© 2005 Nature Publishing Group
*Division of Experimental
Therapy and ‡Division of
The Netherlands Cancer
Correspondence to L.J.v.V.
Cytotoxic chemotherapy and/or
endocrine therapy after surgical
removal and/or radiotherapy of
the primary tumour. Adjuvant
therapy is used to ensure that all
cancer cells are destroyed.
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.
Breast cancer is the most common malignant disease
in Western women. In these patients, it is not the pri-
mary 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 mammo-
graphic 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 patient’s 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 devel-
oping 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
Improving our understanding of the molecu-
lar 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 past4–6, but only the recent data obtained by
gene-expression profiling of human breast carcino-
mas7–9 received broader attention. The DNA-microar-
ray 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
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.
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 591
© 2005 Nature Publishing Group
as well as the current thinking on the metastatic proc-
ess, 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 10–15% of patients with breast cancer
have an aggressive disease and develop distant metas-
tases within 3 years after the initial detection of the
primary tumour. However, the manifestation of metas-
tases 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
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 dis-
tinct, special types of breast tumour, which represent
5–10% 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
The risk of metastasis development increases with
the presence of lymph-node metastasis, a larger-
sized primary tumour and loss of histopathological
differentiation (grade)14–18, which are the estab-
lished breast cancer prognostic markers TABLE 2.
In patients with tumour-negative axillary lymph
nodes, vessel invasion is an additional predictor
for distant recurrence19,20 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 metas-
tases 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
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 remain-
ing 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 addi-
tional prognostic markers that characterize patients
with breast cancer who are at the highest risk of
• Current prognostic criteria only poorly predict the metastasis risk for an individual
breast cancer patient. Therefore, many women receive cytotoxic chemotherapy
• 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
• 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, poor-
prognosis’ stem cell of origin.
Table 1 | Histopathological types of invasive breast carcinoma
Histopathological type of invasive
Invasive ductal carcinoma, not otherwise
Invasive lobular carcinoma5–15% 35–50%
Mixed type, lobular and ductal features 4–5% 35–50%
Tubular/invasive cribriform carcinoma 1–6%90–100%
Mucinous carcinoma<5% 80–100%
Invasive papillary carcinoma <1–2%Unknown
Invasive micropapillary carcinoma<3%Unknown
Metaplastic carcinoma<5% Unknown
Adenoid cystic carcinoma0.1% Unknown
Invasive aprocrine carcinoma 0.3–4%Unknown
Neuroendocrine carcinoma 2–5%Unknown
Secretory carcinoma 0.01–0.15%Unknown
Acinic-cell carcinoma7 cases Unknown
Glycogen-rich, clear-cell carcinoma1–3%Unknown
Sebaceaous carcinoma 4 casesUnknown
Data from REFS 11,12.
592 | AUGUST 2005 | VOLUME 5
© 2005 Nature Publishing Group
Percentage of 2,050 cases
A statistical test that examines
more than two variables at the
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.
A characteristic of a patient that
is associated with the response
or lack of response to a
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 mark-
ers 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
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 devel-
oped prognostic markers, and what is their potential
to accurately predict metastatic potential and patient
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 tyrosine-
kinase activity. ERBB2 is overexpressed due to gene
amplification in 15–30% 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 asso-
ciated with poor outcome in patients with axillary-
lymph-node metastases, but it is not associated with
poor outcome in patients with tumour-negative lymph
nodes30. However, the ERBB2 status of breast tumours
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 Tumours12. 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 surround-
ing 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 manifesta-
tion of distant metastases might therefore be useful for
patient prognosis. The search for circulating tumour
cells started in the late 1980s, and today both immu-
nohistochemical 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 expres-
sion 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 mRNA35, 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 cytok-
eratin-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
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.
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 593
© 2005 Nature Publishing Group
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 detect-
ing circulating epithelial cells in blood37, which might
be attributed to the low sensitivity of the immunohis-
tochemical 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 corre-
lation between the presence of epithelial cells in bone-
marrow aspirates and reduced disease-free intervals and
overall survival38–41. By contrast, several studies did not
confirm the bone-marrow status to be an independent
prognostic indicator42–44. 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
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 breast49,50.
Table 2 | Breast cancer metastasis prognostic markers
MarkerUse in clinic Metastatic determinantsDetails References
Tumour sizeEstablished Tumours under 2 cm in diameter
have a low risk of metastasis;
tumours of 2–5 cm have a high risk
of metastasis; tumours over 5 cm
have a very high risk of metastasis
Established 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
Related to tumour
Histological grade EstablishedGrade 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
Related to tumour
Angioinvasion Established in
The presence of tumour emboli in
over 3 blood vessels is associated
In patients with
High protein levels of uPA and PAI1
are associated with high metastasis
Low steroid-receptor levels are
associated with metastasis
metastasis risk (5
years); related to
is associated with metastasis
In patients with
A ‘good signature’ of 70 genes is
associated with low metastasis risk;
a ‘poor signature’ of 70 genes is
associated with high metastasis risk
Tested in patients
PAI1, plasminogen activator inhibitor 1; uPA, urokinase-type plasminogen activator.
594 | AUGUST 2005 | VOLUME 5
© 2005 Nature Publishing Group
(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
A study in which data are
collected and analysed after all
measurements, interventions or
events in the participants have
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 oth-
ers, 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 metalloprotei-
nases 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 node-
negative 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 patients55–58. 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 mark-
ers58. 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 metas-
tasis. 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 disease-
free 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 pre-
dict response to therapy, in particular the response to
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 exam-
ined the correlation between only one or a few markers
and clinical outcome TABLE 3. Considering the het-
erogeneity 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 car-
cinomas 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 lumi-
nal 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 oestro-
gen-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 sub-
types have characteristic clinical behaviour, they might
also share the same therapeutic targets.
The second approach to determine gene-expres-
sion patterns that can predict the clinical behaviour of
tumours is the supervised classification method. Such
a classification method was used to identify an expres-
sion 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 can-
cer 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 metasta-
size, or a good-prognosis signature, meaning that the
development of metastases was unlikely.
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 595
© 2005 Nature Publishing Group
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 microenviron-
ment on tumour progression. The pure focus on epithe-
lial 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 lymph-
node-negative and lymph-node-positive tumours8. By
multivariate analysis, the gene-expression signature was
the strongest predictor for metastasis-free survival and
overall survival, and was independent of the other clini-
cal 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 metasta-
sis 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) group8, 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
20–25% 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-expres-
sion profile might therefore be used to tailor therapy
for individual breast cancer patients and might reduce
the number of patients who would receive unneces-
sary adjuvant systemic treatment. This gene-expres-
sion 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
Table 3 | Microarray studies on prognostic gene-expression profiles in breast cancer
Microarray type Validation Informative
Metastasis determinants References
cDNA 456 ‘intrinsic’
‘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 sets63,66
70 genes ‘Good signature’ is related to low metastasis
risk; a ‘poor signature’ is associated with
a high metastasis risk. Sensitivity: 91%,
70 genes ‘Good signature’ versus ‘poor signature’.
Sensitivity: 93%, specificity: 53%
76 genes ‘Good 76-gene signature’ versus ‘poor 76
gene-signature’. Sensitivity: 93%, specificity:
442 genes Expression of ‘serum-activated’ signature
versus no expression. Sensitivity: 91%,
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
596 | AUGUST 2005 | VOLUME 5
© 2005 Nature Publishing Group
Metagenes are linear
combinations of individual
gene-expression values. They
have the potential to classify
and predict cellular phenotypes
resulting from deregulation of
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 retrospec-
tively found that could be used to predict outcome
in patients of all age groups with lymph-node-nega-
tive 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 identi-
fied. 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 metas-
tasis 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 stud-
ied, 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
The comparison of these results to the data dis-
cussed earlier is not straightforward, as different micro-
array 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 pat-
terns of gene expression METAGENES that are associated
with lymph-node status at diagnosis and a 3-year-recur-
rence risk in breast cancer patients of all ages69. Owing to
small sample numbers, cross-validation is used to deter-
mine 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 paraffin-
embedded 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 develop-
ment in a specific organ? Massagué and colleagues
identified a set of genes in a human breast cancer cell
line, the expression of which was associated with metas-
tasis 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 serum-
response signature was tested in 295 patients with
breast cancer; these were the same patients who had
been used to validate the prognostic 70-gene-expres-
sion 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 signa-
ture76. 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 chemo-
therapy. 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 super-
vised 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.
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 597
© 2005 Nature Publishing Group
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.
Models of the metastatic cascade
The metastatic properties of tumour cells were exten-
sively 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 intrave-
nous 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 muta-
tions3 (FIG. 2). This model was actually a sequel to
Leighton’s hypothesis in 1965, which predicted that
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 metasta-
sis 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 tumour4–6,81–83. 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 equal-
sized 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 concern-
ing the putative selective nature of the metastatic
phenotype. The ‘dynamic heterogeneity’ model
proposes that metastatic subpopulations are gen-
erated 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 domi-
nance’ 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 itself86,87 (FIG. 2).
Garcia-Olmo and colleagues called for a change
in the concept of metastasis. Their provocative ‘geno-
metastasis 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 tumour88,89 (FIG. 2).
Recently, it has been shown that the genetic back-
ground from which cancer arises also has an affect
on the capacity of mouse mammary tumour cells to
metastasize90. These findings indicate that the propen-
sity to metastasize is, in part, influenced by the normal
genetic make-up of the host.
The observation that the stromal microenviron-
ment functionally contributes to the development of
breast carcinomas (reviewed in REFS 91,92) indicates
that breast cancer progression might be more com-
plex than the current linear theory of activation and
inactivation of oncogenes and tumour-suppressor
genes. Therefore, the key question, whether the find-
ings made in animal and also in in vitro models can be
directly compared to the metastasis of breast tumours
in patients, remains arguable.
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 metastasis4–6,81,82. 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 itself86,87. 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.
598 | AUGUST 2005 | VOLUME 5
© 2005 Nature Publishing Group
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 pro-
files 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 associ-
ated with metastasis9. Furthermore, it was reported
that pairs of human primary breast carcinomas and
their distant metastases, which developed years later,
are highly similar at their transcriptome level93, as
are pre-malignant, pre-invasive, and invasive
A variation to this model was proposed by Massagué
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 sub-
population 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 meta-
static 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 altera-
tions of disseminated tumour cells in the bone marrow
of patients without clinical evidence of metastases gener-
ally did not resemble those of the primary tumours, in
contrast to tumour cells of patients with manifest metas-
tases. 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
An alternative, attractive model of metastasis is based
on the finding that tumours might contain ‘cancer stem
cells’ — rare cells with indefinite proliferative poten-
tial 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 immunocompro-
mised mice, were found to have the ability to form
new tumours48. Al-Hajj and colleagues were able to
distinguish tumour-initiating cells from most non-
tumorigenic 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
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
Figure 3 | New models of the metastatic process in
breast cancer. a | Gene-expression profiling of human
primary breast tumours can predict metastasis risk (‘poor-
prognosis’ (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 non-
tumorigenic 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.
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 599
© 2005 Nature Publishing Group
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 subpopula-
tions, as proposed by Fidler and colleagues3. These vari-
ant cells that are capable of forming metastases, which
have been shown by LURIADELBRUCK FLUCTUATION ANALYSIS
to be generated at a rate of about 5 × 10–5 per cell per
generation84, might, in fact, be a small population of can-
cer 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 transform-
ing 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 tumori-
genesis. By contrast, genetic events that occur in differ-
entiated 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 propen-
sity 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.
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.
New prognostic markers of breast cancer metas-
tasis 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 ran-
domized trials. It is believed that the current health-
care system cannot take the financial burden to test
all new tumour markers in this way. Alternatively,
based on the great predictive power of gene-expres-
sion 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 identifica-
tion of tissue-specific signatures for metastasis would
not only improve our understanding of the mecha-
nisms 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 stud-
ies, which included 24 and 42 patients respectively,
have so far reported the association between a gene-
expression 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 cru-
cial 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 pro-
filing 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 spe-
cificity, tumorigenesis, and probably also the cancer
stem cells98, is also important. This knowledge will
allow the development of new therapeutic strate-
gies targeting both ‘seed and soil’, which might lead
to more effective intervention strategies for breast
cancer and breast cancer metastasis.
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 poor-
prognosis 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.
600 | AUGUST 2005 | VOLUME 5
© 2005 Nature Publishing Group
1. 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,
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 1–3, 2000. J. Natl Cancer Inst.
93, 979–989 (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, 893–895 (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, 309–315
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, 462–472 (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, 71–76 (1981).
van ‘t Veer, L. J. et al. Gene expression profiling predicts
clinical outcome of breast cancer. Nature 415, 530–536
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,
Ramaswamy, S., Ross, K. N., Lander, E. S. & Golub, T. R.
A molecular signature of metastasis in primary solid
tumours. Nature Genet. 33, 49–54 (2003).
10. Bernards, R. & Weinberg, R. A. A progression puzzle.
Nature 418, 823 (2002).
11. Hellman, S. & Harris, J. R. in Diseases of the Breast (eds
Harris, J. R., Lippman, M. E., Morrow, M. & Osborne, C. K.),
407–423 (Lippincott Williams & Wilkins, Philadelphia, 2000).
Describes the clinical behaviour of untreated breast
cancer, including the incidences of regional lymph-
node and distant metastasis.
12. 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).
13. Lee, Y. T. Breast carcinoma: pattern of metastasis at
autopsy. J. Surg. Oncol. 23, 175–180 (1983).
14. Page, D. L. Prognosis and breast cancer. Recognition of
lethal and favourable prognostic types. Am. J. Surg.
Pathol. 15, 334–349 (1991).
15. Koscielny, S. et al. Breast cancer: relationship between
the size of the primary tumour and the probability of
metastatic dissemination. Br. J. Cancer 49, 709–715
16. 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, 1239–1251 (1989).
17. 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, 181–187 (1989).
18. 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, 403–410 (1991).
19. 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, 41–47 (1994).
20. 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,
21. Hellman, S. Karnofsky Memorial Lecture. Natural history of
small breast cancers. J. Clin. Oncol. 12, 2229–2234
22. James, J. J. et al. Bone metastases from breast
carcinoma: histopathological–radiological correlations and
prognostic features. Br. J. Cancer 89, 660–665 (2003).
23. Borst, M. J. & Ingold, J. A. Metastatic patterns of invasive
lobular versus invasive ductal carcinoma of the breast.
Surgery 114, 637–641 (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, R149–R156 (2004).
25. McGuire, W. L. & Clark, G. M. Prognostic factors and
treatment decisions in axillary-node-negative breast
cancer. N. Engl. J. Med. 326, 1756–1761 (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, 1456–1466 (1996).
27. Ransohoff, D. F. Bias as a threat to the validity of cancer
molecular-marker research. Nature Rev. Cancer 5,
28. Slamon, D. J. et al. Human breast cancer: correlation of
relapse and survival with amplification of the HER-2/neu
oncogene. Science 235, 177–182 (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,
30. Ross, J. S. et al. The Her-2/neu gene and protein in breast
cancer 2003: biomarker and target of therapy. Oncologist
8, 307–325 (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,
32. Ross, J. S. et al. Targeted therapy in breast cancer: the
HER-2/neu gene and protein. Mol. Cell Proteomics 3,
33. Dvorak, H. F. Tumours: wounds that do not heal.
Similarities between tumour stroma generation and wound
healing. N. Engl. J. Med. 315, 1650–1659 (1986).
34. Ring, A., Smith, I. E. & Dowsett, M. Circulating tumour
cells in breast cancer. Lancet Oncol. 5, 79–88 (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, 3404–3412
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,
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, 1392–1400 (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, 1652–1658
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, 27–33 (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, 525–533 (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, 3469–3478 (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, 23–30 (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, 196–201 (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,
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,
46. Fidler, I. J. Metastasis: quantitative analysis of distribution
and fate of tumour embolilabeled with 125 I-5-iodo-2ʹ-
deoxyuridine. J. Natl Cancer Inst. 45, 773–782 (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, 3457–3464 (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,
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, 693–706 (2002).
50. Clarke, R. B. et al. A putative human breast stem cell
population is enriched for steroid receptor-positive cells.
Dev. Biol. 277, 443–456 (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, 1–22 (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, 388–397 (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, 303–312 (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,
55. Foekens, J. A. et al. The urokinase system of plasminogen
activation and prognosis in 2780 breast cancer patients.
Cancer Res. 60, 636–643 (2000).
56. Konecny, G. et al. Her-2/neu and urokinase-type
plasminogen activator and its inhibitor in breast cancer.
Clin. Cancer Res. 7, 2448–2457 (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, 913–920 (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,
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,
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, 4617–4622 (2002).
61. Perou, C. M. et al. Molecular portraits of human breast
tumours. Nature 406, 747–752 (2000).
Classifies primary breast tumours into 5 subtypes
on the basis of similarities in their gene-expression
62. Sorlie, T. et al. Gene expression patterns of breast carcinomas
distinguish tumour subclasses with clinical implications. Proc.
Natl Acad. Sci. USA 98, 10869–10874 (2001).
63. Sorlie, T. et al. Repeated observation of breast tumour
subtypes in independent gene expression data sets. Proc.
Natl Acad. Sci. USA 100, 8418–8423 (2003).
64. Egeblad, M. & Werb, Z. New functions for the matrix
metalloproteinases in cancer progression. Nature Rev.
Cancer 2, 161–174 (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, 3817–3827 (2001).
66. Tuma, R. S. A big trial for a new technology: TransBIG
project takes microarrays into clinical trials. J. Natl Cancer
Inst. 96, 648–689 (2004).
67. Wang, Y. et al. Gene-expression profiles to predict distant
metastasis of lymph-node-negative primary breast cancer.
Lancet 365, 671–679 (2005).
68. Tan, P. K. et al. Evaluation of gene expression
measurements from commercial microarray platforms.
Nucleic Acids Res. 31, 5676–5684 (2003).
69. Huang, E. et al. Gene expression predictors of breast
cancer outcomes. Lancet 361, 1590–1596 (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, 607–616 (2004).
NATURE REVIEWS | CANCER
VOLUME 5 | AUGUST 2005 | 601
© 2005 Nature Publishing Group Download full-text
71. Paik, S. et al. A multigene assay to predict recurrence of
tamoxifen-treated, node-negative breast cancer. N. Engl.
J. Med. 351, 2817–2826 (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, 3315–3319 (2005).
73. Kang, Y. et al. A multigenic program mediating breast
cancer metastasis to bone. Cancer Cell 3, 537–549 (2003).
74. Minn, A. J. et al. Distinct organ-specific metastatic
potential of individual breast cancer cells and primary
tumours. J. Clin. Invest. 115, 44–55 (2005).
Shows that organ-specific metastasis of a breast
cancer cell line is determined by the poor-prognosis
signature and an additional metastasis-specific
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,
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, 3738–3743 (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, 171–178 (2005).
78. Fidler, I. J. Selection of successive tumour lines for
metastasis. Nature New Biol. 242, 148–149 (1973).
79. Leighton, J. Inherent malignancy of cancer cells possibly
limited by genetically differing cells in the same tumour.
Acta Cytol. 9, 138–140 (1965).
80. Stackpole, C. W. Distinct lung-colonizing and lung-
metastasizing cell populations in B16 mouse melanoma.
Nature 289, 798–800 (1981).
81. Vaage, J. Metastasizing potentials of mouse mammary
tumours and their metastases. Int. J. Cancer 41, 855–858
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, 252–259 (1980).
83. Weiss, L., Holmes, J. C. & Ward, P. M. Do metastases
arise from pre-existing subpopulations of cancer cells?
Br. J. Cancer 47, 81–89 (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, 998–1001 (1984).
85. Ling, V., Chambers, A. F., Harris, J. F. & Hill, R. P.
Quantitative genetic analysis of tumour progression.
Cancer Metastasis Rev. 4, 173–192 (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, 1263–1267 (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, 597–629 (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, 1159–1164
89. Garcia-Olmo, D. & Garcia-Olmo, D. C. Functionality of
circulating DNA: the hypothesis of genometastasis. Ann.
NY Acad. Sci. 945, 265–275 (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,
91. Bissell, M. J. & Radisky, D. Putting tumours in context.
Nature Rev. Cancer 1, 46–54 (2001).
92. Bhowmick, N. A., Neilson, E. G. & Moses, H. L. Stromal
fibroblasts in cancer initiation and progression. Nature
432, 332–337 (2004).
93. Weigelt, B. et al. Gene expression profiles of primary
breast tumours maintained in distant metastases. Proc.
Natl Acad. Sci. USA 100, 15901–15905 (2003).
94. Ma, X. J. et al. Gene expression profiles of human breast
cancer progression. Proc. Natl Acad. Sci. USA 100,
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,
96. Reya, T., Morrison, S. J., Clarke, M. F. & Weissman, I. L.
Stem cells, cancer, and cancer stem cells. Nature 414,
97. Singh, S. K. et al. Identification of human brain tumour
initiating cells. Nature 432, 396–401 (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,
99. Dontu, G., El-Ashry, D. & Wicha, M. S. Breast cancer,
stem/progenitor cells and the estrogen receptor. Trends
Endocrinol. Metab. 15, 193–197 (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, 362–369
101. Ayers, M. et al. Gene expression profiles predict complete
pathologic response to neoadjuvant paclitaxel and fluorouracil,
doxorubicin, and cyclophosphamide chemotherapy in breast
cancer. J. Clin. Oncol. 22, 2284–2293 (2004).
102. Quackenbush, J. Computational analysis of microarray
data. Nature Rev. Genet. 2, 418–427 (2001).
103. 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, 14863–14868
104. Golub, T. R. et al. Molecular classification of cancer: class
discovery and class prediction by gene expression
monitoring. Science 286, 531–537 (1999).
105. Al-Hajj, M. & Clarke, M. F. Self-renewal and solid tumour
stem cells Oncogene 23, 7274–7282 (2004).
106. Van’t Veer, L. J., Weigelt, B. Road map to metastasis.
Nature Med. 9, 999–1000 (2003).
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 ver-
sion for details.
The following terms in this article are linked online to:
Entrez Gene: http://www.ncbi.nlm.nih.gov/entrez/query.
cytokeratin-5 | cytokeratin-8 | cytokeratin-17 | cytokeratin-18 |
cytokeratin-19 | ERBB2 | MMP1 | MMP9 | PAI1 | PAI2 | uPA |
National Cancer Institute: http://www.cancer.gov/
Adjuvant Online: http://www.adjuvantonline.com/
Breast International Group: http://www.
Netherlands Cancer Institute: http://www.nki.nl
WHO Classification of Tumours: http://www.iarc.fr/WHO-
See online article: S1 (figure)
Access to this interactive links box is free online
602 | AUGUST 2005 | VOLUME 5