Associations between genetic variation in one-carbon metabolism and LINE-1 DNA methylation in histologically normal breast tissues

Abstract

Genome-wide DNA hypomethylation is an early event in the carcinogenic process. Percent methylation of long interspersed nucleotide element-1 (LINE-1) is a biomarker of genome-wide methylation and is a potential biomarker for breast cancer. Understanding factors associated with percent LINE-1 DNA methylation in histologically normal tissues could provide insight into early stages of carcinogenesis. In a cross-sectional study of 121 healthy women with no prior history of cancer who underwent reduction mammoplasty, we examined associations between plasma and breast folate, genetic variation in one-carbon metabolism, and percent LINE-1 methylation using multivariable regression models (adjusting for race, oral contraceptive use, and alcohol use). Results are expressed as the ratio of LINE-1 methylation relative to that of the referent group, with the corresponding 95% confidence intervals (CI). We found no significant associations between plasma or breast folate and percent LINE-1 methylation. Variation in MTHFR, MTR, and MTRR were significantly associated with percent LINE-1 methylation. Variant allele carriers of MTHFR A1289C had 4% lower LINE-1 methylation (Ratio 0.96, 95% CI 0.93–0.98), while variant allele carriers of MTR A2756G (Ratio 1.03, 95% CI 1.01–1.06) and MTRR A66G (Ratio 1.03, 95% CI 1.01–1.06) had 3% higher LINE-1 methylation, compared to those carrying the more common genotypes of these SNPs. DNA methylation of LINE-1 elements in histologically normal breast tissues is influenced by polymorphisms in genes in the one-carbon metabolism pathway. Future studies are needed to investigate the sociodemographic, environmental and additional genetic determinants of DNA methylation in breast tissues and the impact on breast cancer susceptibility.

Introduction

Genome-wide DNA hypomethylation and gene promoter hypermethylation are early events in the carcinogenic process^1-5 and have been linked to chromosomal instability, gene reactivation, activation of proto-oncogenes, and higher frequencies of mutation and recombination events,^6-10 all of which are characteristics of cancer genomes, including cancers of the breast.^11-13

Long interspersed nucleotide element-1 (LINE-1) is a major constituent of repetitive transposable elements in the human genome.¹⁴ There are approximately 500,000 copies of LINE-1 across the human genome, occurring as hypermethylated sequences under physiologically normal conditions. Percent LINE-1 methylation has been used as a surrogate for genome-wide methylation^15,16 and has been studied as a potential biomarker for breast cancer risk.^17-19 Genome-wide DNA methylation (e.g., measured in repetitive sequences including, Alu, Sat2, and LINE-1) in white blood cells (WBCs) has been shown to be associated with age, gender, race, alcohol exposure, and family history of breast cancer.^20-24 Additionally, compared to histologically normal or normal-appearing tissues (adjacent to breast tumors), tumor tissues consistently have lower levels of DNA methylation (i.e., increased hypomethylation throughout the genome), even early in breast carcinogenesis.^4,5,25 Furthermore, there are reports of significant associations between LINE-1 methylation and breast tumor clinicopathological features,³ as well as breast cancer prognosis.^3,4

Current evidence regarding associations between variants in the folate-mediated one-carbon metabolism pathway and DNA methylation suggests that this relationship is a complex one, with evidence supporting a role of this pathway in breast cancer.^26,27 Disruption of this metabolic pathway, through alterations of the bioavailability of specific enzymes or variation in key genes, could lead to altered genome-wide or gene-specific methylation levels with differing consequences by genomic region.²⁸ There have been reports of an association between single nucleotide polymorphisms (SNPs) in genes encoding enzymes involved in one-carbon metabolism and gene-specific promoter methylation levels in some human cancers,^29,30 although findings for breast cancer have been inconclusive.^29,31 Little is known about the association between variation in one-carbon metabolism genes and genome-wide methylation. Further, while there is evidence that dietary folate intake may be associated with genome-wide methylation in WBCs,^32-34 it remains to be elucidated whether an association also exists for methylation levels in breast tissues with folate concentrations in the breast.

The study of histologically normal tissues from healthy women may reveal processes leading up to early carcinogenic events before a cancer or precancerous lesion is clinically detected. Few studies^{3, 35, 36} have examined LINE-1 methylation patterns in histologically normal breast tissues from healthy subjects, and it is unclear what factors and/or exposures are associated with breast tissue-specific LINE-1 methylation. Relevant risk factors that are known to be important in DNA methylation and one-carbon metabolism include folate concentrations and genetic variation in genes encoding the enzymes that play rate-limiting roles in the one-carbon metabolism pathway.

One study³⁶ examined the correlation between percent LINE-1 methylation in WBCs and matched histologically normal breast epithelial cells and reported no correlation. Similar findings were observed in a study²⁵ which examined the correlation between WBC LINE-1 methylation and LINE-1 methylation in normal-appearing tissues adjacent to breast tumors. These findings suggest that methylation patterns in WBCs may be imprecise surrogates of methylation patterns within the breast. In this study, focusing only on LINE-1 methylation within histologically normal breast tissues, we examined the associations between plasma and breast folate concentrations, genetic variation in one-carbon metabolism, and percent genome-wide DNA methylation (assessed by LINE-1 pryosequencing) in histologically normal breast tissues collected from a sample of women with no prior history of cancer who underwent reduction mammoplasty.

Results

Study participant characteristics and associations with percent LINE-1 methylation

Participant characteristics, with unadjusted associations between the variables of interest and percent LINE-1 DNA methylation, are shown in Table 1. The mean age of subjects was 35.7 y (SD = 12.5 years), 77.2% were premenopausal, the mean BMI was 32.3 kg/m² (SD = 7.4) kg/m²) and there were approximately equal proportions of White and non-White participants included in the study.

Table 1.

Participant characteristics associated with percent LINE-1 DNA methylation in histologically normal breast tissues, N = 121

Characteristic	n (%)	Mean LINE-1 methylation (%)a	Ratio (95% CI)b
Age (years)
16–26	37 (30.6)	73.6	1.00 (referent)
27–36	28 (23.1)	73.8	1.00 (0.97–1.03)
37–46	25 (20.7)	73.3	1.00 (0.70–1.03)
47–76	31 (25.6)	75.2	1.02 (0.99–1.05)
			P trend = 0 .42
Menopausal status
Premenopausal	88 (77.2)	74.0	1.00 (referent)
Postmenopausal	26 (22.8)	73.9	1.00 (0.97–1.03)
BMI (kg/m2), categorical
<25.0, under- and normal weight	18 (15.0)	74.2	1.00 (referent)
25.0–29.99, overweight	31 (25.8)	75.0	1.14 (1.01–1.29)
>30.0, obese	71 (59.2)	73.5	1.12 (0.99–1.26)
			P trend = 0 .29
Racec
White	61 (50.4)	74.9	1.00 (referent)
Non-white	60 (49.6)	73.1	0.98 (0.96–1.00)
Family history of breast cancer
No	81 (77.9)	73.7	1.00 (referent)
Yes	23 (22.1)	74.3	1.01 (0.98–1.04)
Age at menarche (years)
≤11	26 (27.1)	75.0	1.00 (referent)
12	31 (32.3)	73.5	0.98 (0.95–1.01)
13	24 (25.0)	74.2	0.99 (0.96–1.02)
≥14	15 (15.6)	72.9	0.97 (0.93–1.01)
			P trend = 0 .48
Parity
Nulliparous	38 (39.2)	75.0	1.00 (referent)
1–2 children	49 (50.5)	73.5	0.98 (0.96–1.01)
>2 children	10 (10.3)	72.7	0.97 (0.93–1.01)
			P trend = 0 .22
Age at first full-term birth (years)d
≤19	14 (30.4)	71.3	1.00 (referent)
20–24	16 (34.8)	75.1	1.05 (1.00–1.10)
≥25	16 (34.8)	72.9	1.02 (0.98–1.07)
			P trend = 0 .08
Duration of OC use (years)e
Never user	47 (43.9)	72.8	1.00 (referent)
≤5 years	29 (27.1)	74.3	1.02 (0.99–1.05)
>5 years	31 (29.0)	75.5	1.04 (1.01–1.07)
			P trend = 0 .03
Duration of alcohol use (years)f
Never drinker	38 (40.0)	72.9	1.00 (referent)
≤17 years	28 (29.5)	73.6	1.01 (0.98–1.04)
>17 years	29 (30.5)	75.6	1.04 (1.01–1.07)
			P trend = 0 .06
Smoking status
Never smoker	64 (64.6)	74.6	1.00 (referent)
Former smoker	15 (15.2)	74.5	1.00 (0.97–1.03)
Current smoker	20 (20.2)	72.3	0.97 (0.94–1.00)
			P trend = 0 .14

Abbreviations: BMI, body mass index; OC, oral contraceptive

^aPercent LINE-1 DNA methylation natural log transformed for normality, back transformed data are presented.

^bRatios (95% CI) are given as e^β (e^{β±1 .96(SE)}) of percent LINE-1 DNA methylation compared to referent group, unadjusted.

^cRace determined by self-report. Non-White category is composed of those self-identifying as Black (57/60 [95%]), Hispanic (2/60 [3.3%]), and other (1/60 [1.7%]).

^dAmong parous women only.

^eMedian duration of OC among ever users was 5 y

^fMedian duration of alcohol use among ever drinkers was 17 y

Associations between percent LINE-1 DNA methylation and folate biomarkers

Unadjusted percent LINE-1 methylation was not significantly associated with either plasma (r = −0.15, P = 0.20) or breast tissue (r = −0.07, P = 0 .44) folate concentrations. Upon stratification by BMI (≥27.0 kg/m² vs. <27.0 kg/m²), we found that among women with BMI ≥27.0 kg/m², percent LINE-1 methylation was not significantly associated with either plasma folate (r = −0.04, P = 0.79) or breast folate (r = −0.06, P = 0.62); while among women with BMI <27.0 kg/m², the correlation between plasma folate and percent LINE-1 methylation was larger in magnitude and approached statistical significance (r = −0.40, P = 0.07). The correlation between LINE-1 methylation and breast folate was unchanged by BMI stratification (r = −0.09, P = 0.62). Table 2 shows that there is no significant multivariable-adjusted association between plasma folate and percent LINE-1 methylation; however, there was some (borderline) evidence of a dose-response trend for percent LINE-1 methylation with increasing breast tissue folate (P trend = 0.05).

Table 2.

Associations between plasma and breast folate concentrations and percent LINE-1 DNA methylation in histologically normal breast tissues

Biomarker	n (%)	Mean LINE-1methylation (%)a	UnadjustedRatio (95% CI)b	Multivariable-adjustedRatio (95% CI)c
Plasma folate (ng/mL)
Tertile 1	25 (32.4)	75.4	1.00 (referent)	1.00 (referent)
Tertile 2	26 (33.8)	74.8	0.99 (0.96–1.03)	0.99 (0.95–1.03)
Tertile 3	26 (33.8)	73.9	0.98 (0.95–1.01)	0.98 (0.94–1.02)
			P trend = 0 .53	P trend = 0 .74
Breast folate (ng/mg)
Tertile 1	37 (32.8)	73.8	1.00 (referent)	1.00 (referent)
Tertile 2	38 (33.6)	74.3	1.01 (0.98–1.03)	1.04 (1.01–1.07)
Tertile 3	38 (33.6)	73.6	1.00 (0.97–1.03)	1.01 (0.98–1.04)
			P trend = 0 .84	P trend = 0 .05

^aPercent LINE-1 DNA methylation natural log transformed for normality, back transformed data are presented.

^bRatios (95% CI) are given as e^β(e^{β±1 .96(SE)}) of percent LINE-1 DNA methylation compared to the referent group, unadjusted.

^cRatios (95% CI) are given as e^β(e^{β±1 .96(SE)}) of percent LINE-1 DNA methylation compared to the referent group, adjusted for (age, race, history of oral contraceptive use, and history of alcohol use).

Associations between LINE-1 DNA methylation and variation in selected one-carbon metabolism SNPs

Multivariable-adjusted associations between common, functional variants in one-carbon metabolism genes and percent LINE-1 methylation are shown in Table 3. The variant C allele of the MTHFR A1298C polymorphism was significantly associated with lower LINE-1 methylation. Compared to women with the common AA genotype of MTHFR A1298C, those carrying at least one variant C allele had 4% lower methylation (AC/CC genotypes: Ratio 0.96, 95% CI 0.93–0.98). Log-additive models of percent LINE-1 methylation were suggestive of an inverse dose-response trend, with decreasing methylation for each additional variant G allele (P trend = 0.06). The variant G alleles of the MTR A2756G and MTRR A66G polymorphisms were significantly associated with higher levels of LINE-1 methylation (MTR, AG/GG genotypes: Ratio 1.03, 95% CI 1.01–1.06; MTRR, AG/GG genotypes: Ratio 1.03, 95% CI 1.01–1.06). In log-additive models, there was some evidence of a dose-response trend for increasing percent LINE-1 methylation with each additional variant allele of MTR A2756G (P trend = 0.05). We observed no evidence of such a trend for MTRR A66G (P trend = 0.23).

Table 3.

Associations between selected one-carbon metabolism SNPs and LINE-1 DNA methylation in histologically normal breast tissues

Polymorphism	n (%)	Mean LINE-1 methylation (%)a	UnadjustedRatio (95% CI)b	Multivariable-adjustedRatio (95% CI)c
MTHFR C677T (rs1801133)
CC	71 (58.7)	73.8	1.00 (referent)	1.00 (referent)
CT/TT	50 (41.3)	74.2	1.01 (0.98–1.03)	0.99 (0.96–1.01)
MTHFR A1298C (rs1801131)
AA	73 (61.9)	74.3	1.00 (referent)	1.00 (referent)
AC/CC	45 (38.1)	73.4	0.99 (0.96–1.01)	0.96 (0.93–0.98)**
MTR A2756G (rs1805087)
AA	70 (59.8)	73.1	1.00 (referent)	1.00 (referent)
AG/GG	47 (40.2)	75.4	1.03 (1.01–1.05)**	1.03 (1.01–1.06)*
MTRR A66G (rs1801394)
AA	34 (29.1)	72.8	1.00 (referent)	1.00 (referent)
AG/GG	83 (70.9)	74.6	1.03 (1.00–1.05)*	1.03 (1.01–1.06)*
MTHFD1 R134K (rs1950902)
AA	34 (29.3)	74.1	1.00 (referent)	1.00 (referent)
AG/GG	82 (70.7)	73.7	1.00 (0.97–1.02)	1.00 (0.97–1.03)
FTHFD T/C (rs2276731)
AA	65 (54.2)	74.4	1.00 (referent)	1.00 (referent)
AG/GG	55 (45.8)	73.5	0.99 (0.97–1.01)	1.00 (0.97–1.02)
FTHFD T/C (rs2002287)
AA	58 (50.0)	73.7	1.00 (referent)	1.00 (referent)
AG/GG	58 (50.0)	74.5	1.01 (0.99–1.03)	1.00 (0.97–1.02)

^aPercent LINE-1 DNA methylation natural log transformed for normality, back transformed data are presented.

^bRatios (95% CI) are given as e^β(e^{β±1 .96(SE)}) of percent LINE-1 DNA methylation compared to common genotype, unadjusted.

^cRatios (95% CI) are given as e^β(e^{β±1 .96(SE)}) of percent LINE-1 DNA methylation compared to common genotype, adjusted for (race, history of oral contraceptive use, and history of alcohol use).

^*P<0.05;

^**P <0.01

Further adjustment for breast folate did not considerably change the associations between genotypes of MTHFR C677T, MTHFR A1298C, MTHFD1 R134K or either of the FTHFD SNPs. However, upon adjusting for breast folate, we found that associations between percent LINE-1 methylation and the variant alleles of MTR A2756 and MTRR A66G were attenuated (MTR, AG/GG genotypes: Ratio 1.02, 95% CI 1.00–1.05; MTRR, AG/GG genotypes: Ratio 1.03, 95% CI 1.00–1.06). In a mutually adjusted model of all one-carbon SNPs studied here, we found that genotypes of these polymorphisms explained only 8% of the variation in percent LINE-1 methylation (adjusted R² = 0.08, P = 0.03). When we included race, history of oral contraceptive use and history of alcohol use to the model, the combination of SNP genotypes and demographic and lifestyle exposure variables explained an additional 17% of LINE-1 methylation variation (adjusted R² = 0.25, P = 0.002).

Given that factors shown to be associated with percent LINE-1 methylation (in Table 1) could be due to associations with BMI, we further adjusted these multivariable models for BMI and found that the ORs and 95% CIs were unchanged, suggesting that BMI did not confound the reported associations.

Discussion

DNA methylation homeostasis is essential for regulation of gene expression and maintenance of genomic integrity in normal mammalian cells. Individuals undergo extensive changes in DNA methylation patterns during their lifespan, in accordance with the changes in their tissue-specific gene expression patterns related to aging and other physiologic states.³⁷ It is well established that alterations in DNA methylation represent early events in carcinogenesis,^1-8,24 characterized by genome-wide hypomethylation, especially at repetitive elements throughout the genome, such as LINE-1, and gene-specific promoter hypermethylation, which affects expression of tumor suppressor genes.^{1-6, 25,35, 38} Histologically normal tissues are known to maintain a high percent methylation of LINE-1 (ranging from approximately 7585%–). However, little is known about the factors that are associated with this biomarker of genome-wide DNA methylation in healthy breast tissues. In this study, lower breast tissue LINE-1 methylation was observed among variant allele carriers (AC/CC) of the MTHFR A1298C SNP, while higher LINE-1 methylation was observed in the breast tissues of variant allele carriers (AG/GG) of the MTR A2756G and MTRR A66G SNPs.

The observation that carriers of at least one variant allele of MTHFR A1298C had lower breast LINE-1 methylation than those carrying the common genotype has not been previously reported. In contrast, evidence from analysis of the association between the MTHFR A1298C SNP and genome-wide methylation in various sample types, indicates a null association.^17,39-42 Although evidence from a recent meta-analysis also indicates no association between the MTHFR A1298C SNP and breast cancer risk,⁴³ this SNP has been found to be associated with both folate and homocysteine concentrations;^44,45 and the variant homozygous allele, specifically, is associated with lower enzymatic activity compared to the common genotype.⁴⁶ MTHFR is an important enzyme in the one-carbon metabolism pathway and the most frequently studied SNPs in this gene are C677T and A1298C, both reducing the enzyme activity in humans, albeit at different levels.⁴⁶ The C677T SNP is located in the N-terminal catalytic domain and is responsible for the reduction of 75% of the enzyme activity in homozygous individuals as demonstrated by in vivo assays in human lymphocytes, in vitro assays using recombinant proteins and by bioinformatics predictions.^46-48 The A1298C SNP is situated in the C-terminal catalytic domain binding S-adenosylmethionine (SAM), the allosteric inhibitor that negatively regulates MTHFR activity in response to methionine levels, and also acts as the main methyl donor for DNA methylation.⁴⁹ This SNP was associated with a moderate reduction of MTHFR activity (by 39% in human lymphocytes), however this was not confirmed by in vitro studies, suggesting in fact that it does not directly impact the catalytic activity of the enzyme, but rather the ability of SAM to exercise its regulatory function.^46,47 The fact that we observed herein an association only between A1298C and methylation may indicate that feedback control mechanisms in the one-carbon metabolism pathway are more critical for DNA methylation homeostasis. Additional studies are warranted to clarify whether an association between MTHFR A1298C genotype and breast tissue-specific genome-wide methylation exists, and if this association impacts breast cancer risk.

In this study, significantly higher breast tissue LINE-1 methylation among women who were carriers of at least one variant allele of MTR A2756G, compared to the common genotype was observed. This finding and data from a recent case-control study⁵⁰ of breast cancer risk in White and Black women, which showed significantly reduced odds of breast cancer among carriers of 2 variant G alleles (GG vs. AA/AG: OR 0.44, 95% CI 0.24–0.80), support a protective role of variant alleles of MTR A2756G. Methyltetrahydrofolate-homocysteine methyltransferase (MTR) is an important enzyme that plays a role in the biosynthesis of 5-methyltetrahydrofolate and methionine, which are both essential for DNA methylation reactions.⁵¹ In addition, MTR affects the availability of tetrahydrofolate for nucleotide biosynthesis.⁵¹ In one of our earlier analyses in a breast cancer case-control study,³¹ we observed no significant association between MTR A2756G and hypermethylation in select gene promoters (measured in breast tumor tissues), including E-cadherin, p16^INK4a, and RAR-β2. We⁵² and others⁵³ have similarly found no association between variation in this SNP and percent methylation in the promoter regions of some specific genes, in histologically normal breast tissues or in breast tumor tissues.⁵³ However, no study has investigated the association between MTR A2756G and percent LINE-1 methylation in histologically normal breast tissues.

Significantly higher percent LINE-1 methylation in the breasts of women carrying at least one variant allele of the MTRR A66G polymorphism, compared to those carrying the common genotype was also observed. There is evidence that the MTRR A66G variant leads to an amino acid substitution in the 5-methyltetrahydrofolate-homocysteine methyltransferase reductase (MTRR) enzyme, which reduces enzymatic activity, thereby reducing DNA methylation.^54,55 In this study, we investigated genome-wide methylation in histologically normal breast tissues only, where methylation of repetitive elements, such as LINE-1, was maintained at high levels with a fairly small range of percent methylation (65.3–82.8%). On the other hand, LINE-1 methylation in tumor tissues varies considerably, with marked demethylation (average demethylation was of approximately 15% compared to normal tissues).⁵⁶ The fact that there were higher levels of LINE-1 methylation in normal breast tissues among variant allele carriers of the MTR A2756G and MTRR A66G SNPs support the notion that methylation patterns within tissues are genetically regulated, at least in part. Furthermore, these findings support a suggestive protective effect in breast cancer associated with carrying the variant G alleles of these SNPs (as observed among some sub-groups of women),^57-59 through favorable modification of genome-wide methylation patterns (i.e., increasing LINE-1 methylation within the breast). Understanding the role of genome-wide methylation patterns, specifically the balance of hypermethylation and hypomethylation, within tissues of interest would be of importance in understanding the relationship between these epigenetic phenomena, the associated genetic variation, and cancer risk.

This study has some strengths and yielded novel data. The significant associations between genetic variation and genome-wide methylation underscores the potential value of examining tissue-specific methylation patterns in the context of understanding cancer risk. Analysis of histologically normal tissues from a well-described sample of women with no history of breast cancer allowed us to examine biomarkers related to events preceding the appearance of neoplastic transformation in the target organ. This study also had limitations that should be considered in interpretation of these findings. First, this study may suffer from selection bias, which cannot be assessed given that there are no data about women who did not participate in this study. The small sample size was also a limitation. In addition, there may be limitations on the generalizability of the study's findings; women undergoing the reduction mammoplasty procedure may differ from the general population of women in some of their characteristics, particularly BMI. However, our additional analysis (further adjusting for BMI in multivariable models) demonstrated that BMI was not a confounder of the observed associations. Also, although methylation of LINE-1 repetitive elements has been shown to provide results that parallel those obtained from a direct measurement of genomic DNA methylation,⁶⁰ LINE-1 methylation has its limitations as a surrogate measure of genomic methylation and we are not certain that there is a definitive association between percent LINE-1 methylation (in WBCs or breast tissues) and breast cancer risk.⁶¹ Another consideration is that our finding of a lack of a significant association between plasma or breast folate concentrations (which we recently showed are only modestly correlated [r = 0.21, P = 0.07]⁵²) and LINE-1 methylation could be due to measurement error, particularly for the breast folate assay which had a CV >10%. Finally, since there are no known clinically relevant cut-points/thresholds for LINE-1 methylation that are related to breast cancer risk, our findings of 1–4% differences in normal tissues (in association with some factors studied here) might be clinically relevant. Therefore, our findings need to be interpreted with caution and need to be confirmed in larger studies.

In summary, we investigated associations of factors involved in the folate-mediated one-carbon metabolism pathway with genome-wide DNA methylation, as measured by methylation patterns of LINE-1 repetitive sequences, within histologically normal breast tissues. Although we observed no association between plasma or breast folate concentrations and percent LINE-1 methylation, we observed significant associations for variation in one-carbon metabolism (MTHFR A1298C, MTR A2756G, and MTRR A66G). Our findings indicate that DNA methylation of LINE-1 elements in histologically normal breast tissues is influenced by genetic factors and suggest a potential genetic contribution to genome-wide DNA methylation in breast carcinogenesis; future studies using additional biomarkers of genome-wide methylation are clearly warranted to investigate the social, demographic, environmental and additional genetic determinants of DNA methylation in breast tissues and the impact on breast cancer susceptibility.

Materials and Methods

Study sample and biospecimens

Detailed methods of this study sample are available elsewhere.^{52, 62} Briefly, 121 women undergoing reduction mammoplasty, age ≥16 , with no prior history of cancer, were enrolled in a cross-sectional study. Both blood and breast tissue specimens were collected and participants completed a detailed interview. All participants provided written informed consent and the study was approved by the Institutional Review Boards of all participating institutions.

Plasma samples were collected within 24 hours prior to surgery and stored at −80°C. Resected remnant breast tissues were grossly dissected to separate epithelial tissues from adipose tissues, and snap frozen in liquid nitrogen within one hour of surgery and stored at −80°C. Breast tissues were pathologically reviewed and samples exhibiting any breast abnormalities, including benign breast disease, were excluded.

LINE-1 DNA methylation analysis

Genomic DNA was isolated from breast tissues using the Puregene DNA purification kit (Gentra Systems), according to the manufacturer's instructions. Aliquots of 500 ng of DNA were treated with sodium bisulfite in order to convert all non-methylated cytosine residues to uracil, using the EZ DNA methylation kit (Zymo Research). Bisulfite modified DNA was then subjected to PCR amplification using the primers and conditions provided by the manufacturer in the PyroMark LINE-1 kit (Biotage AB, now Qiagen), with one of the primers being biotinylated. PCR products were attached to streptavidin coated sepharose beads, then denatured with NaOH in order to become single stranded and purified by several steps of washing with the aid of a Pyrosequencing Vacuum Prep Tool (Pyrosequencing, Inc..), according the manufacturer's protocol. Single stranded PCR products were then used as the template for pyrosequencing in a PyroMark MD instrument (Biotage AB). The methylation percentages at 4 CpG sites were calculated using the Pyro Q-CpG v.1.0.9 software (Biotage AB) and the average was used as a measure of LINE-1 methylation. For quality control, 10% of samples were assayed in duplicate along with commercially-available methylated and unmethylated control DNA (EpiTect Control DNA, Qiagen). Technical personnel performing the assays were blinded to sample characteristics. The mean coefficient of variation (CV) among blinded replicates was 1.38%.

Plasma and breast folate analysis

Plasma folate concentrations were quantified by commercially-available immunoassays on the Immulite 1000 (Siemens Healthcare), as previously reported.⁵² Sample batches included commercial and blinded plasma control samples to assess laboratory variation. The CV for this assay was 3.1%. For quantification of breast folate, a microbiological microtiter plate assay using Lactobacillus casei was used as previously described.⁶³ Briefly, 10–20 mg of breast tissues were homogenized with folate extraction buffer (2% sodium ascorbate, 2% Bis-Tris, and 0.07% 2-mercaptoethanol). The mixtures were immersed in boiling water (20 minutes), cooled on ice, and then centrifuged at 36,000g (20 minutes; 4°C). Samples were incubated with dialyzed chicken pancreas conjugase (2 hours; 37°C), put into 96-well plates with serial dilutions, and incubated with L. casei (24 hours). Plates were then read at 595 nm. Breast folate concentrations were calculated based on an internal folic acid standard curve, were normalized by total protein concentration, and are reported in nanograms of folate per milligram of protein (ng/mg of protein). The CV for the breast folate assay was 12%.

DNA isolation and genotype analysis of one-carbon metabolism genes

Genomic DNA was isolated from buffy coats using the DNAQuik™ isolation kit (BioServe). Genotyping of SNPs in the genes encoding thymidylate synthase (TYMS T/C rs502396), methyltransferase reductase (MTRR A66G rs1801394), methylenetetrahydrofolate dehydrogenase (MTHFD1 R134K rs1950902), and formyltetrahydrofolate dehydrogenase (FTHFD T/C rs2276731 and rs2002287) was performed using a 96-plex Illumina BeadXpress® Assay (Illumina, Inc.). Separately, allelic discrimination by real time PCR with TaqMan probes (Applied Biosystems, Foster City, CA) was used for genotyping SNPs in the genes encoding methylenetetrahydrofolate reductase (MTHFR C677T rs1801133 and MTHFR A1298C rs1801131) and methyltetrahydrofolate-homocysteine methyltransferase (MTR A2756G rs1805087) using primers, probes and conditions as described on the NCI's Cancer Genome Anatomy Project SNP500 Cancer Database (http://snp500cancer.nci.nih.gov/). All genotype frequencies were in Hardy-Weinberg equilibrium.

Statistical analysis

Percent LINE-1 methylation was log-transformed for normality, although back-transformed (i.e., geometric) means are presented in the tables for ease of interpretation. Means (and standard deviations) and frequencies (and percentages) were used to assess the distributions of percent LINE-1 methylation in histologically normal breast tissues by selected participant characteristics. Associations between selected characteristics (exposure variables), treated as categorical, and percent LINE-1 methylation (outcome variable) were examined using multiple regression models. Results are expressed as e^β, representing the ratio of LINE-1 methylation among women in each corresponding category relative to that of the referent group, and e^β±1.96(SE), the corresponding 95% confidence intervals (CI). The same statistical method was used to examine associations between tertiles of plasma and breast folate concentrations and percent LINE-1 methylation. Pearson correlation coefficients were used to describe correlations between percent LINE-1 methylation and plasma and breast folate concentrations.

The chi-square test with 2 degrees of freedom was used to determine Hardy-Weinberg equilibrium of genotype frequencies of the MTHFR, MTR, MTRR, MTHFD1, and FTHFD SNPs. Genotype analyses were performed under a dominant model: women who were carriers of ≥1 variant alleles were compared to women who carried both common type alleles (for each SNP). Similar to the analyses described above, associations between SNP genotype and percent LINE-1 methylation are given as the ratio of LINE-1 methylation among those carrying variant genotypes relative to those carrying both common alleles for each SNP. Participant characteristics found to be associated with percent LINE-1 methylation in unadjusted analyses (see Table 1), were included for adjustment in regression models examining the associations between plasma and breast folate, one-carbon metabolism SNPs, and LINE-1 DNA methylation. All reported P-values are 2-sided and P < 0.05 was considered statistically significant. Analyses were performed using SAS version 9.4 (SAS Institute).

Funding

This work was supported by the Department of Defense, Alcohol Center of Excellence (Grant Number DOD BC022346). The Genomics & Epigenomics Shared Resource is partially supported by NIH/NCI grant P30 CA051008.

Disclosure of Potential Conflicts of Interest

No potential conflict of interest was disclosed.

Acknowledgments

We thank the Genomics and Epigenomics Shared Resource of the Lombardi Comprehensive Cancer Center at Georgetown University for their contributions to SNP genotyping for this study. We also wish to thank the repository team from Dr. Shields' laboratory that helped obtain and process the specimens stored in the biorepository system, including Leoni Leonidaridis, David Goerlitz, and Jeena Mathew.