Background
It has been hypothesized that prenatal environmental exposures may cause lasting epigenetic changes during child development, leading to adverse health outcomes in later life [
1,
2]. Epigenetics refers to heritable changes regulating gene expression that do not affect the DNA base pair sequence. DNA methylation is the most commonly studied epigenetic mark [
3‐
6]. This modification involves attachment of a methyl group to the cytosine base within cytosine-guanine dinucleotides, also known as ‘CpG methylation’. Higher CpG methylation (hypermethylation) within the promoter region of a gene can reduce gene expression [
7].
Our understanding of how the human methylome is organized is rapidly evolving and current literature outlines a classification system for CpG sites, highlighting that their function may be intimately related to their location in the gene [
8,
9]. For example, approximately 70% of gene promoter regions are thought to contain a CpG island – a region with densely concentrated CpGs that typically have low levels of methylation [
10]. CpG sites flanking the island are located in regions termed north shore (upstream, 5′ end) and south shore (downstream, 3′ end) and are thought to be particularly important in regulating gene expression [
6,
9].
There is a growing interest in examining interactions between environmental and genetic factors on differential CpG methylation. A classic example is from animal studies on the
Agouti mouse, where hypomethylation of the intracisternal A particle (IAP) increases expression of the
Agouti gene, resulting in yellow coat color and obese phenotype. Using this model, Waterland et al. [
11] showed that inheritance of the
Agouti gene was associated with trans-generational amplification of obesity and that maternal methyl-donor supplementation could prevent this effect.
In humans, DNA methylation has been proposed to mediate direct intra-uterine associations between maternal and offspring phenotypes. Differential DNA methylation has been reported when assessing offspring exposed
in utero to extreme maternal undernutrition [
12,
13], maternal morbid obesity [
14] and less extreme maternal underweight and maternal obesity [
15]. However, several important challenges remain. There is an ongoing effort to determine the causal direction between DNA methylation and an offspring phenotype. In their 2016 study, Richmond et al. apply a causal framework to parse out whether
HIF3A methylation has a causal effect on BMI or vice versa [
16]. Their results argue for the potential of a phenotype to affect methylation status and highlight the potential for inter-generational influence of maternal BMI on offspring methylation, possibly confounding the offspring
HIF3A methylation and obesity association.
Another important challenge has been replication of results from epigenome-wide association studies (EWAS) studies. For example, the EWAS study by Sharp et al. identified 28 CpGs in newborns that were associated with maternal pre-pregnancy BMI. Four of these hits had previously been reported in literature, but their results did not replicate the direction and magnitude of the earlier analyses [
17]. Additionally, in their EWAS, Aslibekyan et al. found only 8 CpGs in 3 genes (
CPT1A,
PHGDH,
CD38) associated with body mass index (BMI) in adults that withstood replication and multiple testing adjustment [
18].
To avoid the limitations of multiple testing, candidate genes can be selected
a priori. With respect to obesity development, the peroxisome proliferator-activated receptor γ
(PPARγ) gene may play a critical role, functioning as the only gene that is both necessary and sufficient for fat cell production [
19,
20].
PPARγ upregulation has been linked to improvement of critical metabolism-related hormones (increased adiponectin and decreased leptin) and increased insulin sensitivity at the expense of greater body weight in adults [
21] and animals [
22‐
24]. Importantly, while methylation affects
PPARγ expression in animal and
in vitro studies, only limited human data on
PPARγ methylation, its relationship with obesity and/or with perinatal factors are available [
25,
26].
In the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS) cohort, we have previously examined a subset of
PPARγ CpG sites and their relationship with gene expression in a cohort of children with a high prevalence of obesity [
27]. We reported that hypomethylation of the
PPARγ CpG site cg10499651 was associated with increased
PPARγ expression as measured by both real-time polymerase chain reaction (RT-PCR) and nCounter assays. In the current investigation, we build on this finding and add to current data gaps on
PPARγ methylation and its relationship with obesity. We use the Illumina 450 k assay to examine methylation of 23 CpG sites spanning the
PPARγ promoter and gene body regions in children at birth (
N = 373) and at 9 years (
N = 245) and 1) analyze the correlation structure between the 23
PPARγ CpG sites, 2) characterize associations between perinatal factors, including maternal pre-pregnancy BMI, and
PPARγ methylation at birth and at 9 years, and 3) examine associations between
PPARγ methylation, child birthweight and BMI at 9 years.
Discussion
In this study, we aimed to address several knowledge gaps on the 1) correlation structure of PPARγ methylation, 2) relationships between perinatal factors and PPARγ methylation, and 3) associations between PPARγ methylation, birth weight and child BMI. We found that PPARγ methylation displays a highly conserved pattern and report on two methylation blocks comprised of sites 1–3 (block 1) and 18–23 (block 2) present at both birth and 9-year time points. Additionally, we observed high intra-CpG correlations comparing the birth to 9-year time points for all three north shore CpG sites. With respect to aim 2, we found that none of the perinatal variables examined, including gestational age, parity, maternal age and pre-pregnancy BMI and in addition, for 9 years, weight gain in the first 6 months, were significantly associated with PPARγ methylation at either birth or 9 years. Further, we observed that girls had significantly greater methylation at north shore sites 1–3 compared to boys at both time points. Adjusting for sex, we found that methylation at birth for sites 1 and 20 was significantly and inversely associated with birth weight. Similarly, we found that methylation at these sites at 9 years was also significantly and inversely associated with 9-year BMI z-score. Taken together, these results indicate that PPARγ methylation may be involved in regulating child body size and highlight the potential functional importance of north shore sites.
PPARγ CpG organization is typical of many other genes, with its promoter region containing a CpG island flanked by north and south shore sites [
52]. Additionally, in agreement with studies showing complex inter-CpG correlations over both short and long regions,
PPARγ contained two methylation blocks spanning 1 kb over the north shore (block 1) and 130 kb over the south shore, 5′ UTR, and gene body (block 2) [
45,
53]. Interestingly, we also found that north shore CpG site 1 from block 1 was correlated with methylation at sites 20–23 from block 2. There is a growing understanding that the location of a particular CpG site may be functionally important and several studies have highlighted the role of shore sites in gene expression, tissue differentiation, and overall phenotype [
9,
54,
55]. For example, Doi et al. (2009) showed that CpG shore methylation distinguished between several cell lines, including brain, liver, spleen cells, their pluripotent stem cells and parental fibroblasts [
54]. Similarly, Irizarry et al. (2009) showed that most methylation changes associated with colon cancer occurred in CpG shores [
9]. Our observations of methylation blocks surrounding the
PPARγ CpG island and high correlations between the north shore and gene body sites add evidence that shore sites may be of particular relevance in regulating biological pathways.
With respect to changes in CpG methylation over time, although some reports indicate stable methylation patterns [
53,
56] others do not [
57‐
59]. In their analyses of blood samples from the Netherlands Twin Register, Talens et al. (2010) showed that of 8 regions examined, 5 displayed stable methylation patterns for up to 20 years [
56]. Additionally, using Illumina 450 k data, we have previously shown that methylation across 16 paraoxonase 1 gene (
PON1) shore, shelf, and island sites was highly conserved comparing birth and 9-year time points [
53]. On the other hand, Fraga et al. (2004) showed that while 3-year-old monozygotic (MZ) twins showed relatively few epigenetic differences, there was considerably larger variability in older twin pairs [
57]. Our results indicate that
PPARγ methylation is stable over the birth to 9-year period and that even minute differences between CpG sites are conserved.
Although the pattern of CpG sites remained similar over time (Fig.
1), we found that north shore sites had slightly but significantly higher beta values (7.4%) at 9 years compared to birth. Previous literature has identified both hypo and hyper-methylation changes with age and taken together, these findings suggest that different genomic regions may have varying stability over time [
5,
28,
60,
61]. Additionally, we observed small differences by sex, with girls having slightly higher methylation compared to boys, at both birth and 9-year time points. However, these differences were limited to north shore sites 1–3. Although the significance of this remains unclear, our previous work using 450 k data identified that overall ~ 3% of CpG sites are differentially methylated by sex and are enriched for genes related to nervous system development and behavior [
53]. Interestingly, Hall et al. (2014) showed that genome-wide CpG methylation in pancreatic islets differentially clustered between males and females, suggesting that methylation may be involved in sex-specific metabolic differences [
62]. Our results are in line with this and overall show that
PPARγ CpG methylation is carefully maintained, emphasizing its potentially important role in regulating
PPARγ function.
In addition to data gaps on methylation structure and organization, very little is known about the epigenetic changes that accompany obesity development. Our report of an inverse relationship between
PPARγ methylation and body size is consistent with the idea that higher methylation downregulates
PPARγ, suppressing adipogenesis. To date, few studies have examined these relationships in
PPARγ, providing mixed results [
63,
64]. Yan et al. (2014) examined
PPARγ gene expression and methylation in offspring of dams exposed to polycyclic aromatic hydrocarbons (PAHs), reporting that increased PAH exposure was associated with increased weight, fat mass, higher gene expression of
PPARγ and lower
PPARγ CpG methylation [
64]. In contrast to this inverse relationship between
PPARγ methylation and weight, Drogan et al. (2015) analyzed subcutaneous adipose tissue (SAT) samples, showing that tissues from individuals with higher visceral fat mass had increased
PPARγ CpG methylation [
63]. Additionally, Nilsson et al. (2014) found differential
PPARγ methylation in adipose tissues from subjects with type 2 diabetes compared to controls but did not report on this relationship’s direction [
65]. We did not find that
PPARγ methylation at birth could predict 9-year BMI z-score and more work is needed to further elucidate its role in
PPARγ function and adipogenesis over time. Of note, site 1 was located in the north shore, further emphasizing the potentially critical role of north shore sites in regulating gene expression.
Lastly, there are several important points to consider with respect to our analyses. We measured methylation in blood samples, which can introduce bias if cell heterogeneity affects both methylation and obesity. However, our sensitivity analyses accounting for differences in cell composition did not substantially alter associations between PPARγ methylation and child size. Furthermore, our data displayed a consistent pattern of CpG methylation in blood samples over both birth and 9-year time points suggesting that heterogeneity of blood cell types may not significantly affect PPARγ methylation.
Nevertheless, whether
PPARγ methylation in blood is a suitable marker for its activity in adipocytes remains unknown. Several studies have indicated that molecular changes in blood do reflect pathological changes in the body and gene expression in blood is highly concordant (>80%) with expression in other tissues [
66,
67,
68‐
70]. With respect to body size, Ghosh et al. (2010) used principal components analysis to show that blood-based gene expression signals could distinguish between obese and lean subjects [
71]. Interestingly, Charriere et al. (2003) found that based on transcriptome profiling, pre-adipocytes were more closely related to macrophages than adipocytes [
72]. Further, a large genome-wide association study found that BMI was associated with methylation of
HIF3A in both blood and adipose tissue [
70]. Taken together, these data suggest that assessing
PPARγ function in blood may be biological relevant however more work is needed to determine this in the context of methylation.
Additionally, although we had previously shown that methylation at
PPARγ site 23 (gene body) was associated with
PPARγ gene expression [
27], this site was not significantly associated with child birth weight or BMI. Reasons for this inconsistency remain unclear and further research is warranted to examine relationships between CpG location and potential effects on gene expression. Overall, while our research argues that
PPARγ methylation has a relationship with child body weight and that north shore sites may be of particular functional importance, key questions remain on factors that influence site-specific methylation and whether it can be used to predict metabolic outcomes over time.