Altered DNA Methylation and Differential Expression of Genes Influencing Metabolism and Inflammation in Adipose Tissue From Subjects With Type 2 Diabetes Free

The susceptibility for type 2 diabetes (T2D) increases with age, physical inactivity, and obesity in subjects with a genetic predisposition (1). However, the exact molecular mechanisms causing the disease still remain unknown.

Adipose tissue has a central role in whole-body energy metabolism as a dynamic store of triglycerides, and as an endocrine organ that coordinates energy intake and use by other tissues (2). In individuals with T2D, this function is frequently perturbed by an impaired response of the adipocytes to insulin resulting in elevated lipid levels in circulation and storage in alternative tissues such as liver, muscle, and pancreas (3).

One link between environment and disease is epigenetics influencing gene transcription and subsequently organ function (4). We have previously shown that epigenetic modifications may accumulate during aging (5,6), and that DNA methylation in humans is influenced by diet, birth weight, and exercise (7–10), suggesting that epigenetics could be involved in age-related and life style–related diseases such as T2D. Indeed, studies from our group and others (11–17) have identified epigenetic modifications in patients with T2D; e.g., we found subtle nongenetic methylation differences in adipose tissue from five monozygotic (MZ) twin pairs discordant for T2D using the Infinium 27k array (13). Nevertheless, our knowledge about the role of epigenetics in the growing incidence of T2D remains limited, and the genome-wide expression profile has, to our knowledge, not been linked to the epigenome in adipose tissue of diabetic patients. Therefore, our aim was to investigate genome-wide differences in expression and DNA methylation in adipose tissue from MZ twin pairs discordant for T2D. This study design allows the elimination of confounding factors such as genotype, age, and sex. The causes of incomplete concordance between MZ twins are traditionally explained by nonshared environmental factors. To investigate the importance of our findings in the general population, the expression of selected genes was validated in case-control cohorts. Finally, heritability of DNA methylation was estimated in adipose tissue from nondiabetic twins, and the genome-wide methylation pattern was analyzed in adipose tissue from a case-control cohort of unrelated subjects.

Fourteen MZ twin pairs discordant for T2D were recruited through Scandinavian twin registries (18) (Table 1). Five of the twin pairs have previously been studied using the Illumina (San Diego, CA) 27k array (13). Zygosity was confirmed and copy number variants (CNVs) predicted by analysis of 730,525 genetic markers using HumanOmniExpress arrays (Illumina).

One hundred twenty unrelated subjects with normal glucose tolerance (NGT) or T2D were included for biological validation (Table 1). They have previously been described (19).

The characteristics of 28 NGT and 28 T2D unrelated subjects, pairwise matched for age and sex, and selected from a larger twin cohort (20,21) are shown in Table 1.

Ten MZ and 10 same-sex dizygotic (DZ) twin pairs, all with NGT, were selected from a larger cohort (20,21) and used for heritability estimates (Table 1). Zygosity was confirmed by genetic markers.

All study participants provided written informed consent, and the studies were approved by local ethics committees.

After an overnight fast, participants visited clinics for metabolic characterization, and subcutaneous adipose tissue biopsy samples were obtained under local anesthesia, frozen immediately in liquid nitrogen, and stored at −80°C. Glucose tolerance was measured by a 75-g oral glucose tolerance test.

RNA was extracted using the miRNeasy kit followed by the RNeasy MinElute Cleanup kit (Qiagen, Hilden, Germany) for twins or by TRIzol (Invitrogen, Carlsbad, CA) for nontwins. DNA was extracted using the DNeasy Blood and Tissue kit (Qiagen). DNA/RNA quantity and purity were determined by spectrophotometry (NanoDrop Technologies, Wilmington, DE), and RNA integrity was determined using the Experion system (Bio-Rad, Hercules, CA).

mRNA expression was analyzed in adipose tissue from discordant twins using GeneChip Human Gene 1.0 ST arrays (Affymetrix, Santa Clara, CA) according to the manufacturer’s recommendations. We computed Robust Multichip Average expression measures using the Oligo package in Bioconductor (22).

We applied gene set enrichment analysis (GSEA) (23) to expression array data using KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways. All probes corresponding to transcripts were used and ranked according to the t statistics in a paired t test. GSEA was run with the highest occurrence for genes with multiple probes and considered pathways with 1–500 transcripts.

DNA methylation was analyzed in adipose tissue of discordant twins, case-control cohort 2, and the twin cohort for heritability estimates using Infinium HumanMethylation450 BeadChips (Illumina). This array contains 485,577 probes, which cover 21,231 (99%) RefSeq genes (24). DNA was treated with bisulfite using the EZ DNA Methylation kit (Zymo Research, Orange, CA). Analysis was performed following the Infinium HD Assay Methylation protocol (Illumina). To avoid a batch effect, co-twins and matched pairs were hybridized on the same chip. The bioinformatics analyses were performed as previously described (9). Probes with a mean detection P value of >0.01, non-cytosine guanine dinucleotide (CpG) probes, Y-chromosome probes, and 65 single nucleotide polymorphism (SNP) probes were removed. β-Values, ranging from 0 to 100% methylation, were used to report the outcome.

cDNA was produced using the QuantiTect Reverse Transcription Kit (Qiagen). mRNA levels were detected with the ViiA 7 Real-Time PCR System (Applied Biosystems, Foster City, CA) using probe/primers for ELOVL6 (Hs00907564_m1), GYS2 (Hs00608677_m1), FADS1 (Hs00203685_m1), SPP1 (Hs00959010_m1), CCL18 (Hs00268113_m1), IL1RN (Hs00893626_m1), S100A4 (Hs00243202_m1), and SLC37A2 (Hs00400129_m1). Samples were run in duplicate and normalized to HPRT expression (4326321E; Applied Biosystems). NormFinder (25) software demonstrated stable HPRT expression.

Assays (Hs02596879_g1, Hs02596860_s1 and Hs02596867_s1; Applied Biosystems) targeting ND6, RNR2 and CYTB were used to analyze mitochondrial DNA (mtDNA) content with the ViiA 7 system (Applied Biosystems). Samples were analyzed in duplicate and normalized to the quantity of the nuclear gene RNAseP (4316831; Applied Biosystems).

Comparisons between discordant twins and individuals in case-control cohort 2 are based on paired Wilcoxon statistics. To account for multiple testing, we applied false discovery rate (FDR) analyses. Regression analysis was used to relate expression and methylation in discordant twins taking twin pair relationship into account. Mann-Whitney tests were used to analyze data in case-control cohort 1. CNV calling from whole-genome genotyping data were performed with the PennCNV software tool (26). PennCNV implements a hidden Markov model that integrates signal intensity data and SNP allelic ratio distribution from the Illumina HumanOmniExpress chip. Heritability and common environmental influences (c²) were estimated using DF analyses with double-entry twin data (27).

The characteristics of subjects included in the study cohorts are shown in Table 1. Fasting glucose (f-glucose) levels, glucose tolerance, and HbA_1c levels differed significantly between T2D and nondiabetic subjects. In discordant twins, BMI and fat percentage were significantly increased in subjects with diabetes, but the physical fitness level did not differ between co-twins. In twins used for heritability estimates, there were no differences in age, BMI, glucose levels, or HbA_1c levels between MZ and same-sex DZ twins.

Expression data were obtained from the adipose tissue of 12 twin pairs discordant for T2D. We first tested whether sets of related genes were altered in diabetic versus nondiabetic twins. The GSEA yielded 14 significant gene sets with downregulated expression and 17 gene sets with upregulated expression in adipose tissue from diabetic versus nondiabetic co-twins (q < 0.05; Table 2). Downregulated gene sets include pathways involved in oxidative phosphorylation and amino acid, carbohydrate, and lipid metabolism; and upregulated gene sets include pathways involved in inflammation and glycan metabolism. Genes contributing to the enrichment for significant gene sets are presented in Supplementary Tables 1 and 2. Moreover, selected gene sets representing metabolism and immune system pathways are shown in Fig. 1A.

We next tested whether the expression of individual genes was altered in adipose tissue from diabetic versus nondiabetic co-twins. After FDR correction, 197 individual genes were found to be differentially expressed in diabetic versus nondiabetic co-twins (q < 0.15; Supplementary Table 3). One hundred sixteen of these genes were upregulated, and 81 genes were downregulated in diabetic twins. The expression differences of these genes range from 5% to 43% (Supplementary Table 3). A post hoc power calculation to detect expression differences of ≥5% is presented in Supplementary Table 4. Furthermore, genes with the largest expression differences between diabetic and nondiabetic co-twins (P < 0.05) are shown in Table 3. Six genes were technically validated by quantitative real-time PCR (qPCR) in the adipose tissue of twins with array data. We observed significant correlations between the two methods for each gene (ρ = 0.76–0.97, P < 0.0001; Supplementary Fig. 1), and there were significant differences between diabetic and nondiabetic co-twins with fold differences of the same magnitude for qPCR compared with array data (Fig. 1B and C).

We proceeded to investigate whether genes previously linked to obesity and T2D in published genome-wide association studies (GWASs) (P < 1.0 × 10⁻⁵ [www.genome.gov/gwastudies, accessed 22 August 2012]) were differently expressed in adipose tissue from twins discordant for T2D. These genes are associated with these traits through SNPs, which may be located in intergenic regions, and they do not need to have functional evidence as causal genes for obesity or T2D. We found 27 of 144 obesity genes with differential expression in adipose tissue from discordant twins (e.g., IRS1 and VEGFA, P < 0.05; Supplementary Table 5). This was more than expected (P = 0.02, χ² test). Among the 87 T2D candidate genes, 8 were differentially expressed in discordant twins (e.g., PPARG and GLIS3, P < 0.05; Supplementary Table 6). This was not more than expected (P = 0.4, χ² test). As some candidate genes are associated with both obesity and T2D, the differentially expressed candidate genes correspond to 33 unique genes.

The binding of transcription factors to promoter regions is an important mechanism by which expression is regulated. We searched for over-representation of DNA binding motifs in promoter regions of the 197 differentially expressed genes presented in Supplementary Table 3 using Pscan (28) and the JASPAR database (29). We found a significant enrichment for sites binding to KLF4, SP1, PAX5, EGR1, and INSM1 (Supplementary Table 7). Among these, SP1 showed a 6.1% decreased expression in the adipose tissue of diabetic twins (P = 0.009).

To biologically validate our findings in discordant twins, we analyzed the expression of six selected genes in adipose tissue from unrelated subjects with NGT or T2D (case-control cohort 1). The selected genes have known functions related to fat metabolism (ELOVL6 and FADS1) (30,31), glucose metabolism (GYS2) (32), or inflammation (SPP1 [OPN], CCL18, and IL1RN) (33–35); and were selected from the genes contributing to the enrichment scores of GSEA and/or from the list of most downregulated and upregulated genes in discordant twins. All six genes could be validated, and showed significant expression differences between unrelated patients with diabetes and nondiabetic subjects in the same direction as the discordant twins (Fig. 1B and D). This was also true when excluding obese subjects (BMI >30 kg/m²) and only comparing nonobese NGT with nonobese T2D subjects (Supplementary Table 8). Since the unrelated NGT subjects have a broad BMI range, we further examined the impact of obesity on the expression of these six genes by comparing nonobese and obese NGT subjects. Their characteristics are presented in Supplementary Table 9. ELOVL6 and GYS2 had significantly lower expression, and SPP1 and IL1RN had significantly higher expression in obese versus nonobese NGT subjects (Fig. 1E). Additionally, the expression of all genes correlated significantly with BMI and/or glucose levels in NGT subjects (Supplementary Table 10).

Since the expression of genes representing oxidative phosphorylation is decreased in adipose tissue of T2D twins (Table 2 and Fig. 1A), we quantified mtDNA content to follow up these results. The mtDNA content was reduced in adipose tissue from diabetic versus nondiabetic co-twins (Fig. 1F), which is in line with our expression results.

CNVs may contribute to phenotypic variation in twins (36). We therefore analyzed CNVs in MZ twins discordant for T2D and filtered CNVs present in two or more individuals of the same status. We identified six different CNVs, including three that were present only in diabetic twins and three present only in nondiabetic twins (Table 4).

We next studied the global DNA methylation pattern with the Infinium HumanMethylation450 BeadChip in adipose tissue from 14 twin pairs discordant for T2D. After quality control (QC) and filtering, methylation data were obtained for 480,403 CpG sites. To evaluate the global human methylome in adipose tissue, we calculated the average level of methylation for all sites divided into groups based on either their location in relation to the nearest gene (Fig. 2A) or the location in relation to CpG islands (Fig. 2B) (24). There were no significant differences in average DNA methylation for these regions between diabetic and nondiabetic co-twins. While the average methylation level was high within the gene body, 3′ untranslated region (UTR), and intergenic regions, it was low in TSS1500, TSS200, 5′ UTR, and the first exon (Fig. 2A). Moreover, the methylation level was low within CpG islands and intermediate within shores, whereas shelves and open sea showed the highest methylation levels (Fig. 2B).

To visualize the overall relationship between DNA methylation profiles of individuals, both within and between twin pairs, hierarchical cluster analysis was applied to the methylation data using all CpG sites that passed QC and filtering. Twelve of 14 twin pairs clustered together, suggesting a strong role for underlying genetic or common environmental factors in determining the epigenetic profile in adipose tissue of discordant twin pairs (Fig. 2C). After removing CpG sites directly affected by SNPs (37), 11 of 14 twin pairs still clustered together.

We further tested whether any individual sites exhibit differential methylation in the discordant twins. We found 23,470 (∼5%) of 480,403 analyzed CpG sites with differential methylation in adipose tissue from diabetic versus nondiabetic twins at P < 0.05 (Supplementary Table 11). Of those, 11,494 sites had increased methylation, and 11,976 sites had decreased methylation in diabetic twins. Moreover, 3,670 sites had an absolute difference in methylation of ≥3%. However, none of these differences remained significant after FDR correction, and the number of differences (23,470) is not more than that expected by chance. A post hoc power calculation to detect absolute methylation differences of ≥3% is presented in Supplementary Table 4.

These data suggest either that there are limited differences in methylation between patients with diabetes and nondiabetic subjects or that DNA methylation in human adipose tissue is highly heritable. We therefore proceeded to test whether we could identify methylation differences in adipose tissue from unrelated subjects with T2D compared with nondiabetic subjects, matched for age and sex (case-control cohort 2; Table 1). After QC and filtering, the Infinium array generated methylation data for 481,086 sites. We found 71,074 sites (∼15%) with differential DNA methylation at P < 0.05 in adipose tissue from case-control cohort 2. This is almost three times more than that expected by chance (P < 0.0001, χ² test). After FDR analysis, 15,627 sites representing 7,046 unique genes were found to be differentially methylated in patients with diabetes versus nondiabetic subjects (q < 0.15; Supplementary Table 12). The absolute differences in DNA methylation between patients with diabetes and nondiabetic subjects are illustrated in Fig. 2D and E. While 6,754 sites showed increased methylation, 8,873 sites showed decreased methylation in patients with diabetes (Fig. 2D and E). Moreover, CpG sites with differential methylation were over-represented among sites with 20–70% methylation (Fig. 2F and G), suggesting that sites with an intermediate degree of methylation are more likely to change. We analyzed the distribution of differentially methylated sites in patients with diabetes versus nondiabetic subjects at q < 0.15 (Supplementary Table 12), based either on their relation to gene regions (Fig. 2H) or their relation to CpG islands (Fig. 2I). Differentially methylated sites were over-represented in the gene body and open sea, and under-represented in TSS1500, TSS200, and CpG islands compared with the probe distribution on the array (Fig. 2H and I). Moreover, based on the Illumina annotations, differentially methylated sites were over-represented in enhancer regions (31% of significant sites compared with 21% of analyzed sites, P < 0.0001, χ² test).

To gain biological relevance of methylation differences in adipose tissue from unrelated patients with diabetes versus nondiabetic subjects, we performed a KEGG pathway analysis of the 7,046 genes that exhibit differential methylation at q < 0.15 (Supplementary Table 12) using WebGestalt. Pathways involved in inflammation, glycan metabolism, cancer, Wnt signaling, and mitogen-activated protein kinase signaling were among those significantly enriched (Fig. 2J and Supplementary Table 13).

We also examined whether candidate genes for T2D and/or obesity, previously identified by GWAS (P < 1.0 × 10⁻⁵ [www.genome.gov/gwastudies]), exhibit differential methylation in adipose tissue of unrelated patients with diabetes versus nondiabetic subjects. We found differential methylation in 123 sites (q < 0.15) representing 50 T2D candidate genes including IRS1, PPARG, KCNQ1, and TCF7L2 (Supplementary Table 14 and Fig. 2K). This is more than expected (P = 0.002, χ² test). Moreover, 127 sites representing 65 candidate genes for obesity were differentially methylated in case-control cohort 2 (q < 0.15, P = 0.5, χ² test; Supplementary Table 15).

We then tested whether known risk factors for T2D, including obesity and hyperglycemia, affect DNA methylation of the 15,627 CpG sites differentially methylated in case-control cohort 2, already in nondiabetic subjects. BMI and f-glucose were associated with differential DNA methylation of 1,270 and 512 CpG sites, respectively (Supplementary Tables 16 and 17). Importantly, ∼91% of the CpG sites that exhibit differential DNA methylation due to increased BMI or glucose levels in nondiabetic subjects changed in the same direction as methylation in patients with diabetes.

We proceeded to test whether any of the sites that exhibit differential methylation in case-control cohort 2 (q < 0.15; Supplementary Table 12) also exhibit differential methylation in adipose tissue of the discordant twins at P < 0.05 (Supplementary Table 11). Of 15,627 differentially methylated CpG sites in case-control cohort 2, 1,410 sites were also differentially methylated in the same direction in the discordant twins (P < 0.05; Supplementary Table 12). This overlap is almost twice as much as expected (P < 0.0001, χ² test).

Based on the potential problem with cross-reactive Infinium probes (37), we tested whether any of the probes that detected differential methylation in our cohorts cross-react to alternative genomic locations (Supplementary Tables 11 and 12). Importantly, only 25 and 10 possibly cross-reactive probes, respectively, have a perfect match to another location in the genome in the discordant twins and case-control cohort 2. Additionally, we found fewer SNP probes than expected (Supplementary Table 12; P < 0.0001, χ² test).

We next analyzed the expression of two genes, S100A4 (encoding S100 calcium-binding protein A4) and SLC37A2 (encoding glucose-6-phosphate transporter), in adipose tissue from the unrelated subjects in case-control cohort 2. These genes were selected based on the following criteria: they exhibit differential expression in the discordant twins (q < 0.15; Fig. 3A and C), as well as differential methylation in both the discordant twins (P < 0.05; Fig. 3B and D) and case-control cohort 2 (q < 0.15; Fig. 3B and D). In agreement with our expression data in discordant twins, we also found increased expression of S100A4 and SLC37A2 in patients with diabetes in case-control cohort 2 (Fig. 3A and C).

To further investigate the genetic contribution to global methylation variability, we studied genome-wide DNA methylation in adipose tissue from 10 MZ and 10 same-sex DZ twin pairs with NGT (Table 1). We also included 20 unrelated individuals with NGT, pairwise matched for age and sex, in this analysis. The genome-wide methylation data correlated more strongly in MZ versus DZ twin pairs, while the weakest correlation was found in unrelated individuals (Fig. 4A), supporting a genetic impact on adipose tissue DNA methylation. When we estimated correlations for methylation including only 15,627 sites differentially methylated in case-control cohort 2, correlations were weaker but still strongest in MZ twins (Fig. 4B). We next estimated heritability values (27) for DNA methylation in adipose tissue of 10 MZ and 10 DZ twin pairs with NGT. While heritability was 0.18 when using the average degree of methylation of all analyzed sites on the array, heritability was 0.28 when only the 15,627 differentially methylated sites (identified in case-control cohort 2) were included. This is in agreement with the results of a previous study (38). Also after removing CpG sites directly affected by SNPs (37) from this analysis, heritability values were 0.18 and 0.28, respectively. The influence of shared environmental factors on the methylation pattern was low (c² = 0.07 for all sites; c² = 0.08 for 15,627 differentially methylated sites). This is also in agreement with the findings of Grundberg et al. (38).

Epigenetic modifications have been associated with transcriptional regulation (8,16,39). Therefore, we related mRNA expression with DNA methylation for the 197 differentially expressed genes in Supplementary Table 3. DNA methylation of 266 sites, corresponding to 103 genes, were significantly associated with expression in the discordant twins at q < 0.15 (Supplementary Table 18). Additionally, among the genes with largest expression differences between diabetic and nondiabetic co-twins (Table 2), there were significant associations between methylation and expression for 20 sites corresponding to 11 genes (Supplementary Table 19). Among differentially expressed obesity and T2D candidate genes (Supplementary Tables 4 and 5), there were significant associations between methylation and expression for 65 sites corresponding to 17 genes (Supplementary Table 20). Representative associations between methylation and expression are shown in Fig. 3E and F. DNA methylation is known to regulate expression in different ways, depending on the location of CpG sites (39). Hence, we examined the genomic location of CpG sites where methylation correlated with expression (Fig. 3G–J). Methylation sites showing inverse relationships with expression were over-represented in promoter regions, whereas methylation sites showing positive relationships with expression were over-represented in gene bodies and 3′ UTRs (Fig. 3G and H), which is in agreement with the current literature (39). We next examined whether the expression of genes that may contribute to the regulation of DNA methylation was altered in the discordant twins. While TET1 expression was decreased in patients with diabetes (q < 0.15), numerous other candidate genes were only nominally regulated (Supplementary Table 21).

We finally tested whether we could replicate findings from a previous study (13) in which methylation was analyzed with the Illumina 27k array. After correction for multiple testing, Ribel-Madsen et al. (13) identified two CpG sites with methylation differences of >3% in adipose tissue of five twin pairs discordant for T2D. These sites were also differentially methylated in the discordant twins included in our study (Supplementary Fig. 2).

Despite the awareness of lifestyle factors in modulating the risk of T2D, the molecular mechanisms behind such acquired effects remain largely unknown. In this study, we capitalized on the strengths of a twin study design to present for the first time both genome-wide mRNA and DNA methylation profiles in adipose tissue from MZ twin pairs discordant for T2D. There was a significant enrichment of gene expression disruption to pathways relevant in oxidative phosphorylation; and carbohydrate, amino acid, and lipid metabolism in diabetic co-twins. In addition to decreased expression of numerous genes with key functions in mitochondria, diabetic twins also displayed reduced mtDNA content. Disturbed mitochondrial metabolism in adipose tissue has been suggested to shift lipid storage into other tissues, such as skeletal muscle, liver, and pancreas, resulting in severe insulin resistance (40).

Metabolic disorders related to obesity-associated insulin resistance have been characterized by an increased influx of inflammatory cells into adipose tissue (41). Here, we found increased expression of four gene sets related to the immune system in adipose tissue from diabetic twins. The most over-expressed gene in adipose tissue of diabetic co-twins was SPP1 encoding the inflammatory cytokine osteopontin, which recruits macrophages into adipose tissue and stimulates T-cell proliferation during inflammation (34). Mice lacking osteopontin are protected against developing insulin resistance despite diet-induced obesity (42). Recent human studies (43) suggest a link between gastric inhibitory polypeptide and osteopontin in adipose tissue and insulin resistance. Interestingly, SPP1 was over-expressed in adipose tissue of the obese co-twin in healthy MZ twin pairs discordant for BMI (44), suggesting that this defect is present before the development of overt T2D. This was further supported by our biological validation experiments where SPP1 expression was increased in adipose tissue from obese versus nonobese NGT individuals.

Furthermore, GYS2, ELOVL6, and FADS1 were the most downregulated genes in adipose tissue from diabetic twins. Our follow-up analysis in an independent case-control cohort confirmed the decreased expression of these genes in patients with T2D. GYS2 encodes the rate-limiting enzyme in the synthesis of glycogen in adipose tissue, which permits continued uptake of glucose into cells (32). Adipose tissue glycogen serves as a source of glycerol 3-phosphate, which is required for esterification (or re-esterification) of fatty acids into triglycerides. ELOVL6 encodes a key enzyme involved in the elongation of long-chain fatty acids (31), and FADS1 encodes an enzyme that introduces double bonds into growing fatty acid chains (30). Interestingly, recent GWASs (45–47) have identified SNPs near FADS1 that are associated with conditions of altered metabolism, including f-glucose, dyslipidemia, and circulating sphingolipid levels. Our findings of decreased GYS2, ELOVL6, and FADS1 expression in adipose tissue from patients with diabetes could potentially explain the reduced glucose uptake and impaired ability to store lipids in the adipose tissue of these individuals.

Unsupervised clustering proposed a large genetic contribution to the methylation variability as the affected twin from the pair discordant for T2D was epigenetically “closer” to his/her unaffected co-twin than to the other diabetic twins. In further support of a large genetic contribution, the methylation data correlated strongest within healthy MZ twin pairs compared with DZ twins and unrelated individuals. Our heritability estimates also support genetic effects. This is in line with a previous study (48) in which we showed that T2D SNPs affect DNA methylation in human pancreatic islets. Others have also proposed strong genetic effects on the DNA methylation pattern (13,38,49,50). However, 15,627 CpG sites in/near ∼30% of all genes exhibited differential methylation in adipose tissue from a case-control cohort of unrelated individuals, supporting a key role for epigenetic modifications in T2D patients. Moreover, 1,410 of these 15,627 sites were also differentially methylated in the discordant twins and ∼50% of the differentially expressed genes in discordant twins were associated with DNA methylation. These results support the idea that DNA methylation affects the phenotype in adipose tissue from discordant twins too. Additional epigenetic mechanisms such as histone modifications or microRNAs may also play a role in the twins. However, this remains to be investigated in future studies. The identification of CNVs in the discordant twins provides an additional potential explanation for phenotypic variation within MZ twin pairs.

Protein glycosylation is a common post-translational modification. Seven significantly upregulated gene sets in diabetic twins belong to glycan biosynthesis and metabolism. Interestingly, we also found enrichment of differentially methylated genes related to glycan metabolism in the pathway analysis in adipose tissue from unrelated patients with diabetes versus nondiabetic subjects. N-glycosylation is important for efficient trafficking of GLUT4 to its proper compartments in adipocytes (51), suggesting one mechanism by which dysregulation of these genes could contribute to T2D.

MZ twin pairs discordant for T2D are rare, and our relatively small sample size increases the risk of type II errors. More genes may have been identified in a larger cohort. However, we were able to validate eight of eight of our most differentially expressed genes in unrelated case-control cohorts, suggesting that our results represent true findings with biological relevance in people other than twins. It is difficult to draw conclusions about causality for differentially methylated and expressed genes in case-control cohorts. The ideal study design would be to longitudinally assess methylation and expression changes in adipose tissue during an individual’s transition into disease. However, genes analyzed in our replication cohort already had differential expression in obese versus nonobese NGT subjects. Additionally, BMI and/or glucose levels correlated with expression and/or methylation in nondiabetic subjects. Obesity increases the risk for T2D, and, hence, it is possible that altered expression and/or methylation of these genes may contribute to the development of T2D.

In conclusion, our study adds to the understanding of molecular mechanisms that contribute to T2D. Decreased expression of genes involved in energy metabolism and increased expression of inflammatory genes in adipose tissue accompany T2D. Importantly, our expression results from twins discordant for T2D could be replicated in case-control cohorts of unrelated subjects, showing the importance of our findings to the general population. Finally, our study highlights the importance of epigenetics in T2D as adipose tissue from unrelated subjects with the disease exhibit genome-wide methylation changes.

Acknowledgments. The authors thank Ylva Wessman and Targ Elgzyri (Scania University Hospital, Malmö, Sweden) for skilled technical assistance and collection of clinical material. The authors also thank Maria Sterner (Department of Clinical Sciences, Lund University, Malmö, Sweden) for analysis of genotype data. The authors also thank the Swegene Center for Integrative Biology at Lund University for analyzing global mRNA expression and DNA methylation.

Funding. This work was supported by grants from the Swedish Research Council (Dnr 521-2010-2745 and Dnr 523-2010-1061), the Swedish Medical Research Council (K2008-55X-15358-04-3), the Danish Council for Strategic Research, the Danish Council for Independent Research, a Linnaeus grant (LUDC Dnr 349-2008-6589), and a strategic research area grant (EXODIAB [Excellence Of Diabetes Research in Sweden] Dnr 2009-1039), as well as equipment grants from the Knut and Alice Wallenberg Foundation (2009-0243), the Lundberg Foundation (grant no. 359), Avtal om Läkarutbildning och Forskning, the Novo Nordisk Foundation, the Tore Nilsson Foundation, the Syskonen Svensson Foundation, the Diabetes Foundation, Kungliga Fysiografiska Sällskapet i Lund, the European Foundation for the Study of Diabetes/Lilly Foundation, the Söderberg Foundation, and the Påhlsson Foundation. The group of Swedish twins was recruited from the Swedish Twin Registry, which is supported by grants from the Swedish Department of Higher Education and the Swedish Research Council. The Centre of Inflammation and Metabolism is supported by a grant from the Danish National Research Foundation (DNRF55), and the Centre for Physical Activity Research is supported by a grant from TrygFonden.

Duality of Interest. M.K.S. is currently employed by Amgen AB. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.N. researched the data and wrote the manuscript. P.A.J., M.P., M.K.S., P.P., B.K.P., and C.S. collected the clinical data and reviewed the manuscript. A.P., P.V., P.A., J.F., and T.R. analyzed the data and reviewed the manuscript. R.R.-M. and N.L.P. contributed to the study design and reviewed the manuscript. A.V. designed the study, collected the clinical data, and reviewed the manuscript. C.L. designed the study and wrote and reviewed the manuscript. E.N. and C.L. are guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.