Download Two groups of DNA methylation markers to classify colorectal cancer

Review Article
Two groups of DNA methylation markers to classify
colorectal cancer into three epigenotypes
Atsushi Kaneda1,2,3 and Koichi Yagi1
1Genome Science Division, Research Center for Advanced Science and Technology, University of Tokyo, Meguro-ku, Tokyo; 2PRESTO, Japan Science and
Technology Agency, Tokyo, Japan
(Received April 12, 2010 ⁄ Revised July 30, 2010 ⁄ Accepted August 6, 2010 ⁄ Accepted manuscript online August 12, 2010 ⁄ Article first published online September 6, 2010)
A subgroup of colorectal cancer (CRC) shows non-random accumulation of aberrant DNA methylation, so-called CpG island
methylator phenotype (CIMP), which was associated with microsatellite instability and BRAF mutation. As just one group of
methylation markers was suitable to extract CIMP+ ⁄ CIMP-high,
and had been commonly used in the ‘‘one-panel method’’, it had
been unclear whether another cluster of CRC with DNA methylation accumulation exists in microsatellite-stable CRC. We therefore epigenotyped CRC by a comprehensive approach, that is, the
two-way unsupervised hierarchical clustering method using
highly quantitative methylation data by a single detection
method, MALDI-TOF mass spectrometry, on novel regions selected
genome-widely through methylated DNA immunoprecipitation on
array analysis. CRC was clearly clustered into three DNA methylation epigenotypes, high-, intermediate- and low-methylation epigenotypes (HME, IME, and LME, respectively). Methylation
markers are clustered into two distinct groups: Group-1 methylated specifically in HME and including most reported CIMP-related
markers; and Group-2 methylated both in HME and IME. While
suitable markers to detect a subgroup of CRC with intermediate
methylation and correlation to KRAS mutation have been
expected to be developed, our data indicated that a ‘‘two-panel
method’’ is necessary to properly classify CRC into three epigenotypes, the first panel to extract HME using Group-1 markers, and
the second panel to divide the remaining into IME and LME using
Group-2 markers. Here we review and compare our recent study
and reported CRC classification methods by DNA methylation
information, and propose the use of two panels of methylation
markers as CRC classifiers. (Cancer Sci 2011; 102: 18–24)
It has been reported that a subset of cancer cases shows
aberrant methylation in many genes ⁄ loci and another shows
methylation in less number or none of the genes ⁄ loci at all
(Table 1, Fig. 1).(12,20–23) These non-random accumulations of
methylation were first indicated in CRC in 1999, proposing
‘‘CpG island methylator phenotype’’ as a phenotype of cancer
cases with aberrant methylation of significantly more numbers
of genomic marker regions.(12) Most studies during the last decade have used one group of classifier markers, including the original CIMP markers (namely, the ‘‘one-panel method’’)
(Fig. 1A,B). We recently revealed through unsupervised twoway hierarchical clustering that CRC is clearly classified into
three groups by DNA methylation information only: HME,
IME, and LME (Fig. 1D).(23) Intermediate-methylation epigenotype strongly correlates to KRAS-mutation(+) CRC. It is noteworthy that methylation markers are clustered into two distinct
groups, Group-1 and Group-2 (Fig. 1D). Group-1 markers are
methylated specifically in HME, so are suitable to extract HME
CRC. Group-2 markers are methylated both in HME and
IME, so can divide CRC into IME and LME after exclusion of
HME CRC. Here we review and compare our epigenotyping
model and reported CRC classification methods, and propose a
‘‘two-panel method’’ to classify CRC properly (Fig. 1D).
Previously Reported Methods ⁄ Reports on CIMP
Table 1 summarizes several methods ⁄ reports on CRC classification by methylation accumulation phenotypes.(12,20–24)
Proposal of CIMP+ CRC
E
pigenetic information is modification of DNA and its associated proteins, not the DNA sequence itself, and is heritable in cell division. It is widely accepted now that cancer arises
through accumulation of genetic alterations as well as epigenetic
alterations, such as aberrant DNA methylation to silence gene
expressions, and loss of imprinting to double expression of
imprinted genes.(1–3) The importance of epigenetic alterations is
noted not only in cancer progression, but also in cancer initiation,(3,4) as the alterations can be accumulated in the pre-cancerous ‘‘normal’’ tissues to modify a cancer risk,(5,6) and the risk
might be managed by target therapy.(7)
Gene silencing by DNA methylation of its promoter region is
one of the most important epigenetic mechanisms to inactivate
gene expression.(2) Genome-wide search of genes using aberrant
methylation as markers is useful to identify novel tumor-suppressor genes and methylation markers for cancer classification.(8–13) Several approaches to detect methylated regions have
been developed, originally analyzing limited regions represented
by methylation-sensitive restriction enzymes,(14) and recently
analyzing regions on a genome-wide scale.(15–19)
Cancer Sci | January 2011 | vol. 102 | no. 1 | 18–24
In 1999, Toyota et al. developed MCA to represent methylated
genomic regions using methylation-sensitive restriction enzymes.
Thirty-three genomic DNA clones that aberrantly methylated in
cancer cell line Caco2, but not in normal colon tissue, were identified using a subtraction method, representation difference analysis.(10) The 33 clones were named MINT. Most of the MINT
clones showed methylation not only in CRC, but also in normal
colon mucosa in an age-related manner, thus were called type A.
These type A clones were methylated in most CRC cases, so are
considered to be non-classifier markers. The remaining seven
clones were methylated specifically in cancer, not in normal
mucosa, thus called type C. The type C markers were methylated
in CRC less frequently than type A, and classified CRC cases sharply into two groups: a group with methylation in three or more
loci; and a group with extremely rare methylation. The former
CRC subset was considered as CIMP+, and these type C MINT
clones were considered as CIMP markers (Table 1, Fig. 1A).(12)
3To whom correspondence should be addressed.
E-mail: [email protected]
doi: 10.1111/j.1349-7006.2010.01712.x
ª 2010 Japanese Cancer Association
Table 1.
Reports on colorectal cancer (CRC) classification by methylation information
Toyota et al.(12)
Yamashita et al.(24)
Year
1999
2003
Marker selection (sample used)
MCA-RDA (Caco2)
MS-AFLP (32 CRCs)
Weisenberger
et al.(20)
No. of phenotypes
2
Marker panel
Methylation
Shen et al.(22)
Yagi et al.(23)
2005
2006
2007
2010
MethyLight
markers
Reported
markers
Reported
markers
MeDIP-chip &
expression array
(HCT116, SW480)
0,
+8
0,
+27
1311 fi 44,
+16
MethyLight
Pyrosequence
COBRA
MCA
MSP
MassARRAY
No. of original markers, +no. of previously reported markers
33 fi 7,
203 fi 30,
195 fi 88,
+0
+6
+0
Quantitative methylation analysis method
COBRA
MS-AFLP
MethyLight
MSP
Classification method
Methylation
frequency
Methylation phenotypes
CIMP+
CIMP)
Ogino et al.(21)
Methylation
frequency
Hierarchical
clustering
Methylation
frequency
Hierarchical
clustering
Hierarchical
clustering
(Absent)
CIMP+
CIMP)
CIMP-high
CIMP-low
CIMP-0
CIMP1
CIMP2
CIMP-negative
HME
IME
LME
0
2
3
3
3
NA
Methylation
Methylation
Genetic and
methylation
Methylation
1
1
NA
2
Fig. 1A
Fig. 1B
Fig. 1C
Fig. 1D
No. of methylation marker panels
1
NA
See also
Fig. 1A
NA
CIMP, CpG island methylator phenotype; COBRA, combined bisulfite restriction analysis; HME, high-methylation epigenotype; IME, intermediatemethylation epigenotype; LME, low-methylation epigenotype; MCA, methylated CpG island amplification; MeDIP, methylated DNA
immunoprecipitation; MS-AFLP, methylation-sensitive amplified fragment length polymorphism; MSP, methylation-specific PCR; NA, not
applicable; RDA, representation difference analysis.
Hierarchical Clustering Using Quantitative Data to Detect
CIMP+ CRC
In 2006, Weisenberger et al.(20) used quantitative methylation
data to classify CRC using a powerful classification method,
unsupervised hierarchical clustering, and successfully confirmed
the existence of CIMP. They carried out a first screen of all the
195 MethyLight markers available at that time against 10 pairs
of normal colon and tumor tissue samples to select 92 cancerspecific methylation markers. Second, the quantitative methylation status of the 92 markers in 48 pairs of normal colon and
tumor tissue samples was analyzed by MethyLight. Hierarchical
clustering using these data revealed that there is a cluster of CRC
showing high methylation and strong correlation to BRAF mutation and MSI. They unfortunately did not find another cluster of
CRC showing intermediate methylation at this second step. Then
a third independent set of 187 tumors were analyzed, and a new,
more suitable panel of five methylation markers to determine
CIMP+ was proposed, comprising CACNA1G, IGF2, NEUROG1, RUNX3, and SOCS1 (Table 1, Fig. 1A). CACNA1G
possesses MINT31, one of the original CIMP markers by Toyota
et al.(12), at 2 kb upstream of its transcription start site.(23)
Lack of Group-2 Markers in CIMP+ Model: Importance of
Comprehensive Search for Candidate Markers
MINT clones identified by MCA were considered to belong to
Group-1 (Fig. 2A), so were actually suitable to extract
Kaneda and Yagi
CIMP+ ⁄ HME CRC. But unfortunately MINT clones did not
include Group-2 markers, so could not detect IME.(23) If more
than one CRC cell line (Caco2) had been analyzed by MCA, or
a more genome-wide approach had been available in the analysis of Caco2, Group-2 markers might have been identified.
Although p16CDKN2A and hMLH1 were often analyzed with
MINT clones, hMLH1 is a Group-1 marker showing no methylation in IME or LME (Fig. 2A). p16CDKN2A shows only slightly
higher methylation in IME than LME, so still belongs to Group1.(23)
NEUROG1 is a Group-2 marker, but shows even higher average methylation levels in HME than IME (See Fig. 3A,
H > I > L), so was still suitable to be selected in the marker
panel for CIMP+. The other four Weisenberger markers are
methylated specifically in HME, and show low and very similar
methylation levels in IME and LME (Fig. 3A). Overall, Weisenberger’s panel is very suitable to detect CIMP+ ⁄ HME, but cannot identify IME. Two possibilities can be raised why the IME
cluster was not identified in their study. First, the 48 tumor samples in the second screen might not have been enough to make
IME cluster clearly, although as many as 19 KRAS-mutation(+)
CRC were included in the 48. Second, their 195 MethyLight
markers(20) might not have included enough Group-2 markers
to make IME cluster. In fact, the 26 Group-2 markers we
identified(23) were not included in the 195 MethyLight markers
other than NEUROG1. A comprehensive genome-wide
search for marker candidates is suggested to be important and
necessary.
Cancer Sci | January 2011 | vol. 102 | no. 1 | 19
ª 2010 Japanese Cancer Association
(A)
(C)
(D)
(B)
Fig. 1. Models for phenotypes of methylation accumulation. X-axes, colorectal cancer (CRC) cases; Y-axes, markers ⁄ genes. Grey boxes indicate
methylation(+). (A) CpG island methylator phenotype (CIMP)+ ⁄ ).(12,20) The representative marker panel contains five markers. CIMP+, colorectal
cancer (CRC) methylated in 3–5 among five markers, which correlates to microsatellite instability (MSI)-high and BRAF-mutation(+); CIMP), CRC
methylated in 0–2 among five markers. MSS, microsatellite stable. (B) CIMP-high ⁄ -low ⁄ -0.(21,28) Similar to the CIMP+ ⁄ ) model, only one marker
panel is used, and the representative marker panel contains eight markers. CIMP-0, CRC with no methylation; CIMP-high, CRC methylated in 6–8
among eight markers; CIMP-low, CRC methylated in 1–5 among eight markers. CIMP-low was reported to correlate possibly with KRAS
mutation(+). (C) CIMP1 ⁄ 2 ⁄ -negative.(22) Genetic markers are also used in determining CIMP status, but they perform better than methylation
markers in classification. CIMP1, cluster with BRAF mutation and CIMP1 marker methylation; CIMP2, cluster with KRAS mutation and CIMP2
marker methylation; CIMP-negative, cluster with p53 mutation and absence of methylation. (D) High-methylation epigenotype (HME),
intermediate-methylation epigenotype (IME), and low-methylation epigenotype (LME).(23) HME, CRC with methylation of Group-1 and Group-2
markers, correlating strongly to MSI-high and BRAF-mutation(+). In MSS CRC, there is no clear subset showing random methylation of Group-1
markers like the CIMP-high ⁄ -low ⁄ -0 model. Instead, there is another clear subset IME showing accumulation of methylation of Group-2 markers
only, and correlating to KRAS-mutation(+). Two-step classification using two marker panels is proposed: the first panel to extract CRC with
Group-1 marker methylation as HME; and the second panel to divide the remaining into CRC with Group-2 marker methylation as IME and the
others as LME. There is no need to use genetic markers in this classification if using our novel Group-2 methylation markers.
Absence of CIMP? Importance of Quantitative Analysis
and Classifier Marker Selection
When methylation profiles of hMLH1, MGMT, p16CDKN2A,
p14ARF, APC, and CDH1 in 207 CRC were analyzed by
Yamashita et al.(24) using MSP, the number of methylated loci
per tumor showed a non-bimodal distribution. The result denied
non-random accumulation of methylation in a subgroup of CRC,
and suggested the absence of CIMP (Table 1). The concept and
definition of CIMP had been arbitrary at that time, as they
pointed out, and they successfully made a stir in the argument
on CIMP. However, MSP is not a quantitative analysis and
could give false positive data by detecting very infrequent methylation at few CpG sites. In fact, MGMT seemed to be the most
frequently methylated gene in their MSP analysis, which caused
24 (12%) additional CRC cases with one methylated locus, that
is, a decrease of 24 CRC cases without methylation.(24) But our
quantitative methylation data indicated that MGMT is an outlier,
non-classifier gene belonging to neither Group-1 nor Group-2
due to its low methylation score in any epigenotypes of CRC.(23)
In Yamashita’s(25) report, 203 methylated regions were also
identified by methylation-sensitive amplified fragment length
polymorphism analysis of 32 CRCs, and methylation of 30 chosen markers did not show a non-random pattern but distributed
evenly in CRC cases again.(24) This suggested to us that a panel
of randomly chosen markers could give no cluster of CRC cases
due to contamination with many non-classifier markers. In
Weisenberger’s classification, non-cancer-specific MethyLight
markers were excluded before hierarchical clustering analysis,
and most cancer-specific markers were further excluded in
developing the best CIMP panel.(20) Two-way hierarchical
20
clustering in our study revealed the existence of two major marker groups, Group-1 and Group-2, but also the existence of outlier, non-classifier genes.(23) Methylation epigenotype, or CIMP,
is not a phenotype of simultaneous methylation of all the CpG
islands in a genome. We must develop a decision panel by
selecting classifier genes cautiously.
Proposal of CIMP-Low
An additional subset, such as CIMP-intermediate or CIMP-low,
has been described to embrace tumors with intermediate levels
of methylation.(26,27) In 2006, for example, Ogino et al.(21) proposed a CIMP-low subset of CRC, showing 1–3 methylated promoters among CACNA1G, p16CDKN2A, CRABP1, hMLH1, and
NEUROG1. They reported that the CIMP-low CRC is associated
with male sex and KRAS mutations, and thus suggested that
CIMP-low CRC is a different subtype from CIMP-high and
CIMP-0. This classification is not ideal, but might work as NEUROG1 is a Group-2 marker showing higher methylation in IME
than LME (Fig. 2). Colorectal cancer with methylation of NEUROG1 is regarded as CIMP-low, although CRC with random
methylation of other genes is also regarded as CIMP-low. In
another report, they suggested using eight markers (CACNA1G,
p16CDKN2A, CRABP1, IGF2, hMLH1, NEUROG1, RUNX3, and
SOCS1) and defining CRC with 1 ⁄ 8–5 ⁄ 8 methylated promoters
as CIMP-low, 0 ⁄ 8 as CIMP-0, and 6 ⁄ 8–8 ⁄ 8 as CIMP-high
(Fig. 1B).(28,29)
Ogino et al.(30) noted, however, that a difference between
CIMP-low and CIMP-0 is not clear-cut as these methylation
markers were specific for CIMP-high and not ideal for identification of CIMP-low, and that sensitive and specific markers for
doi: 10.1111/j.1349-7006.2010.01712.x
ª 2010 Japanese Cancer Association
(A)
(B)
Fig. 2. Classification of methylation markers. (A) LOX or many CpG island methylator phenotype (CIMP) markers (e.g. methylated-in-tumor
[MINT]1 and MINT2) show methylation specifically to high-methylation epigenotype (HME; H) only, and belong to Group-1 (H >> I = L). Other
CIMP markers (e.g. CACNA1G and MINT17) show slightly higher methylation in intermediate-methylation epigenotype (IME; I), but hardly
distinguish IME and low-methylation epigenotype (LME; L) because of the tiny difference (H >> I > L). Group-2 markers show significantly higher
methylation in IME than LME, thus can distinguish the two epigenotypes. Some Group-2 markers (e.g. HAND1 and ELMO1) show even higher
methylation in HME (H > I > L), and some (e.g. THBD and ADAMTS1) show similar, very high methylation levels in both HME and IME (H = I > L).
Among reported CIMP markers, only NEUROG1 belongs to Group-2. There are non-classifier genes (e.g. CDO1, CIDEB, and RASSF1A) (H = I = L).
(B) If a marker shows a high methylation level specifically to IME or LME only, it would not be necessary to use two panels at two steps (Fig. 1D)
and classification would be much easier. These specific markers were not identified, however, at least by our previous studies. There were
reports suggesting that RASSF2 or IGFBP3 might be specific markers for intermediate methylation phenotype,(29,35) but these genes were shown
to belong to the H = I > L pattern (Fig. 3A) in our study.
Fig. 3. Quantitativity of MALDI-TOF mass spectrometry (MassARRAY). (A) MassARRAY is highly
quantitative
and
reproducible.
Duplicated
methylation control samples (methylation 0%, 25%,
50%, 75%, 100%) showed a linear standard curve
with high correlation coefficient (R2). (B) Because of
the sequence difference between methylated and
unmethylated DNA after bisulfite treatment, their
amplification efficiencies could be different and
PCR bias is often observed. Methylated alleles
would tend to be more efficiently amplified, as
shown. When R2 was low (£0.9), the primer pair
was excluded or redesigned.
CIMP-low need to be determined if CIMP-low really exist. This
is a good support for our two-panel method. They reported that
BRAF-mutation(+) CRC showed a non-random pattern of CpG
island methylation, whereas KRAS-mutation(+) CRC showed a
Kaneda and Yagi
random pattern, by analyzing 16 regions.(31) This also indicated
that these previous markers specific for HME ⁄ CIMPhigh, BRAF-mutation(+) CRC could show methylation in
IME ⁄ CIMP-low, KRAS-mutation(+) CRC only randomly. If
Cancer Sci | January 2011 | vol. 102 | no. 1 | 21
ª 2010 Japanese Cancer Association
(A)
(B)
MINT loci to be highly specific to CIMP2 and to predict CIMP2
better than NEUROG1, by analyzing methylation using a competitive PCR method MCA. The MINT loci, however, should
show no or very small differences in average methylation levels
between IME and LME and belong to Group-1 markers (Fig. 2),
so were considered to be difficult to classify CIMP2 and CIMPnegative.(23) In any classification study, all the methylation data
are to be completed using a quantitative method.(32)
Our CRC Epigenotyping
Fig. 4. Models of mechanisms correlating between oncogene
mutation and methylation profile. M in a circle means methylation.
(A) During the tumorigenesis step, oncogene activation might
somehow induce specific methylation epigenotypes. (B) In an early
step of tumorigenesis, promoter methylation might be accumulated
and be requisite to escape from oncogene-induced senescence.
Without methylation, BRAF or KRAS mutation might lead to cellular
senescence, and other aberrations might be necessary to develop
tumor. With Group-2 methylation accumulated, cells might escape
from activated KRAS-induced senescence and could lead to tumor
development, although Group-2 methylation might not be enough
for cells to escape from activated BRAF-induced senescence. With
Group-1 and Group-2 methylation, cells might escape from activated
BRAF-induced senescence and could lead to tumor development. In
this model, accumulated methylation of Group-1 and Group-2 genes
could be a risk for colorectal cancer (CRC) with oncogene activation.
HME, high-methylation epigenotype; IME, intermediate-methylation
epigenotype; LME, low-methylation epigenotype.
Group-2 markers showing a non-random pattern of methylation
in KRAS-mutation(+) CRC are used in a panel, the difference
between IME ⁄ CIMP-low and LME ⁄ CIMP-0 will be clear-cut.
Classification into Three Subsets
In 2007, Shen et al.(22) proposed through genetic and epigenetic
analysis that colon cancer was classified into three subsets,
CIMP1, CIMP2, and CIMP-negative (Table 1, Fig. 1C). This
report successfully showed the existence of three clusters of
CRC with different molecular characteristics: one with MSIhigh, BRAF-mutation(+) and methylation, another with KRASmutation(+) and different methylation, and the other with p53
mutation and an absence of these methylation.
Genetic markers performed equally well or better than epigenetic markers in their report, so they highlighted the importance
of integrated genetic and epigenetic analysis.(22) KRAS mutation
itself was used in the CIMP2 marker panel, for example. The
questions whether CRC is classified into three subsets by methylation accumulation phenotypes only, and whether there are
methylation markers suitable to detect intermediate methylation
subgroup or KRAS-mutation(+) CRC, were therefore not clearly
answered at this stage.
The authors analyzed 27 previously reported regions using
four different methylation detection methods: 13 regions by Pyrosequencing; 7 by COBRA; 6 by MCA; and 1 by MSP.(22)
They found methylation of NEUROG1 to be highly specific to
CIMP2, so their hierarchical clustering might be quite similar to
our classification. They also reported methylation of several
22
Attention in our methods. We therefore carried out epigenotyping of CRC very carefully by comprehensive approach, i.e.,
two-way unsupervised hierarchical clustering using highly quantitative methylation data by a single detection method, MALDITOF mass spectrometry (MassARRAY; Sequenom, San Diego,
CA, USA),(33) using genome-widely selected novel regions
through MeDIP-chip analysis.(23)
Although MeDIP is known to be innaccurate for analyzing
low-CpG regions, it is a powerful tool to identify methylated
high-CpG regions, for example, promoter CpG islands.(16,34) We
used MeDIP on two CRC cell lines, MSI-high HCT116 and
microsatellite-stable SW480. The MeDIPed samples underwent
unbiased amplification by in vitro transcription before hybridization, and samples were hybridized to Human Promoter 1.0R tiling array covering 4 071 296 CpG sites around over 25 500
promoter regions comprehensively.(16,23)
To select genes whose promoter methylation causes gene
silencing, we also carried out expression array analysis of the
CRC cell lines with ⁄ without treatment by 5-aza-2¢-deoxycytidine and trichostatin A, and selected 55 genes showing expression in normal colon, silencing in methylated cell lines, and
re-expression after 5-aza-2¢-deoxycytidine and trichostatin A
treatment, from 1311 candidates.(23)
To obtain highly quantitative data for these 55 genes and previously reported 19 regions, quantitativity of 74 primer pairs for
MassARRAY was carefully validated using methylation control
samples (Fig. 3).(23) Because of the sequence difference between
methylated DNA and unmethylated DNA after bisulfite treatment, amplification efficiencies of the two could be different
and PCR bias is often observed. A linear standard curve was
drawn and the correlation coefficient (R2) was calculated at each
CpG unit. When there was no PCR bias, methylated and unmethylated alleles were amplified equally, and R2 would be high
(>0.9) (Fig. 3A). When there was PCR bias, R2 would be low
(£0.9), usually because methylated alleles were amplified more
efficiently than unmethylated alleles (Fig. 3B). Only when three
or more CpG units with R2 > 0.9 existed, the primer pairs were
considered to be highly quantitative enough and used for further
analyses. Primers for 60 markers including 44 novel markers
and 16 reported genes ⁄ loci were successfully developed.
Clinical CRC samples were microscopically examined for
determination of cancer cell contents, and were dissected to
enrich cancer cells to 40% or more when necessary.(23)
Three epigenotypes of CRC cases and two clusters of
markers. Two-way unsupervised hierarchical clustering classi-
fied CRC cases into three distinct epigenotypes: HME, IME, and
LME (Fig. 1D). The high-methylation epigenotype strongly
correlates to BRAF mutation and MSI-high, and IME strongly
correlates to KRAS mutation.(23) Without genetic markers, CRC
is clearly classified into three subclasses using methylation
information only. It is also noteworthy that IME KRASmutation(+) CRC showed significantly worse prognoses.
Genes were also clustered into two major groups, Group-1 and
Group-2, forming several outlier ⁄ non-classifier genes as mentioned above (Figs 1D, 2A).(23) Markers methylated specifically
in IME or LME have not been identified (Fig. 2B). Two marker
panels are therefore necessary to classify CRC at two steps: the
doi: 10.1111/j.1349-7006.2010.01712.x
ª 2010 Japanese Cancer Association
first panel to extract HME using Group-1 markers; and the second
panel to divide the remaining into IME and LME (Fig. 1D). If we
answer to Ogino’s question,(30) Group-2 markers could be the
suitable markers for the intermediate methylation phenotype,
although combined use of Group-1 and Group-2 markers are necessary. Previous CIMP-related markers were all categorized into
Group-1 markers except NEUROG1. Genome-wide search for
novel markers enabled us to avoid bias of available markers to
Group-1 in the previous studies.
Nagasaka et al.(35) analyzed methylation of seven canonical
CIMP markers and seven new markers in CRC, and reported
that methylation frequency of seven canonical CIMP markers
was high in BRAF-mutation(+) CRC but low in KRAS-mutation(+) CRC, whereas methylation frequency of the additional
markers was high in both BRAF-mutation(+) and KRAS-mutation(+) CRC. Their argument could be similar to ours.
Why do oncogene mutations correlate to epigenotypes? The
strong correlations between HME and BRAF(+), and between
IME and KRAS(+), are interesting, and possibly suggest a causal
link between methylation epigenotypes and oncogene mutation.(23,35) Oncogene activation might somehow induce different
epigenetic, genome-wide alterations (Fig. 4A). In fact, silencing
of genes occurred in Kras-transformed NIH3T3 but not in
untransformed NIH3T3, and essential effectors for the epigenetic
gene silencing in Kras-transformed NIH3T3 were reported.(36)
Another possibility is that promoter methylation might be accumulated in an early step of tumorigenesis.(3) Oncogene activation
is known to induce a premature form of cellular senescence and
fall into irreversible arrest to block cellular proliferation.(37)
While cells without methylation might senesce, cells with Group1 and ⁄ or Group2 gene silencing might escape from oncogeneinduced senescence and become tumor when an oncogene(s) is
activated (Fig. 4B), as disruption of p16 and ⁄ or p53 leads to
escape from oncogene-induced senescence.(38) For example,
IGFBP3, reported to be induced in replicatively senescent
HUVEC,(39) showed a high methylation rate in IME and HME
CRC (Fig. 2A), although colon epithelium cells are different
from endothelial cells. Further studies are necessary to establish
which gene ⁄ signal activation is essential in oncogene-induced
senescence, and which is actually inactivated by promoter methylation in tumor with a specific methylation epigenotype.
References
14 Ushijima T. Detection and interpretation of altered methylation patterns in
cancer cells. Nat Rev Cancer 2005; 5: 223–31.
15 Weber M, Davies JJ, Wittig D et al. Chromosome-wide and promoter-specific
analyses identify sites of differential DNA methylation in normal and
transformed human cells. Nat Genet 2005; 37: 853–62.
16 Hayashi H, Nagae G, Tsutsumi S et al. High-resolution mapping of DNA
methylation in human genome using oligonucleotide tiling array. Hum Genet
2007; 120: 701–11.
17 Lister R, Pelizzola M, Dowen RH et al. Human DNA methylomes at base
resolution show widespread epigenomic differences. Nature 2009; 462: 315–
22.
18 Gu H, Bock C, Mikkelsen TS et al. Genome-scale DNA methylation mapping
of clinical samples at single-nucleotide resolution. Nat Methods 2010; 7: 133–6.
19 Laird PW. Principles and challenges of genome-wide DNA methylation
analysis. Nat Rev Genet 2010; 11: 191–203.
20 Weisenberger DJ, Siegmund KD, Campan M et al. CpG island methylator
phenotype underlies sporadic microsatellite instability and is tightly associated
with BRAF mutation in colorectal cancer. Nat Genet 2006; 38: 787–93.
21 Ogino S, Kawasaki T, Kirkner GJ, Loda M, Fuchs CS. CpG island methylator
phenotype-low (CIMP-low) in colorectal cancer: possible associations with
male sex and KRAS mutations. J Mol Diagn 2006; 8: 582–8.
22 Shen L, Toyota M, Kondo Y et al. Integrated genetic and epigenetic analysis
identifies three different subclasses of colon cancer. Proc Natl Acad Sci U S A
2007; 104: 18654–9.
23 Yagi K, Akagi K, Hayashi H et al. Three DNA methylation epigenotypes in
human colorectal cancer. Clin Cancer Res 2010; 16: 21–33.
24 Yamashita K, Dai T, Dai Y, Yamamoto F, Perucho M. Genetics supersedes
epigenetics in colon cancer phenotype. Cancer Cell 2003; 4: 121–31.
25 Yamamoto F, Yamamoto M, Soto JL et al. Notl-Msell methylation-sensitive
amplied fragment length polymorhism for DNA methylation analysis of
human cancers. Electrophoresis 2001; 22: 1946–56.
26 Toyota M, Ahuja N, Suzuki H et al. Aberrant methylation in gastric cancer
associated with the CpG island methylator phenotype. Cancer Res 1999; 59:
5438–42.
27 Ueki T, Toyota M, Sohn T et al. Hypermethylation of multiple genes in
pancreatic adenocarcinoma. Cancer Res 2000; 60: 1835–9.
1 Feinberg AP, Tycko B. The history of cancer epigenetics. Nat Rev Cancer
2004; 4: 143–53.
2 Jones PA, Baylin SB. The epigenomics of cancer. Cell 2007; 128: 683–92.
3 Baylin SB, Ohm JE. Epigenetic gene silencing in cancer – a mechanism for
early oncogenic pathway addiction? Nat Rev Cancer 2006; 6: 107–16.
4 Feinberg AP, Ohlsson R, Henikoff S. The epigenetic progenitor origin of
human cancer. Nat Rev Genet 2006; 7: 21–33.
5 Sakatani T, Kaneda A, Iacobuzio-Donahue CA et al. Loss of imprinting of
Igf2 alters intestinal maturation and tumorigenesis in mice. Science 2005; 307:
1976–8.
6 Kaneda A, Feinberg AP. Loss of imprinting of IGF2: a common epigenetic
modifier of intestinal tumor risk. Cancer Res 2005; 65: 11236–40.
7 Kaneda A, Wang CJ, Cheong R et al. Enhanced sensitivity to IGF-II signaling
links loss of imprinting of IGF2 to increased cell proliferation and tumor risk.
Proc Natl Acad Sci U S A 2007; 104: 20926–31.
8 Hayashizaki Y, Hirotsune S, Okazaki Y et al. Restriction landmark genomic
scanning method and its various applications. Electrophoresis 1993; 14: 251–
8.
9 Ushijima T, Morimura K, Hosoya Y et al. Establishment of methylationsensitive-representational difference analysis and isolation of hypo- and
hypermethylated genomic fragments in mouse liver tumors. Proc Natl Acad
Sci U S A 1997; 94: 2284–9.
10 Toyota M, Ho C, Ahuja N et al. Identification of differentially methylated
sequences in colorectal cancer by methylated CpG island amplification.
Cancer Res 1999; 59: 2307–12.
11 Kaneda A, Kaminishi M, Yanagihara K, Sugimura T, Ushijima T.
Identification of silencing of nine genes in human gastric cancers. Cancer Res
2002; 62: 6645–50.
12 Toyota M, Ahuja N, Ohe-Toyota M, Herman JG, Baylin SB, Issa JP. CpG
island methylator phenotype in colorectal cancer. Proc Natl Acad Sci U S A
1999; 96: 8681–6.
13 Kaneda A, Wakazono K, Tsukamoto T et al. Lysyl oxidase is a tumor
suppressor gene inactivated by methylation and loss of heterozygosity in
human gastric cancers. Cancer Res 2004; 64: 6410–15.
Kaneda and Yagi
Conclusion
Comprehensive analysis through the selection of novel markers
on a genome-wide scale and highly quantitative data allowed us
to classify CRC, using methylation information only, into three
epigenotypes: HME, IME, and LME. Methylation markers are
classified into two clusters: Group-1 showing methylation in
HME; and Group-2 showing methylation in HME and IME.
These two groups of markers are proposed to be used at two
steps in classifying CRC properly.
Acknowledgments
This work was supported by PRESTO program from Japan Science and
Technology Agency and a Grant-in-Aid for Cancer Research from the
Ministry of Health, Labour and Welfare, Japan.
Abbreviations
CIMP
CRC
HME
IME
LME
MCA
MeDIP
MINT
MSI
MSP
CpG island methylator phenotype
colorectal cancer
high-methylation epigenotype
intermediate-methylation epigenotype
low-methylation epigenotype
methylated CpG island amplification
methylated DNA immunoprecipitation
methylated-in-tumor
microsatellite instability
methylation-specific PCR
Cancer Sci | January 2011 | vol. 102 | no. 1 | 23
ª 2010 Japanese Cancer Association
28 Ogino S, Kawasaki T, Kirkner GJ, Suemoto Y, Meyerhardt JA, Fuchs CS.
Molecular correlates with MGMT promoter methylation and silencing support
CpG island methylator phenotype-low (CIMP-low) in colorectal cancer. Gut
2007; 56: 1564–71.
29 Kawasaki T, Nosho K, Ohnishi M et al. IGFBP3 promoter methylation in
colorectal cancer: relationship with microsatellite instability, CpG island
methylator phenotype, and p53. Neoplasia 2007; 9: 1091–8.
30 Ogino S, Goel A. Molecular classification and correlates in colorectal cancer.
J Mol Diagn 2008; 10: 13–27.
31 Nosho K, Irahara N, Shima K et al. Comprehensive biostatistical analysis of
CpG island methylator phenotype in colorectal cancer using a large
population-based sample. PLoS ONE 2008; 3: e3698.
32 Ogino S, Cantor M, Kawasaki T et al. CpG island methylator phenotype
(CIMP) of colorectal cancer is best characterised by quantitative DNA
methylation analysis and prospective cohort studies. Gut 2006; 55: 1000–6.
33 Ehrich M, Nelson MR, Stanssens P et al. Quantitative high-throughput
analysis of DNA methylation patterns by base-specific cleavage and mass
spectrometry. Proc Natl Acad Sci U S A 2005; 102: 15785–90.
24
34 Weber M, Hellmann I, Stadler MB et al. Distribution, silencing potential and
evolutionary impact of promoter DNA methylation in the human genome. Nat
Genet 2007; 39: 457–66.
35 Nagasaka T, Koi M, Kloor M et al. Mutations in both KRAS and BRAF may
contribute to the methylator phenotype in colon cancer. Gastroenterology
2008; 134: 1950–60.
36 Gazin C, Wajapeyee N, Gobeil S, Virbasius CM, Green MR. An elaborate
pathway required for Ras-mediated epigenetic silencing. Nature 2007; 449:
1073–7.
37 Serrano M, Lin AW, McCurrach ME, Beach D, Lowe SW. Oncogenic ras
provokes premature cell senescence associated with accumulation of p53 and
p16INK4a. Cell 1997; 88: 593–602.
38 Sharpless NE, DePinho RA. Cancer: crime and punishment. Nature 2005;
436: 636–7.
39 Kim KS, Kim MS, Seu YB, Chung HY, Kim JH, Kim JR.
Regulation of replicative senescence by insulin-like growth factor-binding
protein 3 in human umbilical vein endothelial cells. Aging Cell 2007; 6:
535–45.
doi: 10.1111/j.1349-7006.2010.01712.x
ª 2010 Japanese Cancer Association