Abstract

   Obesity has become a major public health issue which is caused by a
   combination of genetic and environmental factors. Genome-wide DNA
   methylation studies have identified that DNA methylation at
   Cytosine-phosphate-Guanine (CpG) sites are associated with obesity.
   However, subsequent functional validation of the results from these
   studies has been challenging given the high number of reported
   associations. In this study, we applied an integrative analysis
   approach, aiming to prioritize the drug development candidate genes
   from many associated CpGs. Association data was collected from previous
   genome-wide DNA methylation studies and combined using a
   sample-size-weighted strategy. Gene expression data in adipose tissues
   and enriched pathways of the affiliated genes were overlapped, to
   shortlist the associated CpGs. The CpGs with the most overlapping
   evidence were indicated as the most appropriate CpGs for future
   studies. Our results revealed that 119 CpGs were associated with
   obesity (p ≤ 1.03 × 10^−7). Of the affiliated genes, SOCS3 was the only
   gene involved in all enriched pathways and was differentially expressed
   in both visceral adipose tissue (VAT) and subcutaneous adipose tissue
   (SAT). In conclusion, our integrative analysis is an effective approach
   in highlighting the DNA methylation with the highest drug development
   relevance. SOCS3 may serve as a target for drug development of obesity
   and its complications.

   Keywords: DNA methylation, obesity, association, gene expression, CpG

Introduction

   Since 1980, the incidence of obesity has increased throughout the world
   (Stevens et al., [38]2012; Ng et al., [39]2014). The onset of obesity
   involves the interaction between genetic and environmental factors
   (Contaldo and Pasanisi, [40]2004; Ussar et al., [41]2015). Genome Wide
   Association Studies (GWASs) have successfully identified many genetic
   variations associated with human complex diseases and provide crucial
   new insights about underlying molecular mechanisms (De La Vega et al.,
   [42]2011; Fall and Ingelsson, [43]2014; Winham et al., [44]2014;
   Evangelou et al., [45]2018). Until now, the largest obesity GWAS study
   has identified 97 body mass index (BMI) associated loci (P ≤ 5 × 10^−8)
   from up to 339,224 individuals. However, most of the genetic
   susceptibility remains unclear (Locke et al., [46]2015).

   Existing evidence suggests that obesity is a result of interactions
   between genetic and environmental factors (Marti et al., [47]2008). DNA
   methylation provides a molecular mechanism for the interaction between
   the environment and obesity, in that it may affect individual
   susceptibility to obesity by altering the gene expression. In recent
   years, the association between DNA methylation and obesity has
   intensively been studied (van Dijk et al., [48]2015; Dhana et al.,
   [49]2018; Wang et al., [50]2018). For example, a genome-wide DNA
   methylation association study in obesity that recruited 5,387
   individuals, identified 278 CpGs associated with BMI (Wahl et al.,
   [51]2017). The associated CpGs have provided wider insight in addition
   to previous genetic studies. On the other side, the numerous associated
   CpGs has made it difficult for functional investigations using cell and
   animal models.

   In this study, we applied an integrative analysis approach, to
   prioritize genes with more relevance from several associated CpGs.
   Using this approach, we identified SOCS3 as a promising candidate for
   mechanism studies and drug development. This approach can also be
   adapted to genome-wide DNA methylation studies of other diseases.

Methods

   The integrative analysis approach included three components. The first
   component was to nominate the candidate CpGs by combining the
   association results from previous studies of peripheral blood samples
   (Steps 1–4, Figure [52]1). The second component was to estimate the
   functional relevance of the candidates through pathway enrichment
   analysis (Step 5). The third component was to validate that the genes
   affiliated with candidate CpGs were differentially expressed in adipose
   tissues (Step 6). Finally, the evidence from these components were put
   together and the genes with positive support from all components were
   considered and prioritized by our approach (Step 7).

Figure 1.

   [53]Figure 1
   [54]Open in a new tab

   The flow chart of the intergrative analysis. The circled numbers
   represent the steps in the pipeline.

Literature Search

   The literature search was conducted in the PubMed database using the
   keywords “CpG”, “DNA methylation” and “obesity” to capture all articles
   published from 2014 to 2018. We applied an English language restriction
   to our search results.

Inclusion Criteria and Data Extraction

   Both cohort studies and case-control studies reporting the association
   between DNA methylation and obesity (as measured by BMI) were included
   in this meta-analysis. Studies that used samples from cancer patients
   were not included. We further excluded the studies that used non-human
   subjects.

   The full text of each article was carefully read to determine whether
   studies should be included. Once included, data were extracted from the
   articles, including the publication year, participant characteristics,
   sample size, association p-value, and the effect size.

Meta-Analysis

   We employed a sample-size weighted strategy to combine the p values
   reported in the included studies, taking into consideration the
   direction of the association effect size. This strategy was implemented
   using R software ([55]https://www.r-project.org/). In this
   meta-analysis the CpG site with p value less than 1.03 × 10^−7
   (Bonferroni correction based on 485,577 CpGs designed in Illumina
   HM450K array) and with effect sizes consistent with the direction
   across all included studies, were considered as significant.

Pathway Enrichment Analysis

   We investigated the enrichment of the affiliated genes in the Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathways, using the Metascape
   online software ([56]http://metascape.org; Tripathi et al., [57]2015).
   The genes were annotated using the default resources provided by
   Metascape. KEGG pathways were reduced using the default settings (the
   number of gene hits ≥3, enrichment p-value ≤ 0.05 and enrichment
   statistics ≥1.5). A FDR p-value ≤ 0.05 was taken to declare a
   significant enrichment.

Differential Expression Analysis in Adipose Tissues

   We aimed to investigate whether the associated genes were
   differentially expressed in the SAT and VAT of obesity patients, by
   comparing their gene transcription levels with normal individuals. This
   analysis was performed using the GEO2R tool
   ([58]https://www.ncbi.nlm.nih.gov/geo/geo2r/) on two datasets,
   [59]GSE2508 (10 obese vs. 10 lean) and [60]GSE88837 (15 obese vs. 15
   lean) for the SAT and VAT, respectively. The gene transcription levels
   were assayed using Affymetrix Human Genome U95 V2 and U133 arrays. The
   differential gene expression in obese samples was identified using the
   Bayesian estimation by GEO2R. Transcription level data of each sample
   was queried from the GEO database (Davis and Meltzer, [61]2007).
   Empirical Bayes statistics were calculated using the R package “limma”
   (Smyth, [62]2004; Ritchie et al., [63]2015). The fold change of DNA
   methylation was calculated using the group mean. P value ≤ 0.05 and
   |log[2] (fold change)| ≥1 were used as criteria for differentially
   expressed genes. CpGs which were differentially expressed in both
   tissue types were identified as relevant loci.

The DNA Methylation Associated With Obesity in Human VAT and Liver Tissue

   The DNA methylation of the included studies was all measured in
   peripheral blood, but the DNA methylation in peripheral blood may be
   different from that in the metabolic tissues. To test whether the
   association in peripheral blood samples can be transferred into obesity
   related tissue, we tested the association of the significant CpGs in
   human VAT and liver tissue, using two GEO datasets, [64]GSE88940 (10
   obese vs. 10 normal VAT samples) and [65]GSE65057 (8 obese vs. 7 normal
   liver samples), respectively.

Results

Characteristic of Individual Studies

   According to the keywords “CpG”, “DNA methylation” and “obesity”, a
   total of 350 related articles were retrieved. Two hundred and seventy
   studies were excluded based on the title and abstract, as they were
   inconsistent with inclusion criteria, leaving 80 articles. Of those, 67
   articles were excluded after a full-text review. As a result, 13
   articles were included in the analysis. The reason for the exclusion of
   most articles was because they were functional studies in cells or
   animals. The basic characteristics of the included studies are detailed
   in Table [66]1.

Table 1.

   Characteristics of the included genome-wide DNA methylation studies.
   References Corhort source Ethnicity Subjects Body mass