Methylation of the Polycomb Group Target Genes Is a Possible Biomarker for Favorable Prognosis in Colorectal Cancer

CRC is the third most common type of cancer in the worldwide and the most common type in males in the Kingdom of Saudi Arabia (1). The age-standardized incidence rates (per 100,000) of CRC in Kingdom of Saudi Arabia vary between 9.8 in females to 14.3 in males (1) Although low compared with Western countries, CRC incidence rate in Kingdom of Saudi Arabia has almost tripled in less than 8 years (2). This increase in CRC incidence rates coincides with a shift towards Western world lifestyle including diet and daily activities (3, 4).

CRC is a heterogeneous disease with different molecular characteristics associated with the sites from which the tumors originate. Such heterogeneity is compounded by the multitude of genetic and epigenetic variations acting as passengers or drivers of the tumor (5). Majority of CRC develop via chromosomal instability (CIN) pathway. CIN is often exacerbated by inactivation of the Wnt signaling pathway “master regulator” APC gene (6), activating mutations of KRAS or BRAF oncogenes (7), or deletions of the 18q (8), and 17p (9) chromosomal regions with deleterious effects on the tumor suppressor genes TP53 and DCC. Defective mismatch repair (MMR) pathway results in a subtler form of genomic instability, namely microsatellite instability (MSI). High levels of MSI (or MSI-H) in sporadic CRC are usually caused by hypermethylation of the MLH1 promoter (10). In terms of methylation, the CpG island methylator phenotype (CIMP) pathway is the second most common pathway in sporadic CRC (11).

CRC tumors with high levels of CIN have a poor prognosis, especially if they are in stage II or III (12). Conversely, MSI-H tumors have a better clinical outcome compared with microsatellite-stable (MSS) tumors (13). CIMP-positive (CIMP+) CRC tumors are usually associated with the proximal colon of older females. CIMP+ tumors also show positive association with BRAF mutations (14). CIMP+ CRC tumors have better prognosis if the tumors are also MSI-H. However, CIMP+ CRC tumors that are MSS have poor clinical outcome.

In this study, we investigate the methylation frequency of several genes including the CIMP markers (IGF2, CACNA1G, NEUROG1, RUNX3, and SOCS1), the MMR genes (MSH2 and MLH1), tumor suppressor genes (APC, RASSF1A, and CDKN2A) in addition to MGMT DNA repair gene. The hypermethylation-mediated silencing of the polycomb group target (PCGT) genes (15) in cancer has been recently shown to be a hallmark of carcinogenesis (16). Although the methylation of such genes have been shown in ageing normal colon mucosa, their methylation is much more widespread and pronounced in cancerous samples (16–18). We have recently shown that methylation of SLIT2, SFRP1, HIC1, and MYOD1 is frequent in sporadic breast cancer from Saudi Arabia with methylation of the latter a possible marker for poor prognosis (19). The contribution of the inactivation of MMR genes in CRC for this population is yet to be shown. Similarly, such information is lacking for the other genes analyzed in this cohort.

The material of the present study consist of a series of 120 CRC specimens, retrospectively collected from the archives of Anatomical Pathology Laboratory in King Abdulaziz University Hospital (Jeddah, Kingdom of Saudi Arabia), covering the period from January 2005 to December 2009. Serial sections were cut from paraffin blocks, stained with hematoxylin and eosin for routine histologic examination, classification, grading, and staging following the American Joint Committee on Cancer (AJCC) staging system (20). The pertinent clinicopathologic data (gender, age, grade, and lymph node status), and follow-up results were retrieved from the patients' records after obtaining the relevant ethical approvals. DNA was extracted from 10 μm thin formalin-fixed paraffin-embedded slices using the Qiagen QIAMP Formalin-fixed Paraffin-embedded Tissue DNA extraction kit, following the manufacturer's guidelines. KRAS and BRAF mutational status were determined according to the previously published reports (21). The microsatellite instability was determined according to Berg and colleagues (22).

Up to 0.5 microgram of DNA was used for bisulfite conversion using the Qiagen Epitect Bisulfite Conversion kit. DNA methylation analysis was conducted using MethyLight as described elsewhere (23). The methylation levels of RASSF1A, APC, MGMT, CDKN2A, SLIT2, SFRP1, MYOD1, HIC1, MSH2, MLH1 and the CIMP markers IGF2, SOCS1, RUNX3, CACNA1G, NEUROG1 were analyzed using the primer-probe combinations listed in Table 1 which were made according to previously published reports (24–27). A probe targeting bisulfite-modified Alu repeat sequences was used to normalize for input DNA. The specificity of the reaction was ascertained using sssl-treated and bisulfite-modified positive control DNA (Qiagen) and the negative control DNA (Qiagen). The percentage of fully methylated reference (PMR) was calculated by dividing the gene:Alu ratio of a sample by the gene:Alu ratio of the positive control DNA and multiplying by 100. Samples with PMR more than 10 were considered positive for methylation, whereas samples with PMR less than 10 were considered negative (i.e., unmethylated). The PMR more than 10 is considered positive as it indicates a very likely hypermethylation-mediated loss of expression for the genes analyzed.

All statistical tests were carried out using IBM SPSS Statistics version 19. Fisher exact test was used to identify statistical significance of correlation between methylation events and clinicopathologic factors. The primary endpoints of the study included overall disease-free survival (DFS) calculated from the date of diagnosis to the appearance of disease recurrence or the last recorded date of being alive or death caused by CRC. In calculating DFS, patients who died of other or unknown causes were excluded. All survival times were calculated by univariate Kaplan–Meier analysis, and equality of the survival functions between the strata was tested by log-rank (Mantel–Cox) test. Multivariate Cox regression analysis was conducted to disclose independent predictors of DFS. All tests were 2-sided, and P values < 0.05 were considered statistically significant. K-means clustering was conducted using the Gene CLUSTER 3.0 program and visualized using JavaTree software (28).

We analyzed 120 patients with colorectal cancer selected on the basis of the availability of tissue material and clinical data. Mean age was 58 years (range, 24–96 years) with 34 patients (28.3%) being under 50 years. This cohort consisted of 72 male patients (60%) and 48 female patients (40%). Thirty eight (31.7%) tumors were in the right colon, 36 (30%) tumors were in the left colon, and 46 (38.3%) tumors were in the rectum. There is a significant association between the male gender and tumors from the right colon (P = 0.016). In addition, tumors from the left colon were more predominant in females (P = 0.027). Tumors from the right side are more likely to involve the lymph nodes (P = 0.039). There are no other significant differences between right, left, and rectal colon cancer in terms of age, grade, or recurrence.

The mutation status of BRAF codon 600 and KRAS at codon 12,13 was determined by sequencing. BRAF mutations were rare in our cohort (n = 120, 2.5%). KRAS mutations were more frequent (n = 108, 24.1%). There was no significant association between KRAS mutation and tumor location, age, sex, or grade. However, metastatic tumors are more likely to harbor a KRAS mutation (P = 0.007). The microsatellite instability (MSI) status was determined for 66 cases. A total of 22.7% patients had microsatellite stable (MSS) tumors, whereas 34.8% exhibited MSI-low status and 42.4% were MSI-high. In this cohort, it is more likely for the rectal cancer cases to be microsatellite stable (P = 0.012). We have determined the CpG island methylator phenotype (CIMP) by analyzing the methylation frequency of 5 genes; IGF2, SOCS1, RUNX3, CACNA1G, and NEUROG1. A case is considered CIMP+ if methylation of 2 or more genes can be detected. Overall, CIMP+ tumors were only 14.2% of the total cohort and are significantly associated with male patients (P = 0.014).

The methylation frequency of the 15 genes analyzed are shown in Fig. 1. The highest overall methylation frequency observed was for SFRP1 (70%) and the lowest was for MLH1 and MSH2 (1.2% and 0%, respectively). The methylation levels for the PCGT genes are consistently higher in tumorous tissues compared with matching nonmalignant counterparts (Supplementary Fig. S1). There is no significant association between the methylation of any gene and tumor location (with the exception of the CIMP markers). When stratified according to tumor location, metastasis is associated with SLIT2, SFRP1, and RASSF1A methylation in tumors originating from the rectum (P = 0.011, P = 0.003, and P = 0.039, respectively). Also in the rectal tumors, MYOD1 methylation is positively associated with KRAS mutations (P = 0.003). Rectal cancers from male patients exhibit a significant association with APC methylation (P = 0.007). Moreover, APC methylation positively associated with MSI-H rectal tumors (P = 0.012). Tumors originating from the left colon exhibit a positive association between SFRP1 methylation (P = 0.022), HIC1 methylation (P = 0.022), RASSF1A methylation (P = 0.018), and male patients. MSI-H right-sided tumors are positively associated with RASSF1A methylation (P = 0.019).

We have conducted K-means clustering analysis based on the methylation status of 13 genes and CIMP status (Fig. 2) to distinguish the subgroups of our cohort based on methylation events. As shown in Fig. 2, 3 distinct subgroups can be identified by K-means clustering. Group I (methylation-low, n = 36) is characterized by over representation of rectal cancer (50%) and being MSS or MSI-low (combined percentage is 70%). Group II (n = 67) is characterized by the prominent methylation of the PCGT genes (SFRP1, SLIT2, HIC1, MYOD1) in addition to the hypermethylation of MGMT, RASSF1A, and APC. Male patients were 59.7% of group II patients (Table 2). Group III (methylation-high, n = 17) is characterized by the predominance of male patients who in addition to methylation of the PCGT genes show positive CIMP status and frequent methylation of RASSF1A, APC, MGMT, and CDKN2A.

Group I exhibits week association with the female gender (P = 0.07) and tend to be well-differentiated tumors (P = 0.052). In addition, KRAS mutations show negative association with group I cases (P = 0.053). Group II is weekly but significantly associated with MSI-H status (P = 0.048). Group III is strongly associated with the male gender (P = 0.014) and poorly differentiated tumors (P = 0.048). None of the new groups show statistically significant association with age, tumor location, or KRAS mutation.

Next, we evaluated overall DFS in the 3 groups by univariate Kaplan–Meier analysis. When conducted on all patients, there was no significant effect of belonging into any of the groups on DFS. However, when stratified into young (<50 years old) versus old (>50 years old), worse DFS could be seen if the patient is older than 50 years old and displays group III methylation pattern (P = 0.058), Fig. 3A. When stratified according to the gender, group I patients display worse DFS in males only (P = 0.073). Interestingly, being a male patient displaying group II methylation pattern carries a favorable prognosis as significantly better DFS is observed (P = 0.027; Fig. 3B).

Multivariate Cox regression analysis was conducted with the variables like KRAS status, metastasis, age, sex, and tumor location in addition to grouping by K-means cluster analysis. As expected, metastasis is the strongest poor prognosis indicator with P < 0.0001 and HR of 8.837 [95% confidence interval (CI), 3.787–20.619]. However, group II methylation pattern is a good prognosis indicator with P = 0.013 and HR of 0.269 (95% CI, 0.095–0.761).

In this study, we have analyzed 120 cases of sporadic colorectal cancer (CRC) originating from the right colon, left colon, and the rectum for the presence of KRAS, BRAF mutations, MSI, and CIMP. We have additionally analyzed the methylation frequency of the polycomb group target genes (PCGT) that are normally silenced in stem cells. Furthermore, we have determined the methylation frequency of RASSF1A, APC, CDKN2A, MGMT, MLH1, and MSH2 in the same cohort.

The mutation rates of KRAS (24.1%) and BRAF (2.5%) were found to be lower than worldwide average, which may reflect possible ethnic differences affecting the mutation rates of these 2 oncogenes (29). CIMP frequency (14.2%) is similar to previous reports (11). However, we could not detect any statistically significant association between CIMP+ status and mutations of BRAF, KRAS, or presence of MSI. Moreover, CIMP+ tumors were found to be associated with the male gender in our cohort (P = 0.014). Right-sided tumors were also more prominent in males (P = 0.016). Therefore, there is a correlation between CIMP and right-sided tumors although this association is not statistically significant (P = 0.164).

The most common events in our cohort were the methylation of the PCGT genes as well as the methylation of MGMT gene. Interestingly, it has been suggested the MGMT is also a target of the polycomb complex (15). The methylation of the MMR genes was rare in our cohort. This is could be because CRC follows a different route to tumorigenesis and we cannot exclude the presence of inactivating mutations in this pathway

We have conducted K-means clustering analysis to stratify our samples based on their methylation signature. Our cohort can be separated into 3 distinct groups, group I (low methylation), group II (intermediate methylation), and group III (high methylation). Low-methylation frequency in all the genes studied is the hallmark of group I which is mostly, but not exclusively, represented by tumors originating from the rectum. Group I cases are likely to be well-differentiated tumors harboring wild-type KRAS (P = 0.052 and P = 0.053, respectively). Group II cases are MSI-H (P = 0.048); however, they are exclusively CIMP negative. Group II cases do not show any statistically significant association with any other clinicopathologic parameter despite being the group with best representation in our cohort. CRC tumors exhibiting the most frequent methylation amongst the genes analyzed clustered together in Group III. These poorly differentiated and CIMP+ tumors (P = 0.048) originated mostly from male patients (P = 0.014) but had no other significant associations. Belonging to any of the 3 groups can serve as a potential prognostic marker for overall survival depending on age or sex. Patients with CRC who are more than 50 years old and display group III methylation pattern have worse overall DFS compared with younger counterparts or patients displaying other methylation patterns. This is in line with previous reports of poor prognosis in CIMP+ cases that are not MSI-H (30). A potentially poorer clinical outcome is observed for male patients displaying group I methylation pattern. The reason for this observation is unknown as the presence and nature of genetic alterations in these tumors is not determined in this study. Unfortunately, the sample size of this group is too small to allow for a more definitive conclusion. Interestingly, however, is the observation that male patients with group II methylation pattern have a better clinical outcome compared with patients displaying other methylation pattern. This is perhaps not entirely surprising as MSI-H tumors generally have good prognosis (30). Multivariate Cox regression analysis confirmed this observation.

No potential conflicts of interest were disclosed.

Conception and design: A. Dallol, J. Al-Maghrabi, A. Buhmeida, M.S. Al-Ahwal, M.H. Al-Qahtani

Development of methodology: A. Dallol, J. Al-Maghrabi, A.G. Chaudhary

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Dallol, J. Al-Maghrabi, A. Buhmeida, M.A. Gari, A.G. Chaudhary, M.S. Al-Ahwal, A. Sibiany

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Dallol, J. Al-Maghrabi, H.-J. Schulten

Writing, review, and/or revision of the manuscript: A. Dallol, J. Al-Maghrabi, A. Buhmeida, A.G. Chaudhary, M.S. Al-Ahwal, M.H. Al-Qahtani

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Dallol, J. Al-Maghrabi, A.M. Abuzenadah, M.S. Al-Ahwal, M.H. Al-Qahtani

Study supervision: A. Dallol, J. Al-Maghrabi, A. Buhmeida, A.M. Abuzenadah, M.S. Al-Ahwal, M.H. Al-Qahtani

The authors thank the Ministry of Higher Education and King Abdulaziz City for Science and Technology (KACST) and the Scientific Chair for Colorectal Cancer for their financial support to this research. The authors also thank Ms. Shylu Mathews and Ms. Manar Ata for their significant technical assistance.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.