Methylation biomarkers of polybrominated diphenyl ethers (PBDEs) and association with breast cancer risk at the time of menopause

Abstract

Background:

Exposure to polybrominated diphenyl ethers (PBDEs) may influence risk of developing post-menopausal breast cancer. Although mechanisms are poorly understood, epigenetic regulation of gene expression may play a role.

Objectives:

To identify DNA methylation (DNAm) changes associated with PBDE serum levels and test the association of these biomarkers with breast cancer risk.

Methods:

We studied 397 healthy women (controls) and 133 women diagnosed with breast cancer (cases) between ages 40 and 58 years who participated in the California Teachers Study. PBDE levels were measured in blood. Infinium Human Methylation EPIC Bead Chips were used to measure DNAm. Using multivariable linear regression models, differentially methylated CpG sites (DMSs) and regions (DMRs) associated with serum PBDE levels were identified using controls. For top-ranked DMSs and DMRs, targeted next-generation bisulfite sequencing was used to measure DNAm for 133 invasive breast cancer cases and 301 age-matched controls. Conditional logistic regression was used to evaluate associations between DMSs and DMRs and breast cancer risk.

Results:

We identified 15 DMSs and 10 DMRs statistically significantly associated with PBDE levels (FDR<0.05). Methylation changes in a DMS at BMP8B and DMRs at TP53 and A2M-AS1 were statistically significantly (FDR < 0.05) associated with breast cancer risk.

Conclusion:

We show for the first time that serum PBDE levels are associated with differential methylation and that PBDE-associated DNAm changes in blood are associated with breast cancer risk.

Introduction

An increasing body of evidence supports the critical role of environmental exposures in the development of breast cancer (Knower et al. 2014, Rodgers et al. 2018, Terry et al. 2019). Yet, the mechanisms by which environmental chemicals promote breast cancer are not fully established. Polybrominated diphenyl ethers (PBDEs), a class of over 200 organo-halogenated compounds widely used in plastics, foams, textiles, electronic devices and building materials, are some of the most pervasive environmental pollutants detected in the environment, including in wildlife and human tissue (Darnerud et al. 2001, Hites 2004, Betts 2008, Lorber 2008). PBDEs are endocrine disrupting chemicals (EDCs) that bind to hormone receptors (i.e., estrogen receptor) and act as both agonists and antagonists (He et al. 2008, Mercado-Feliciano and Bigsby 2008, Mercado-Feliciano and Bigsby 2008, Kanaya et al. 2019). Increasing evidence suggests that EDC exposures may cause disease through epigenetic regulation of gene expression (Pouliot et al. 2015, Rojas et al. 2015).

Based on their role as EDCs, PBDE exposure is potentially a risk factor for development of breast cancer, particularly during the menopausal transition. In the population, the rate of estrogen receptor positive (ER+) breast cancer incidence decreases at menopause when women’s natural levels of estrogens and progesterones are declining (Yasui and Potter 1999). For women taking hormone replacement therapy (HRT), levels of estrogen and progesterone are supplemented. As found in the Women’s Health Initiative study, women taking HRT had significantly increased risks of developing ER+ breast cancer (Rossouw et al. 2002). A biologically-based breast tumor growth rate model developed by Santen and colleagues (Santen et al. 2013) explained the increased risks. In this model, HRT promoted the growth of pre-existing occult lesions and minimally initiated de novo tumors leading to clinically detectable breast cancers. We hypothesize that PBDEs behave similarly to HRT, by acting as weak estrogen mimetics, in promoting growth of occult cancers.

The few previous case-control studies to directly assess the association of PBDE serum levels and risk of breast cancer have been inconsistent and largely negative. This may be because the samples were collected after diagnosis and treatment of breast cancer so that the PBDE measurement at the time of collection did not reflect past exposures, which is important given that PBDE levels, although persistent, are declining after being banned. Cancer treatment possibly could further affect PBDE levels. Due to the persistent and cumulative effect of DNA methylation (DNAm) from an environmental exposure, DNAm is a more sensitive and reliable biomarker than a one-time chemical measurement, as it reflects the history of cumulative exposure (Alavian-Ghavanini and Ruegg 2018, Meehan et al. 2018), whereas a single PBDE measurement may not represent the history of past exposure.

Blood DNAm has been demonstrated as a suitable biomarker of breast cancer risk (Brennan et al. 2012, van Veldhoven et al. 2015, Xu et al. 2020). In this study, we hypothesized that PBDE serum levels would affect DNAm. We conducted an epigenome-wide association study (EWAS) to first determine the association of PBDE levels and DNAm changes in blood from healthy controls. After identifying DNAm biomarkers of PBDE serum levels, we then tested their association with risk of developing breast cancer.

1. Materials and Methods

1.1. Participants.

Women between the ages of 40 and 58 years, the time of the menopausal transition, were selected from the nested breast case-control study within the CTS, an on-going prospective cohort study of breast cancer among female California public school professionals initiated in 1995 (Bernstein et al. 2002). Questionnaires were completed at baseline with subsequent follow-up questionnaires. At the time of blood draw, a short questionnaire was administered including menopausal status on the date of blood draw. Phlebotomists visited participants to collect the blood samples for cases after breast cancer diagnosis and for women without breast cancer frequency matched to the cases by 5-year age group, race/ethnicity, and broad geographic region (corresponding to the three field study collection sites) (Hurley et al. 2019). Between 2011 and 2015, a serum separator tube of blood was drawn to be used for assays of environmental chemicals. This subset of women with the tube of blood for environmental studies formed the eligible pool of participants from which we selected 550 participants who were between the ages of 40 to 58 years of age at blood draw including 144 invasive breast cancers diagnosed between the ages of 40 to 57 years and 406 healthy controls. After excluding samples with missing PBDE measurements or poor DNA quality, there were a total of 397 healthy controls and 138 invasive breast cancer cases available for study. In total, 397 controls and 133 breast cancer cases were used (Figure 1).

Figure 1.

Schema of sample use for healthy controls and invasive breast cancer cases between the ages of 40 and 58 years

1.2. Measurement of PBDE levels.

Serum PBDE concentrations were measured using gas chromatography/high resolution mass spectrometry (Hurley et al. 2017) at the California Department of Toxic Substances Control Lab. The serum PBDE concentrations were lipid-adjusted; the raw PBDE concentration in ng/ml was transformed into a lipid-adjusted concentration in ng/g total lipids based on Phillips’ formula (Phillips et al. 1989). For this study, we focused on the three predominant PBDE congeners, 2,2’,4,4’,5,5’-hexabromodiphenyl ether (BDE-153), 2,2’,4,4’,6-pentabromodiphenyl ether (BDE-100), and 2,2’,4,4’-tetrabromodiphenyl ether (BDE-47). The average limit of detection (LOD) and the detection frequencies for each of these congeners ranged from 0.008 to 0.032 ng/ml and 80 to 90%, respectively (Table S1). Concentrations below the LOD for which no signal was detected were estimated by single imputation from a log-normal probability distribution based on the observed distribution of quantified measurements, following the method suggested by Lubin et al (Lubin et al. 2004).

1.3. Measurement of DNA methylation (DNAm)

DNA from whole blood was extracted using the QIAamp DNA Blood Maxi Kit (Qiagen, CA). For measuring DNAm, the Infinium Human Methylation EPIC Bead Chip (Illumina Inc., CA, USA), which interrogates epigenome-wide DNA methylation for over 860,000 CpG sites, was run on 500 ng DNA/sample in the City of Hope Integrative Genomics Core on 320 control samples. Not all samples were assayed because of budget issues. Out of the 397 control samples, we randomly selected 317 plus 3 duplicates for verifying plate identification. Four of the samples had no PBDE measurements, so in total, we obtained results from 313 healthy controls (Figure 1). Summary statistics for relevant variables for those healthy controls are shown in Table 1. To minimize the risk of introducing bias that can confound downstream analyses, DNA samples were randomized using a random number generator and placed in 96-well plates according to sorted random numbers. After processing the chips, raw methylation image files were processed using the minfi and EnMix (Xu et al. 2016)packages (Aryee et al. 2014) in R. The “preprocessNoob” method (Triche et al 2013) was used for normalization and probe failures due to low intensities were removed by the criterion of detection p value greater than 0.01. The type I and type II probe bias was adjusted using the method of regression on correlated probes(Niu et al. 2016) and batch effects related to chip position and sample processing plate were adjusted using the ComBat method from the sva package (Johnson et al. 2007). The rmSNPandCH function in the DMRcate package was adopted to remove cross-hybridizing probes (Pidsley et al. 2016), sex-chromosome probes and probes within 2 bp distance to single nucleotide polymorphisms (SNPs) with minor allele frequency greater than 0.05. The white blood cell composition (CD8+, CD4+, Natural Killer cells, Monocytes, B-cells and nucleated red blood cells) was estimated from DNAm measurements using the Houseman method (Houseman et al. 2012).

Table 1.

Summary of selected characteristics for 313 healthy control participants with methylation data from the EPIC bead chip.

Variables	Summary
Age, mean (range)	51.7 (40 – 58)
Body mass index (BMI), mean (range)	24.5 (17.1 – 53.7)
Menopausal status	N (%)
Pre-menopausal	49 (15.7)
Peri-menopausal	38 (12.1)
Post-menopausal	226 (72.2)
Race	N (%)
Asian	36 (11.5)
Black	23 (7.3)
Latina	61 (19.5)
White	184 (58.8)
Unknown	9 (2.9)
Age at parity (years)	N (%)
Nulliparous	125 (40.0)
≤ 24	45 (14.4)
25–29	72 (23.0)
≥30	69 (22.0)
Unknown	2 (0.6)
Smoking	N (%)
Never	270 (86.3)
Former	34 (10.9)
Current	8 (2.5)
Unknown	1 (0.3)
Alcohol	N (%)
None	112 (35.8)
<20g/day	176 (56.2)
≥20g/day	7 (2.2)
Unknown	18 (5.8)
Hormone replacement therapy (HRT)	N (%)
No	224 (71.6)
Past	52 (16.6)
Current	37 (11.8)
PBDE concentration in ng/g total lipids	Median (range)
BDE-47	15.3 (2.7 – 343.6)
BDE-100	2.8 (0.4 – 72.6)
BDE-153	5.7 (1.9 – 284.9)

1.4. DNAm and PBDE levels.

Differential methylation sites (DMSs) were identified using robust linear regression with heteroskedasticity-consistent estimators to model the methylation levels of each individual CpG (M values) as the dependent variable and a single PBDE exposure as the main predictor while adjusting for covariates, including cell type composition (numeric), race/ethnicity (categorical), menopausal status (categorical), body mass index (BMI; numeric), age at parity (categorical), smoking (categorical), alcohol (categorical), hormone replacement therapy (HRT; categorical), sample collection center (categorical), sample collection year (categorical), and age at blood draw (numeric). In the linear model, a PBDE serum level was considered as either a continuous variable in the primary analysis or a categorical variable in the secondary analysis comparing samples in the bottom quartile of PBDE level to samples in the top quartile of PBDE level. This DMS analysis was implemented using the “CpGassoc (Barfield et al. 2012) R package. In addition to the conventional genomic inflation factor (lambda), the bacon package (van Iterson et al. 2017) was used to compute the test-statistic inflation in each association test.

Multivariable linear regression models, implemented in the DMRcate R package, were used to identify statistically significant DMRs (regional FDR ≤ 0.05) (Peters et al. 2015). A candidate DMR was identified as a group of adjacent (within 1 Kb) CpG sites in a chromosome region that had a consistent linear trend of association (p value < 0.01 and the same direction of association) in the DMS analysis. A statistically significant DMR was determined using a false discovery rate (FDR) < 0.05, which was corrected for the total number of tested chromosome regions using the Benjamini and Hochberg procedure (Benjamini and Hochberg 1995). Ingenuity Pathway Analysis (IPA) was used to predict pathways enriched in the genes nearby DMSs and DMRs.

To investigate the possible regulatory effect of DNAm changes on gene expression, we used RNAseq data and EPIC methylation data generated for peripheral blood monocytes collected from 79 healthy controls in the Mood and Methylation Study (MMS) (Zhu et al. 2019). Details of measurements of DNAm and gene expression for the healthy controls have been previously described (Zhu et al. 2019). We estimated the Spearman correlation between DNAm of individual CpG sites within a DMR and expression level of its nearby gene by comparing the fold change in mean methylation with the fold change in gene expression between those in the 25% and 75% methylation percentiles of 79 healthy controls in the MSS.

1.5. Analysis of association of PBDE-DNAm biomarkers and risk of developing breast cancer

To assess the association between risk of developing invasive breast cancer and the DMS/DMR biomarkers of PBDE serum levels described above, we matched cases with controls (two to four controls per case) on age and BMI; 301 controls and 133 invasive breast cancers were matched (Figure 1). Of the 301 matched controls, 217 were part of the first experiment to identify biomarkers of PBDE serum levels (Figure 1). The characteristics of the participants are shown in Table 2.

Table 2.

Characteristics of the 434 participants included in the analysis of the association of biomarkers of PBDE serum levels and risk of developing breast cancer.

Characteristics	Cases	Controls	P-value
Participants (n)	133	301
Age (mean and range)	50.2 (40–57)	50.4 (40–57)	0.72
Body mass index (mean and range)	25.7 (17.6–40.4)	25.7 (17.5 – 41.3)	0.93
Serum PBDE level in ng/g total lipids
BDE-47 (median, range)	11.3 (2.73–116)	15.0 (2.66–344)	0.312
BDE-100 (median, range)	2.23 (0.48–33.0)	2.71 (0.41–72.4)	0.227
BDE-153 (median, range)	5.3 (1.37–110)	5.85 (1.16–285)	0.136
Race/ethnicity, n (%)
Asian	8 (6)	44 (15)	< 0.001
Black	0	22 (7)
Latina	10 (8)	52 (17)
White	111 (83)	174 (58)
Unknown	4 (3)	9 (3)
Hormone replacement therapy, n (%)
Yes (past or current)	36 (27)	58 (19)	0.077
No	97 (73)	243 (81)
Age at parity
Nulliparous	56 (42)	124 (41)	0.499
≤ 24	15 (11)	29 (10)
25–29	29 (22)	84 (27)
≥ 30	33 (25)	62 (21)
Unknown	0	2 (1)
Smoking, n (%)
Never	113 (85)	257 (85)	0.893
Former	17 (13)	38 (13)
Current	3 (2)	5 (2)
Unknown	0	1
Alcohol, n (%)
None	44 (33)	117 (39)	0.046
< 20g/day	72 (54)	160 (53)
≥ 20g/day	11 (8)	9 (3)
Unknown	6 (5)	15 (5)
Sample collection site, n (%)
COH	46 (35)	113 (38)	0.001
CPIC	54 (41)	155 (51)
UCI	33 (25)	33 (11)
Year of blood draw, n (%)
2011	4 (3)	54 (18)	< 0.001
2012	65 (49)	120 (40)
2013	37 (28)	75 (25)
2014	19 (14)	30 (10)
2015	8 (6)	22 (7)

Column percentage included in the parentheses; COH: City of Hope; CPIC: Cancer Prevention Institute of California; UCI: University of California, Irvine; Age at Parity age: age at first full term pregnancy.

To measure DNA methylation for the DMSs and DMRs using targeted next generation bisulfite sequencing (tNGBS), we sent 500 ng genomic DNA of each sample to EpigenDx Inc. (Hopkinton, MA). Briefly, EZ DNA methylation-Gold Kit (Zymo Rsearch, CA) was used for bisulfite conversion of genomic DNA. To amplify top DMSs and DMRs, multiplex PCR was performed using the Qiagen Hot-Star Taq system for multiple in-silica designed targeted PCR assays with compatible amplicon size, primer Tm, GC content, and ΔG values. The Kapa Library kit (Kapa Biosystems, MA) was employed to add barcode and sequencing adaptor to PCR fragments. Enriched, template-positive library molecules were then sequenced on the Ion S5™ sequencer using Ion 530™ sequencing chips (Thermo Fisher, MA). FASTQ files from the Ion Torrent S5 server were aligned to the local reference database using open-source Bismark Bisulfite Read Mapper with the Bowtie2 alignment algorithm. Methylation values for each measured CpG site were calculated in Bismark by dividing the number of methylated reads by the total number of reads, considering only CpG sites covered by a minimum of 100 total reads. For tNGBS, we selected all DMRs with FDR < 0.05 and the DMSs with FDR < 0.05 and at least 1 adjacent CpG site within the distance of 100 bp either upstream or downstream of the DMS. Because we only would keep CpG sites with a minimum of 100 reads, the additional nearby CpGs in the 200 bp amplicons allow us to analyze a surrogate CpG if needed; the adjacent CpG site to the selected DMS may not have been included in the EPIC array.

Concordance of EPIC array and tNGBS data.

First, we calculated Spearman correlations to investigate the correlation between methylation values from the EPIC arrays and tNGBS using the 217 samples in common. Second, for the CpG sites with moderate to high methylation correlation, we used linear regression models to test the association of PBDE levels and methylation in the 434 samples in the case-control study. In the model, methylation measured by bisulfite sequencing was the outcome variable and the log2(BDE-153) or log2(BDE-100) level was the primary predictor variable, adjusting for breast cancer status, age, race, center, BMI, year of blood draw, age at parity, smoking, alcohol, and HRT use. For CpG sites with validated (P < 0.05) associations of PBDE levels and methylation, conditional logistic regression models, implemented in the clogit function from the survival R package, were used to test association between breast cancer status and DNA methylation of each DMS and each CpG site within a DMR. In the model, breast cancer status was the dependent variable and methylation of one CpG site at a time was the primary predictor, adjusting for covariates (race/ethnicity, sample collection center, year of blood draw, BMI, age at parity, smoking, alcohol, and HRT use). Benjamini-Hochberg procedure (Benjamini and Hochberg 1995) was used to compute the FDR.

2. Results

2.1. Identification of DMSs and DMRs associated with serum PBDE levels in healthy controls

Serum concentrations of the three PBDEs were right skewed and log₂-transformations were done so that the transformed values approximately follow a normal distribution for linear regression analysis (Figure S1, panels A-C). Serum levels of the three PBDEs showed moderate to strong positive correlations with the strongest correlation (Spearman correlation rho value = 0.91) observed between BDE-47 and BDE-100 (Figure S1, panel D). Of the more than 860,000 CpG sites measured, 775,216 CpG sites remained for final statistical analysis after exclusion of problematic probes.

In the primary analysis of continuous levels of the PBDEs, we evaluated associations between methylation of individual CpG sites and serum levels of each PBDE. From the association with levels of log2 (BDE-153), we identified 7 DMSs with FDR ≤ 0.05 (Figure 2, Manhattan plot). In the secondary analysis comparing methylation differences between the top and bottom 25% of samples determined by the lower and upper quartile of BDE-153 concentration, respectively, an additional six DMSs had an FDR < 0.05. For BDE-47, only 2 DMSs had an FDR < 0.05 in the primary analysis. For BDE-100, no DMSs met the FDR threshold of < 0.05. The Quantile-Quantile (QQ) plots of p values from these associations and corresponding test-statistic inflation estimated by lambda and bacon indicate that the test-statistic genomic inflation in the tests was modest (Figure S2). The p values, effect sizes, and nearby genes for these 15 DMSs associated with BDE153 and BDE47 are shown in Table 3.

Figure 2.

Manhattan plot of p-values from the primary linear regression model of methylation values of individual CpG sites to log₂(BDE-153). The model was adjusted for cell type composition, race/ethnicity, menopausal status, BMI, smoking, alcohol, HRT, age at parity, center, year of blood draw, and age at blood draw. The horizontal line represents the FDR threshold of 0.05; gene names for DMSs with FDR < 0.05) were labeled (cg1255997 in intergenic region)

Table 3:

Statistically significant DMSs (FDR < 0.05) associated with PBDE serum levels in samples from 313 healthy controls.

Cpg IDa	Chr	Location	Gene	Gene region	P-value	FDR	Effect sizeb	Association	Analysisc
cg17748822	1	40234041	BMP8B	Body	1.77E-08	0.014	−0.005	BDE-153	Primary
cg16663424	1	54562396	TCEANC2	3’UTR	2.75E-08	0.021	0.042	BDE-153	Secondary
cg09347135	2	71890349	DYSF	Body	9.41E-08	0.024	−0.046	BDE-153	Secondary
cg21115527	7	75931838	HSPB1	TSS200	7.08E-08	0.024	0.005	BDE-153	Secondary
cg00856432	3	53755477	CACNA1D	Body	2.14E-07	0.028	0.042	BDE-153	Secondary
cg06113552	1	54562389	TCEANC2	3’UTR	1.48E-07	0.028	0.060	BDE-153	Secondary
cg16646600	1	54562543	C1orf83	3’UTR	1.83E-07	0.028	0.049	BDE-153	Secondary
cg09785344	15	101389974			5.86E-08	0.036	0.006	BDE-47	Primary
cg15646096	1	155160365	MUC1	Body	9.19E-08	0.036	−0.008	BDE-47	Primary
cg05371417	4	139113574	SLC7A11	Body	1.85E-07	0.036	−0.007	BDE-153	Primary
cg07103072	20	32951050	ITCH	TSS200	1.82E-07	0.036	0.001	BDE-153	Primary
cg23416540	15	45571777	SLC28A2-AS1	TSS1500	1.40E-07	0.036	0.005	BDE-153	Primary
cg10826200	19	4247502	CCDC94	Body	2.46E-07	0.038	0.003	BDE-153	Primary
cg22117172	7	91764530	CYP51A1	TSS1500	3.39E-07	0.044	0.007	BDE-153	Primary
cg16115867	17	46755566			4.07E-07	0.045	−0.009	BDE-153	Primary

^aCpG IDs with bold font were selected for validation by next generation bisulfite sequencing

^beffect size is on the scale of M values in the linear regression model; in the primary analysis, the effect size is M value changes per 1 unit of log2(BDE) increase (per doubling of PBDE concentration); in the secondary analysis, the effect size is the mean M value difference between two sample groups determined by the 25% and 75% BDE percentiles

^cIn the linear regression model, M value methylation = BDE + covariates, BDE was either a continuous predictor variable log2(BDE) in the primary analysis or a categorical variable (two sample groups determined by 25% and 75% BDE percentiles) in the secondary analysis.

For chromosome regions where there was a minimum of four CpGs and at least two were DMSs with p ≤ 0.01, we tested the associations between PBDE serum levels and differential methylation across the region. We identified a total of 14 statistically significant (FDR < 0.05) DMRs overlapping with gene promoters and harboring at least 4 DMSs that demonstrated consistent direction of association in a region (Table 4). Two DMRs were statistically significantly associated with BDE-47, 4 DMRs with BDE-100, and 8 DMRs with BDE153 serum levels. The DMR nearby LRRC27 was statistically significantly associated with serum levels of the three PBDE congeners. The DMRs nearby UCN3 and LINC01168 were associated with both BDE-100 and BDE-153 serum levels. The majority of the DMRs (7 of 10 unique DMRs, 78%) were negatively associated (hypomethylated) with PBDE levels.

Table 4.

Statistically significant DMRs (FDR < 0.05) associated with PBDE serum levels in samples from 313 healthy controls.

Associationa	Chrb	DMR start	DMR end	CpGc	Nearby gene	Gene region	Peak Pd	FDRe	Meanf diff
BDE-153	12	9217079	9217907	10	A2M-AS1	TSS200	8×10−6	0.0002	0.0109
BDE-153	15	45669010	45671347	19	GATM	TSS1500	1×10−3	0.0379	−0.0029
BDE-153	10	134777785	134779406	8	LINC01168	TSS1500	3×10−6	0.0056	−0.0090
BDE-100	10	134777785	134779406	8	LINC01168	TSS1500	1×10−4	0.0214	−0.0081
BDE-47	10	134777785	134778648	6	LINC01166	TSS1500	3×10−3	0.0213	0.0067
BDE-100	10	134150119	134151130	12	LRRC27	TSS200	6×10−6	0.0049	−0.0378
BDE-47	10	134150119	134151130	12	LRRC27	TSS200	2×10−5	0.0030	−0.0388
BDE-153	10	134150119	134151130	12	LRRC27	TSS200	3×10−3	0.0164	−0.0252
BDE-153	22	18463400	18463646	4	MICAL3	5’UTR	1×10−5	0.0224	−0.0094
BDE-100	12	124873347	124874007	5	NCOR2	TSS1500	7×10−4	0.0056	0.0299
BDE-153	1	163037784	163039184	16	RGS4	5’UTR	2×10−5	0.0399	−0.0172
BDE-153	1	3585225	3585587	5	TP73	TSS1500	2×10−3	0.0051	−0.0181
BDE-100	10	5406143	5407594	13	UCN3	TSS200	2×10−4	0.0178	−0.0113
BDE-153	10	5406143	5407594	12	UCN3	TSS200	4×10−4	0.0140	−0.0029

^aAssociation between a BDE level and multiple CpG sites in a region tested by DMRcate method

^bChromosome position is based on GRCh37/(hg19)

^cNumber of CpG sites in a DMR(differentially methylated region)

^dsmallest p-value among multiple CpG sites in a region associated with BDE exposure from identification of individual differentially methylation site

^eoverall regional significance of a DMR, Benjamini-Hochberg adjusted for the total number of tested regions agglomerated from groups of individual significant CpG sites

^fMean methylation changes across all CpG sites in a region in relation to one unit of log2(BDE) in the primary analysis or mean methylation difference between samples with top and bottom 25% BDE concentration in the secondary analysis(italic font).

From IPA analysis, 6 canonical pathways were enriched in the 21 genes nearby the DMSs and DMRs (Table S2). Three genes (HSPB1, NCOR2, and TP73) are in the Aryl hydrocarbon receptor (AhR) pathway and one gene (MICAL3) is in the Pregnenolone biosynthesis pathway.

In order to assess whether DNA methylation changes in the DMRs were regulating gene expression, we tested the correlations between DNAm of individual CpG sites within a DMR and corresponding gene expression using DNAm and RNAseq data sets from the MSS (Zhu et al. 2019). Statistically significant and consistent negative correlations (adjusted p-value < 0.05) between gene expression and DNAm across multiple CpG sites within a DMR were observed in six of the nine tested DMRs (bold font of fold change in gene expression in the Table S3). Statistically significant positive correlations were observed in one of five CpG sites within the DMR nearby NCOR2, whereas in the the DMRs nearby UCN3 and GATM, both statistically significant positive and negative correlations were observed in CpG sites within the DMR (Table S3). For CpG sites with consistent negative correlations between methylation and gene expression, a small change of methylation (less than 10% or fold change < 1.1 in methylation between the top and bottom quartile of methylation in the MSS set) resulted in a greater than two-fold change in gene expression between the sample groups (Table S3, fold change in gene expression is less than 0.5 for genes with negative correlations between methylation and gene expression).

2.2. Testing PBDE-associated methylation changes as biomarkers of breast cancer risk

We selected all 10 unique DMRs in Table 4 and the 6 DMSs (FDR < 0.05) in Table 3 in bold font which had at least one adjacent CpG site within 100 bp in either direction for bisulfite sequencing. In addition, we included DMS cg27113430 which was associated with both BDE-153 and BDE-100 at FDR ≤ 0.10. Six of seven DMSs and six of 10 DMRs passed the in-silico PCR assay design at EpigenDX Inc. An example of an in-silico PCR assay design for a DMR is shown in Figure S3. Using tNGBS, DNA methylation at these 12 regions was measured in blood DNA samples from 133 breast cancer cases and 301 matched health controls (Table 2). Of the 301 controls, EPIC array data were available for 217 samples (Figure 1), so we investigated the correlation between methylation values from the EPIC arrays and tNGBS. Nine of the 12 sites had high correlations (Spearman correlation > 0.6), 2 sites (cg17748822 in the BMP8B and cg05371417 in SLC7A11) had moderate correlations (Spearman correlation between 0.3 and 0.4), and 1 site (cg07039560 in the HSPB1) had a poor correlation (Spearman correlation < 0.3) Table S4) and so was excluded from further analysis. Of the 11 regions with moderate to high methylation correlation, the association of BDE-153 level and methylation was validated (P < 0.05) for the DMS at BMP8B and DMRs at TP73 and A2M-AS1 for the 434 samples in case control study (Table S5).

In those sites where there was a significant association of PBDE levels and tNGBS methylation, conditional logistic regression was used to test associations between breast cancer status and methylation. A total of 21 CpG sites were at DMS cg17748822 (in intron 3 of BMP8B) and DMRs in the promoter regions of A2M-AS1 and TP73. For these three sequenced regions, methylation changes in at least one CpG were statistically significantly associated with breast cancer risk (FDR < 0.05) (Table 5; additional test results in Table S6). As an example, DNAm differences between breast cancer cases and controls at TP73 and BMP8B are shown in a combined figure of the logistic regression curve, histogram, and boxplot (Figure 3). Compared to controls, hypomethylation was observed in invasive breast cancer cases in CpGs in the promoter of TP73 and in intron 3 of BMP8B.

Table 5.

Statistically significant associations (FDR < 0.05) between DMS/DMR and breast cancer risk.

DMS/DMR	Gene	Chr:locationa	Gene Region	P	FDRb	Odds Ratio (95%CI)c	Median cased	Median control
DMR	A2M-AS1	12:9217602	TSS200	0.0007	0.007	0.94(0.91–0.98)	14.1	17.6
DMS	BMP8B	1:40234042	Intron 3	0.0010	0.009	0.91(0.86–0.96)	80.6	82.4
DMR	TP73	1:3585587	TSS1500	0.0066	0.019	0.89(0.82–0.97)	6.0	6.6

^aChromosome position is based on GRCh37/(hg19)

^bBenjamini-Hochberg adjusted P value

^cConditional logistic regression models were used to evaluate association between DNAm and breast cancer risk for 1 DMSs and 2 DMRs (21 CpG sites), adjusting for covariates of race/ethnicity, center of blood collection, year of blood draw, BMI, parity, alcohol, smoking, and HRT

^dMedian methylation percentage in case group.

TSS200 = within 200 bp upstream of transcription start site; DMR = differentially methylated region; DMS = differentially methylated site.

Figure 3.

Combination graph of logistic regression curve, histogram, and boxplot to display association between DNA methylation and breast cancer status for each gene promoter region. Fitted logistic regression curves (black) show probability of developing breast cancer based on DNA methylation. Histograms represent frequency (counting #) of cases (top histogram) or controls (bottom histogram) in each DNA methylation (x axis) bin. Panel A: TP73 promoter region, breast cancer cases exhibited hypermethylation compared to controls. Panel B: At the intron 3 BMP8B region, breast cancer cases exhibited hypomethylation compared to controls.

3. Discussion

PBDEs have been shown to act as EDCs through altering hormone signaling pathways both in vitro and in vivo (Diamanti-Kandarakis et al. 2009, Vuong et al. 2018, Kanaya et al. 2019). Effects may be direct or through alteration of epigenetic regulation features such as DNAm and noncoding RNA expression (Alavian-Ghavanini and Ruegg 2018). We hypothesized that, similar to the effects of exposure to hormone replacement therapy and increased risks of breast cancer (Rossouw et al. 2002, Santen et al. 2013), women at the time of menopausal transition may be particularly vulnerable to endocrine disruptors such as PBDEs. This is the first study to test the association of methylation biomarkers of PBDE serum levels on risk of developing breast cancer and suggests that they may potentially be a mechanism underlying the effect of PBDE serum levels on risk of developing breast cancer. The PBDE exposure may perturb epigenetic patterning of specific genes which then alters expression of those genes which then affects pathways that may alter risk of developing breast cancer. Our approach was to first conduct an EWAS to identify DNA methylation changes associated with serum PBDE levels in healthy women at the time of the menopausal transition (between 40 and 58 years at the time of blood draw), and second, use these DNA methylation biomarkers of body burden to test their association with risk of developing breast cancer.

For the 3 PBDE congeners, we identified 14 DMRs at promoter regulatory regions with FDR < 0.05 (Table 4) and 15 DMSs with FDR < 0.05 (Table 3). Of the three PBDEs, the vast majority of the DMRs and DMSs were associated with BDE-153 serum levels rather than BDE-100 and BDE-47 serum levels. Multiple studies have reported interactions of PBDEs with nuclear receptors (NRs) including hormone receptors (i.e., estrogen, progesterone, and thyroid hormone receptors) (Diamanti-Kandarakis et al. 2009, Vuong et al. 2018, Kanaya et al. 2019) and aryl hydrocarbon receptors (AhRs) (Kanaya, Bernal et al. 2019). NRs as well as AhRs can regulate DNA methylation patterns by recruiting DNA methyltransferases or thymine DNA glycosylase to specific genomic sites (Garcia-Carpizo et al. 2011, Hassan et al. 2017). Of the 10 genes regulated by the DMRs shown in Table 4, 3 genes are in hormone-related pathways based on IPA (Table S2); NCOR2 and TP73 are involved in AhR signaling and MICAL3, regulated by ER-alpha, is involved in the biosynthesis of pregnenolone, a precursor for all steroid hormones. These pathways are implicated in steroid and xenobiotic metabolism and may play important roles in metabolism of PBDEs (Alavian-Ghavanini and Ruegg 2018, Meehan et al. 2018). NCOR2 plays an important role in DNA methylation, chromatin remodeling, and histone modifications by silencing mediators of retinoic acid and thyroid hormone receptors (Perissi et al. 2010). TP73, which belongs to the TP53 family, is involved in regulation of cell proliferation and cell death and is frequently over-expressed in tumors compared to normal tissue (Gomez et al. 2018). In the MSS study, we observed a statistically significant negative correlation (Spearman correlation of −0.4, adjusted p value of 0.00014, Table S3) of methylation in the promoter of TP73 with gene expression. We observed that the promoter region of TP73 was statistically significantly associated with breast cancer risk (Table 5), with hypomethylation in invasive breast cases, and thus likely higher expression, compared to healthy controls. This observation suggests that interference of steroid receptors with PBDE exposure may perturb epigenetic patterning of relevant genes in these pathways and alter susceptibility for development of breast cancer.

Three of the 14 DMRs are located in promoter regions of noncoding RNAs, including one antisense RNA (A2M-AS1) and two LincRNAs (LINC01168 and LINC01166) (Table 4). We found a statistically significant positive association (hyper-methylation) of the DMR in the A2M-AS1 promoter and PBDE levels (Table 4). A2M-AS1 is a cis-acting antisense RNA transcribed from the 3-prime end of A2M. Cis-acting antisense RNA and its targeted mRNA can form a duplex that either degrades targeted mRNA by recruiting RNAase H or stops the translation of its targeted gene by blocking ribosome from binding to mRNA (Pelechano and Steinmetz 2013). In the MCC study, a highly statistically significant negative correlation between A2M-AS1 and A2M gene expression was observed (Spearman correlation −0.54, p value 1.01 × 10⁻⁸), consistent with the repression of antisense RNA on transcription. Moreover, a highly statistically significant negative correlation between DNAm of the four DMR CpG sites in the A2M-AS1 promoter and A2M-AS1 gene expression was observed, whereas a highly statistically significant positive correlation was observed between methylation of A2M-AS1 and A2M gene expression (Table S3). For these CpG sites, the regulatory effects of 5% to 10% DNA methylation changes on gene expression were greater than two-fold (Table S3). Given these correlations, it is likely that hyper-methylation of the A2M-AS1 promoter region from higher PBDE levels causes down-regulation of A2M-AS1 gene expression, which in turn results in upregulation of its target, A2M. A2M is in the IL-6 and IL-10 signaling pathway (IPA, Table S2) where it functions as a protease inhibitor and IL10 transporter (Acuner-Ozbabacan et al. 2014).

For LINC01168, a highly significant negative correlation (Spearman correlation of − 0.58, adjusted p value of 1.8E-08) between DNAm and gene expression was observed (Table S3). However, no regulatory targets have been reported. Although we do not have RNA in order to sequence LincRNA, the DNAm results suggest that higher serum levels of PBDE affected LincRNA expression levels, which then regulated transcription of corresponding coding RNA.

Of nine DMRs tested with the MSS data (Zhu et al. 2019), we found that for six DMR, modest methylation changes (0.1% to 10% changes between samples in the top quartile and the bottom quartile of methylation) resulted in greater than two-fold changes in gene expression (the fold change of gene expression with bold font in the Table S3). Our results are consistent with other environmental exposure studies where methylation differences between environmental chemical exposed and unexposed groups are generally small to moderate on the scale of 1 to 10% (Breton et al. 2017, Latchney et al. 2018). Although small magnitude differences, these statistically significant associations with exposure are validated across different populations and age groups (review by (Breton et al. 2017)). One explanation for observing only small DNAm differences in relation to chemical serum levels may be that the DNAm changes are only in a specific cell type in the blood (Latchney et al. 2018). One study suggested that methylation in T-cells may serve as a history of environmental exposure (Meehan et al. 2018). Bauer et al confirmed that the smoking-associated hypomethylation of a CpG site within GPR15, replicated by many studies, was due to a higher proportion of CD3+ T cells in smokers than non-smokers (Bauer et al. 2015). We used the Housemen approach (Houseman et al. 2012) to estimate blood cell composition and did not observe statistically significant differences in cell type proportion between samples with high and low PBDE concentrations (see Figure S4 as one example). In order to assess whether PBDE-influenced methylation changes are from a specific cell type within the blood, one would need to sort for different cell types in blood followed by lineage resolution analysis to examine effects of PBDE serum levels on DNA methylation for each specific cell type at high resolution.

Epigenetic changes in blood cells may represent adaptive intermediary responses to environmental exposure (Lappalainen and Greally 2017). In order to focus on methylation changes from serum levels of PBDEs, we first focused on healthy women to identify these biomarkers of body burden. By not including the breast cancer cases, for whom blood was drawn after diagnosis, we could circumvent potential issues of reverse causation of changes due to the cancer or treatment. Therefore, with our approach, the PBDE-associated DMSs (Table 3) and DMRs (Table 4) that we identified were free of cancer influence and reflect changes due to the PBDE levels rather than cancer. In the three DMS/DMRs that had a significant association of DNAm and BDE-153 levels in the tNGBS data, we then tested the association of the DNAm biomarkers of PBDE levels and breast cancer risk. At the DMS in intron 3 of BMP8B and the two DMRs at promoter regions of TP73 and A2M-AS1, methylation changes in at least one CpG were statistically significantly associated with breast cancer risk (FDR < 0.05) (Table 5; additional test results in Table S5). Compared to controls, breast cancer cases exhibited more hypomethylation (Table 5), which was associated with higher BDE-153 levels in the EWAS of 313 cancer-free controls for BMP8B (Table 3) and the DMR in the promoter of TP73 (Table 4), whereas the association direction was not consistent for A2M-AS1 (Table 4) where hypermethylation was associated with higher BDE-153 levels. One reason for this difference for the A2M-AS1 DMR may be that the observed association is due to effects of breast cancer rather than the association with PBDE given that the breast cancer samples were taken post-diagnosis; in the association of DNAm as outcome in the 434 samples, the coefficient for breast cancer status was more highly associated with DNAm than was the BDE-153 level. This would be consistent with results from the prospective Sister Study where they found that a number of methylation changes likely reflected cancer already present (Xu et al. 2020). TP73, as described previously, and BMP8B are plausible candidates to be associated with breast cancer. Hypomethylation in intron 3 of BMP8B was associated with higher levels of BDE-153 and with risk of developing breast cancer. Bone morphogenetic proteins (BMPs) are involved in extracellular signaling as part of the TGF-beta superfamily. Higher BMP8B expression has been associated with poorer overall survival in breast cancer (Alarmo and Kallioniemi 2010).

Few epidemiologic studies have investigated the association of levels of PBDEs and risk of developing breast cancer. Two studies investigated the association of PBDE levels in adipose tissue and results were inconsistent. In the Chinese study of 209 breast cancer cases and 165 controls, adipose tissue from breast cancer cases had significantly higher levels than the adipose tissue from controls, with the majority of the control tissue from the abdomen (He et al. 2018). In the Northern California study of 78 breast cancer cases and 56 controls, adipose levels in the breast were not significantly different than in the controls (Hurley et al. 2011). Interestingly, levels of PBDE were similar in the cases in both studies. In PBDEs measured in serum of Alaska Natives (n = 75 cases and 95 controls), a statistically significant difference (p = 0.04) was found for the one PBDE measured (BDE-47), although in multivariable analysis, it was not significant (Holmes et al. 2014). In a large study of 902 invasive breast cancers and 936 controls, between the ages of 41 and 87 years of age at blood draw, from the nested case control study of the CTS, there was no significant association of PBDEs and risk of developing breast cancer (Hurley et al. 2019). Similarly, in our study for which some samples overlap with the Hurley et al study (Hurley et al. 2019), there was no direct association of serum PBDE levels and breast cancer risk (data not presented). There may be several reasons why there were no associations seen when directly comparing body burden levels of exposure and breast cancer risk. First, the samples for the breast cancer cases were all drawn at or after the time of breast cancer diagnosis such that PBDE measurements at that time point may not represent pre-diagnostic and chronic exposures. Examination of DNA methylation as biomarkers of body burden may be a more appropriate way to identify possible effects of PBDE levels on risk of developing breast cancer in that chemical exposures may cause long-lasting, cumulative, and persistent epigenomic changes (Alavian-Ghavanini et al 2018 and Meehan et al 2018) and thus DNA methylation changes may better reflect long-term exposures than case-control association studies of a one-time serum PBDE measure.

Strengths and limitations.

There were several strengths of the study. We focused on the menopausal transition as a window of susceptibility for the development of breast cancer. This is important given the growing recognition that the breast may be especially vulnerable to environmental insults at critical times during the life course (Birnbaum and Fenton 2003, Rice et al. 2003, Terry et al. 2019). Second, we used a study design where we first identified the methylation biomarkers of PBDE serum levels in controls and then only tested those biomarkers in a case control design, thus preventing potential confounding with breast cancer status. Limitations include the small number of cases for the case-control study and that not all PBDE-associated DMS biomarkers were tested, so there may be additional DMS biomarkers of PBDE levels associated with breast cancer. Additionally, this was a cross-sectional study, so we can measure association but not assign causality. Lastly, no replication set was available with measured PBDE levels and EWAS data.

4. Conclusion

We show for the first time that PBDE levels are associated with DNAm changes in blood and that a number of these biomarkers of PBDE serum levels are associated with risk of developing breast cancer at the time of menopause. Although methylation changes were modest, of 20 DMRs tested in an independent data set, 10 DMRs were associated with more than 2-fold changes in gene expression. These biomarkers of PBDE serum levels associated with invasive breast cancer at the time of menopause need to be validated in a larger group of samples collected near the time of menopause and before breast cancer diagnosis.

Highlights.

• Conducted an EWAS of PBDE serum levels

• Identified DNA methylation biomarkers of PBDE blood serum levels

• PBDE-associated blood DNA methylation associated with breast cancer risk at menopause

• Environmental chemicals and risk of breast cancer at the menopausal transition

• PBDE serum levels and association with methylation changes in hormone-pathway genes