Evidence for heritable predisposition to epigenetic silencing of MLH1
Version of Record online: 17 JAN 2007
Copyright © 2007 Wiley-Liss, Inc.
International Journal of Cancer
Volume 120, Issue 8, pages 1684–1688, 15 April 2007
How to Cite
Chen, H., Taylor, N. P., Sotamaa, K. M., Mutch, D. G., Powell, M. A., Schmidt, A. P., Feng, S., Hampel, H. L., Chapelle, A. d. l. and Goodfellow, P. J. (2007), Evidence for heritable predisposition to epigenetic silencing of MLH1. Int. J. Cancer, 120: 1684–1688. doi: 10.1002/ijc.22406
- Issue online: 21 FEB 2007
- Version of Record online: 17 JAN 2007
- Manuscript Accepted: 14 SEP 2006
- Manuscript Received: 1 JUN 2006
- US National Institutes of Health. Grant Numbers: CA71754, CA067941, CA16058, CA091842
- epigenetic silencing;
Epigenetic silencing of MLH1 is the most common cause of defective DNA mismatch repair in endometrial and colorectal cancers. We hypothesized that variation in the MLH1 gene might contribute to the risk for MLH1 methylation and epigenetic silencing. We undertook a case-control study to test for the association between MLH1 variants and abnormal MLH1 methylation. Eight MLH1 SNPs were typed in the normal DNA from women with endometrial carcinoma. For these studies, the cases were women whose cancers exhibited MLH1 methylation (N = 98) and the controls were women whose cancers had no MLH1 methylation (N = 219). One MLH1 SNP, rs1800734, located in the MLH1 CpG island at −93 from the translation start site, was significantly associated with MLH1 methylation as were age at diagnosis and patient body mass index. In validation experiments, a similar-sized cohort of colorectal carcinoma patients (N = 387) showed a similar degree of association with the −93 SNP; a smaller cohort of endometrial carcinomas (N = 181) showed no association. Combining all 3 cohorts showed an odds ratio of 1.61 (95% CI: 1.20–2.16) for the AA or AG vs. GG genotype at the −93 SNP. Identification of risk alleles for MLH1 methylation could shed light on mechanisms of epigenetic silencing and may ultimately lead to new approaches to the prevention or treatment of malignancies associated with MLH1 inactivation. © 2007 Wiley-Liss, Inc.
Loss of DNA mismatch repair leading to microsatellite instability (MSI) is seen in approximately 28% of endometrial cancers.1 Epigenetic silencing of MLH1 is the most common cause of defective DNA mismatch repair.2, 3, 4, 5, 6, 7, 8 In our large series of endometrial carcinomas, >20% have MLH1 promoter methylation.1 We previously showed that MLH1 methylation is associated with increased age at diagnosis.1 Furthermore, MSI+ endometrial cancers appear to be associated with lower body mass index (BMI).9 As is the case for endometrial cancers, MSI with MLH1 methylation is associated with older age at diagnosis in colorectal cancers.6, 10 MSI+ colorectal cancers tend to be right-sided tumors.11 To date, there have been no genetic factors associated with the risk for epigenetic silencing of MLH1 in tumors.
Murrell et al. reported an association between polymorphisms in the IGF2 gene and Beckwith-Wiedemann syndrome, an overgrowth syndrome in which loss of imprinting of IGF2 and KCNQ1 is frequently observed.12 Variants in the IGF2 differentially methylated region were found to be significantly associated with sporadic Beckwith-Wiedemann syndrome in which loss of maternal allele-specific methylation of the differentially methylated region KvDMR1 in the KCNQ1 gene was observed.12 This was the first evidence that sequence variation within a gene may be causally associated with abnormalities in methylation (loss of methylation).
Unlike the loss of methylation in Beckwith-Wiedemann syndrome, the MLH1 methylation defect in endometrial cancer is a gain of methylation in sequences that are normally unmethylated. We hypothesized that germline variation in the MLH1 locus might contribute to the risk for MLH1 methylation. We undertook a case-control association study to test for MLH1 variants associated with MLH1 methylation and epigenetic silencing in sporadic endometrial cancer.
Material and methods
Cancer patient and control populations
Three cancer populations (a St. Louis, Missouri endometrial cancer cohort, a Columbus, Ohio colorectal cancer cohort and a Columbus, Ohio endometrial cancer cohort) and 1 cohort of cancer free controls were investigated. All 3 cancer and the control populations were Caucasians only. For the endometrial cancer cohorts, only those patients with endometrioid tumors were investigated. The cancer free controls were from the St. Louis, Missouri region. All were female Caucasians, >65 years of age, with no history of cancer and with an intact uterus. All cancer patients and controls were consented to Institutional Review Board approved protocols. Clinical data, including age at diagnosis, BMI (St. Louis endometrial cancer cohort only) and tumor location (colorectal cohort only) were extracted from patients' clinical charts. Normal cellular DNA was prepared from peripheral blood leukocytes, or for a minority of cases uninvolved (nonneoplastic) uterine or colon tissues.
Characterization of tumors
All tumors were assessed for MSI using 5 microsatellite repeats (BAT25, BAT26, D2S123, D5S346 and D17S250 or D18S69) as described previously.1 For the St. Louis endometrial cancer cohort (N = 317), tumors were assessed for MLH1 methylation at cytosines at −250 and −252 relative to the initiation codon using combined bisulfite restriction analysis with BstUI.1, 13 For these studies, cases were Caucasian women with sporadic endometrial cancers whose tumors had MLH1 methylation and controls were Caucasian women with cancers that had no MLH1 methylation (2 controls were not evaluated for methylation because of inadequate tumor DNA quality).
The Columbus, Ohio colorectal cancer cohort included 387 patients, part of a larger series previously described.7 Among the 387 subjects, there were 99 cases whose tumors had MLH1 methylation at positions −250 and −252 (BstUI digestion combined bisulfite restriction analysis; all 99 were positive). There were 288 MSI-negative tumors (controls). This cohort represents sporadic cases that were unselected for age or other criteria. (HNPCC cases were, however, eliminated.) Although the MSI-negative tumors were not assessed for MLH1 methylation, all were shown to express MLH1 based on immunohistochemistry. Earlier studies showed that only approximately 5% of MSI-negative colon cancers exhibit MLH1 methylation.14 The Columbus, Ohio endometrial cancer cohort included 181 women, part of a larger series previously described.8 There were 55 cases (MSI+ endometrial cancers with MLH1 methylation at positions −250 and −252) and 126 controls (MSI-negative endometrial cancers). The MSI-negative controls were not evaluated for methylation. However, we assumed that approximately 95% would be unmethylated based on the results for the St. Louis endometrial cancers analyzed. These endometrial cancer patients were unselected for age or other criteria (HNPCC cases, however, were excluded) and are believed to have sporadic disease.
Normal (constitutional) DNA was genotyped for the 8 SNPs using Pyrosequencing. Pyrosequencing assays were designed using Primer3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) and Pyrosequencing AB version 1.0.1 software. PCR was performed in 96-well plates (20 μl volume, 11 μl of AmpliTaq Gold® PCR Master Mix, 0.5 μl of 4 μM forward and reverse primers and 20 ng of genomic DNA). PCR was carried out as follows: denaturation of 95°C for 30 sec, annealing at 55–60°C for 30 sec and extension at 72°C for 30 sec, for a total of 49 cycles, after an initial 93°C 20 min hot start. After PCR amplification, streptavidin-coated beads (Biotage, AB) were added to the PCR products and bound to biotinylated primers on the template strand DNA. A Pyrosequencing magnet was used to capture beads and move them into NaOH solution. The single strand template DNA binding beads were captured again by magnet and moved to annealing buffer for washing, followed by Pyrosequencing (PSQ™96 MA instrument).
Chi-square tests were used to test Hardy-Weinberg equilibrium in both cases and controls for each marker. Multivariate analyses were performed using logistic regression using SAS 8.0 (SAS Institute, Cary, NC). For linkage disequilibrium analysis, the normalized coefficients of disequilibrium D′ were calculated using PowerMarker software15 for intermarker pair-wise linkage disequilibrium comparisons.
St. Louis endometrial cancer cohort and control populations
Three hundred seventeen sporadic endometrial cancer patients were included in the MLH1 methylation association studies. All the patients were Caucasian and had a diagnosis of endometrioid endometrial cancer. Of these 317 cancer patients, 98 had tumors with methylation of the MLH1 promoter. These 98 women were classified as cases and the remaining 219 were controls. Eight MLH1 SNPs (Fig. 1) were typed in the endometrial cancer population with an overall genotyping success rate of 99.96%. The genotypes in both cases and controls were in Hardy-Weinberg equilibrium. The 8 SNPs were in high linkage disequilibrium (Fig. 1b). Representative examples of the Pyrosequencing™ genotyping and MLH1 promoter methylation analysis are shown in Figure 2. Bisulfite sequencing confirmed biallelic MLH1 methylation as previously reported.16 Analysis of 7 COBRA-positive tumor DNAs from −93 heterozygotes revealed methylation of numerous surrounding CpG sites (data not shown).
Patient age at diagnosis, BMI and −93 SNP were significantly associated with MLH1 methylation (Table I). The mean age of women with endometrial cancers in which MLH1 was methylated was 5.2 years greater than the mean age of women whose tumors lacked MLH1 methylation (p < 0.001). On average, women whose tumors had MLH1 methylation were thinner than those women whose tumor had no methylation (mean BMI 31.4 vs. 34.8, p = 0.005). The frequency of the minor allele (A) at the −93 SNP was higher in the cases than in controls (27.6% vs. 19.0%, p = 0.005). The −93 SNP is immediately downstream (157 bp and 159 bp) of the cytosines evaluated for methylation (Fig. 1a). A permutation analysis (10,000 random assignments for case/control status) gave an empirical P-value of 0.01, indicating that the value obtained for the −93 SNP (0.005) was unlikely to have occurred by chance. Furthermore, using a conservative Bonferroni correction accounting for the 8 SNPs, age and BMI variables in the model, the −93 SNP association remained significant.
|Variable||Alleles (major/minor)||Position||Location1||Minor allele frequency||p2|
|Cases (N = 98)||Controls (N = 219)|
|Age (range)||67.7 (46–92)||62.5 (26–91)||<0.001|
|BMI (range)||31.4 (16.9–53.6)||34.8 (13.8–86.6)||0.005|
Analysis of 286 cancer free controls from the US Central Midwest (all women over age 65 with an intact uterus) revealed a frequency of 21.2% for the −93 A allele, consistent with the frequency seen in the endometrial cancer control group (Table I). Furthermore, the odds ratio for the AA or AG genotypes versus GG genotype was 1.64 (95% CI: 1.04–2.61, p = 0.023) for the St. Louis endometrial cancer cases (all with MLH1 methylated tumors) versus St. Louis cancer free controls.
None of the remaining 7 polymorphisms showed significant association with MLH1 methylation (Table I).
Validation studies: Colorectal and endometrial cancer cohorts
We undertook studies to validate the association between the −93 SNP and MLH1 methylation in additional patient cohorts. Three hundred eighty-seven sporadic Caucasian colorectal cancer patients from the Columbus, Ohio region were studied. These patients are part of a larger population previously investigated for DNA mismatch repair abnormalities.7 Patient age, tumor location (right-sided disease) and the −93 SNP were significantly associated with MLH1 methylation in the colorectal cancer cohort (Table II). Again, the −93 SNP minor allele frequency was higher in the cases than in the controls (31.8% vs. 23.7%, p = 0.038).
|N = 99||N = 55||N = 288||N = 126|
|Age (range)||73 (38–98)||62 (23–86)||<0.001|
|Tumor location2 (mean score)||0.794||0.296||<0.001|
|−93 A allele||0.318||0.237||0.038|
|Age (range)||65 (30–88)||60 (17–94)||0.041|
|−93 A allele||0.236||0.246||0.999|
We then went on to evaluate a smaller endometrial cancer cohort (N = 181) from the Columbus, Ohio area. No association was seen between MLH1 methylation and the MLH1 SNP investigated. The failure to replicate the association observed for the St. Louis endometrial cancer cohort could be attributable to the power of the replication sample. On the basis of the number of patients available, the statistical power of the second endometrial cancer population was approximately 50% of the original St. Louis endometrial cancer cohort. The probability of detecting any true association signal(s) of the same effective size in the Ohio cohort was therefore only approximately half of that in the St. Louis cohort (under the assumption that the 2 populations are the same). The Ohio endometrial cancer patients who had MLH1 promoter methylation in their tumors were, however, significantly older than those women whose tumors lacked methylation (Table II). BMI data were not available for the majority of the endometrial patients and consequently we are unable to comment on whether the BMI/MLH1 methylation association seen in the St. Louis patient population is also seen in the Ohio endometrial cancer patients.
The −93 SNP A allele was associated with the MLH1 methylation phenotype in both the St. Louis, Missouri endometrial cancer cohort and the Columbus, Ohio colorectal cancer cohort. The strength of the associations by genotype in the 2 populations was consistent with the A allele having a dominant effect (data not shown). The odds ratio for the AA or AG genotypes versus GG genotype was 2.0 for the St. Louis, Missouri endometrial cancer population and 1.78 for the Columbus, Ohio colorectal cancer population (Table III). The association remains highly significant when all 3 populations are combined (2 endometrial cancer and 1 colorectal), p = 0.001, with an odds ratio of 1.61, 95% CI: 1.20–2.16. When all 3 cohorts were combined, 51% of cases carried an A allele whereas only 39% of controls had the A allele.
|AA+AG||GG||Odds ratio (95% CI)|
|St. Louis EC|
The association between the −93 A allele and MLH1 methylation in the St. Louis, Missouri endometrial cancer cohort and the Columbus, Ohio colorectal cancer cohort represents the first report of an inherited genetic factor influencing risk for epigenetic silencing in tumors. Germline epimutation of the MLH1 gene has, however, been described in several patients with multiple primary tumors and/or early onset MSI−+ colorectal cancer.17, 18, 19, 20 These patients exhibited methylation of MLH1 in normal tissues. In 4 of 5 informative cases, the homologue carrying the −93 SNP G allele was methylated whereas the A homologue was largely unmethylated.17, 18, 19, 20 Although the association between the −93 SNP and tumor-specific methylation does not explain the inherited epimutations that have been described,17, 18, 19, 20 it is consistent with cis factors contributing to somatic epimutation.
The association between MLH1 methylation in endometrial cancers and increasing age at diagnosis and lower BMI may point to distinct etiologies for sporadic endometrioid cancers with normal DNA mismatch repair and those with MLH1 methylation and epigenetic silencing. Increased age at diagnosis and lower BMI associated with MLH1 methylation likely reflects lower estrogen levels in women with methylated tumors relative to those women whose tumors are unmethylated. It is widely accepted that endometrioid endometrial cancers are more common in women exposed to excess levels of estrogen. Fat tissue is the major source of estrogen in post menopausal women and increased BMI correlates with higher estrogen levels.21 Given the observed association between lower BMI and increased age at diagnosis and MLH1 methylation in endometrial cancer in the St. Louis cohort, it may be that estrogen is less of a risk factor in women whose tumors exhibit MLH1 methylation when compared to women whose tumors have no methylation. The fact that the association between the MLH1 SNP and MLH1 methylation remains significant when the age and BMI effects are considered strongly supports the notion that the genetic factor is important in the St. Louis patient population. The failure to observe a similar effect in the Columbus Ohio endometrial cancer population could be taken to mean that the initial positive association was spurious. Alternatively, the lack of replication in the Columbus endometrial cancer population could be due to the small sample size (approximately 50% of the power for the St. Louis cohort), differences in environmental factors and potentially confounding genetic factors. However, the demonstration that the −93 SNP is associated with MLH1 methylation in a colorectal cancer cohort from the Columbus region strongly suggests that the genetic effect seen in the St. Louis patient is reproducible and not unique to endometrial cancers. It should be noted that the strength of the association and modest odds ratio (1.61 for 3 different populations combined) indicates clearly that other factors contribute to MLH1 methylation in endometrial and colorectal cancers.
It is noteworthy that the −93 (rs1800734) A allele was previously associated with risk for squamous cell lung cancers in Korean patients.22 The lung cancers were not, however, evaluated for MLH1 methylation and epigenetic silencing. Recent studies in Chinese and American non-small cell lung cancers have shown that methylation of the MLH1 promoter and/or epigenetic silencing of MLH1 is a frequent event.23, 24 It is conceivable that the −93 A allele (or sequence variation in linkage disequilibrium with the −93 A allele) is also associated with risk for MLH1 methylation in lung cancers and possibly in other malignancies. How variation in the 5′ region of the MLH1 gene influences risk for epigenetic silencing is yet to be determined. On the basis of footprinting experiments and reporter constructs, the associated variant (rs1800734 at −93) is part of the 5′ MLH1 sequences that bind an unknown factor and are required for optimal expression.25 It is not known whether variation at position −93 has a direct effect on gene expression, methylation of the promoter region and/or chromatin modification, or importantly, if other variants in linkage disequilibrium might explain the observed association with epigenetic silencing of MLH1 in endometrial and colorectal cancers. The observation that the association with the −93 A allele is seen for both AA homozygotes and AG heterozygotes might be explained by dominant trans effects akin to transvection.26 Our bisulfite sequencing analysis confirmed that as previously reported,16 biallelic MLH1 promoter methylation is common. Recent studies have suggested that site-specific repressors of transcription may recruit DNA methyltransferases, leading to de novo gene methylation and epigenetic silencing.27, 28 It is possible that variation in the MLH1 regulatory region could contribute to abnormal methylation and gene silencing as a result of altered transcription factor binding. Further studies to validate the association between the −93 SNP and MLH1 methylation, analysis of additional variants within or near the MLH1 locus and characterization of the role the −93 variant plays in transcription factor binding and expression should help clarify if and how promoter sequence variation is causally associated with risk for epigenetic silencing of MLH1.
The work is supported in part by the US National Institutes of Health (CA71754 to P.J.G, CA067941 to A.d.l.C, CA16058 to the Ohio State University Comprehensive Cancer Center and CA091842 to the Siteman Cancer Center, Barnes-Jewish Hospital Washington University in support of the Biostatistics, Hereditary Cancer and Tissue Procurement Cores). We thank Jenny Panescu, Mary Ann Mallon and Patricia Werner for their assistance with these studies.
- 9Body mass index: relationship to clinical pathologic and molecular characteristics of endometrial cancer. Gynecol Oncol 2005; 96: 996., , , , , .