Download Genomics and proteomics—the way forward

Annals of Oncology16 (Supplement 2): ii30 – ii44, 2005
doi:10.1093/annonc/mdi728
Genomics and proteomics—the way forward
J. P. A. Baak1,2,3, E. A. M. Janssen1, K. Soreide1 & R. Heikkilæ4
Departments of 1Pathology and 4Medical Hemato-Oncology, Stavanger University Hospital, Stavanger, Gade Institute, 2University of Bergen,
Norway; 3Free University, Amsterdam, The Netherlands
The Post-Genomic Era is characterized by huge numbers. This will only increase in the future when
more will become known about proteins, which not only outnumber the genes, but also can be functionally altered in various ways. These "big numbers" have two consequences. Studies, which were
despisingly called Fishing Expeditions only a few years ago are now respectfully dubbed Discovery
Science, which answers questions and generate new hypotheses. Secondly, it is not always easy to
select the way forward in Genomics and Proteomics. Therefore emerging technologies are updated
and the hallmarks of cancer cells briefly discussed. Promising applications for screening/early
disease detection, diagnosis and tumour classification, prognostication, predictive response and
tailoring therapy are described.
Strategic choices as to certain types of study, which diseases or organ sites should be analyzed or
techniques used, are mainly political and very much determined by loco-regional and national interests. However, in post-genomic research the following practical points must be considered. To
avoid non-informative noise, homogeneous groups (preferably of small lesions) should be analyzed.
Adequate sampling will be even more important than in the past to minimize the risk of non-informative benign cells inclusion. Accurate, biologically relevant, well reproducible quantitative morphological information is of the utmost importance to define the lesions, as mixtures of functionally
different cells will obscure important signals. RNA and proteins degrade quickly under hypoxic conditions, so that for reliable results the interval between tumour excision and freezing must be minimized and standardized. Promising results must always be independently validated as the use of
relatively few patients combined with the enormous number of variables analyzed, carries a serious
risk of selection bias and too optimistic results. An integrated collaboration structure of multiple
disciplines becomes increasingly important.
Introduction
A mere 50 years have passed since the 1953 landmark
description of the DNA double helix [1], yet the parallel
efforts of the Human Genome Project and Celera Genomics
have successfully sequenced the human genome [2, 3]
(officially completed on April 14, 2003), thus introducing the
‘post-genomic era’ [4]. These are major achievements in
biology and medicine in general, contributing not least to the
understanding of carcinogenesis. New molecular insights and
technologies in oncology have given clues to the initiation,
early detection and possible prevention of neoplasia, prognostic and predictive markers, and targets for early detection
and therapy.
The search for cancer-causing alterations within the currently known genes of the whole genome (‘genomics’, the
study of the human genome) is complicated by the different
ways the genes may be transcribed (transcriptomics) into a
variety of functionally different proteins (‘proteomics’, the
analysis of the protein complement of the genome), which can
themselves undergo essential functional changes.
q 2005 European Society for Medical Oncology
The definition of proteomics has changed greatly over time
[5]. Originally, it was coined to describe the large-scale, highthroughput separation and subsequent identification of proteins
resolved by 2-dimensional poly-acrimide gel electrophoresis
(2DE). Currently, proteomics denotes nearly any type of technology focusing upon proteins analysis, ranging from a single
protein to thousands in one experiment. Proteomics thus has
replaced the phrase ‘protein science’.
The post-genomic era is characterized by the generation of
massive data numbers from studies with descriptive
approaches. Although previously despisingly called ‘fishing
expeditions’, today the term ‘discovery science’ is respectfully
used for such large-scale studies, crediting the answers and
new hypotheses being generated. Alongside the optimism
fuelled by new discoveries, however, there is also the sensation of being overwhelmed by their complexity and size.
Such a sensation is underlined by the notion that according to
our existing knowledge, only 1.1% of the genome consists of
exons coding for proteins, 24% is intronic sequences and the
remaining 75% consists of intergenic DNA, currently without
a known function in RNA transcription or protein translation.
ii31
Moreover, compared with evolutionary lower organisms,
human beings have only twice to three times as many genes
as the fruit fly and the mustard plant. This indicates that the
functional complexity, rather than the absolute number of
genes, is essential for the human phenotype.
To clarify where we are in the post-genomic era, we could
say that we now have the letters (the sequence) and have thus
far found a few sentences (genes that we know of), but we
have only merely begun reading the chapter contents (how
genes may be transcribed), while the books (the proteins and
the metabolites) will keep human mankind reading for many
centuries to come. In more scientific terms, the outline of the
genome has enabled the study of gene products that are the
focal point of proteomic studies, the effectors of the DNA [6].
Thus, considering the complexities of human nature and disease processes, to illustrate where cancer science is today: we
have just become aware of the alphabet-code for a vast library
of knowledge.
We must therefore accept the idea that we are only at the
beginning of a scientific discovery journey, a journey that may
not be finished for many decades, or even centuries, to come.
One may ask, how do we grasp this unfolding knowledge?
This article will attempt to give a short overview of where we
are and how the way forward might be chosen.
The genome lays a foundation for the
understanding of cancer as a genetic disease
The completion of the Human Genome Project gives us more
insight into the genetic variations that are causally implicated
in oncogenesis. A 2004 consensus report described that more
than 1% of genes are causally involved in oncogenesis [7].
Some of these genetic mutations can be transmitted through
the germ-line and result in hereditary cancers (5–10% of all).
Many hereditary cancers have been extensively described, so
that the focus now is more on the susceptibility for cancers in
certain families or populations. This has resulted in a new
approach, molecular epidemiology: searching for low penetrant cancer susceptibility genes that may give rise to much
smaller increases in individual risk [8]. These genes may
interact with environment and lifestyle factors and as a result,
cancer risk is not equally elevated in all persons exposed to an
environmental factor, and the risk can change. One example
is the change in cancer incidence in Chinese and Japanese
immigrants to the USA [9]. Another example is relatives of
patients with early-onset lung cancer, who have a genetic
predisposition and elevated risk for developing lung cancer, in
contrast to relatives of patients that develop lung cancer later
in life [10]. It was recently suggested that a T27C polymorphism in CYP17 (cytochrome P450 c17a) is associated with
elevated sex hormone levels, and interacts with insulin levels
and diet to affect breast density levels and potential breast
cancer risk [11]. Such genetic susceptibility is currently being
intensively investigated by DNA fingerprinting with the use of
single nucleotide polymorphisms (SNPs) in large populations.
From genotype to phenotype: the hallmarks of
human cancer cells
Research over the past decades has provided us with increased
insight that cancer is a genetic disease. Knudson [12] formulated the so-called ‘two-hits hypothesis’ to explain the development of retinoblastoma. A larger, stepwise number of
acquired genetic mutations were later detected in the early
colorectal cancer studies [13–15]. Today, cancer is often perceived as a genetic disease driven by the time sequence of
these DNA changes and is phenotypically determined by a
limited number of underlying rules [16, 17]. These genetic
alterations and rules are not the same for all tumors even
within a certain organ site (e.g. lung cancer). This functional
perception of cancer cells has clinical implications concerning
the biological behavior, natural history and potential therapeutic tumor targets.
Briefly, the acquired set of functional capabilities of a cancer cell are: (i) self-sufficiency in growth signals; (ii) insensitivity to anti-growth signals; (iii) evasion of apoptosis; (iv)
limitless replicative potential; (v) sustained angiogenesis; and
(vi) tissue invasion and metastasis [16].
Furthermore, extracellular matrix and its components (fibroblasts, stromal cells, signal substances, inflammatory cells)
influence cancer cell proliferation, invasion and metastasis
[18–21]. Fibroblasts have a more profound influence on the
development and progression of carcinomas than was previously appreciated [19], e.g. in oral epithelium differentiation
[22].
Molecular techniques
The molecular techniques for genomics and proteomics
studies are developing rapidly (for an overview, see Baak et al.
[23]). A short description follows here. Table 1 shows important widely established techniques with their respective
strengths and weaknesses. Some newer promising methods
will be described in detail.
Established DNA techniques
Conventional karyotyping. The development of chromosome
banding techniques in 1969 was a breakthrough, as all
chromosomes could be individually recognized (‘conventional
karyotyping’) and chromosomal rearrangements characterized
(a classical example is the Philadelphia chromosome in
chronic myeloid leukaemia).
Fluorescence in situ hybridization. A small DNA fragment of
known origin (a probe) is fluorescently labeled and hybridized
to a metaphase chromosome spread or interphase nuclei. The
probe binds to homologous sequences within the chromosomes and this can be visualized by fluorescence microscopy
to identify chromosomes, centromeres, aberrations in interphase tumor nuclei and others (Figure 1).
Comparative genomic hybridization (CGH). CGH provides
information on the number of copies of chromosome parts
throughout the whole tumor genome to identify genetic
ii32
Table 1. Molecular techniques, their strong and weaker points
Advantages
Limitations
Provides data on
Karyotyping
Whole chromosome
analysis
Laborious
Structural large chromosomal changes
Fluorescence in situ
hybridization
Gene specific
Maximum four genes at
a time
Gains and losses of
specific genes in cells
and tissue
Comparative genome
hybridization
Whole genome
analysis
Resolution limited to 10 Mbp
Gains and losses throughout
the whole genome
Spectral karytyping
Whole genome
analysis
Laborious
Structural chromosomal changes
Microarray
Whole genome
analysis
Fresh tissue necessary for
cDNA isolation
Differential expression of thousands
of genes
Microsatellite instability
Allele specific
Large numbers of microsatellites
needed to be informative
Chromosome stability, effectiveness of
DNA repair mechanisms
Loss of heterozygosity
Allele specific
No correlation with mRNA
expression
Losses at gene level
Single nucleotide
polymorphism (SNP)
Stable
Large number of SNPs
needed to be informative
Susceptibility for disease development
and therapy response
Abundant through
genome
SNPs map not finished
yet
Gene specific
Many false-negative and -positive
results
Gene function under well
described circumstances
Separation of complex
protein mixtures
Hydrophobic and membrane proteins
do not enter well
Differential protein expression profiling
of complex protein mixtures
Sensitivity
(micromolar, 106)
Posttranscriptional
modifications can be
detected
Only 2000 spots per
gel
Not suitable for small
clinical samples
Detects new protein
interactions
in vivo
False positives
Genomic techniques
RNA interference
Proteomic techniques
Two-dimensional
SDS–PAGE
Yeast two-hybrid
Protein–ligand interactions
Laborious
Protein microarray
High throughput
Data analysis
Very diverse both function
and detection
Limited amount of antibodies and
recombinant proteins available
Fluorescence microscopy
(FRET, FLIP, FRAP,
FLIM, BRET, PRIM)
Very specific, under well
described circumstances
in vivo and in time
Laborious
Spatial and temporal dynamics
of protein–ligand interactions in the
living cell
Mass spectometry and
combinations with TOF
High specificity and sensitivity
(femtomolar, 1015)
Often extensive sample preparation
Protein identification/characterization,
mass fingerprinting
Surface-enhanced laser
desorption/ionization
High speed, high resolution
separations of complex
protein mixtures
Detection >30 kDa
Differential protein expression profiling
of complex protein mixtures
Small samples
Low mass reproducibility
Tissue microarray
Little sample preparation
Large protein profile CVs
High throughput
Formalin-fixed material
Morphologic information intact
Possible on archival material
Protein–ligand interactions and protein
expression in complex protein
mixtures
Protein expression and location
in correlation with patient
and therapy data
ii33
Table 1. (Continued)
Liquid chromatography
nuclear magnetic
resonance
Advantages
Limitations
Provides data on
Quantitative
Only metabolites [amino acids,
organic acids and bases,
nucleotides, carbohydrates,
osmolytes, lipids (broad,
non-specific resonances)]
can be detected
Metabolite target analysis, metabolite
profiling, metabolic fingerprinting,
metabolic
profiling
High throughput
Expensive
Complete sample
recovery
Little sample
preparation
Complex protein
mixtures
changes (deletions or amplifications). The resolution of
chromosome CGH is limited to 10 million base pairs
(Mbp).
Array CGH. Array CGH does not require karyotyping and the
resolution can be 0.5 Mbp or better. Figure 2 explains array
CGH and shows the results of chromosome and array CGH of
the same chromosome combined in one figure.
Spectral karyotyping (SKY). SKY gives all chromosomes a
unique fluorescent color by hybridizing with chromosomespecific probes each labeled with a different combination of
Figure 1. Fluorescence in situ hybridization (FISH) examples. Top, left: interphase cytogenetics. Normal diploid (left, two spots per chromosome) and
malignant (right, several spots per chromosome, indicating its aneuploidy) urothelial cells from a urinary bladder washing hybridized with centromere
probes to chromosomes 3 (red), 7 (green) and 17 (white), and a locus-specific probe for p16 at 9p21 (yellow). Bottom: FISH using whole chromosome X
and chromosome 3 probe on tumor cell line scc040, showing that a part of chromosome X is attached to chromosome 12, and showing that the addition
on chromosome 3 is in fact other chromosome 3 material. Reprinted with permission from Baak et al. [23].
ii34
Figure 2. Top left: schematic view of the array comparative genomic hybridization (CGH) technique and higher sensitivity and resolution of array CGH
over chromosome CGH. Top right: example of a hybridization of green-labeled tumor DNA and red-labeled normal reference DNA onto an array of 2400
BAC clones representing the whole genome with an average spacing of 1 Mbp. Bottom: chromosome CGH detected low level gain of a large part of
chromosome arm 20q; array CGH identified a narrow high level amplification in band 20q13.2. Reprinted with permission from Baak et al. [23].
Figure 3. Spectral karyotyping (SKY) karyogram of tumor cell line scc040. For every pixel in the image the spectrum from 450 to 750 nm is measured,
and with this spectral information the unique color combination of each chromosome is identified. The addition on chromosome 3 is composed of chromosome 3 material, the addition on chromosome 12 comes from a part of chromosome X, and the addition on chromosome 18 consists of chromosome 8
material (the third chromosome 9 is lying in overlap with chromosome 4). Reprinted with permission from Baak et al. [23].
ii35
fluorochromes, and is useful for the genome-wide detection of
structural chromosomal changes. Translocations are readily
visible (Figure 3). SKY does not require prior knowledge of
chromosomal breakpoints.
Gene expression (cDNA) arrays. High-density oligonucleotide
cDNA microarrays measure many thousands of gene-specific
mRNAs in a single tissue sample in parallel (Figure 4). Largescale gene expression analysis has proved to be a valid strategy for developing gene expression profiles, or ‘signatures’, to
classify prognostic subgroups. Unfortunately the method
requires fresh tissue for optimal results.
Loss of heterozygozity (LOH), microsatellite instability (MSI).
One strategy for screening the genome used a panel of 150
polymorphic microsatellite markers from throughout the
whole genome with LOH. By using microsatellites one can
also check the ability of cells to repair DNA replication errors.
In tumors with MSI, both genetic and epigenetic modifications
of mismatch repair genes were identified.
Proteomics techniques
Two-dimensional electrophoresis. 2DE [16] is a powerful
technique for protein separation. The digestion of spots in the
gel makes it possible to further analyse proteins with mass
spectrometry. Peptide spectra obtained in this way can be used
to search protein sequence databases.
Yeast two-hybrid system. The two-hybrid system can generate
vast amounts of new data, but each and every complex has to
be tested and confirmed separately afterwards.
Protein microarrays. Most protein microarrays are affinitybased. Although powerful, these techniques have several
drawbacks: lack of available antibodies, lack of purified
recombinant proteins and cross-reactivity with affinity agents.
Figure 4. Genes which distinguish normal from malignant endometrium (P, proliferative; S, secretory; T, tumors) endometrium (Permax <0.50, three-fold,
100 difference). Columns show individual tissues, rows represent genes. Color scale shows standard deviation from the mean expression value for each
gene. Dendrograms on the margin show agglomerative hierarchical clustering (Wards linkage, euclidean distances) of genes (right) and tissues (bottom)
(courtesy of Dr G. L. Mutter). Reprinted with permission from Baak et al. [23].
ii36
Fluorescence resonance energy transfer (FRET). FRET monitors macromolecular interactions and resolves the spatial and
temporal dynamics of protein–protein interactions in the living cell, but in principle it can also be applied to tissue sections of fixed material.
Surface-enhanced laser desorption/ionization (SELDI). SELDI
is an affinity-based mass spectometry method in which proteins (<20 kDa) are selectively absorbed to a chemically modified surface, and impurities are removed by washing with
buffer.
Tissue microarrays (TMAs). TMAs (Figure 5) consist of hundreds of small core (0.6 –2.0 mm) tissue sections arrayed on a
glass slide for immunohistochemistry (IHC) (protein), in situ
hybridization (DNA, RNA), and are useful for evaluation of
new antibodies and large-scale outcome studies. Because of
the very small sample size per case (with two cylinders of
1 mm diameter per tumor of 2 2 cm; typically <0.5% per
section), TMAs only give a reliable impression of a tumor
characteristic if large numbers of cases are analyzed, and even
then, the significance of rare events may easily be overlooked
(see below under ‘Critical remarks and need for validation’).
Evolving molecular methods
Single nucleotide polymorphisms. SNPs are the most abundant
form of DNA polymorphisms in the human genome. Each
SNP has a defined position in a chromosome at which base
pairs differ among individuals with significant frequency
(>1%). Human SNPs are not very polymorphic, contrasting
CA repeats (see below), and therefore SNPs are often less
informative than other genetic markers such as simple
sequence-length polymorphisms and microsatellites. On the
other hand, SNPs occur abundantly in the whole genome and
provide great potential for automated genotyping. SNPs can
therefore be used as genetic markers to detect disease genes in
genetic linkage studies. Thus high-density SNP mapping can
be as informative as current strategies with simple sequencelength polymorphisms and microsatellites. As an example,
genetic polymorphisms in T27C in CYP17 have been linked
to the risk of developing breast cancer [24].
Loss of heterozygosity. Huang et al. [25] recently developed a
high-density oligonucleotide array-based SNP genotyping
method, whole genome sampling analysis (WGSA), to identify
genome-wide chromosomal gains and losses at high resolution. WGSA simultaneously genotypes over 10 000 SNPs by
allele-specific hybridization to perfect match and mismatch
probes synthesized on a single array. The coupling of LOH
analysis, via SNP genotyping, with copy number estimations
using a single array provides additional insight into the
structure of genomic alterations. With median inter-SNP
euchro- matin distances of 199 kilobases, this method affords
a resolution that is not easily achievable with non-oligonucleotide-based experimental approaches.
RNA interference (RNAi). RNA silencing [26] is a posttranscriptional gene silencing process that is based on sequencespecific interactions of small interfering RNAs with targeted
mRNA molecules. Silencing is initiated when double-stranded
Figure 5. An example of a tissue microarray (TMA) consisting of 1 mm diameter cylinders. Ki-67 staining. (A) Paraffin block for TMA; (B) histological
TMA section, with three control markers at the upper left; (C, D) examples of different immunostains.
ii37
RNA is processed into small RNAs, which triggers a number
of enzymatic reactions resulting in the degradation of the targeted mRNA molecules [27]. RNAi-based genome-wide functional analysis for mammalian cells [28] is well underway.
Combining RNAi with transfected cell array (TCA) holds
enormous potential for drug-discovery oriented research and
therapy.
Transfected cell array. The principle of TCA is based on the
transfection of DNA or RNA molecules. These are immobilized on a solid surface upon which cells are cultured. Cells
growing on top of the DNA/RNA spots become transfected,
resulting in the expression or silencing of specific
genes/proteins in spatially distinctive groups of cells. Consequently, the physiological effects caused by the introduction of
these foreign nucleic acids can be studied [29]. Densities of up
to 8000 cell clusters per standard slide can be achieved [30].
Microfluidics. Microfluidics technology utilizes a network of
channels and wells that are etched onto glass or polymer chips
to build ‘laboratories-on-chips’. Pressure or electrokinetic
forces move pico- or nanoliter volumes in a finely controlled
manner through the channels. These microfluidic circuits can
be designed to accommodate virtually any analytic biochemical process. For example, a lab-on-a-chip for immunological
assays could integrate sample input, dilution, reaction and separation; or one designed to map restriction enzyme fragments
might have an enzymatic digestion chamber followed by a
separation column. Labs-on-chips are well suited for highthroughput analyses. Their small dimensions reduce processing times and the amount of reagents necessary per assay,
and have high reproducibility owing to standardization and
automation [31].
Applications of genomics and proteomics
The long interval between a discovery and a clinical
application
It takes a long time before the real impact of a new discovery
is realized. An example is that the invention of the microscope
by Leeuwenhoeck occurred in the 17th century, yet the era of
‘cellular pathology’ was not fully introduced until the teaching
of Virchow, nearly 200 years later! Similarly, even though
p53 was discovered two decades ago, for the first 10 years it
was believed to be an oncogene. Only later was it vindicated
as a tumor-suppressor gene and the ‘guardian of the genome’
[32]. DNA contains four variables only: adenine, cytosine,
guanine and thymidine, yet it took many decades to understand its biology, and that process is far from finished. Proteins are much more complex: they contain 20 different amino
acids and the number of possible permutations thus is exponentially larger. In addition, there are more than 100 known
different possible posttranslational modifications [33] and
each one will be able to change the function and location of a
protein drastically over time and under different conditions.
The interval between the development of genomic and
proteomic technologies, and the actual paradigm shift in
the understanding of disease and development of new treatments may be very long indeed!
The conclusion of the above is that we are at the very
beginning of clinical genomic and proteomic applications.
Nonetheless, the enormous expansion of molecular cancer
research to date has created applications that are useful in the
daily care of patients. The following description can by no
means be complete, but aims to mention typical examples,
including weak points, and how new discoveries have been
made by the combined use of different technologies. Here we
will concentrate on some established or promising applications
in (i) early disease detection/screening, (ii) diagnosis and
tumor classification, (iii) prognostication, (iv) prediction of
response, and (v) tailoring of therapy.
Screening
In many countries, classical cytological screening for cervical
cancer is being amplified with testing for human papilloma
virus [34]. International genomic collaborations to detect
minimal residual cancer have produced important results in
leukemia [35]. Proteomics technologies have also been
announced as future biomarkers for screening and early
detection of cancer disease [36–38]. However, the early promising results with near 100% sensitivity and specificity for
detection of ovarian cancer with proteomics [39] have been
seriously criticized, as Baggerly et al. [40] showed that the
method performed no better than chance for classifying the
second dataset. In agreement with this, the reproducibility of
the proteomic profiling approach remains to be established.
Yet, others still hope that proteomic profiling will enable
protein-screening in adjunct with endoscopic surveillance for
improved and more effective detection of early (pre)malignant
disease [36, 41].
Tumor classification
Histological typing and grading of tumors is widely used, but
is notoriously subjective and lacks intra- and inter-observer
reproducibility [42, 43]. This is unacceptable, as the therapeutic and social consequences of neighboring grades for the
individual patient are often enormous: e.g. adjuvant
chemotherapy or radiotherapy in FIGO 1 ovarian cancer ‘grade
I–II’; and total colectomy in ulcerative colitis patients ‘with
some epithelial dysplasia’ [44]. Molecular analysis allows for
subgrouping based on genomic or proteomic (including IHC)
profiles together with histopathology evaluation in colorectal
cancer, breast cancer, lung cancer, lymphomas and others
[45–52]. A well known example is that BCL-2 (an anti-apoptotic protein) is overexpressed in follicular lymphomas, principally as a result of the t(14;18)(q32;q21), and useful in
distinguishing follicular lymphoma (usually BCL-2-positive)
from follicular hyperplasia (BCL-2-negative), although with
certain exceptions [53]. Multiplex polymerase chain reaction
(PCR) assays have successfully been developed and standardized for the detection of clonally rearranged immunoglobulin
(Ig) and T-cell receptor (TCR) genes and the chromosome
ii38
aberrations t(11;14) and t(14;18) [54], and clonality testing in
lymphomas has been widely adopted as a routine diagnostic
method. Moreover, ‘undifferentiated’ (poor prognosis) laryngeal cancers and gastric appeared to be large-cell lymphomas
(with good prognosis) when IHC panels were applied. The
classification of the diagnosis ‘unknown primary’ (in the past
applied to 20% or more of metastases) can be reduced considerably with the today’s application of accurate multi-panel
immunopathology. Hopefully, molecular techniques will render even more exact subclassifications in adjunct to histomorphological evaluation. Eventually, the diagnosis ‘unknown
primary’ may become obsolete [55]. An example is DNA
methylation mapping, which showed that unique profiles of
hypermethylated CpG islands exist, defining each neoplasia
[56], as in, for example, prostate cancer [57]. These results for
subclassification of tumors are promising, but validation and an
appropriate link to clinical outcomes is always mandatory.
Prognostic value, prediction of response and subgroup
analysis
Many articles have had a prognostic goal, as staging (like
grading and typing) is practically useful but may have a low
prognostic accuracy. Prognostic gene expression signatures
are promising, as, for example, in breast cancer [48]. CGH
analysis in lymph node-negative breast cancers led to the discovery that a gain in chromosome 3q is a much stronger prognosticator than classical features [58]. The most common
region of overlap in this study was 3q26. The PIK3CA gene
[59] located here is mutated in up to 40% of primary breast
cancers [60, 61] (Figure 6). Overexpression of PIK3CA can
result in an increased activation of the Akt pathway after
activation of ErbB2 upon binding with a growth factor
(e.g. estradiol or heregulin) [28], and lead to excessively
increased proliferation, invasion and cell motility (Figure 8).
Proliferation is the strongest prognosticator in node-negative
breast cancer [62–64]. However, many other factors, like
noey2 deletion, may also cause increased proliferation [65].
Noey2 is located on chromosome 1p, but 1p deletion was not
prognostic [58]. We hypothesize that there are two types of
high proliferation in breast cancer, a PIK3CA-dependent one
(strongly prognostic) and a second caused by genetic changes
in other genes (which are much less, or not at all, prognostic).
It is important that the tumor suppressor gene PTEN opposes
the action of PIK3CA. PTEN deficiencies occur in up to 35%
of breast tumors, and may be an important predictor of trastuzumab resistance in breast cancers with ErbB2 overexpression
[66]. Loss of PTEN tumor suppressor function is observed in
tumors of breast, prostate, thyroid and endometrial origin
[67–69]. Allelic losses in the proximity of the PTEN locus
(10q23) also occur in sporadic colorectal cancers [70]. PTEN
therefore may be a general key progression determinant in
(pre)cancers. In endometrial hyperplasias (EHs), a frequent
pre-neoplastic lesion, progression to cancer only occurs in
PTEN-negative and not in the PTEN-positive EHs [68]
(Figure 7). However, only those PTEN-null EHs with an unfavorable morphometric D-Score progressed [71], greatly
increasing the positive predictive value of PTEN-null glands
(Figures 8 and 9). This again shows the importance of
combined molecular and morphological technology to get the
strongest prognostic information.
Targeted tumor therapies
Targeted therapies are now being developed that work at
the level of the proteome. Examples of these agents
include imatinib mesylate (Gleevecw) targeting the Bcr-ABL
tyrosine kinase in chronic myelogenous leukemia and mutated
c-KIT- or PDGFRa-tyrosine kinase in gastrointestinal stromal
tumors [72], trastuzumab (Herceptinw) for (her2-neu amplified)
breast cancer [73, 74] and gefitinib (Iressaw) for EGFR-mutated
lung cancers [75]. These are all targeted therapies in that they
are monoclonal antibodies (imatinib, trastuzumab) or small
molecules (gefitinib) directed at distinct defects in the tumor
cells. Even though early clinical trials have shown
positive results, resistance patterns have been reported for gefitinib [76, 77], imatinib [78, 79] and trastuzumab [66, 67]. Consequently, several pathways should perhaps be simultaneously
addressed if long-term response or remission is to be achieved.
The problem is that such combinations would be expected to
increase the toxicity, which undermines the whole idea of
targeted therapy, but combination analysis may also prevent
overtreatment. Her2-neu and proliferation testing is a good
example. Volpi et al. [64] showed that in node-negative breast
cancer patients, HER2 expression was a significant discriminant
of prognosis, but only in the subgroup of patients with rapidly
proliferating cancers. PTEN-negative patients are more prone
to trastuzumab resistance [66]. A problem with Her2-neu is the
testing method and which patients to select for treatment. Generally accepted is to treat all IHC strongly positive (3+) cancers,
although several have no gene overexpression while others with
gene amplification do not show IHC [80]. The cause and the
clinical value of these discordances are not yet known.
The way forward
What course shall we take with genomics and proteomics in
cancer research and treatment? Which strategic and practical
choices should be made?
Strategic choices
Just as ‘All roads lead to Rome’, so the answer to ‘Genomics
and Proteomics—The Way Forward’ may be similarly easy,
but at the same time complex, for there are certainly many
possible ways to go in the post-genomic era. Eventually,
many will hopefully lead to a better understanding of human
disease, its causes, prevention and potential therapeutic
targets. Clearly, one way will lead there sooner than another;
we do not yet have a final blueprint and need to rely on
cooperation and experience in getting to our target. Strategic
choices regarding types of study, which diseases or organ sites
should be analyzed or which techniques used, are mainly
political and very much determined by locoregional and
national interests.
ii39
Growth
Factor
Receptor
Tyr
Kinase
PIP3
Ras
GTP
PIP2
PTEN
P-Tyr
p85
SH2
Phosporylation
of BRCA1
p110
PIK3CA
AKT
Mitogenesis
ACG kinases
Apoptosis
Caspase 9, BAD
Cell Motility
RAC, BTK kinase
NFkappaB, uPa
Figure 6. Schematic overview of some of the Akt pathway effectors. Overexpression of the catalytic subunit of PI-3 kinase may result in an increased
activation of the Akt-pathway, in reaction to a growth factor binding to cell surface receptor ErbB2. PTEN promotes the transition from PIP3 to PIP2,
thereby suppressing PIK3CA.
However, the following practical points must be considered
in research on ‘The way forward’.
Analyze homogeneous groups
Large tumors are often geno- and phenotypically very heterogeneous, hypoxic and necrotic. The essential information one
is searching for can be heavily blurred by the additional noise
coming from these epiphenomenal properties. Thus, the
utmost care is needed with the interpretation of results from
large tumors. It is not always immediately evident that the
material presented in an article is biased for large tumors, as,
for example, in studies using frozen tissue for RNA and protein chips. With many years of daily practical experience in
pathology laboratories, we can guarantee that frozen tumor
material often comes from large tumors! Other examples
where heterogeneity has a major influence are when tumors
from different stages are included, or mixtures of carcinoma
in situ, small and large cancers, or different histological subtypes or different age groups with varying genetic and
prognostic information. Unfortunetely, including such mixtures may be the rule rather than the exception.
Sample accurately
Adequate sampling is always important, but even more so
when the cancers are small, as is increasingly the case. In
small samples there is the added risk of including proportionally significant numbers of benign epithelial and
other cells. Consequently, accurate sampling by means of, for
Figure 7. PTEN-expressing (brown) and -null glands (blue) in (A) ‘normal’ proliferative and (B, C) hyperplastic endometrium. Note that adjacent glandular cells can be positive and negative (B). Stroma cells are invariably positive (reprinted from Baak et al. [104]).
ii40
100
D-Score>1 or
D-Score<1/PTEN+
Progression 0/89
% Without Cancer
In Follow-Up
80
60
P<0.00001
40
D-Score<1 &
PTEN Null
Progression 7/14
20
0
0
24
48
72
96
120
Follow Up (months)
Figure 8. Cancer-progression free survival curves of endometrial hyperplasia, according to combined application of the two different technologies: the morphometric D-Score and PTEN expression (reprinted from
Baak et al. [104]).
Figure 10. The bench-to-bedside approach.
breast cancer. A formal count is strongly prognostic but
impressions are not (Baak J.P.A., van Diest P.J., Janssen
E.A.M. and Voorhorst F., 2005, unpublished results).
Poor reproducibility owing to different techniques or
poor test methods
Figure 9. RNA and proteins are very sensitive to hypoxia, caused by
excision of the tumor.
example laser dissection, becomes even more important than
in the past. In pre-cancers, the precise localization in the
mucosa of a prognostic factor is often highly important to
extract its value. In a cervical squamous intraepithelial lesion,
for example, the concentration of retinoblastoma protein is
predictive of progression, but only in the deep layer of the
epithelium. Mixtures of superficial, deep and basal cells (as
occurs with biochemical tests) will blur the results [81].
Morphology, accurate quantitation and keeping the
definitions
It is important to emphasize that in many areas, molecular
insights are only possible with very careful morphological analyses, as for example in gastrointestinal lymphomas [52] and
endometrial hyperplasias [82]. Accurate quantitation of microscopic features is also essential; qualitative impressions are
often simply too indefinite [83]. Not adhering to definitions of
certain features may have enormous consequences for the prognostic value of a prognostic factor, where for example formal
mitotic activity counts versus general mitotic impressions in
Up to 50% of breast cancers have been reported to present
with loss of PTEN function [66], but others using IHC [84]
found just 8% PTEN negativity and almost 70% showed weak
positivity. This raises the important question: which determination technique best predicts the clinical outcome? A typical
example of a poorly defined test method is the use of IHC
without strict quality and assurance control or standardization;
the international NEQAS network for IHC quality testing has
shown that a high percentage of IHC determinations in different laboratories were suboptimal [85]. In the future, an
international quality standard for IHC cancer studies should be
introduced.
Minimize and standardize the interval between excision
of a tumor and freezing
This is essential for RNA and proteins, but less urgent for
DNA (which is much more resistant to degradation) (Figure 9).
Critical remarks and need for validation
We have seen that post-genomic era studies are characterized
by huge numbers [86–91]. The traditional reductionist
approach with hypothesis-driven research that focuses on one
gene at a time is now being challenged by high-technology,
hypothesis-generating ‘Omic’ approaches [92]. The discoverybased research currently applied to the Omic-technologies
calls for a change in reasoning [93]. The steps in development
(‘Will it work?’), evaluation (‘Does it work?’), and
explanation (‘How does it work?’) are separate. Many of the
current Omic-discoveries are based on large-scale detection of
genes/gene products from where we do not yet know. One
might argue that a test may be useful if patterns can reliably
ii41
discriminate disease from no disease, regardless of whether
they can be understood or explained. However, before such
profiles are to be clinically implicated the results need to be
validated, according to good laboratory practice guidelines
[44, 94].
In other words, microarray techniques have made it easier
to find ‘the needle in the haystack’. However, as these results
often come up with tens or hundreds of genes that are differentially expressed in tumors compared with normal cells, we
now find ourselves looking at a haystack of needles. The challenge for the future is to sift out the genetic information from
non-informative noise and find clinically useful data by means
of sound experimental design [95]. Testing very many variables at the same time has a serious risk, i.e. that the learning
(training) sets are very small compared with the large number
of features analyzed. This can result in far too optimistic
results that cannot be reproduced in subsequent new patient
groups (a phenomenon called ‘overtraining’ of the initial
learning set). For example, the differences in the same
tumor types can be more striking than similar, as shown in
microarray studies for lung cancer [46, 96]. We have mentioned above that serum proteomics tests [39] reported to be
nearly 100% sensitive and specific for ovarian cancer, 3 years
later are yet to be duplicated. Verification by an independent
method [92] or in independent patient groups (see Baak [44])
is mandatory. Recently, Michiels et al. [97] reanalyzed data
from the seven largest published studies that have attempted
to predict prognosis of cancer patients on the basis of DNA
microarray analysis. By using multiple random sets the stability of the molecular signature and the proportion of misclassifications were studied. They found that the list of genes
identified as predictors of prognosis was highly unstable;
molecular signatures depended strongly on the selection of
patients in the training sets. For all but one study, the
proportion misclassified decreased as the number of patients
in the training set increased. Because of inadequate validation,
the chosen seven studies published overoptimistic results. Five
of the seven studies did not classify patients better than chance
[97]. In agreement with Baggerly et al. [40] and Michiels et al.
[97] we believe that any genomic and proteomic study should
be considered prone to error before it is further validated by
methods or patient groups that are independent of the original
study.
Large scale analyses on many patients may be required.
One such analysis used a statistical method, ‘comparative
metaprofiling’, which identifies and assesses the intersection
of multiple gene expression signatures from a diverse collection of microarray datasets [98]. The results were encouraging.
From 40 published cancer microarray datasets, comprising 38
million gene expression measurements from >3700 cancer
samples, a common transcriptional profile emerged that is
universally activated in most cancer types relative to the
normal tissues from which they arose, likely reflecting
essential transcriptional features of neoplastic transformation
[98].
Consequences of the post-genomic era
The variety of consequences of the post-genomic era is not
yet clear, but certain points have emerged.
Will genomics and proteomics replace morphological
science? When gene expression arrays became available some
6 years ago, it was predicted that the end of morphological
science was near, but it now seems unlikely that this will happen soon. The studies by Mutter et al. on endometrial pre-cancers are classical examples of the development as follows. In
1985, the basis was laid for the WHO94 classification of endometrial hyperplasias, but a molecular basis was lacking. By
1995, studies could not detect a strong correlation between
WHO94 and genetic clonality. However, subsequent computerized morphometrical analysis revealed a strong correlation
between clonality and the prognostic morphometric D-Score
[82]. Further analyses comparing the prognostic value of the
WHO94 and D-Score showed that the latter was superior
[71, 99], but the morphometric technology required for the
D-Score is not widely available. This led to a subjective variant, the Endometrial Intraepithelial Neoplasia (EIN) classification [100], which in our study is well reproducible [101].
Thus, pathologists come back to the standard microscopic
image, but with renewed thinking, and are now armed with
new knowledge about reactive endometrial changes (called
benign hyperplasias) versus genetically monoclonal expansive
growing lesions with a high cancer risk (called EIN). This
new approach greatly strengthens the prognostic evaluations
and therapeutic choices. Of course, compared with
genetic testing microscopic evaluations are currently very
cost-effective and time-efficient, but this may change over
time when more efficient automated molecular tests become
available.
It is beyond the purpose of this article to discuss in detail
the expected changes in the curricula of medical students,
basic scientists, clinicians and technicians, and the consequences for the staff members of cancer-related departments.
Bioinformatics (the application of computer science and
informatics to molecular biology) becomes essential for the
study of DNA, RNA and proteins, and will help to map
sequences to databases, create models for molecular interaction, evaluate structural compatibility, find differences
between host and pathogen DNA and identify conservation
motifs in protein structure [102].
For research, the development of a cooperative framework
among basic researchers, medical specialists, bio-mathematicians, technology experts and producers is essential for realizing the revolutionary promise that genomics and proteomics
hold for drug development, regulatory science, medical
practice and public health.
The hospital structure may also change from speciality
driven (surgery, radiology, medical oncology, pathology) to
specialized multidisciplinary treatment teams encompassing
certain technologies, e.g. anti-angiogenesis or gene transfer.
The legal aspects of biobanks are becoming rapidly more
important. Publications on obscure material may soon no
ii42
longer be acceptable. The ownership of the material and permission from the patient to use the material is already legalized in a number of countries.
Concluding remarks
It follows from the above that we agree with Hall and Lowe
[103] that modern cancer research is about the interplay
between disciplines. New findings must be carefully validated.
A lesson from several successful studies is that genomic and
proteomic insights only came through very careful morphological studies, analysis of homogeneous groups and the use
of accurate, reproducible quantitative methods.
Finally, a piece of sound advice to everyone on the way forward: don’t sell your microscope!
References
1. Watson JD, Crick FH. Molecular structure of nucleic acids; a structure for deoxyribose nucleic acid. Nature 1953; 171: 737 –738.
2. Venter JC, Adams MD, Myers EW et al. The sequence of the human
genome. Science 2001; 291: 1304– 1351.
3. Lander ES, Linton LM, Birren B et al. Initial sequencing and analysis of the human genome. Nature 2001; 409: 860 –921.
4. Guttmacher AE, Collins FS. Welcome to the genomic era. N Engl J
Med 2003; 349: 996 –998.
5. Conrads TP, Zhou M, Petricoin EF 3rd et al. Cancer diagnosis using
proteomic patterns. Expert Rev Mol Diagn 2003; 3: 411–420.
6. Tyers M, Mann M. From genomics to proteomics. Nature 2003; 422:
193 –197.
7. Futreal PA, Coin L, Marshall M et al. A census of human cancer
genes. Nat Rev Cancer 2004; 4: 177–183.
8. Chen YC, Hunter DJ. Molecular epidemiology of cancer. CA Cancer
J Clin 2005; 55: 45–54.
9. Sasco AJ. Migration and cancer. Rev Med Interne 1989; 10:
341 –348.
10. Jonsson S, Thorsteinsdottir U, Gudbjartsson DF et al. Familial risk
of lung carcinoma in the Icelandic population. JAMA 2004; 292:
2977–2983.
11. Hong CC, Thompson HJ, Jiang C et al. Association between the
T27C polymorphism in the cytochrome P450 c17alpha (CYP17)
gene and risk factors for breast cancer. Breast Cancer Res Treat
2004; 88: 217– 230.
12. Knudson AG Jr. Mutation and cancer: statistical study of retinoblastoma. Proc Natl Acad Sci USA 1971; 68: 820 –823.
13. Kinzler KW, Vogelstein B. Lessons from hereditary colorectal cancer. Cell 1996; 87: 159–170.
14. Vogelstein B, Fearon ER, Hamilton SR et al. Genetic alterations
during colorectal-tumor development. N Engl J Med 1988; 319:
525 –532.
15. Fearon ER, Vogelstein B. A genetic model for colorectal tumorigenesis. Cell 1990; 61: 759–767.
16. Hanahan D, Weinberg RA. The hallmarks of cancer. Cell 2000; 100:
57 –70.
17. Hahn WC, Weinberg RA. Rules for making human tumor cells. N
Engl J Med 2002; 347: 1593–1603.
18. Mueller MM, Fusenig NE. Friends or foes–bipolar effects of the
tumour stroma in cancer. Nat Rev Cancer 2004; 4: 839 –849.
19. Bhowmick NA, Neilson EG, Moses HL. Stromal fibroblasts in cancer initiation and progression. Nature 2004; 432: 332–337.
20. Jacks T, Weinberg RA. Taking the study of cancer cell survival to a
new dimension. Cell 2002; 111: 923–925.
21. Bhowmick NA, Moses HL. Tumor-stroma interactions. Curr Opin
Genet Dev 2005; 15: 97–101.
22. Costea DE, Loro LL, Dimba EA et al. Crucial effects of
fibroblasts and keratinocyte growth factor on morphogenesis of
reconstituted human oral epithelium. J Invest Dermatol 2003; 121:
1479– 1486.
23. Baak JP, Path FR, Hermsen MA et al. Genomics and proteomics in
cancer. Eur J Cancer 2003; 39: 1199– 1215.
24. Han W, Kang D, Park IA et al. Associations between breast cancer
susceptibility gene polymorphisms and clinicopathological features.
Clin Cancer Res 2004; 10: 124–130.
25. Huang J, Wei W, Zhang J et al. Whole genome DNA copy number
changes identified by high density oligonucleotide arrays. Hum
Genomics 2004; 1: 287 –299.
26. Fire A, Xu S, Montgomery MK et al. Potent and specific genetic
interference by double-stranded RNA in Caenorhabditis elegans.
Nature 1998; 391: 806–811.
27. Hannon GJ, Rossi JJ. Unlocking the potential of the human genome
with RNA interference. Nature 2004; 431: 371–378.
28. Elbashir SM, Harborth J, Lendeckel W et al. Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells.
Nature 2001; 411: 494–498.
29. Ziauddin J, Sabatini DM. Microarrays of cells expressing defined
cDNAs. Nature 2001; 411: 107–110.
30. Erfle H, Simpson JC, Bastiaens PI, Pepperkok R. siRNA cell arrays
for high-content screening microscopy. Biotechniques 2004; 37:
454–458. 460, 462.
31. Hong JW, Studer V, Hang G et al. A nanoliter-scale nucleic acid processor with parallel architecture. Nat Biotechnol 2004; 22: 435–439.
32. Lane DP. Cancer. p53, guardian of the genome. Nature 1992; 358:
15–16.
33. Krishna RG, Wold F. Posttranslational modifications. In Angeletti
RH (ed.): Proteins–Analysis and Design. San Diego: Academic
Press 1988; 121– 206.
34. Zielinski GD, Bais AG, Helmerhorst TJ et al. HPV testing and monitoring of women after treatment of CIN 3: review of the literature
and meta-analysis. Obstet Gynecol Surv 2004; 59: 543 –553.
35. van der Velden VH, Bruggemann M, Hoogeveen PG et al. TCRB
gene rearrangements in childhood and adult precursor-B-ALL: frequency, applicability as MRD-PCR target, and stability between
diagnosis and relapse. Leukemia 2004; 18: 1971–1980.
36. Feldman AL, Espina V, Petricoin EF 3rd et al. Use of proteomic patterns to screen for gastrointestinal malignancies. Surgery 2004; 135:
243–247.
37. Petricoin E, Wulfkuhle J, Espina V, Liotta LA. Clinical proteomics:
revolutionizing disease detection and patient tailoring therapy. J Proteome Res 2004; 3: 209–217.
38. Wulfkuhle JD, Liotta LA, Petricoin EF. Proteomic applications for
the early detection of cancer. Nat Rev Cancer 2003; 3: 267–275.
39. Petricoin EF, Ardekani AM, Hitt BA et al. Use of proteomic patterns
in serum to identify ovarian cancer. Lancet 2002; 359: 572 –577.
40. Baggerly KA, Morris JS, Edmonson SR, Coombes KR. Signal in
noise: evaluating reported reproducibility of serum proteomic tests
for ovarian cancer. J Natl Cancer Inst 2005; 97: 307–309.
41. Posadas EM, Simpkins F, Liotta LA et al. Proteomic analysis for the
early detection and rational treatment of cancer—realistic hope? Ann
Oncol 2005; 16: 16–22.
42. Baak JP, Oort J. The case for quantitative pathology. In Baak JP
(ed.): Manual of Cancer Diagnosis and Prognosis. New York:
Springer 1991; 12 –31.
ii43
43. Sloane JP, Amendoeira I, Apostolikas N et al. Consistency achieved
by 23 European pathologists from 12 countries in diagnosing breast
disease and reporting prognostic features of carcinomas. European
Commission Working Group on Breast Screening Pathology. Virchows Arch 1999; 434: 3–10.
44. Baak JP. The framework of pathology: good laboratory practice by
quantitative and molecular methods. J Pathol 2002; 198: 277– 283.
45. Chaurand P, Sanders ME, Jensen RA, Caprioli RM. Proteomics in
diagnostic pathology: profiling and imaging proteins directly in tissue sections. Am J Pathol 2004; 165: 1057–1068.
46. Bhattacharjee A, Richards WG, Staunton J et al. Classification of
human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses. Proc Natl Acad Sci USA 2001; 98:
13790–13795.
47. Sorlie T, Perou CM, Tibshirani R et al. Gene expression patterns of
breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 2001; 98: 10869–10874.
48. van de Vijver MJ, He YD, van’t Veer LJ et al. A gene-expression
signature as a predictor of survival in breast cancer. N Engl J Med
2002; 347: 1999–2009.
49. Zhou W, Goodman SN, Galizia G et al. Counting alleles to predict
recurrence of early-stage colorectal cancers. Lancet 2002; 359:
219 –225.
50. Lossos IS, Czerwinski DK, Alizadeh AA et al. Prediction of survival
in diffuse large-B-cell lymphoma based on the expression of six
genes. N Engl J Med 2004; 350: 1828–1837.
51. Centeno BA, Enkemann SA, Coppola D et al. Classification of
human tumors using gene expression profiles obtained after microarray analysis of fine-needle aspiration biopsy samples. Cancer Cytopathology 2005; In press.
52. Isaacson PG, Du MQ. Gastrointestinal lymphoma: where morphology meets molecular biology. J Pathol 2005; 205: 255–274.
53. Schraders M, de Jong D, Kluin P et al. Lack of Bcl-2 expression in
follicular lymphoma may be caused by mutations in the BCL2 gene
or by absence of the t(14;18) translocation. J Pathol 2005; 205:
329 –335.
54. van Dongen JJ, Langerak AW, Bruggemann M et al. Design and
standardization of PCR primers and protocols for detection of clonal
immunoglobulin and T-cell receptor gene recombinations in suspect
lymphoproliferations: report of the BIOMED-2 Concerted Action
BMH4-CT98-3936. Leukemia 2003; 17: 2257–2317.
55. Bloom G, Yang IV, Boulware D et al. Multi-platform, multi-site,
microarray-based human tumor classification. Am J Pathol 2004;
164: 9– 16.
56. Esteller M, Corn PG, Baylin SB, Herman JG. A gene hypermethylation profile of human cancer. Cancer Res 2001; 61: 3225–3229.
57. Jeronimo C, Usadel H, Henrique R et al. Quantitation of GSTP1
methylation in non-neoplastic prostatic tissue and organ-confined
prostate adenocarcinoma. J Natl Cancer Inst 2001; 93: 1747–1752.
58. Janssen EA, Baak JP, Guervos MA et al. In lymph node-negative
invasive breast carcinomas, specific chromosomal aberrations are
strongly associated with high mitotic activity and predict outcome
more accurately than grade, tumour diameter, and oestrogen receptor. J Pathol 2003; 201: 555 –561.
59. Shayesteh L, Lu Y, Kuo WL et al. PIK3CA is implicated as an
oncogene in ovarian cancer. Nat Genet 1999; 21: 99 –102.
60. Campbell IG, Russell SE, Choong DY et al. Mutation of the
PIK3CA gene in ovarian and breast cancer. Cancer Res 2004; 64:
7678–7681.
61. Lee JW, Soung YH, Kim SY et al. PIK3CA gene is frequently
mutated in breast carcinomas and hepatocellular carcinomas. Oncogene 2005; 24: 1477– 1480.
62. Silvestrini R, Benini E, Veneroni S et al. p53 and bcl-2 expression
correlates with clinical outcome in a series of node-positive breast
cancer patients. J Clin Oncol 1996; 14: 1604–1610.
63. van Diest PJ, van der Wall E, Baak JP. Prognostic value of proliferation in invasive breast cancer: a review. J Clin Pathol 2004; 57:
675–681.
64. Volpi A, Nanni O, De Paola F et al. HER-2 expression and cell proliferation: prognostic markers in patients with node-negative breast
cancer. J Clin Oncol 2003; 21: 2708– 2712.
65. Yu Y, Xu F, Peng H et al. NOEY2 (ARHI), an imprinted putative
tumor suppressor gene in ovarian and breast carcinomas. Proc Natl
Acad Sci USA 1999; 96: 214– 219.
66. Nagata Y, Lan KH, Zhou X et al. PTEN activation contributes to
tumor inhibition by trastuzumab, and loss of PTEN predicts trastuzumab resistance in patients. Cancer Cell 2004; 6: 117 –127.
67. Pandolfi PP. Breast cancer—loss of PTEN predicts resistance to
treatment. N Engl J Med 2004; 351: 2337–2338.
68. Mutter GL, Lin MC, Fitzgerald JT et al. Altered PTEN expression
as a diagnostic marker for the earliest endometrial precancers. J Natl
Cancer Inst 2000; 92: 924–930.
69. Bruni P, Boccia A, Baldassarre G et al. PTEN expression is reduced
in a subset of sporadic thyroid carcinomas: evidence that PTENgrowth suppressing activity in thyroid cancer cells mediated by
p27kip1. Oncogene 2000; 19: 3146–3155.
70. Goel A, Arnold CN, Niedzwiecki D et al. Frequent inactivation of
PTEN by promoter hypermethylation in microsatellite instabilityhigh sporadic colorectal cancers. Cancer Res 2004; 64: 3014–3021.
71. Baak JP, Orbo A, van Diest PJ et al. Prospective multicenter evaluation of the morphometric D-score for prediction of the outcome of
endometrial hyperplasias. Am J Surg Pathol 2001; 25: 930– 935.
72. Verweij J, Casali PG, Zalcberg J et al. Progression-free survival in
gastrointestinal stromal tumours with high-dose imatinib: randomised trial. Lancet 2004; 364: 1127–1134.
73. Eisenhauer EA. From the molecule to the clinic—inhibiting HER2
to treat breast cancer. N Engl J Med 2001; 344: 841 –842.
74. Slamon DJ, Leyland-Jones B, Shak S et al. Use of chemotherapy
plus a monoclonal antibody against HER2 for metastatic breast
cancer that overexpresses HER2. N Engl J Med 2001; 344:
783–792.
75. Herbst RS, Fukuoka M, Baselga J. Gefitinib—a novel targeted
approach to treating cancer. Nat Rev Cancer 2004; 4: 956–965.
76. She QB, Solit D, Basso A, Moasser MM. Resistance to gefitinib in
PTEN-null HER-overexpressing tumor cells can be overcome
through restoration of PTEN function or pharmacologic modulation
of constitutive phosphatidylinositol 30 -kinase/Akt pathway signaling.
Clin Cancer Res 2003; 9: 4340–4346.
77. Kurata T, Tamura K, Kaneda H et al. Effect of re-treatment with
gefitinib (‘Iressa’, ZD1839) after acquisition of resistance. Ann
Oncol 2004; 15: 173–174.
78. Hochhaus A, Hughes T. Clinical resistance to imatinib: mechanisms
and implications. Hematol Oncol Clin North Am 2004; 18: 641–656, ix.
79. Debiec-Rychter M, Cools J, Dumez H et al. Mechanisms of resistance to imatinib mesylate in gastrointestinal stromal tumors and
activity of the PKC412 inhibitor against imatinib-resistant mutants.
Gastroenterology 2005; 128: 270 –279.
80. Lal P, Salazar PA, Chen B. HER-2 testing in breast cancer using
immunohistochemical analysis and fluorescence in situ hybridization: a single-institution experience of 2,279 cases and comparison
of dual-color and single-color scoring. Am J Clin Pathol 2004; 121:
631–636.
81. Kruse AJ, Skaland I, Janssen EA et al. Quantitative molecular parameters to identify low-risk and high-risk early CIN lesions: role of
ii44
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
markers of proliferative activity and differentiation and Rb availability. Int J Gynecol Pathol 2004; 23: 100 –109.
Mutter GL, Baak JP, Crum CP et al. Endometrial precancer diagnosis by histopathology, clonal analysis, and computerized morphometry. J Pathol 2000; 190: 462 –469.
Baak JP, Janssen E. DNA ploidy analysis in histopathology. Morphometry and DNA cytometry reproducibility conditions and clinical
applications. Histopathology 2004; 44: 603–614.
Panigrahi AR, Pinder SE, Chan SY et al. The role of PTEN and its
signalling pathways, including AKT, in breast cancer; an assessment
of relationships with other prognostic factors and with outcome.
J Pathol 2004; 204: 93–100.
Rhodes A, Jasani B, Balaton AJ, Miller KD. Immunohistochemical
demonstration of oestrogen and progesterone receptors: correlation of
standards achieved on in house tumours with that achieved on
external quality assessment material in over 150 laboratories from
26 countries. J Clin Pathol 2000; 53: 292–301.
Roberts AB, Wakefield LM. The two faces of transforming growth
factor beta in carcinogenesis. Proc Natl Acad Sci, USA 2003; 100:
8621–8623.
Piek E, Roberts AB. Suppressor and oncogenic roles of transforming
growth factor-beta and its signaling pathways in tumorigenesis. Adv
Cancer Res 2001; 83: 1 –54.
Sharpless NE, DePinho RA. p53: good cop/bad cop. Cell 2002; 110:
9 –12.
Attardi LD. The role of p53-mediated apoptosis as a crucial antitumor response to genomic instability: lessons from mouse models.
Mutat Res 2005; 569: 145–157.
Oren M. Lonely no more: p53 finds its kin in a tumor suppressor
haven. Cell 1997; 90: 829 –832.
Levine AJ, Finlay CA, Hinds PW. P53 is a tumor suppressor gene.
Cell 2004; 116: S67–S69. 1 p following S69.
92. Liang P, Pardee AB. Analysing differential gene expression in cancer. Nat Rev Cancer 2003; 3: 869–876.
93. Ransohoff DF. Evaluating discovery-based research: when biologic
reasoning cannot work. Gastroenterology 2004; 127: 1028.
94. Hall PA, Going JJ. Predicting the future: a critical appraisal of cancer prognosis studies. Histopathology 1999; 35: 489 –494.
95. Yeatman TJ. The future of cancer management: translating the genome, transcriptome, and proteome. Ann Surg Oncol 2003; 10: 7–14.
96. Garber ME, Troyanskaya OG, Schluens K et al. Diversity of gene
expression in adenocarcinoma of the lung. Proc Natl Acad Sci USA
2001; 98: 13784–13789.
97. Michiels S, Koscielny S, Hill C. Prediction of cancer outcome with
microarrays: a multiple random validation strategy. Lancet 2005;
365: 488–492.
98. Rhodes DR, Yu J, Shanker K et al. Large-scale meta-analysis of cancer microarray data identifies common transcriptional profiles of
neoplastic transformation and progression. Proc Natl Acad Sci USA
2004; 101: 9309–9314.
99. Baak JP, Mutter GL. EIN and WHO94. J Clin Pathol 2005; 58: 1 –6.
100. Mutter GL. Endometrial intraepithelial neoplasia (EIN): will it bring
order to chaos? The Endometrial Collaborative Group. Gynecol
Oncol 2000; 76: 287–290.
101. Hecht JL, Ince TA, Baak JP et al. Prediction of endometrial carcinoma by subjective endometrial intraepithelial neoplasia diagnosis.
Mod Pathol 2005; 18: 324–330.
102. Debes JD, Urrutia R. Bioinformatics tools to understand human
diseases. Surgery 2004; 135: 579 –585.
103. Hall PA, Lowe SW. Molecular and cellular themes in cancer
research: II. J Pathol 2005; 205: 121 –122.
104. Baak JP, van Diermen B, Steinbach A et al. Lack of PTEN
expression in endometrial intraepithelial neoplasia (EIN) is correlated with cancer progression. Hum Pathol 2005; In press.