DNA methylation in peripheral blood measured by LUMA is associated with breast cancer in a population-based study

Abstract

Our purpose was to identify epigenetic markers of breast cancer risk, which can be reliably measured in peripheral blood and are amenable for large population screening. We used 2 independent assays, luminometric methylation assay (LUMA) and long interspersed elements-1 (LINE-1) to measure “global methylation content” in peripheral blood DNA from a well-characterized population-based case-control study. We examined associations between methylation levels and breast cancer risk among 1055 cases and 1101 controls and potential influences of 1-carbon metabolism on global methylation. Compared with women in the lowest quintile of LUMA methylation, those in the highest quintile had a 2.41-fold increased risk of breast cancer (95% confidence interval: 1.83–3.16; P, trend<0.0001). The association did not vary by other key tumor characteristics and lifestyle risk factors. Consistent with LUMA findings, genome-wide methylation profiling of a subset of samples revealed greater promoter hypermethylation in breast cancer case participants (P=0.04); higher LUMA was associated with higher promoter methylation in the controls (P=0.05). LUMA levels were also associated with functional sodium nitroprusside in key 1-carbon metabolizing genes, MTHFR C677T (P=0.001) and MTRR A66G (P=0.018). LINE-1 methylation was associated with neither breast cancer risk nor 1-carbon metabolism. Our results show that global promoter hypermethylation measured in peripheral blood was associated with breast cancer risk.—Xu, X., Gammon, M. D., Hernandez-Vargas, H., Herceg, Z., Wetmur, J. G., Teitelbaum, S. L., Bradshaw, P. T., Neugut, A. I., Santella, R. M., Chen, J. DNA methylation in peripheral blood measured by LUMA is associated with breast cancer in a population-based study.

Breast cancer is a manifestation of abnormal genetic as well as epigenetic changes (1). While alterations in methylation status within promoter regions may affect gene control by impairing transcription, global loss of DNA methylation may affect chromatin structure, induce chromosomal instability and aneuploidy, and increase mutation rates (2, 3). Because the methylation profiles of the human genome are tissue specific (4) and because of the high cost of whole-genome methylation profiling, global methylation levels, rather than specific profiles in surrogate tissues, such as blood, have gained promise in epidemiologic studies in cancer (5–7). In this study, we used 2 independent but complementary methods to assess global methylation levels with 2 distinct focuses. While long interspersed elements-1 (LINE-1) methylation approximates global methylation levels of repetitive elements or transposons, which play key roles in maintaining genomic stability (8), the luminometric methylation assay (LUMA; refs. 9, 10) measures levels of 5-^mC in the C^mCGG motif, which is over-represented in gene promoters in the genome (11).

One-carbon metabolism is involved in DNA methylation, as it provides the universal methyl donor, S-adenosylmethionine (SAM). We have systematically studied the association of 1-carbon metabolism and breast cancer in the Long Island Breast Cancer Study Project (LIBCSP; refs. 12–17). We found that intakes of 1-carbon related nutrients, such as folate and related B vitamins, were inversely associated with breast cancer risk (12, 16), and functional polymorphisms of the key genes in the 1-carbon metabolic pathway modified these associations. However, our knowledge on whether these associations act through an epigenetic pathway, or more generally whether environmental or modifiable factors influence DNA methylation, is extremely limited. In this study, we performed both LUMA and LINE-1 methylation assays on >2100 peripheral blood samples from the LIBCSP participants. We systematically evaluated the association of methylation dysregulation with breast cancer risk and the influence of 1-carbon metabolism on these measurements.

MATERIALS AND METHODS

Study population

We utilized the resources of the LIBCSP, a population-based study. Details of the study design have been described in detail previously (18, 19). Briefly, case participants were women residing in Nassau and Suffolk counties of Long Island, NY, who were newly diagnosed with a first primary breast cancer between August 1, 1996 and July 31, 1997. Control participants were matched by frequency to the expected age distribution of the cases; they were identified through random digit dialing for those younger than 65 yr and through the Center for Medicare and Medicaid Services rosters for those older than 65 yr. Participant information was obtained as part of the in-person, interviewer-administered interview and medical record abstraction that occurred in 1998. During the structured interview, information on demographic characteristics and breast cancer-related factors was ascertained. Blood samples were collected by trained field staff from 73.1% of the case and 73.3% of the control participants at the time of the case-control interview, which occurred shortly after diagnosis and before the completion of the first course of treatment for the majority of case participants (18). DNA was isolated from blood specimens using the methods previously described (19). The study protocol was approved by the institutional review boards of the collaborating institutions.

LUMA

The principle of the LUMA has been described in detail elsewhere (10). Briefly, 200 ng genomic DNA was digested for 4 h by HpaII + EcoRI or MspI + EcoRI in 2 separate reactions, which were set up in a 96-well plate format. Then, 15 μl annealing buffer (20 mM Tris acetate and 2 mM Mg-acetate, pH 7.6) was added to the digestion product, and samples were analyzed with a PyroMark Q24 system (Qiagen, Valencia, CA, USA) with the following dispensation order: GTGTCACATGTG. The method has been validated using samples with known DNA methylation status (20). We followed the modified protocol described by Bjornsson et al. (20). 5-Aza-dC demethylated and CpG methylated Jurkat genomic DNA (New England Biolabs, Ipswich, MA, USA) were used as unmethylated and methylated controls in the assay. The LUMA methylation level was expressed as a percentage obtained from the following equation: methylation (%) = [1 − (HpaII ΣG/ΣT)/(MspI ΣG/ΣT)] * 100 (20). Randomly selected samples were replicated to examine the batch effect or variation between different runs. The corresponding CV% was <1%. Cases and controls were assayed at the same time with laboratory personnel blinded to both case-control and quality control status.

LINE-1

A prevalidated pyrosequencing-based methylation assay (21, 22) was used to assess 4 CpG sites in the promoter region of LINE-1. Genomic DNA (500 ng) was bisulfite treated and purified using Zymo DNA methylation kits (Zymo Research, Irvine, CA, USA) and eluted in 20 μl elution buffer. Each 50-μl PCR contained 10× PCR buffer, 3.0 mM MgCl₂, 200 μM dNTPs, 0.2 μM primers, 1.25 U DNA polymerase (HotStar; Qiagen), and ∼10 ng of bisulfite-converted DNA. The polymerase was activated by incubation at 95°C for 10 min, followed by 34 cycles of 95°C for 15 s, 45°C for 30 s, and 72°C for 30 s. The reaction was then allowed to extend for 5 min at 72°C. A universal biotinylated primer was used in the initial PCR reaction to allow for isolation of the amplicon, followed by denaturation and release of a single-strand product for pyrosequencing. PCR products (10 μl) were sequenced using the Pyrosequencing PSQ96 HS System (Qiagen). Methylation status at each of the 4 loci was analyzed individually as a T/C single nucleotide polymorphism (SNP) using QCpG software (Qiagen). Methylation status data at all 4 loci were averaged to provide an overall percentage 5-^mC status. The assay was performed at EpigenDX (Worcester, MA, USA).

Whole-genome methylation profiling

As functional validation to our measurements on global methylation levels, we performed genome-wide methylaton profiling by Infinium HumanMethylation27 BeadChip (Illumina, Inc., San Diego, CA, USA) on a subset of 24 study samples (12 case/control pairs; 6 pairs each with LUMA values in the lowest and highest quintile). A total of 500 ng of genomic DNA was treated with sodium bisulfite using the EZ DNA methylation-gold kit (Zymo Research). Before bead array analysis, the quality of DNA modification was verified by pyrosequencing. The Infinium methylation assays were performed in accordance with the manufacturer's instructions. The assay generates DNA methylation data for 27,578 individual CpG sites spanning 14,495 well-annotated, unique gene promoters (from −1500 to +1500 from the transcription start site). The assay information is available online (http://www.illumina.com). All 24 samples were randomized, modified, labeled, hybridized, scanned, and analyzed in the same batch after randomization to the two beadchips to minimize the variation introduced by experimental procedure. The laboratory personnel were blinded to case/control status and LUMA level. Methylation data on the X chromosome were excluded from the data analyses. BeadStudio 3.2 was used for obtaining the signal values (AVG-β) corresponding to the ratio between the methylated allele and the sum of the signals of both methylated and unmethylated alleles. After filtering, GraphPad Prism software (GraphPad Software, La Jolla, CA, USA) was used for further analysis, using the AVG-β values. The Mann-Whitney test and the Wilcoxon signed rank test were used for unpaired and paired analysis comparing average methylation between classes, respectively.

Genotyping and dietary assessment

Genotyping was conducted on 13 functional polymorphisms in the 1-carbon metabolism pathway using methods we described previously (12, 14). These 13 polymorphisms are as follows: methylenetetrohydrofolate reductase (MTHFR) rs1801133 and rs1801131; thymidylate synthetase (TYMS) 3R/2R tandem repeat; dihydrofolate reductase (DHFR) 19-bp deletion; methionine synthase (MTR) rs1805087; methionine synthase reductase (MTRR) rs1801394; betaine-homocysteine S-methyltransferase (BHMT) rs3733890; reduced folate carrier-1 (RFC1) rs1051266; cytoplasmic serine hydroxymethyltransferase (cSHMT) rs1979277; phosphatidylethanolamine N-methyltransferase (PEMT) rs7946 and rs12325817; and choline dehydrogenase (CHDH) rs12676 and rs9001. Dietary intake values for 1-carbon related micronutrients and compounds were calculated based on data collected as part of the case-control interview, assessed using a modified block food frequency questionnaire, which was self-completed by the study participants (23, 24).

Statistical analysis

Analyses were undertaken to estimate the associations between LUMA and the risk of developing breast cancer among case and control participants. The relationships between 1-carbon nutrients/polymorphisms and LUMA were investigated only among controls to minimize the potential influence by the onset of breast cancer. The same strategies were applied to the LINE-1 data analyses. All statistical analyses were performed using SAS 9.2 statistical software (SAS Institute, Cary, NC, USA).

First, for the case-control analyses, methylation levels (LUMA and LINE-1) were first used as a continuous variable, expressed as a percentage. After being tested for normality, methylation differences between cases and controls were compared by t test. Methylation level was then categorized into quintiles (Q1–Q5, Q1 is the lowest) based on the distribution among controls and entered into an age-adjusted unconditional logistic regression model to examine the association between LUMA methylation and the risk of developing breast cancer. Tests for linear trend of odds ratios (ORs) were calculated by treating methylation levels as a 5-level ordinal variable in the model. To rule out the influence of factors that may affect methylation levels, we performed subgroup analyses restricted to the study population with the following criteria: no chemotherapy, no radiotherapy, no horm1 therapy, and no family history. Further, to examine possible subgroup effects, we considered potential effect modification by menopausal status, and other modifiable breast cancer risk factors, using stratified analyses. We also conducted analyses by breast cancer subtypes as defined by cancer type (in situ vs. invasive) and hormone receptor status [estrogen receptor (ER) and progesterone receptor (PR) as assessed by immunohistochemistry, with the positive group defined as ER and PR both positive (ER⁺/PR⁺) and negative group defined as all others (ER⁺/PR⁻, ER⁻/PR⁺, and ER⁻/PR⁻].

Second, among controls only, we explored the influence of 1-carbon factors (diet or genetic) on methylation levels using ANOVA to compare the mean methylation levels among different nutrient intake levels and different genotypes. ANOVA was also used to compare mean methylation levels across categories of other covariates, including demographic characteristics and other factors that may affect the development of breast cancer.

Potential confounding effects were evaluated by adjustment in multivariate models. Potential confounders include age (at diagnosis among case participants and at identification among controls), menopausal status (pre- vs. post-), race, family history of breast cancer, and history of benign breast. If eliminating a covariate from the full model changed the effect estimate by ≥10%, the covariate was considered a confounder and kept in the model. Otherwise, that covariate was dropped from the multivariate model. None of the breast cancer risk factors tested was considered a confounder using this criterion; thus, only results from the age-adjusted model are presented. Potential interactions were tested on a multiplicative scale by performing likelihood ratio tests to compare the difference of log likelihood statistics for a model with or without a cross-product term for 2 main effect variables.

RESULTS

LUMA methylation is positively associated with breast cancer risk

LUMA methylation was measured on 2156 blood samples. The mean level of LUMA of 57.3% (sd: 15.7%; IQR: 20.5%) among the 1055 breast cancer case participants was significantly higher than the mean of 52.4% (sd: 16.7%; IQR: 22.9%) among the 1101 control participants (P<0.0001 for t test). The percentage represents the proportion of 5-^mC within the CCGG motif throughout the genome. Logistic regression analyses (Fig. 1) revealed a positive association between LUMA methylation level and breast cancer risk in a dose-dependent fashion (P<0.0001 for trend), where a >2-fold increase in breast cancer risk [OR 2.41, 95% confidence interval (CI): 1.83–3.16] was observed among individuals in the highest quintile of LUMA methylation (Q5) compared with those in the lowest (Q1).

Figure 1.

LUMA level and breast cancer risk in the LIBCSP. Methylation levels were categorized into quintiles (Q1–Q5, Q1 is the lowest) based on the distribution among controls. ORs presented are age adjusted.

To consider possible influences from factors that may affect methylation levels, such as clinical therapy undergone for the first primary breast cancer and family history of breast cancer, we performed subgroup analyses restricted to the study population without these factors. As shown in Table 1, subgroup analyses were restricted to the following: case participants who donated blood before receiving any chemotherapy (78.8%), radiation therapy (79.1%), and hormonal therapy (71.8%), or who had only undergone surgery before donating blood (47.4%); or among case and control participants without a family history of breast cancer (83.8%). Positive associations between LUMA methylation and breast cancer risk, similar to those observed in the overall population, persisted in all subgroups, with a consistent >2-fold elevation in risk in Q5 of LUMA methylation (Table 1).

Table 1.

Subgroup analyses of LUMA level and breast cancer risk in the LIBCSP

Subgroup	LUMA	Case	Control	OR (95% CI)	P, trend
Case-only characteristics
Prechemotherapy	Q1	100	216	1.00 (ref.)	<0.0001
	Q2	118	219	1.16 (0.84–1.61)
	Q3	134	219	1.32 (0.96–1.82)
	Q4	207	215	2.08 (1.53–2.82)
	Q5	272	219	2.68 (1.99–3.61)
Preradiation	Q1	118	216	1.00 (ref.)	<0.0001
	Q2	115	218	0.97 (0.70–1.33)
	Q3	156	218	1.31 (0.97–1.78)
	Q4	194	218	1.63 (1.21–2.19)
	Q5	251	218	2.11 (1.58–2.81)
Prehormonal	Q1	97	209	1.00 (ref.)	<0.0001
	Q2	109	205	1.15 (0.82–1.60)
	Q3	142	204	1.50 (1.09–2.07)
	Q4	171	209	1.76 (1.29–2.41)
	Q5	238	205	2.50 (1.84–3.39)
Surgery only	Q1	58	208	1.00 (ref.)	<0.0001
	Q2	75	203	1.33 (0.89–1.96)
	Q3	92	203	1.63 (1.11–2.38)
	Q4	121	206	2.11 (1.46–3.04)
	Q5	154	204	2.71 (1.89–3.87)
Case and control characteristics
No family history	Q1	105	185	1.00 (ref.)	<0.0001
	Q2	122	188	1.14 (0.82–1.59)
	Q3	144	192	1.32 (0.96–1.82)
	Q4	198	186	1.88 (1.37–2.56)
	Q5	263	177	2.62 (1.93–3.56)

Treatment information was the status of the case participants' first course of treatment at the time of phlebotomy, which was undertaken shortly after the diagnosis of the first primary breast cancer. ORs presented are age adjusted.

To investigate possible heterogeneity of results in the LUMA-breast cancer risk association when considering breast cancer subtype, we performed analyses with the case groups defined by selected tumor characteristics, including ER/PR status (ER⁺/PR⁺ vs. other ER/PR status) and breast cancer type (in situ vs. invasive; Table 2). The positive association between LUMA methylation and breast cancer risk did not differ by ER/PR status; within each stratum, the highest methylation category (Q5) was associated with >2-fold risk. However, the methylation association with breast carcinoma in situ appeared to be more prominent, with the highest methylation quintile conferring a >6-fold risk, although the confidence interval is wide (OR: 6.18; 95% CI: 3.34–11.43).

Table 2.

LUMA level and breast cancer risk by tumor characteristics, menopausal status, and lifestyle factors in the LIBCSP

Characteristic	LUMA	Case	Control	OR (95%CI)	Case	Control	OR (95%CI)
Tumor characteristics (case only)
ER/PR status		Negative			Positive
	Q1	35	221	1.00 (ref.)	57	221	1.00 (ref.)
	Q2	38	220	1.09 (0.66–1.79)	71	220	1.25 (0.84–1.86)
	Q3	49	220	1.41 (0.88–2.26)	75	220	1.32 (0.89–1.96)
	Q4	61	220	1.75 (1.11–2.76)	103	220	1.82 (1.25–2.64)
	Q5	80	220	2.30 (1.48–3.56)	113	220	1.99 (1.38–2.88)
	P			<0.0001			<0.0001
Cancer type		In situ			Invasive
	Q1	13	221	1.00 (ref.)	128	221	1.00 (ref.)
	Q2	29	220	2.24 (1.14–4.42)	122	220	0.96 (0.70–1.31)
	Q3	16	220	1.24 (0.58–2.63)	168	220	1.32 (0.98–1.77)
	Q4	43	220	3.32 (1.74–6.35)	202	220	1.59 (1.19–2.12)
	Q5	80	220	6.18 (3.34–11.4)	254	220	1.99 (1.50–2.65)
	P			<0.0001			<0.0001
Menopausal characteristics (case and control)
Menopausal status		Pre-			Post-
	Q1	46	80	1.00 (ref.)	94	131	1.00 (ref.)
	Q2	42	70	1.04 (0.62–1.77)	106	139	1.06 (0.74–1.53)
	Q3	74	74	1.74 (1.07–2.83)	105	140	1.05 (0.73–1.51)
	Q4	78	71	1.91 (1.18–3.10)	161	138	1.63 (1.15–2.31)
	Q5	100	80	2.17 (1.36–3.47)	224	133	2.35 (1.67–3.30)
	P			<0.0001			<0.0001
Lifestyle characteristics (case and control)
Smoking status		Never			Ever
	Q1	72	114	1.00 (ref.)	69	107	1.00 (ref.)
	Q2	80	126	1.01 (0.67–1.51)	71	94	1.17 (0.76–1.80)
	Q3	99	127	1.23 (0.83–1.83)	85	93	1.42 (0.93–2.16)
	Q4	130	122	1.69 (1.15–2.48)	115	98	1.82 (1.21–2.73)
	Q5	190	116	2.59 (1.78–3.77)	144	104	2.15 (1.45–3.18)
	P			<0.0001			<0.0001
Alcohol drinking		Never			Ever
	Q1	48	67	1.00 (ref.)	93	154	1.00 (ref.)
	Q2	62	88	0.98 (0.60–1.61)	89	131	1.13 (0.78–1.63)
	Q3	74	79	1.31 (0.80–2.13)	110	141	1.29 (0.90–1.85)
	Q4	90	82	1.53 (0.95–2.47)	155	138	1.86 (1.32–2.63)
	Q5	123	86	2.00 (1.26–3.17)	211	134	2.61 (1.86–3.65)
	P			0.0015			<0.0001
Multivitamin use		No			Yes
	Q1	76	100	1.00 (ref.)	63	121	1.00 (ref.)
	Q2	74	119	0.82 (0.54–1.24)	74	98	1.45 (0.95–2.23)
	Q3	83	99	1.10 (0.73–1.67)	101	120	1.62 (1.08–2.42)
	Q4	126	93	1.78 (1.19–2.66)	113	123	1.76 (1.19–2.63)
	Q5	148	102	1.91 (1.29–2.82)	180	111	3.11 (2.12–4.58)
	P			<0.0001			<0.0001

ORs presented are age adjusted. All P values are for trend.

We also examined whether the association between LUMA methylation and breast cancer risk may be modified by menopausal status and several key modifiable lifestyle factors, including cigarette smoking, alcohol intake, and multivitamin supplement use. Stratified analyses did not reveal any significant variation of effects with respect to menopausal status (pre- vs. post-), alcohol (nondrinkers vs. drinkers), or smoking (never vs. ever) status (Table 2). For multivitamin supplement use, the positive association between LUMA methylation and breast cancer risk appeared to be more prominent among multivitamin supplement users (Q5 vs. Q1 OR: 3.11; 95% CI: 2.12–4.58) compared with nonusers (Q5 vs. Q1 OR: 1.99; 95% CI: 1.29–2.82), although the test for interaction did not reach statistical significance (P=0.11 for interaction).

LINE-1 promoter methylation was also measured in the same population. The mean level of LINE-1 methylation was 78.8% (range: 69.0–89.2%) among 1064 breast cancer case participants and 78.8% (range: 66.0–94.4%) among 1100 control participants. The methylation level did not differ by case-control status (P=0.94), nor was it associated with risk of developing breast cancer (Q5 vs. Q1: OR 1.00, 95% CI: 0.76–1.30). There was no correlation between LINE-1 and LUMA global methylation in peripheral blood DNA in the study population (r²=0.0004; P=0.33). No association between LINE-1 and breast cancer risk was observed (Supplemental Table S1A). Lack of association was also observed between LINE-1 and key demographic and clinical features (Supplemental Table S1B).

LUMA measures global promoter methylation levels

To validate that LUMA, measuring the C^mCGG target motif, overrepresents CpG islands and, in turn, approximates methylation levels of the gene promoters rather than overall CpG methylation content of the genome, we performed genome-wide methylation profiling of a subset (12 case/control pairs with extreme LUMA values) of DNAs from our study samples. We used the Infiniuim HumanMethylation27 BeadChip, which interrogates 27,578 CpG loci located within the proximal promoter regions of transcription start sites of 14,475 consensus coding sequence (24) in the U.S. National Center for Biotechnology Information Database (Genome Build 36). After filtering, signals from 27,173 CpG sites were analyzed. Among these, 19,851 CpG sites were located in the gene promoters containing CpG islands (CpGIPs), while the remaining 7322 were in the promoters without CpG islands (non-CpGIP). Consistent with our LUMA findings, the average methylation levels of all CpG sites was higher in cases than in controls (P=0.04), as well as for CpG sites located in CpG-island promoters (P=0.01; Fig. 2A). When we compared the promoter CpG methylation levels between the Q1 group (with lowest LUMA values) and Q5 group (with highest LUMA values), those with high LUMA values had higher CpG-island methylation; however, the association was significant only among controls (Fig. 2B). LINE-1 methylation was not correlated with any 27K BeadChip measurements, including overall CpG or CpGIP (data not shown).

Figure 2.

LUMA correlates with CpG island promoter methylation in the LIBCSP. Average β values were obtained after bead array methylation analysis using Illumina Infinium 27K in a subset of 12 cases and 12 controls with low (Q1) and high (Q5) LUMA values. A) Case-control comparison of methylation level of all CpG sites as well as CpG sites located in CpG island promoter. B) Comparison of methylation levels of CpG sites located in CpG island promoter between lowest quintile (Q1) and highest quintile (Q5) of LUMA values among cases and controls, respectively.

One-carbon metabolism may influence LUMA methylation

We explored whether genetic and dietary factors involved in 1-carbon metabolism may influence LUMA methylation levels in our control population. We first examined the potential influence of 13 functional polymorphisms in the 1-carbon metabolic pathway on LUMA methylation (Supplemental Table S2), and significant associations were observed for MTHFR C677T (rs1801133) and MTRR A66G (rs1801394) (Fig. 3). Compared with the MTHFR 677CC homozygotes genotype, individuals with 677TT genotype had significantly reduced levels of LUMA methylation (mean: 54.4 vs. 49.1%; P=0.0012). For the MTRR A66G polymorphism, the GG genotype was associated with higher LUMA levels compared with AA genotype (54.9 vs. 50.7%; P=0.018). When the potential influence of 1-carbon nutrients (including dietary intake of folate, choline, methionine, betaine, and vitamins B₂, B₆, and B₁₂) as well as alcohol intake, cigarette smoking, and multivitamin supplement use was examined, no significant associations were observed (Supplemental Table S3). When analyses similar to LUMA were carried out with LINE-1, no significant associations were observed (Supplemental Tables S2 and S3).

Figure 3.

Pattern of significant association between two 1-carbon metabolism SNPs on LUMA methylation levels among controls in the LIBCSP.

DISCUSSION

Aberrant DNA methylation has been shown to play important roles in breast cancer carcinogenesis (1). In this study, we demonstrated that LUMA methylation, a global measurement of promoter methylation, in peripheral blood DNA was strongly associated with the risk of developing breast cancer. Our study, which is based on data drawn from a large population-based sample, is the first to use LUMA methylation in the context of breast cancer in an epidemiologic setting. Notably, the risk associated with this marker was not influenced by known breast cancer risk factors, including family history, and menopausal status, nor did it vary by the hormone receptor status of the case participants' tumor. Our findings that methylation levels in surrogate tissues (blood) are positively associated with breast cancer risk are novel. Therefore, we performed thorough and vigorous quality control to rule out possible bias that may be associated with the case-control study design. We ruled out the possible influence of therapy on global methylation by restricting analyses to the pretreatment (i.e., chemotherapy, hormonal therapy, and radiation therapy) population and demonstrated that the positive association between LUMA methylation and breast cancer risk persisted.

While the methylation-cancer risk association is remarkable, the underlying biological mechanism is less clear. The principle question is how and why methylation status between target and surrogate tissue (e.g., blood) correlate or how the methylation status in surrogate tissue reflects the epignome or gene regulation in the target tissue. So far, the evidence for the correlation of blood-derived DNA methylation measurements with tissue-specific methylation is limited. It is plausible that methylation patterns in DNA obtained from blood may be more “plastic” compared with those of other tissues, due to proximity to environmental influences, such as nutrition via blood supply to the gut. Thus, the effects of environmental influences on DNA methylation patterns in blood may be short-term and exaggerated compared with those in target tissues. An animal study showed that the correlation between blood-derived and tissue-specific DNA methylation pattern was complex; on different nutrition intervention, blood-derived DNA methylation measurements did not always reflect methylation of other tissues (25). Further investigations are warranted to understand the correlation of mehylation patterns between target and surrogate tissues.

Nevertheless, there is great interest in identifying epigenetic markers in blood that can be used to identify high risk populations because blood-based biomarkers are easier to obtain and amendable to population screening. Studies have generated promising, although inconclusive, results. Different study designs and various sample sizes, together with the existence of multiple assays and lack of a gold standard for global methylation measurement, make comparisons among studies difficult, which has limited the potential utility of this epigenetic marker as a disease predictor. For example, one study investigated methylation of ALU repetitive elements in peripheral blood cells and found no correlation with breast cancer risk in 169 case and 180 control participants, while promoter hypermethylation of a panel of breast cancer related genes in the blood was associated with increased risk of breast cancer (26). Another hospital-based case-control study of ∼180 case and ∼180 control participants showed decreased deoxycytosine (5-^mC) as measured by mass spectroscopy in leukocyte DNA among breast cancer cases (7). In this same study, LINE-1 methylation using pyrosequencing was also measured in 37 individuals and no correlation was found either with the 5-^mC measurement or with breast cancer risk (7). Cho et al. (27) used MethyLight assay to measure 3 repetitive elements (LINE1, Sat2, and Alu) in white blood cells (WBCs) among 40 case and 40 control participants and found that only Sat2 level differs between cases and controls. They also examined WBC DNA promoter hypermethylation of 8 tumor suppressor genes but did not find any association with breast cancer risk. Brooks et al. (28) used quantitative methylation-specific PCR to assessed blood promoter methylation of 4 genes and found that the methylation status was unable to distinguish case and control groups. Several studies measuring overall WBC global DNA methylation in other cancer types including colon, bladder, stomach, and head and neck cancer have found an increased risk for cancer between those in the lowest quantile of global DNA methylation compared with those in the highest quantile of risk (29). A few existing studies investigating gene-specific methylation in WBCs and cancer risk have mostly focused on selected genes for breast cancer and colon cancer (29); data are still very limited and systematic, and unbiased approaches to the selection of genes have not been routinely employed. As methylation status is likely cell type specific, it would be ideal to assess the methylation status of individual cell populations. However, it is important to point out that peripheral blood samples from epidemiologic studies almost always consist of mixed cell populations (such as buffy coat), and fractionation of cell populations is simply not feasible or practical in an epidemiologic study setting.

Unlike LINE-1 or Alu methylation, LUMA methylation is not gene specific and does not reflect methylation states in nongenic, repetitive elements. The recognition sequence of the isoschizomer pair for LUMA, HpaII/MspI, is 5′-CCGG-3′; as a result, only CpG dinucleotides residing in the CCGG motif are assessed. Less than 1% of the genome is comprised of CpG islands and/or first exons, whereas >7% of the CpG dinucleotides are found in these regions. The genome appears to have methylated and unmethylated domains, and CpG islands and/or first exons comprise 34% of unmethylated domains (11, 30). LUMA analysis of CCGG methylation assays 8% of all CpG sites. Because of the additional C and G flanking the CpG, LUMA oversamples high GC content, unmethylated domains predominantly found in the promoter region of the genes. Our results from the whole-genome methylation profiling clearly support this notion. The HumanMethylation27 BeadChip interrogates 27,578 CpG loci located within the gene proximal promoter regions of transcription start sites. Consistent with LUMA findings, we observed significant genome-wide promoter hypermethylation in blood DNA from breast cancer cases. Furthermore, the association between LUMA hypermethylation and breast cancer risk corroborates the notion that the regulatory regions of the genome are usually hypermethylated in cancer cells. Nevertheless, it is important to note that we chose a small subset of samples with extreme LUMA values for functional validation using the HumanMethylation27 BeadChip. Such strategy is to maximize our ability to use a minimal number of samples to demonstrate the difference in promoter methylation potentially associated with LUMA. As expected, the average β values of these CpG sites located in CpGIP showed a clear difference between high (Q5) and low (Q1) values among controls (Fig. 2B). The absence of a significant difference in case participants may be the result of higher variability compared with the controls, which may reflect high variability of gene expression levels usually observed in tumor tissues. Methylation in blood samples obtained from case participants may also indicate a different stage of immune response/ignorance against a variety of breast cancer stages, grades, and histological subtypes (31).

There is a marked difference between LUMA ranges in different quartiles and mean methylation levels obtained by microarray analysis. These differences could be explained by the fact that the LUMA assay and Illumina 27K chip do not have fully overlapping CpG coverage. With the use of the HpaII/MspI isoschizomers, LUMA assays a wide range of CpG sites (exclusively in the context of GGCC sequence) throughout the genome. On the other hand, Illumina 27K microarrays are strongly focused on gene promoters. LUMA is biased to CpG islands, but 40% of CpG islands are outside of 5′ gene regions; whereas the chip is biased to gene promoters, but 30% of promoters do not have a CpG island. Therefore, many promoters that are assayed by the LUMA are not covered by the 27K chip and vice versa. Nevertheless, it is important to note that despite the lack of concordance between the magnitudes of methylation changes in different quartiles, the trends in methylation status are consistent between these two methods. The discrepancy between magnitude, but consistency in trend, suggests that while LUMA reflects promoter methylation, it also detects methylation outside promoters (not measured by 27K chip) such as repetitive elements, enhancers, intergenic, intragenic, and 3′-CpG islands, and they may control, for example, the expression of noncoding RNAs (32).

Epigenetic alterations do not alter nucleotide sequence; thus, they are potentially reversible (33). There is an urgent need to identify environmental or lifestyle factors that may result in epigenetic dysregulation or have the potential to restore it. One-carbon metabolism is a promising pathway to study these effects because it plays a key role in supplying 1-carbon units for methylation reactions (34). Genetic polymorphisms involved in 1-carbon metabolism have been studied with respect to measurements of global DNA hypomethylation (35, 36). A population-based gastric cancer study reported that possession of the MTHFR 677TT genotype was significantly associated with genomic DNA hypomethylation (37). Null results were reported for bladder cancer (6) and in normal colon tissue (38). In this study, we found that polymorphism in 2 key 1-carbon metabolic genes, MTHFR and MTRR, modified LUMA methylation levels (Supplemental Table S2). Methionine synthase reductase (MTRR) regenerates a functional methionine synthase via reductive methylation. A defect in reductive activation of methionine synthase would impair SAM synthesis and affect the methylation reaction (39). The finding that the MTHFR 677T allele associated with lower global methylation is consistent with previous findings. However, the downstream effect on breast cancer risk and its interpretation is less straightforward. MTHFR is involved in multiple processes (both DNA methylation and nucleotide synthesis) and its effects may be tissue specific. Indeed, functional effects of MTHFR 677T mutation have been shown to differ by cell types (colon vs. breast; ref. 40); under a sufficient folate supply, the 677T allele was associated with increased DNA methylation in colon cells and decreased methylation in a breast cells (consistent with our findings). In addition, the MTHFR 677T variant is associated with decreased uracil misincorporation in colon cancer cells but increased uracil misincorporation in breast cancer cells, which may explain the increased risk associated with breast cancer.

While a significant influence of the genetic polymorphisms of 1-carbon metabolizing genes was observed, the complete lack of influence of 1-carbon nutrients on LUMA methylation is surprising (Supplemental Table S3), even though the lack of dietary influence on 5-^mC content was previously reported in bladder and breast cancer studies (6, 7). One possible explanation is that the exposure assessment methods that are commonly used to ascertain usual dietary intake in large epidemiologic studies such as ours do not reflect intake over the life course (41). In contrast, functional polymorphisms of 1-carbon genes are likely to reflect lifelong methyl capacity in the body, and ultimately result in significant changes in the epigenome. Also important, the temporality of global methylation measurement in general, not only by LUMA but also by other methods, is unknown and needs to be systematically evaluated.

In summary, our results show that LUMA hypermethylation in peripheral blood is strongly associated with the risk of developing breast cancer. Despite the several merits of our study including a large population-based sample, we acknowledge the limitation of the observational design, which cannot assure causality. A prospective cohort design with blood samples collected before the development of disease and with serial LUMA measurements would enhance our ability to make stronger inferences regarding the relationship between global methylation status and breast cancer risk. Nevertheless, the consistency and strength of our study findings regarding the association with breast cancer offers promise that LUMA methylation measurements may ultimately contribute to improving our ability to prevent disease development.

Supplementary Material

This article includes supplemental data. Please visit http://www.fasebj.org to obtain this information.