Access the full text.
Sign up today, get DeepDyve free for 14 days.
Background: It has been proposed that maternal folic-acid supplement use may alter the DNA-methylation patterns of the offspring during the in-utero period, which could inﬂu- ence development and later-life health outcomes. Evidence from human studies sug- gests a role for prenatal folate levels in inﬂuencing DNA methylation in early life, but this has not been extended to consider persistent effects into adulthood. Methods: To better elucidate the long-term impact of maternal folic acid in pregnancy on DNA methylation in offspring, we carried out an epigenome-wide association study (EWAS) nested within the Aberdeen Folic Acid Supplementation Trial (AFAST—a trial of two different doses: 0.2 and 5 mg, folic acid vs placebo). Offspring of the AFAST partici- pants were recruited at a mean age of 47 years and saliva samples were proﬁled on the Illumina Inﬁnium Human Methylation450 array. Both single-site and differentially methy- lated region analyses were performed. –9 Results: We found an association at cg09112514 (p¼ 4.0310 ), a CpG located in the 5’ untranslated region of PDGFRA, in the main analysis comparing the intervention arms [low- (0.2 mg) and high-dose (5 mg) folic acid combined (N¼ 43)] vs placebo (N¼ 43). Furthermore, a dose–response reduction in methylation at this site was identiﬁed in rela- tion to the intervention. In the regional approach, we identiﬁed 46 regions of the genome V The Author(s) 2018. Published by Oxford University Press on behalf of the International Epidemiological Association. 928 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 3 929 that were differentially methylated in response to the intervention (Sidak p-value <0.05), including HLA-DPB2, HLA-DPB1, PAX8 and VTRNA2–1. Whereas cg09112514 did not rep- licate in an independent EWAS of maternal plasma folate, there was suggested replica- tion of differential methylation in PAX8. Conclusions: The results of this study suggest that maternal folic-acid supplement use is associated with changes in the DNA methylation of the offspring that persist for many years after exposure in utero. These methylation changes are located in genes implicated in embryonic development, immune response and cellular proliferation. Further work to investigate whether these epigenetic changes translate into detectable phenotypic differ- ences is required. Key words: epigenetic, AFAST, randomized–controlled trial, longitudinal, epigenome-wide association study, DNA methylation Key Messages We investigated the impact of a folic-acid supplementation trial that enrolled pregnant women in the late 1960s on long-term epigenetic changes in their offspring by assessing differences in DNA-methylation levels of their offspring at a mean age of 47 years. In saliva samples obtained from the offspring 47 years after the trial was conducted, we identiﬁed 45 regions of the genome that were differentially methylated in response to the intervention. The results of the study suggest that maternal folic-acid supplement use is associated with changes in DNA methyla- tion that persist for many years after in-utero exposure, but further work is needed to investigate whether these epi- genetic changes translate into detectable phenotypic differences. Introduction impact of nutritional exposures on the epigenome are often confounded, e.g. by other highly correlated macro/micronu- Folate is an essential micronutrient that plays an important trients or socio-economic factors not adequately captured in role in fetal development, with the potential for lifelong con- these previous studies. The strongest evidence relating prenatal sequences. It is a key player in one-carbon metabolism that nutrition to offspring methylation derives from intervention is closely linked to the provision of methyl groups for the studies (randomized–controlled trials or natural experiments). methylation of DNA —an epigenetic process that is crucial in In these studies, large differences in nutritional status occur in early development. Therefore, it has been proposed that ma- the study population (largely) at random and are therefore un- ternal folic acid may alter the methylation patterns of the off- 7,10,11 likely to be associated with confounding factors. spring during the in-utero period, which could impact health To better elucidate the long-term impact of maternal folic outcomes in later life. This was demonstrated in the Agouti acid in pregnancy on DNA methylation in the offspring, we mouse model, where methyl donor supplements (including carried out an epigenome-wide association study (EWAS) folate) given to pregnant dams resulted in increased DNA nested within the Aberdeen Folic Acid Supplementation methylation in the offspring at the Agouti allele, which had Trial (AFAST). AFAST was a randomized–controlled trial of phenotypic consequences of shifting offspring coat colour two different doses of folic acid (0.2 or 5 mg per day vs pla- and reducing the risk of obesity and tumorigenesis. cebo) starting at booking for antenatal care at <30 weeks’ Evidence from human studies suggests a role for prenatal 12–14 gestation that was performed in the late 1960s. folate levels in influencing DNA methylation in neonates and 6–9 Offspring of women who participated in AFAST, born dur- children, but this has not been extended to consider the per- ing the trial, were identified and invited to participate in the sistent effects of such exposures into adulthood. To examine present study at a mean age of 47 years. Their saliva samples effects in adults, studies with long-term follow-up are were obtained for DNA-methylation profiling. required. Furthermore, observational studies investigating the Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 930 International Journal of Epidemiology, 2018, Vol. 47, No. 3 Methods Institute of Applied Health Sciences, University of Aberdeen (http://www.abdn.ac.uk/iahs/research/obsgynae/ Parent study amnd/index.php). For this study, the ‘affected’ offspring of Study design trial participants (i.e. the children born during the mothers’ 12–14 AFAST has been described in detail elsewhere. Briefly, participation in the trial) were traced using the Community from June 1966 to June 1967, 3187 potentially eligible Health Index (CHI) and those living in the Grampian area women (women booking for antenatal care at <30 weeks’ were approached for participation by mail. Multiple births gestation who were resident in Aberdeen, UK) were invited were excluded. A total of 692 offspring were invited to to participate in a trial to examine the effects of folic-acid participate (Supplementary Figure 1, available as supplement use on pregnancy outcomes. Any woman for Supplementary data at IJE online) and sent an information whom folic acid had been prescribed previously was leaflet and consent form. excluded from the study. In all, 2928 women were randomized by alternate allocation to receive either 0.2 mg Follow-up data collection folic acid/day (n¼ 466, 15.6%), 5 mg folic acid/day Participants who consented to participate (N¼ 265, (n¼ 485, 16.6%) or a placebo (n¼ 1977, 67.5%). Trial Supplementary Figure 1, available as Supplementary data compliance was assessed by self-report and by measurement at IJE online) were mailed a short questionnaire to collect of folate status. In the placebo group, 1.9% reported that information on sex, age, self-reported height and weight, they had not taken their tablets regularly, compared with education, ethnicity, their health (e.g. current medications 1.7% in the group taking 0.2 mg folic acid and 3.2% in the and health conditions) and current and past smoking status group taking 5 mg. Prior to allocation, serum folate concen- and alcohol intake. A saliva sample collection kit trations were similar in the three groups and a dose–response (Oragene, DNA Genetek, Kanata, Ontario, Canada) was relationship was seen after allocation until the post-partum provided and participants were asked to collect a saliva period, indicating that the tablets were regularly taken from sample and return it through the post; 197 participants re- the time of recruitment (mean gestational age at book- turned a saliva sample and 196 completed a questionnaire, ing¼ 17 weeks) until the end of pregnancy (mean gestational representing a response rate of 28% (197/692). age at delivery¼ 40 weeks). Among 2093 parous women, The original trial treatment status of the study partici- the incidence of a positive history of congenital malforma- pant’s mother and other relevant trial data were provided tion in a previous pregnancy was 2%. by the Aberdeen Maternity and Neonatal Databank and linked to the offspring data. The linked anonymized data were given to researchers for analysis. Baseline data collection At the booking visit, the age of the mother, her gestation, DNA methylation parity, weight and blood pressure were recorded. The oc- Of the 196 individuals who returned a saliva sample and cupations of husbands/partners recorded on the study form questionnaire, 180 individuals had DNA extracted from at the time of delivery were used to determine the social saliva that passed quality control (QC). Of these, 170 were class of the women based on the Classification of female whereas only 10 were male. To minimize sex ef- Occupations 1966. The trial database was linked to the fects, we restricted profiling to females only and over- Aberdeen Maternity and Neonatal Databank to add fur- sampled based on intervention status (111 individuals: 66 ther demographic information on maternal smoking and placebo, 21 low-dose folic acid, 24 high-dose folic acid). height of mother. Additional information on mothers’ Genome-wide DNA-methylation profiling was performed weight and blood pressure at booking were obtained from on samples from 111 individuals using the Illumina the original obstetric records. Serum folate was measured Infinium HumanMethylation 450 array, run as described as previously described for 99.7% of women at the ante- previously. natal booking visit, 82.8% at approximately 30 weeks’ Details of sampling handling and DNA-methylation gestation, 37.2% at 36 weeks’ gestation and 63.4% in the profiling are outlined in the Supplementary Material, avail- postpartum period. able as Supplementary data at IJE online. For this analysis, investigating the effect of intervention on methylation, we included 43 placebo and 43 intervention (20 low-dose and Offspring study 23 high-dose) individuals to obtain a 1:1 placebo:interven- Identification and recruitment of participants tion selection and reduce the effects of batch Data from AFAST are archived within the Aberdeen (Supplementary Material and Figure 1, available as Maternity and Neonatal Databank records held by the Supplementary data at IJE online). Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 3 931 Ethics approval surrogate variables were generated using the ‘SVA’ pack- age in R and included in models to adjust for technical Ethics approval was given by the NRES Committee South batch and cell-type mixture given the absence of meas- West–Central Bristol REC. Approval to obtain addresses ured cell types in these samples. EWAS were performed of the offspring through the CHI was obtained from the using the ‘CpGassoc’ package implemented in R, Caldicott Guardian, the Medical Director of NHS which is designed to perform flexible analyses of methyla- Grampian. This study was conducted in accordance with tion array data and to test for an association between the Research Governance Framework for Health and methylation at CpG sites across the genome and pheno- Social Care and Good Clinical Practice and under the types of interest, adjusting for relevant covariates. sponsorship of the University of Bristol. All samples were Sites were annotated using the information provided by used and stored in accordance with the UK Human Tissue Illumina. Act 2004. Regional approach Statistical analysis Adjacent probes on the HM450 array are often highly cor- related and differentially methylated regions (DMRs) may We first aimed to assess whether the baseline characteris- be more biologically important than individual CpGs. tics of the subset of individuals included in our analysis ap- Therefore, as well as our single-site (CpG) analysis, we peared to be equally distributed with regard to a number also assessed differential methylation across larger regions of maternal and offspring variables outlined earlier. of the genome in response to the intervention. For this, we Continuous baseline demographic characteristics across used ‘Comb-P’ to identify regions enriched for low p-val- the three treatment groups were summarized as means and ues, corrected for auto-correlation with neighbouring standard deviations, and tested for overall trend using one- CpGs within 500 base pairs using the Stouffer-Liptak way analysis of variance (ANOVA). Categorical baseline method and adjusted for multiple testing using the Sidak variables were summarized as percentages and numbers in correction. each of the three treatment groups and an overall trend was tested by using the chi-squared test for trend. Functional analysis To explore the function of any identified DMRs, we used EWAS the missMethyl R package to test for enrichment for We next conducted an EWAS to investigate the long-term any gene ontology (GO) classification terms or the Kyoto impact of the randomized folic-acid supplement-use inter- Encyclopaedia of Genes and Genomes (KEGG) path- vention on offspring methylation in adulthood by evaluat- ways. The method applies Fisher tests, while correcting ing the association between DNA methylation (normalized for biases in the genomic coverage of the Illumina b value at 470 617 CpG sites on the array) and folic-acid Infinium HumanMethylation450 BeadChip array. All supplement use. CpGs on the array were used as background. We also We first combined both the 0.2- and 5-mg treatment used Fisher tests to test whether CpGs within our DMRs groups to form a ‘folic acid supplement use’ group and car- were enriched for CpGs within epialleles. Again, the ried out linear regression models to test the associations be- background was all CpGs on the array. P-values for all tween the normalized b values at each CpG site as the enrichment analyses were adjusted for multiple testing dependent variable and folic-acid supplement use as the in- using the FDR method. dependent variable. Secondary analysis was then per- formed to determine associations between low-dose supplement use vs placebo, high-dose supplement use vs Replication placebo and an ordinal model of high dose, low dose and For replication, we performed a look-up of epigenome- placebo. wide significant CpG sites from the single-site analysis in We adjusted for multiple testing using false discovery an EWAS meta-analysis of maternal plasma folate and rate correction (FDR) and also investigated CpGs with a DNA methylation (N¼ 1996) using summary-level data –5 p-value <110 . These analyses were adjusted for from this study obtained through dbGAP (dbGAP methylation array batch and also adjusted for gestational phs001059.v1.p1). Using these summary data from the age at booking and age of the offspring at the time of EWAS meta-analysis, where effect estimates, standard data and sample collection in the main analysis (folic-acid errors and p-values were available for each CpG site, we supplement use vs placebo), given findings of a difference also used Comb-P to identify overlap with the DMRs ob- in these covariates between the treatment groups. Ten tained from the EWAS in AFAST. Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 932 International Journal of Epidemiology, 2018, Vol. 47, No. 3 Table 1. Baseline characteristics of the mothers of participants in this study, collected as part of the original AFAST (1966–67) (n¼ 86) Variable Category Placebo (n¼ 43) Folic-acid supplement P 0.2 mg/day (n¼ 20) 5 mg/day (n¼ 23) Categorical n (%) n (%) n (%) Chi Age at delivery (years) (N¼ 86) <20 5 (11.6) 3 (15.0) 4 (17.4) 20–24 10 (23.3) 8 (40.0) 10 (43.5) 25–29 16 (37.2) 4 (20.0) 5 (21.7) 30 12 (27.9) 5 (25.0) 4 (17.4) 0.50 Parity (N¼ 86) 0 16 (37.2) 5 (25.0) 11 (47.8) 1 or 2 19 (44.2) 12 (60.0) 9 (39.1) 3 8 (18.6) 3 (15.0) 3 (13.0) 0.53 Smoking in pregnancy (N¼ 83) No 21 (51.2) 13 (68.4) 11 (47.8) Yes 20 (48.8) 6 (31.6) 12 (52.2) 0.36 Social class (N¼ 85) Non-manual 10 (23.3) 4 (20.0) 6 (27.3) Manual 33 (76.7) 16 (80.0) 16 (72.7) 0.86 Pre-eclampsia (N¼ 86) No 31 (72.1) 16 (80.0) 17 (73.9) Mild 12 (27.9) 4 (20.0) 6 (26.1) 0.80 Continuous Mean (SD) Mean (SD) Mean (SD) ANOVA BMI in pregnancy (kg/m )(N¼ 84) 23.6 (3.2) 23.3 (3.7) 23.7 (3.2) 0.92 Gestational age at booking (weeks) (N¼ 86) 16.4 (4.3) 16.3 (4.5) 20.2 (5.9) 0.006 Serum folate at booking (ng/ml) (N¼ 86) 6.5 (3.3) 6.5 (3.0) 5.9 (3.3) 0.79 Results differences between the offspring enrolled in this study compared with the original sample in terms of their Baseline characteristics birthweight or gestational age at delivery (Supplementary The baseline characteristics of the pregnant women in the Table 1, available as Supplementary data at IJE online), al- three treatment groups were broadly comparable though we were unable to assess differences between char- (Table 1), with the exception of gestational age at booking, acteristics in adulthood, which were absent for those where women in the high-dose (5 mg) folic-acid supple- individuals who were not followed up. ment group were enrolled at a later gestation than in the other two groups. However, this trend was not apparent in the larger sample of pregnant women in the trial EWAS (Supplementary Table 1, available as Supplementary data 14,23 at IJE online), indicating that this difference is likely We next conducted an EWAS to investigate the long-term attributable to chance. An evaluation of baseline character- impact of the randomized folic-acid supplement-use inter- istics showed no clear differences between the mothers vention on offspring methylation in adulthood by evaluat- of offspring enrolled in this study compared with the ing the association between DNA methylation (normalized original sample (Supplementary Table 1, available as b value at each of the 460 617 CpG sites on the array) and Supplementary data at IJE online). folic-acid supplement use. We found an association at just The characteristics of the female offspring from mothers one CpG site, cg09112514, which withstood the FDR cor- in the three treatment groups were also similar (Table 2), rection where the intervention was associated with a 0.8% with the exception of age at sample and data collection, [95% confidence interval (CI)¼ 0.4, 1.2] reduction in whereby offspring in the placebo group were slightly older methylation at this site (Figure 1). A further 14 CpG sites on average, and body mass index (BMI), which was higher were found to surpass a less conservative p-value threshold –5 in the intervention groups. Given these differences, we of 110 and effects were generally not attenuated with included gestational age and age at sample and data collec- additional adjustment for gestational age and age at tion as covariates in subsequent models. As BMI may be a follow-up as covariates (Supplementary Table 2, available possible outcome or mediator of the intervention and as Supplementary data at IJE online). We observed a re- methylation change, it was not considered as a covariate. duction in methylation at 9 of these 14 CpG sites in the An evaluation of baseline characteristics showed no clear folic-acid supplement group vs placebo. Furthermore, Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 3 933 Table 2. Characteristics of participants included in this study (AFAST offspring, N¼ 86) Variable Category Placebo (n¼ 43) Folic-acid supplement 0.2 mg/day (n¼ 20) 5 mg/day (n¼ 23) Categorical n (%) n (%) n (%) Chi Age at follow-up (years) (N¼ 86) 46 6 (14.0) 14 (70.0) 6 (26.1) 47 33 (76.7) 6 (30.0) 17 (73.9) 48 4 (9.3) – – <0.001 Current smoking (N¼ 86) No 26 (60.5) 9 (45) 14 (60.9) Yes 17 (39.5) 11 (55) 9 (39.1) 0.47 Education (N¼ 84) O-level 19 (46.3) 10 (50) 12 (52.2) A-level/university 22 (53.7) 10 (50) 11 (47.8) 0.90 Alcohol intake Daily/weekly 19 (44.2) 12 (60.0) 12 (52.2) 0.56 Monthly 20 (46.5) 8 (40.0) 10 (43.5) Not at all 4 (9.30) 0 (0) 1 (4.4) Folic-acid supplements Yes 1 (2.3) 1 (5.0) 0 (0) No 42 (97.7) 19 (95.0) 23 (100) 0.56 Current medication Yes 31 (72.1) 15 (75.0) 17 (73.9) 0.97 No 12 (27.9) 5 (25.0) 6 (26.1) Health problems Yes 25 (58.1) 12 (60.0) 14 (60.9) 0.98 No 18 (41.9) 8 (40.0) 9 (39.1) Continuous Mean (SD) Mean (SD) Mean (SD) ANOVA BMI (kg/m )(N¼ 85) 24.2 (3.8) 26.5 (6.4) 27.8 (7.2) 0.04 Length of gestation (weeks) (N¼ 86) 40.9 (1.1) 39.9 (2.6) 40.3 (1.5) 0.07 Birthweight (g) (N¼ 86) 3333 (506) 3093 (620) 3269 (493) 0.25 Figure 1. Manhattan plot for the EWAS of in-utero folic-acid supplement use (low and high dose combined vs placebo) (N¼ 86). Solid line¼ FDR threshold for association to account for multiple testing; Dotted line¼ Bonferroni corrected threshold for association to account for multiple testing. Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 934 International Journal of Epidemiology, 2018, Vol. 47, No. 3 3a, available as Supplementary data at IJE online). Furthermore, for the high dose vs placebo model and low dose vs placebo model, 28 DMRs (Supplementary Table 7 and Figure 3b, available as Supplementary data at IJE on- line) and 2 DMRs (Supplementary Figure 3c and Table 8, available as Supplementary data at IJE online) were identi- fied, respectively. Notable regions included HLA-DPB2 and HLA-DPB1, which had low regional p-values in all models; PAX8, which was found to have low regional p- values in both the main and high-dose models; and VTRNA2–1, which was found to have the lowest regional p-value in the high-dose model. Functional analysis CpGs within DMRs identified using the main model (folic- acid supplement use vs placebo; 303 CpGs; Supplementary Table 6, available as Supplementary data at IJE online) Figure 2. Box plot for methylation at PDGFRA (cg09112514) in the differ- were most enriched for KEGG pathways relating to cancer ent intervention groups (N¼ 86). and regulation of the actin cytoskeleton (FDR-adjusted p- the same CpG site, cg09112514, was found to be most value for enrichment¼ 0.002) and GO terms related to kid- strongly associated with both the low and high doses when ney development, although it should be noted that no GO considered in separate models and was also strongly asso- terms were enriched after correction for multiple testing. ciated in the ordinal model of high dose, low dose and pla- Similarly, CpGs within DMRs identified using the high- and –7 cebo (p¼ 4.4710 )(Supplementary Tables 3–5, low-dose models were not enriched for any KEGG path- available as Supplementary data at IJE online) and illus- ways or GO terms after FDR correction (Supplementary trated a dose–response with regard to the intervention arm Tables 9 and 10, available as Supplementary data at IJE (Figure 2). online). We next investigated whether any of the CpG sites that CpGs within DMRs identified using the main model –5 surpassed the p-value threshold of 110 were identified as (303 CpGs; Supplementary Table 6, available as being either single-nucleotide polymorphism (SNP)-con- Supplementary data at IJE online) were highly enriched for founded or cross-hybridizing based on a comprehensive assess- CpGs within epiallelic regions (133 CpGs; Supplementary ment reported by Naeem et al. Two CpG sites, cg25455598 Table 11, available as Supplementary data at IJE online): and cg13682325, identified in the main analysis were flagged six CpGs within DMRs were also within epiallelic regions –10 by this study as lower-quality probes (Supplementary Table 2, (Chi-Square 353.89; p¼ 3.610 ). All six CpGs were available as Supplementary data at IJE online). within a DMR mapping to PAX8 (Chr2: 113992762– 113993314). CpGs within DMRs identified using the high- dose model (high dose vs placebo; 170 CpGs; DMRs Supplementary Table 7, available as Supplementary data Given the low power available in this study of just 86 indi- at IJE online) were also enriched for epialleles: 16 CpGs viduals to identify strong site-specific signals, we con- within DMRs were within epiallelic regions (Chi-square –36 sidered taking a regional approach to assess DMRs of the 5131.91; p¼ 4.310 ). Six out of 16 CpGs were in a genome in response to the intervention. This was further DMR mapping to PAX8 and the remaining 10 CpGs were supported by the Q-Q and Volcano plots of the site- in a DMR mapping to VTRNA2–1 (Chr5: 135414858– specific EWAS analysis that showed an inflation of p-val- 135416614). ues above that expected by chance in the main analysis of intervention vs placebo and particularly for the high dose Replication vs placebo model (Supplementary Figure 2, available as Supplementary data at IJE online). In the DMR analysis, Using summary findings from a neonatal EWAS of mater- we identified 46 DMRs with a Sidak p-value (multiple test- nal plasma folate, we performed a look-up of cg09112514 ing corrected)< 0.05 in the main model (folic-acid supple- (PDGFRA) and found that this did not replicate in that ment use vs placebo) (Supplementary Table 6 and Figure study (p¼ 0.96). We also performed a DMR analysis of Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 3 935 EWAS summary findings and then identified overlap be- Nevertheless, we attempted to replicate our findings in tween the DMRs identified in AFAST. A DMR at PAX8 the largest EWAS of maternal folate conducted to date –6 9 with a regional p-value of 2.0810 in AFAST had a re- (N¼ 1996). We found no clear association between ma- –10 gional p-value of 2.4610 in this independent replica- ternal cg09112514 and maternal plasma folate levels in tion sample (Supplementary Figure 4 and Table 12, this study. This lack of replication between studies may re- available as Supplementary data at IJE online). flect methylation profiling in different tissues (saliva vs cord blood), differences in the timing of methylation as- sessment (adults vs newborn infants), differences in the ex- Discussion posure measure (folic-acid supplement use vs maternal The results of this study, conducted within the context of a plasma folate) or other differences in the study design and randomized–controlled trial, suggest that maternal folic- populations investigated. Alternatively, the lack of repli- acid supplement use is associated with changes in DNA cation might indicate that this signal represents a false- methylation that persist for many years after in-utero ex- positive finding, given the small sample size of our study. posure. In saliva samples obtained from the offspring To combat the low power in our EWAS, we also took a 47 years after the trial was conducted, an effect of folic-acid regional approach to assess DMRs of the genome in re- supplement use on DNA methylation was identified at sponse to the intervention. We also assessed replication of –9 9 cg09112514 (p¼ 4.0310 ), a CpG site in the 5’ UTR of the DMRs in the results from the previous EWAS and this PDGFRA, in the main single-site EWAS analysis compar- time found some suggested replication of differential ing the intervention arms [low (0.2 mg) and high (5 mg) methylation at PAX8 in relation to maternal folate. In fur- dose folic acid] (N¼ 43) vs placebo (N¼ 43). Furthermore, ther support for the robustness of the DMR findings, both a dose–response reduction in methylation at this site was PAX8 and VTRNA2–1 are notable, as they are gene re- identified with regard to the intervention. We also identi- gions in which deemed ‘metastable epialleles’ have previ- fied 46 regions of the genome that were differentially ously been identified in relation to peri-conceptional 11,36 methylated in response to folic-acid supplement use, includ- nutrition. Metastable epialleles are defined as those ing HLA-DPB2, HLA-DPB1, PAX8 and VTRNA2–1. that are influenced by the in-utero environment, occur sys- PDGFRA encodes a platelet-derived growth factor re- temically and are highly stable over many years, which ceptor that has been linked with congenital neural tube de- are reflective of the methylation changes observed in this fects (NTDs) and isolated cleft palate. In particular, study. However, whereas these previous studies highlight mouse models have indicated that deregulated expression the importance of the periconceptional environment for es- of this gene leads to NTD formation and specific haplo- tablishing methylation marks at these metastable epialleles, types of the PDGFRA P1 promoter strongly affect rates of in this study, the intervention was initiated at an average 27,28 NTD genesis. Furthermore, methylation in this gene gestational age of 16 weeks. Similarly, our top site in region has recently been linked with subtypes of orofacial PDGFRA is implicated in NTDs but the critical period for cleft. It is therefore interesting that we identified differen- folate status on risk of NTDs is thought to be periconcep- tial methylation at a site in PDGFRA in relation to tional. Nonetheless, our results are consistent with previ- folic-acid supplement use, given the well-established link ous findings suggesting that environmentally induced between folate status in pregnancy and risk of such birth DNA-methylation change may not be limited to the peri- 30–32 37 defects. Although the effect size was small (0.8% re- conceptional period. duction in methylation at this site in the folic-acid supple- We observed a reduction in methylation at 10 of the 15 ment-use group), this does not preclude biological top CpG sites in the folic-acid supplement use vs placebo plausibility of this methylation difference, which may have groups (Supplementary Table 2, available as Supplementary subtle effects on health outcomes. data at IJE online) as well as a reduction in methylation at Differences in genome-wide DNA methylation have 287 of the 303 CpG sites contributing to the top DMRs in been evaluated in relation to maternal folate and other the folic-acid supplement use vs placebo groups micronutrient exposures in candidate gene studies and (Supplementary Table 6, available as Supplementary data at EWAS. However, unlike the study conducted here, most IJE online). Furthermore, there was widespread (although previous studies investigating maternal folate have meas- low-magnitude) hypomethylation among those CpGs not ured methylation only in newborn infants, with just one surpassing multiple testing correction in response to the study evaluating methylation at a later time point in in- intervention (Supplementary Figure 2, available as fancy. Therefore, our study is novel in investigating methy- Supplementary data at IJE online). These findings of hypo- lation change into adulthood in relation to this prenatal methylation in relation to folic-acid exposure are consistent 8,9,38 exposure. with previous findings, despite the fact that folate is a Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 936 International Journal of Epidemiology, 2018, Vol. 47, No. 3 methyl donor (and therefore might be anticipated to in- health outcomes remains to be determined. Further work to crease methylation levels at these CpG sites). Nonetheless, investigate whether these epigenetic changes translate into de- as was discussed previously, folic acid has been shown to tectable phenotypic differences is required. disturb the intracellular one-carbon metabolism by inhibit- ing methylenetetrahydrofolate reductase (MTHFR) activity 39 Supplementary data that may decrease DNA methylation and so our findings Supplementary data are available at IJE online are not inconsistent with respect to known biological pathways. Key strengths of this study include the experimental de- Funding sign in which this study was nested, with random alloca- This work was supported by the NIHR Bristol Biomedical Research tion, adequate concealment and evidence of good 14 Centre at the University Hospitals Bristol NHS Foundation Trust compliance. In addition, given the identified role of folic- and the University of Bristol. The views expressed in this publication acid supplement use in the prevention of NTDs, an RCT to are those of the authors and not necessarily those of the NHS, the determine the long-term effects of in-utero exposure to National Institute for Health Research or the Department of Health. folic acid vs placebo would no longer be ethical. This his- R.C.R., G.C.S., N.K., T.G., G.D.S. and C.L.R. work in a unit that receives funds from the University of Bristol and the UK Medical torical study therefore provides a unique opportunity to in- Research Council (MC_UU_12013/1, MC_UU_12013/2 and vestigate the impact of folic-acid supplements in pregnancy MC_UU_12013/8). This work was also supported by CRUK (grant on long-term DNA-methylation changes in a trial setting. number C18281/A19169) and the ESRC (grant number ES/ It also illustrates a successful attempt of enrolling individ- N000498/1). C.M.T. is supported by a Wellcome Trust Career Re- uals into a study through data-record linkage approxi- entry Fellowship (grant number 104077/Z/14/Z). mately 47 years after the initial trial, which allowed us to Conflict of interest: None declared. look at long-term effects of in-utero exposure to folic acid. Furthermore, participants included in this study were simi- References lar to the original study sample with respect to the baseline 1. Scholl TO, Johnson WG. Folic acid: inﬂuence on the outcome of characteristics, indicating that the randomized nature of pregnancy. Am J Clin Nutr 2000;71(Suppl 5):1295S–303. the intervention was preserved. This study also highlights 2. Barua S, Kuizon S, Junaid MA. Folic acid supplementation in the value of saliva as a non-invasive sample on which to pregnancy and implications in health and disease. J Biomed Sci perform DNA-methylation profiling and the value of 2014;21:77. methylation profiles as a biosocial archive for historical 3. Anderson OS, Sant KE, Dolinoy DC. Nutrition and epigenetics: an interplay of dietary methyl donors, one-carbon metabolism exposure. and DNA methylation. J Nutr Biochem 2012;23:853–59. Limitations include the small sample size of this study, 4. Waterland RA, Michels KB. Epigenetic epidemiology of the de- which might have generated spurious associations at the velopmental origins hypothesis. Annu Rev Nutr 2007;27: single-CpG level in the EWAS as a result of low power, al- 363–88. though the replication of DMRs in independent studies 5. Wolff GL, Kodell RL, Moore SR, Cooney CA. Maternal epigen- supports the robustness of these findings. A further limita- etics and methyl supplements affect agouti gene expression in tion relates to the fact that this study was conducted in fe- Avy/a mice. FASEB J 1998;12:949–57. 6. Steegers-Theunissen RP, Obermann-Borst SA, Kremer D et al. male offspring only and individuals residing within the Periconceptional maternal folic acid use of 400 microg per day is Grampian area, meaning results may not be generalizable. related to increased methylation of the IGF2 gene in the very As the offspring response rate was 28%, selection bias young child. PLoS One 2009;4:e7845. could have played a role in our results, although an evalu- 7. Cooper WN, Khulan B, Owens S et al. DNA methylation proﬁl- ation of baseline characteristics revealed no clear differ- ing at imprinted loci after periconceptional micronutrient sup- ences between the mothers or offspring of those enrolled in plementation in humans: results of a pilot randomized controlled this study compared with the original sample. trial. FASEB J 2012;26:1782–90. Overall, the results of this study suggest that maternal 8. Amarasekera M, Martino D, Ashley S et al. Genome-wide DNA methylation proﬁling identiﬁes a folate-sensitive region of differ- folic-acid supplement use, even after the periconceptional ential methylation upstream of ZFP57-imprinting regulator in period, is associated with changes in DNA methylation in the humans. FASEB J 2014;28:4068–76. offspring that persist for many years after in-utero exposure. 9. Joubert BR, den Dekker HT, Felix JF et al. Maternal plasma fol- These methylation changes are located in genes implicated in ate impacts differential DNA methylation in an epigenome-wide pathways related to cancer, metabolism and infection, and meta-analysis of newborns. Nat Commun 2016;7:10577. therefore may mediate long-term effects of folic-acid exposure 10. Heijmans BT, Tobi EW, Stein AD et al. Persistent epigenetic dif- 14,42,43 in pregnancy. However, the causal relevance of these ferences associated with prenatal exposure to famine in humans. methylation changes with regard to these developmental and Proc Natl Acad Sci USA 2008;105:17046–49. Downloaded from https://academic.oup.com/ije/article/47/3/928/4931210 by DeepDyve user on 16 July 2022 International Journal of Epidemiology, 2018, Vol. 47, No. 3 937 11. Waterland RA, Kellermayer R, Laritsky E et al. Season of 28. Zhu H, Wicker NJ, Volcik K et al. Promoter haplotype combin- conception in rural gambia affects DNA methylation at ations for the human PDGFRA gene are associated with risk of putative human metastable epialleles. PLoS Genet 2010;6: neural tube defects. Mol Genet Metab 2004;81:127–32. e1001252. 29. Sharp GC, Ho K, Davies A et al. Distinct blood DNA methylation 12. Hall MH, Pirani BB, Campbell D. The cause of the fall in serum proﬁles in subtypes of orofacial cleft. Clin Epigenetics 2017;9:63. folate in normal pregnancy. Br J Obstet Gynaecol 1976;83: 30. Group MVSR. Prevention of neural tube defects: results of the 132–36. Medical Research Council Vitamin Study. Lancet 1991;338: 13. Hall MH. Folic acid deﬁciency and congenital malformation. 131–37. J Obstet Gynaecol Br Commonw 1972;79:159–61. 31. Czeizel AE, Dudas I. Prevention of the ﬁrst occurrence of neural- 14. Charles DHM, Ness AR, Campbell D, Smith GD, Whitley E, tube defects by periconceptional vitamin supplementation. N Hall MH. Folic acid supplements in pregnancy and birth out- Engl J Med 1992;327:1832–35. come: re-analysis of a large randomised controlled trial and up- 32. Relton CL, Wilding CS, Lafﬂing AJ et al. Low erythrocyte folate date of Cochrane review. Paediatr Perinat Epidemiol 2005;19: status and polymorphic variation in folate-related genes are asso- 112–24. ciated with risk of neural tube defect pregnancy. Mol Genet 15. General Register Ofﬁce. Classiﬁcation of Occupations. London: Metab 2004;81:273–81. HMSO, 1966. 33. van Mil NH, Bouwland-Both MI, Stolk L et al. Determinants of 16. Ayorinde AA, Wilde K, Lemon J, Campbell D, Bhattacharya S. maternal pregnancy one-carbon metabolism and newborn Data resource proﬁle: the Aberdeen Maternity and Neonatal human DNA methylation proﬁles. Reproduction 2014;148: Databank (AMND). Int J Epidemiol 2016;45:389–94. 581–92. 17. Dedeurwaerder S, Defrance M, Calonne E, Denis H, Sotiriou C, 34. Richmond RC, Joubert BR. Contrasting the effects of intra- Fuks F. Evaluation of the inﬁnium methylation 450K technol- uterine smoking and one-carbon micronutrient exposures on off- ogy. Epigenomics 2011;3:771–84. spring DNA methylation. Epigenomics 2017;9:351–67. 18. Relton CL, Gaunt T, McArdle W et al. Data resource proﬁle: 35. Khulan B, Cooper WN, Skinner BM et al. Periconceptional ma- Accessible Resource for Integrated Epigenomic Studies (ARIES). ternal micronutrient supplementation is associated with wide- Int J Epidemiol 2015;44:1181–90. spread gender related changes in the epigenome: a study of a 19. McGregor K, Bernatsky S, Colmegna I et al. An evaluation of unique resource in the Gambia. Hum Mol Genet 2012;21: methods correcting for cell-type heterogeneity in DNA methyla- 2086–101. tion studies. Genome Biol 2016;17:84. 36. Silver MJ, Kessler NJ, Hennig BJ et al. Independent genomewide 20. Barﬁeld RT, Kilaru V, Smith AK, Conneely KN. CpGassoc: an R screens identify the tumor suppressor VTRNA2-1 as a human function for analysis of DNA methylation microarray data. epiallele responsive to periconceptional environment. Genome Bioinformatics 2012;28:1280–81. Biol 2015;16:118. 21. Pedersen BS, Schwartz DA, Yang IV, Kechris KJ. Comb-p: soft- 37. Tobi EW, Lumey LH, Talens RP et al. DNA methylation differ- ware for combining, analyzing, grouping and correcting spatially ences after exposure to prenatal famine are common and timing- correlated P-values. Bioinformatics 2012;28:2986–88. and sex-speciﬁc. Hum Mol Genet 2009;18:4046–53. 22. Phipson B, Maksimovic J, Oshlack A. missMethyl: an R package 38. Gonseth S, Roy R, Houseman EA et al. Periconceptional folate for analyzing data from Illumina’s HumanMethylation450 plat- consumption is associated with neonatal DNA methylation form. Bioinformatics 2016;32:286–88. modiﬁcations in neural crest regulatory and cancer development 23. Taylor CM, Atkinson C, Penfold C et al. Folic acid in pregnancy genes. Epigenetics 2015;10:1166–76. and mortality from cancer and cardiovascular disease: further 39. Smith DE, Hornstra JM, Kok RM, Blom HJ, Smulders YM. follow-up of the Aberdeen folic acid supplementation trial. Folic acid supplementation does not reduce intracellular homo- J Epidemiol Community Health 2015;69:789–94. cysteine, and may disturb intracellular one-carbon metabolism. 24. Naeem H, Wong NC, Chatterton Z et al. Reducing the risk of false Clin Chem Lab Med 2013;51:1643–50. discovery enabling identiﬁcation of biologically signiﬁcant genome- 40. Teschendorff AE, Yang Z, Wong A et al. Correlation of wide methylation status using the HumanMethylation450 array. smoking-associated DNA methylation changes in buccal cells BMC Genomics 2014;15:51. with DNA methylation changes in epithelial cancer. JAMA 25. Rattanasopha S, Tongkobpetch S, Srichomthong C, Siriwan P, Oncol 2015;1:476–85. Suphapeetiporn K, Shotelersuk V. PDGFRa mutations in humans 41. Relton CL, Hartwig FP, Davey Smith G. From stem cells to the with isolated cleft palate. Eur J Hum Genet 2012;20:1058–62. law courts: DNA methylation, the forensic epigenome and the 26. Helwig U, Imai K, Schmahl W et al. Interaction between undu- possibility of a biosocial archive. Int J Epidemiol 2015;44: lated and patch leads to an extreme form of spina-biﬁda in 1083–93. double-mutant mice. Nat Genet 1995;11:60–63. 42. Campbell NRC. How safe are folic acid supplements? Arch 27. Joosten PH, Toepoel M, Mariman EC, Van Zoelen EJ. Promoter Intern Med 1996;156:1638–44. haplotype combinations of the platelet-derived growth factor 43. Burdge GC, Lillycrop KA. Folic acid supplementation in preg- alpha-receptor gene predispose to human neural tube defects. nancy: are there devils in the detail? Br J Nutr 2012;108: Nat Genet 2001;27:215–17. 1924–30.
International Journal of Epidemiology – Oxford University Press
Published: Jun 1, 2018
Keywords: follow-up; saliva
Access the full text.
Sign up today, get DeepDyve free for 14 days.