The characteristics of subjects used in these analyses (n
= 380) are shown in Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103927
). Because subjects were selected based on SGA status and matched to non-SGA infants on sex, gestational age, and maternal age, we observed no significant differences in those characteristics among birth weight groups (SGA, AGA, LGA). Significant differences were noted between groups for maternal ethnicity and maternal tobacco smoking during pregnancy, with a greater proportion of non-Caucasian infants (p
= 0.01) and infants with exposure to tobacco smoke in the SGA category (p
= 0.03). The vast majority of mothers reported prenatal vitamin use; only three subjects reported alcohol consumption during pregnancy.
Associations between infant growth and repetitive element methylation.
DNA methylation and infant growth.
Mean LINE-1 and AluYb8 methylation levels differed significantly by birth weight status [see Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103927
< 0.0001]. Mean LINE-1 and AluYb8 methylation levels were also found to be weakly to moderately correlated, with a Pearson correlation of 0.29 [95% confidence interval (CI): 0.20, 0.38]. To examine whether “global methylation” represented as an average of gene-associated methylation identified in a genomewide manner was associated with infant growth, we tested for an association between birth weight percentile and mean methylation across the 26,486 autosomal loci measured using the Illumina Infinium array in a subset of 184 placenta samples. Mean methylation across the 26,486 autosomal CpG loci was not significantly associated with the birth weight groups (see Supplemental Material, ). There were no significant differences in the distributions of the demographic characteristics of this subsample and the larger population examined for LINE-1 and AluYb8 methylation extent (data not shown).
To further evaluate the association between infant birth weight and repetitive element methylation identified in the univariate analyses, controlled for confounders, we used multiple linear regression. These models demonstrate that with a 10% increase in LINE-1 mean methylation levels, birth weight percentile significantly increased by 9.7 (95% CI: 2.9, 16.6), and with a 10% increase in AluYb8 mean methylation levels, birth weight percentile significantly increased by 14.5 (95% CI: 4.9, 24.0) (). Both models were adjusted for infant sex and maternal age, BMI, ethnicity, and tobacco, alcohol and prenatal vitamin use during pregnancy. Infant sex and maternal ethnicity were also significant predictors of birth weight percentile (); all other covariates were not significant.
DNA methylation and in utero exposures.
presents the results of our exploration of the differences in the measures of DNA methylation by various in utero exposures. Only AluYb8 methylation levels differed by infant sex (p = 0.04), with male infants having slightly higher mean AluYb8 methylation (mean ± SD = 65.3 ± 3.3) than did female infants (64.6 ± 3.3). Furthermore, we found that mean AluYb8 levels differed significantly by maternal tobacco use during pregnancy (p < 0.01), whereas mean LINE-1 levels significantly differed only by maternal alcohol use during pregnancy (p = 0.02; ). Array-based mean methylation across the 26,486 CpG loci did not differ by any of the in utero exposures explored.
Relationships between repetitive element and gene associated methylation.
To investigate the relationship between repeat element methylation and gene-associated methylation, we clustered the 26,486 autosomal CpG loci interrogated on the Infinium HumanMethylation27 BeadArray into 16 methylation classes based on their methylation pattern using an RPMM [for the heat map of the 16 classes, see Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103927
)]. This data-driven clustering allowed us to define stable classes of CpG loci demonstrating similar patterns of methylation, which we used to examine associations between methylation of these classes of CpGs and LINE-1 or AluYb8 methylation. By clustering, we hoped to uncover biologically meaningful groups of CpG loci, with similar methylation patterns, which may share biological characteristics such as comparable functionality in certain developmental pathways or location in similar genomic regions.
Figure 1 Association of LINE-1 (A) and AluYb8 (B) methylation with mean RPMM class methylation. The colored dots indicate the degree of average CpG class methylation, as indicated by the key. The red dashed lines represent the null limits for the permutation distribution (more ...)
The association between LINE-1 or AluYb8 methylation and average methylation within the RPMM-derived classes for each of the 184 individuals in the study is presented in . There was an overall significant relationship between AluYb8 methylation and RPMM class methylation (p = 0.01); no association was observed between LINE-1 and RPMM class methylation. When considering class-specific relationships, a positive association was noted between AluYb8 methylation and methylation of CpGs encompassing the low to intermediately methylated class 9 (). Additionally, classes 7, 11, and 12 were marginally positively associated with AluYb8 methylation levels, with CpGs exhibiting low to intermediate methylation.
To examine similarities in the genomic context of the CpGs making up these data-driven RPMM classes, we examined the proportion of CpG loci associated with specific DNA sequence features within each class, including location within a repetitive element (LINE-1, LINE-2, Alu, or MIR), within a CGI, or having the associated gene considered a PcG protein target gene (). As expected, a greater proportion of CpGs in the relatively methylated (rightmost) classes 9–12 are localized in repetitive elements (). Conversely, a high proportion of CpGs in the relatively unmethylated classes 1–6 and 8 are in CGIs (). There was also interclass variation in the frequency of CpGs association with PcG target genes, with classes 7 and 9 in particular having a much higher proportion than the others (). Class 9, which was associated with AluYb8 methylation extent, contained a high proportion of CpG loci located within CGIs (81%) and whose genes are considered PcG target genes (32%).
Figure 2 Frequency of sequence features associated with RPMM class CpG loci: percentage of CpG loci found in LINE-1 (A), LINE-2 (B), Alu (C), and MIR (D), within a CGI (E), or having the associated gene considered a PcG protein target gene (F). The colored dots (more ...)
Motivated by the variation in sequence features (bioinformatic attributes) observed between RPMM-based classes, we next performed a bioinformatically informed clustering of CpG loci into classes, grouping CpGs according to their sequence context: presence within a CGI, LINE-1, LINE-2, Alu, or MIR element, or PcG target gene and proximity (≤ 1,000 bases) to any TFBS. This resulted in 41 distinct bioinformatically derived classes, containing at least one CpG, based on bioinformatic attributes [for the distribution of CpG loci by bioinformatically derived class, see Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103927
)]. The association between LINE-1 and AluYb8 repetitive element methylation and aggregate methylation values of the autosomal CpG loci within each of the 41 bioinformatic classes is depicted in . There was an overall significant relationship between AluYb8 methylation extent and bioinformatically derived CpG class methylation (p
< 0.0001) but not between LINE-1 methylation and bioinformatically derived CpG class methylation. When considering class-specific relationships, there was a positive association between AluYb8 methylation levels and mean methylation across loci allocated within a PcG target gene and LINE-2 element and proximal to a TFBS (class PcG/LINE-2/TFBS) and loci located within a CGI and PcG target gene and proximal to a TFBS (class CpG/PcG/TFBS).
Figure 3 Association of LINE-1 (A) and AluYb8 (B) methylation with bioinformatically derived CpG class methylation. The colored dots indicate the degree of average class methylation. The red dashed lines represent the null limits for the permutation distribution (more ...) GSEA analysis.
To expand on the bioinformatically informed analysis, which suggested that methylation of CpG loci proximal to a TFBS were associated with AluYb8 methylation, we performed a GSEA using specific TFBS, obtained from the UCSC Genomes Browser. Specifically, each of the 26,486 autosomal CpG loci are simultaneously evaluated for their association with LINE-1 or AluYb8 methylation extent and the genomic locations or TFBS that are overrepresented within 1 kb of the loci [for the GSEA results, see Supplemental Material, (http://dx.doi.org/10.1289/ehp.1103927
)]. Only three TFBS genes were overrepresented among loci whose methylation was associated with LINE-1 methylation: MEF2
, and POU2F1
. For those overrepresented among loci associated with AluYb8 methylation, there were a number of developmentally related TFBS genes (CUX1
, MYOD, OCT), as well as those involved in cell processes (ACTR3B
), gene expression (NFE2L
), lipid/sterol homeostasis (NR3C1
), and immune modulation (GATA1
). Only POU2F1
was found to be overrepresented among loci associated with both LINE-1 and AluYb8 methylation.