Study population

Peripheral blood DNA samples were available for 333 mothers along with cord blood DNA samples from 454 neonates, including 137 mother-child pairs. These subjects were recruited as part of the NCCGP, between 1996 and 2003, at a single maternity unit in West Cumbria, UK [21]. Mothers completed a detailed questionnaire and provided blood samples during their first antenatal clinic (mean (SD) gestation = 11.3 (5.4) weeks). Cord blood from neonates was collected at birth (mean (SD) gestation = 39.5 (1.4) weeks). Written informed consent was obtained from all participating mothers recruited during pregnancy. The consent obtained included use of their own biological samples and those of their child (including DNA) for epidemiological studies. Ethical approval to undertake this study was obtained from the Cumbria and Lancashire Local Research Ethics Committee.

Genotype analysis for DNMT1 and DNMT3B

Eleven SNPs spanning the DNMT1 and DNMT3B genes were genotyped across all study samples by K-Bioscience using their proprietary KasPar system (www.kbioscience.co.uk/). These included 2 pairwise tagging SNPs for DNMT1 (rs2290684, rs2241531) and 8 pairwise tagging SNPs for DNMT3B (rs6119954, rs992472, rs2424928, rs2424932, rs6058897, rs437302, rs406193) as described previously by Cebrian et al [15]. In addition, two DNMT3B SNPs (rs1569686, rs2424913) located within the promoter/5' gene region and previously associated with altered promoter activity were selected for investigation [16,19,25,26]. Polymorphisms demonstrating genotyping success rates less than 80% and minor allele frequencies (MAF) less than 5% were removed prior to analysis. Hardy Weinberg Equilibrium (HWE) was assessed for each SNP; those demonstrating deviations from HWE (P < 0.05) were removed prior to analysis.

LUMA assay to determine global DNA methylation

Genomic DNA methylation was measured as part of our previously published study [27] and was available for a subset of individuals in the current analysis (Table 1). The luminometric methylation assay (LUMA) protocol has been described in detail previously [27,28]. Briefly, 200ng DNA was incubated for 4h at 37°C with 1) methylated (EcoRI plus MspI) and 2) unmethylated (EcoRI plus HpaII) restriction enzyme sets in 20µl volume reactions containing 5 units of each enzyme (New England Biolabs) and 2µl of Tango buffer (Fermentas). Digests were carried out in triplicate for each sample and analysed by Pyrosequencing on a Pyromark™ MD system. Data are presented as a methylation ratio (defined as (HpaII/EcoRI)/(MspI/EcoRI)) with a higher ratio indicative of less methylated DNA.

Bisulfite Pyrosequencing for candidate gene DNA methylation analysis

In addition to global methylation three candidate genes, namely IGF2, IGFBP3 and ZNT5, were selected for analysis. These genes were selected as they represent three contrasting degrees of methylation; IGF2 is an imprinted locus with mean methylation ~50%, IGFBP3 is constitutively methylated at low levels (~5%) and ZNT5 is constitutively methylated at high levels (~90%). Furthermore, we have previously shown that DNA methylation levels at these three loci are variable and influenced by a combination of genetic and environmental factors [27].

Methylation levels at the three candidate loci were measured using bisulfite Pyrosequencing as part of our previously published study [27] and made available for the current analysis. Table 1 reports the number of CpG sites and the number of samples assessed for the three loci. Pyrosequencing and quality control (QC) methods were described in detail previously [27]. Briefly, bisulfite conversion of 2µg of genomic DNA was performed using EZ DNA Methylation Gold™ kit (Zymo Research) following the manufacturer’s protocol. Quantitative bisulphite Pyrosequencing (Qiagen, UK) with Pyro Q-CpG™ Software (version 1.0.6.) was subsequently used to determine the percentage methylation at individual CpG sites within the differentially methylated region (DMR) of IGF2 (NG_008849.1; 6098-6375), the IGFBP3 promoter (NT_007819.17; 45951336-45951104) and the ZNT5 promoter (NT_006713.15; 18983340-18983714). 0.2µg of bisulfite treated DNA was added to the first PCR reaction with 12.5µl Hot Star Taq mastermix (Qiagen) and optimised primer concentrations. Complete primer and PCR conditions are described in Table S1 in File S1. Subsequent sequencing reactions followed the manufacturer’s protocol. Assays were assessed for amplification bias and reliability as described previously [27,29]. Zero and 100% in vitro methylated controls were run routinely alongside samples as internal controls.

CpG sites with poor success rates across the study population (<80% success) were removed prior to analysis. Similarly, CpG sites demonstrating extreme measures of methylation with limited variability across the study population (median methylation = 0% or 100%) were dropped prior to analysis. The genotypic and methylation data can be made available upon request to the authors.

Data analysis

Linkage disequilibrium (LD) between genetic variants was assessed using Haploview software [30]. Correlation in methylation levels at individual CpG sites spanning the 3 candidate loci was assessed using non-parametric Spearman’s rank correlation. Those loci demonstrating at least modest correlation (rho > 0.6) across individual CpG sites were also analysed using the mean percentage methylation for that locus. Non-parametric tests for trend were performed to explore relationships between methylation levels and SNPs under an additive model. Polymorphisms with low MAF (i.e. <0.15) were analysed under a dominant model, with respect to the minor allele, using Kruskal Wallis tests. Comparisons between methylation levels in maternal-infant pairs were made by non-parametric Spearman’s rank correlation and Wilcoxon signed-rank tests. Uncorrected p-values are presented throughout with p ≤ 0.05 considered statistically significant. Corrections were applied post hoc when interpreting the results to assess whether nominally significant observations were robust to correction for multiple testing. Bonferroni as well as Benjamini and Hochberg corrections were applied in this context. All statistical analyses were performed in Stata (version 10, Statacorp).

Power of the study to detect associations between DNMT genotype and DNA methylation

With 333 maternal samples, there was 94% power to detect a 1.4 fold change in global methylation assuming an additive model with a 5% MAF at the 0.05 significance level. This power calculation was based on data reported previously [27] in which the mean (SD) methylation ratio across maternal samples was 0.46 (0.3). Under the same model there was 84% power to detect an absolute change of 4% in IGF2 methylation, 87% power for an absolute change of 4% in IGFBP3 methylation and 81% power for an absolute change of 6% in ZNT5 methylation. Again these calculations were based on previously reported data in which the mean (SD) methylation percentages for IGF2, IGFBP3 and ZNT5 were 45 (8)%, 7 (7)%, 90 (12%), respectively [27]. There was greater power to detect the same effects in the larger neonate cohort (n=454). Power calculations were performed in Quanto [31,32].

Cohort characteristics

Peripheral blood DNA samples were available for 333 mothers along with cord blood DNA samples from 454 neonates. The average age at the time of sample collection from the mothers was 29 years and 52% of the neonates were male (see Table S2 in File S1). In addition, these samples comprised 137 mother-child pairs. These pairs were representative of their larger sample groups (see Table S2 in File S1). SNPs mapping to the DNMT1 and DNMT3B genes were genotyped across all samples. Pyrosequencing data was previously generated on smaller subsets of these samples (Table 1).

Genotype data across the DNMT1 and DNMT3B genes

Eleven SNPs, with MAF >0.05, were genotyped successfully in both neonates and adult women (Figure 1). Three of these SNPs (DNMT1 rs2241531, DNMT3B rs437302 and rs406193) had relatively low MAF (<0.15) and underwent exploratory analyses using a dominant rather than an additive model. One DNMT3B SNP (rs6119954) did not conform to HWE (p = 0.003) in the neonate group and was excluded from further analysis. The two putative functional SNPs rs1569686 and rs2424913 demonstrated strong LD (r²>0.8) with the DNMT3B rs992472 and rs2424928 tagging SNPs (Figure 1). In total, 10 SNPs were taken forward for analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Linkage disequilibrium and SNP frequencies across DNMT genes in mothers and infants.

Alleles = Major: Minor, GT % = genotyping success rate (mothers total N = 333, infants total N = 454), MAF = minor allele frequency, HWE p = Hardy Weinberg Equilibrium p-value. Single underscore = putative functional SNPs; strikethrough = excluded from analysis.

https://doi.org/10.1371/journal.pone.0076506.g001

Correlation between CpG sites within the IGF2, IGFBP3 and ZNT5 genes

In addition to global methylation, methylation of 3 CpG sites in the DMR of the imprinted gene IGF2 and 5 CpG sites in both the IGFBP3 and ZNT5 promoter regions was quantified (Table 1). CpG site 4 in ZNT5 was very highly methylated with little inter-individual variation (median methylation = 100%) and was excluded from further statistical analysis. CpG sites 5 and 6 in ZNT5 were dropped from further statistical analysis due to poor success rates (<80%) in both maternal and neonate samples. The remaining CpG sites (3 within IGF2, 5 within IGFBP3 and 2 with ZNT5) demonstrated high success rates in both maternal and neonate samples (Table 1) and were taken forward for analysis. In addition, there were modest to strong correlations (average rho: 0.64-0.82) demonstrated between methylation levels at the CpG sites mapping to the IGF2 and the IGFBP3 loci in both mothers and infants (Table 1). Hence, mean methylation levels across these two loci were calculated and these means were also used in further analyses. Weaker correlations (rho<0.6) were found between methylation levels of the CpG sites measured within the ZNT5 locus therefore only methylation values for individual CpG sites were assessed further. Measurement of global DNA methylation was less successful in the maternal samples (74%) compared to that in the neonate samples (90%). However, we chose to retain this measure for analysis but acknowledge that any findings must be viewed with caution.

Predictors of maternal methylation

Table 2 summarises data for those variables for which there were statistically significant (p≤0.05) associations between maternal genetic factors and maternal methylation levels. See Table S3 in File S1 for all results. The minor G allele of the DNMT1 SNP rs2290684 was associated with increased methylation of CpG site 3 in IGF2 following an additive model. Similar trends were demonstrated at other IGF2 CpG sites, including the mean methylation across all CpG sites measured, but these did not reach the p≤0.05 threshold for statistical significance (Table S3 in File S1). Rs2290684 was also associated with ZNT5 CpG site 3 methylation levels.

Predictors of infant methylation: infant genotypes

Table 2 summarises data for those variables for which there were statistically significant (p≤0.05) associations between infant genetic factors and infant methylation levels. See Table S4 in File S1 for all results. Increased methylation at CpG sites 3 and 5 of the IGFBP3 locus was associated with the minor G allele of the DNMT1 rs2290684 variant, whereas carriers of the minor G allele of rs2241531 had lower methylation at CpG site 5 compared with wild-types (Table 2). Several SNPs within the infants’ DNMT3B gene were associated with infant IGFBP3 methylation (Table 2). In particular, carriage of the minor alleles for the putative functional SNPs (rs1569686, rs2424913) along with the minor alleles of the pairwise tagging SNPs rs992472 and rs2424923 was accompanied by increased methylation at CpG sites 3 and/or 5. The most statistically significant association being demonstrated between the putative functional SNP rs2424913 and IGFBP3 CpG site 3 (p=0.012). In contrast, carriage of the minor alleles at both rs6058897 and rs406193 were associated with decreased methylation at CpG sites 3 and 5.

Predictors of infant methylation: maternal methylation and genotypes

Greater variability was demonstrated for measures of global DNA methylation across the maternal samples compared to their paired infant samples (Table 3). Small yet notable differences were observed in methylation levels at CpG sites mapping to IGF2 and ZNT5 across the maternal and infant paired samples (Table 3). For both genes methylation levels were consistently higher in infant compared to maternal samples. Nonetheless, there was limited pairwise correlation (rho < 0.6) in methylation levels, both global and loci specific, demonstrated between mothers and their infants within the 137 mother-infant pairs (Table 3). Similarly, there were no observable differences in methylation levels at CpG sites mapping to IGFBP3 between the mothers and their infant pair.

The impact of maternal genotypes on infant methylation levels was explored and statistically significant findings are presented in Table 2. See Table S5 in File S1 for all results. Maternal genotypes at DNMT1 rs2241531 and DNMT3B rs6058897 were associated with infant global DNA methylation levels. In both instances, the minor alleles were associated with lower DNA methylation levels. The minor alleles for the DNMT3B putative functional SNPs (rs1569686, rs2424913) and pairwise tagging SNPs rs992472 and rs2424923 were associated with increased IGFBP3 methylation levels at CpG sites 1 and 5 in the infant. In particular the associations involving rs1569686, rs2424913 and rs2424928 with methylation levels at CpG site 5 were equally significant at p = 0.003, p = 0.002 and p = 0.001, respectively. However, given the large number of tests performed (10 SNPs x 13 methylation measures = 130 tests, not accounting for the 3 individual subgroups) none of these associations remain significant at the conservative Bonferroni corrected significance threshold of p = 3x10^-4 (i.e. 0.05/130). Similarly, these associations are no longer significant following the less conservative Benjamini & Hochberg false discovery correction (corrected p-value = 0.13 for each of the three SNPs).