PERSPECTIVES

be overexpressed on a one-dimensional pro-

TIMELINE
tein gel when normal cells were compared
with those that were infected with simian
virus 40 (SV40) DNA tumour virus5. It
Analysing differential gene turned out that the increase in the expres-
sion of the p53 protein was caused by the
expression in cancer stabilization of the protein through its inter-
action with SV40 large T antigen.
The development, by Patrick O’Farrell, of
Peng Liang and Arthur B. Pardee two-dimensional (2D) protein gel elec-
trophoresis, which separates proteins by both
Analysis of messenger RNA and proteins is cancer1 that are controlled by many genes. size and charge, allowed a more complete
widely used to compare patterns of gene Compounding the complexity is the fact that visualization of cellular protein expression6.
expression between cells or tissues of many of the so-called oncogenes or tumour- Later, Robert Croy and Arthur Pardee found
different kinds and under different suppressor genes are signalling molecules that a transformed cell differs from its normal
conditions; for example, between normal themselves, each of which functions to control parental cell in expression of only a few of the
and cancer cells. The goal of the individuals the expression of a subset of downstream more than 1,000 proteins that are detected on
who are developing these methods has genes2,3. So, the analysis of differential gene 2D protein gels. One of these proteins, named
been to enable faster, simpler, more expression — known as expression genetics or p68, was shown to be expressed at a much
sensitive and systematic analyses, and over functional genomics — has become one of the higher level in carcinogen-transformed
the past few decades techniques have most widely used strategies for discovering and murine fibroblasts7. This result reinforced the
become increasingly more sophisticated. understanding the molecular circuitry under- notion that subtle changes in gene expression
This timeline article reviews the evolution of lying cancer. Throughout this review, we pro- might hold the key to the understanding of
these technologies as well as strategies for vide examples of gene discovery in cancer cancer. However, frustration often followed
identifying differentially expressed genes in research made by studying differential gene such studies, owing to the inability to recover
normal and cancer cells. It also highlights expression mainly at the mRNA level, with a sufficient amounts of the differentially
their use for the search for target genes of special emphasis on genes that are regulated by expressed protein species for further molecu-
the tumour suppressor p53. the tumour suppressor p53. lar characterization. Also, it became clear that
Over the past two decades, several meth- this method was not sensitive enough, detect-
In the past 50 years, we have made remark- ods have been developed to allow comparative ing only about 2,000 of the estimated 10,000
able progress in understanding genetics. studies of gene expression between normal or more different proteins that are expressed
Although the central genetic dogma is and cancer cells. Starting with simple in a cell. Newer techniques for the analysis of
defined as the flow of genetic information approaches that used gel electrophoresis to protein expression — collectively known as
from DNA to messenger RNA and then to compare protein expression, methods that proteomics — have been developed in recent
protein, the complete sequencing of the focus on mRNA analysis have evolved and years and are mainly powered by the use of
3-billion base-pair human genome has shed become increasingly sophisticated, as a result mass spectrometry to greatly improve sensi-
little light so far on precisely how such unidi- of the inventions of recombinant DNA, tivity and allow the characterization of small
rectional information flow from tens of DNA sequencing and PCR technologies (see quantities of protein.
thousands of genes is programmed in a cell. TIMELINE)4. To better understand the princi-
Decoding such information from the human ples behind some of the main methods, we Differential hybridization
genome is likely to be even more challenging can group them into six categories. With the advent of recombinant DNA tech-
and time-consuming than the sequencing of nology in the late 1970s, studies of compara-
the genome. Although classical genetics has Protein gel electrophoresis tive gene expression quickly shifted from
been a powerful tool for dissecting molecular Perhaps the earliest and arguably the most looking at proteins to the analysis of mRNA
diseases that are affected by the gain or loss of successful example of studying differential expression using complementary DNA. The
function of a protein encoded by a single gene expression in cancer was the discov- earliest approach was differential hybridiza-
gene, such a strategy has proved to be less ery of the p53 tumour-suppressor protein tion, in which the pair of mRNA samples to
fruitful for understanding diseases such as in the late 1970s. The protein was found to be compared (for example, from normal

NATURE REVIEWS | C ANCER VOLUME 3 | NOVEMBER 2003 | 8 6 9

PERSPECTIVES

Timeline | Techniques that are used to study differential gene expression in cancer
Zimmermann and Timberlake Peng Liang and Arthur Pardee
Studies began with overcame some of the problems developed the differential display (DD) The complementary DNA
differential hybridization with DNA-probe complexity with microarray was pioneered GeneChip oligo
technique15 — an ‘open system’ that
using recombinant their subtractive hybridization by Patrick Brown and microarrays were
requires no previous knowledge of
DNA technology. technique. David Botstein27. developed by Affymetrix 28.
messenger RNA or gene sequences.

1980 1981 1990 1992 1995 1996 2000

2D protein gel electrophoresis Expressed sequence tag (EST) John Welsh and Michael Serial analysis of gene expression was The beads-based EST sequencing
was developed by Patrick sequencing was developed by Craig McClelland invented a method developed by Kenneth Kinzler and Bert approach was developed by
O’Farrell6. Venter 50. This technique has had a that is related to DD called RNA Volgstein’s group51. Sydney Brenner 54.
significant role in gene discovery. arbitrarily primed PCR16.

versus cancer cells) were radioactively expressed genes (FIG. 2). The resulting cDNA
labelled as cDNA probes with 32P by reverse probes with reduced complexity were then
transcription with oligo-dT primers that individually checked by northern blot for
anneal to the polyadenylic chains (polyA confirmation. The discovery of T-cell
tails) present at the 3′ termini of all eukary- receptors in the mid 1980s by Mark Davis Normal cells Tumour cells
otic mRNAs. The resulting two cDNA and colleagues — when they compared the
probes were then differentially hybridized differences in mRNA expression between
to duplicate filters, which had on them tens T and B cells using such strategies — is one mRNA
of thousands of plaques from a phage of the best examples of gene discovery
cDNA library 8. Comparison of the through the analysis of differential gene
hybridization pattern to cDNA-containing expression12. Dr. Davis recalled in his accep- Reverse transcription
phage plaques between two mRNA probes tance speech for his General Motor Cancer and labelling with
allowed the identification of genes that Research Award that he would not have fluorescent dyes or
radioactivity
were uniquely expressed in one but not the achieved the feat had he relied on differen-
other RNA sample (FIG. 1). Although this tial hybridization of dot blots with complex cDNA
strategy has implicated several differentially cDNA probes13.
expressed genes that are involved in the This exciting discovery fuelled a flood of
hormone responsiveness of human breast biomedical research using gene-expression Radioactive label
cancer cells 9 and that are overexpressed analysis as a basic strategy to understand a
during infection by human T-cell wide variety of biological systems. The dis- ss cDNA molecules
are used to hybridize
leukaemia/lymphoma virus10, it was soon covery of the cyclin-dependent kinase cDNA phage plaques
realized that such a ‘reverse northern’ inhibitor WAF1 (also known as p21) as a tar- lifted onto duplicate
fillers
approach of using complex cDNA probes get gene of p53 by Bert Vogelstein’s group
(BOX 1) would not be able to detect most provided another great example of the power
genes, which are expressed at a low level8. As of this conceptually simple and logical
a result, differential screening quickly gave method14. Since then, several PCR-based sub-
way to hybridization methods that use tractive-hybridization strategies have been
cDNA probes with reduced complexity after developed, including representational differ-
a ‘subtraction’ process. ence analysis (RDA) and suppression PCR, Differentially
which allow a smaller amount of mRNA expressed gene

Subtractive hybridization samples to be analysed. Figure 1 | Differential hybridization.

Realizing the problems with cDNA-probe All messenger RNAs that are expressed in the
complexity for differential hybridization, in Differential display cells being compared are first labelled either
the early 1980s C. Zimmermann and col- To speed up the identification of differen- radioactively or fluorescently by copying them
into single-stranded complementary DNAs
leagues devised an ingenious approach tially expressed genes, we developed differ-
(ss cDNAs) using reverse transcriptase. The
known as subtractive hybridization to ential display (DD) in the early 1990s, with resulting complex cDNA probes (BOX 1) are then
enrich for cDNA probes that represent the aim of overcoming the limitations of used to hybridize to the cDNA phage plaques
mRNAs that are uniquely expressed in one previous methods 15. John Welsh and that are lifted onto duplicate nylon membranes.
cell but not the other 11. This method Michael McClelland invented a related The key difference between differential screening
removed most of the cDNAs that repre- method known as RNA arbitrarily primed and microarrays (FIG. 4) is that phage cDNA
plaques contain some degree of redundancy in
sented the genes that are commonly PCR (RAP-PCR) using only primers of
sequences (for example, a highly expressed gene
expressed in both cells being compared, random sequence16. A sensitive method was might be represented by more than one plaque
and left behind only single-stranded required so that it could be applied to on a filter), whereas microarrays usually strive to
cDNAs that represented a few differentially systems for which scarce biological samples minimize such redundancy.

870 | NOVEMBER 2003 | VOLUME 3 www.nature.com/reviews/cancer

 PERSPECTIVES

are available, and by which all mRNAs, Box 1 | cDNA probe complexity
whether scarce or abundant, can be repre-
sented. Also, the method needed to be sys- This term is used to specify the number of complementary DNA species (or messenger RNA
tematic, so that a complete search of all the species) and their relative concentrations within a cDNA probe. For differential screening and
expressed genes in a cell was possible. Based micoarrays, a cDNA probe is made by reverse transcription of all the mRNAs that are expressed
on these crucial criteria, differential display in a cell or within a tissue specimen using an oligo-dT primer, which targets the polyA tails that
was developed by integrating two of the are present in most eukaryotic mRNAs. In fact, such a cDNA probe is so complex that it consists
most simple, powerful and commonly used of as many as 10,000 different species, each ranging from a few to thousands of copies per cell —
this approach is used in both differential hybridization (FIG. 1) and microarrays (FIG. 4). For
molecular biological methods; namely, PCR
subtractive hybridization, the complexity of a subtracted cDNA probe is greatly reduced by
and DNA sequencing by gel electrophore-
removing most of the commonly expressed mRNAs for a pair of RNA samples being compared
sis15. In essence, DD works by systematically
(FIG. 2), whereas differential display displays 50–100 mRNAs at a time using different primer
amplifying the 3′ termini of eukaryotic pairs (FIG. 3) and serial analysis of gene expression (SAGE) counts 25–75 mRNAs during each
mRNA by reverse transcription-PCR using sequencing run of SAGE tags (FIG. 5).
one of the three anchored oligo-dT primers Clearly, when so complex a cDNA probe is used, one of the main challenges during microarray
(that is, the run of Ts ending with a C, G analysis is to be certain that a hybridization signal is specific and quantitative to a known gene
or A) in combination with a set of short sequence that is laid on a ‘chip’. One simple control experiment could clearly illustrate the potential
primers of arbitrary sequences (FIG. 3). Based problems in hybridization with a complex cDNA probe; instead of labelling all the mRNAs by
on the finding that each arbitrary primer reverse transcription with oligo-dT primers, only one mRNA at a time could be labelled for several
would recognize its corresponding mRNA genes that are highly or rarely expressed that are represented on an array with a corresponding
targets with a minimum of seven matching gene-specific primer (for example, a primer that anneals just upstream of the polyA of a gene of
bases, mathematical models have been pro- interest). These single gene-specific probes, when hybridized to the microarrays individually, will
posed to predict the relation between the provide an accurate glimpse of the actual sensitivity and specificity of microarrays, in comparison
number of arbitrary primers and the cover- with complex cDNA probes labelled with an oligo-dT primer. If an experiment fails to detect only
age of expressed genes in any given eukary- one gene specifically at a time on an array that contains tens of thousands of other genes when a
otic cell 17. One of the main advantages of gene-specific cDNA probe is used, how can we be certain that all the ‘colourful’ signals seen on an
DD is its technical simplicity and accessibil- array or ‘chip’ truly reflect an accurate and sensitive snapshot of global gene expression within a cell?
ity, which makes it and several related
methods the most widely used approaches
for studying differential gene expression18.
Unlike microarrays, DD does not require
any previous knowledge of mRNA or gene
Normal cells Tumour cells
sequences, making it an ‘open’ system that
is applicable to any eukaryotic organism.
Many oncogene targets have been identified mRNA
by DD, including genes that are regulated by
RAS19,20, v-REL21 and ERBB22. One of the RAS
target genes was shown to be a new cytokine,
now known as interleukin-24 (IL-24) (REF. 23). Reverse transcription and labelling with
Many important target genes of the p53 fluorescent dyes or radioactivity
tumour suppressor have also been discov- mRNA cRNA
ered by DD, which will be discussed in Fluorescent dye
more detail below.
Matching mRNA and cDNA
Initially, DD suffered from a high rate of molecules are allowed to hybridize
‘false positives’ (also known as ‘noise’ in ref-
erence to microarrays), but technical
improvements as well as care in experimen-
tal design 24 have greatly reduced the
number of false positives to allow truly dif-
Double-stranded cDNA/RNA
ferentially expressed genes to be steadily hybrids are isolated and
identified18. Despite the great impact of the discarded
method on biomedical research, most work Subtracted cDNAs are used as probes
involving DD has used limited PCR reac- to isolate tumour-associated gene(s)

tions to take a ‘shot-gun’ approach that

identifies only a few genes at a time, giving
DD a low-tech, low-precision and low-
throughput image. It is only recently that Figure 2 | Subtractive hybridization. All messenger RNAs that are expressed in a cell are first converted
efforts have been made for the automation to single-stranded complementary DNAs (ss cDNAs), which are then hybridized to an excess of all the
mRNAs that are expressed in the other cell type being compared. Genes that are commonly expressed
of DD technology with fluorescent digital
in both cells will form cDNA/mRNA double-stranded duplexes, whereas cDNA that is uniquely expressed
data acquisition and analysis to increase its in the first cell will be in single-stranded form, which can be separated from most double-stranded
throughput and accuracy for systematic cDNA/mRNA species. The resulting ‘subtracted cDNAs’ are then further characterized to determine if
gene-expression analysis25,26. they indeed represent genes that are uniquely expressed in the first cell.

NATURE REVIEWS | C ANCER VOLUME 3 | NOVEMBER 2003 | 8 7 1

PERSPECTIVES

Microarrays cDNA plaques are replaced with spotted into more easily interpretable biological path-
More than a decade after differential cDNAs or oligos, and radioactive labels are ways for the understanding and classification
hybridization was introduced came two replaced with fluorescent ones. The promises of cancer. Indeed, DNA microarrays have
‘hot’ technologies — cDNA microarrays, of these methods29,30 are based on their poten- been used to profile gene-expression patterns
pioneered by Patrick Brown and David tial in being able to simultaneously analyse of almost all of the main cancers — including
Botstein27, and GeneChip arrays (oligo the expression of mRNAs from tens of thou- leukaemia31,32, lymphoma33, adenocarcinoma
arrays), developed at Affymetrix28 (FIG. 4). sands of genes, which can then be further of the lung 34,35, breast 36 and prostate 37 — and
These techniques are, in essence, based on the analysed using computers, in the hope that promise to change the way cancer is diagnosed,
differential-hybridization strategy, in which gene-expression patterns can be transformed classified and treated in the clinic. However, the
realization of these potentials will be a consid-
erable challenge, as the differences between
studies of the same tumour types can often be
more striking than their similarities38–41. The
best such examples are two large microarray
Normal cells Tumour cells studies on lung cancer 34,35.
One of the greatest advantages of
microarrays over other methods, including
mRNA population Same process
as for normal DD, is that each spot on a microarray con-
CAAAAAAAAAAA-An cells tains a known sequence. So, once a signal is
GAAAAAAAAAAA-An detected, the nature of the gene is known.
UAAAAAAAAAAA-An However, the hidden catch of such a benefit
is that it also makes array-based methods
Reverse transcription
5′-AAGCTTTTTTTTTTTG-3′ ‘closed’ systems that are only able to cover
known gene sequences. This might be best
exemplified by the fact that more than 80%
CAAAAAAAAAAA-An of the more well-documented p53 tumour-
GTTTTTTTTTTTCGAA suppressor gene targets were identified by
PCR amplification other, open-system-based methods such as
DD and serial analysis of gene expresion
5′-AAGCTTGATTGCC-3′
(SAGE), which can identify both known and
5′ AAGCTTTTTTTTTTTG-3′
novel genes (TABLE 1).
The inherent complexity of the cDNA
probes that are used in differential-
AAGCTTGATTGCC
hybridization strategies remains the root
GTTTTTTTTTTTCGAA
cause of the lack of signal sensitivity and
AAGCTTGATTGCC specificity for most low-abundance
GTTTTTTTTTTTCGAA mRNAs8,42,43. Without a doubt, all human
genes can eventually be condensed on a
single array or ‘chip’, but uncertainty
remains as to whether each of these tens of
AAGCTTGATTGCC thousands of cDNA probes will hybridize
GTTTTTTTTTTTCGAA to only their corresponding target template
and to nothing else on the chip. Not sur-
Denaturing polyacrylamide gel prisingly, more and more researchers are
and image scanning
cautious about the accuracy of microarray
RNA sample: Normal Tumour data, but most studies place the blame only
Negative electrode on inadequate bioinformatical and statisti-
cal tools for ‘data mining’ (the analysis of
noisy data)42–49, rather than on the funda-
Differentially expressed cDNA mental problem of the complexity of
cDNA probes (BOX 1). Therefore, as with
Positive electrode any other method for the analysis of differ-
Figure 3 | Differential display. In this approach the 3′ termini of eukaryotic messenger RNA are ential gene expression, data from micro-
systematically amplified by reverse transcription-PCR using one of the three anchored oligo-dT primers array experiments should be considered
(in this example, 5′-AAGCTTTTTTTTTTTG -3′) in combination with a set of short primers of arbitrary with caution, unless each data point can be
sequences (in this example, 5′-AAGCTTGATTGCC-3′). The length of the arbitrary primers is designed verified by an independent method such as
such that each will recognize about 50–100 mRNAs under a given PCR condition. As a result, mRNA northern-blot analysis.
3′ tails, defined by any given pair of anchored primer and arbitrary primer, are amplified. By changing
primer sequences, different subsets of mRNA can be analysed and displayed by denaturing
polyacrylamide gel electrophoresis. Side-by-side comparisons of such complementary DNA patterns Expressed sequence tags and SAGE. In the
between or among relevant RNA samples indicate differences in gene expression. Differentially expressed early 1990s, when the human genome project
cDNA bands can be retrieved and sequenced for further molecular characterization. was taking shape, Craig Venter realized that

872 | NOVEMBER 2003 | VOLUME 3 www.nature.com/reviews/cancer

 PERSPECTIVES

the National Institutes of Health (NIH) A p53 profile

Cancer Genome Anatomy Program, which Cancer research has taken a great leap for-
provides a convenient source of cDNA ward in terms of the number of cancer-
Normal cells
clones for functional studies of genes that related genes that have been identified by the
Tumour cells have been identified by methods for com- different methods for analysing differential
parative analysis of gene expression. Because gene expression. To provide a glimpse of the
of the high cost and labour-intensive nature progress that has been made over the past
mRNA of comprehensive EST sequencing, the decade in the understanding of cancer
method itself is rarely used directly to through the study of differential gene
Reverse transcription identify differentially expressed genes. expression, we summarize the identification
and labelling with
flourescent dyes
In the mid 1990s, SAGE was developed of key p53 target genes that have been func-
by Kenneth Kinzler and Bert Vogelstein’s tionally characterized, as well as the methods
ss cDNA
group51. Unlike the original EST sequencing that were used for their identification.
strategy, in which cDNA clones were ran- As mentioned earlier, p53 was discovered
domly picked from cDNA libraries, SAGE more than 20 years ago using one-dimen-
Combine in equal technology measures the level of gene sional protein gel electrophoresis. However,
amounts expression based on the frequency of it was not until a decade later that Bert
Hybridize probe to occurrence of the 3′ signature SAGE tags of Vogestein’s group showed that p53 is a
microarray containing
thousands of known 10–14 bases in length that might be unique nuclear-DNA-binding protein and that it
gene sequences to each transcript (FIG. 5). Because of the functions as a transcription factor55. Since
minimal sequence information that is nec- then, the search for p53 target genes has
Scan
essary to define an expressed gene or intensified. The p21 cyclin-dependent kinase
mRNA, a dozen or more SAGE tags from 2 inhibitor was the first prototype p53 target
different genes can be obtained and to be identified using the subtractive-
sequenced at a time, which greatly speeds hybridization technique14. Proof that p21
up the EST counting process. Like DD, alone was not sufficient to mediate the
SAGE is an open-system-based gene- tumour-suppressor function of p53 came
discovery tool. However, because of the when mice with a p21 knockout failed to
limited sequence information that is con- show a tumour-prone phenotype56. This rev-
Figure 4 | DNA micrarrays. All messenger
RNAs that are expressed in the cells being
tained in 10–14-base SAGE tags, adequate elation fuelled the search for additional p53
compared are first labelled fluorescently by gene assignment using SAGE methods still target genes, especially those that might
copying them into complementary DNAs using a requires an extensive bioinformatic sup- mediate the apoptotic function of p53.
reverse transcriptase. The resulting complex port, such as an extensive collection of Increasing numbers of such candidate p53
cDNA probes (BOX 1) are then used to hybridize cDNA sequences for the biological system target genes are being identified57 (TABLE 1),
to the cDNA templates or gene-specific oligos under investigation. The NIH Cancer but, so far, no single gene has been shown to
either directly synthesized or spotted on a glass
surface to determine the expression of
Genome Project (CGAP) now maintains a be the bona fide p53 target based on either
thousands of genes simultaneously. Red and comprehensive SAGE database for various human or mouse genetic evidence 3. An
green spots represent cDNAs that are only normal and cancer cell lines and tissues, emerging thought has been that p53 might
expressed in normal or tumours cells, with a web interface known as SAGE control a network of tissue-specific genes,
respectively. Yellow fluorescence indicates Genie52,53, which allows a quick glimpse of rather than a single target, which, together,
cDNAs that are expressed in both samples. the expression pattern for a gene of interest. carry out the full p53 tumour-suppressor
Beads-based EST sequencing is a recent functions3. Most of the newer potential p53
method that was developed by Sydney target genes listed in TABLE 1 have been
sequencing expressed genes (cDNA), Brenner and marketed by Lynx Therapeutics. partially characterized and their expression
instead of the whole genome (in which up The method, also known as massively confirmed, at least, by the ‘gold’ standard —
to 98% of DNA sequences are non-coding parallel signature sequencing (MPSS) com- northern-blot analysis. Functionally, these
‘junk’ sequences), would be a more sensible bines non-gel-based signature sequencing p53 target genes fall mainly into four cate-
approach. His strategy, by a single run of with in vitro cloning of millions of templates gories: cell-cycle or growth regulators (such
sequencing of the 3′ ends of randomly on separate 5-micron-diameter microbeads54. as p21); proapoptotic factors (such as BAX);
picked cDNA clones from a cDNA library, MPSS captures and identifies transcript DNA-damage repair proteins (such as
generated a comprehensive collection of sequences and analyses the level of expression GADD45 and TP53TG1); and negative regu-
such expressed sequence tags (ESTs)50. The of expressed genes in a sample by counting lators of p53 (such as MDM2 and PIRH2).
EST sequencing not only resulted in the the number of individual mRNA molecules About half of these better understood p53
discovery of many novel genes but also pro- that represent each gene. Individual mRNAs target genes were identified by DD, and the
vided information on the relative abun- are identified through the generation of a other half by SAGE and subtractive
dance in expression of each gene based on 17–20-base signature sequence, immediately hybridization (TABLE 1). Using a human gene
the number of times a corresponding adjacent to the 3′ end of the 3′-most Sau3A chip from Affymetrix that contained 6,000
cDNA sequence was represented in a cDNA restriction site in cDNA sequences. (For a genes, Arnold Levine’s group identified 107
library from either normal or tumour cells. more detailed description of this methodol- upregulated and 54 downregulated p53 tar-
The EST sequencing strategy has had a key ogy, see Beads-based EST sequencing in get genes58. This result extrapolates to at least
role in gene discovery and cataloging for online links box) 500 upregulated and 260 downregulated p53

NATURE REVIEWS | C ANCER VOLUME 3 | NOVEMBER 2003 | 8 7 3

PERSPECTIVES

Table 1 | Key potential p53 target genes identified by different technologies and further characterized
Gene Definition/function Method* References
MDM2 p53 negative regulator Candidate 64
WAF1 CDK2 inhibitor SH, SAGE 14
14-3-3 σ Growth inhibition SAGE 57
GADD45 DNA repair SH 57
BAX Apoptosis Candidate 65
Cyclin G Cell-cycle regulator DD 66
IGFBP3 IGF binding protein, growth inhibition SH 67
PIG3 NADPH-quinone oxidoreductase SAGE 68
KILLER/DR5 Apoptosis SH, SAGE 69
EI24/PIG8 Novel gene, apoptosis DD, SAGE 68,70
PAG608 Novel zinc-finger protein, apoptosis DD 71
DDA3 Novel gene, growth inhibition DD 72
TP53TG1 Novel gene, DNA damage DD 73
TP53TG3 Novel gene, cell-cycle checkpoint DD 74
p53R2 Ribonucleotide reductase DD 75
PERP Novel gene, pro-apoptotic SH 76
PIR121 Novel gene, RNA binding Array 77
NOXA Novel gene, pro-apoptotic BH3 protein DD 78
PIDD Novel gene, death-domain protein DD 79
p53AIP1 Novel gene, apoptosis, p53 phosphorylation DD 80
p53DINP1 Novel gene, apoptosis, p53 phosphorylation DD 81
PUMA Novel gene, pro-apoptotic BH3-protein SAGE, Array 82,83
PIRH2 Ubiquitin ligase, p53 negative regulator DD 84
PAC1 Protein phosphatase, pro-apoptotic Array 85
FAS/APO1 Cell-death receptor Candidate 86
APAF1 Apoptosis Array 87
PTEN Tumour suppressor Candidate 88
BID Apoptosis Array 89
*Differential display (DD), DNA-microarray (Array), serial analysis of gene expression (SAGE), subtractive hybridization (SH) and candidate gene approach
(Candidate), which evaluates an individual gene of an investigator’s choosing. APAF1, apoptotic protease activating factor; BAX, BCL2-associated X protein; BID,
BH3 interacting domain death agonist; CDK2, cyclin-dependent kinase 2; DDA3, differential display and activated by p53; GADD45, growth arrest and DNA-
damage-inducible; IGF, insulin-like growth factor; IGFBP3, insulin-like growth factor binding protein 3; PAG608, p53-activated gene 608; PTEN, phosphatase and
tensin homologue.

target genes, given a minimum estimate of Future directions No matter which gene-discovery meth-
30,000 genes in the human genome. We now have an arsenal of gene-expression ods are used for hunting down the mecha-
Another recent cDNA microarray screening analysis technologies at our disposal to study nisms of cancer, ultimately it will be the
for p53 target genes that used more than differential gene expression. New methods, functional characterization of each gene —
33,000 unique human cDNAs or ESTs such as beads-based EST sequencing, are con- identified by genetic, cell-biological and
implicated more than 1,500 potential p53 tinually being invented and tested4. Among biochemical methods — that will shed light
targets 59. It is interesting to note that these, however, two fundamentally different on the its true relevance. However, these
although many candidate p53 target genes approaches and schools of thought are diverg- processes can be painstakingly long and
have been isolated by microarrays, few of ing. The traditional reductionist approach hard. Remember that the p53 tumour-
these genes have been confirmed by north- with hypothesis-driven research that focuses suppressor gene was discovered more than
ern-blot analysis or further characterized on one gene at a time is now being challenged two decades ago and has been worked on
functionally, which could take some time to by high-technology, hypothesis-generating since by tens of thousands of laboratories
sort out. Without further confirmation and genomic approaches. Two Nobel laureates throughout the world. For nearly 10 years
functional characterization of these genes, have been so concerned about this genomic after its discovery, p53 was believed to be a
it is not yet clear whether the large number approach — which requires huge resources culprit oncogene. Only later was it vindi-
of p53 target genes that have been identi- and generates enormous amounts of data that cated as a tumour-suppressor gene and the
fied by arrays are in fact ‘noise’ (false are hard to analyse60 — that one, Sydney ‘guardian of the genome’. Although much
positives) of the method, or whether the Brenner, dubbed it ‘Sillycon valley fever’61, and progress has been made in understanding
screenings by other methods were not the other, Walter Gilbert, called the ‘omics’ era the function of p53 as a transcription factor
comprehensive enough. the death of molecular biology 62. that controls the cell-cycle progression and

874 | NOVEMBER 2003 | VOLUME 3 www.nature.com/reviews/cancer

 PERSPECTIVES

apoptosis in response to DNA damage, the

exact molecular nature of how p53 acts as
this crucial tumour suppressor still remains
elusive. The recently developed RNA inter-
ference techniques should help to speed up Normal cells Tumour cells
gene characterizations.
Any gene-expression profiling studies,
Same process
whether they are carried out by DD, SAGE, mRNA AAAAA as for normal
AAAAA
DNA microarrays or proteomics, should be AAAAA
cells
considered prone to error without further AAAAA
Reverse transcription
confirmation of the expression of each gene AAAAA
AAAAA
by methods that are independent of the AAAAA
original study — in other words, guilt AAAAA
should not be assumed without a true asso- ds cDNAs
AAAAA
ciation being corroborated. As a judiciary TTTTT
analogy, guilt by association, racial profiling Cut with an AAAAA
anchoring restriction
and the assumption of guilt until innocence enzyme
TTTTT
is proven are all flawed legal practices. AAAAA
TTTTT
Similarly, we believe that identifying the cul-
AAAAA
prit genes for cancer through the analysis of TTTTT
differential gene expression should be held AAAAA
TTTTT
to the same standard. Like a crime suspect,
each gene should be treated as innocent AAAAA
TTTTT
until proven guilty by further biological and AAAAA
functional studies. The profiling of only TTTTT
candidate gene sequences followed by ‘clus- AAAAA
TTTTT
ter analysis’, without any subsequent confir-
mation, should be considered to be as prone
to error as racial profiling. Individuals tags
In the preface to a method book on pro-
tein purification, Nobel Laureate Arthur TTTTT
Kornberg quoted an admonition of Efraim Linker
Racker — “Don’t waste clean thinking on
dirty enzymes” — to illustrate the impor-
tance of good biochemical practice in enzy-
mology 63. A similar doctrine — “Don’t
waste clear thinking on dirty data” — will Linked tags
certainly continue to help produce a better
quality of science and contribute to steady
Ligated into a cloning vector
progress in the field of gene-expression
analysis of cancer. Let’s move beyond gene
listing and get down to the biology. Tag sequencing and quantitation
Peng Liang is at the Vanderbilt–Ingram Cancer
Center, Department of Cancer Biology, School of
Medicine, Vanderbilt University, Nashville,
Tennessee 37232, USA.
Arthur B. Pardee is at the Department of Adult Patterns of gene- Normal Tumour
Oncology, Dana–Farber Cancer Institute, 44 expression analysis
Binney Street, Boston, Massachusetts 02115, USA.
Expression

Expression

10 10
8 8
6 6
Correspondence to A.B.P. or P.L. 4 4
level

level

e-mail: [email protected]; 2 2
[email protected] 0 0
A B C D E F G H A B C D E F G H
doi: 10.1038/nrc1214 Gene product Gene product

1. Knudson, A. G. Two genetic hits (more or less) to cancer.

Nature Rev. Cancer 1, 157–162 (2001). Figure 5 | SAGE. (Serial analysis of gene expresion). All messenger RNAs that are expressed in a cell or from
2. Sager, R. Expression genetics in cancer: shifting the a tissue specimen are first reverse transcribed with an oligo-dT primer and converted to double-stranded
focus from DNA to RNA. Proc. Natl Acad. Sci. USA 94, cDNAs (ds cDNAs), which are then cut with an anchoring restriction enzyme. The most 3′ fragment of each
952–955 (1997)
3. Vogelstein, B., Lane, D. & Levine, A. J. Surfing the p53
cDNA is ligated to a linker that contains a type IIS restriction site that allows a tagging restriction enzyme to
network. Nature 408, 307–310 (2000). cut the tagged cDNA 10–14 bases downstream. The resulting transcript-specific SAGE tags that are
4. Lorkowski, S. & Cullen, P. (eds.) Analyzing Gene released are concatemerized (25–75 tags) and ligated into a cloning vector for sequencing. After sequencing
Expression. A Handbook of Methods: Possibilities many such concatemerized SAGE tags the level of gene expression can be estimated by computer analysis
and Pitfalls (Wiley–VCH, Weinheim,
2002). that is based on the occurrence of each SAGE tag having a unique sequence.

NATURE REVIEWS | C ANCER VOLUME 3 | NOVEMBER 2003 | 8 7 5

PERSPECTIVES

5. Linzer, D. I. & Levine, A. J. Characterization of a 54K 34. Garber, M. E. et al. Diversity of gene expression in 66. Okamoto, K. & Beach, D. Cyclin G is a transcriptional
dalton cellular SV40 tumor antigen present in SV40- adenocarcinoma of the lung. Proc. Natl Acad. Sci. USA target of the p53 tumor suppressor protein. EMBO J.
transformed cells and uninfected embryonal carcinoma 98, 13784–13789 (2001). 13,4816–4822 (1994).
cells. Cell 17, 43–52 (1979). 35. Bhattacharjee, A. et al. Classification of human lung 67. Buckbinder, L. et al. Induction of the growth inhibitor
6. O’Farrell, P. H. High resolution two-dimensional carcinomas by mRNA expression profiling reveals distinct IGF-binding protein 3 by p53. Nature 377, 646–649
electrophoresis of proteins. J. Biol. Chem. 250, adenocarcinoma subclasses. Proc. Natl Acad. Sci. USA. (1995).
4007–4021 (1975). 98, 13790–13795 (2001). 68. Polyak, K. et al. A model for p53-induced apoptosis.
7. Croy, R. G. & Pardee, A. B. Enhanced synthesis and 36. Sorlie, T. et al. Gene expression patterns of breast Nature 389, 300–305 (1997).
stabilization of Mr 68,000 protein in transformed BALB/ carcinomas distinguish tumor subclasses with clinical 69. Wu, G. S. et al. KILLER/DR5 is a DNA damage-inducible
c-3T3 cells: candidate for restriction point control of cell implications. Proc. Natl Acad. Sci. USA 98, p53-regulated death receptor gene. Nature Genet. 17,
growth. Proc. Natl Acad. Sci. USA. 80, 4699–8703 10869–10874 (2001). 141–143 (1997).
(1983). 37. Dhanasekaran, S. M. et al. Delineation of prognostic 70. Gu, Z. et al. ei24, a p53 response gene involved in
8. Sargent, T. D. Isolation of differentially expressed genes. biomarkers in prostate cancer. Nature 412, 822–826 growth suppression and apoptosis. Mol. Cell. Biol. 20,
Methods Enzymol. 152, 423–432 (1987). (2001). 233–241 (2000).
9. Masiakowski, P. et al. Cloning of cDNA sequences of 38. Wooster, R. Cancer classification with DNA microarrays: 71. Israeli, D. et al. A novel p53-inducible gene, PAG608,
hormone-regulated genes from the MCF-7 human is less more? Trends Genet. 16, 327–329 (2000). encodes a nuclear zinc finger protein whose
breast cancer cellline. Nucl. Acids Res. 10, 7895–7903 39. Petricoin, E. F. et al. Medical applications of microarray overexpression promotes apoptosis. EMBO J. 16,
(1982). technologies: a regulatory science perspective. Nature 4384–4392 (1997).
10. Manzari, V. et al. Abundant transcription of a cellular Genet. 32, S474–S4799 (2002). 72. Lo, P. K. et al. Identification of a novel mouse p53 target
gene in T cells infected with human T-cell leukemia- 40. Goodman, N. Microarrays: hazardous to your science. gene DDA3. Oncogene 18, 7765–7774 (1999).
lymphoma virus. Proc. Natl Acad. Sci. USA 80, 11–15 Genome Technol. April, 42–45, (2003). 73. Takei, Y. et al. Isolation of a novel TP53 target gene from
(1983). 41. Ring, B. Z. & Ross D. T. Microarrays and molecular a colon cancer cell line carrying a highly regulated wild-
11. Zimmermann, C. R. et al. Molecular cloning and selection markers for tumor classification. Genome Biol. 3, 2005 type TP53 expression system. Genes Chromosom.
of genes regulated in Aspergillus development. Cell 21, (2002). Cancer 23, 1–9 (1998).
709–715 (1980). 42. Kuo, W. P. et al. Analysis of matched mRNA 74. Ng, C. et al. Isolation and characterization of a novel
12. Hedrick, S. M. et al. Isolation of cDNA clones encoding measurements from two different microarray TP53-inducible gene, TP53TG3. Genes Chromosom.
T cell-specific membrane-associated proteins. Nature technologies. Bioinformatics 18, 405–412 (2002). Cancer 26, 329–335 (1999).
308, 149–153 (1984). 43. Kothapalli, R. et al. Microarray results: how accurate are 75. Tanaka, H. et al. A ribonucleotide reductase gene
13. Davis, M. On the Trail of T-Cell Receptors. they? BMC Bioinformatics 3, 22 (2002). involved in a p53-dependent cell-cycle checkpoint for
Accomplishment in Cancer Research. 87–94 (General 44. Goryachev, A. B, Macgregor, P. F. & Edwards, A. M. DNA damage. Nature 404, 42–49 (2000).
Motor Cancer Research Foundation, 1996). Unfolding of microarray data. J. Comput. Biol. 8, 76. Attardi, L. et al. PERP, an apoptosis-associated target of
14. El-Deiry, W. S. et al. WAF1, a potential mediator of 443–461 (2001). p53, is a novel member of the PMP-22/gas3 family.
p53 tumor suppression. Cell 75, 817–825 (1993). 45. King, H. C. & Sinha, A. A. Gene expression profile Genes Dev. 14, 704–718 (2000).
15. Liang, P. & Pardee, A. B. Differential display of eukaryotic analysis by DNA microarrays: promise and pitfalls. JAMA 77. Saller, E. et al. Increased apoptosis induction by 121F
mRNA by means of the polymerase chain reaction. 286, 2280–2288 (2001). mutant p53. EMBO J. 18, 4424–4437 (1999).
Science 257, 967–971 (1992). 46. Shedden, K. & Cooper, S. Analysis of cell-cycle-specific 78. Oda, E. et al. Noxa, a BH3-only member of the Bcl-2
16. Welsh, J. et al. Arbitrarily primed PCR fingerprinting of gene expression in human cells as determined by familiy and candidate mediator of p53-induced
RNA. Nucl. Acids Res. 20, 4965–4970 (1992). microarrays and double-thymidine block synchronization. apoptosis. Science 288, 1053–1058 (2000).
17. Liang, P. et al. Analysis of altered gene expression by Proc. Natl Acad. Sci. USA 99, 4379–4384 (2002). 79. Lin, Y., Ma, W. & Benchimol, S. Pidd, a new death-
differential display. Methods Enzymol. 254, 304–321 47. Cooper, S. Cell cycle analysis and microarrays. Trends domain-comtaining protein is induced by p53 and
(1995). Genet. 18, 289–290 (2002). promotes apoptosis. Nature Genet. 26, 124–127
18. Liang, P. A decade of differential display. Biotechniques 48. Jenssen T.-K. et al. Analysis of repeatability in spotted (2000).
33, 338–346 (2002). cDNA microarrays. Nucl. Acids Res. 30, 3235–3244 80. Oda, E. et al. p53AIP1, a potential mediator of p53-
19. McCarthy, S. A. et al. Rapid induction of heparin-binding (2002). dependent apoptosis, and its regulation by Ser-46-
epidermal growth factor/diphtheria toxin receptor 49. Quackenbush, J. Microarray data normalization and phosphorylated p53. Cell 102, 849–862 (2000).
expression by Raf and Ras oncogenes. Genes Dev. 9, transformation. Nature Genet. 32, S496–S501 81. Okamura, S. et al. p53DINP1, a p53-inducible gene,
1953–1964 (1995). (2002). regulates p53-dependent apoptosis. Mol. Cell 8, 85–94
20. Zhang, R. et al. Identification of a novel ligand-receptor 50. Adams M. D. et al. Sequence identification of 2,375 (2001).
pair constitutively activated by Ras oncogenes. J. Biol. human brain genes. Nature 355, 632–634 82. Yu, J. et al. PUMA induces the rapid apoptosis of
Chem. 275, 24436–24443 (2000). (1992). colorectal cancer cells. Mol. Cell 7, 673–682 (2001).
21. You, M. et al. ch-IAP1, a member of the inhibitor-of- 51. Velculescu, V. E. et al. Serial analysis of gene expression. 83. Nakano, K. & Vousden, K. H. PUMA, a novel
apoptosis protein family, is a mediator of the Science 270, 484–487 (1995). proapoptotic gene, is induced by p53. Mol. Cell 7,
antiapoptotic activity of the v-Rel oncoprotein. Mol. Cell. 52. Boon, K. et al. An anatomy of normal and malignant gene 683–694 (2001).
Biol. 17, 7328–7341 (1997). expression. Proc. Natl Acad. Sci. USA 99, 11287–11292 84. Leng, R. P. et al. Pirh2, a p53-induced ubiquitin-protein
22. Park, B.-W. et al. Induction of the Tat-binding protein 1 (2002). ligase, promotes p53 degradation. Cell 112, 779–791
gene accompanies the disabling of oncogenic erbB 53. Liang, P. SAGE Genie: a suite with panoramic view of (2003).
receptor tyrosine kinases. Proc. Natl Acad. Sci. USA 96, gene expression. Proc. Natl Acad. Sci. USA 99, 85. Yin, Y. et al. PAC1 phosphatase is a transcription target
6434–6438 (1999). 11547–11548 (2002). of p53 in signalling apoptosis and growth suppression.
23. Wang, M. et al. Interleukin-24 (Mob-5/Mda-7) signals 54. Brenner, S. et al. Gene expression analysis by massively Nature 422, 527–531 (2003).
through two heterodimeric receptors, IL-22R1/IL-20R2 parallel signature sequencing (MPSS) on microbead 86. Owen-Schaub, L. B. et al. Wild-type human p53 and a
and IL-20R1/IL-20R2. J. Biol. Chem. 277, 7341–7347 arrays. Nature Biotechnol. 18, 630–634 2000). temperature-sensitive mutant induce Fas/APO-1
(2002). 55. Kern, S. E. et al. Oncogenic forms of p53 inhibit p53- expression. Mol. Cell. Biol. 15, 3032–3040 (1995).
24. Liang, P. Factors ensuring successful use of differential regulated gene expression. Science 256, 827–830 87. Kannan, K. et al. DNA microarray analysis of genes
display. Methods 16, 361–364 (1998). (1992). involved in p53 mediated apoptosis: activation of Apaf-1.
25. Shimkets, R. A. et al. Gene expression analysis by 56. Deng, C. et al. Mice lacking p21CIP1/WAF1 undergo Oncogene 20, 3449–3455 (2001).
transcript profiling coupled to a gene database query. normal development, but are defective in G1 checkpoint 88. Stambolic, V. et al. Regulation of PTEN transcription by
Nature Biotechnol. 17, 798–803 (1999). control. Cell 82, 675–684 (1995). p53. Mol Cell. 8, 317–325 (2001).
26. Cho, Y. et al. Multi-color fluorescent differential display. 57. El-Deiry, W. S. Regulation of p53 downstream genes.
Biotechniques 30, 562–572 (2001). Semin. Cancer Biol. 8, 345–357 (1998).
27. Schena, M. et al. Quantitative monitoring of gene 58. Zhao, R. et al. Analysis of p53-regulated gene expression Online links
expression patterns with a complementary DNA patterns using oligonucleotide arrays. Genes Dev. 14,
microarray. Science 270, 467–470 (1995). 981–993 (2000). DATABASES
28. Chee, M. et al. Accessing genetic information with high- 59. Wang, L. et al. Analyses of p53 target genes in the The following terms in this article are linked online to:
density DNA arrays. Science 274, 610–614 (1996). human genome by bioinformatic and microarray Cancer.gov: http://cancer.gov/
29. Ramaswamy, S. et al. Multiclass cancer diagnosis using approaches. J. Biol. Chem. 276, 43604–43610 breast cancer | leukaemia | lung cancer | lymphoma | prostate
tumor gene expression signatures. Proc. Natl Acad. Sci. (2001). cancer
USA 98, 15149–15154 (2001). 60. Gibbs, W. W. Shrinking to enormity: DNA microarrays are LocusLink: http://www.ncbi.nlm.nih.gov/LocusLink/
30. Chung, C. H., Bernard, P. S. & Perou, C. M. Molecular reshaping basic biology but scientists fear they may soon BAX | ERBB | GADD45 | IL-24 | MDM2 | p53 | p68 | PIRH2 |
portraits and the family tree of cancer. Nature Genet. 32, drown in the data. Sci. Am. 284, 33–34 (2001). RAS | v-REL | WAF1
S533–S540 (2002). 61. Brenner, S. Sillycon valley fever. Curr. Biol. 9, R671
31. Golub, T. R. et al. Molecular classification of cancer: class (1999). FURTHER INFORMATION
discovery and class prediction by gene expression 62. Gilbert, W. Life after the helix. Nature 421, 315–316 Access to this interactive links box is free online.
monitoring. Science 286, 531–537 (1999). (2003). Beads-based EST sequencing: http://www.lynxgen.com
32. Yeoh, E. J. et al. Classification, subtype discovery, and 63. Kornberg, A. Why purify enzyme? Methods Enzymol. cDNA microarray: http://cmgm.stanford.edu/pbrown/
prediction of outcome in pediatric acute lymphoblastic 182, 1–5 (1990). Differential display: http://www.differentialdisplay.com
leukemia by gene expression profiling. Cancer Cell 2, 64. Wu, X. et al. The p53-mdm-2 autoregulatory feedback GeneChip array: http://www.affymetrix.com
133–143 (2002). loop. Genes Dev. 7, 1126–1132 (1993). International Symposia on Differential Gene Expression:
33. Alizadeh, A. A. et al. Distinct types of diffuse large B-cell 65. Miyashita, T. & Reed, J. C. Tumor suppressor p53 is a http://medschool.mc.vanderbilt.edu/GeneXP/
lymphoma identified by gene expression profiling. Nature direct transcriptional activator of the human bax gene. SAGE: http://www.sagenet.org
403, 503–511 (2000). Cell 80, 293–299 (1995). SAGE Genie: http://cgap.nci.nih.gov/SAGE

876 | NOVEMBER 2003 | VOLUME 3 www.nature.com/reviews/cancer