If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
First Department of Internal Medicine, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, JapanDepartment of Molecular Biology, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, Japan
Address reprint requests to Hiromu Suzuki, M.D., Ph.D., Department of Molecular Biology, Sapporo Medical University, South-1, West-17, Chuo-ku, Sapporo 060-8556, Japan
First Department of Internal Medicine, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, JapanDepartment of Molecular Biology, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, Japan
First Department of Internal Medicine, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, JapanDepartment of Molecular Biology, the Research Institute for Frontier Medicine, Sapporo Medical University, Sapporo, Japan
The concept of the CpG island methylator phenotype (CIMP) in colorectal cancer (CRC) is widely accepted, although the timing of its occurrence and its interaction with other genetic defects are not fully understood. Our aim in this study was to unravel the molecular development of CIMP cancers by dissecting their genetic and epigenetic signatures in precancerous and malignant colorectal lesions. We characterized the methylation profile and BRAF/KRAS mutation status in 368 colorectal tissue samples, including precancerous and malignant lesions. In addition, genome-wide copy number aberrations, methylation profiles, and mutations of BRAF, KRAS, TP53, and PIK3CA pathway genes were examined in 84 colorectal lesions. Genome-wide methylation analysis of CpG islands and selected marker genes revealed that CRC precursor lesions are in three methylation subgroups: CIMP-high, CIMP-low, and CIMP-negative. Interestingly, a subset of CIMP-positive malignant lesions exhibited frequent copy number gains on chromosomes 7 and 19 and genetic defects in the AKT/PIK3CA pathway genes. Analysis of mixed lesions containing both precancerous and malignant components revealed that most aberrant methylation is acquired at the precursor stage, whereas copy number aberrations are acquired during the progression from precursor to malignant lesion. Our integrative genomic and epigenetic analysis suggests early onset of CIMP during CRC development and indicates a previously unknown CRC development pathway in which epigenetic instability associates with genomic alterations.
Colorectal cancer (CRC) is a leading cause of cancer mortality worldwide, but the incidence of CRC can be reduced through detection and removal of colorectal adenomas. Notably, however, most small colorectal polyps do not progress to malignancy; thus, the identification of precursor lesions likely to become cancerous is also extremely important for reducing CRC mortality.
CRCs are thought to arise, in part, through the accumulation of genetic changes, including mutations of oncogenes and tumor suppressor genes.
It is also generally accepted that CRCs can exhibit either of two genetic instabilities: chromosomal instability (CIN) or microsatellite instability (MSI).
In addition to these genetic changes, epigenetic alterations, including DNA methylation and histone modification, play critical roles in the development of CRC; in addition, an increasing number of genes involved in cell cycle control, DNA repair, tumor invasiveness, and the response to growth factors have been identified as targets of hypermethylation in CRC.
These epigenetic alterations are thought to be the main driving force in a subset of CRCs exhibiting concurrent hypermethylation of multiple loci, which is termed the CpG island methylator phenotype (CIMP).
CIMP-positive CRCs show characteristic clinicopathological and molecular features, including proximal tumor location, female sex, older age, high tumor grade, wild-type TP53, frequent BRAF and KRAS mutations, and MSI. In addition, several studies support the hypothesis that CRCs can be categorized into three subclasses based on aberrant CpG island methylation: CIMP-high (CIMP-H; also known as CIMP1), CIMP-low (CIMP-L; also known as CIMP2), and CIMP-negative (CIMP-N). CIMP-H CRCs are significantly associated with a BRAF mutation, MLH1 methylation, and subsequent MSI.
but much remains unknown about their respective molecular and clinicopathological features.
Reducing the incidence of cancers, such as CRC, will require a better understanding of the mechanisms underlying carcinogenesis and the molecular alterations occurring in premalignant lesions. For example, although experimental evidence has confirmed the presence of CIMP in CRCs, the role of CIMP in the progression of precancerous lesions toward cancer is not yet fully understood. In recent years, sessile serrated adenomas (SSAs) have been the origin of MSI-positive/CIMP-H cancers, which account for approximately 10% to 15% of sporadic CRCs.
From these and other studies of many tumors, a model was suggested in which CIN and epigenetic instability (CIMP) represent the two major pathways of CRC development, with up to 50% of CRCs being characterized as CIMP.
This means that the origin of a large fraction of CIMP cancers remains unclear.
High-resolution magnifying colonoscopy is a powerful diagnostic tool for detecting premalignant lesions. According to Kudo's classification, the pit patterns of nonneoplastic lesions are classified as type I (normal colon) or type II [hyperplastic polyp (HP)], whereas the pit patterns of neoplastic lesions are classified as types III, IV, and V.
Recently, we performed an integrative analysis of the morphological, pathological, and molecular signatures in colorectal precancerous lesions and identified a novel pit pattern (type II, open pits) that was specific to SSAs.
Those results depict an important relationship between morphological characteristics and molecular alterations that will significantly improve our ability to detect premalignant lesions. In the present study, our aim was to uncover the molecular evolution of CIMP cancers through an integrative analysis of many precursor and malignant colorectal lesions. Based on their genetic and epigenetic signatures, we propose a model in which CRCs develop via four distinct pathways. We also provide evidence that CIMP and CIN are not completely mutually exclusive, so that chromosomal aberrations may play important roles in a subset of CIMP cancers. These findings will improve our understanding of the pathogenesis of CRCs and could potentially contribute to better clinical management of premalignant lesions.
Materials and Methods
Study Population and Tissue Specimens
Colorectal tumor tissues were collected from Japanese patients who underwent endoscopic or surgical resection of a colorectal tumor at Akita Red Cross Hospital (Akita, Japan). A total of 368 specimens from 192 precursor lesions, 38 noninvasive carcinomas [carcinoma in situ (CIS)], 100 CRCs, and 38 samples of adjacent normal tissue were analyzed in this study. Informed consent was obtained from all patients before collection of the specimens. Approval of this study was obtained from the Institutional Review Board of Akita Red Cross Hospital and Sapporo Medical University (Sapporo, Japan). By using the standard phenol/chloroform procedure, genomic DNA was extracted from biopsy specimens obtained before endoscopic or surgical resection. CRC cell lines were maintained and treated with 5-aza-2′-deoxycytidine, as previously described.
High-resolution magnifying endoscopes (CF260AZI; Olympus, Tokyo, Japan) were used for all colonoscopic analyses. Colorectal subsite locations were defined as right side colon proximal to the splenic flexure (cecum, ascending colon, hepatic flexure, and transverse colon), left side colon distal to the splenic flexure (splenic flexure, descending colon, and sigmoid colon), and rectum (rectosigmoid and rectum). All detected colorectal tumors were observed at high magnification after staining with indigo carmine dye and 0.05% crystal violet. Surface microstructures were classified according to Kudo's pit pattern classification system.
Most often, one biopsy specimen was collected from each lesion for the extraction of genomic DNA. However, when two or more pit patterns were found in a single lesion (eg, adenoma to carcinoma transition), biopsy specimens were obtained for each respective pit pattern (see Supplemental Figure S1 at http://ajp.amjpathol.org). Thereafter, the lesions underwent endoscopic mucosal resection, endoscopic submucosal dissection, or surgical resection, after which histological analyses were performed (see Supplemental Figure S1 at http://ajp.amjpathol.org). Conventional adenomas, such as tubular adenoma and tubulovillous adenoma, were diagnosed using standard criteria. Serrated lesions, including HP, traditional serrated adenoma (TSA), and SSA, were classified based on criteria previously described by Torlakovic et al.
Serrated lesions that did not satisfy the criteria for SSA or TSA were defined as HP. Tumors were classified into three categories: precursor lesions (HP, tubular adenoma, tubulovillous adenoma, TSA, and SSA), CIS, and CRCs.
MCAM Data
Methylated CpG island amplification microarray (MCAM) analysis was performed as previously described.
A BioPrime Plus Array CGH Genomic Labeling System (Life Technologies, Carlsbad, CA) was used to label MCA amplicons from tumor samples with Alexa Fluor 647 (Life Technologies, Carlsbad, CA), and those from a pooled mixture of normal colonic tissue were labeled with Alexa Fluor 555 (Life Technologies). Labeled MCA amplicons were then hybridized to a custom human CpG island microarray (G4497A; Agilent Technologies, Santa Clara, CA), which included 15,134 probes covering 6157 unique genes. After washing, the array was scanned using an Agilent DNA Microarray Scanner (Agilent Technologies), and the data were processed using Feature Extraction software version 10.7 (Agilent Technologies), and analyzed using GeneSpring GX version 11 (Agilent Technologies). Unsupervised hierarchical clustering and statistical analyses were then performed.
Methylation Analysis by Bisulfite Pyrosequencing
Bisulfite pyrosequencing was performed as previously described.
Briefly, genomic DNA (1 μg) was modified with sodium bisulfite using an EpiTect Bisulfite Kit (Qiagen, Hilden, Germany). Pyrosequencing was then performed using a PSQ96 system with a PyroGold reagent Kit (Qiagen), after which the results were analyzed using Q-CpG software version 1.0.9 (Qiagen). Unsupervised hierarchical clustering, principal component analysis, and correspondence analysis of validation data were performed using JMP version 8 (SAS Institute Inc., Cary, NC). For the statistical analysis, quantitative methylation levels of each gene were converted to z scores, which were defined as follows: (Methylation Level in Each Sample - Mean Methylation Level in All Samples)/(SD of Methylation Levels in All Samples). Primer sequences are listed in Table 1.
Mutations in codon 600 of BRAF and codons 12 and 13 of KRAS were examined by pyrosequencing using BRAF and KRAS pyrokits (Qiagen), respectively, according to the manufacturer's instructions. Mutation of PIK3CA, AKT2, and PDK1 was analyzed by direct sequencing, as previously described.
Mutation of TP53 was initially detected by PCR–single-stranded conformational polymorphism analysis, followed by direct sequencing, as previously described.
A National Cancer Institute Workshop on Microsatellite Instability for cancer detection and familial predisposition: development of international criteria for the determination of microsatellite instability in colorectal cancer.
MSI was defined by the presence, in the tumor sample, of bands that were abnormal in size, compared with a corresponding normal sample. A tumor sample was defined as MSI positive when two or more markers showed instability.
Array-Based Comparative Genomic Hybridization
Array-based comparative genome hybridization (array CGH) analysis was performed as previously described.
Briefly, 500 ng of genomic DNA from colorectal tumor specimens and sex-matched reference DNA from noncancerous colonic mucosa were labeled with Cy5 and Cy3, respectively, using a Genomic DNA Enzymatic Labeling Kit (Agilent Technologies), and were then hybridized to Human Genome CGH Microarray Kit 180A (G4449A; Agilent Technologies). DNA copy number aberrations (CNAs) were identified using the ADM-2 algorithm included in the Agilent Genomic Workbench software version 5 (Agilent Technologies). A copy number loss was defined as a log 2 ratio of <−0.5, and a copy number gain was defined as a log 2 ratio of >0.5. Unsupervised hierarchical clustering of CNAs was performed using JMP, version 8. The microarray data in this study have been submitted to the Gene Expression Omnibus, and the accession number is GSE35534.
Statistical Analysis
To compare differences in continuous variables between groups, t-tests or an analysis of variance with a post hoc Tukey's test was performed. A Fisher's exact test or a χ2 test was used for analysis of categorical data. P < 0.05 (two sided) was considered statistically significant. Statistical analyses were performed using JMP, version 8, and SPSS statistics 18 (IBM Corporation, Somers, NY).
Results
Three Methylation Subclasses in Precancerous and Malignant Colorectal Tumors
To clarify the epigenetic changes occurring early during colorectal tumorigenesis, we first performed global methylation analysis using MCAM in a series of normal colonic tissues (n = 16), precursor lesions (HP, n = 3; SSA, n = 5; TSA, n = 3; tubular adenoma, n = 6; tubulovillous adenoma, n = 10), CISs (n = 14), and CRCs (n = 28). Hierarchical clustering analysis using the MCAM results identified 1010 probe sets that detected frequent hypermethylation in tumor tissues (see Supplemental Figure S2 at http://ajp.amjpathol.org). Subsequent K-means clustering analysis using these probe sets revealed that the samples could be clearly categorized into three subclasses based on the level of their methylation (Figure 1A); subclasses with high and intermediate methylation were presumed to reflect CIMP-H and CIMP-L tumors, respectively. Among the 1010 probe sets, 538 unique genes were in the high-methylation group, whereas 259 genes were in the intermediate-methylation group (see Supplemental Figure S2C and Supplemental Tables S1 and S2 at http://ajp.amjpathol.org). A subset of the precursor lesions could also be categorized into these subclasses, indicating that CIMP-like methylation patterns were already established early during carcinogenesis.
Figure 1Genome-wide methylation analysis of CpG islands and selected marker genes reveals three distinct methylation classes in precursor lesions and CRCs. A: MCAM results were obtained from 28 precursor lesions, 14 CISs, 28 CRCs, and 16 samples of normal colonic tissue. A set of 1010 probes showing tumor-specific methylation were selected, after which K-means clustering analysis was performed (K = 3). B: Unsupervised hierarchical clustering of the methylation data for 25 markers in a set of 368 specimens consisting of 192 precursor lesions, 38 CISs, 100 CRCs, and 38 samples of normal colonic tissue. Pathological diagnosis and mutation of KRAS and BRAF are shown (top panel). C: Frequencies of KRAS and BRAF mutations in precursor (left panel) and advanced (right panel) lesions with the indicated CIMP status. The number of the samples in each portion is also shown. D: Principle component analysis of the methylation of 25 genes in precursor (left panel) and advanced (right panel) lesions. Two-dimensional scatter plots using the variances of the first and second components are shown. Each dot represents a single lesion; and red, green, and blue dots represent CIMP-H, CIMP-L, and CIMP-N lesions, respectively.
To further characterize the genes that acquired methylation progressively during CIMP pathway tumorigenesis, we next performed MCAM analysis with a series of precursor lesions in which CIMP-N flat components were present, along with CIMP-positive protruding components within the same lesions (see Supplemental Figure S2D at http://ajp.amjpathol.org). Because both components were presumed to arise from the same origin, these lesions could represent an ideal model for analyzing the molecular progression of CIMP cancers. CIMP status was defined using classic CIMP markers (MINT1, MINT2, MINT12, MINT31, and CDKN2A), and tumors with methylation of three or more markers were defined as CIMP. When we analyzed three pairs of precursor lesions using MCAM, we identified 36 unique genes that were differentially methylated between CIMP-positive and CIMP-N precursor lesions (see Supplemental Table S3 at http://ajp.amjpathol.org). These genes were potentially the earliest targets of aberrant methylation during CIMP pathway tumorigenesis, and most of them overlapped with the genes identified in the initial MCAM analysis (see Supplemental Figure S2E at http://ajp.amjpathol.org).
Methylation Profiling Identified CIMP in Precancerous Lesions
Based on the results previously summarized, we selected a series of marker genes to characterize the methylation profile of precursor and malignant lesions. We initially selected 14 genes (LRP1B, CDKN2A, WNT5A, MEOX2, ZNF569, GALNT14, SOX5, DFNA5, DLX4, SFRP2, WIF1, FZD10, KCNV1, and IGF2BP1) identified in the MCAM analysis (see Supplemental Figure S2 at http://ajp.amjpathol.org). Among them, IGF2BP1, KCNV1, DLX4, GALNT14, and ZNF569 were not previously reported to be methylated in CRCs, but RT-PCR analysis using multiple CRC cell lines confirmed that they were frequent targets of epigenetic silencing in CRC (see Supplemental Figure S3 at http://ajp.amjpathol.org). In addition, we selected 11 well-characterized markers (MLH1, SFRP1, IGFBP7, DKK2, MIR34B, MINT1, MINT2, MINT12, MINT31, RASSF2, and RASSF5) used for methylation analysis.
We performed quantitative bisulfite pyrosequencing of the 25 markers in a total of 330 specimens consisting of 192 precursor lesions, 38 CISs, and 100 CRCs (Table 2). Consistent with the MCAM results, unsupervised hierarchical clustering using the pyrosequencing data revealed that, in addition to malignant lesions (CISs and CRCs), precursor lesions could also be divided into three relative subclasses (M-clusters 1 to 3; Figure 1B). The BRAF mutation was significantly enriched in M-cluster 1 tumors, in which most of the genes were highly methylated, suggesting that this subclass corresponded to CIMP-H (Figure 1B). M-cluster 2 included tumors with intermediate levels of methylation and a prevalent KRAS mutation, which corresponded to CIMP-L, whereas M-cluster 3 tumors exhibited the lowest methylation levels, which corresponded to CIMP-N. Among the precursor lesions, 37 (19.3%) exhibited a BRAF mutation, and most of those [24 (64.9%) of 37] were categorized as CIMP-H (Figure 1C). In CISs and CRCs, the KRAS mutation was most enriched in the CIMP-L group, but this tendency was less apparent among the precursor lesions (Figure 1C).
Table 2Clinicopathological Features of the Colorectal Tumors Used in this Study
Feature
Total (N = 192)
CIMP-H (n = 42)
CIMP-L (n = 46)
CIMP-N (n = 104)
P value
Precursor Lesion
Age (years)
69.61 ± 9.68
72.14 ± 8.01
72.43 ± 9.82
67.34 ± 9.73
<0.05
Sex
F
73 (38.02)
21 (50)
26 (56.52)
26 (25)
<0.001
M
119 (61.98)
21 (50)
20 (43.48)
78 (75)
Location
Right
92 (47.92)
32 (76.19)
27 (58.7)
33 (31.73)
<0.001
Left
40 (20.83)
5 (11.9)
7 (15.22)
28 (26.92)
Rectum
60 (31.25)
5 (11.9)
12 (26.09)
43 (41.35)
Size (mm)
12.98 ± 9.61
13.73 ± 8.44
16.67 ± 12.55
8.38 ± 6.48
<0.001
Morphological characteristics
Protruding type
82 (42.71)
17 (40.48)
21 (45.65)
44 (42.31)
Flat type
110 (57.29)
25 (59.52)
25 (54.35)
60 (57.69)
Depressed type
0
0
0
0
Histological features
HP
28 (14.58)
2 (4.76)
5 (10.87)
21 (20.19)
<0.001
SSA
29 (15.1)
26 (61.90)
1 (2.17)
2 (1.92)
TSA
25 (13.02)
6 (14.29)
5 (10.87)
14 (13.46)
Tubular adenoma
53 (17.6)
2 (4.76)
7 (15.22)
44 (42.31)
Tubulovillous adenoma
57 (29.69)
6 (14.29)
28 (60.87)
23 (22.12)
CIS
(N = 38)
(n = 8)
(n = 5)
(n = 25)
Age (years)
66.71 ± 13.77
75.75 ± 7.78
73.6 ± 5.5
62.44 ± 14.61
<0.05
Sex
F
13 (34.21)
3 (37.5)
2 (40)
8 (32)
M
25 (65.79)
5 (62.5)
3 (60)
17 (68)
Location
Right
18 (47.37)
6 (75)
4 (80)
8 (32)
<0.05
Left
9 (23.68)
0
1 (20)
8 (32)
Rectum
11 (28.95)
2 (25)
0
9 (36)
Size (mm)
17.66 ± 9.75
20.63 ± 12.96
19.6 ± 10.21
16.32 ± 8.64
Morphological features
Protruding type
20 (52.63)
6 (75)
1 (20)
13 (52)
Flat type
16 (42.11)
2 (25)
4 (80)
10 (40)
Depressed type
2 (5.26)
0
0
2 (8)
CRC
(N = 100)
(n = 17)
(n = 15)
(n = 68)
Age (years)
67.81 ± 12.48
71 ± 9.54
72.13 ± 12.78
66.06 ± 12.82
Sex
F
43 (43)
14 (82.35)
8 (53.33)
21 (30.88)
<0.01
M
57 (57)
3 (17.65)
7 (46.67)
47 (69.12)
Location
Right
48 (48)
16 (94.12)
6 (40)
26 (38.24)
<0.001
Left
22 (22)
1 (5.88)
3 (20)
18 (26.47)
Rectum
30 (30)
0 (0)
6 (40)
24 (35.29)
Stage (UICC)
I
38 (38)
8 (47.06)
3 (20)
27 (39.71)
II
31 (31)
4 (23.53)
5 (33.33)
22 (32.35)
III
23 (23)
5 (29.41)
7 (46.67)
11 (16.18)
IV
8 (8)
0
0
8 (11.76)
F, female; M, male; UICC, Union for International Cancer Control.
We next used principle component analysis to further evaluate our bisulfite pyrosequencing results and found that the first and second components accounted for 53.1% of the total variance (Figure 1D; see also Supplemental Table S4 at http://ajp.amjpathol.org). Two-dimensional plotting then showed that the characteristic pattern for each M-cluster was shared by the precursor and malignant lesions (Figure 1D; see also Supplemental Figure S4A at http://ajp.amjpathol.org).
Our hierarchical clustering analysis also showed that the marker genes could be categorized into three subgroups (Figure 1B). Group A genes, which were methylated in most of the samples, corresponded to the type A (age-related) genes originally proposed by Toyota et al.
Group B and C genes appeared to correspond to type C (cancer-specific) genes, whereas group C genes in this study were more strongly associated with CIMP-H. Interestingly, among BRAF-mutant precursor lesions, adenomas showed much higher levels of methylation of groups B and C genes than HPs (Figure 2). Similar results were obtained with KRAS-mutant precursors, although the difference was not statistically significant (Figure 2). By contrast, methylation was minimally up-regulated in adenomas in which both BRAF and KRAS were wild type, suggesting that accumulation of aberrant methylation, in concert with a BRAF or a KRAS mutation, may promote the progression from benign tumors to precancerous lesions.
Figure 2Association between methylation and BRAF/KRAS mutations. The levels of methylation of groups A to C genes in normal colon, precursor, and malignant lesions are shown as z scores. Colorectal lesions are divided into three groups, according to their BRAF or KRAS mutation status. Among the BRAF mutants, levels of methylation of group B and C genes are significantly higher in adenomas than HPs but are not further up-regulated in advanced lesions. *P < 0.05, ***P < 0.001.
Clinicopathological Features of CIMP-Positive Precursor Lesions
The associations between the clinicopathological characteristics and the CIMP status of the precursor lesions, CISs, and CRCs are summarized in Table 2. Age, sex, and tumor location were matched among the three groups. As with CRCs, CIMP-positive precursor lesions were more prevalent than CIMP-N precursor lesions among female and older patients. Interestingly, CIMP-L was associated with larger diameters among precursor lesions, although this tendency was not apparent among malignant lesions. Most of the CIMP-H precursor lesions were SSAs, whereas most of the CIMP-L precursor lesions were tubulovillous adenomas (Table 2; see also Supplemental Figure S4B at http://ajp.amjpathol.org). Levels of groups B and C gene methylation were higher in SSAs than in the other precancerous lesions, whereas methylation of group A genes was higher in all precursor types than in normal colonic tissue (see Supplemental Figure S4C at http://ajp.amjpathol.org).
CNAs Are Late Events in Colorectal Tumorigenesis
Several studies have shown an inverse relationship between CIMP and CIN in CRC.
18q Loss of heterozygosity in microsatellite stable colorectal cancer is correlated with CpG island methylator phenotype-negative (CIMP-0) and inversely with CIMP-low and CIMP-high.
For that reason, we next used array CGH to analyze CNAs in 40 precursor lesions, 25 CISs, and 19 CRCs (Table 3). Unsupervised hierarchical clustering analysis using the array CGH data revealed that the tumors could be categorized into three subclasses, according to their CNA status (C-clusters 1 to 3): C-cluster 1 was enriched in tumors with the fewest CNAs, whereas C-cluster 2 tumors were characterized by frequent copy number gains on chromosomes 7 and 19; both gains and losses were prevalent among tumors in C-cluster 3 (Figure 3A; see also Supplemental Figure S5A at http://ajp.amjpathol.org). Much of the precursor lesions [33 (82.5%) of 40] were enriched in C-cluster 1, whereas most of the malignant lesions (CISs and CRCs) were enriched in C-cluster 2 or 3, suggesting that CNAs occurred late during colorectal tumorigenesis (Figure 3A). Most of the CIMP-positive precursors and malignant lesions were categorized as C-cluster 1 or 2 (Figure 3A; see also Supplemental Figure S6A at http://ajp.amjpathol.org), and the methylation levels of the groups B and C genes were similar between C-cluster 1 and 2 tumors (see Supplemental Figure S6B at http://ajp.amjpathol.org). These observations suggested that CIMP and CNAs were inversely correlated in many colorectal tumors and that a subset of the CIMP-positive tumors (C-cluster 2) exhibited frequent copy number gains, particularly on chromosomes 7 and 19. As was previously seen, we found that most BRAF-mutant precursors and malignant lesions were enriched in C-cluster 1 (Figure 3A). By contrast, although most of the KRAS-mutant precursors were enriched in C-cluster 1, KRAS-mutant malignant lesions were equally distributed among all three C-clusters (Figure 3A; see also Supplemental Figure S6C at http://ajp.amjpathol.org).
Table 3Clinicopathological Features of the Colorectal Tumors Analyzed Using Array CGH
Figure 3Distinct subclasses of precursor and malignant colorectal lesions are defined based on their CNAs. A: Unsupervised hierarchical clustering analysis using array CGH data from 40 precursor lesions, 25 CISs, and 19 CRCs. Lesions could be categorized into three subclasses (C-clusters 1 to 3). CIMP status and gene mutations are indicated (top panel), as are chromosome (Chr) numbers (left panel). Ratios of precursor lesions, CISs, and CRCs in each C-cluster are shown (bottom panel). B: Ratios of genetic defects in AKT/PIK3CA pathway genes and TP53 mutations in precursor (left panel) and advanced (right panel) lesions with the indicated CNA status.
The results of our integrative genetic and epigenetic analysis of precursor lesions were indicative of several distinct molecular pathways leading to CRC development. Notably, CIMP-positive/BRAF-mutant CRCs did not exhibit more CNAs than did pre-invasive lesions with the same BRAF mutations and CIMP-positive methylation profile, suggesting that such pre-invasive lesions may progress to CRC without additional CNAs. By contrast, CIMP-positive/KRAS-mutant precursors appeared to develop via CNA-independent and CNA-dependent pathways. The CNA-dependent pathway was characterized by frequent amplification of BRAF and EZH2 on chromosome 7q and amplification of AKT2/PAK4 and DNMT1 on chromosome 19 (see Supplemental Figure S5B at http://ajp.amjpathol.org). More important, we found that most C-cluster 2 tumors exhibited genetic defects (mutations and/or CNAs) in genes whose products were implicated in the AKT/PIK3CA pathway, including AKT1, AKT2/PAK4, PDK1, and PIK3CA (Figure 3B; see also Supplemental Figure S5B at http://ajp.amjpathol.org).
Dynamics of the Molecular Signatures during the Progression of Colorectal Tumorigenesis
Our results suggested that acquisition of CNAs was essential for BRAF wild-type precursors to progress to more advanced tumors. To confirm this finding, we analyzed a series of colorectal lesions in which precursor components were present together with more advanced lesions within the same tumors. According to Kudo's classification, the aberrant pit patterns observed using magnifying colonoscopy are hallmarks of malignant tumors, and enabled us to distinguish between the precursor and advanced components (see Supplemental Figure S1 at http://ajp.amjpathol.org).
We first analyzed the precursor lesions (n = 22), in which portions with early pit patterns (type II or III) were present, along with more advanced pits (type IV), although both components were histologically premalignant (Figure 4A). Progression from precursor lesions with early pits to lesions with advanced pits was associated with the accumulation of DNA methylation, whereas genetic alterations (mutations and CNAs) were rarely acquired (Table 4; see also Supplemental Figures S7 and S8 at http://ajp.amjpathol.org). By contrast, progression from precursor (type II, III, or IV pit) to malignant (type V pit) lesions (n = 27) was accompanied by the occurrence of a wide variety of genetic changes, whereas methylation levels remained largely unchanged (Figure 4B and Table 4; see also Supplemental Figures S7 and S8 at http://ajp.amjpathol.org). For example, CIMP-H adenomas with a BRAF mutation acquired MSI as they developed into CISs, suggesting that inactivation of MLH1 and subsequent genetic instability were late events in the CIMP-H pathway. In addition, CIMP-L and CIMP-N adenomas acquired mutations and CNAs as they developed into CISs and CRCs. Most malignant lesions that exhibited C-cluster 2–type CNAs were derived from tubulovillous adenomas and were characterized by a KRAS mutation and CIMP-L (Table 4). On the other hand, most advanced lesions with C-cluster 3–type CNAs were CIMP negative, and more than half of those lesions were derived from tubular adenomas (Table 4).
Figure 4Endoscopic and histological findings in a set of mixed colorectal lesions. A: Endoscopic and histological findings from a representative precursor lesion in which a flat portion with an early pit pattern (type II, early portion) is present, along with a protruding portion with advanced pits (type IV, advanced portion). Both components are histologically premalignant (HP and TSA). Biopsy specimens were obtained from the respective portions (blue and red boxes), after which the molecular profiles are analyzed. B: Endoscopic and histological findings from a representative lesion in which a precursor portion (type IV, pit pattern) is present, along with a malignant portion (type V, pit pattern). Biopsy specimens were obtained from the respective portions (blue and red boxes), after which the molecular profiles are analyzed.
In the present study, we performed integrated genetic and epigenetic analyses with many colorectal neoplasias, including premalignant and malignant lesions. Because of the tight association between CIMP and the clinicopathological features of CRCs, it was anticipated that epigenetic profiling of premalignant lesions would provide important information that would aid in selecting appropriate therapeutic options and predicting clinical outcomes.
Numerous studies have confirmed that CRCs arise through the accumulation of both genetic and epigenetic alterations; however, the interactions between these alterations early during carcinogenesis remained largely uninvestigated. In addition, only a small fraction of colorectal adenomas may develop into malignant tumors, and progression from adenoma to cancer generally takes >10 years.
Thus, the identification of genetic and/or epigenetic alterations that directly correlate with the malignant potential of precursor lesions could facilitate risk assessment and enable prevention of CRCs.
Our comprehensive methylation analysis revealed that the aberrant methylation patterns characterizing CIMP are established early during colorectal tumorigenesis. We found that CIMP-H precursor lesions are strongly associated with SSA and BRAF/KRAS mutations, whereas CIMP-L precursor lesions are associated with tubulovillous adenomas and frequent KRAS mutations. Recently, Yagi et al
reported that colorectal adenomas could be classified into high-, intermediate-, and low-methylation epigenotypes, and that the intermediate-methylation epigenotype correlated significantly with a KRAS mutation, which is consistent with our observations. These results are indicative of the important relationship between the histological type and the molecular features of premalignant colorectal lesions. By contrast, CIMP-N precursor lesions contained various histological types, including HP, tubular adenoma, and tubulovillous adenoma. Recent reports have shown that a subset of HPs with BRAF or KRAS mutations progress to SSAs or TSAs, but the time at which aberrant methylation occurs remains unclear.
Advanced colorectal polyps with the molecular and morphological features of serrated polyps and adenomas: concept of a “fusion” pathway to colorectal cancer.
Our present findings indicate that, among KRAS- or BRAF-mutant precursors, the methylation levels of several genes were significantly increased during the progression from HP to adenoma. Concurrent increases in the methylation of multiple genes were further confirmed in mixed lesions containing HP and adenoma components (Table 4).
CIN is an important driving force promoting colorectal tumorigenesis, and recent studies have shown an inverse relationship between CIMP and CIN.
18q Loss of heterozygosity in microsatellite stable colorectal cancer is correlated with CpG island methylator phenotype-negative (CIMP-0) and inversely with CIMP-low and CIMP-high.
Our genome-wide CNA analysis of precursor and malignant lesions revealed that most CNAs are acquired during the progression from adenomas to CISs/CRCs. Consistent with earlier reports, CIMP-H tumors showed few CNAs, whereas both copy number gains and losses were prevalent among CIMP-N tumors.
18q Loss of heterozygosity in microsatellite stable colorectal cancer is correlated with CpG island methylator phenotype-negative (CIMP-0) and inversely with CIMP-low and CIMP-high.
Furthermore, we discovered that a subset of KRAS-mutant/CIMP-L tumors exhibited a unique CNA pattern characterized by frequent copy number gains at chromosomes 7 and 19, with relatively few copy number losses.
In contrast to the tight association between a BRAF mutation and CIMP-H in a subset of CRCs, the relationship between a KRAS mutation and CIMP status is not fully understood, which likely reflects the molecular complexity of KRAS-mutant CRCs.
Our analysis suggests that KRAS-mutant precursors progress to CRCs via three distinct pathways (Figure 5). First, such as BRAF-mutant/CIMP-H tumors, a subset of KRAS-mutant tumors, derived from SSAs, exhibit high levels of methylation and few CNAs. Although methylation of MLH1 is essential for BRAF-mutant/CIMP-H adenomas to develop into cancers, MLH1 methylation was infrequent among KRAS-mutant/CIMP-positive tumors. SSAs with a KRAS mutation are presumed to be the origin of MSI-negative/CIMP-H CRCs, although the molecular mechanisms underlying the progression from precursors to malignant lesions remain unknown.
Figure 5Model for development of CRCs via four distinct molecular pathways.
Second, we identified a subclass of KRAS-mutant/CIMP-positive cancers originating from tubulovillous adenomas or TSAs, in which alteration in AKT/PIK3CA signaling was crucially involved. In these tumors, genes associated with an AKT/PIK3CA signaling pathway were commonly affected during the progression from adenomas to malignant lesions. Several studies have shown aberrant AKT/PIK3CA signaling to be critical for CRC development, and mutations in PIK3CA, AKT1, AKT2, and PDK1 and amplification of AKT2/PAK4 are all reportedly associated with a poor prognosis.
In the present study, concurrent amplification of AKT2/PAK4 was specifically observed in KRAS-mutant/CIMP-positive cancers. Interestingly, we also observed frequent amplification of BRAF in this type of tumor. Although further study is needed to clarify its functional role in tumorigenesis, recent studies have shown that BRAF amplification promotes acquired resistance to MAPK/ERK kinase 1/2 inhibitors in CRC cells.
Amplification of the driving oncogene, KRAS or BRAF, underpins acquired resistance to MEK1/2 inhibitors in colorectal cancer cells [Erratum appears in Sci Signal 2011, 4: er2].
The molecular profiles of mixed precursor lesions suggest that CIMP is acquired during the progression from flat to protruding-type adenomas (Figure 4A; see also Supplemental Figures S2D, S7A, and S8A at http://ajp.amjpathol.org), whereas the KRAS mutation status was unchanged between the two components. This means that a KRAS mutation precedes CIMP in the KRAS/CIMP pathway, and that the acquisition of epigenetic changes, in addition to the KRAS mutation, may promote adenoma cell proliferation.
By contrast, KRAS-mutant/CIMP-N tumors are derived from tubular adenomas or tubulovillous adenomas and are characterized by a frequent TP53 mutation and high levels of CNA. An analysis of CIMP-N mixed lesions revealed that a KRAS mutation is found only in some advanced components, suggesting that a KRAS mutation occurs late during tumorigenesis in this pathway. Thus, KRAS mutations appear to play multiple roles during colorectal carcinogenesis. Further study will be required to fully characterize the functional diversity of KRAS mutations in CRCs; however, based on the genetic and epigenetic alterations found in CRC and the timing of the occurrence of KRAS mutations, we propose that CRCs develop via the four distinct pathways illustrated in Figure 5.
The results of several recent studies support the two-colon concept, which suggests that MSI-positive, CIMP-positive, and BRAF-mutant CRCs occur more frequently in the proximal colon.
Assessment of colorectal cancer molecular features along bowel subsites challenges the conception of distinct dichotomy of proximal versus distal colorectum.
recently reported that the frequencies of CIMP-H, MSI-high, and BRAF mutations in CRCs increased gradually along colorectal subsites from the rectum to the ascending colon. Because of the relatively few samples, we could not confirm this continuum concept in our present study. However, given its potentially significant impact on both basic and clinical research in CRC, testing this theory through further study of a larger population would seem warranted.
Our findings in this study have important implications for translating the molecular basis of carcinogenesis into a clinical benefit. The detection of high-risk precursor lesions is essential for preventing CRCs, and pit pattern observation using magnifying endoscopy enables us to detect neoplastic lesions with malignant potential.
We have dissected the morphological, histological, and molecular alterations in precancerous lesions of the colorectum and determined that they are directly linked to one another. Moreover, we provide strong evidence that aberrant pit patterns reflect histological changes and genetic and epigenetic defects in the precursor lesions, and that intratumoral variation in pit patterns could be predictive of the extent of the molecular abnormalities in a given tumor. Such a microstructure-based diagnostic system is readily available to the clinician, although specific skills are required for detailed pit pattern analysis. Further advances in the algorithm for pattern recognition may lead to the development of an innovative diagnostic system for detecting premalignant lesions and a reduction in CRC mortality.
Acknowledgments
We thank Dr. Yutaka Kondo for technical advice on MCAM analysis, Tomo Hatahira for technical assistance, and Dr. William F. Goldman for editing the manuscript.
Colonoscopic and pathological findings for mixed lesions. A: Colonoscopic views of a representative protruding-type tumor with (right panel) and without (left panel) indigo carmine dye. Magnified views of potions indicated by the blue and red boxes are shown. Pit patterns in these portions are types IV and V, respectively, indicating that this lesion consists of tubulovillous adenoma and CIS. Two biopsy specimens are obtained from each portion to extract genomic DNA, after which the lesion is treated with endoscopic mucosal resection. B: Histological appearance of the tumor in A. Biopsy sites are indicated by blue and red arrows. Histological findings for the portions with types IV and V pits in A are confirmed to be tubulovillous adenoma and CIS, respectively (bottom panel).
Genome-wide CpG island methylation in precursor and malignant colorectal lesions. A: Workflow of MCAM analysis and selection of methylation marker genes. B: Unsupervised hierarchical clustering analysis of MCAM results. Each row represents a probe, and each column represents a sample (28 precursor lesions, 14 CISs, 28 CRCs, 16 samples of normal colonic tissue, and 2 CRC cell lines). We selected 1010 probe sets (indicated by a red box) that showed tumor-specific methylation patterns for subsequent K-means clustering analysis (see Figure 1A). C: Volcano plot analysis using the MCAM data shown in Figure 1A to identify genes methylated in the high- and intermediate-methylation subclasses. Genes for which there are significant differences between the two clusters (P < 0.05, fold change, more than twofold) are shown in red. D: MCAM analysis of precursor lesions in which CIMP-positive portions are present, along with CIMP-N portions. Top panel, Endoscopic findings from precursor lesions. Flat portions are CIMP negative, whereas protruding portions are CIMP positive, in all cases. Bottom panel, Hierarchical clustering analysis of the MCAM data obtained from the lesions shown above and three samples of normal colonic tissue. Thirty-six unique genes differentially methylated between the CIMP-positive and CIMP-N components (fold change, more than twofold; P < 0.05) are selected, after which hierarchical clustering analysis is performed. E: Venn diagram analysis of the methylated genes identified by MCAM analysis in C and D.
RT-PCR analysis of genes newly identified by MCAM analysis in a set of CRC cell lines, with and without 5-aza-2′-deoxycytidine treatment. Results from DNMT1−/−DNMT3B−/− HCT116 cells (DKO2) and a normal colonic tissue sample are also shown. Gene expression is restored by demethylating treatment in several CRC cell lines.
Methylation profiling in early and advanced colorectal lesions and its association with the histological findings. A: Summary of the principle component analysis (PCA) shown in Figure 1D. Variances of the first (left panel) and second (right panel) components in specimens with the indicated CIMP status (M-cluster 1, CIMP-H; M-cluster 2, CIMP-L; and M-cluster 3, and CIMP-N) are shown. Precursor and advanced lesions (CISs and CRCs) exhibit similar PCA patterns, suggesting that aberrant methylation is established early during colorectal tumorigenesis. B: Scatter plot of the corresponding analysis to compare the methylation profiles with the histological characteristics of colorectal lesions. Samples are divided into three subclasses according to their methylation profiles (M-clusters 1 to 3, red boxes) or five subclasses according to their histological type (circles). Significant similarities are found between M-cluster 1 (CIMP-H) and SSA and between M-cluster 2 (CIMP-L) and tubulovillous adenoma. C: Levels of methylation of group A to C genes in colorectal lesions of the indicated histological type. Methylation levels are shown as z scores.
Analysis of CNAs using array CGH. A: Summary of array CGH data showing the frequencies of gains (red) and losses (green) in colorectal lesions with the indicated CNA profiles (C-clusters 1 to 3; see also Figure 3A). B: Representative genes frequently amplified or deleted in colorectal tumors. Frequencies of gain or loss of the indicated genes in each C-cluster are shown. Chr, chromosome.
Association between copy number profiles, CIMP status, and BRAF/KRAS mutations. A: Ratios of the respective CIMP status in the precursor (left panel) and advanced (right panel) lesions with the indicated CNA profiles (C-clusters 1 to 3). B: Methylation of group A to C genes in colorectal lesions in the indicated C-clusters. Methylation levels in lesions with their respective CNA profiles are shown as z scores. The levels of methylation of group A and B genes are significantly lower in C-cluster 3 tumors than in other subclasses. NS, not significant. C: Ratios of the respective C-cluster subclasses in precursor and advanced lesions. Samples are divided into three groups according to their BRAF or KRAS mutation status, indicated at the top. Most precursor lesions with KRAS mutations are categorized as C-cluster 1, in which CNAs are lowest. By contrast, advanced lesions with KRAS mutations are distributed among all three C-cluster subclasses, suggesting that a subset of precursor lesions acquire CNAs during their progression to CISs and CRCs.
Changes in molecular signature during the progression of colorectal lesions. A: Left panel, Endoscopic and histological findings from a representative precursor lesion in which a flat portion with an early pit pattern (type II) is present, along with a protruding portion with advanced pits (type IV). Both components are histologically premalignant (tubular adenoma and tubulovillous adenoma). Biopsy specimens are obtained from the respective portions (blue and red boxes), after which methylation profiles are analyzed. Right panel, Two-dimensional scatter plot of the methylation levels of the indicated genes in the early and advanced potions. Methylation levels are significantly higher in the advanced portion than in the early portion. Bottom panel, Array CGH analysis of the early and advanced portions. CNAs are infrequent in both subcomponents. B: Left panel, Endoscopic and histological findings from a representative lesion in which an adenomatous portion is present, along with a dysplastic portion. Biopsy specimens are obtained from the respective portions (blue and red boxes), after which methylation profiles are analyzed. Right panel, Two-dimensional scatter plot of the methylation levels of the indicated genes in the adenomatous and dysplastic portions. Methylation levels are similar between the two components. Bottom panel, Array CGH analysis for the adenomatous (blue) and dysplastic (red) portions. Regions with gains are shown at the top, and those with losses are indicated at the bottom. Both gains and losses are significantly greater in the cancer portion than the adenomatous portion. Chr, chromosome.
A: Levels of methylation of group A to C genes in precursor lesions containing both early and advanced pit patterns. Methylation levels are compared between portions with early pits and those with advanced pits. Methylation levels are significantly higher in the advanced portions than in the early portions, suggesting that aberrant methylation accumulates during the progression of colorectal tumorigenesis. B: Methylation in colorectal lesions in which precursor portions are present, along with malignant portions (CISs or CRCs). There are no significant (NS) differences between the methylation levels in the two portions.
A National Cancer Institute Workshop on Microsatellite Instability for cancer detection and familial predisposition: development of international criteria for the determination of microsatellite instability in colorectal cancer.
18q Loss of heterozygosity in microsatellite stable colorectal cancer is correlated with CpG island methylator phenotype-negative (CIMP-0) and inversely with CIMP-low and CIMP-high.
Advanced colorectal polyps with the molecular and morphological features of serrated polyps and adenomas: concept of a “fusion” pathway to colorectal cancer.
Amplification of the driving oncogene, KRAS or BRAF, underpins acquired resistance to MEK1/2 inhibitors in colorectal cancer cells [Erratum appears in Sci Signal 2011, 4: er2].
Assessment of colorectal cancer molecular features along bowel subsites challenges the conception of distinct dichotomy of proximal versus distal colorectum.
Supported by a Grant-in-Aid for the Third-Term Comprehensive 10-Year Strategy for Cancer Control (M.T. and H.S.), a Grant-in-Aid for Cancer Research from the Ministry of Health, Labor, and Welfare, Japan (M.T. and H.S.), the A3 foresight program from the Japan Society for Promotion of Science (H.S.), Grants-in-Aid for Scientific Research (A) from the Japan Society for Promotion of Science (K.I.), The Japanese Foundation for Research and Promotion of Endoscopy Grant (E.Y.), and a Research Grant from the Princess Takamatsu Cancer Research Fund (09-24119 to M.T.).
00603-7&id=gr1.jpg){kind=link}
00603-7&id=gr2.jpg){kind=link}
00603-7&id=gr3.jpg){kind=link}
00603-7&id=gr4.jpg){kind=link}
00603-7&id=gr5.jpg){kind=link}