Introduction

There are many well-established environmental and genetic factors associated with breast cancer risk but whether DNA methylation measured in blood is associated with breast cancer risk remains to be determined. The rationale for such an association is that blood DNA methylation levels could be a surrogate for breast tissue methylation [1,2,3,4], represent immunity/inflammation at the target tissue itself, indicate altered molecular pathways involved in carcinogenesis, or reflect past endogenous or exogenous exposures such as hormone levels [5, 6], alcohol [7], body mass index (BMI) [8, 9], smoking [10, 11], or ionizing radiation [12]. To better understand the possible relationship between blood DNA methylation and breast cancer risk, large, prospective studies with good exposure information and blood samples collected prior to diagnosis are needed.

Evidence for an association between blood-based DNA methylation and breast cancer risk is inconsistent [13]. A recent systematic review summarized the most promising findings of DNA methylation as a biomarker of disease [14], concluding that hypermethylation at BRCA1 and RASSF1A gene promoters, and hypermethylation of the body of the ATM gene were more common in breast cancer cases [1,2,3,4, 15,16,17,18,19,20,21]. An analysis published after the systematic review reported CpG sites cg06418238 in RPTOR, cg00736299 in MGRN1, and cg27466532 in RAPSN to be more commonly hypomethylated in breast cancer cases [22], although these associations were not replicated by the Melbourne Collaborative Cohort Study (MCCS) [23,24,25]. A nested case-control study (298 cases and 612 controls) within the Sister Study [26] reported that methylation level at 250 CpG sites of the 27 K methylation array was associated with breast cancer. Global methylation refers to the total level of 5-methylcytosine (5mC) content in a sample relative to total cytosine [27] and global hypomethylation has been related to genomic instability and cancer development [9, 28]. This global measure has also been investigated as a potential biomarker for breast cancer risk, producing inconsistent findings [14, 29]. This could be partly explained by the use of different technologies and proxies to measure global methylation [27]. Most recently, methylation arrays have been used to measure selected methylation sites across the genome, and some studies have analyzed the average methylation across these sites as a surrogate measure of global methylation. Using this measure of global methylation derived for the HumanMethylation450 BeadChip array, two prospective studies, the MCCS [23] and the Italian cohort of the European Prospective Investigation into Cancer and Nutrition (EPIC-Italy) [30], reported that lower global methylation levels in blood DNA were associated with a higher risk of breast cancer. However, this association was not observed by the Norwegian Women and Cancer (NOWAC) study [30].

In this study, we examined the association between DNA methylation in blood and breast cancer risk in the largest to-date study, involving 1663 incident invasive breast cancer cases and 1885 controls combined from four prospective cohort studies.

Methods

We combined results from four prospective studies: the MCCS [23], EPIC-Italy [30], the IARC cohort of the European Prospective Investigation into Cancer and Nutrition (EPIC-IARC) [31], and the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) [32]. Details of the four cohorts and the case-control studies nested therein can be found in Additional file 1: Supplemental Note.

Methylation array

Three of the cohorts (MCCS, EPIC-Italy, EPIC-IARC) measured DNA methylation using the Illumina Infinium HumanMethylation450 BeadChip (HM450K) array while PLCO used the MethylationEPIC BeadChip, which includes 90% of the sites in the HM450K array [33]. The four studies processed methylation data using the same protocol. Raw intensity data were imported into the R programming software using the minfi Bioconductor package [34]. Data were pre-processed for background subtraction and control normalization using the preprocessIllumina function. Subset-quantile within array normalization (SWAN) was performed to correct for type I and II probe bias [35]. Each study excluded samples for which more than 5% of CpG sites had a detection P value > 0.01. Also, each study excluded CpGs with a P value > 0.01 in more than 20% of the samples, and CpGs labeled as cross-reactive [36, 37], or located on the Y chromosome. The meta-analysis was performed on the 365,145 included CpGs common to the four studies. Beta values were used for analysis.

Statistical analysis

The four studies individually performed conditional (MCCS, EPIC-Italy, EPIC-IARC) or unconditional (PLCO) logistic regression to estimate the odds ratio (OR) and associated standard error (SE) per one standard deviation increase in methylation for each of the 365,145 CpG probes, adjusting where appropriate for the matching variables specific to each study (see Additional file 1: Supplemental Note), cell-type proportions estimated with the Houseman algorithm (percentage CD8T+, CD4T+, NK, B cell, monocytes, granulocytes) [38], and other variables to account for batch effects, such as plate or surrogate variable analysis (SVA) [39, 40]. These analyses within each study were repeated for subsets of individuals defined by the following case characteristics: age at diagnosis (< 50, ≥ 50 years old), estrogen receptor (ER) status (ER+/−), stage (early and late, see Additional file 1: Supplemental Note; associations with individual CpG probes were done in the PLCO only) and time since from blood collection (< 2 years, ≥ 2 years; < 5, 5–10, > 10 years). Estimates of the pooled ORs and their associated SE were calculated using fixed-effects meta-analysis. The Wald test was used to calculate P values. Statistical significance was defined by a false discovery rate (FDR)-adjusted P value, or q value, less than 0.05, which means that 5% of significant tests (i.e., tests with q < 0.05) were expected to be false positives. Heterogeneity in the odds ratio across studies or subgroups was assessed using the I2 statistic and the P value for the Cochran’s Q statistic.

To assess enrichment at specific genomic regions, we considered the average of the ORs for the CpGs located in those regions. We defined the average log (OR), βavg, by βavg = \( \frac{1}{\mathrm{K}}\sum {\beta}_k \) and approximated its standard error, SEavg, by SEavg = \( \frac{1}{K^2} \) wTΦθΦw, where K is the number of CpGs in the region of interest, βk and σk the estimated log (OR), and SE for probe k, θ the K × K correlation matrix of the K probes among controls, Φ the K × K diagonal matrix with the SEs along the diagonal and w a column vector of values equal to 1. We then report exp(βavg) as the average OR and calculate the 95% confidence interval under normality assumption and P values using the Wald test.

Each of the four studies also performed adjusted logistic regression to assess the association between a one standard deviation increase in global DNA methylation and the risk of breast cancer. Global methylation was defined for each study participant as the average methylation level across all 365,145 CpGs. In addition, we computed average measures in CpG islands, CpG shores, CpG shelves, and “open sea,” as well as at regulatory regions defined using the UCSC classification (promoter, gene body, 3′ UTR or intergenic) [41]. CpG sites were considered to be in a promoter region if they were located within 1500 bp of the transcription start site (TSS), in the 5′UTR or in the first exon [41].

All analyses were conducted using the R software, version 3.4.1.

Results

The characteristics of the 1663 cases and 1885 controls included in the meta-analysis are described in Table 1.

Table 1 Characteristics of the nested case-control studies included in the meta-analysis
Full size table

Associations between individual CpG sites and breast cancer risk

We found no evidence of association between blood DNA methylation measured at individual CpG sites and breast cancer risk (Fig. 1, q > 0.59) overall, or after stratification by age (< 50, > 50 years old), ER status (ER+/−), stage (early/late), or time since blood collection (< 5, 5–10, > 10 years) (data not shown). To account for possible subclinical effects from occult disease at the time of blood collection, we also investigated risk associations < 2 years versus ≥ 2 years since blood draw. We observed putative associations within 2 years of blood draw with cg00899463 (promoter of SGMS2, q < 10−8) and cg07145930 (promoter of G3BP1, q = 5.78 × 10−5), but there was substantial heterogeneity across studies for both CpGs (I2 = 100%), with each association driven by a single study.

Fig. 1
figure 1

Overall associations between methylation and breast cancer risk. QQ-plot (a) and a volcano plot (b) for the overall associations between methylation values at individual CpG sites and breast cancer risk

Full size image

We next focused on CpGs and genes reported to be associated with breast cancer risk in previous studies. Of the 250 CpGs reported by Xu et al. [26], 205 were included in our meta-analysis; we found no evidence of an association between these CpGs and breast cancer risk (q value > 0.17; q value calculation based on those 205 CpGs). Overall, methylation at the promoter region of BRCA1 was not associated with breast cancer risk (P = 0.88); of the 26 CpGs in this region that we examined, pooled ORs were consistent in direction with previous reports for 14, with cg20185525 having the largest OR (OR = 1.11, 95% CI = 0.99, 1.25). With respect to gene RASSF1A, 43 CpGs were in its promoter region but only 7 of them were associated with breast cancer risk in the direction previously reported. Overall, there was no association with the promoter region of RASSF1A overall (enrichment P = 0.061). Of the 9 CpG sites located in the body of ATM, only two were associated with breast cancer risk in the direction previously reported and there was no enrichment (P = 0.26). Previously identified CpGs in RPTOR (cg06418238), MGRN1 (cg00736299) and RAPSN (cg27466532) were not associated with breast cancer risk (cg06418238, P = 0.237, I2 = 0%; cg00736299, P = 0.779, I2 = 58%; cg27466532, P = 0.636, I2 = 0%).

Associations between global methylation and breast cancer risk

We found no evidence of an association between global methylation levels and breast cancer risk. We estimated an OR per one standard deviation increase in average methylation levels of 0.94 (95% CI 0.85, 1.05; P = 0.26; P for study heterogeneity = 0.86; I2 = 0%; Fig. 2). Neither adding previously published results from the NOWAC study (Additional file 1: Figure S1) nor stratifying by age at blood draw, ER status or time since blood draw qualitatively changed the result (Table 2). Global methylation was associated with breast cancer risk in women diagnosed with late-stage disease when analyses were not adjusted for cell type (OR = 0.83, 95% CI 0.71, 0.97), but not when cell types were included in the model (OR = 0.96, 95% CI 0.79, 1.13). Global methylation measures computed by the CpG region (island, shores, shelves, or open sea) were not associated with breast cancer risk. With respect to global methylation in CpG in promoters, a significant association was observed when the analysis was not adjusted for cell type, but it was not statistically significantly associated (P = 0.92) after adjustment (Table 2). Global methylation at other genomic regions, such as gene bodies, 3′UTR, or between genes, was not statistically significantly associated with breast cancer risk.

Fig. 2
figure 2

Average methylation and breast cancer risk. Forrest plot for the associations between average methylation and breast cancer risk when they are not adjusted (a) and adjusted (b) for cell type composition (CD8T+, CD4T+, NK, B cell, monocytes, granulocytes)

Full size image
Table 2 Stratified analysis of average methylation levels and breast cancer risk by several characteristics
Full size table

Discussion

No association between DNA methylation in blood and breast cancer risk was observed in this meta-analysis, either at individual sites or globally. Associations with CpG sites that had previously been reported to be differentially methylated between breast cancer cases and controls by smaller studies were not replicated by our larger study.

BRCA1 promoter methylation in blood has previously been reported to be higher in breast cancer patients compared with controls by some studies [1,2,3, 20], but not others [17, 18, 21]. In our study, we found no evidence of an association between BRCA1 methylation and breast cancer risk. It is important to note that blood collection in many of these previous studies was done at the time of cancer diagnosis, so DNA methylation could have been affected by the developing tumor; other studies included BRCA1 mutation carriers that were oversampled/represented in the cases but not in the controls. This differs from this meta-analysis in which all studies were prospective and did not select on family history (less than 20% of participants reported having a family history of breast cancer in the PLCO, which was the only one of the four cohorts with available information). When analyses were restricted to subjects with a family history of breast cancer (108 cases and 107 controls), no significant association was observed. In addition, the assays used in the aforementioned studies were mostly based on targeted regions, for which a single measurement of methylation for an entire genomic region is obtained, and not genome-wide array-based assays, as in our study, for which methylation measurements are performed at individual CpG sites. However, we conducted enrichment tests and still did not observe associations. Similar results were observed for RASSF1A, despite previously reported associations [22, 42,43,44].

A few studies have reported hypermethylation of repetitive elements within the body of ATM to be associated with breast cancer risk [4, 15]. None of the CpG sites in ATM were statistically significantly associated with breast cancer risk in our study. However, the HM450 assay used in our analysis is gene-centered, with a special focus on gene promoters, and has less coverage of other areas of the genome, particularly repetitive elements. Our analysis included only nine CpG sites in the body of ATM so that we could not thoroughly assess gene-body methylation.

Previous findings for cg06418238 in RPTOR, cg00736299 in MGRN1, and cg27466532 in RAPSN that were recently reported to be associated with breast cancer risk [22] were not replicated by our study. One of the cohorts included in the meta-analysis (MCCS) had previously published the lack of replication of the CpG sites [24], and this was confirmed by this meta-analysis.

A previous analysis of the Sister Study reported a methylation signature for breast cancer consisting of 250 differentially methylated CpG sites in blood identified in a prospective study of 298 cases and 612 cancer-free women [26]. We found no evidence of association with risk for the 205 CpG sites we could examine in our larger study. A number of differences between the two studies could potentially explain the discrepant findings. First, the Sister Study was enriched for family history, with all cases and controls being relatives of breast cancer cases, whereas the four cohorts included in the meta-analysis were sampled from the general population. Another difference is that the average time from blood draw to diagnosis for cases in the Sister Study was 1.3 years, suggesting that some of these women may have been developing breast cancer at the time of blood draw. The average time to diagnosis for the cases in our study was over 5 years. None of the sites reported by the Sister Study were associated with risk. We also performed a sensitivity analysis to rule out the influence of potential subclinical disease.

There was no association between global DNA methylation and breast cancer risk in our meta-analysis. Previously, two of the cohorts (MCCS and EPIC-Italy) included in our study reported significant associations with breast cancer risk [23, 30]. However, EPIC-Italy included additional case-control pairs for the present analysis (248 pairs compared to 162 pairs in the previous publication [30]), and in this study, the association was no longer significant. Only one of the cohorts (MCCS) included in the meta-analysis showed a significant association, similar to their previously result, but only when the analysis was not adjusted for cell-type composition. No evidence of association was observed individually in either of the two largest, unpublished studies included in this meta-analysis (EPIC-IARC and PLCO), as was the case in the NOWAC study [30].

Previous studies that assessed associations between global DNA methylation and breast cancer risk have reached mixed conclusions. Global methylation (percentage of 5-methyldeoxycytosine, i.e., %5-dmC, with respect to the total cytosine content) was measured using different methods, at different genomic locations (LINE-1, CmCGG common in promoters, etc.), and various populations and study designs (retrospective versus prospective) were used, which makes it difficult to compare results. Several studies have also suggested that blood global hypomethylation is associated with increased cancer risk in general [9, 45,46,47], not necessarily just breast cancer risk. The reasons for any such association are not clear, but it may represent changes in the regulation of methyl transfer reactions due to age or certain exposures (residual confounding), the presence of the cancer years before diagnosis, circulating tumor DNA, or a false positive.

Our analysis has several strengths. First, it is the largest evaluation to date of blood DNA methylation and breast cancer risk. The large sample allowed us not only to investigate new associations, but also to test associations previously reported by smaller studies. Another important aspect of our analysis is the prospective design of all the studies included, which avoids the influence of reverse causation (methylation changes due to the presence of cancer), as well as potential effects on blood methylation of treatment given prior to blood collection. Our analysis also has some limitations. First, we had limited statistical power to detect small effects of methylation on breast cancer risk, despite having a sample size which was larger than that of previous studies which reported significant associations. Another limitation is that we restricted the analysis to sites in the Illumina HM450 BeadChip, which accounts for less than 2% of the 28 million CpG sites in the genome [48]. This array is also gene-centered and particularly lacks coverage in intergenic regions. Therefore, our analysis may have missed genomic regions that could be relevant to breast cancer risk. Finally, DNA methylation is cell-specific, and therefore, average methylation levels in blood cells could reflect cell heterogeneity by case-control status. None of the included studies had collected information on direct blood cell types, so we used a standard algorithm to infer blood cell counts and account for this source of heterogeneity.

Conclusions

In summary, we found no evidence in this meta-analysis of four prospective cohort studies that DNA methylation measured in blood is associated with substantial variation in breast cancer risk. Larger studies may be required to determine if modest to weak associations exist between blood DNA methylation and breast cancer risk. Studies with greater methylation coverage of the genome may also have the potential to uncover novel associations.