Lifetime Postnatal Exposure to Perfluoroalkyl Substance Mixture and DNA Methylation at Twelve Years of Age

Abstract

Per- and polyfluoroalkyl substance (PFAS) exposure has been linked to DNA methylation changes in neonates and adults. We previously reported that prenatal PFAS exposure may have a durable impact on DNA methylation from birth to adolescence. However, few studies have examined the association of postnatal PFAS exposure with alterations in DNA methylation. We examined the associations of lifetime postnatal PFAS mixture exposure with leukocyte DNA methylation in 154 adolescents from the HOME Study (2003–2006; Cincinnati, Ohio). Lifetime postnatal PFAS mixture exposure was estimated using latent profile analysis of four PFAS concentrations measured at birth, and ages 3, 8, and 12 years. We measured DNA methylation in peripheral leukocytes at 12 years using the Illumina HumanMethylation EPIC BeadChip. We estimated covariate-adjusted associations between postnatal PFAS mixture concentrations and DNA methylation measures using linear regression, and used KEGG enrichment analysis to identify molecular pathways. Four significant differentially methylated positions were observed in the higher vs. lower PFAS profile (FDR p-value <0.05). These PFAS-associated CpG sites annotated to gene regions related to various cancers, cognition, and cardiometabolic health. We identified 17 pathways (FDR p-value <0.05), which indicates possible mechanism linking PFAS exposure to several health effects.

Graphical Abstract

Introduction

Per- and polyfluoroalkyl substances (PFAS) are fluorinated organic chemicals that have been produced since the 1940s and used widely in industrial processes and consumer products, including stain and water resistant textile coatings, food container coatings, non-stick cookware, floor polish, fire-fighting foams, and industrial surfactants.^1–3 The production of some PFAS was voluntarily phased-out in the United States (US) beginning in the 2000s,^4–6 but many PFAS are resistant to environmental degradation and bioaccumulative.⁷ As a result, PFAS, such as perfluorooctanoate (PFOA), perfluorooctane sulfonate (PFOS), perfluorohexane sulfonate (PFHxS) and perfluorononanoic acid (PFNA) still contaminate drinking water, soil, and food, leading to detection of these PFAS in the blood of nearly all pregnant women, children, and adults in the US.^7–9

A number of studies have linked PFAS exposure to a wide range of health conditions, including altered vaccine response, cognition, puberty, cardiometabolic disorders, some cancers, and thyroid disease.^{6, 10–12} Although the underlying mechanisms that explain the associations between PFAS and health are not well characterized, PFAS associated epigenetics changes has been proposed as one of the potential mechanisms. Prior studies show that prenatal and concurrent serum PFAS concentrations are associated with altered DNA methylation in newborns,^13–15 children,^{16, 17} and adults.^{18, 19} However, we are unaware of studies examining the associations of lifetime postnatal PFAS mixture exposure with changes in DNA methylation in adolescence.

Examining the impact of postnatal PFAS is important, as infants and young children may be more susceptible to the effects of the mixture of PFAS that are often present in human milk or indoor dust.²⁰ Thus, we sought to address this gap in the literature by using data from the Health Outcomes and Measures of the Environment (HOME) Study to evaluate associations of lifetime postnatal exposure to a mixture of PFOA, PFOS, PFNA, and PFHxS with measures of DNA methylation at twelve years of age.

Methods and Materials

Study Participants

We used data from the HOME Study, which enrolled pregnant women from 2003 to 2006 in Cincinnati, Ohio. As previously described, of 1,263 eligible women, 468 enrolled and 410 remained in the study to deliver live-born infants.²¹ Of these, 347 child participants had at least one PFAS concentration measurement postnatally. Among these, 160 participants had valid DNA methylation measurements and 154 of these with complete covariate information were included in the final analysis.

The study was approved by the institutional review board of Cincinnati Children’s Hospital Medical Center (CCHMC). All caregivers provided written informed consent and minor study participants provided written informed assent.

PFAS Measurements

Serum concentrations of PFOA, PFOS, PFHxS, and PFNA were measured in blood samples collected at birth and ages 3, 8, and 12 years using online solid phase extraction-high performance liquid chromatography-isotope dilution with tandem mass spectrometry method at the Centers for Disease Control and Prevention (CDC).^22–24 Limits of detection (LOD) were 0.1 ng/mL (PFOA and PFHxS), 0.2 ng/mL (PFOS), and 0.082 ng/mL (PFNA). We assigned concentrations below the LOD to LOD /√2.

DNA Methylation Analysis and Cell Deconvolution

Trained staff collected fasting blood samples from study participants at 12 years of age, which were stored at −80 °C prior to DNA extraction and methylation analysis. We extracted genomic DNA from 100 μl whole blood using DNeasy Blood & Tissue Kit (Qiagen; Hilden, DEU) with RNase A (100 mg/ml; cat. no. 19101) following the manufacturer’s suggested protocol.¹⁶ DNA methylation was interrogated for each sample using the Infinium HumanMethylationEPIC BeadChip (EPIC) array (Illumina, Inc.; San Diego, CA, USA). Briefly, 250 ng genomic DNA was sodium bisulfite converted using the EZ DNA Methylation Kit (Zymo Research; Irvine, CA, USA). Bisulfite-converted DNA was then amplified, enzymatically fragmented, and hybridized to the Infinium MethylationEPIC BeadChip (Illumina). EPIC assays were performed at the Genomics, Epigenomics and Sequencing Core at the University of Cincinnati College of Medicine following Illumina’s recommended protocols.

Normalization and quality control of raw DNA methylation data was performed according to standard protocols.^{25, 26} Samples with >5% high detection p-values (p-value >1×10⁻⁷; n= 4) and probes with at least one high detection p-value were excluded. Sex chromosomes, loci associated with single-nucleotide polymorphisms (SNPs) and cross-hybridizing probes were also excluded.²⁷ As a result, 669,622 probes for autosomal loci were included in the analysis. Noob (preprocessNoob: minfi Bioconductor package) was used for background correction and dye bias normalization. Beta MIxture Quantile dilation (BMIQ; wateRmelon) was applied to correct probe design bias ²⁸ and ComBat was used to adjust for any batch effects.²⁶ We used logit-transformed $\text{M-values}=lo{g}_2\left(\frac{bet{a}_i}{1-bet{a}_i}\right)$ for all statistical analyses, where beta is the ratio of intensities measured by methylated probes and the sum of intensities measured by methylated and unmethylated probes.²⁹

Relative proportions of cell types in blood samples were assessed using the R package ‘FlowSorted.BloodExtended.EPIC’.³⁰ The proportions of twelve cell types (neutrophils, eosinophils, basophils, monocytes, naïve and memory B cells, naïve and memory CD4 + and CD8 + T cells, natural killer, and T regulatory cells) were calculated in the adolescents using an extended adult reference dataset.³⁰

Statistical Analysis

We used latent profile analysis (LPA) models to identify subgroups of participants with similar lifetime exposure to a mixture of four PFAS (i.e., PFOA, PFOS, PFNA, and PFHxS), assessed at birth, age 3, 8 and 12 years. LPA is designed to identify unobservable subgroups within a population.³¹ Unlike conventional approaches that are variable centered, LPA is a person-centered modeling technique which is more salient to identifying subgroups of persons with similar sources of PFAS exposure.³¹ Thus, LPA assumes that participants can be classified into subgroups with varying degrees of probability that have different profiles of PFAS exposures.³² As complete PFAS data at all four timepoints were required for performing LPA, we first performed multiple imputation by chained equations to impute any missing postnatal PFAS concentrations. Using all observed PFAS concentrations, the imputation process results in 50 imputed data sets, which were used to fit a series of LPA models. Within each imputed data set, we evaluated 1-, 2-, and 3-profile solutions using robust maximum likelihood estimation, 5,000 random starts, and assuming equal variances across classes for model identification. To determine the best fitting model, we calculated the sample-size adjusted Bayesian information criterion (SSABIC), model entropy, and bootstrapped likelihood ratio tests (BLRT) within each imputed dataset and model. Based on these assessments, we selected the 2-profile solution as the best fitting model. After selecting the two profiles as the best fitting solution, we re-ran the LPA model using all 50 imputed datasets, with estimated parameters being pooled via Rubin’s rules. Then, we re-ran the 2-profile LPA model in each imputed dataset, fixing parameters to those from the pooled model, so that we could assign individuals to their most likely latent profile within each imputed dataset via modal assignment. As such, the uncertainty in profile assignment was propagated to later analyses.³³ The LPA groupings (higher vs. lower PFAS profile) were identified based on average log₂-transformed PFAS concentrations in each profile. In this study, 65% of children were assigned to the first profile and 35% of children were assigned to the second profile (Figure S1). Compared to the first profile, geometric mean serum PFOA, PFOS, PFNA, and PFHxS concentrations were consistently greater in the second profile at each time point. We hereafter refer to the first profile as the “lower PFAS profile” and the second profile as the “higher PFAS profile”.

We examined univariate features of covariates by lifetime postnatal PFAS profile assignment. We assessed the average difference in DNA methylation levels for each CpG site by comparing higher PFAS profile to lower PFAS profile using linear models with robust methods that allow for a small percentage of outliers (the R package limma), adjusting for child age (continuous), sex (female vs. male), and race (non-Hispanic White vs. non-Hispanic Black and other), pregnancy serum cotinine concentrations (continuous), breastfeeding duration (continuous), maternal annual household income (continuous), and cell type composition at age 12 years. We selected these confounders using directed acyclic graphs (DAGs) (Figure S2) and based on prior knowledge.¹⁶ All the statistical analyses were performed using logit-transformed $\text{M-values}=lo{g}_2\left(\frac{bet{a}_i}{1-bet{a}_i}\right)$, where beta is the ratio of intensities measured by methylated probes and the sum of intensities measured by both methylated and unmethylated probes.

To account for multiple comparisons across DNA methylation loci, we used the Benjamini-Hochberg procedure to control false discovery rate (FDR) at the 5% level across CpG sites for lifetime postnatal PFAS exposure; any FDR p-value <0.05 was considered significant.³⁴ Genomic features associated with each CpG site were annotated using the UCSC Genome Browser database for human reference genome GRCh37,³⁵ and CpG sites assigned to TSS200 or TSS1500 (200 or 1500 bases from transcription start site) were annotated as being located in a promoter region.

To better understand the potential biological functions behind the microarray data, KEGG enrichment analysis (i.e., gene set analysis) was performed on genes annotated from the top 500 CpG sites with the lowest FDR p-values for lifetime postnatal PFAS to identify putative biological pathways. We used the gometh function in the R package ‘missMethyl’ to perform the enrichment test, which accounts for the different number of probes per gene on EPIC array and CpG sites annotated to multiple genes.³⁶

In secondary analysis, we evaluated the associations between lifetime postnatal PFAS mixture concentrations and cell type composition in adolescent blood at age 12 years by comparing higher PFAS profile to lower PFAS profile as PFAS have been identified as immunotoxicants.³⁷

We used Stata v15.1 (Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC) for multiple imputation, Mplus v8.6 (Muthén and Muthén) to estimate the LPA models, and R (version 4.2.2; R Development Core Team) for all other analyses.

Results and Discussion

Among 154 participants included in our final analysis, 100 were assigned to the lower PFAS profile and 54 were assigned to the higher PFAS profile (Table S1). The characteristics of included participants were similar to those of participants from the full population (Table S2). Adolescent age at the methylation measurement did not significantly differ between the lower and higher PFAS profile. Compared to the lower PFAS profile, the higher PFAS profile included a higher percentage of children who were non-Hispanic White, whose mothers were not exposed to tobacco smoke, whose family had annual household income>$80,000/year, and who had longer breastfeeding duration. Postnatal concentrations of PFOA, PFOS, PFNA, and PFHxS in our study (Figure S1) were similar to those of children aged 3–11 years participating in National Health and Nutrition Examination Survey (NHANES) 2013–2014.⁸

Four CpG sites were associated with higher vs. lower PFAS profile at FDR p-value <0.05 after accounting for confounders and multiple comparisons (Table 1 and Figure 1). These differentially methylated CpG sites were associated with four unique genes (GTF2A2, CAV1, EHMT2 and PDE5A) (Table 1). However, none of these CpG sites overlapped with those reported in our prior study for evaluating the association between gestational PFAS exposure and DNA methylation at birth and age 12 years.¹⁶ Three of the four CpG sites were hypomethylated and one was in a CpG island, and three were associated with a promoter. These CpG sites annotated to genes associated with PFAS related health conditions, including various cancers,^38–42 cognition,³⁹ cardiometabolic^43–45 and cardiovascular health⁴⁶ (Table S3).

Table 1.

Leukocyte CpG sites at age 12 years associated with lifetime postnatal PFAS mixture concentrations at FDR p-value <0.05 (higher vs. lower PFAS profile) in the HOME Study.

CpG	Chr	Position	CpG Location	Gene	Gene Region	β (p-value)	FDR p-value
cg00368989	15	59949771	Island	GTF2A2	TSS200	−0.282 (1.75E-07)	0.039
cg01265597	7	116164633	Shore	CAV1	TSS1500	−0.201 (8.36E-08)	0.039
cg15635876	6	31865578	Shore	EHMT2	TSS200	−0.346 (1.48E-07)	0.039
cg08957091	4	120448071	Open sea	PDE5A	Body	0.477 (2.57E-07)	0.043

Note: none of these CpG sites were associated with an enhancer identified by the FANTOM5 project. FANTOM5 enhancers are active, in vivo-transcribed enhancers. Chr, chromosome; Body, between the ATG and stop codon; TSS, transcription start site; TSS200, 200 bases from TSS; TSS1500, 1500 bases from TSS; CpG, cytosine–guanine dinucleotide; FDR, false discovery rate; HOME, Health Outcomes and Measures of the Environment. Statistically significant CpG sites were defined as having FDR p-values <0.05; CpG sites were annotated to a promoter region if they were assigned toTSS200 or TSS1500. All models were adjusted for child age (continuous), sex (female vs. male), and race (non-Hispanic White vs. non-Hispanic Black and other), pregnancy serum cotinine concentrations (continuous), breastfeeding duration (continuous), maternal annual household income (continuous), and cell type composition at age 12 years.

Figure 1.

Volcano and Manhattan plots for the epigenome-wide adjusted associations of serum concentrations of lifetime postnatal PFAS mixture (higher vs. lower PFAS profile) with DNA methylation at age 12 years in the HOME Study (Cincinnati, Ohio; enrolled 2003–2006). These models were adjusted for child age (continuous), sex (female vs. male), and race (non-Hispanic White vs. non-Hispanic Black and other), pregnancy serum cotinine concentrations (continuous), breastfeeding duration (continuous), maternal annual household income (continuous), and cell type composition at age 12 years. The left panel is a volcano plot showing the difference in leukocyte DNA methylation (magnitude of effect on M-value: x-axis) associated with lifetime postnatal PFAS mixture concentrations (higher vs. lower PFAS profile) for each CpG site plotted against its negative log₁₀-transformed p-value (y-axis). Triangles represent the CpG sites with FDR p-value <0.05. The right panel is a Manhattan plot showing negative log₁₀-transformed p-values for the associations between postnatal PFAS mixture concentrations (higher vs. lower PFAS profile) and DNA methylation across chromosomes. Statistically significant CpG sites were defined as having an FDR p-value <0.05. Horizontal lines denote with FDR p-value <0.05. Note: CpG, cytosine–guanine dinucleotide; FDR, false discovery rate; HOME, Health Outcomes and Measures of the Environment.

Prior epigenome-wide association studies (EWAS) have examined the associations between prenatal exposure to PFAS and changes in DNA methylation among newborns^13–15 and children,¹⁶ while one study examined the associations of postnatal PFAS exposure with DNA methylation in children.¹⁷ The latter cross-sectional study of serum PFAS concentrations and DNA methylation among 63 Swedish children aged 7–11 years from the Ronneby Biomarker Cohort found 12 differentially methylated positions comparing the high PFAS exposure group (serum PFOS range: 190–464 ng/mL) to the low PFAS exposure group (serum PFOS range: 1–5 ng/mL).¹⁷ None of these 12 CpG sites overlapped with those four CpG sites reported by our study; however, three of these CpGs reported by Ronneby Biomarker Cohort had the same direction of the associations observed in our study with raw p-values >0.4. The inconsistent findings reported by ours and Ronneby Biomarker Cohort study may be explained by differences in the study source population and design, sample size, concentration ranges of PFAS exposure group, timing of PFAS assessments, and the timing of examining DNA methylation. Although we were not able to identify the same CpG sites observed by the Ronneby Biomarker Cohort study, the CpG sites identified by both studies were linked to genes that are associated with similar PFAS related health outcomes (e.g., cancers and cognition).

After performing enrichment analysis on genes annotated from the top 500 CpG sites with lowest FDR p-values for lifetime postnatal PFAS, we observed significant enrichment of gene sets annotating to 17 unique KEGG pathways with an FDR p-value <0.05 (Table 2). The pathway with the smallest FDR p-value for lifetime postnatal PFAS was viral carcinogenesis (FDR p-value=0.002), followed by pathways (FDR p-values=0.028 or 0.041) related to inflammation and cancers (e.g., Ras, Wnt and TNF signaling pathways). These findings are consistent with current literature showing PFAS as immunotoxicants³⁷ and carcinogens.⁴⁷

Table 2.

KEGG pathways significantly enriched (FDR p-value <0.05) among top 500 CpG sites associated with lifetime postnatal PFAS mixture concentrations (higher vs. lower PFAS profile) ordered by FDR p-values in the HOME Study (Cincinnati, Ohio; enrolled 2003–2006).

Pathway ID	Description	# of genes in pathway	# of differentially methylated genes in pathway	p-value	FDR p-value
path:hsa05203	Viral carcinogenesis	193	14	6.57E-06	0.002
path:hsa04014	Ras signaling pathway	235	13	2.24E-04	0.028
path:hsa05100	Bacterial invasion of epithelial cells	77	7	3.52E-04	0.028
path:hsa05160	Hepatitis C	157	10	3.91E-04	0.028
path:hsa05213	Endometrial cancer	58	6	4.64E-04	0.028
path:hsa05162	Measles	139	9	6.72E-04	0.028
path:hsa05169	Epstein-Barr virus infection	199	11	6.75E-04	0.028
path:hsa04530	Tight junction	169	10	6.96E-04	0.028
path:hsa04310	Wnt signaling pathway	170	10	7.28E-04	0.028
path:hsa04668	TNF signaling pathway	114	8	7.84E-04	0.028
path:hsa04917	Prolactin signaling pathway	70	6	1.27E-03	0.041
path:hsa00520	Amino sugar and nucleotide sugar metabolism	49	5	1.48E-03	0.041
path:hsa05165	Human papillomavirus infection	330	14	1.79E-03	0.041
path:hsa05020	Prion disease	259	12	1.81E-03	0.041
path:hsa05163	Human cytomegalovirus infection	225	11	1.82E-03	0.041
path:hsa04728	Dopaminergic synapse	131	8	1.92E-03	0.041
path:hsa05167	Kaposi sarcoma-associated herpesvirus infection	194	10	1.97E-03	0.041

Note: Top 500 CpG sites for lifetime postnatal PFAS ranked by FDR p-values with corresponding genes were included in the pathway analysis. At FDR p-value <0.05, we identified 17 significant KEGG pathways for lifetime postnatal PFAS. FDR, false discovery rate; HOME, Health Outcomes and Measures of the Environment; ID, identifier.

In secondary analyses, we examined the associations between lifetime postnatal PFAS mixture profile assignment and cell type composition in adolescent blood samples. Generally, the higher PFAS profile was not associated with estimated immune cell type composition (Table S4). However, we observed an increased percentage of natural killer cells among adolescents in the higher PFAS profile compared to the lower PFAS profile (p-value = 0.05). A prior report using the HOME Study found that prenatal PFAS was associated with decreased B-lymphocytes in cord blood.¹⁶ In addition, PFAS can decrease B cell populations in animals,⁴⁸ and some PFAS are associated with decreased monocytes and natural killer cell levels in adults,⁴⁹ reduced vaccine antibodies in children,⁵⁰ and increased COVID-19 severity in adults.⁵¹ More studies are needed to improve our understanding on the immunotoxic effects of PFAS exposure.

Strengths of our study include the application of LPA to estimate lifetime exposure to a mixture of four PFAS from birth to age 12 years. In addition, we measured DNA methylation using the EPIC array, which interrogates almost twice the measured CpG sites than 450K array and provides higher coverage within enhancer regions and distal regulatory elements.⁵² However, there were also several limitations. The sample size was moderate to investigate associations between serum PFAS concentrations by comparing higher to lower PFAS profile and DNA methylation, and thus may have reduced statistical power. We also measured the levels of DNA methylation in leukocytes, which may not accurately represent the DNA methylation patterns in other tissues. We were not able to evaluate the impact of PFAS exposure on DNA hydroxymethylation, which may independently affect the associations between PFAS and child health.⁵³ Moreover, the results from our study may not be generalizable to populations with different genetic admixtures or PFAS levels. Finally, our results were not validated in another cohort.

Higher lifetime postnatal PFAS mixture profile was associated with differences in several CpG sites at age 12 years in this cohort. These CpG sites were in or near genes linked to PFAS-associated health conditions. To the best of our knowledge, this is the first study to examine the association between lifetime postnatal exposure to PFAS mixture and alterations in DNA methylation in adolescence. Future studies with large sample size are needed to confirm our findings.

Synopsis:

Exposure to a mixture of four PFAS from birth to twelve years of age may affect DNA methylation in adolescence.