Profiling cancer Marco Ciro, Adrian P Bracken and Kristian Helin

213
Profiling cancer
Marco Ciro, Adrian P Bracken and Kristian Helin
In the past couple of years, several very exciting studies have
demonstrated the enormous power of gene-expression profiling
for cancer classification and prediction of patient survival. In
addition to promising a more accurate classification of cancer
and therefore better treatment of patients, gene-expression
profiling can result in the identification of novel potential targets
for cancer therapy and a better understanding of the molecular
mechanisms leading to cancer.
parameters, such as patient history, tumour histology and
the expression of often not very reliable tumour markers.
Third, since cancer therapy is based on cancer classification, it is a difficult task to choose the optimal treatment.
In the future, if oncologists are to optimally tailor the
treatment of each patient, they will need information
regarding the diagnosis and prognosis of the cancer based
on an understanding of specific genetic alterations.
Addresses
European Institute of Oncology, Department of Experimental Oncology,
via Ripamonti 435, 20141 Milan, Italy
Correspondence: Kristian Helin; e-mail: khelin@ieo.it
Since cancer is a genetic disease, and it is believed that
the specific genetic changes in the cancer determine the
phenotype, it should be possible to base cancer classification on the specific expression pattern of cellular genes
(the transcriptome). The invention of DNA microarray
technology [3–5] has made it possible to test this prediction and, as we will describe in this review, several studies
have shown that gene-expression profiling can be used to
accurately diagnose tumours and to predict clinical outcome. In a few cases, the use of microarray technology has
been instrumental in identifying novel genes involved in
cancer, and although the technology as such may not give
us an understanding of the molecular mechanisms that
result in cancer, it will provide us with several testable
hypotheses. The hope is that the use of microarray
technology in combination with classical gene function
characterisation will result in the development of highly
specific drugs tailored to the treatment of specific subsets
of cancers.
Current Opinion in Cell Biology 2003, 15:213–220
This review comes from a themed issue on
Cell regulation
Edited by Pier Paolo di Fiore and Pier Giuseppe Pelicci
0955-0674/03/$ – see front matter
ß 2003 Elsevier Science Ltd. All rights reserved.
DOI 10.1016/S0955-0674(03)00007-3
Abbreviations
ALL
acute lymphoblastic leukaemia
AML
acute myelogenous leukaemia
BRCA1
breast cancer 1
CGH
comparative genomic hybridisation
DLBCL
diffuse large-B-cell lymphoma
EWS/FLI Ewing’s sarcoma transforming fusion protein
EZH2
enhancer of zeste homologue 2
PDGFRa platelet-derived growth factor receptor a
pRB
retinoblastoma protein
siRNA
small interfering RNA
In this review, we briefly summarise the major recent
advances obtained through the determination of cancer
gene-expression patterns, and discuss how these advances
could result in a better understanding of tumour biology.
Classifying cancer
Introduction
Cancer occurs because of genetic alterations that select
for cells that escape the regulatory mechanisms restricting
normal cell growth [1]. These genetic alterations include
point mutations, and chromosomal deletions, amplifications and translocations [2]. The large clinical heterogeneity of these alterations poses several challenges to
cancer researchers, pathologists and oncologists. First,
it is an enormous task to identify the molecular mechanisms underlying the genesis of cancer, and although a
large number of genes and molecular pathways central to
the biology of cancer have been identified, there is an
urgent need to accelerate the discovery of the key events
leading to cancer. Second, it is difficult to accurately
classify cancer, and currently used classifications are often
not based on an understanding of the genetics of cancer.
Instead, cancer classification is mostly based on empirical
www.current-opinion.com
Accurate cancer classification is essential since any effective cancer treatment is based on a detailed knowledge of
the primary tissue of origin and its histopathological
appearance. Until now, cancer classification methods
have been based on the morphology of the tumour
samples, the presence or absence of metastases and the
degree of differentiation. In some rare cases, tumour
subclasses have been delineated, for example the division
of acute leukaemia into acute myelogenous leukaemia
(AML) and acute lymphoblastic leukaemia (ALL). However, these classical methods are often inadequate, since
many morphologically similar tumours classified in this
way are found to have dramatically different clinical
outcomes and responses to treatment. Recent reports
have shown that the determination of the gene-expression profiles of tumours using microarrays is a promising
alternative approach.
Current Opinion in Cell Biology 2003, 15:213–220
214 Cell regulation
The first reports to show that gene-expression profiling
can be used for the discovery and prediction of relevant
tumour classes came from studies of leukaemias and
lymphomas [6,7]. Golub and colleagues [6] applied microarrays to a set of leukaemia samples belonging to the
AML or ALL classes. Solely on the basis of gene-expression profiles, they found that samples were clustered into
two groups, corresponding to the known AML and ALL
classes. In addition, they identified a cluster of 50 genes
that best distinguished AMLs from ALLs. Significantly,
this cluster of 50 genes was sufficient to classify AMLs
and ALLs in an independent set of leukaemia samples.
This study therefore demonstrated for the first time that
gene-expression profiling could be applied to predict
tumour classes in the absence of any previous knowledge.
of co-ordinately expressed genes well correlated with the
mitotic index. They also found clusters of genes related to
specific cell types, such as stromal cells, B and T cells,
endothelial cells, and macrophages, reflecting the potential of microarray screenings in uncovering the histological complexity of breast tumours. Importantly, the same
group recently showed that these classes could be correlated to different patients’ survival [8] and their work
therefore demonstrated that breast tumours could be
potentially classified into clinically relevant classes based
on gene expression analysis.
More recently, several other groups have applied microarray technology to study different types of tumours,
including melanomas, lymphomas and leukaemias, and
breast, lung, brain, and prostate cancers [6–9,10,11,12,
13,14,15,16,17,18,19,20]. Because of space limitations in this review, we will illustrate some of the more
interesting recent advances by focusing on a few studies
relating to breast cancer.
In breast cancer patients, the detection of lymph node
metastases at the time of surgery is currently used to
determine whether the cancer has spread or not [21]. The
determination of metastases or the likelihood of its occurrence is an essential parameter for oncologists to decide
whether a breast cancer patient should receive further
(‘adjuvant’) treatment following removal of the primary
tumour. Since adjuvant therapy is associated with several
toxic side effects, it is extremely valuable to identify
markers that can reliably predict whether an in situ breast
cancer will develop metastases or not. This prompted
van’t Veer and colleagues [14] to perform gene-expression profiling of 78 breast cancers from lymph-nodenegative patients who had not received adjuvant therapy.
The authors looked for differences in the gene-expression profiles between patients who subsequently developed metastases and patients who did not. Then, they
generated a list of 70 differentially expressed genes,
which they termed a ‘poor prognosis signature’. Significantly, they validated this prognosis signature on an
independent group of lymph-node-negative patients
and found that the predictor identified the disease outcome with high accuracy. They therefore concluded that
a signature for poor prognosis already exists in primary
breast tumours at the time of surgery and that it can be
used to precisely predict survival. Importantly, the predictive power of the proposed ‘prognosis classifier’ is
much stronger than currently used methods, especially
for identifying those patients who did not relapse after
surgery, but would have been unnecessarily treated with
adjuvant therapy.
Breast cancer patients have different clinical outcomes
and vary in their responsiveness to treatment, despite an
overall similarity in tumour morphology. Therefore,
gene-expression profiling could represent a very promising tool, not only in delineating breast cancer molecular
profiles as compared with normal cellular profiles, but also
in deriving new clinical and therapeutically relevant
classes. Perou et al. [9] derived a molecular profile of
breast cancer by characterising 65 tumour samples from
42 different patients. Hierarchical clustering allowed the
identification of groups of samples based on gene expression only. They termed the largest cluster of genes
identified as the ‘proliferation cluster’, defined as a set
As a final example of the power of microarray technology
in the classification of breast cancer, Hedenfalk and
colleagues demonstrated that there are distinct geneexpression profiles that differentiate tumour samples with
mutations in the BRCA1 (breast cancer 1) and BRCA2
tumour suppressor genes, as compared with sporadic
cancers lacking these mutations [10]. Interestingly,
one tumour sample defined as a ‘sporadic tumour’ by
classical diagnostic methods was by contrast classified,
according to its molecular profile, as a carrier of a BRCA1
mutation. Further controls revealed that although the
BRCA1 gene was intact in this tumour, the promoter
was in fact silenced by DNA methylation. These
Subsequently, another study showed that gene-expression profiling could in addition be used to predict clinical
outcome [7]. In this study, gene-expression profiles were
determined on samples from diffuse large-B-cell lymphoma (DLBCL) patients, who are known to have highly
variable clinical outcomes combined with varying therapeutic responses. On the basis of gene-expression profiling, the tumour samples were clustered into two classes
related to the different stages of B cell differentiation,
one resembling normal germinal centre B cells, and the
other sharing the molecular profile of in-vitro-activated
B cells. Significantly, the two classes showed strong
correlation with clinical outcomes, demonstrating that
patients with germinal centre B-like DLBCL had a
better overall survival. On the basis of these results,
the authors proposed that the two classes could be
referred to as two separate diseases with probably different clinical outcomes.
Current Opinion in Cell Biology 2003, 15:213–220
www.current-opinion.com
Profiling cancer Ciro, Bracken and Helin 215
observations strongly point to the efficacy of expression
profiling in detecting changes in gene expression in the
absence of germline information. This is of particular
interest, as it is becoming increasingly clear that epigenetic events are very important determinants of tumour
development [22].
Identifying key players in human cancer
Although several studies, including those mentioned
above, have shown the potential use and power of
gene-expression profiling for the classification of cancer,
there are significantly fewer reports in which geneexpression profiling has been applied successfully to
identify key genes critical for tumourigenesis. Indeed,
thus far we have learned relatively little about the molecular mechanisms leading to cancer by studying the
prognosis classifiers identified in the various studies.
Despite this limited success, the hope is that geneexpression profiling will contribute to the identification
of the key genes critical for tumourigenesis. Identification of such genes is of great importance, since these
genes are potential targets for the development of
tumour-specific drugs with fewer toxic side-effects.
Moreover, the identification of these genes will lead to
a better understanding of tumour biology. Below, we
describe a few success stories in which microarray technology has led to the identification of potential key
regulators of human cancer.
The first story comes from the work of Trent and colleagues. A new taxonomy of melanomas was recently proposed, with possible clinical and therapeutic implications
[11]. Cluster analysis based on gene-expression data
classified 31 melanoma samples into two main groups.
These groups were very different in the expression of
genes involved in cell motility and invasion, suggesting
that the new classes differ in their metastatic potential, as
was subsequently confirmed by in vitro scratch healing
assays. Interestingly, Wnt5a, a member of the Wnt family
of ligands [23], was among the genes that strongly correlated with enhanced motility and invasiveness. In agreement with a possible involvement of Wnt5a in metastasis,
the same laboratory showed that the overexpression of
Wnt5a increased the metastatic phenotype of melanoma
cells and that an antibody to the Wnt5a receptor,
Frizzled-5, was effective in inhibiting the invasive capacity of aggressive melanoma cells in vitro [24]. Therefore,
in addition to suggesting a gene-expression profile for
metastatic melanomas, these studies also pointed to the
Wnt5a signalling pathway as an attractive target for therapeutic inhibition of tumour invasivity in melanomas.
Medulloblastoma is a highly metastatic brain tumour of
childhood that is difficult to distinguish from other brain
tumours using the classical histopathological methods.
The main predictor of patient survival is the metastatic
spread and, as in the cases of melanoma and breast cancer,
www.current-opinion.com
there is an urgent need for reliable genetic markers that
can clearly differentiate the metastatic from the nonmetastatic states. In a recent study, MacDonald et al.
used gene-expression profiling to identify 85 genes differentially expressed between metastatic and non-metastatic medulloblastomas [25]. They found that the
PDGFRa (platelet-derived growth factor receptor a)
and members of the Ras/MAPK (mitogen-activated protein kinase) downstream pathway were slightly overexpressed in metastatic tumours, both at the mRNA and
protein levels. Furthermore, they showed that an antibody against PDGFRa or specific inhibitors of the Ras/
MAPK pathway reduced the cell adhesion and migration
properties of one medulloblastoma cell line in vitro.
These findings suggest that this signalling pathway is
involved in metastatic spread and may suggest that inhibitors of PDGFRa or of the Ras/MAPK pathway can be
used in the treatment of medulloblastoma.
As a final example of how microarray studies have helped
the identification of key markers in cancer progression,
Dhanasekaran et al. have successfully applied geneexpression profiling to the study of prostate cancer
[20]. Prostate tumours are not only one of the most
common malignancies in males, but are also among the
most clinically and morphologically heterogeneous cancers, being almost incurable if they present metastatic
spread. The authors described a molecular signature of
prostate cancer progression by analysing cDNA arrays
from normal prostate and from localised and metastatic
prostate specimens. They found 55 genes to be significantly upregulated in metastatic prostate cancer. Interestingly, one of the strongest markers in the list was
the Polycomb protein enhancer of zeste homologue 2
(EZH2). In a follow-up to this study, the same group
confirmed that EZH2 is highly expressed in metastatic
prostate cancer, as compared with benign tissues [26].
Significantly, high EZH2 expression was found to correlate with poor clinical outcome. In addition, inhibition of
EZH2 expression by small interfering RNA (siRNA)
resulted in a significant impairment in cellular growth
of prostate cancer cells. These findings strongly suggest
that EZH2 is a marker for prostate cancer progression
with a potential role in the regulation of cell growth.
Further studies will be required to determine if EZH2 is
actually causal of this cancer progression in addition to
being an excellent marker.
Current limitations in gene-expression
profiling
It is clear that gene-expression profiling has been highly
successful in predicting the tumour gene subclasses and
in identifying potential novel key regulatory genes
involved in the genesis of cancer. However, it is important
to realise that the technique is also very limited, since it
exclusively relies on measuring changes in gene expression. For example, none of the above screens identified
Current Opinion in Cell Biology 2003, 15:213–220
216 Cell regulation
Figure 1
Uncontrolled
proliferation
Cancer phenotypes
Resistance to apoptosis
and growth arrest
New
Metastasis
RAS
p16
BMI1
CycD
TBX2
CDK4
?
WNTs
E-cadherin
CycE
CDK2
p14
?
Pathway
pRb
MDM2
E2F
p53
β-catenin
TCF
?
PKC
?
Motility
Target genes
CDC6
p21
CCND1
...
CCNE1
FAS
MYC
...
CDC25A
BAX
...
...
MCMs
MDM2
...
...
p14
APAF1
...
APAF1
...
Current Opinion in Cell Biology
From gene expression to functional pathways? The acquisition of specific cancer phenotypes, such as uncontrolled proliferation and resistance to
apoptosis, is a consequence of specific genetic alterations that determine a global change in gene expression. In the past decade, classical biology
and genetics have shown that alterations in the ARF ! MDM2 ! p53 and the p16 ! Cyclin D ! pRB ! E2F pathways are common in most
human cancers, resulting in insensitivity to antigrowth stimuli and aberrant proliferation, respectively. The critical players of these pathways are shown.
Genes that may be found to have altered gene-expression levels or already have been detected as overexpressed by microarray analysis are shown in
red. Significantly, important oncogenes such as RAS or key tumour suppressors such as p53, pRB or p14ARF (p14) have not been identified by geneexpression analysis of tumours. Moreover, alterations in the levels of genes such as APAF1 or CDKN1A (p21), whose expression is suppressed in
certain cancers, are difficult to detect by microarray analysis. The limitation of microarray technology in identifying key players in cancer is due to the
fact that the technology relies on detection of gene expression only. Oncogenes can be activated by point mutations without affecting the level of
expression, and point mutations or subtle changes in the level of tumour suppressor gene expression are sufficient to result in tumourigenesis, but will
be undetected. Therefore, we believe that the major contribution of microarray technology in the current effort to identify key molecular players in
cancer will be in the identification of functional clusters of genes shared between different cancer types. By surveying these gene clusters, traditional
biology and genetic approaches may gain insights into novel cancer phenotypes and key pathways central to human cancer, with new potential
oncogenes and tumour suppressor genes. As an example, the WNT5a gene, initially identified by gene-expression profiling as a robust marker of
metastatic behaviour in melanomas, has been proposed recently as a critical regulator of tumour invasiveness in melanoma cells through the activation
of the protein kinase C (PKC) pathway. CCND1, cyclin D1; CCNE1, cyclin E1; CDK, cyclin-dependent kinase; CycD, cyclin D; MCM, mini chromosome
maintenance deficient; TCF, T-cell-specific transcription factor.
the RAS oncogene as being an important player in the
genesis of tumours (see also Figure 1). The reason for this
is that RAS is activated by point mutations, and not
overexpression [27], and the microarrays currently in
use will not pick up such changes. Decreased levels of
tumour suppressor genes such as TP53 (encoding p53) or
RB1 (encoding retinoblastoma protein [pRB]) have also
not been identified in expression profiling, since often
only one of the alleles of these genes is lost in cancer and
the other mutated allele is expressed.
Current Opinion in Cell Biology 2003, 15:213–220
Although gene-expression profiling is sensitive enough
to detect even twofold changes in expression levels,
most studies have been performed using heterogeneous
tumour tissue, which contains non-tumour cells in addition to the tumour cells. Therefore, most gene-expression profiles published so far have not been able to
detect twofold changes in the tumour cells themselves.
To solve this problem several researchers are in the
process of using microdissected tumour tissue [28],
and it will be interesting to see whether gene-expression
www.current-opinion.com
Profiling cancer Ciro, Bracken and Helin 217
profiling of the isolated tumour cells will lead to the
identification of the classical tumour suppressor genes as
part of prognosis classifiers.
Another problem for the identification of the critical
alterations of important pathways in cancer is posed by
the high degree of genetic instability observed in the
tumours that results in a global change in gene expression,
affecting many cellular processes at a time. This makes
the finding of potential key genes even more difficult.
Therefore, to make sense of the complex molecular
profiles of cancers, researchers need to build cellular
models that reflect what might occur in the tumour. In
this way, it could be possible to identify potentially
interesting genes, which subsequently can be tested in
functional assays to assess their importance in cancer
progression.
To illustrate this point, Lessnick and colleagues [29]
have recently demonstrated the utility of linking data
from transcriptional profiling to traditional functional
assays in Ewing’s sarcoma. They performed microarray
expression profiling of a cell line expressing an inducible
form of the EWS/FLI (Ewing’s sarcoma transforming
fusion protein) oncogene fusion protein. In this setting,
cells underwent growth arrest upon induction of the
oncogene, presumably due to the activation of a tumour
suppressor. Interestingly, many genes linked to growth
suppression were upregulated by EWS/FLI. Among
these genes, TP53 was identified as the most likely
candidate to mediate growth arrest. Subsequently, the
authors demonstrated that the EWS/FLI oncogene stress
response caused an early p53-dependent but pRB-independent growth arrest. Therefore, they proposed p53 as a
potential tumour suppressor in Ewing’s sarcoma. In
addition, since the EWS/FLI-expressing cells were
found to eventually undergo a late growth inhibition
— even if the p53 or pRB pathways were impaired —
the authors also speculated that additional growth inhibitory pathways, and therefore tumour suppressors,
remain to be identified.
It is likely that in the very near future the genome-wide
approach to gene expression will be improved by new
array-derived technologies [30], overcoming some of the
current limitations. In particular, new technologies are
being developed that will allow screening for global
changes in protein expression [31,32] or for protein–
ligand interactions [33]. Furthermore, the combination
of existing techniques will most probably result in a better
understanding of cancer genetics. For example, by implementing comparative genomic hybridisation (CGH) with
cDNA microarray profiling, Pollack and colleagues were
able to link alterations in copy number to specific geneexpression profiles in breast tumours [34,35]. Finally, a
new array-based technique has been developed that
allows for the simultaneous analysis of alterations in gene
www.current-opinion.com
expression and promoter methylation on the basis of the
identification of expressed CpG-island sequence tags
(ECIST) [36]. This technique is important, since DNA
methylation is a common epigenetic event that affects
gene transcription in cancer cells.
Conclusions
Undoubtedly, the ability of microarray technology to
survey the complex molecular profiles of cancers has
enormous potential to improve the methods currently
used by clinicians to classify cancers and to predict
patient survival. However, the potential application of
gene-expression profiling as a routinely used diagnostic
method is an open question, in part because of feasibility
and cost. Carefully designed and standardised experimental approaches together with the generation of special facilities to which hospitals can send tumour tissues
for gene-expression analyses should make gene-expression profiling feasible. Meanwhile, the development of
mass-produced arrays for specific tumours and more
competition among the producers of chips should
dramatically decrease the cost.
Another issue that still needs to be addressed is the
precision of the various ‘prognosis clusters/signatures’.
This will require the testing of larger numbers of samples,
the development of common protocols for sample preparation and data mining, and the possibility for researchers to compare independent datasets from different
laboratories. There are several essential steps to reach
this goal which, in addition to better interactions between
clinicians, pathologists and researchers, would require the
generation of databases where researchers are invited to
deposit their primary datasets. From these public databases, scientists will be able to derive invaluable clues for
their research.
To obtain more knowledge regarding the molecular
mechanisms that lead to cancer, gene-expression profiling will certainly have an important role in the coming
years. When any biologist considers the fact that all
cancer gene-expression profiles show high levels of
proliferation-related genes, it is of course not surprising
that the growth-control pRB pathway is found deregulated in most human cancers. However, in addition to
uncontrolled growth, it is becoming increasingly evident that additional common sets of altered phenotypes
are shared among all cancer forms, for instance the
ability to metastasise and resistance to apoptosis. This
is despite the fact that thousands of different cancers
exist with complex morphologies and various genetic
changes [1]. Therefore, it is logical to think that by
surveying the available gene-expression databases, new
functional gene clusters can be identified that may
reveal new cellular pathways that are important for
cancer progression (Figure 1). Upon further study using
classical biological and genetic approaches, these new
Current Opinion in Cell Biology 2003, 15:213–220
218 Cell regulation
monitoring by hybridization to high-density oligonucleotide
arrays. Nat Biotechnol 1996, 14:1675-1680.
pathways will probably reveal new oncogenes and
tumour suppressors.
6.
Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M,
Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA et al.:
Molecular classification of cancer: class discovery and class
prediction by gene expression monitoring. Science 1999,
286:531-537.
7.
Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A,
Boldrick JC, Sabet H, Tran T, Yu X et al.: Distinct types of diffuse
large B-cell lymphoma identified by gene expression profiling.
Nature 2000, 403:503-511.
8.
Sørlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H,
Hastie T, Eisen MB, van de Rijn M, Jeffrey SS et al.: Gene
expression patterns of breast carcinomas distinguish tumor
subclasses with clinical implications. Proc Natl Acad Sci USA
2001, 98:10869-10874.
9.
Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS,
Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA et al.:
Molecular portraits of human breast tumours. Nature 2000,
406:747-752.
Update
To further validate the prognostic potential of the breastcancer-specific gene-expression profile identified by van’t
Veer et al., the same group recently extended the study to
a cohort of 295 patients [37]. This time, both lymphnode-negative or lymph-node-positive breast cancer were
analysed in patients younger than 53 years. The authors
classified 115 patient to the good prognosis group and 180
patients to the bad prognosis group, based on the expression of the 70 predictor genes previously identified. The
prognosis signature best performed in predicting the risk
of distant metastases and the overall survival within the
first five and ten years. In addition, Kaplan-Meyer analysis
of the probability of remaining metastasis-free showed
that the gene-expression profiling is a far better predictor
than other currently used criteria based on histological
and clinical characteristics.
Interestingly, the predictive power of the molecular prognosis profile didn’t correlate with the lymph-node status.
This remarkable observation implies that the ability of a
breast tumour to develop distant metastasis is an earlyacquired capability detectable in the primary tumour by
gene-expression profiling, and that lymph-node positivity
at the time of diagnosis is not a good indicator of later
metastasis. These data strongly suggest that gene-expression profiling at the moment of surgery is a far better
outcome predictor for breast cancer than any other currently used criteria and may also be a very powerful tool
for forecasting patients who would benefit most from
adjuvant therapy.
Acknowledgements
We thank Claire Attwooll and Fraser McBlane for helpful comments on the
manuscript. The work in the authors’ laboratory is supported by grants from
the Italian Association for Cancer Research (AIRC), the Italian Foundation
for Cancer Research (FIRC), the Human Science Frontiers Science
Programme, the EU’s Fifth Framework Programme, the Association for
International Cancer Research, and the Italian Health Ministry.
References and recommended reading
Papers of particular interest, published within the annual period of
review, have been highlighted as:
of special interest
of outstanding interest
1.
Hanahan D, Weinberg RA: The hallmarks of cancer. Cell 2000,
100:57-70.
2.
Lengauer C, Kinzler KW, Vogelstein B: Genetic instabilities in
human cancers. Nature 1998, 396:643-649.
3.
Schulze A, Downward J: Navigating gene expression using
microarrays — a technology review. Nat Cell Biol 2001,
3:E190-E195.
4.
Schena M, Shalon D, Davis RW, Brown PO: Quantitative
monitoring of gene expression patterns with a complementary
DNA microarray. Science 1995, 270:467-470.
5.
Lockhart DJ, Dong H, Byrne MC, Follettie MT, Gallo MV, Chee MS,
Mittmann M, Wang C, Kobayashi M, Horton H et al.: Expression
Current Opinion in Cell Biology 2003, 15:213–220
10. Hedenfalk I, Duggan D, Chen Y, Radmacher M, Bittner M, Simon R,
Meltzer P, Gusterson B, Esteller M, Kallioniemi OP et al.: Geneexpression profiles in hereditary breast cancer. New Engl J Med
2001, 344:539-548.
The authors performed gene-expression profiling of 21 primary breast
carcinomas and identified 176 genes that were differentially expressed in
tumours with BRCA1 mutations and tumours with BRCA2 mutations.
Fifty-one of these genes were found to best differentiate BRCA1-mutation-positive, BRCA2-mutation-positive and sporadic cases of primary
breast cancer. Although a limited number of specimens were used and
independent datasets were not analysed, the study indicates that geneexpression profiles can increase the specificity of the molecular
classification of breast cancer.
11. Bittner M, Meltzer P, Chen Y, Jiang Y, Seftor E, Hendrix M,
Radmacher M, Simon R, Yakhini Z, Ben-Dor A et al.: Molecular
classification of cutaneous malignant melanoma by gene
expression profiling. Nature 2000, 406:536-540.
12. Schoch C, Kohlmann A, Schnittger S, Brors B, Dugas M,
Mergenthaler S, Kern W, Hiddemann W, Eils R, Haferlach T: Acute
myeloid leukemias with reciprocal rearrangements can be
distinguished by specific gene expression profiles. Proc Natl
Acad Sci USA 2002, 99:10008-10013.
13. Pomeroy SL, Tamayo P, Gaasenbeek M, Sturla LM, Angelo M,
McLaughlin ME, Kim JY, Goumnerova LC, Black PM, Lau C et al.:
Prediction of central nervous system embryonal tumour
outcome based on gene expression. Nature 2002, 415:436-442.
This is an outstanding study, in which the authors first showed the
feasibility of distinguishing medulloblastomas from other embryonal
CNS (central nervous system) tumours on the basis of gene-expression profiling. They also demonstrated that expression analysis could
be used to identify the desmoplastic medulloblastoma subclass
among other medulloblastomas. Finally, they performed gene-expression profiling on 60 medulloblastomas with the aim of identifying a
prognosis classifier. By a supervised learning approach they identified
a prognosis classifier containing eight genes that with high significance could predict patient survival (only 13 out of 60 classification
errors). Using the same set of medulloblastomas, the authors showed
that the prognosis classifier substantially improved the currently available prognostic parameters for survival. An independent dataset
confirming these studies could result in a very useful clinical tool to
predict patient survival.
14. van’t Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M,
Peterse HL, van der Kooy K, Marton MJ, Witteveen AT et al.: Gene
expression profiling predicts clinical outcome of breast
cancer. Nature 2002, 415:530-536.
This outstanding paper offers a wonderful example of the enormous
potential of gene-expression profiling for the prediction of clinical outcome of cancer patients. van’t Veer and colleagues performed geneexpression profiling of 78 lymph-node-negative primary breast cancers
and identified a group of 70 genes, which they called a ‘poor prognosis’
signature. They showed that this signature outperforms all currently used
clinical parameters in predicting disease outcome and they suggested
that if independent studies confirm their data, the poor prognosis signature will dramatically help in the determination of patients who would
benefit from adjuvant therapy.
www.current-opinion.com
Profiling cancer Ciro, Bracken and Helin 219
15. Shipp MA, Ross KN, Tamayo P, Weng AP, Kutok JL, Aguiar RC,
Gaasenbeek M, Angelo M, Reich M, Pinkus GS et al.: Diffuse large
B-cell lymphoma outcome prediction by gene-expression
profiling and supervised machine learning. Nat Med 2002,
8:68-74.
Similar to [7], this study used gene-expression profiling to subclassify B-cell
lymphomas. A reliable 30-gene predictor was shown to correctly classify 71
out of 77 tumours (58 DLBCLs and 19 follicular lymphomas). Furthermore,
the authors identified a 13-gene predictor that was able to predict five-year
overall survival rates with high significance. Interestingly, three genes from
this predictor were also identified in [7] to correlate with poor prognosis.
16. Armstrong SA, Staunton JE, Silverman LB, Pieters R, den Boer ML,
Minden MD, Sallan SE, Lander ES, Golub TR, Korsmeyer SJ: MLL
translocations specify a distinct gene expression profile that
distinguish a unique leukemia. Nat Genet 2002, 30:41-47.
This paper shows that the subset of human acute leukaemias with a
translocation of the mixed-lineage leukaemia gene (MLL) have a specific
gene-expression signature significantly different from the AML and ALL
profiles. Gene-expression profiling was applied to 20 ALL, 20 AML and 17
MLL patients. The authors showed that MLL has a distinct molecular
profile from AML and ALL and expresses many myeloid- and lymphocytespecific genes. In addition, the authors developed a three-class predictor
based on 100 genes that correctly classified 10 out of 10 independent
samples. The authors suggested that MLL leukaemia should be referred
to as a distinct disease.
17. Singh D, Febbo PG, Ross K, Jackson DG, Manola J, Ladd C,
Tamayo P, Renshaw AA, D’Amico AV, Richie JP et al.: Geneexpression correlates of clinical prostate cancer behavior.
Cancer Cell 2002, 1:203-209.
18. Beer DG, Kardia SL, Huang CC, Giordano TJ, Levin AM, Misek DE,
Lin L, Chen G, Gharib TG, Thomas DG et al.: Gene-expression
profiles predict survival of patients with lung adenocarcinoma.
Nat Med 2002, 8:816-824.
The importance of this study is the identification of a prognosis identifier
for early-stage lung adenocarcinomas that can predict patient survival.
This prognosis identifier would, when introduced in the clinic, allow the
oncologist to selectively propose adjuvant therapy for patients with poor
prognosis.
19. Khan J, Wei JS, Ringner M, Saal LH, Ladanyi M, Westermann F,
Berthold F, Schwab M, Antonescu CR, Peterson C et al.:
Classification and diagnostic prediction of cancers using gene
expression profiling and artificial neural networks.
Nat Med 2001, 7:673-679.
20. Dhanasekaran SM, Barrette TR, Ghosh D, Shah R, Varambally S,
Kurachi K, Pienta KJ, Rubin MA, Chinnaiyan AM: Delineation of
prognostic biomarkers in prostate cancer. Nature 2001,
412:822-826.
Prostate-specific antigen (PSA) is currently used as a prognostic marker for the diagnosis of prostate cancer. However, nonmalignant conditions such as benign prostatic hyperplasia (BPH) also result in
elevated levels of PSA. To identify better markers for prognosis, the
authors of this paper performed gene expression profiling and successfully identified several genes that were specifically expressed in
metastatic prostate cancer.
21. Shek LL, Godolphin W: Model for breast cancer survival: relative
prognostic roles of axilliary nodal status, TNM stage, estrogen
receptor concentration, and tumor necrosis. Cancer Res 1988,
48:5565-5569.
22. Jones PA, Baylin SB: The fundamental role of epigenetic events
in cancer. Nat Rev Genet 2002, 3:415-428.
23. Polakis P: Wnt signaling and cancer. Genes Dev 2000,
14:1837-1851.
24. Weeraratna AT, Jiang Y, Hostetter G, Rosenblatt K, Duray P, Bittner
M, Trent JM: Wnt5a signaling directly affects cell motility and
invasion of metastatic melanoma. Cancer Cell 2002, 1:279-288.
In previous work from the same group [11], WNT5a was identified as one of
the best markers for the metastatic behaviour of human melanomas. In this
paper the authors show that the Wnt5a/Frizzled-5 signalling pathway is a
critical determinant of the motility and invasion capabilities of melanoma
cells. In fact, WNT5a overexpression in melanoma cell lines caused alteration of cell morphology with increased cell motility and invasion, and
increased levels of activated protein kinase C (PKC). Furthermore, an
antibody raised against Frizzled-5 was effective in inhibiting PKC activation
and cell motility. These findings were supported by the observation that
high levels of WNT5a protein were found in high-grade tumours.
www.current-opinion.com
25. MacDonald TJ, Brown KM, LaFleur B, Peterson K, Lawlor C, Chen
Y, Packer RJ, Cogen P, Stephan DA: Expression profiling of
medulloblastoma: PDGFRA and the RAS/MAPK pathway as
therapeutic targets for metastatic disease. Nat Genet 2001,
29:143-152.
In this paper the authors analysed 10 metastatic and 13 non-metastatic
medulloblastomas and built a class predictor based on the 85 genes that
best discriminated between the two groups. Once tested on an independent set of samples, four out of five medulloblastomas were correctly
classified according to their diagnosis. Despite the low change in gene
expression, members of the PDGRA signalling pathway were picked up
for further studies and shown to be essential for the invasive potential in a
medulloblastoma cell line.
26. Varambally S, Dhanasekaran SM, Zhou M, Barrette TR, Kumar
Sinha C, Sanda MG, Ghosh D, Pienta KJ, Sewalt RG, Otte AP et al.:
The Polycomb group protein EZH2 is involved in progression of
prostate cancer. Nature 2002, 419:624-629.
In this paper, which is a continuation of [20], the authors confirmed that
the levels of one of the genes identified in their previous screen, EZH2, are
highly expressed at both the mRNA and protein levels in metastatic
prostate cancer as compared with the benign state. They correlated
the high levels of EZH2 in clinically localised prostate cancers with poor
prognosis. In agreement with a causal role of EZH2 in the formation of
metastatic tumours, abrogation of EZH2 expression by small interfering
RNA resulted in less proliferation of the prostate cancer cell line. However,
control normal cells or non-metastatic prostate cancer cells were not
tested in the experiments, and it is therefore not known if EZH2 is
specifically required for metastatic prostate cells to grow or if it is required
for normal cell growth.
27. Bos JL: The ras gene family and human carcinogenesis.
Mutat Res 1988, 195:255-271.
28. Maitra A, Wistuba II, Gazdar AF: Microdissection and the study of
cancer pathways. Curr Mol Med 2001, 1:153-162.
29. Lessnick SL, Dacwag CS, Golub TR: The Ewing’s sarcoma
oncoprotein EWS/FLI induces a p53-dependent growth arrest
in primary human fibroblasts. Cancer Cell 2002, 1:393-401.
This paper is a good example of how microarray technology can be
applied to model cellular systems with the aim of identifying cooperating
mutations in human cancer. Ewing’s sarcomas contain a small group of
chromosomal translocations, of which the most common gives rise to
the EWS/FLI fusion protein. Gene-expression profiling was performed
using human diploid fibroblasts expressing EWS/FLI, and the authors
found that the profiles resembled closely the profile of the cancer itself.
The expression of EWS/FLI induced a growth arrest, suggesting that
other genetic changes were required for tumourigenesis. Significantly,
their gene-expression profiles revealed that p53 is transcriptionally
upregulated upon EWS/FLI expression. Inhibition of p53 resulted in
the abrogation of the growth arrest, supporting a role for p53 as a
tumour suppressor in Ewing’s sarcoma. This approach may be applied
to several cancer models with the hope of identifying several other key
cooperative mutations.
30. Mohr S, Leikauf GD, Keith G, Rihn BH: Microarrays as cancer
keys: an array of possibilities. J Clin Oncol 2002, 20:3165-3175.
31. Han DK, Eng J, Zhou H, Aebersold R: Quantitative profiling of
differentiation-induced microsomal proteins using isotopecoded affinity tags and mass spectrometry. Nat Biotechnol
2001, 19:946-951.
32. Shiio Y, Donohoe S, Yi EC, Goodlett DR, Aebersold R, Eisenman
RN: Quantitative proteomic analysis of Myc oncoprotein
function. EMBO J 2002, 21:5088-5096.
33. Mirzabekov A, Kolchinsky A: Emerging array-based technologies
in proteomics. Curr Opin Chem Biol 2002, 6:70-75.
34. Pollack JR, Perou CM, Alizadeh AA, Eisen MB, Pergamenschikov A,
Williams CF, Jeffrey SS, Botstein D, Brown PO: Genome-wide
analysis of DNA copy-number changes using cDNA
microarrays. Nat Genet 1999, 23:41-46.
35. Pollack JR, Sørlie T, Perou CM, Rees CA, Jeffrey SS, Lonning PE,
Tibshirani R, Botstein D, Borresen-Dale AL, Brown PO: Microarray
analysis reveals a major direct role of DNA copy number
alteration in the transcriptional program of human breast
tumors. Proc Natl Acad Sci USA 2002, 99:12963-12968.
In this paper the authors show a global analysis of genomic alterations
across 6691 genes in 44 breast tumour and 10 breast cancer cell lines
using CGH analysis. This approach allowed for the precise mapping of
recurrent regions of DNA amplification and loss. In addition, by combining
Current Opinion in Cell Biology 2003, 15:213–220
220 Cell regulation
these data with gene-expression levels from a previous report [9], they
showed that 62% of the amplified genes were also highly expressed.
Furthermore, they report that 12% of all the variations in mRNA levels
among breast tumours could be attributed to variation in gene copy
number. Therefore, these data point to a stronger than previously
expected influence of DNA copy number on the overall deregulated
levels of gene expression in tumours.
36. Shi H, Yan PS, Chen CM, Rahmatpanah F, Lofton-Day C, Caldwell
CW, Huang TH: Expressed CpG island sequence tag microarray
for dual screening of DNA hypermethylation and gene silencing
in cancer cells. Cancer Res 2002, 62:3214-3220.
Current Opinion in Cell Biology 2003, 15:213–220
37. van de Vijver M, He YD, van’t Veer LJ, Dai H, Hart AAM, Voskuil DW,
Schreiber GJ, Peterse JL, Roberts C, Marton MJ et al.: A geneexpression signature as a predictor of survival in breast cancer.
New Engl J Med 2002, 347:1999-2009.
This study is an extension of van’t Veer et al. (2002) [14]. The expression
profiles of tumours from 295 patients were analysed (including the 78
used in [14]), and the gene-expression ratios for the 70 poor prognosis
genes were examined in more detail. This study shows that the previously
identified 70 poor prognosis genes are more reliable as a prognosis
predictor than previously used criteria, and suggest that the determination of the expression of these 70 genes might be a useful tool to decide
on which patients would benefit from adjuvant therapy.
www.current-opinion.com