We conducted this study to show a distinct genetic profile of colorectal cancer with CpG island methylator phenotype (CIMP). Using quantitative DNA methylation analysis, we have shown a clear bimodal distribution of 80 MSI‐H tumours according to the number of methylated promoters, and that CIMP is associated with high BRAF and low KRAS mutation rates. In addition, even within MSI‐H and MSS tumours, CIMP tumours have significantly higher BRAF mutation frequencies than non‐CIMP counterparts. Therefore, our data indicate that CIMP is a distinct biological subtype of colorectal cancer, independent of MSI status.
The term CIMP has been used for a subset of colorectal cancers with promoter methylation in multiple genes.10,11,12,13,14,15
As most previous studies showed unimodal distributions of the numbers of methylated loci,11,14,20,21
it has been unclear how many loci should be methylated for a given tumour to be classified as CIMP. It has also been suggested that CIMP is merely within a spectrum of predicted distributions of stochastic methylation events.20,21
However, there were a number of limitations of these previous studies.
Firstly, these previous studies used MSP to determine promoter methylation in multiple genes.10,11,12,13,14,15,20,21
MSP based methylation assays have major limitations in assay quality, and cannot reliably distinguish high from very low levels of methylation with little or no biological significance. We have demonstrated that most colorectal cancers with low levels of promoter methylation in MLH1
(p16), or MGMT
(PMR <4 by MethyLight) show intact protein expression, indicating little or no biological significance of low levels of DNA methylation in these loci.22
Thus studies to determine promoter methylation and CIMP status by MSP should be evaluated with caution.
Secondly, choice of gene promoters to determine CIMP status is important. Aberrant CpG island methylation has been shown to occur in a non‐random fashion and the pattern of methylation is tumour specific.42
Thus a carefully selected and validated panel of methylation markers should be used to determine CIMP status. We evaluated each of the five gene promoters (that is, CACNA1G
, and NEUROG1
) as well as MGMT
for suitability in the CIMP panel. There were previous studies that included MGMT
in CIMP panels.20,21
However, our data indicate that MGMT
show low sensitivity (62%) and specificity (66%) for determination of CIMP and may be better excluded from a CIMP panel. The other five markers, excluding MGMT
, show very high (>90%) sensitivity and/or specificity, and methylation in each of these five markers correlates very well with overall CIMP status (that is, methylation in multiple markers) (see appendix). Therefore, inclusion of these five markers into the CIMP panel can be justified. The fact that MGMT
methylation is not a good predictor of CIMP does not mitigate the biological importance of MGMT
methylation as it has been associated with G>A mutations in KRAS
and may contribute to field effects (field defects) that might lead to colorectal cancer development.43,44
Thirdly, in most previous studies, the numbers of subjects were limited and not population based. The importance of population based studies, particularly prospective cohort studies, cannot be overemphasised to avoid obvious sources of bias. There has been only one large population based retrospective study on over 800 cases15
which revealed unique associations of CIMP with various clinicopathological features by multivariate analyses.15
However, no previous study has been based on large prospective cohorts.
To resolve the issues mentioned above, we measured CpG island methylation in five gene promoters by quantitative real time PCR (MethyLight) assays on 460 colorectal cancer cases in large prospective cohort studies. We have demonstrated a clear bimodal distribution of the number of methylated loci among 80 MSI‐H tumours (fig 1). Most previous investigators arbitrarily defined CIMP as having methylation in
50% markers. Our data indicate that CIMP can be defined as having methylation in
4/5 markers when using a panel of five carefully selected CIMP specific markers. The distribution of 34 MSI‐H BRAF
mutated tumours, all of which showed 4/5 or more methylated loci, further supports this higher cut off for CIMP (table 5).
Previous studies using MSP based assays designated approximately 30–35% of colorectal cancers as CIMP positive.10,11,12,13
However, MSP may detect very low levels of biologically insignificant methylation. Hence the previously reported frequencies of CIMP might be overestimates. Our quantitative DNA methylation data indicate that CIMP is less frequent (only 17%) than previously reported. In fact, when a lower PMR cut off of 1 was used and CIMP was defined as having methylation in
3/6 loci (including MGMT
in the CIMP panel), the frequency of CIMP increased to 32% (
148/460) (detailed data not shown). This CIMP frequency estimate matches with previous reports,11,14,20,21
suggesting that 10% or more tumours might have been misclassified as CIMP in previous studies. Misdiagnosis of CIMP might obscure associations of CIMP with various clinical and pathological features. We recommend that quantitative DNA methylation analysis should be validated and used for research in cancer epigenetics.
CIMP has been previously associated with MSI.11,13,14,15,20
CIMP is more common in sporadic MSI‐H tumours than in MSI‐H tumours in the setting of hereditary non‐polyposis colorectal cancer.45
We have shown a strong association of MSI‐H with CIMP, and striking bimodal distributions of the number of methylated loci among 80 MSI‐H tumours (fig 1), supporting different pathogenetic mechanisms for the two distinct types (CIMP and non‐CIMP) of MSI‐H tumours, with different BRAF
mutation frequencies (63% v
0%) and KRAS
mutation frequencies (1.9% v
31%). Likewise, BRAF
mutations were present in 54% of CIMP MSS tumours in contrast with only 6.6% of non‐CIMP MSS tumours (fig 2), also supporting two distinct types (CIMP and non‐CIMP) of MSS tumours. Thus our data confirm a striking link between BRAF
mutations and CIMP, as in previous studies.16,17,18
All of the 34 BRAF
mutant MSI‐H tumours were CIMP with
4/5 methylated loci (table 5).
Previous studies suggested a link between KRAS
mutation and CIMP.13,15
The link might be due to hypermethylation and silencing of MGMT
, a mismatch repair gene, predisposing to KRAS
However, in other studies, CIMP tumours showed lower KRAS
mutation frequencies than non‐CIMP tumours.11,14,16,17,47
Our data support an inverse association of KRAS
mutations with CIMP (table 4). We also examined whether tumours with 1/5–3/5 methylated loci were associated with KRAS
mutations. We showed that 1/5–3/5 methylated tumours had higher frequencies of KRAS
mutations (48%) and BRAF
mutations (11%) than the 0/5 methylated tumours (35% and 2.9%, respectively; both p<0.02). Further investigation will be necessary to determine whether the 1/5–3/5 methylated tumours should be called “CIMP‐low” (analogous to “MSI‐low”), which is distinct from 0/5 methylated tumours. None the less, our results could explain conflicting data on KRAS
mutation frequencies among CIMP colorectal cancers in previous studies11,13,14,15,16,17,47
where MSP assays and looser criteria for CIMP might have included some “CIMP‐low” cases (with frequent KRAS
mutations) into CIMP tumours (with a low frequency of KRAS
In conclusion, CpG island methylator phenotype (CIMP) of colorectal cancer is best characterised by quantitative DNA methylation analysis and carefully selected markers. CIMP is a distinct epigenetic subtype of colorectal cancer, tightly associated with BRAF mutations and MSI‐H. The frequency of CIMP is likely to be 20% or less, which is lower than previously reported. Quantitative DNA methylation assays should be validated and used for research and clinical practice in cancer epigenetics.