receptor-bound protein 7; and the normal-breast-like group, which expresses genes of adipose-cell and other non-epithelial-cell origin. The oestrogen-receptor-positive tumours that were originally found to be one group61 have latterly been separated into at least two distinct groups: the luminal A subtype, which expresses high levels of cytokeratin-8 and cytokeratin-18 and other breast luminal genes; and the luminal B subtype, which expresses only low levels of these genes62. Importantly, these five subtypes also represent clinically distinct subgroups of patients. For example, the basal-like and the ERBB2+ oestrogen-receptor-negative subtypes are associated with the shortest survival times, whereas the oestrogen-receptor-positive luminal-A subtype tumours have the best outcome of all subtypes62. These findings have been confirmed in independent gene-expression data sets63. As the tumours that belong to the different subtypes have characteristic clinical behaviour, they might also share the same therapeutic targets. The second approach to determine gene-expression patterns that can predict the clinical behaviour of tumours is the supervised classification method. Such a classification method was used to identify an expression profile of 70 genes that predicted the likelihood of distant metastases in young patients (<55 years of age) with lymph-node-negative tumours7. This RETROSPEC TIVE STUDY was particularly informative as the patients had not received adjuvant therapy, which is likely to modify outcome, and were diagnosed with breast cancer between 1983 and 1994, making a follow-up of 10 years or more possible. The primary breast tumours were classified as having either a poor-prognosis signature, which means they were likely to metastasize, or a good-prognosis signature, meaning that the development of metastases was unlikely.

ENZYMELINKED IMMUNOSORBENT ASSAY

(ELISA). A sensitive test to quantitatively determine small amounts of a particular protein in a solution. In an ELISA the interaction between the protein of interest and a specific antibody is detected by an enzyme that is linked to the antibody and converts a colourless substrate to a coloured product.
RETROSPECTIVE STUDY

A study in which data are collected and analysed after all measurements, interventions or events in the participants have taken place.

NATURE REVIEWS | C ANCER

VOLUME 5 | AUGUST 2005 | 595

2005 Nature Publishing Group

REVIEWS

Table 3 | Microarray studies on prognostic gene-expression profiles in breast cancer

Microarray type cDNA Validation Informative genes 456 intrinsic genes Crossvalidation Independent training set 534 intrinsic genes 70 genes Metastasis determinants Luminal A tumours have a better outcome than luminal B tumours. Worst outcome is for basal-like and ERBB2+ tumours Repeated finding in independent data sets Good signature is related to low metastasis risk; a poor signature is associated with a high metastasis risk. Sensitivity: 91%, specificity: 73% Good signature versus poor signature. Sensitivity: 93%, specificity: 53% Accuracy: 90% Good 76-gene signature versus poor 76 gene-signature. Sensitivity: 93%, specificity: 48% Expression of serum-activated signature versus no expression. Sensitivity: 91%, specificity: 29% References 62

cDNA Oligonucleotide

63,66 7

Oligonucleotide Oligonucleotide Oligonucleotide

Independent training set Crossvalidation Independent training set Independent training set

70 genes metagenes 76 genes

8 69 67

Oligonucleotide

442 genes

75,76

The poor-prognosis signature included genes involved in the cell cycle, invasion and metastasis, angiogenesis and signal transduction. Interestingly, it also comprised genes that are almost exclusively expressed by the stromal cells that surround the epithelial cells in a tumour. For example, these include MMP1 and MMP9, which are required for ECM degradation and tumour invasion64. The upregulation of genes that are highly expressed by stromal cells in a prognosis signature for breast cancer metastasis, and their defined role in invasion, again underlines the influence of the tumour microenvironment on tumour progression. The pure focus on epithelial cells using microdissection for the understanding of breast cancer progression and detection of prognostic markers is therefore most likely not sufficient to provide a successfully prognostic gene signature. The 70-gene signature was validated in a cohort of 295 patients that included patients with both lymphnode-negative and lymph-node-positive tumours8. By
Box 1 | Microarray platforms At present, multiple microarray platforms exist that use varying parameters. These parameters include distinct sets of genes, either cDNAs of variable lengths or small oligonucleotide sequences, and the use of two different methods to determine gene activity. One approach is to apply a single test set of fluorescently labelled cDNA from cancer cells or tissue samples to the array, the other is to hybridize both a test and a reference set of differentially labelled cDNAs to a single microarray, and measure the ratio. This might be the reason why different commercially available microarray platforms have been found to show considerable divergence in the composition of genes in a gene-expression signature68. For the analysis and interpretation of the microarray data, a range of computational tools are available102. The two basic approaches are unsupervised hierarchical clustering analyses, which orders both tumours and genes on the basis of their similarity of gene expression103, and supervised methods, which identify gene-expression patterns that discriminate tumours on the basis of pre-defined clinical information104.

multivariate analysis, the gene-expression signature was the strongest predictor for metastasis-free survival and overall survival, and was independent of the other clinical and pathological prognostic markers. In the group of 151 patients who had lymph-node-negative disease, 60% of the patients were classified as having a high metastatic risk (poor prognosis) and 40% of the patients as having a low metastasis risk (good prognosis). After a follow-up period of 10 years, 56% of the poor-prognosis patients developed a metastasis, whereas only 13% of the good-prognosis patients did. Classifying the patients who are at risk of metastasis on the basis of traditional clinical parameters, the St Gallen criteria65 assigned 15% of the 151 patients with lymph-node-negative breast cancer to the low metastasis risk (good prognosis) group 8 , and the National Institutes of Health (NIH) criteria2 assigned only 7% of these patients to this group. After 10 years of follow-up in these 151 patients, approximately 2025% of the good-prognosis patients (as classified by the St Gallen or NIH criteria) had recurrences at distant sites, and only approximately 45% of the high-risk patients experienced metastasis8. These results indicate that the present criteria misclassify a significant number of patients, which results in their overtreatment or undertreatment. The 70-gene-expression profile might therefore be used to tailor therapy for individual breast cancer patients and might reduce the number of patients who would receive unnecessary adjuvant systemic treatment. This gene-expression profile is currently being tested in retrospective series from other hospitals as well as in a well-designed prospective study by a large translational research consortium (known as the Translational Research Breast International Group, or TransBIG). This study will independently determine whether the prognostic power of the 70-gene signature is reproducible in a

596 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer

REVIEWS
more diverse population in a multicentre setting, and whether it can replace traditional clinicopathological data, as described earlier, in the near future66. Recently, using a different microarray platform, a gene-expression signature of 76 genes was retrospectively found that could be used to predict outcome in patients of all age groups with lymph-node-negative breast cancer67. In a group of 115 breast cancer patients who had not received adjuvant systemic therapy, a 76-gene profile consisting of 60 genes for oestrogen-receptor-positive patients and 16 genes for oestrogen-receptor-negative patients was identified. As for the 70-gene signature described above, genes involved in cell death, cell cycle and proliferation, DNA replication and repair, and immune response were represented in the identified profile. In the same study, the 76-gene signature was validated by a set of 171 lymph-node-negative breast cancer patients. This signature represents an independent prognostic marker strongly associated with a higher risk of tumour metastasis and shortened overall survival. Also, this signature would result in a reduction of the number of patients who would be recommended to have systemic adjuvant therapy that was probably unnecessary. According to the 76-gene signature, in the group of patients studied, 52% of the patients with a low risk of metastasis development would be eligible for adjuvant treatment, compared with 90% and 89% by the St Gallen65 and NIH2 guidelines. The comparison of these results to the data discussed earlier is not straightforward, as different microarray platforms as well as mathematical algorithms were used. This is probably the reason why only three genes overlap between the two signatures. It has been shown that different microarray platforms might reveal different gene sets, but are actually reporting the same biological processes68. Nevertheless, the finding of a second signature by an independent research group confirms the existence of a gene-expression prognosis profile in patients with primary breast carcinoma. A third supervised approach identified aggregate patterns of gene expression METAGENES that are associated with lymph-node status at diagnosis and a 3-year-recurrence risk in breast cancer patients of all ages69. Owing to small sample numbers, cross-validation is used to determine the accuracy of the 3-year-recurrence predictor instead of a second independent set of tumour samples. Therefore, further studies are required to validate this metagene classifier for breast cancer recurrence. Recently, two studies also reported novel markers to predict distant metastasis risk and clinical outcome in patients with oestrogen-receptor-positive breast tumours who have been treated with adjuvant tamoxifen. Ma et al. showed that the expression ratio of the two genes HIXB13 and IL17BL accurately predicts metastatis development70, whereas Paik et al. identified 21 genes that can be detected by RT-PCR analysis in paraffinembedded tumour tissue to predict distant recurrence71. However, neither the two-gene expression ratio nor the 21-gene panel is effective in predicting clinical outcome of adjuvantly untreated breast cancer patients70,72. Can we determine a signature that not only predicts poor outcome, but also the risk of metastasis development in a specific organ? Massagu and colleagues identified a set of genes in a human breast cancer cell line, the expression of which was associated with metastasis to bone in mice73. Subsequently, 63 primary breast cancer samples were profiled to determine whether this gene-expression pattern could be used to identify those that had metastasized to bone in patients74. Hierarchical clustering, however, could not distinguish between tumours that had metastasized to bone and those that had not. Only when the analysis was restricted to those tumours that were known to have metastasized could the profile weakly discriminate a bone-metastasis cluster from a lung-metastasis cluster. This indicates that the data obtained from this mouse model cannot directly be transferred to the human situation. Interestingly, using published DNA-microarray data, a gene-expression signature was developed that is associated with the serum response in fibroblasts, and it was able to predict metastasis risk in different kinds of human tumour, including breast, prostate, lung, gastric and hepatocellular carcinomas75. In a subsequent study, the predictive power of the serumresponse signature was tested in 295 patients with breast cancer; these were the same patients who had been used to validate the prognostic 70-gene-expression profile as described earlier. It was shown that both overall survival and metastasis-free survival are markedly diminished in patients whose tumours expressed the serum-induced gene-expression profile compared with those that did not express this signature76. This signature approximately identified 90% of patients who developed metastases, and at the same time would have spared 30% of women who did not develop metastasis from exposure to cytotoxic chemotherapy. These results illustrate the potential utility and improved metastasis risk stratification (that is, the more accurate risk assignment) of the signature, independently of clinical or pathological risk factors. In addition, the observation that the transcriptional signature of the response of fibroblasts to serum can predict human breast tumour metastasis again reveals a possible important contribution of stromal fibroblasts to tumour progression. Epithelial tumour cells might therefore activate some of the normal wound-healing responses that lead to metastasis33. Finally, it must be stressed that the membership of a gene in a prognostic list that is determined by a supervised classification method is not necessarily indicative of the importance of that gene in cancer pathology. This is because such lists are strongly influenced by the subset of patients who are used for gene selection77. In conclusion, gene-expression profiling might refine the prognostic classification of breast cancer, allowing researchers to more accurately identify patients who are at metastasis risk than the present conventional prognostic markers. Moreover, the genes that are deregulated in the molecularly defined classes with poor outcome might also constitute novel targets for therapy.

METAGENE

Metagenes are linear combinations of individual gene-expression values. They have the potential to classify and predict cellular phenotypes resulting from deregulation of oncogenic pathways.

NATURE REVIEWS | C ANCER

VOLUME 5 | AUGUST 2005 | 597

2005 Nature Publishing Group

REVIEWS
metastases arise from definite genetically determined subpopulations in primary tumours79. However, multiple passages of this heterogeneous melanoma cell line, both in animals and in cell culture, provide sufficient opportunity for variant genotypic cell types to arise80. Therefore, metastasis might be more accurately studied using spontaneous metastasis models, and cells that were not carried in culture. In such models the formation of natural metastases from the primary tumours of the mice is investigated. Indeed, tumour cells derived from these metastases did not have greater metastatic capacity than those isolated from the corresponding primary tumour46,8183. This was also true for the spontaneous metastases that arose from B16 tumours5 these cells only had increased metastatic potential when placed in the experimental context developed by Fidler et al.3,78. The data from the spontaneous METASTASIS ASSAYS indicate that metastases are a random representation of disseminated tumour cells, all of which have the ability to form a metastasis (FIG. 2). Additionally, Weiss et al. showed that equalsized fragments isolated from large or small tumours showed no difference in their metastatic potential in mice83, indicating that there is no apparent relationship between metastatic potential and tumour size. Different hypotheses attempted to reconcile the discrepancies in the experimental findings concerning the putative selective nature of the metastatic phenotype. The dynamic heterogeneity model proposes that metastatic subpopulations are generated at high rates in a primary tumour, but that these variants are relatively unstable, resulting in a dynamic equilibrium between generation and loss of metastatic variants84,85 (FIG. 2). The clonal dominance theory of metastasis, however, proposes that once a metastatic subclone emerges within a primary tumour, the progeny of this subclone overgrow and dominate the tumour mass itself 86,87 (FIG. 2). Garcia-Olmo and colleagues called for a change in the concept of metastasis. Their provocative genometastasis hypothesis, which is based on in vitro work, proposes that metastasis occurs through transfection of susceptible cells in distant organs with dominant, plasma-circulating oncogenes that are derived from the primary tumour 88,89 (FIG. 2). Recently, it has been shown that the genetic background from which cancer arises also has an affect on the capacity of mouse mammary tumour cells to metastasize90. These findings indicate that the propensity to metastasize is, in part, influenced by the normal genetic make-up of the host. The observation that the stromal microenvironment functionally contributes to the development of breast carcinomas (reviewed in REFS 91,92) indicates that breast cancer progression might be more complex than the current linear theory of activation and inactivation of oncogenes and tumour-suppressor genes. Therefore, the key question, whether the findings made in animal and also in in vitro models can be directly compared to the metastasis of breast tumours in patients, remains arguable.

Primary tumour

Metastasis

Figure 2 | Models of the metastatic process. a | The traditional model of metastasis suggests that only subpopulations of tumour cells (red) aquire metastatic capacity late in tumorigenesis3. b | Spontaneous metastasis assays indicate that all tumour cells have the capability to develop a metastasis46,81,82. c | The dynamic heterogeneity model proposes that the frequency with which metastatic variants arise within the primary tumour determines its metastatic potential84,85. d | The clonal dominance theory proposes that metastatic subclones within a primary tumour can overgrow and dominate the tumour mass itself 86,87. e | The genometastasis hypothesis proposes that metastasis occurs through transfection of susceptible cells in distant organs with circulating oncogenes88,89. Adapted from REF. 105 Nature Medicine (2003) Macmillan Magazines Ltd.

Models of the metastatic cascade

EXPERIMENTAL METASTASIS ASSAY

Assays in which cultured tumour cells or minced tumours are introduced into blood vessels or organs of mice and examined for their in vivo growth and tumour formation.

The metastatic properties of tumour cells were extensively investigated in the late 1970s and early 1980s by means of experimental metastasis assays. By studying the metastatic behaviour of cultured B16 melanoma cells that were injected intravenously into mice, Fidler showed that cells derived from outgrowths of these cells (metastases) have a higher metastatic potential than those derived from the original cell line78. In vitro clones from the parent B16 culture varied greatly in their ability to produce lung metastases after intravenous injection into mice3. These observations led to a metastasis model, which proposed that most primary tumour cells have a low metastatic potential, and that during later stages of tumorigenesis rare cells acquire metastatic capacity through additional somatic mutations3 (FIG. 2). This model was actually a sequel to Leightons hypothesis in 1965, which predicted that

598 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer

REVIEWS
are highly similar at their transcriptome level93, as are pre-malignant, pre-invasive, and invasive breast cancers94. A variation to this model was proposed by Massagu and colleagues73. As described above, a human breast cancer cell line was shown to harbour, besides a poor-prognosis signature, an additional gene set that mediated osteolytic bone metastasis73. These findings were interpreted to bridge the gap between the subpopulation metastasis model3 and the one based on the microarray data of human tumours7,9,10. The authors propose the intriguing model that primary tumours with metastatic capacity possess the poor-prognosis signature and, in addition, subpopulations of cells also have a superimposed tissue-specific gene-expression profile that predicts the site of metastasis73 (FIG. 3). In contrast to the microarray studies on primary breast carcinomas, the analysis of human disseminated breast cancer cells led to a model proposing that metastatic disease evolves independently from the primary tumour95 (FIG. 3). This theory clearly challenges the paradigm that tumour progression to metastasis occurs through clonal genomic evolution. The genomic alterations of disseminated tumour cells in the bone marrow of patients without clinical evidence of metastases generally did not resemble those of the primary tumours, in contrast to tumour cells of patients with manifest metastases. However, the authors are not able to distinguish between true metastatic cells and tumour cells that are not capable of proliferating at distant sites, and cannot exclude the possibility that the cytokeratin-positive cells might also be of epithelial origin that is unrelated to cancer. An alternative, attractive model of metastasis is based on the finding that tumours might contain cancer stem cells rare cells with indefinite proliferative potential that drive the formation and growth of tumours96. Experimentally, only a minority of human breast cancer cells, which were derived from fresh human tumours and grown in mammary fat pads of immunocompromised mice, were found to have the ability to form new tumours48. Al-Hajj and colleagues were able to distinguish tumour-initiating cells from most nontumorigenic cancer cells based on cell-surface-marker expression. These findings might provide an explanation for clinical observations in breast cancer patients such as the phenomena interpreted as tumour dormancy and the lack of prognostic significance of disseminated tumour cells in bone marrow one of the newer potential prognostic markers as discussed earlier in this article. Cancer stem cells have also been identified in other human cancers such as brain tumours97.
Integrative model of breast cancer metastasis

Primary tumour

Metastasis

Figure 3 | New models of the metastatic process in breast cancer. a | Gene-expression profiling of human primary breast tumours can predict metastasis risk (poorprognosis (red) versus good-prognosis (pink) signature), which indicates that the capacity to metastasize might be acquired early during tumorigenesis7,8,10. b | Primary tumours with metastasizing capacity display the poor-prognosis signature and an additional tissue-specific expression profile predicting the site of metastasis (green, bone; blue, liver; purple, lung)73,74. c | The parallel evolution model proposes that the dissemination of metastatic cancer cells occurs early in oncogenesis and independently from tumour cells at the primary site95. d | Only breast cancer stem cells, not the nontumorigenic bulk of the tumour, have the ability to metastasize and form new tumours48. See also REF. 105. Adapted from REF. 106 Nature Medicine (2003) Macmillan Magazines Ltd.

Gene-expression analysis of breast cancer metastasis. Findings from DNA-microarray studies have revived the discussion about the metastatic process. As described above, the ability of gene-expression profiles of human primary breast carcinomas to predict the metastatic potential7,8 indicates that the ability to metastasize is an early and inherent property of the breast tumour (FIG. 3). These results indicate that breast cancer is both a local, and a systemic, disease. Furthermore, the data challenge the idea that variant cells arise that give rise to metastases during the late stages of tumour progression3. Several independent lines of evidence seem to support this concept. A study by Ramaswamy and colleagues shows that different types of human primary adenocarcinoma harbour the same gene-expression signature that is associated with metastasis9. Furthermore, it was reported that pairs of human primary breast carcinomas and their distant metastases, which developed years later,

Studies with prognostic markers such as uPA/PAI1, the 70-gene expression profile, and the fibroblast serum-response signature have all demonstrated that the tumour microenvironment seems to significantly contribute to tumorigenesis, which supports numerous recent in vivo and in vitro studies (reviewed in REFS 92,98). The integration of this knowledge and the proposed

NATURE REVIEWS | C ANCER

VOLUME 5 | AUGUST 2005 | 599

2005 Nature Publishing Group

REVIEWS
When breast cancer aetiology has followed the metastasis route from the start, there is a high likelihood of clinically defined systemic disease. By contrast, the disease remains local in patients with non-metastatic, good-prognosis types of tumour.
Future directions
Stroma

Primary tumour

Metastasis

Breast stem cell Differentiated progenitor cell

Breast cancer stem cell

Figure 4 | An integrative model of breast cancer metastasis. Oncogenic mutations occuring in a breast stem cell (red) can cause the transformation to a breast cancer stem cell, generating poor-prognosis tumours (orange). Mutations occurring in differentiated progenitor cells (yellow) might form a non-metastatic good-prognosis breast carcinoma (pink). In the metastatic poorprognosis tumours, under the influence of stromal fibroblasts, only the population of breast cancer stem cells has the ability to metastasize. There might be variant cancer stem cells that differ in their tissue selectivity for metastasis, expressing an additional tissue-specific profile (for example: green, bone; purple, lung). At the site of metastasis, the disseminated cancer stem cells would again induce a similar stromal response as in the primary breast tumour.

LURIADELBRUCK FLUCTUATION ANALYSIS

A method to estimate mutation rates in cell populations, originally designed for bacteria.

metastasis models allow us to speculate about a new model of the metastatic process (FIG. 4). In this model, primary breast carcinomas with metastatic potential can be distinguished from those that have a low likelihood of metastasis by their gene-expression profiles the poor- and the good-prognosis signatures, respectively, as determined by the 70-gene expression profile7. Metastatic-type tumours, under the influence of the stromal fibroblasts, might harbour seeding subpopulations, as proposed by Fidler and colleagues3. These variant cells that are capable of forming metastases, which have been shown by LURIADELBRUCK FLUCTUATION ANALYSIS to be generated at a rate of about 5 105 per cell per generation84, might, in fact, be a small population of cancer stem cells48,96 that gives rise to the non-tumorigenic bulk of the tumour. So, the initiating event of metastasis might be a breast stem cell that undergoes transforming oncogenic mutations, leading to a cancer stem cell that generates poor-prognosis tumours. Specifically, a self-renewing population of (stem) cells is necessary to accumulate the mutations that are required for tumorigenesis. By contrast, genetic events that occur in differentiated progenitor cells might display a non-metastatic, good-prognosis breast carcinoma99. So, mutations that occur at different stages of tumour differentiation control the capacity of the tumour cell to metastasize. Additionally, there might be variant cancer stem cells that differ in their propensity to metastasize to a specific tissue, and therefore express an additional tissue-specific profile73. At the site of metastasis, the disseminated cancer stem cells would then again need, or induce, a stromal response similar to that of their primary breast tumour, as well as the formation of new blood vessels.

New prognostic markers of breast cancer metastasis are urgently needed to avoid overtreatment or undertreatment of newly diagnosed patients. Microarray gene-expression analysis has shown promise as a useful prognostic marker. But do these studies provide us with new tumour markers that can be routinely used for newly diagnosed breast cancer patients? Clearly, to allow microarray testing in all hospitals, the technology and access to it needs to be improved. Furthermore, the current system to get these gene-expression signatures to the highest level of clinical use26 requires large prospective randomized trials. It is believed that the current healthcare system cannot take the financial burden to test all new tumour markers in this way. Alternatively, based on the great predictive power of gene-expression signatures, several well-designed retrospective studies might be sufficient for their introduction into the clinic. Gene-expression signatures might also be used to predict the site of human tumour metastasis these can currently be predicted in mice. The identification of tissue-specific signatures for metastasis would not only improve our understanding of the mechanisms by which tumours spread to specific tissues, but would also identify new therapeutic targets. In addition, we await the identification of predictive gene-expression profiles that will enable us to tailor adjuvant therapy choices to individuals. With the exception of a few small pilot studies, only two studies, which included 24 and 42 patients respectively, have so far reported the association between a geneexpression signature and drug sensitivity to docetaxel or to a combination regimen containing paclitaxel, fluorouracil, doxorubicin and cyclophosphamide in breast cancer patients100,101. Based on the proposed integrative model of the metastatic process, it might be worthwhile focusing more research on the characterization of the crucial population of cancer stem cells. Breast cancer stem cells make an attractive therapeutic target, as tumours would be targeted by their cell of origin. The isolation of the cancer-stem-cell population and the analysis of these cells by gene-expression profiling should facilitate the identification of specific pathways that are important for their growth and survival. Further elucidation of the contribution of the tumour environment to regulating tissue specificity, tumorigenesis, and probably also the cancer stem cells98, is also important. This knowledge will allow the development of new therapeutic strategies targeting both seed and soil, which might lead to more effective intervention strategies for breast cancer and breast cancer metastasis.

600 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer

REVIEWS

1.

2.

3.

4.

5.

6.

7.

8.

9.

10. 11.

12.

13. 14.

15.

16.

17.

18.

19.

20.

21.

22.

Early Breast Cancer Trialists Collaborative Group (EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 365, 16871717 (2005). Gives an overview of the results of 194 randomized trials of adjuvant chemotherapy or hormonal therapy among women with early breast cancer. Eifel, P. et al. National Institutes of Health Consensus Development Conference Statement: adjuvant therapy for breast cancer, November 13, 2000. J. Natl Cancer Inst. 93, 979989 (2001). Provides an assessment of available data between January 1995 and July 2000 regarding the use of adjuvant therapy for breast cancer. Fidler, I. J. & Kripke, M. L. Metastasis results from preexisting variant cells within a malignant tumour. Science 197, 893895 (1977). The experimental data establish the longstanding model of metastasis, which indicates that metastatic cells are rare and arise late during tumorigenesis. Milas, L., Peters, L. J. & Ito, H. Spontaneous metastasis: random or selective? Clin. Exp. Metastasis 1, 309315 (1983). Giavazzi, R., Alessandri, G., Spreafico, F., Garattini, S. & Mantovani, A. Metastasizing capacity of tumour cells from spontaneous metastases of transplanted murine tumours. Br. J. Cancer 42, 462472 (1980). Mantovani, A., Giavazzi, R., Alessandri, G., Spreafico, F. & Garattini, S. Characterization of tumour lines derived from spontaneous metastases of a transplanted murine sarcoma. Eur. J. Cancer 17, 7176 (1981). van t Veer, L. J. et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530536 (2002). Shows that a 70-gene expression profile of primary breast tumours can predict the later appearance or absence of distant metastases. van de Vijver, M. J. et al. A gene-expression signature as a predictor of survival in breast cancer. N. Engl. J. Med. 347, 19992009 (2002). Ramaswamy, S., Ross, K. N., Lander, E. S. & Golub, T. R. A molecular signature of metastasis in primary solid tumours. Nature Genet. 33, 4954 (2003). Bernards, R. & Weinberg, R. A. A progression puzzle. Nature 418, 823 (2002). Hellman, S. & Harris, J. R. in Diseases of the Breast (eds Harris, J. R., Lippman, M. E., Morrow, M. & Osborne, C. K.), 407423 (Lippincott Williams & Wilkins, Philadelphia, 2000). Describes the clinical behaviour of untreated breast cancer, including the incidences of regional lymphnode and distant metastasis. Tavassoli, F. A. & Devilee, P. (eds) World Health Organization Classification of Tumours. Pathology and Genetics of Tumours of the Breast and Female Genital Organs. (IARC press, Lyon, 2003). Lee, Y. T. Breast carcinoma: pattern of metastasis at autopsy. J. Surg. Oncol. 23, 175180 (1983). Page, D. L. Prognosis and breast cancer. Recognition of lethal and favourable prognostic types. Am. J. Surg. Pathol. 15, 334349 (1991). Koscielny, S. et al. Breast cancer: relationship between the size of the primary tumour and the probability of metastatic dissemination. Br. J. Cancer 49, 709715 (1984). Rosen, P. P., Groshen, S., Saigo, P. E., Kinne, D. W. & Hellman, S. Pathological prognostic factors in stage I (T1N0M0) and stage II (T1N1M0) breast carcinoma: a study of 644 patients with median follow-up of 18 years. J. Clin. Oncol. 7, 12391251 (1989). Carter, C. L., Allen, C. & Henson, D. E. Relation of tumour size, lymph node status, and survival in 24,740 breast cancer cases. Cancer 63, 181187 (1989). Elston, C. W. & Ellis, I. O. Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up. Histopathology 19, 403410 (1991). Pinder, S. E. et al. Pathological prognostic factors in breast cancer. III. Vascular invasion: relationship with recurrence and survival in a large study with long-term follow-up. Histopathology 24, 4147 (1994). de Mascarel, I. et al. Obvious peritumoural emboli: an elusive prognostic factor reappraised. Multivariate analysis of 1320 node-negative breast cancers. Eur. J. Cancer 34, 5865 (1998). Hellman, S. Karnofsky Memorial Lecture. Natural history of small breast cancers. J. Clin. Oncol. 12, 22292234 (1994). James, J. J. et al. Bone metastases from breast carcinoma: histopathologicalradiological correlations and prognostic features. Br. J. Cancer 89, 660665 (2003).

23. Borst, M. J. & Ingold, J. A. Metastatic patterns of invasive lobular versus invasive ductal carcinoma of the breast. Surgery 114, 637641 (1993). 24. Arpino, G., Bardou, V. J., Clark, G. M. & Elledge, R. M. Infiltrating lobular carcinoma of the breast: tumour characteristics and clinical outcome. Breast Cancer Res. 6, R149R156 (2004). 25. McGuire, W. L. & Clark, G. M. Prognostic factors and treatment decisions in axillary-node-negative breast cancer. N. Engl. J. Med. 326, 17561761 (1992). 26. Hayes, D. F. et al. Tumour marker utility grading system: a framework to evaluate clinical utility of tumour markers. J. Natl Cancer Inst. 88, 14561466 (1996). 27. Ransohoff, D. F. Bias as a threat to the validity of cancer molecular-marker research. Nature Rev. Cancer 5, 142149 (2005). 28. Slamon, D. J. et al. Human breast cancer: correlation of relapse and survival with amplification of the HER-2/neu oncogene. Science 235, 177182 (1987). 29. Slamon, D. J. et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. N. Engl. J. Med. 344, 783792 (2001). 30. Ross, J. S. et al. The Her-2/neu gene and protein in breast cancer 2003: biomarker and target of therapy. Oncologist 8, 307325 (2003). 31. Yamauchi, H., Stearns, V. & Hayes, D. F. When is a tumour marker ready for prime time? A case study of c-erbB-2 as a predictive factor in breast cancer. J. Clin. Oncol. 19, 23342356 (2001). 32. Ross, J. S. et al. Targeted therapy in breast cancer: the HER-2/neu gene and protein. Mol. Cell Proteomics 3, 379398 (2004). 33. Dvorak, H. F. Tumours: wounds that do not heal. Similarities between tumour stroma generation and wound healing. N. Engl. J. Med. 315, 16501659 (1986). 34. Ring, A., Smith, I. E. & Dowsett, M. Circulating tumour cells in breast cancer. Lancet Oncol. 5, 7988 (2004). 35. Stathopoulou, A. et al. Molecular detection of cytokeratin-19-positive cells in the peripheral blood of patients with operable breast cancer: evaluation of their prognostic significance. J. Clin. Oncol. 20, 34043412 (2002). 36. Zach, O., Kasparu, H., Wagner, H., Krieger, O. & Lutz, D. Prognostic value of tumour cell detection in peripheral blood of breast cancer patients. Acta Med. Austriaca 59, 3234 (2002). 37. Pierga, J. Y. et al. Clinical significance of immunocytochemical detection of tumour cells using digital microscopy in peripheral blood and bone marrow of breast cancer patients. Clin. Cancer Res. 10, 13921400 (2004). 38. Diel, I. J. et al. Micrometastatic breast cancer cells in bone marrow at primary surgery: prognostic value in comparison with nodal status. J. Natl Cancer Inst. 88, 16521658 (1996). 39. Landys, K., Persson, S., Kovarik, J., Hultborn, R. & Holmberg, E. Prognostic value of bone marrow biopsy in operable breast cancer patients at the time of initial diagnosis: results of a 20-year median follow-up. Breast Cancer Res. Treat. 49, 2733 (1998). 40. Braun, S. et al. Cytokeratin-positive cells in the bone marrow and survival of patients with stage I, II, or III breast cancer. N. Engl. J. Med. 342, 525533 (2000). 41. Wiedswang, G. et al. Detection of isolated tumour cells in bone marrow is an independent prognostic factor in breast cancer. J. Clin. Oncol. 21, 34693478 (2003). 42. Molino, A. et al. Bone marrow micrometastases in 109 breast cancer patients: correlations with clinical and pathological features and prognosis. Breast Cancer Res. Treat. 42, 2330 (1997). 43. Ismail, M. S. et al. Detection of micrometastatic disease and monitoring of perioperative tumour cell dissemination in primary operable breast cancer patients using real-time quantitative reverse transcription-PCR. Clin. Cancer Res. 10, 196201 (2004). 44. Gebauer, G., Fehm, T., Merkle, E., Jaeger, W. & Mitze, M. Micrometastases in axillary lymph nodes and bone marrow of lymph node-negative breast cancer patients prognostic relevance after 10 years. Anticancer Res. 23, 43194324 (2003). 45. Funke, I. & Schraut, W. Meta-analyses of studies on bone marrow micrometastases: an independent prognostic impact remains to be substantiated. J. Clin. Oncol. 16, 557566 (1998). 46. Fidler, I. J. Metastasis: quantitative analysis of distribution and fate of tumour embolilabeled with 125 I-5-iodo-2deoxyuridine. J. Natl Cancer Inst. 45, 773782 (1970). 47. Gangnus, R., Langer, S., Breit, E., Pantel, K. & Speicher, M. R. Genomic profiling of viable and proliferative micrometastatic cells from early-stage breast cancer patients. Clin. Cancer Res. 10, 34573464 (2004).

48. Al-Hajj, M., Wicha, M. S., Benito-Hernandez, A., Morrison, S. J. & Clarke, M. F. Prospective identification of tumorigenic breast cancer cells. Proc. Natl Acad. Sci. USA 100, 39833988 (2003). Describes the prospective isolation of candidate breast tumour stem cells and their ability to form tumours in mice. 49. Gudjonsson, T. et al. Isolation, immortalization, and characterization of a human breast epithelial cell line with stem cell properties. Genes Dev. 16, 693706 (2002). 50. Clarke, R. B. et al. A putative human breast stem cell population is enriched for steroid receptor-positive cells. Dev. Biol. 277, 443456 (2005). 51. Andreasen, P. A., Kjoller, L., Christensen, L. & Duffy, M. J. The urokinase-type plasminogen activator system in cancer metastasis: a review. Int. J. Cancer 72, 122 (1997). 52. Umeda, T., Eguchi, Y., Okino, K., Kodama, M. & Hattori, T. Cellular localization of urokinase-type plasminogen activator, its inhibitors, and their mRNAs in breast cancer tissues. J. Pathol. 183, 388397 (1997). 53. Janicke, F., Schmitt, M. & Graeff, H. Clinical relevance of the urokinase-type and tissue-type plasminogen activators and of their type 1 inhibitor in breast cancer. Semin. Thromb. Hemost. 17, 303312 (1991). 54. Bajou, K. et al. Host-derived plasminogen activator inhibitor-1 (PAI-1) concentration is critical for in vivo tumoural angiogenesis and growth. Oncogene 23, 69866990 (2004). 55. Foekens, J. A. et al. The urokinase system of plasminogen activation and prognosis in 2780 breast cancer patients. Cancer Res. 60, 636643 (2000). 56. Konecny, G. et al. Her-2/neu and urokinase-type plasminogen activator and its inhibitor in breast cancer. Clin. Cancer Res. 7, 24482457 (2001). 57. Janicke, F. et al. Randomized adjuvant chemotherapy trial in high-risk, lymph node-negative breast cancer patients identified by urokinase-type plasminogen activator and plasminogen activator inhibitor type 1. J. Natl Cancer Inst. 93, 913920 (2001). 58. Harbeck, N., Kates, R. E. & Schmitt, M. Clinical relevance of invasion factors urokinase-type plasminogen activator and plasminogen activator inhibitor type 1 for individualized therapy decisions in primary breast cancer is greatest when used in combination. J. Clin. Oncol. 20, 10001007 (2002). 59. Look, M. P. et al. Pooled analysis of prognostic impact of urokinase-type plasminogen activator and its inhibitor PAI-1 in 8377 breast cancer patients. J. Natl Cancer Inst. 94, 116128 (2002). Confirms the prognostic value of uPA and PAI1 in primary breast cancer in a re-analysis of independent patient data provided by the members of the EORTC Receptor and Biomarker Group. 60. Harbeck, N. et al. Enhanced benefit from adjuvant chemotherapy in breast cancer patients classified high-risk according to urokinase-type plasminogen activator (uPA) and plasminogen activator inhibitor type 1 (n = 3424). Cancer Res. 62, 46174622 (2002). 61. Perou, C. M. et al. Molecular portraits of human breast tumours. Nature 406, 747752 (2000). Classifies primary breast tumours into 5 subtypes on the basis of similarities in their gene-expression patterns. 62. Sorlie, T. et al. Gene expression patterns of breast carcinomas distinguish tumour subclasses with clinical implications. Proc. Natl Acad. Sci. USA 98, 1086910874 (2001). 63. Sorlie, T. et al. Repeated observation of breast tumour subtypes in independent gene expression data sets. Proc. Natl Acad. Sci. USA 100, 84188423 (2003). 64. Egeblad, M. & Werb, Z. New functions for the matrix metalloproteinases in cancer progression. Nature Rev. Cancer 2, 161174 (2002). 65. Goldhirsch, A., Glick, J. H., Gelber, R. D., Coates, A. S. & Senn, H. J. Meeting highlights: international consensus panel on the treatment of primary breast cancer. Seventh international conference on adjuvant therapy of primary breast cancer. J. Clin. Oncol. 19, 38173827 (2001). 66. Tuma, R. S. A big trial for a new technology: TransBIG project takes microarrays into clinical trials. J. Natl Cancer Inst. 96, 648689 (2004). 67. Wang, Y. et al. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 365, 671679 (2005). 68. Tan, P. K. et al. Evaluation of gene expression measurements from commercial microarray platforms. Nucleic Acids Res. 31, 56765684 (2003). 69. Huang, E. et al. Gene expression predictors of breast cancer outcomes. Lancet 361, 15901596 (2003). 70. Ma, X. J. et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell 5, 607616 (2004).

NATURE REVIEWS | C ANCER

VOLUME 5 | AUGUST 2005 | 601

2005 Nature Publishing Group

REVIEWS

71. Paik, S. et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N. Engl. J. Med. 351, 28172826 (2004). 72. Esteva, F. J. et al. Prognostic role of a multigene reverse transcriptase-PCR assay in patients with node-negative breast cancer not receiving adjuvant systemic therapy. Clin. Cancer Res. 11, 33153319 (2005). 73. Kang, Y. et al. A multigenic program mediating breast cancer metastasis to bone. Cancer Cell 3, 537549 (2003). 74. Minn, A. J. et al. Distinct organ-specific metastatic potential of individual breast cancer cells and primary tumours. J. Clin. Invest. 115, 4455 (2005). Shows that organ-specific metastasis of a breast cancer cell line is determined by the poor-prognosis signature and an additional metastasis-specific gene set. 75. Chang, H. Y. et al. Gene expression signature of fibroblast serum response predicts human cancer progression: similarities between tumours and wounds. PLoS Biol. 2, E7 (2004). 76. Chang, H. Y. et al. Robustness, scalability, and integration of a wound-response gene expression signature in predicting breast cancer survival. Proc. Natl Acad. Sci. USA 102, 37383743 (2005). 77. Ein-Dor, L., Kela, I., Getz, G., Givol, D. & Domany, E. Outcome signature genes in breast cancer: is there a unique set? Bioinformatics 21, 171178 (2005). 78. Fidler, I. J. Selection of successive tumour lines for metastasis. Nature New Biol. 242, 148149 (1973). 79. Leighton, J. Inherent malignancy of cancer cells possibly limited by genetically differing cells in the same tumour. Acta Cytol. 9, 138140 (1965). 80. Stackpole, C. W. Distinct lung-colonizing and lungmetastasizing cell populations in B16 mouse melanoma. Nature 289, 798800 (1981). 81. Vaage, J. Metastasizing potentials of mouse mammary tumours and their metastases. Int. J. Cancer 41, 855858 (1988). 82. Eccles, S. A., Heckford, S. E. & Alexander, P. Effect of cyclosporin A on the growth and spontaneous metastasis of syngeneic animal tumours. Br. J. Cancer 42, 252259 (1980). 83. Weiss, L., Holmes, J. C. & Ward, P. M. Do metastases arise from pre-existing subpopulations of cancer cells? Br. J. Cancer 47, 8189 (1983). 84. Hill, R. P., Chambers, A. F., Ling, V. & Harris, J. F. Dynamic heterogeneity: rapid generation of metastatic variants in mouse B16 melanoma cells. Science 224, 9981001 (1984). 85. Ling, V., Chambers, A. F., Harris, J. F. & Hill, R. P. Quantitative genetic analysis of tumour progression. Cancer Metastasis Rev. 4, 173192 (1985). 86. Kerbel, R. S., Waghorne, C., Man, M. S., Elliott, B. & Breitman, M. L. Alteration of the tumorigenic and metastatic

properties of neoplastic cells is associated with the process of calcium phosphate-mediated DNA transfection. Proc. Natl Acad. Sci. USA 84, 12631267 (1987). 87. Kerbel, R. S., Waghorne, C., Korczak, B., Lagarde, A. & Breitman, M. L. Clonal dominance of primary tumours by metastatic cells: genetic analysis and biological implications. Cancer Surv. 7, 597629 (1988). 88. Garcia-Olmo, D., Garcia-Olmo, D. C., Ontanon, J., Martinez, E. & Vallejo, M. Tumour DNA circulating in the plasma might play a role in metastasis. The hypothesis of the genometastasis. Histol. Histopathol. 14, 11591164 (1999). 89. Garcia-Olmo, D. & Garcia-Olmo, D. C. Functionality of circulating DNA: the hypothesis of genometastasis. Ann. NY Acad. Sci. 945, 265275 (2001). 90. Lancaster, M., Rouse, J. & Hunter, K. W. Modifiers of mammary tumour progression and metastasis on mouse chromosomes 7, 9, and 17. Mamm. Genome 16, 120126 (2005). 91. Bissell, M. J. & Radisky, D. Putting tumours in context. Nature Rev. Cancer 1, 4654 (2001). 92. Bhowmick, N. A., Neilson, E. G. & Moses, H. L. Stromal fibroblasts in cancer initiation and progression. Nature 432, 332337 (2004). 93. Weigelt, B. et al. Gene expression profiles of primary breast tumours maintained in distant metastases. Proc. Natl Acad. Sci. USA 100, 1590115905 (2003). 94. Ma, X. J. et al. Gene expression profiles of human breast cancer progression. Proc. Natl Acad. Sci. USA 100, 59745979 (2003). 95. Schmidt-Kittler, O. et al. From latent disseminated cells to overt metastasis: genetic analysis of systemic breast cancer progression. Proc. Natl Acad. Sci. USA 100, 77377742 (2003). 96. Reya, T., Morrison, S. J., Clarke, M. F. & Weissman, I. L. Stem cells, cancer, and cancer stem cells. Nature 414, 105111 (2001). 97. Singh, S. K. et al. Identification of human brain tumour initiating cells. Nature 432, 396401 (2004). 98. Bissell, M. J. & Labarge, M. A. Context, tissue plasticity, and cancer; are tumour stem cells also regulated by the microenvironment? Cancer Cell 7, 1723 (2005). 99. Dontu, G., El-Ashry, D. & Wicha, M. S. Breast cancer, stem/progenitor cells and the estrogen receptor. Trends Endocrinol. Metab. 15, 193197 (2004). 100. Chang, J. C. et al. Gene expression profiling for the prediction of therapeutic response to docetaxel in patients with breast cancer. Lancet 362, 362369 (2003). 101. Ayers, M. et al. Gene expression profiles predict complete pathologic response to neoadjuvant paclitaxel and fluorouracil,

102. 103.

104.

105. 106.

doxorubicin, and cyclophosphamide chemotherapy in breast cancer. J. Clin. Oncol. 22, 22842293 (2004). Quackenbush, J. Computational analysis of microarray data. Nature Rev. Genet. 2, 418427 (2001). Eisen, M. B., Spellman, P. T., Brown, P. O. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc. Natl Acad. Sci. USA 95, 1486314868 (1998). Golub, T. R. et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531537 (1999). Al-Hajj, M. & Clarke, M. F. Self-renewal and solid tumour stem cells Oncogene 23, 72747282 (2004). Vant Veer, L. J., Weigelt, B. Road map to metastasis. Nature Med. 9, 9991000 (2003).

Acknowledgements
We apologize to those authors whose work we could not cite directly due to space constraints. We would like to thank A. Berns, R. Bernards, R. Kortlever and S. Rodenhuis for advice and critical reading of the manuscript. B.W. and L.J.v.V. are supported by the Dutch Cancer Society, the Netherlands Genomics Initiative and the EU 6th Framework Programme.

Competing interests statement

The authors declare competing financial interests: see web version for details.

Online links
DATABASES The following terms in this article are linked online to: Entrez Gene: http://www.ncbi.nlm.nih.gov/entrez/query. fcgi?db=gene cytokeratin-5 | cytokeratin-8 | cytokeratin-17 | cytokeratin-18 | cytokeratin-19 | ERBB2 | MMP1 | MMP9 | PAI1 | PAI2 | uPA | uPAR National Cancer Institute: http://www.cancer.gov/ breast cancer FURTHER INFORMATION Adjuvant Online: http://www.adjuvantonline.com/ Breast International Group: http://www. breastinternationalgroup.org/ EORTC: http://www.eortc.be/ Netherlands Cancer Institute: http://www.nki.nl WHO Classification of Tumours: http://www.iarc.fr/WHOBlueBooks/ SUPPLEMENTARY INFORMATION See online article: S1 (figure) Access to this interactive links box is free online

602 | AUGUST 2005

| VOLUME 5
2005 Nature Publishing Group

www.nature.com/reviews/cancer