CpG Island Methylator Phenotype in Colorectal Cancer

Abstract

Aberrant methylation of promoter region CpG islands is associated with transcriptional inactivation of tumor-suppressor genes in neoplasia. To understand global patterns of CpG island methylation in colorectal cancer, we have used a recently developed technique called methylated CpG island amplification to examine 30 newly cloned differentially methylated DNA sequences. Of these 30 clones, 19 (63%) were progressively methylated in an age-dependent manner in normal colon, 7 (23%) were methylated in a cancer-specific manner, and 4 (13%) were methylated only in cell lines. Thus, a majority of CpG islands methylated in colon cancer are also methylated in a subset of normal colonic cells during the process of aging. In contrast, methylation of the cancer-specific clones was found exclusively in a subset of colorectal cancers, which appear to display a CpG island methylator phenotype (CIMP). CIMP+ tumors also have a high incidence of p16 and THBS1 methylation, and they include the majority of sporadic colorectal cancers with microsatellite instability related to hMLH1 methylation. We thus define a pathway in colorectal cancer that appears to be responsible for the majority of sporadic tumors with mismatch repair deficiency.

Full text

Proc. Natl. Acad. Sci. USA
Vol. 96, pp. 8681–8686, July 1999
Medical Sciences
CpG island methylator phenotype in colorectal cancer
MINORU TOYOTA,NITA AHUJA,MUTSUMI OHE-TOYOTA,JAMES G. HERMAN,STEPHEN B. BAYLIN,
AND JEAN-PIERRE J. ISSA*
The Johns Hopkins Oncology Center, 424 North Bond Street, Baltimore, MD 21231
Communicated by John W. Littlefield, Johns Hopkins University School of Medicine, Baltimore, MD, May 26, 1999 (received for review
February 19, 1999)
ABSTRACT Aberrant methylation of promoter region
CpG islands is associated with transcriptional inactivation of
tumor-suppressor genes in neoplasia. To understand global
patterns of CpG island methylation in colorectal cancer, we
have used a recently developed technique called methylated
CpG island amplification to examine 30 newly cloned differ-
entially methylated DNA sequences. Of these 30 clones, 19
(63%) were progressively methylated in an age-dependent
manner in normal colon, 7 (23%) were methylated in a
cancer-specific manner, and 4 (13%) were methylated only in
cell lines. Thus, a majority of CpG islands methylated in colon
cancer are also methylated in a subset of normal colonic cells
during the process of aging. In contrast, methylation of the
cancer-specific clones was found exclusively in a subset of
colorectal cancers, which appear to display a CpG island
methylator phenotype (CIMP). CIMPⴙ tumors also have a
high incidence of p16 and THBS1 methylation, and they
include the majority of sporadic colorectal cancers with
microsatellite instability related to hMLH1 methylation. We
thus define a pathway in colorectal cancer that appears to be
responsible for the majority of sporadic tumors with mis-
match repair deficiency.
In the development of colorectal cancers (CRCs), a series of
tumor-suppressor genes such as APC, p53, and genes on
chromosome 18q (DCC, SMAD2, and DPC4兾SMAD4) are
inactivated by mutations and chromosomal deletions (1, 2). A
subset of CRCs also show a characteristic mutator phenotype
which causes microsatellite instability (MSI) and mutations at
other gene loci such as TGF
␤
RII (3) and BAX (4). This
phenotype usually results from inactivation of mismatch repair
(MMR) genes such as hMSH2 and hMLH1 (5). Another
molecular defect described in CRC is CpG island (CGI)
methylation. CGIs are short sequences rich in the CpG dinu-
cleotide and can be found in the 5⬘ region of about half of all
human genes (6). Methylation of cytosine within 5⬘ CGIs is
associated with loss of gene expression and has been seen in
physiological conditions such as X chromosome inactivation
and genomic imprinting (7). Aberrant methylation of CGIs has
been detected in genetic diseases such as the fragile-X syn-
drome (8), in aging cells (9), and in neoplasia (10, 11). About
half of the tumor-suppressor genes that have been shown to be
mutated in the germ line of patients with familial cancer
syndromes have also been shown to be aberrantly methylated
in some proportion of sporadic cancers, including Rb, VHL,
p16, hMLH1, and BRCA1 (10, 11). Tumor-suppressor gene
methylation in cancer is usually associated with (i) lack of gene
transcription and (ii) absence of coding region mutation. Thus
it has been proposed that CGI methylation serves as an
alternative mechanism of gene inactivation in cancer.
The causes and global patterns of CGI methylation in human
cancers remain poorly defined. We have previously reported
that aging could play a factor in this process because methyl-
ation of several CGIs could be detected in an age-related
manner in normal colon mucosa as well as in CRC (9). In
addition, aberrant methylation of CGIs has been associated
with the MSI phenotype in CRC (12) as well as specific
carcinogen exposures (13). However, an understanding of
aberrant methylation in CRC has been somewhat limited by
the small number of CGIs analyzed to date. We have used a
recently developed PCR-based methylation screening tech-
nique, methylated CpG island amplification (MCA), to deter-
mine the methylation status of multiple CGIs in a relatively
large number of samples. In this report, we analyzed the
methylation status of 30 new loci and 3 known tumor-
suppressor genes in a panel of 50 primary CRCs and 15 colonic
adenomas. We find that (i) the majority of CGI methylation
events in CRCs are related to incremental hypermethylation in
normal colon as an age-related phenomenon; and (ii) virtually
all the other methylation events occur in a distinct subset of
CRCs and adenomas which appear therefore to have a new
phenotype, which we termed CpG island methylator pheno-
type (CIMP). These data shed additional light on the global
patterns of CGI methylation in human cancer and delineate a
distinct pathway involving tumor-suppressor gene hypermeth-
ylation in the evolution of CRC.
MATERIALS AND METHODS
Samples and Cell Lines. Fifty CRCs and 15 colorectal
adenoma samples were obtained from The Johns Hopkins
Hospital. All patients gave informed consent prior to specimen
collection according to institutional guidelines. CRCs used in
this study were characterized previously for MSI status (12).
Presence or absence of MSI was determined according to strict
criteria, requiring band shifts at both dinucleotide and mono-
nucleotide tracts. Of these, 43 cancers were randomly selected
and 7 cancers had previously been shown to display MSI. Nine
of 43 (21%) randomly selected samples showed MSI. There-
fore, in total, 16 samples were MSI positive and 34 samples
were MSI negative.
Genes Examined. In this study, we examined 30 of 33
differentially methylated CpG islands cloned by MCA as
described (MINT1–33; ref. 14). The remaining three clones
were not examined because of high background or size too
small. In addition, we examined the methylation status of p16,
THBS1, and hMLH1 by MCA (p16) or bisulfite-PCR (THBS1
and hMLH1).
Detection of Aberrant Methylation by MCA. MCA was
performed essentially as described (14). A detailed protocol is
available at http:兾兾www.med.jhu.edu兾methylation兾MCA.
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore be hereby marked ‘‘advertisement’’ in
accordance with 18 U.S.C. §1734 solely to indicate this fact.
PNAS is available online at www.pnas.org.
Abbreviations: CRC, colorectal cancer; MSI, microsatellite instability;
CGI, CpG island; MCA, methylated CpG island amplification; CIMP,
CpG island methylator phenotype.
*To whom reprint requests should be addressed. e-mail: jpissa@jhmi.
edu.
8681

html. Presence or absence of methylation in cancers was
determined by comparing the signals in the tumor vs. normal
lanes. Each sample was blotted in duplicate. Each filter
included mixtures of positive control (Caco2) and a negative
control (normal colon mucosa from an 18-year-old individual).
All 30 MINT clones, as well as p16 (15) were examined by
MCA.
Bisulfite-PCR. Bisulfite treatment and PCR reactions were
performed as described previously (16, 17). A detailed proto-
col is available at http:兾兾www.med.jhu.edu兾methylation兾
protocols.html. In total, 12 MINT clones (MINT 1, 2, 4, 6, 7,
11, 12, 23, 25, 29, 31, 32), as well as THBS1, and hMLH1 were
examined for methylation status in this way. Primer sequences,
conditions for PCR, and restriction enzymes used are available
at http:兾兾 www.med.jhu.edu兾methylation兾primers.htm.
Southern Blot Analysis. Five micrograms of DNA was
digested with 20–100 units of restriction enzymes as specified
by the manufacturer (New England Biolabs). DNA fragments
were separated by agarose gel electrophoresis and transferred
to a nylon membrane (Zeta-probe, Bio-Rad). Filters were
hybridized with
32
P-labeled probes and washed at 65°C with 2⫻
SSC兾0.1% SDS for 10 min twice, and 0.1⫻ SSC兾0.1% SDS for
20 min. Filters were then exposed to a phosphor screen for
24–72 hr and analyzed by using a PhosphorImager (Molecular
Dynamics).
RESULTS
Methylation Analysis of Multiple CpG Islands by Using
MCA and Bisulfite-PCR. We have recently developed a PCR-
based method termed MCA (14). In MCA, the methylation
status of CRC and normal colon mucosa samples can be
determined by the presence or absence of a hybridization
signal. Using MCA coupled with representational difference
analysis, we have isolated 33 sequences (MINT1–33) differ-
entially methylated in Caco2, a CRC cell line, and have
confirmed the MCA results by Southern blot analysis. These
included 29 CGIs, and several fragments were identical to
known gene sequences. Thus, they appear to be fairly repre-
sentative of hypermethylation events in cancer. The cloning of
a large number of CGIs from a CRC cell line allowed us to
study the global patterns of hypermethylation in this neoplasm.
To determine the methylation status of these clones in
primary tissues, we have used MCA for 30 of the 33 clones (3
clones could not be accurately studied because of high back-
ground (MINT19 and MINT29) or small size (MINT33). As
shown in Fig. 1, MCA provides semiquantitative methylation
data and yields results that are essentially identical to Southern
blot analysis or bisulfite-PCR. In 12 cases (MINT1, 2, 4, 6, 7,
11, 12, 23, 25, 29, 31, 32), MCA results were confirmed by
bisulfite-PCR for all samples, and the two techniques were
concordant in 98% of the cases (examples in Figs. 1 and 2).
Of the 30 clones analyzed, 26 (87%) were also found to be
methylated in some primary CRCs (examples in Figs. 1–3).
The four clones methylated only in the cell line Caco2 were (i)
MINT14, a LINE element; (ii) MINT14 and MINT18, se-
quences that had a very low CpG frequency and did not qualify
as CpG Islands; and (iii) MINT16 (intron 1 of
␣
-Tubulin). The
other 26 clones all qualified as CpG islands on the basis of the
criteria ⬎200 bp, G⫹C content ⬎50%, and CpG兾GpC ⬎ 0.6
(18). Thus, almost all (26 of 27) nonrepetitive CpG islands
recovered from the Caco2 cell line were also methylated in
some primary tumors.
Two Types of Methylation in CRC. By examining the
methylation status of several known genes in colorectal tu-
mors, we have previously demonstrated that some genes tend
to be methylated in an age-dependent manner in normal colon
(9). In the present study of 26 CpG islands, we find that
hypermethylation patterns in CRC fell into two distinct cate-
gories. A majority of the clones (19兾26, 73%) were found to be
frequently methylated (average 75%, ranging from 30% to
100% of the tumors) in the tumors tested, and a slight amount
of methylation was also detected in normal colon mucosa (Fig.
2A). For all of these clones, the normal colon mucosa obtained
from young patients showed less methylation compared with
the normal mucosa from older patients (Fig. 2 B–D). For this
age-related methylation, identical results were obtained by
MCA, Southern blot analysis, and bisulfite-PCR analysis,
suggesting that it is not related to the technique used to study
methylation. Thus, the majority of CGIs hypermethylated in
CRC are methylated in normal colon mucosa as well, in an
age-related manner. We propose to name this methylation
type A for aging-specific methylation.
The remaining 7 clones were methylated exclusively in CRC:
no methylation was observed in normal colon by any technique
(see Figs. 1 and 2), and their frequency of methylation was
significantly lower than type A methylation (ranging from 10%
to 50%, see below). We propose to name this type of meth-
ylation type C for cancer-specific. There was no significant
FIG. 1. Comparison of detection of aberrant methylation by using
Southern blotting, MCA, or bisulfite-PCR analysis. (A) Methylation of
MINT2 detected by Southern blot analysis. Genomic DNA from
normal colon mucosa of an 18-year-old person and the CRC cell line
Caco2 was digested with restriction endonucleases (H, HindIII; HS,
HindIII ⫹ SmaI; HX, HindIII ⫹ XmaI), electrophoresed, blotted, and
hybridized with MINT2 probe. Caco2 DNA fails to digest completely
with SmaI, indicating methylation of one or both SmaI sites, but cuts
down with XmaI, ruling out polymorphisms at the SmaI sites. (B)
Semiquantitative feature of MCA. DNAs from the Caco2 cell line and
normal colon were mixed in various proportions before MCA, and
methylation of MINT2 was analyzed by dot-blot hybridization (Upper).
% refers to the relative proportion of Caco2 DNA. The signal intensity
of each dot was determined by PhosphorImager densitometry, and it
increased linearly relative to the proportion of tumor cells in the mix.
(C) Detection of MINT2 methylation by MCA. We blotted 0.1
␮
gof
MCA products from colorectal tumors and corresponding normal
colon mucosa in duplicate onto a nylon filter and hybridized them with
a MINT2 probe. Tumors 391, 467, 874, 709, and 351 are methylated
at this site. Sample numbers are shown above each lane. N, normal
colon; T, colon tumor. (D) Detection of methylation by bisulfite-PCR.
Genomic DNA was treated with bisulfite and amplified with primers
specific to MINT2. Twenty percent of the PCR products were digested
with BstUI and electrophoresed in 3% agarose gels. Arrows indicate
bands reflecting methylation of the BstUI site. BstUI cleaves only the
methylated alleles, yielding 115- and 88-bp bands (seen in a CRC cell
line, RKO, and tumors 391, 467, 874, and 709). The coexistence of
methylated and unmethylated bands reflects the fact that all these
tumors are not microdissected and contain various amounts of con-
taminating normal tissues. Sample numbers are shown above. N,
normal colon; T, colon tumor.
8682 Medical Sciences: Toyota et al. Proc. Natl. Acad. Sci. USA 96 (1999)

difference in the G⫹C content between type A and type
C clones (average, 57.1% for type A vs. 58.3% for type C,
P ⫽ 0.36).
CIMP in CRC. We next analyzed in detail the methylation
status of the type A and type C MINT clones in a panel of 50
primary colorectal cancers and 15 adenomatous polyps. The
data obtained by MCA were verified by bisulfite-PCR for 12
clones, and the results obtained with the two techniques were
98% concordant. As mentioned above, all 19 type A clones
were frequently methylated in cancers (average 75%, ranging
from 30% to 100% of the tumors, see examples in Figs. 2 A and
3A and summary in Fig. 4). These frequencies are similar to
what we had previously observed for ER (9), MyoD, and N33
(19). When we considered the methylation status of the 7 type
C clones however, a remarkable pattern emerged (examples in
Figs. 1, 2A, and 3B, summarized in Fig. 4). The 50 cancers fell
into two distinct groups: (i) a group with a high level of type
C methylation, whereby all the tumors had methylation of 3 or
more loci simultaneously (average 5.1 loci per tumor) and (ii)
a group where methylation of any type C clone is extremely
rare (average 0.3 locus per tumor). In sharp contrast, type A
methylation was not significantly different between these two
groups of tumors (Figs. 2A,3A, and 4). Thus, the first group
of tumors appears to display a novel process, which we propose
to call CpG island methylator phenotype (CIMP). These
tumors would then be prone to transcriptional silencing linked
to promoter methylation, and they would have the potential to
inactivate multiple tumor-suppressor genes simultaneously.
We next studied the relationship of CIMP to known clini-
copathological factors in the progression of colorectal neo-
plasia. In this limited series, CIMP did not appear to correlate
with age, gender, or stage of the CRC. However a majority of
proximal (cecum, ascending colon) cancers were affected by
this pathway (18兾22, 82%), which was significantly different
from distal (descending, recto-sigmoid) cancers (10兾27, 37%;
P ⫽ 0.003, Fisher’s exact t test). Furthermore, CIMP was also
found in preneoplastic adenomatous polyps. Overall, 15 ade-
nomas were studied, and 7兾15 were CIMP⫹. In 6 cases, both
FIG. 2. Methylation of the MINT clones in CRC and mucosa. (A) Examples of MINT clone methylation in colorectal cancer. MCA products
from colon tumors (T) and corresponding normal colon mucosa (N) were blotted on a nylon membrane and hybridized with one of the MINT clones
(indicated on the left), as well as a p16 probe. MINT6, 24, and 32 are examples of type A methylation, whereas MINT2, 12, and p16 are examples
of type C methylation. Tumors 850 and 874 are CIMP⫹, whereas tumors 835, 872, and 733 are CIMP⫺ (see text). Also shown is the Caco2 cell
line (right). (B) Examples of age-related methylation in normal colon as detected by MCA using MINT5, 8, and 21 as probes. MCA products from
normal colon mucosa from patients of various ages (indicated at the top in years) were hybridized with the probe indicated on the left. The signal
intensity for each sample was determined by densitometry, and a ratio of mucosa兾Caco2 (right) is indicated below each sample as a percentage.
In each case, methylation is more prominent in DNA from older individuals. (C) Southern blot analysis of clones showing type A methylation in
normal colon. DNA from normal colon mucosa from patients of various ages (top) was digested with HindIII and the methylation-sensitive enzyme
SmaI and probed with MINT 5 (Left) and MINT22 (Right). Arrows indicate the methylated alleles. The relative proportion of methylated alleles
was determined by densitometry and is indicated below each lane. (D) Bisulfite-PCR analysis of clones showing type A and type C methylation
in normal colon. DNA from normal colon mucosa from patients of various ages (top) was bisulfite treated, amplified by PCR, and digested with
restriction enzymes specific for sites that are created after bisulfite treatment (if the CpG sites are methylated). The arrows indicate the methylated
alleles. Percentage methylation was determined by densitometry and is indicated below each lane. In type A loci, the percentage of methylated
alleles from older people (⬎70) was significantly higher than that of younger people (⬍30) (6.4 ⫾ 1.1 vs. 22.9 ⫾ 1.2, P ⬍ 0.000001 for MINT23;
1.3 ⫾ 0.3 vs. 16.1 ⫾ 4.0, P ⬍ 0.01 for MINT32). In type C loci no methylation was detected in normal colon mucosa. The loci analyzed and restriction
enzymes used are indicated below each gel.
Medical Sciences: Toyota et al. Proc. Natl. Acad. Sci. USA 96 (1999) 8683

an adenoma and a cancer from the same patient were exam-
ined. In 1 of these, CIMP was detected in both the adenoma
and the cancer; in 3 cases, CIMP was detected in the cancer but
not in the adenoma; in 2 cases, CIMP was detected in neither
the adenoma nor the cancer. These results suggested that
CIMP is an early event in colorectal carcinogenesis.
To determine whether CIMP indeed affects the methylation
status of known tumor-suppressor genes important to tumor
progression, we studied the p16 gene (20), which is one of the
most frequently altered genes in human neoplasms, and the
THBS1 gene, which encodes for an angiogenesis-inhibitor with
tumor-suppressor properties (21, 22). p16 was studied by both
MCA and Southern blot analysis, and THBS1 was studied by
Southern blotting and bisulfite-PCR. There was an excellent
concordance between methylation of both genes and the
presence of CIMP: all 20 cancers and 3 adenomatous polyps
methylated at the p16 CpG island had been classified as
CIMP⫹ by using our type C MINT clones (Fig. 4). Similarly,
all 9 tumors hypermethylated at THBS1 also belonged to the
CIMP⫹ group (Fig. 4). Thus, CIMP is not limited to the clones
recovered by MCA兾RDA, but truly reflects a methylator
phenotype in these tumors.
Microsatellite Instability Is Linked to CIMP in CRC. In a
previous study (12), we reported a link between microsatellite
instability and a hypermethylator phenotype in sporadic CRC.
Relatively few mutations in mismatch repair genes have been
reported in sporadic MSI⫹ cancers, but hMLH1 methylation
has recently been observed in some cases (23–25) and was
suggested to be a primary cause of the MSI phenotype in
sporadic cancers. To determine the relation between CIMP
and MSI in colorectal cancer, we measured hMLH1 methyl-
ation by using bisulfite兾PCR in our panel of CRC, which had
also been previously typed for the presence of MSI (Fig. 3).
hMLH1 was studied by bisulfite-PCR only because it does not
have 2 SmaI sites in its CGI. Overall, 16 of 50 (32%) cancers
had evidence of MSI. Among the 29 CIMP⫹ cases, 12 had
evidence of hMLH1 methylation, suggesting that hMLH1 is
one of the targets of global hypermethylation in CRC. All of
these 12 tumors had MSI. By contrast, hMLH1 methylation
was detected in only one of the 21 CIMP⫺ cases. These data
establish a strong link between the CIMP phenotype, hMLH1
methylation, and MSI in CRC. Two lines of evidence suggest
that MSI may follow, and be caused by, CIMP and hMLH1
methylation. First, CIMP is detectable in about half of colonic
adenomas, but none of these tumors have hMLH1 methylation,
and MSI is rare in this preneoplastic lesion (26, 27). Second,
CIMP is not simply caused by mismatch repair defects because,
MSI is absent in more than half of the CIMP⫹ cases, and
CIMP was absent in 4 of the 16 cancers with MSI. Overall,
these data suggest that, in sporadic CRC, the majority (12兾16,
75%) of cases with MSI may be caused by CIMP followed by
hMLH1 methylation, loss of hMLH1 expression, and resultant
mismatch repair deficiency (24). Interestingly, mismatch repair
deficiency in colorectal cancer is also clustered in proximal
tumors, similar to CIMP.
DISCUSSION
Recently, several reports have suggested that aberrant meth-
ylation of CGIs may play an important role in cancer devel-
opment (10, 11). However, there is little integrated informa-
tion on aberrant CGI methylation in cancer at multiple loci,
probably because of the lack of a method to detect methylation
in a large number of samples for unselected CGIs throughout
the genome. Furthermore, it has been shown that cultured cell
lines have a high degree of CGI methylation (28) but it was not
known to what extent this reflects methylation in primary
cancers. In the present study, we have been able to address
some of these issues by studying 30 differentially methylated
loci in a panel of colorectal cancers, a study facilitated by the
relatively quantitative and high-throughput features of MCA.
Despite the fact that all sequences were initially recovered
from a colon cancer cell line, 26兾27 nonrepeated CpG islands
proved to be methylated in some primary colon cancers.
Therefore, there appears to be relatively little ‘‘artifactual’’
methylation in cell lines. Nevertheless, cell lines often have
more frequent hypermethylation of selected loci than do
primary tumors, and it appears probable that this reflects
continuous selection in culture.
Analysis of the 26 clones methylated in primary tumors
revealed two distinct types of hypermethylation in cancer (type
A for aging-specific and type C for cancer-specific), which may
have distinct causes, and different roles in cancer development.
Type A methylation was seen in the majority of these clones:
19 of 26 (73%) clones were methylated in an age-related
manner in normal colon, and hypermethylated at high fre-
quency in colorectal cancer, as we have shown for the ER gene
(9) and others (19, 29). These results suggest that a large
number of CGIs in the human genome are incrementally
methylated during the aging process and, for many genes, this
methylation correlates with reduced gene expression as shown
for ER (9) and Versican (14). Although we do not know the
mechanism of type A methylation, it likely results from
physiological processes rather than a genetic alteration be-
cause (i) it is very frequent and affects large numbers of cells;
(ii) it is present in all individuals, not just patients with cancer;
and (iii) this process is gene and tissue specific (19). Because
the methylation status at a given CGI is thought to be related
to positive (methylator) factors (30–33) and negative (protec-
tor) factors (34–37), it is possible that, for some genes, this
balance slightly favors de novo methylation, and that this is
reflected by progressive hypermethylation after repeated cell
divisions.
In contrast to type A methylation, type C methylation is
relatively infrequent in primary colorectal cancer, and is never
observed in normal colon mucosa. Furthermore, detailed
analysis of type C methylation in CRC revealed a striking
pattern, suggesting the presence of a hypermethylator pheno-
type in a subset of these tumors: CIMP⫹ cases are character-
ized by frequent concordant methylation of the type C clones
examined. By contrast, type C methylation is virtually nonex-
istent in tumors without CIMP. This concordance cannot be
due to simple experimental variation or artifacts because (i)
methylation was verified by using separate methods (MCA,
bisulfite-PCR, and Southern blotting); (ii) the concordance
was not limited to the MINT clones, since it also affected the
FIG. 3. Methylation analysis of CpG islands in CRC by using
bisulfite-PCR. Bisulfite-treated DNA from CRC with (Left) or without
(Right) CIMP and paired normal colon mucosa were amplified and
digested with restriction enzymes that cleave only the methylated CpG
sites. The arrows indicate the methylated alleles. The loci analyzed and
restriction enzymes used are indicated below each gel. MINT6 is an
example of type A methylation, whereas hMLH1 exemplifies type C
methylation.
8684 Medical Sciences: Toyota et al. Proc. Natl. Acad. Sci. USA 96 (1999)

p16, THBS1, and hMLH1 genes; and (iii) there was no signif-
icant difference in type A methylation between CIMP⫹ and
CIMP⫺ tumors. Through its ability to silence multiple genes
simultaneously, CIMP would then be functionally equivalent
to genetic instability, resulting in the rapid accumulation of
molecular alterations with a potential to accelerate the neo-
plastic process. Because many genes are potential candidates
for inactivation through promoter methylation (10, 11), CIMP
may have profound pathophysiological consequences in neo-
plasia through inactivation of tumor-suppressor genes, metas-
tasis-suppressor genes, angiogenesis inhibitors, and others. In
fact, our data suggest that CIMP could also result in mismatch
repair deficiency through methylation and inactivation of the
hMLH1 promoter, and it may explain up to 75% of cases of
sporadic CRC with MSI.
The causes of type A and type C methylation are probably
different because the latter is detected only in a limited number
of cases and the genes affected are distinct. This defect could
be either aberrant de novo methylation (through a mutation in
DNA-methyltransferase for example), or loss of protection
against de novo methylation, through the loss of a trans-
activating factor (34, 35). Because DNA-methyltransferase
activity is similar in the two groups (data not shown), we favor
the latter hypothesis. Thus, at least in CRC, it appears likely
that type C methylation (an epigenetic error) is actually caused
by a genetic event that results in an increased chance of
methylating a subset of CGIs. Ironically, this epigenetic defect
may then result in additional genetic lesions through the
induction of mismatch-repair deficiency.
On the basis of these data, we propose the following
speculative model integrating CGI methylation into CRC
development (Fig. 5). In this model, CGI methylation plays
two distinct roles, and it appears to arise through distinct
mechanisms. Initially, type A methylation arises as a function
of age in normal colorectal epithelial cells. By potentially
affecting genes that regulate the growth and兾or differentiation
of these cells, such methylation could account, in part, for the
hyperproliferative state that is thought to precede tumor
formation in the colon (38). Such hyperproliferation is known
to arise with age in colorectal epithelium (39, 40) and to be
marked in patients with colorectal cancer. Furthermore, mod-
ulation of type A methylation may provide one possible
explanation for the reduction in CRC tumorigenesis by reduc-
ing levels of DNA-methyltransferase (41). A second major role
for CGI methylation appears later, possibly at the transition
between small and large adenomas in the colon (on the basis
of 15 adenomas presented here and preliminary data on 50
additional cases). This methylation (type C) affects only a
FIG. 4. Methylation profile of colorectal cancer. The methylation
status of all type C MINT clones, as well as p16, THBS1, and hMLH1,
was determined in a panel of primary CRC and adjacent mucosa. None
of the normal tissues studied showed any degree of methylation for
these genes. Also shown are representative type A clones (Right). Each
column represents a separate gene locus (indicated on top). Each row
is a primary CRC (samples above the bold solid line) or adenomatous
polyp (below the bold solid line). Black rectangles, methylated tumor;
FIG. 5. A model integrating CGI methylation in colorectal carci-
nogenesis. See text for details.
white rectangles, unmethylated tumor. MSI denotes the presence of
microsatellite instability and was not determined for the adenomas. In
the Center are known genes methylated in CRC (p16, THBS1,
hMLH1).
Medical Sciences: Toyota et al. Proc. Natl. Acad. Sci. USA 96 (1999) 8685

subset of tumors, which then evolve along a pathway of global
hypermethylation. We propose then that CIMP leads to cancer
development through the simultaneous inactivation of multi-
ple tumor-suppressor genes such as p16, and induction of
mismatch repair deficiency through inactivation of hMLH1.
This model may be applicable to most human malignancies. In
fact, we have observed evidence for type A and type C
methylation in brain tumors (42), and we have preliminary
evidence suggesting the presence of CIMP in multiple types of
neoplasms, including stomach cancers, brain tumors, and
hematopoietic malignancies. Deciphering the mechanisms un-
derlying these phenomena should facilitate the early detection,
prevention, and therapy of various neoplasms.
We thank Dr. Stanley Hamilton for providing the colon tumor
specimens and Dr. Bert Vogelstein for reviewing the manuscript. This
work was supported by grants from the National Institutes of Health
(National Cancer Institute Grants CA77045 and CA54396 and Colon
Cancer Spore Grant CA62924). M.T. is a Postdoctoral Fellow of the
Japan Society for the Promotion of Science. N.A. is supported by
National Institutes of Health Training Grant 1-T32-DK07713. J.-P.J.I.
is a Kimmel Foundation Scholar.
1. Kinzler, K. W. & Vogelstein, B. (1996) Cell 87, 159–170.
2. White, R. L. (1998) Cell 92, 591–592.
3. Markowitz, S., Wang, J., Myeroff, L., Parsons, R., Sun, L.,
Lutterbaugh, J., Fan, R. S., Zborowska, E., Kinzler, K. W.,
Vogelstein, B., et al. (1994) Science 268, 1336–1338.
4. Rampino, N., Yamamoto, H., Ionov, Y., Li, Y., Sawai, H., Reed,
J. C. & Perucho, M. (1997) Science 275, 967–969.
5. Peltomaki, P. & de la Chapelle, A. (1997) Adv. Cancer Res. 71,
93–119.
6. Bird, A. P. (1986) Nature (London) 321, 209–213.
7. Latham, K. E. (1996) Trends Genet. 112, 134–138.
8. Hansen, R. S., Gartler, S. M., Scott, C. R., Chen, S. H. & Laird,
C. D. (1992) Hum. Mol. Genet. 8, 571–578.
9. Issa, J.-P., Ottaviano, Y. L., Celano, P., Hamilton, S. R., David-
son, N. E. & Baylin, S. B. (1994) Nat. Genet. 7, 536–540.
10. Baylin, S. B., Herman, J. G., Graff, J. R., Vertino, P. M. & Issa,
J.-P. (1998) Adv. Cancer Res. 72, 141–196.
11. Jones, P. A. (1996) Cancer Res. 56, 263–267.
12. Ahuja, N., Mohan, A. L., Li, Q., Stolker, J. M., Herman, J. G.,
Hamilton, S. R., Baylin, S. B. & Issa, J.-P. (1997) Cancer Res. 57,
3370–3374.
13. Issa, J.-P., Baylin, S. B. & Belinsky, S. A. (1996) Cancer Res. 56,
3655–3658.
14. Toyota, M., Ho, C., Ahuja, N., Jair, K.-W., Li, Q., Ohe-Toyota,
M., Baylin, S. B. & Issa, J.-P. J. (1999) Cancer Res. 59, 2307–2312.
15. Herman, J. G., Merlo, A., Mao, L., Lapidus, R. G., Issa, J.-P.,
Davidson, N. E., Sidransky, D. & Baylin, S. B. (1995) Cancer Res.
55, 4525–4530.
16. Sadri, R. & Hornsby, P. J. (1996) Nucleic Acids Res. 24, 558–559.
17. Xiong, Z. & Laird, P. W. (1997) Nucleic Acids Res. 25, 2532–2534.
18. Gardiner-Garden, M. & Frommer, M. (1987) J. Mol. Biol. 196,
261–282.
19. Ahuja, N., Li, Q., Mohan, A. L., Baylin, S. B. & Issa, J.-P. (1998)
Cancer Res. 58, 5489–5494.
20. Kamb, A. (1995) Trends Genet. 11, 136–140.
21. Dameron, K. M., Volpert, O. V., Tainsky, M. A. & Bouck, N.
(1994) Science 265, 1582–1584.
22. Sheibani, N. & Frazier, W. A. (1995) Proc. Natl. Acad. Sci. USA
92, 688–692.
23. Kane, M. F., Loda, M., Gaida, G. M., Lipman, J., Mishra, R.,
Goldman, H., Jessup, J. M. & Kolodner, R. (1996) Cancer Res.
57, 808–811.
24. Herman, J. G., Umar, A., Polyak, K., Graff, J. R., Ahuja, N., Issa,
J.-P. J., Markowitz, S., Willson, J. K., Hamilton, S. R., Kinzler,
K. W., et al. (1998) Proc. Natl. Acad. Sci. USA 95, 6870–6875.
25. Cunningham, J. M., Christensen, E. R., Tester, D. J., Kim, C. Y.,
Roche, P. C., Burgart, L. J. & Thibodeau, S. N. (1998) Cancer Res.
58, 3455–3460.
26. Samowitz, W. S. & Slattery, M. L. (1997) Gastroenterology 112,
1516–1519.
27. Konishi, M., Kikuchi-Yanoshita, R., Tanaka, K., Muraoka, M.,
Onda, A., Okumura, Y., Kishi, N., Iwama, T., Mori, T., Koike, M.,
et al. (1997) Gastroenterology 111, 307–317.
28. Antequera, F., Boyes, J. & Bird, A. (1990) Cell 62, 503–514.
29. Issa, J.-P., Vertino, P. M., Boehm, C. D., Newsham, I. F. & Baylin,
S. B., (1996) Proc. Natl. Acad. Sci. USA 93, 1757–1762.
30. Mummaneni, P., Bishop, P. L. & Turker, M. S. (1993) J. Biol.
Chem. 268, 552–558.
31. Mummaneni, P., Walker, K. A., Bishop, P. L. & Turker, M. S.
(1995) J. Biol. Chem. 270, 788–792.
32. Magewu, A. N. & Jones, P. A. (1994) Mol. Cell. Biol. 14,
4225–4332.
33. Vertino, P. M., Yen, R. W., Gao, J. & Baylin, S. B. (1996) Mol.
Cell. Biol. 16, 4555–4565.
34. Macleod, D., Charlton, J., Mullins, J. & Bird, A. P. (1994) Genes
Dev. 8, 2282–2292.
35. Brandeis, M., Frank, D., Keshet, I., Siegfried, Z., Mendelsohn,
M., Nemes, A., Temper, V., Razin, A. & Cedar, H. (1994) Nature
(London) 371, 435–438.
36. Turker, M. S. & Bestor, T. H. (1997) Mutat. Res. 386, 119–130.
37. Chen, F. Y., Harris, L. C., Remack, J. S. & Brent, T. P. (1997)
Proc. Natl. Acad. Sci. USA 94, 4348–4353.
38. Lipkin, M. (1988) Cancer Res. 48, 235–245.
39. Holt, P. R. & Yeh, K. Y. (1988) Gastroenterology 95, 1556–1563.
40. Roncucci, L., Ponz de Leon, M., Scalmati, A., Malagoli, G.,
Pratissoli, S., Perini, M. & Chahin, N. J. (1988) Cancer 62,
2373–2377.
41. Laird, P. W., Jackson-Grusby, L., Fazeli, A., Dickinson, S. L.,
Jung, W. E., Li, E., Weinberg, R. A. & Jaenisch, R. (1995) Cell
81, 197–205.
42. Li, Q., Jedlicka, A., Ahuja, N., Gibbons, M. C., Baylin, S. B.,
Burger, P. C. & Issa, J.-P. J. (1998) Oncogene 16, 3197–3202.
8686 Medical Sciences: Toyota et al. Proc. Natl. Acad. Sci. USA 96 (1999)