Introduction

Data from U.S. population surveys demonstrate a significant increase in obesity prevalence among children age 2–19 years old, from 5.5% in 1976–19801 to 16.9% in 2007–20101,2, with obesity defined as body mass index (BMI) ≥ 95th percentile using the Centers for Disease Control and Prevention (CDC)3. Severe obesity is the most rapidly growing paediatric obesity subgroup, and recent estimates suggest that this disease afflicts up to 6% of all children and adolescents in the United States4. Compared to youth with BMI in the obese range, those with severe obesity have higher rates of immediate and long-term metabolic and cardiovascular comorbidities5. It stands to reason that youth with obesity and severe obesity may also differ in aetiological factors and consequences, including epigenetic.

There is growing evidence that DNA methylation might contribute to obesity. Indeed, candidate gene methylation studies in animal models and humans have demonstrated methylation changes in promoters of various genes that are implicated in obesity, appetite control and/or metabolism, insulin signaling, immunity, growth and circadian clock regulation6,7,8,9. For example, the methylation percentage of insulin-like growth factor 2 (IGF2) promoter was higher in overweight infants than in lean infants8. The methylation of peroxisomal proliferator activated receptor-γ-co-activator-1α promoter in children blood predicts adiposity at adolescence independently of sex, age, pubertal timing and activity6.

To identify novel genes and pathways related to obesity and obesity-induced complications, epigenome-wide association studies (EWAS) are needed. Two previous studies using the HumanMethylation27 BeadChip with 27,000 CpGs, primarily targeting gene promoters and CpG islands (CGIs), examined blood leukocytes of obese and lean adolescents10,11. Wang et al. discovered two obesity-associated inflammatory genes Ubiquitin Associated And SH3 Domain Containing A and Tripartite Motif Containing 3 (UBASH3A and TRIM3), and Almen et al. identified 20 CpGs differentially methylated between groups. A larger study looking at the impact of BMI on DNA methylation in different tissues using the HumanMethylation 450 BeadChip by Dick et al. found five probes correlated with BMI, three of which were in the intron of HIF3A (Hypoxia Inducible Factor 3, Alpha Subunit)12. In children, two recent papers also showed that specific DNA methylation profiles in blood differ between lean and obese subjects13,14. Using the NimbleGen Human DNA Methylation 385 K Promoter Plus CpG Island Microarray, Ding et al. found 575 demethylated CGIs and 277 hypermethylated CGIs in obese children. Finally, Huang et al. identified 129 differentially methylated CpGs associated with 80 unique genes. None of the obesity-associated CpGs was common to all studies, which may be due to differences between the study populations, diverse genetic backgrounds, or heterogeneous metabolic phenotypes between different BMI categories.

Given the paucity of research on the different BMI categories, the purpose of this study were twofold: (i) investigate DNA methylation marks in all obese children compare to lean controls; (ii) identify the differentially methylated CpGs associated with severe obesity in childhood. We hypothesized that patients with severe obesity could help at identifying epigenetic changes, as extreme phenotypes improve genetic association studies.

Results

Identification of differentially methylated CpGs associated with childhood obesity

The genome-wide methylation analysis was conducted in 20 obese children (BMI Z-score > 2.5,) and 17 controls (Table 1). Three controls were excluded due to bad quality DNA and arrays.

Table 1 Characteristic of obese and control children.
Full size table

Because DNA methylation varies by cell type and could bias EWAS results conducted in blood samples, we estimated the cell type compositions in each sample using minfi, and found that cellular composition was not similar between cases and control subjects (Supplemental Table 1). To correct for these differences, we used next the Houseman’s correction algorithm15.

After correction for cellular heterogeneity and multiple testing, thirty-one CpGs were differentially methylated (FDR ≤ 0.05) between obese children and control subjects, 10 were hypermethylated, and 21 were hypomethylated in obese compared to lean children (Table 2). The largest difference was observed for cg26834418 located in the promoter region (TSS1500) of the CHORDC1 gene (+13% in control subjects compare to obese children). This gene also showed two other probes differentially methylated in obese compare to lean children. Because of the relatively small numbers in the study and small differences in methylation levels for most of the significant CpG sites, we analysed the confounding variables age and sex. We found that the methylation of 5/31 and 1/31 of the identified CpG sites was correlated with respectively age or sex (Figs 1 and 2).

Table 2 Differentially methylated CpGs in obese children.
Full size table
Figure 1: Correlation between DNA methylation and age in 20 obese children and 17 controls at the 31 associated CpGs.
figure 1

*Significant correlation (p < 0.05), DNA methylation value in the y-axis and age in years in the x-axis.

Full size image
Figure 2: Association between DNA methylation and sex in 20 obese children and 17 controls at the 31 associated CpGs.
figure 2

*Significant correlation (p < 0.05), DNA methylation value in the y-axis and sex (M = male, F = female) in the x-axis.

Full size image

The distribution of the 31 CpGs showed that DNA methylation variation was distributed over the CpG island shores, and that although CGI are enriched on the array (31% of Illumina probes are in CGI); only 16% of our obesity–associated CpGs were located in CGI, compared to 38% in shores (22% in S-Shore and 16% in N-shore) (p = 0.016, Pearson’s Chi-squared test) (Fig. 3A). The genomic distribution of the 31 CpGs in comparison to all the probes located on the 450 K BeadChip array with respect to gene structure is shown in Fig. 3B. We also found an enrichment of differentially methylated CpGs outside promoter and gene body (p = 2.10−4, Pearson’s Chi-squared test) (Fig. 3B).

Figure 3
figure 3

Distribution of differentially methylated CpGs versus all CpGs sites on the Infinium HumanMethylation450 BeadChip in relation to (A) CpG island regions; (B) the nearest gene regions. Chi-square analysis was performed to test over- or under-representation of sequence features among the differentially methylated CpGs, *P.value < 0.05. (C) Venn diagram of the identified CpGs in obese or severely obese children, below the 13 common CpGs.

Full size image

Since obesity is an extremely heterogeneous disease, we then chose to focus only on the extremely obese children (BMI z-score ≥ 3.5).

Identification of differentially methylated CpGs associated with severe childhood obesity

We analysed next DNA methylation marks only in the 11 severely obese children of the group. We found 151 differentially methylated CpGs (q.value ≤ 0.05), 69 were hypermethylated, and 82 were hypomethylated in severely obese patients compared to lean controls (Supplemental Table 2). Of these 151 differentially methylated CpGs, ten had a greater than 10% difference in methylation between the case and control groups. The most significant difference was observed for a cg27590049 located in the LMX1A (LIM Homeobox Transcription Factor 1, Alpha)16; and the largest difference was observed for a cg07944420 located on the gene body of ACSF3 (Acyl-CoA Synthetase Family Member 3) that showed a decreased by 17% of methylation level in control subjects compared to obese children. Thirteen of the 151 CpGs were common to our first analysis comparing obese children all together, 18 seems specific to moderately obese children and 138 to severely obese children (Fig. 3C).

Next, we performed a gene set enrichment analysis (GSEA) to explore the potential of shared biologically relevant pathways among the obesity-associated methylation events. Seven pathways showed a significant enrichment including “IRS1 Target Genes” and different cancer traits (Table 3).

Table 3 Gene set enrichment analysis of severe obese children associated CpGs.
Full size table

The genomic distribution of these 151 differentially methylated CpGs in relation to CpG density (CGIs, shores, shelves, and open sea) was not clearly different from the whole array CpG distribution and there was no significant enrichment within specific gene regions (data not shown).

Discussion

In this study we aimed to identify obesity related methylation marks in peripheral blood leukocytes using a genome wide approach in youth obese children. The primary finding of this study was that most of the epigenetic marks are different in moderate and severe obesity. We identified respectively 18 and 138 differentially methylated CpGs between moderate or severe obese children and lean controls. As observed for genetic association studies, sampling individuals with extreme phenotypes can enrich the presence of epigenetic variations and can therefore lead to an increase in detection of these differences. Moreover, most of the differentially methylated CpGs was found within open seas or intergenic regions, consistently with previous findings showing that DNA methylation may be more dynamic outside CGIs.

Compare to previous studies10,11,12,13,17,18, we replicated the association between DNA methylation level and obesity at 10 gene loci (ANKRD11: ankyrin repeat domain 11, AVPI1: arginine vasopressin-induced 1, CDK19: cyclin-dependent kinase 19, FOXK2: forkhead box K2, HDAC4: histone deacetylase 4, IFT140: intraflagellar transport 140, KIAA0753, LTBP1: latent transforming growth factor beta binding protein 1, MYOM2: myomesin 2, TBC1D8: TBC1 domain family, member 8). Among these genes, HDAC4 is an interesting candidate since recently, Abu-Farha et al. showed a reduced expression of HDAC mRNA and protein in human obese subjects both in blood cells and adipose tissue19. This is consistent with clinical data in humans that associated the haploinsufficiency of HDAC4 with obesity20. In our study, we found a HDAC methylation level higher in obese children than in controls.

Numerous genes found in our severe obese analysis were also associated with cancer. Many epidemiological and clinical studies have demonstrated that early obesity is an established risk factor for many cancers in later life21. Cross talk between macrophages, adipocytes, and epithelial cells occurs via obesity-associated hormones, growth factor signalling, inflammation, vascular integrity processes, microenvironmental perturbations, and other mediators, which could enhance the cancer risk and/or progression22. Thus, the methylation changes in childhood obesity could increase the risks for later cancer susceptibility.

Most of the identified genes are not expressed or do not have a relevant function in blood cells; whether these epigenetic marks in blood may reflect or correlate with methylation in more relevant tissues is not known. However, several studies showed that DNA methylation measured in whole blood is a marker for less accessible tissues that are directly involved in disease. For instance, Murphy et al. have shown for example that methylation across the H19 DMRs (Differentially Methylated Regions) was similar across several tissues from divers embryonic origins23. Likewise, in non-imprinted loci, Talens et al. also found that DNA methylation levels did not differ in blood and buccal cells, from mesodermal and ectodermal embryonic tissues, respectively24. The recent work of Huang et al.25 identified 1,285 discordant and 1,961 concordant genes for methylation between blood and adipose tissue; the discordant genes are enriched in biological functions related to immune response, leukocyte activation or differentiation, and blood coagulation. Moreover, epigenetic marks associated with type-2 diabetes26 or adiposity7 have also been identified in peripheral tissues.

There were several limitations to this study. The first limitation was that we used DNA from whole blood. To correct our methylation data for this weakness, we used a Houseman correction algorithm27. It must also be noted that adjusting for cell composition makes impossible the process of replication and validation of the identified CpGs by pyrosequencing. Replication could also be accomplished if there exists an independent replication population; in this case the models could be re-applied. We tried to replicate our findings by running the available datasets for common obesity (GEO DataSets: GSE25301, GSE43975, GSE44763, GSE73103) under the RefFreeEwas procedure, but we failed to find any associated CpGs after correction for cell heterogeneity, even those previously identified by the authors without cell heterogeneity correction.

The second limitation is that we cannot conclude about the existence of these epigenetic marks before the establishment of the obesity. While this lack of interpretability is inherent to the design of the cross-sectional case control study, the finding of methylation marks associated with the early stages of severe obesity in young patients may be of pathogenic relevance to certain features or complications of this disease. Indeed, the marks that have been found here could be used in a longitudinal study of the young patients in order to gain both biomarker and mechanistic insights. The longitudinal sampling of cells from adolescence to adulthood should further allow which of these epigenetic changes follow the development of the long term overt phenotype of severe obesity and its complications.

Our major strength was that all studied participants were children, aged from 3 to 13 years old, less subject to cofounding factors like medication or comorbidities, very common in adult obese patients.

In conclusion, the identification of methylation changes in specific genes will provide important targets for further study into the underlying mechanisms and the therapeutic potential for childhood obesity.

Methods

Study participants

Twenty obese children (5 to 13 years old) and equal numbers of control children (3 to 13 years old) were included in the study. The BMI for obese children (10 male) and lean children (11 male) was 26.0 ± 4.8 kg/m2 and 16.7 ± 2.2 kg/m2, respectively (Table 1). Age and gender-specific BMI Z-scores were determined by using the growth charts form the World Health Organization with a mean BMI Z-score of 3.6 ± 0.8 for obese children and 0.2 ± 1.1 for controls. Patients with monogenic or syndromic forms of obesity were excluded. To limit the risk of population stratification, all recruited children are of Caucasian ancestry assessed by family history and grandparents’ birthplace. All methods were carried out in accordance with relevant guidelines and regulations. Patients and controls were included in the study according to the French bioethics law with families being carefully informed and having signed a detailed informed consent. All protocols were agreed by French ethic boards (CODECOH DC-2013-1977, CPP C0-13-004, CCTIRS n°14-116bis, CNIL n°91 4228).

Infinium humanMethylation450 beadchip array

DNA was extracted from whole blood cells of 20 case and 20 control subjects using Gentra DNA extraction kit (Qiagen). Genomic DNA (1 ug) from each of the 40 subjects was bisulphite-converted using Zymo EZ DNA Methylation-Gold kit (ZymoResearch) and the DNA was analysed using the Infinium HumanMethylation450 platform (Illumina, Inc.) by The Genotyping National Center (CNG, CEA, Evry, France).

Infinium HumanMethylation450 BeadChip array data processing

DNA methylation status of case and control subjects was established using Illumina Infinium HumanMethylation450 BeadChips that cover 485,764 cytosine positions of the human genome. Preprocessing and normalization involved steps of probe filtering, color bias correction, background subtraction and subset quantile normalization as previously described28. After these intra-sample normalization procedures, β-values were calculated. To avoid batch effect, all samples were processed together. Obese subjects were compared to control subjects, using t-tests.

Assessment of cell composition differences

The R package minfi29 was used to estimate the fraction of CD8T-, CD4T-, NK- and B-cells, monocytes, and granulocytes in our 40 samples. This package allows for estimating cell fractions in Illumina 450 K methylation data from whole blood, based on the methylation data published for flow-sorted cells30, and algorithms27.

Cell heterogeneity correction for the methylation data analysis

To correct our methylation data analysis for cell heterogeneity between samples, we used R package RefFreeEwas15. This package allows for conducting EWAS while deconvoluting DNA methylation arising as mixtures of cell types. This method is similar to surrogate variable analysis, except that it makes additional use of a biological mixture assumption.

Additional Information

How to cite this article: Fradin, D. et al. Genome-Wide Methylation Analysis Identifies Specific Epigenetic Marks In Severely Obese Children. Sci. Rep. 7, 46311; doi: 10.1038/srep46311 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.