Differences in Genome-wide DNA Methylation Profiles in Breast Milk by Race and Lactation Duration

Black women are disproportionately affected by early-onset, highly aggressive breast cancer, more specifically triple-negative breast cancer (TNBC; ref. 1). Black women also consistently have the highest breast cancer–related mortality rates compared with other races (1, 2). It is not well-understood why black women have higher incidence rates of TNBC; however, differences in certain exposures, such as obesity or age at first live birth, may have differential effects on risk for different subtypes of breast cancer, and may contribute to the racial disparities identified in breast cancer (3). Breastfeeding can also lower a woman's risk of breast cancer, and black women tend to breastfeed at lower rates compared with white and Hispanic women (4, 5). A better understanding of how race and lactation duration affect breast cancer risk among healthy women is greatly needed. A molecular understanding of how known breast cancer risk factors may influence the healthy breast microenvironment and potentially influence breast cancer development could provide important etiologic insights.

DNA methylation, the addition of a methyl group on a cytosine, can affect gene expression levels, and is thought to be linked to tumor development in the breast and other sites (6, 7). DNA methylation is reversible, making it an ideal target for cancer prevention (8, 9). DNA hypermethylation in promoter regions of tumor-suppressor genes may silence these genes in cancer, while DNA hypomethylation might increase oncogene expression (8, 10). It is hypothesized that this modification occurs early in tumor development and could even affect tumor phenotypes and prognosis (10).

Previous work in healthy individuals has shown associations between tissue and blood DNA methylation and various clinical covariates, such as race, age, body mass index (BMI), alcohol consumption, and environmental exposures (11–14). Researchers have also observed race-specific differences in DNA methylation at specific genes in healthy individuals that are also associated with various cancers, including breast, colorectal, pancreatic, and prostate cancers, all of which experience racial disparities in incidence or mortality (8, 11, 15, 16). Because DNA methylation patterns are tissue- and site-specific, acquiring tissue specimens may require burdensome and invasive techniques, especially for healthy individuals.

Breast milk is one noninvasive specimen that represents the breast environment (17). Breast milk contains epithelial cells, leucocytes, cytokines, proteins, and hormones, each of which may be targeted and interrogated for understanding breast cancer development (17, 18). Much of the work performed in breast milk has focused on better understanding its health benefits for newborn children (19, 20), with more recent findings correlating breast milk microbiome with maternal weight (21). We and others have demonstrated the feasibility of measuring DNA methylation in breast milk (22–24) and identified associations between promoter DNA methylation levels and age (15).

The goal of this study was to identify differences in genome-wide DNA methylation levels in breast milk of healthy lactating women by race and other breast cancer risk factors during a unique time in their reproductive life cycle (postpartum) that is known to affect breast cancer risk. Identifying epigenetic changes in breast milk may enable identification of biomarkers that are associated with breast cancer risk and that mediate risk factors and protective factors, including breastfeeding itself.

Lactating women (age 18 years and older) were recruited through national and local media to provide breast milk samples (∼100 mL of pumped or hand-expressed breast milk), a health and lifestyle questionnaire, and written consent for research to the Breastmilk Laboratory at the University of Massachusetts at Amherst. Participants included 83 white and 61 black uniparous women who had not undergone a breast biopsy, were cancer-free at the time of donation, donated breast milk between 2006 and 2014, and had complete information from the questionnaire on pregnancy-related variables as well as smoking and BMI.

Women who lived or visited within 100 miles of Amherst, MA had their breast milk samples and completed questionnaires collected by a researcher at home who immediately delivered the specimens to the laboratory at ambient temperature for processing. For women living outside of 100 miles of Amherst, MA, breast milk samples were shipped with an ice pack via a prepaid UPS breast milk collection kit (17). A total of 52% of milk samples were expressed in the morning (between 5 am and 11:59 am), and 20% were expressed at other times throughout the day. The remaining 28% of samples had no expression time data. All milk samples were shipped on the same day that they were expressed.

Covariate data were obtained from paper questionnaires completed at the time of donation and included questions about reproductive health (i.e., parity, breastfeeding history, and oral contraceptive use), general health (i.e., previous cancer diagnosis and subsequent cancer treatment, history of breast biopsy, prescription medication use, over-the-counter pain reliever use, over-the-counter vitamin or supplement use, and recent cold of flu symptoms), demographic information (i.e., smoking status, current age, race, ethnicity, occupation, income, current residence, current height and weight, general diet information, and general physical activity information), and family history of breast or ovarian cancer. Variables were selected a priori based on completeness as there were large amounts of missing data for some variables, including general diet information, as well as occupation and income. Lactation duration was defined as the age of the current baby breastfed in days. This study was conducted in accordance with recognized ethnical guidelines (U.S. Common Rule) and approved by the Institutional Review Boards of University of Massachusetts at Amherst (Amherst, MA) and at the NIH.

DNA was extracted from milk samples using the phenol-chloroform method as described previously (15). One mL from each milk sample was put into a 2.0 mL tube. Lysis buffer (100 mmol/L Tris, 1 mmol/L EDTA, 0.5% SDS, 200 mmol/L NaCl, and 5 μL of 200 ng/mL Proteinase K) was added to each tube. All tubes were then placed in a 56°C water bath overnight. After an additional 2-hour incubation at 56°C with 6 μl extra 200 ng/mL proteinase K, each lysed sample was divided into two aliquots of 622 μL each. An equal volume of phenol/chloroform/isoamyl alcohol (25:24:1) was added to each tube. Samples were vortexed vigorously for 30 seconds and centrifuged at 15,000 × g for 10 minutes. The aqueous phase was transferred to two new tubes and an equal amount of chloroform/isoamyl alcohol (Sigma) was added. Samples were vortexed vigorously for 30 seconds and centrifuged at 15,000 × g for 10 minutes. The aqueous phase (1,200 μL) was transferred to three new tubes (400 μL in each tube) and 40 μL (0.1 volume) of 3 mol/L sodium acetate, 1 mL (2.5 volumes) of ice-cold ethanol, and 1 μL of glycol blue were added to each tube. This was left overnight at −20°C. The precipitate for one tube of each sample was then spun at 15,000 × g for 10 minutes. The supernatant from the first tube of each sample was discarded and the precipitation mix of a second aliquot was added to this tube. Centrifugation was repeated. Supernatant of the second tube was disposed then the third aliquot was added, centrifuged once more, and the supernatant removed. Once the final supernatant was removed, the pellet was washed with 70% ethanol, dried, and eluted in 22 μL of elution buffer (Qiagen).

Genomic DNA was quantified using a real-time TaqMan PCR assay targeting ALU repetitive elements. The forward primer sequence was: 5′-ATC ACG AGG TCA GGA GAT CGA G-3′; the reverse primer sequence was: 5′-CCG GCT AAT TTT TGT ATT TTT AGT AGA GA-3′ and the probe sequence was: 5′-6FAM-ATC CCG GCT AAC ACG GTG AAA CCC-BHQ-1-3′. Primer and probes were synthesized by Biosearch Technologies. Genomic DNA (1 μL) of each sample in triplicate was evaluated using serial dilutions of white blood cell DNA (Promega) as a standard curve to determine DNA amounts. ALU PCRs were performed in a 30 μL total reaction volume as described in Campan and colleagues for MethyLight assays (25).

Purified DNA from breast milk specimens was sent to the University of Southern California Molecular Genomics Core for Illumina HumanMethylation450 (HM450) BeadChip analysis. The total amount of DNA from each breast milk sample was bisulfite treated with the Zymo EZ DNA Methylation Kit (Zymo Research) and 1 μL aliquots were used for MethyLight quality control analyses to determine the completeness of conversion and the amount of converted DNA available for the HM450 assay (26). The remaining bisulfite converted DNA samples were further processed using the Illumina FFPE Restoration Solution (Illumina) as specified by the manufacturer. The restoration solution repairs degraded DNAs for use in genome-scale genotyping and DNA methylation assay platforms. The entire restored sample was then used as a substrate for the Illumina HM450 BeadArrays, as recommended by the manufacturer and described previously (27). BeadArrays were scanned using Illumina iScan readers and the raw signal intensities were extracted from the *.IDAT files and normalized using the R package sesame (28, 29), a recently developed R package that masks problematic probes (i.e., probes for which DNA methylation is invalid because they overlap SNPs or repeats).

Subject characteristics were compared by race using a t test for continuous variables and a χ² test for categorical variables.

Questionnaire and DNA methylation data were integrated into one file for statistical analysis in R (version 3.3.2). Of the original 144 women, 5 were removed because their overall analytical signal rates were below 85%. Of the 482,421 total probes on the HM450 Beadchip, 176,386 probes were excluded because they were (i) located at or within 10 bp of known SNPs, (ii) known to be cross-reactive (30), or (iii) missing in 50% or more of the observations. After these exclusions 306,035 probes were included in the final analysis. Of these 306,035 probes, 138,363 CpG probes in promoter regions [i.e., TSS200, TSS1500, 5′ untranslated region (UTR), and first Exon], as defined by Sandoval and colleagues, were used for promoter-specific analyses.

Generalized linear regression models were used to identify relationships between breast milk DNA methylation and race as well as other breast cancer risk factors. DNA methylation beta values were treated as a continuous outcome. Bonferroni corrections (P < 1.63 × 10⁻⁷ for the full list of probes and P < 3.61 × 10⁻⁷ for analysis restricted to probes in the promoter region) were used to adjust P values unless otherwise noted. The adjusted model included race, lactation duration, age, BMI, smoking history, and donation year. Principal component analysis of the 10,000 most variable methylated CpG sites did not reveal any batch effects, and surrogate variable analysis did not identify any additional variables to be included for adjustment in the final multivariable model. Black and white women differed by some covariates, including lactation duration, over-the-counter pain medication use, shipping status, and donation year, thus we performed analyses stratified by race to determine their effects.

Biological significance of the genes corresponding to the significantly differentially methylated probes from the promoter-based analysis was determined using the Ingenuity Pathway Analysis (IPA) software and knowledge base. Only significant probes with a mean beta value difference of 0.1 or greater were included in the IPA. This threshold was applied to filter out small differences reflecting minor shifts in the composition of originating cell type populations. For lactation duration, this difference was calculated between the first (<125 days) and last (>269 days) categories. P values were calculated using the right-tailed Fisher exact test, which were then converted to P scores [−log₁₀(P value)]. For example, a P value of 1 × 10⁻¹⁰ would be equivalent to a score of 10.

The 139 healthy lactating women (82 white and 57 black) who donated milk for this study had a mean age of 30.2 years, ranging from 19 to 44 years (Table 1). Compared with white women, black women less frequently reported past week use of over-the-counter pain medications (P < 0.01) and reported ever smoking less often (P = 0.02). A higher percentage of black women were recruited >269 days after giving birth (P = 0.03), were more likely to have had breast milk shipped to the laboratory (P < 0.01), and to have donated after 2010 (P < 0.01). Other factors assessed, including time of day of milk expression, did not differ significantly by race.

DNA methylation levels significantly differed by race for 284 probes independent of lactation duration, age, smoking status, BMI, and donation year, and were scattered throughout the genome (Fig. 1; Supplementary Table S1). The top 10 CpG sites most significantly associated with race are listed in Table 2 along with gene annotation. A complete list of significant differentially methylated CpG sites by race is provided in Supplementary Table S2. Of the 284 significant CpG probes, 242 probes (85%) showed increased DNA methylation among black women (Supplementary Table S1). In addition, 80 (28%) probes were located in CpG islands, 84 (30%) were located in shores, 29 probes (10%) were located in shelves, and the remaining 91 (32%) probes were not located in or near CpG islands. Finally, 74 CpG sites were located in promoter region, of which 65 (86%) probes displayed increased DNA methylation among black women as compared with white women (Supplementary Table S3). A total of 116 (41%) CpG sites were located in the gene body regions, of which 106 (91%) probes displayed increased DNA methylation among black women compared with white women.

Analyses stratified by race revealed no statistically significant associations between DNA methylation and past week use of over-the-counter pain medication, smoking history, family history of breast cancer, age at first birth, number of pregnancies, menses age, or BMI, after the Bonferroni correction (1.63 × 10⁻⁷; Supplementary Table S5). There were also no significant associations between DNA methylation and shipping status for white women; there were not enough black women who did not ship their breast milk sample to evaluate the association between DNA methylation and shipping.

We identified 227 CpG probes for which their DNA methylation levels were significantly and inversely associated with the lactation duration (Supplementary Table S1). These probes were independent of race, age, smoking status, BMI, and donation year, and were scattered throughout the genome (Fig. 2). The top 10 CpG sites most significantly associated with lactation duration are listed in Table 3 along with gene annotation. A complete list of significant differentially methylated CpG sites by lactation duration is provided in Supplementary Table S4. Of these 227 significant CpG probes, 18 (8%) were located in CpG Islands, 59 (26%) were located in shores, 28 (12%) were located in shelves, while the remaining 122 (54%) probes were not located in or near CpG islands. A total of 67 probes (30%) were in gene promoter regions, 111 probes were located in the gene body, and the remaining nine probes were in the 3′-UTRs (Supplementary Table S3).

Analyses stratified by categorical lactation duration (<125 days, 125–269 days, and >270 days) revealed significant associations (after Bonferroni correction) between DNA methylation and race, and consistently showed increased DNA methylation in black women, which confirms the robustness of this finding in our adjusted model (Supplementary Table S1).

DNA methylation levels varied by race at 94 probes targeting promoter regions, including 74 (79%) with higher levels among black women. For lactation duration, 75 probes showed decreased methylation (Supplementary Table S6). The full list of significant probes in promoter regions associated with race and lactation duration are presented in Supplementary Tables S7 and S8.

We performed IPA (31) on the gene lists from differentially methylated probes with a mean beta value difference of 0.1 or greater to identify their potential biological relevance. This analysis revealed 19 unique genes from the 19 differentially methylated probes that showed differential DNA methylation by race and met the difference threshold, which were enriched for the following networks: (i) amino acid metabolism, molecular transport, and small molecule biochemistry (P score = 3), (ii) cancer, organismal injury and abnormalities, and RNA post-transcriptional modification (P score = 3), and (iii) amino acid metabolism, cancer, and carbohydrate metabolism (P score = 3; Supplementary Table S9A). Of these 19 genes, seven (SRMS, GSE1, ABCC4, DHRS4, STAB2, RPS16, and IFNGR2) had a disease or function annotation category that indicated “cancer” in the Ingenuity Knowledge Base (IKB), however, none of them were specific to breast cancer. Two additional proteins (from the full list of 94 promoter CpG sites), ALDH2 and EPAS1 were also identified as being associated with cancer according to the Cancer Gene Census (cancer.sanger.ac.uk; ref. 32).

For lactation duration, IPA revealed 48 unique genes from the 56 differentially methylated probes by lactation duration that met the difference threshold and were enriched for the following networks: (i) cellular movement, cellular growth and proliferation, and cell signaling (P score = 19), (ii) cell death and survival, cellular movement, and cardiac enlargement (P score = 17), and (iii) cellular development, cellular growth and proliferation, and hematological system development and function (P score = 4; Supplementary Table S9B). Of the 48 genes, eight (CYP19A1, G0S2, VEGFA, VDR, CDC42EP3, LIMA1, CD33, and PRKCD) had a disease or function annotation category that indicated “Cancer” in the IKB. Five (of the eight) proteins were implicated specifically in breast cancer: CYP19A1, G0S2, VEGFA, VDR, CDC42EP3. Two additional proteins (of the 48 genes), CANT1 and CREB3L1 were also identified as being associated with cancer according to the Cancer Gene Census (cancer.sanger.ac.uk; ref. 32).

In this first study to explore the relationship between genome-wide DNA methylation levels in breast milk and race as well as other breast cancer risk factors, we identified 284 CpG probes differentially methylated by race, and 227 CpG probes differentially methylated by lactation duration. Of the CpG probes differentially methylated by race, 85% of them indicated DNA hypermethylation associated with black race, while all probes differentially methylated as a function of lactation duration indicated reduced DNA methylation with increasing lactation duration, including when analyses were restricted to the promoter region. Furthermore, IPA revealed networks believed to be important to the development of cancer.

Our results provide new evidence of the impact of race and lactation duration on breast milk DNA methylation, which may inform breast cancer etiology. Previous studies have identified differentially methylated regions between breast tumors and healthy tissue (33), while other studies have identified differentially methylated regions by race and breast cancer subtype (34). In particular, research has shown that there are more differentially methylated sites in estrogen receptor–negative breast tumors in black women compared with breast tumors from white women (33–35), suggesting that DNA methylation could impact expression of genes involved in breast cancer subtype carcinogenesis and potentially explaining the racial disparities observed not only for breast cancer overall (i.e., highest incidence rates in white women; ref. 36 and highest mortality rates in black women; ref. 1), but also in the proportion of breast cancer subtypes observed for these groups of women (i.e., highest proportion of TNBC in black women; ref. 1). Our findings suggest that DNA methylation states at this developmentally important time in a woman's reproductive life cycle might contribute to our understanding of breast cancer etiology and/or racial disparities in breast cancer incidence.

In a previous study that explored racial differences (n = 61 white and n = 22 black) in genome-wide DNA methylation in breast tissues from women undergoing breast reduction surgery, 485 CpG sites were differentially methylated between black and white women after adjusting for age and BMI, with 58% being hyper-methylated in black women (13). These tissues were blunt dissected, prior to freezing, to remove adipose tissue. The mean age for black women was 34.4 years and 40.7 for white women. We compared our list of 284 CpG sites with their list of 485 and found 17 CpG sites in common, all of which showed DNA hypermethylation in the same direction, with all except one showing DNA hypermethylation in black women (Supplementary Table S10). Four CpG sites (of the 17) were found in gene promoter regions. Some differences between the two studies that might account for such low concordance include, bio-specimen used (breast milk vs. healthy breast tissue), the starting number of CpG sites tested (our 306,035 probes compared with their 247,456), our 139 (82 white and 57 black) samples compared with their 83 samples (61 white and 22 black), and the correction method for multiple tests (our Bonferroni, 1.63 × 10⁻⁷, compared with their Benjamin and Hochberg FDR, 1.35 × 10⁻⁴). In addition, DNA methylation likely reflects a complex interplay between genetic and environmental exposures, which may be unaccounted for or different between our respective study populations. Despite these differences, we still observed some overlap between the IPA network results, which could indicate that the DNA methylation states in breast milk are robust, and in fact representative of the health of the breast tissue and not just the lactation state.

The IPA network results revealed that race and lactation length were associated with DNA methylation levels for genes in networks that may be relevant to carcinogenesis (i.e., cellular development, cellular growth, and proliferation, etc.). These results suggest that race and lactation duration may affect DNA methylation states and, potentially, subsequently the expression of genes involved in cancer early in life. For race, there were seven genes (SRMS, GSE1, ABCC4, DHRS4, STAB2, RPS16, and IFNGR2) that the IKB (31) indicated were associated with cancer, but not specifically breast cancer. For lactation duration, there were eight genes (CYP19A1, G0S2, VEGFA, VDR, CDC42EP3, LIMA1, CD33, and PRKCD) that the IKB indicated were associated with cancer, with five (CYP19A1, G0S2, VEGFA, VDR, and CDC42EP3) specifically related to breast cancer, and an additional two genes (CANT1 and CREB3L1) from the Cancer Gene Census. Whether differentially methylated probes and pathways affected by race have implications for early onset breast cancer is an area for future investigation.

Strengths of this study include the use of samples from cancer-free participants to understand epigenetic differences by exposure status and the use of novel and more rigorous normalization process to mask deleted and hyperpolymorphic regions. A special recruitment effort was employed to obtain samples from black women, which allowed us to oversample this demographic for this study and thus, provided additional power to detect differences between black and white women. However, some limitations to our study include the relatively small sample size and potential residual confounding as there are factors that may differ by race that we did not collect, such as diet, socio-economic information, and breastfeeding practices. Finally, we also recognize that DNA methylation levels in this study may reflect that of the lactation state rather than the long-term health of the breast, and additional studies comparing intra- and interwoman variation in methylation profiles observed both during lactation as well as the post-lactational period are needed. Despite these limitations, evaluating DNA methylation profiles among black and white women during a critical postpartum window may provide important etiologic information to better understand how lactation protects against breast carcinogenesis. In addition, we were able to derive IPA networks consistent with a previous study performed in healthy breast tissue (13). Confirmation and extension of these findings could provide insights into markers and mechanisms related to the effects of pregnancy and breastfeeding on breast cancer risk.

This study identified associations between genome-wide DNA methylation levels in breast milk and race and lactation duration, suggesting that these two exposures influence epigenetics of the healthy postpartum breast. It is well known that race and breastfeeding can influence breast cancer risk (3); however, the mechanisms by which these two exposures play a role in breast carcinogenesis are not well understood. The epigenetic differences by race and lactation duration provide some clues as to how these exposures might affect breast cancer risk.

P.W. Laird has received speakers bureau honoraria from Progenity, Inc. and has ownership interest (including patents) in AnchorDx. D.J. Weisenberger is a consultant (paid consulting) at Zymo Research, has an unpaid consultant/advisory board relationship with Zymo Research. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B.C. Davis Lynn, R.M. Pfeiffer, J. Murphy, M.E. Sherman, K.F. Arcaro, G.L. Gierach

Development of methodology: B.C. Davis Lynn, C. Bodelon, D.J. Weisenberger, E.P. Browne, M.E. Sherman, K.F. Arcaro

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.W. Laird, M. Campan, D.J. Weisenberger, E.P. Browne, M.E. Sherman, K.F. Arcaro

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.C. Davis Lynn, C. Bodelon, R.M. Pfeiffer, H.P. Yang, H.H. Yang, M. Lee, P.W. Laird, J. Murphy, J.N. Sampson, D.L. Anderton, M.E. Sherman, G.L. Gierach

Writing, review, and/or revision of the manuscript: B.C. Davis Lynn, C. Bodelon, R.M. Pfeiffer, H.P. Yang, P.W. Laird, J. Murphy, J.N. Sampson, E.P. Browne, D.L. Anderton, M.E. Sherman, K.F. Arcaro, G.L. Gierach

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Bodelon, H.P. Yang, J. Murphy, E.P. Browne, D.L. Anderton

Study supervision: M.E. Sherman, K.F. Arcaro, G.L. Gierach

A special thank you to Drs. Mingyi Wang and Bin Zhu at the NCI's Cancer Genomics Research Laboratory for their expertise in bioinformatics. This research is supported by the Intramural Research Program and the Cancer Prevention Fellowship Program of the NCI at the NIH and a NIH Bench-to-Bedside Award from the Office of Research on Women's Health. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.