SNP rs16906252C>T Is an Expression and Methylation Quantitative Trait Locus Associated with an Increased Risk of Developing MGMT-Methylated Colorectal Cancer

The DNA repair protein O⁶-methylguanine DNA methyltransferase (MGMT) plays a crucial role in protecting cells from the carcinogenic effects of alkylating agents by removing mutagenic alkyl adducts from DNA. MGMT molecules are degraded during the repair process, so need constant replenishment via transcription. The most frequent mechanism of transcriptional inactivation of MGMT in cancer is aberrant methylation of its CpG island promoter (1, 2). MGMT methylation or reduced MGMT mRNA levels in tumors are predictors of chemosensitivity to alkylating chemotherapy in patients with cancer (3–6). Therefore, it would be of translational interest to determine the factors, including genetic risk variants, which affect the expression and/or the methylation status of MGMT, as these could also influence the outcomes of treatment with alkylating chemotherapy (7, 8).

Colorectal cancer is heterogeneous and may be subdivided into 4 major molecular subtypes, based on gene expression and methylation profiles, and enrichment for particular genetic alterations, consistent with 4 distinct neoplastic pathways (9, 10). One subtype exhibits the CpG island methylator phenotype (CIMP)-high involving methylation of numerous CpG islands genome-wide, BRAF V600E mutation, and high rates of microsatellite instability (11, 12). Another subtype features CIMP-low, in which a subset of CIMP-high loci is methylated, and enrichment for KRAS mutations (12, 13). The remaining subtypes are CIMP-negative with chromosomal instability (9, 10). MGMT methylation occurs in around 25% to 40% of colorectal cancer (1, 2) and has been associated with KRAS mutations (2, 14–20), particularly G>A transitions (14, 15, 17, 18). However, no consistent correlation between MGMT methylation and CIMP status has been found (2, 9, 12, 20). MGMT methylation thus cross-sects the major molecular subtypes of colorectal cancer (9, 12), and the mechanisms underlying it remain undefined.

We and others have found a close association between carriage of the T variant of the germline SNP rs16906252C>T within the MGMT promoter enhancer region and the presence of MGMT methylation in colorectal cancer (2, 21), lung adenocarcinoma (22), pleural mesothelioma (23), glioblastoma (8, 24), and large cell B lymphoma (25). We and others have also reported low levels of MGMT methylation within normal tissues of cancer cases and healthy controls in association with the T variant (2, 26). The initial finding of low-level MGMT methylation within the macroscopically normal colorectal mucosa (NCM) of colorectal cancer cases led to speculation this may serve as a field defect preceding and predisposing to colorectal cancer (16), although the association with rs16906252C>T was unrealized at that time. Consistent with the concept of field cancerization (27), we showed that low-level MGMT methylation within the NCM of colorectal cancer cases was more commonly detected in those whose tumor was correspondingly methylated (2). Furthermore, we demonstrated that low-level MGMT methylation within the NCM of both colorectal cancer cases and controls was associated with the rs16906252C>T genotype (2). Low-level MGMT methylation has also been detected in the peripheral blood lymphocytes (PBL) of healthy controls, wherein the methylation was associated with rs16906252C≥T and linked specifically to alleles bearing the T variant (26).

The functional mechanism underlying the association between the rs16906252C>T genotype and MGMT methylation remains unknown. Of potential functional relevance, rs16906252C>T is a synonymous SNP located in exon 1 within a cis-acting enhancer element that is essential for MGMT upregulation (28, 29). Reduced transcription has been identified as one mechanism underlying permanent epigenetic inactivation via the accumulation of repressive histone modifications, and ultimately, promoter methylation (30). Promoter reporter assays performed in lung adenocarcinoma and glioblastoma cell lines have shown reduced activity from the T, compared with the wild-type C, allele at rs16906252C>T (8, 22). The effect of rs16906252C>T on transcriptional activity has not been systematically investigated ex vivo, although reduced levels of transcripts derived from the T allele was reported in bronchial epithelial cells from 2 heterozygous cancer-free individuals (22).

We hypothesized that the rs16906252C>T variant is associated with an elevated risk of developing an MGMT-methylated subtype of CRC, functionally mediated via constitutively reduced transcription from the T allele and its susceptibility to somatic methylation. Using a molecular pathologic epidemiology retrospective case–control study design, we tested our hypotheses in a test sample of 1,054 colorectal cancer cases, which we stratified by MGMT methylation status of the tumors, and 451 controls from Sydney, Australia, and replicated key observations in an independent sample of 612 cases and 245 controls from the Australasian Colon Cancer Family Registry (ACCFR). We also sought to elucidate the mechanism underlying the association between rs16906252C>T and MGMT methylation in colorectal cancer by investigating constitutive allelic methylation and expression profiles within normal tissues from subsets of cases and controls.

An overview of the study design and inclusion of subjects and specimens is provided in Fig. 1. Specimens of primary colorectal cancer and subject-matched NCM and/or PBL were drawn from 2 independent series of colorectal cancer cases from Australia. The “Sydney CRC series” comprised a sample of 1,054 colorectal cancer cases drawn from 1,178 patients who had undergone surgical resection of a primary colorectal cancer at St Vincent's Hospital Sydney, between 1993 and 2008 (Supplementary Table S1), including 1,039 patients in whom we previously reported a close association between the presence of MGMT methylation in colorectal cancer and rs16906252C>T (2). Cases identified with germline mutations within the mismatch repair genes, APC and MUTYH were omitted. The “ACCFR CRC series” comprised 612 cases diagnosed with incident colorectal cancer between 1997 and 2008, for whom both blood and tumor were available, recruited via the Victorian Cancer Registry, a population registry for the state of Victoria (Supplementary Table S2; ref. 31). For patients who developed 2 or more primary colorectal cancer, one cancer was selected at random for inclusion in this study. PBL was obtained from 2 groups of control subjects without a personal history of cancer recruited from the same catchment area as the respective colorectal cancer series. Controls for the “Sydney CRC series” comprised 451 blood donors recruited at St Vincent's Hospital Sydney via the Australian Red Cross Blood Service (ARCBS), whose rs16906252C>T genotypes we have previously reported (8). Controls for the “ACCFR CRC series” comprised 245 individuals randomly recruited from the general population via Victorian Electoral Rolls and were additionally age- and sex-frequency matched to the cases (31). All subjects were derived from predominantly Caucasian populations.

For ex vivo functional analyses of the rs16906252C>T SNP, subsets of cases from the “Sydney CRC series” and “ARCBS controls” were selected on the basis of genotype. NCM from 3 cancer-free subjects who underwent colonic resection at St Vincent's Hospital Sydney for indications other than colorectal cancer and who we previously identified as rs16906252C>T heterozygotes (2) were included as controls.

This study was approved by Human Research Ethics Committees of The University of New South Wales St Vincent's Campus (H02/022, H07/002), The University of Melbourne (#020702), and ARCBS (2003#08). All subjects provided signed informed consent prior to participation.

Genotyping of rs16906252C>T was performed on DNA extracted from PBL or NCM, either by PCR amplification followed by HhaI restriction fragment length polymorphism (RFLP) analysis, as previously described (2, 24) or by using Sequenom's iPLEX Gold. The genotyping assay was designed using MassARRAY Assay Design 3.0 software (Sequenom, Inc.) and performed according to the manufacturer's instructions. Extension product sizes were determined using Sequenom's Compact MALDI-TOF mass spectrometer. The resulting mass spectra were converted to genotype data using SpectroTYPER-RT software.

DNA was converted with sodium bisulphite using the EZ DNA Methylation Gold kit (Zymo Research). For tumors, MGMT methylation status was determined by COBRA in the Sydney CRC series and by MethyLight in the ACCFR CRC series, as previously described (2, 11, 19). Tumors with methylation levels of ≥5% (COBRA) or a percentage of methylated reference (PMR) value >4.0 (MethyLight) were considered “methylated,” as this threshold correlates most closely with protein loss (32).

As previously described (2), a sensitive real-time methylation-specific PCR (qMSP) assay was used to measure MGMT methylation in matched PBL, NCM, and colorectal cancer from a subset of colorectal cancer cases and PBL from a subset of controls (Fig. 2A). Tumors were categorized as “methylated” when the PMR > 4.0. PBL and NCM were categorized as “methylation-low” when PMR ≥ 0.01 (the analytical sensitivity threshold for this assay).

To determine whether methylation-low detected by q-MSP within normal tissues of rs16906252C>T heterozygotes was linked to a specific rs16906252 allele, an allelic-MSP assay encompassing rs16906252 was performed using primers 5′-GGTTGTTATCGTTTCGAGGGAGAGTT-3′ and 5′-CCGCGCCCCGAATATACTAAA-3′ (Fig. 2A). Amplicons were cloned using the pGEMTeasy PCR cloning system (Promega), and the inserts from 12 to 24 colonies sequenced.

Luciferase promoter reporter constructs containing the MGMT promoter sequence, with either the wild-type C (HAP1) or the variant T (HAP4) allele at rs16906252 inserted into the pGL2-basic vector (22), were transiently transfected into human embryonal kidney (HEK293) and colorectal cancer (HCT116) cell lines. The levels of luciferase reporter activity were measured by luminometry (Supplementary Methods).

To measure the relative levels of transcripts derived from the C and T alleles at rs16906252C>T, allele quantification (AQ) assays were performed in cDNAs from NCM of selected colorectal cancer cases and controls either directly by HhaI AQ-RFLP analysis of rs16906252 at c.66 or indirectly at the linked rs1803965C>T SNP at c.252 within exon 3 by AQ-pyrosequencing, and subsequently ascertaining the rs16906252;rs1803965 haplotypes by clonal sequencing of cDNA fragments encompassing both SNPs (Supplementary Methods and Supplementary Fig. S1). SNPs are located with reference to GenBank accession NM_002412.3. To ascertain haplotypes between rs16906252C>T of functional interest and the downstream rs1803965C>T SNP assayed, RT-PCR was performed across exons 1 and 3 encompassing both SNPs in cDNAs from double heterozygotes using primers 5′-CCGGATATGCTGGGACAG-3′ and 5′-CTGGTGAACGACTCTTGCTG-3′. Products were cloned in pGEMTEasy to isolate individual transcripts, and the inserts sequenced using vector primers.

For both AQ assays, the fold difference in the levels of rs16906252C:T transcripts were normalized against the levels of the respective alleles in genomic DNA and calculated as an allelic expression ratio (AER) by: (C_cDNA/T_cDNA)/(C_DNA/T_DNA) (33). Following haplotyping of double heterozygotes, the AERs obtained by AQ-pyrosequencing at rs1803965C>T in exon 3 were inverted, where necessary, to reflect the C:T AER at rs16906252C>T upstream.

MGMT methylation status was analyzed as a categorical variable except where stated. P values for unadjusted ORs were estimated using the χ² test. Multivariate unconditional logistic regression was used to determine independent factors among significant, unadjusted covariates. Where more than one factor was associated with presence of MGMT methylation, logistic regression was used to obtain adjusted ORs. A multinomial logistic regression model was used to compare the subsets of colorectal cancer cases stratified by MGMT methylation status, with healthy controls as the referent group. Nonnormally distributed, continuous PMR and AER values were compared between groups using a Mann–Whitney U test or between subject-matched samples using a related-samples Wilcoxon signed rank test. Luciferase expression values were compared using the Student t test. A correlation between AER and PMR values was estimated using Spearman ρ. All statistical tests were 2-sided.

The minor allele frequency (MAF) of rs16906252C>T was obtained from data accessed from dbSNP via the NCBI website http://www.ncbi.nlm.nih.gov/projects/SNP/ and the 1000 Genomes Project at http://browser.1000genomes.org/index.html.

In our test series of 1054 Sydney CRC cases, 292 tumors (27.7%) were positive for MGMT methylation. We found a clear association between the presence of MGMT methylation in colorectal cancer and carriage of the T variant at rs16906252C>T in univariate analysis [OR, 6.23; 95% confidence interval (CI), 4.3–8.9; P < 0.0001; Table 1], with strong evidence for an independent association in multivariate analyses (OR, 8.0; 95% CI, 5.3–12.1; P < 0.0001; Supplementary Table S1), recapitulating our previous findings among 1,039 of these cases (2). In our validation series of 612 colorectal cancer cases from the ACCFR (Supplementary Table S2), 129 tumors (21.1%) were MGMT-methylated, and a clear association between MGMT methylation and the rs16906252C>T genotype was found (OR, 7.12; 95% CI, 4.3–11.7; P < 0.001; Table 1). In both series, these associations remained significant when adjusted for age and gender (Table 1). Our consistent findings in 2 independent colorectal cancer series provide confirmatory evidence for a close association between rs16906252C>T and MGMT methylation in colorectal cancer. The rs16906252C>T genotype had low sensitivity but high specificity for an MGMT-methylated tumor (Supplementary Table S3).

To determine whether the rs16906252C>T genotype was associated with an increased risk of colorectal cancer development, we firstly compared the distribution of genotypes between the cases and controls in the test sample (Fig. 1A). The MAF of rs16906252C>T in the Sydney CRC series was 7.5%, and 14.9% of patients carried the T variant (CT/TT genotypes combined). In the respective ARCBS controls, the MAF was 6.1%, and 12% carried the T variant (8). The SNP genotype frequencies were in Hardy–Weinberg equilibrium in both the colorectal cancer (P = 0.21) and the control (P = 0.86) groups. Carriage of the T variant was not associated with risk for colorectal cancer overall (Table 3). Next, in a molecular pathologic epidemiology study design, we determined whether the rs16906252C>T genotype was associated with risk of developing colorectal cancer specifically exhibiting MGMT methylation in a subtype-specific manner. We stratified the colorectal cancer cases by the MGMT methylation status of their tumor into MGMT-methylated or MGMT-unmethylated categories and compared the distribution of genotypes in these 2 categories with the control group. Carriage of the T variant was associated with an increased risk for MGMT-methylated colorectal cancer (OR, 3.3; 95% CI, 2.0–5.3; P < 0.0001), coupled with a reduced risk for an MGMT-unmethylated colorectal cancer (OR, 0.5; 95% CI, 0.3–0.9; P = 0.01; Table 2).

To confirm these observations, we repeated the analyses in the validation sample (Fig. 1A). The MAF in the ACCFR CRC series was 7.0%, and 13.6% carried the T variant. The MAF in the matched control group was 6.3%, and 12.7% carried the T variant (Table 2). The SNP genotype frequencies were in Hardy–Weinberg equilibrium in both cases (P = 0.99) and controls (P = 0.29). The rs16906252C>T genotype was not associated with risk for colorectal cancer overall; however, rs16906252C>T was associated with an elevated risk for MGMT-methylated colorectal cancer (OR, 4.0; 95% CI, 2.4–6.8; P < 0.0001) and a reduced risk for MGMT-unmethylated colorectal cancer (OR, 0.5; 95% CI, 0.3–0.9; P = 0.02; Table 2). Hence, the associations we observed between the rs16906252C>T genotype and elevated risk of developing MGMT-methylated colorectal cancer and reduced risk of MGMT-unmethylated colorectal cancer in the test sample were replicated with similar magnitude in the validation sample.

We next sought to elucidate the mechanistic basis for the risk association identified between rs16906252C>T and development of an MGMT-methylated cancer. We determined whether (i) low-level MGMT methylation was detectable systemically or specifically within the NCM of colorectal cancer cases; (ii) this was associated with rs16906252C>T genotype; and (iii) methylation or genotype were predictive of the MGMT methylation status of the corresponding tumor. We used sensitive qMSP (Fig. 2A) to measure the MGMT methylation levels in matched PBL, NCM, and tumor tissues from a subset of cases from the Sydney CRC series. To maximize the number of cases with the rs16906252C>T genotype, all heterozygotes from this series were considered eligible for inclusion and those with availability of all 3 tissue types were included (n = 30; Fig. 1B). A similar number (n = 32) of C/C homozygotes were then selected from the series at random.

(A) Low levels of MGMT methylation were detected in either or both PBL and NCM of a proportion of cases. Figure 2B shows the actual PMR scores for the matched tissues from each case segregated by rs16906252 genotype. (B) In categorical analyses, methylation-low (PMR ≥ 0.01) was detected in the PBL of 9.4% of C/C and 73.3% of C/T cases. In NCM, methylation-low was detected in 15.6% of C/C and 93.3% of C/T cases (Table 3). A strong association between rs16906252C>T genotype and presence of methylation was identified for all 3 tissue types (P < 0.0001; Supplementary Table S4). (C) To determine whether the presence of methylation-low in PBL and/or NCM, or rs16906252C>T genotype, was predictive of MGMT methylation status in the corresponding tumor, we compared the strengths of their associations. Methylation-low in PBL or NCM and rs16906252C>T genotype showed moderate sensitivity and specificity for methylated tumors (Supplementary Table S5). Both the T genotype and the presence of methylation in PBL or NCM were strongly associated with tumor methylation (ORs, 11.9, 9.9, and 7.7, respectively; Table 3). However, the association for methylation-low in PBL and NCM, individually, declined significantly when adjusted for genotype (adjusted OR, 3.7, P = 0.08 and adjusted OR, 1.6, P = 0.6, respectively), whereas the risk for genotype adjusted by methylation-low status showed little change (PBL: adjusted OR, 5.8; NCM: adjusted OR, 8.3; Table 3). When compared with patients with unmethylated normal tissues, those with methylation-low in both NCM and PBL showed a strong association with methylated tumors (OR, 13.9). This association also diminished significantly when adjusted for genotype (adjusted OR, 3.3), whereas the genotype association remained borderline statistically significant when adjusted by methylation status (adjusted OR, 6.4; Table 3).

On the basis of this subset of colorectal cancer cases, rs16906252C>T genotype appears to be the primary driver of MGMT methylation in colorectal cancer. Nevertheless, when actual PMR scores in normal tissues were analyzed as a continuous variable segregated by rs16906252C>T genotype, C/T cases had significantly higher levels of MGMT methylation compared with C/C cases in both PBL (P < 0.0001) and NCM (P < 0.0001) (Fig. 2B). Furthermore, within subjects, the levels of MGMT methylation were significantly higher in NCM compared with PBL among C/T cases (P < 0.0001), but not in C/C cases (Fig. 2B). The latter findings support the hypothesis that, in the setting of colorectal cancer, the T variant confers susceptibility to MGMT methylation within normal tissues, and this occurs at higher levels in NCM than in PBL.

To determine definitively whether MGMT methylation-low within normal tissues of colorectal cancer cases was attributable to selective methylation of the T allele, allelic-MSP encompassing rs16906252C>T (Fig. 2A), followed by clonal sequencing of the products, was performed in C/T heterozygotes who had exhibited methylation-low by qMSP. Among the 28 C/T cases in whom methylation-low had been detected in NCM and/or PBL, 20 NCM and 5 PBL samples from 24 cases were positive by allelic-MSP. The allelic-MSP products from 18 NCM and 4 PBL samples were successfully cloned and sequenced. As illustrated for representative samples (Fig. 2C), methylation was dense and monoallelic on the T allele in 17/18 NCM, and biallelic, but predominant on the T allele, in one NCM. Dense methylation specific to the T allele was observed in all 4 PBL (Fig. 2C).

To determine whether MGMT methylation-low was detectable and associated with rs16906252C>T in the PBL of cancer-free subjects, we screened 164 ARCBS controls of known genotype (134 C/C, 30 C/T) by qMSP. Methylation-low was detected in 11.9% of C/C and 50% of C/T subjects and was clearly associated with the T variant (P < 0.0001, Supplementary Table S4). When actual PMR values were analyzed as a continuous variable, methylation levels were higher in C/T (median, 0.01% ± 0.19%) than in C/C subjects (median, 0.00% ± 0.39%; P < 0.0001, Fig. 2D). The frequency and levels of methylation in PBL were similar in cases and controls when stratified by rs16906252C>T genotype (Supplementary Table S4; Fig. 2D). These findings indicate that low-level MGMT methylation in PBL is dependent on rs16906252C>T genotype and is unrelated to cancer status.

To assess the effect of the T variant on MGMT promoter activity, we performed in vitro luciferase promoter reporter assays using constructs containing the MGMT promoter with either the wild-type C or the variant T allele at the rs16906252 site. In parallel transient transfections of the promoter reporter constructs, the T allele was expressed at levels of 65% and 63% relative to the C allele, in HEK293 and HCT116 cells, respectively (Fig. 3A).

To determine whether the T variant was associated with reduced expression ex vivo, the relative levels of transcripts derived from the C and T alleles of rs16906252C>T were measured in the NCM of rs16906252C>T heterozygous subjects, including 24 cases from the Sydney CRC series and 3 controls from St Vincent's Hospital, by allele quantification (AQ) RFLP at rs16906252 (Fig. 1C). Here, allelic expression levels were measured by densitometry of allele-specific bands post-electrophoresis (Supplementary Fig. S1A). AERs for the C allele relative to the T allele ranged from 1.4 to 4.8 (median, 2.3 ± 0.83) in the cases and from 1.2 to 4.6 (median 1.7 ± 1.84) in the controls, with no significant difference between cases and controls (Fig. 3B). Thus, a significant allelic expression imbalance (AEI) was observed between the two rs16906252 alleles in NCM, irrespective of cancer status, with expression from the T allele at 21% to 83% of that from the C allele.

To confirm these findings, we used an indirect AQ approach (Supplementary Fig. S1B). Using AQ pyrosequencing, the relative levels of allelic transcripts at a downstream reference SNP, rs1803965C>T located within exon 3, were measured in the NCM of a subset of 8 of the rs16906252C>T heterozygous cases who were also heterozygous for rs1803965C>T, and a comparison group of 9 rs1803965C>T heterozygous cases who were homozygous C/C at rs16906252 (Fig. 1C). Considerable variation in rs1803965C:T AERs were observed among the double-heterozygote group, with either genetic allele of the assayed rs1803965C>T SNP expressed at the higher level, whereas balanced allelic expression was observed among the cases homozygous C/C at rs16906252 (Supplementary Fig. S2). To adjust the AERs to reflect the transcriptional activity associated with rs16906252C>T in the double heterozygotes, we next determined the rs16906252;rs1803965 haplotypes by RT-PCR across exons 1 and 3 encompassing both SNPs, followed by clonal sequencing of the cDNA fragments (Supplementary Fig. S1B). The AERs for the double heterozygotes were then adjusted to rs16906252C:T AERs, irrespective of which allele showed higher expression at rs1803965 downstream. To minimize any statistical bias incurred by this adjustment, the rs1803965 AERs in the rs16906252 C/C homozygote group were adjusted to ≥1.0, irrespective of which allele of this assayed SNP showed the highest expression level. The range of adjusted rs16906252C:T AERs in the double heterozygote group was 1.8–4.5 (median ± SD: 2.5 ± 0.78), indicating reduced expression associated with the rs16906252 T variant in all 8 cases (Fig. 3C). The adjusted AER values obtained for these 8 cases by AQ-RFLP and AQ pyrosequencing showed strong concordance (Supplementary Fig. S2C). The range of adjusted AERs among the rs16906252 C/C homozygotes was 1.08–1.54 (median ± SD: 1.23 ± 0.18; Fig. 3C). The degree of AEI in rs16906252C>T cases was significantly higher than in the C/C homozygotes (P = 0.003; Fig. 3C). The significant transcriptional reduction associated with the rs16906252 T variant was not inversely correlated with methylation levels in the NCM samples, therefore not a consequence of it (Fig. 3D). Collectively, these findings provide strong evidence that the rs16906252 T allele is constitutively expressed at a reduced level in NCM, irrespective of cancer status.

We firstly confirm the close association between rs16906252C>T genotype and MGMT methylation in primary colorectal cancer. Including our study, this association has now been independently observed in 3 colorectal cancer series (2, 21). Similar findings at additional cancer sites suggest rs16906252C>T is a key determinant of MGMT methylation pan-cancer (8, 22–25).

A key novel finding of this study was evidence for an association between rs16906252C>T and an elevated risk of developing colorectal cancer exhibiting MGMT methylation, and conversely, a reduced risk of developing MGMT-unmethylated colorectal cancer. These risk associations were only identified by virtue of the molecular pathologic epidemiology approach taken, whereby risk associations between rs16906252C>T genotype and colorectal cancer was performed in a molecular subtype–specific manner after stratifying the cases by the MGMT methylation status of their tumor. This illustrates the power of molecular pathologic epidemiology, an emerging field that integrates pathology and epidemiology to take molecular heterogeneity of cancer into account in deciphering risk associations that may otherwise be imperceptible using standard molecular epidemiology approaches (34–37). The lack of an association between rs16906252C>T and overall risk for colorectal cancer is unsurprising, given that no cancer risk association has been reported for tag-SNPs around MGMT in genome-wide association studies (GWAS). rs16906252 is not represented on SNP arrays commonly employed for mass genotyping due to its low MAF in the general population, thus has not specifically been interrogated in GWAS of cancer.

One limitation of our case–control study was the lack of self-reported race/ethnicity data, hence for geographic origin matching purposes, our control groups were selected from the same catchment areas as the case series. The MAF reported for rs16906252C>T in dbSNP and the 1000 Genomes Project databases ranges from 4% to 9% for Europeans, 4% for South Asians, and 0% for Africans and East Asians. The MAFs of 6.1% to 6.3% in our controls and 7.0% to 7.5% in our cases is consistent with the high rate of Caucasians in the populations from which our subjects were recruited. A large prospective population-based study would be required to demonstrate whether the rs16906252C>T genotype predisposes to the development of cancers exhibiting MGMT deficiency.

In seeking to elucidate the mechanistic basis for the interaction between rs16906252C>T and risk for MGMT-methylated colorectal cancer, we detected low-level MGMT methylation in the PBL and NCM of colorectal cancer cases and in the PBL of healthy controls in association with rs16906252C>T. These findings are consistent with prior findings by our group in NCM from both colorectal cancer cases and controls (2) and by others in PBL from controls (26). Among colorectal cancer cases, genotype was a stronger predictor of an MGMT-methylated tumor than the presence of low-level methylation within either normal tissue. Nevertheless, among rs1696252C>T heterozygous cases, methylation levels were higher in NCM than in PBL, consistent with a role for this genotype-dependent MGMT methylation in field cancerization. Although global levels of methylation in these normal tissues were low, clonal sequencing revealed dense methylation of individual T alleles, consistent with methylation occurring in a small proportion of cells, predisposed by rs16906252C>T. The frequency and levels of MGMT methylation in PBLs did not differ between colorectal cancer cases and controls following stratification by rs16906252 genotype, further indicating that methylation was dependent on genotype, not cancer status or treatment. These findings provide firm evidence that rs16906252C>T represents a functional genetic variant that acts in cis to alter the somatic methylation state of the MGMT promoter, termed a methylation quantitative trait locus (meQTL). A recent study demonstrated a proportion of GWAS SNPs associated with cancer risk represent meQTLs, providing supportive evidence that SNP-associated secondary somatic epigenetic alteration may mediate cancer risk (38). We also demonstrated rs16906252 was associated with differential levels of allelic MGMT transcription, thus also represents an expression QTL (eQTL). Our luciferase promoter reporter assays showed that the T allele diminished promoter activity in HEK293 and HCT116 cells, consistent with prior findings in other cancer cell lines (8, 22), suggesting that rs16906252C>T directly effects transcriptional regulation of MGMT. More relevant, we showed that the T allele is expressed around half the level of the C allele in the NCM of both colorectal cancer cases and controls. This magnitude of transcriptional reduction was not attributable to MGMT methylation, which was present at low levels if at all, thus may be regarded as constitutive. Interestingly, rs16906252C>T predicted prolonged survival in glioblastoma patients treated with temozolomide, irrespective of whether their tumor was methylated or unmethylated at MGMT (8). Reduced transcription from the T allele provides one possible explanation for the extended survival experienced by T allele carriers whose glioblastoma was MGMT-unmethylated.

On the basis of our collective results, we propose a step-wise model to explain the risk association between rs169062562C>T and development of MGMT-methylated colorectal cancer. In this model, the T variant underlies constitutively reduced transcription (possibly by abrogating binding of transcriptional activators to the MGMT enhancer). This predisposes the T allele to chromatin remodeling, ultimately resulting in somatic promoter methylation within normal tissues. In turn, this stable epigenetic silencing predisposes to the development of MGMT-methylated cancer through field cancerization and clonal amplification of cells harboring MGMT methylation. The epigenetic alteration associated with rs16906252C>T within normal tissues likely serves as an intermediary linking this genotype with increased risk of developing an MGMT-methylated tumor.

No potential conflicts of interest were disclosed.

The content of this manuscript does not necessarily reflect the views or policies of the NCI or any of the collaborating centers in the Colon Cancer Family Registry (CCFR), nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government or the CCFR.

Conception and design: J. Kuroiwa-Trzmielina, D.D. Buchanan, C. Rosty, M.C. Southey, J.L. Hopper, M.P. Hitchins

Development of methodology: J. Kuroiwa-Trzmielina, D.D. Buchanan, C. Rosty, M.C. Southey, J.L. Hopper, M.P. Hitchins

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Kuroiwa-Trzmielina, R.W. Rapkins, R.L. Ward, D.D. Buchanan, M. Clendenning, C. Rosty, M.C. Southey, I. Winship, J.L. Hopper, M.A. Jenkins, N.J. Hawkins, M.P. Hitchins

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Kuroiwa-Trzmielina, F. Wang, R.W. Rapkins, D.D. Buchanan, A.K. Win, M. Clendenning, C. Rosty, J.L. Hopper, J. Olivier, N.J. Hawkins, M.P. Hitchins

Writing, review, and/or revision of the manuscript: J. Kuroiwa-Trzmielina, R.W. Rapkins, R.L. Ward, D.D. Buchanan, A.K. Win, M. Clendenning, C. Rosty, M.C. Southey, I. Winship, J.L. Hopper, M.A. Jenkins, J. Olivier, N.J. Hawkins, M.P. Hitchins

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Kuroiwa-Trzmielina, D.D. Buchanan, C. Rosty, M.C. Southey, J.L. Hopper, N.J. Hawkins, M.P. Hitchins

Study supervision: R.W. Rapkins, D.D. Buchanan, C. Rosty, M.C. Southey, J.L. Hopper, N.J. Hawkins, M.P. Hitchins

Other (funding): M.P. Hitchins

We gratefully acknowledge the Jeremy Jass Memorial Pathology Bank for the ACCFR pathology samples and data used in this study. MGMT promoter vectors were gifted by Drs. Belinski and Leng, Lovelace Respiratory Research Institute, Albuquerque, NM. We thank Robert W. Haile for critical reading of this article.

This work was supported by Australian National Health and Medical Research Council (NHMRC) grants APP568811 (M.P. Hitchins, R.L. Ward, N.J. Hawkins), APP1025799 (D.D. Buchanan, A.K. Win, M.A. Jenkins), APP1006498 (Career Development Fellowship II to M.P. Hitchins), APP1073395 (Early Career Fellowship to A.K. Win), APP1020493 (Senior Research Fellowship to M.A. Jenkins), APP1023434 (Senior Principle Research Fellowship to J.L. Hopper), Cancer Institute NSW grant RG104075 (Career Development Fellowship Top-Up Award to M.P. Hitchins), Reserve Talents of Universities Overseas Research Program of Heilongjiang grant [2013]350 (Travel Fellowship to F. Wang), and grant UM1 CA167551 from the National Cancer Institute (NCI) and through a cooperative agreement with the Australasian Colorectal Cancer Family Registry (U01 CA074778 and U01/U24 CA097735).

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.