altered dna methylation and differential expressionofgenes...

Emma Nilsson,1,2 Per Anders Jansson,3 Alexander Perfilyev,1 Petr Volkov,1 Maria Pedersen,4

Maria K. Svensson,5 Pernille Poulsen,6 Rasmus Ribel-Madsen,2 Nancy L. Pedersen,7 Peter Almgren,8

João Fadista,8 Tina Rönn,1 Bente Klarlund Pedersen,4 Camilla Scheele,4 Allan Vaag,2 and Charlotte Ling1

Altered DNA Methylation andDifferential Expression of GenesInfluencing Metabolism andInflammation in Adipose TissueFrom Subjects With Type 2DiabetesDiabetes 2014;63:2962–2976 | DOI: 10.2337/db13-1459

Genetics, epigenetics, and environment may together af-fect the susceptibility for type 2 diabetes (T2D). Our aimwas to dissect molecular mechanisms underlying T2Dusing genome-wide expression and DNA methylation datain adipose tissue from monozygotic twin pairs discordantfor T2D and independent case-control cohorts. In adiposetissue from diabetic twins, we found decreased expres-sion of genes involved in oxidative phosphorylation; carbo-hydrate, amino acid, and lipid metabolism; and increasedexpression of genes involved in inflammation and glycandegradation. The most differentially expressed genes in-cluded ELOVL6, GYS2, FADS1, SPP1 (OPN), CCL18, andIL1RN. We replicated these results in adipose tissue froman independent case-control cohort. Several candidategenes for obesity and T2D (e.g., IRS1 and VEGFA) weredifferentially expressed in discordant twins. We founda heritable contribution to the genome-wide DNA methyl-ation variability in twins. Differences in methylation be-tween monozygotic twin pairs discordant for T2D weresubsequently modest. However, 15,627 sites, representing7,046 genes including PPARG, KCNQ1, TCF7L2, and IRS1,

showed differential DNA methylation in adipose tissuefrom unrelated subjects with T2D compared with controlsubjects. A total of 1,410 of these sites also showed dif-ferential DNA methylation in the twins discordant for T2D.For the differentially methylated sites, the heritability esti-mate was 0.28. We also identified copy number variants(CNVs) in monozygotic twin pairs discordant for T2D.Taken together, subjects with T2D exhibit multiple tran-scriptional and epigenetic changes in adipose tissue rele-vant to the development of the disease.

The susceptibility for type 2 diabetes (T2D) increases withage, physical inactivity, and obesity in subjects with a geneticpredisposition (1). However, the exact molecular mecha-nisms causing the disease still remain unknown.

Adipose tissue has a central role in whole-body energymetabolism as a dynamic store of triglycerides, and as anendocrine organ that coordinates energy intake and useby other tissues (2). In individuals with T2D, this functionis frequently perturbed by an impaired response of the

1Epigenetics and Diabetes, Department of Clinical Sciences, Lund UniversityDiabetes Centre, Lund University, Clinical Research Centre, Malmö, Sweden2Department of Endocrinology, Diabetes and Metabolism, Rigshospitalet, Copen-hagen, Denmark3Wallenberg Laboratory, Sahlgrenska University Hospital, Gothenburg, Sweden4The Centre of Inflammation and Metabolism and the Centre for PhysicalActivity Research, Department of Infectious Diseases, Rigshospitalet, Universityof Copenhagen, Copenhagen, Denmark5Institute of Medicine, Sahlgrenska University Hospital, University of Gothenburg,Gothenburg, Sweden6Global Development, Novo Nordisk A/S, Bagsværd, Denmark7Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stock-holm, Sweden

8Diabetes and Endocrinology, Department of Clinical Sciences, Lund Univer-sity Diabetes Centre, Clinical Research Centre, Lund University, Malmö,Sweden

Corresponding authors: Emma Nilsson, [email protected], andCharlotte Ling, [email protected].

Received 20 September 2013 and accepted 15 April 2014.

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db13-1459/-/DC1.

© 2014 by the American Diabetes Association. Readers may use this article aslong as the work is properly cited, the use is educational and not for profit, andthe work is not altered.

See accompanying article, p. 2901.

2962 Diabetes Volume 63, September 2014

METABOLISM

http://crossmark.crossref.org/dialog/?doi=10.2337/db13-1459&domain=pdf&date_stamp=2014-08-11

mailto:[email protected]

mailto:[email protected]

http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db13-1459/-/DC1


adipocytes to insulin resulting in elevated lipid levels incirculation and storage in alternative tissues such as liver,muscle, and pancreas (3).

One link between environment and disease is epigeneticsinfluencing gene transcription and subsequently organ func-tion (4). We have previously shown that epigenetic modifi-cations may accumulate during aging (5,6), and that DNAmethylation in humans is influenced by diet, birth weight,and exercise (7–10), suggesting that epigenetics could beinvolved in age-related and life style–related diseases suchas T2D. Indeed, studies from our group and others (11–17)have identified epigenetic modifications in patients withT2D; e.g., we found subtle nongenetic methylation differ-ences in adipose tissue from five monozygotic (MZ) twinpairs discordant for T2D using the Infinium 27k array(13). Nevertheless, our knowledge about the role of epige-netics in the growing incidence of T2D remains limited, andthe genome-wide expression profile has, to our knowledge,not been linked to the epigenome in adipose tissue ofdiabetic patients. Therefore, our aim was to investigategenome-wide differences in expression and DNAmethylationin adipose tissue from MZ twin pairs discordant for T2D.This study design allows the elimination of confounding fac-tors such as genotype, age, and sex. The causes of incompleteconcordance between MZ twins are traditionally explained bynonshared environmental factors. To investigate the impor-tance of our findings in the general population, the expres-sion of selected genes was validated in case-control cohorts.Finally, heritability of DNA methylation was estimated inadipose tissue from nondiabetic twins, and the genome-

wide methylation pattern was analyzed in adipose tissuefrom a case-control cohort of unrelated subjects.

RESEARCH DESIGN AND METHODS

Study Participants

Discordant TwinsFourteen MZ twin pairs discordant for T2D were recruitedthrough Scandinavian twin registries (18) (Table 1). Five ofthe twin pairs have previously been studied using the Illu-mina (San Diego, CA) 27k array (13). Zygosity was confirmedand copy number variants (CNVs) predicted by analysisof 730,525 genetic markers using HumanOmniExpressarrays (Illumina).

Case-Control Cohort 1One hundred twenty unrelated subjects with normalglucose tolerance (NGT) or T2D were included for bi-ological validation (Table 1). They have previously beendescribed (19).

Case-Control Cohort 2The characteristics of 28 NGT and 28 T2D unrelatedsubjects, pairwise matched for age and sex, and selectedfrom a larger twin cohort (20,21) are shown in Table 1.

Twin Cohort for Heritability EstimatesTen MZ and 10 same-sex dizygotic (DZ) twin pairs, allwith NGT, were selected from a larger cohort (20,21) andused for heritability estimates (Table 1). Zygosity wasconfirmed by genetic markers.

Table 1—Clinical characteristics of study subjects included in the discordant twin cohort, the two case-control cohorts, and thetwin cohort for heritability estimates

Characteristics

Discordant twins Case-control cohort 1 Case-control cohort 2Twin cohort for heritability

estimates

Nondiabetic T2Da NGT T2Db NGT T2D MZ DZ

N (male/female) 14 (9/5) 14 (9/5) 70 (32/38) 50 (26/24) 28 (15/13) 28 (15/13) 20 (10/10) 20 (10/10)

Age (years) 67.6 6 7.7 67.6 6 7.7 53.4 6 7.4 55.5 6 6.3 74.3 6 4.3 74.5 6 4.2 71.0 6 5.4 70.6 6 5.9

BMI (kg/m2) 29.8 6 6.8 32.0 6 7.1c 29.4 6 5.9 29.6 6 5.8 27.0 6 3.6 27.4 6 3.6 26.7 6 2.1 26.5 6 2.9

Fat% (n = 9 pairs)d 30.5 6 8.8 33.6 6 9.4c

Physical fitness (a.u.)(n = 6 pairs)e 2.6 6 0.8 2.2 6 1.3

Fasting plasmaglucose (mmol/L) 6.0 6 0.5 9.3 6 2.6f 5.6 6 0.5 9.3 6 3.0f 5.5 6 0.5 7.1 6 1.9f 5.4 6 0.3 5.4 6 0.4

2-h glucose (mmol/L) 8.3 6 1.8 16.1 6 5.2c 6.0 6 1.3 16.5 6 5.0f 6.6 6 0.7 15.1 6 5.0f 6.1 6 1.2 5.9 6 1.0

HbA1c

% 5.9 6 0.4 7.5 6 1.8c 5.5 6 0.3 6.8 6 1.1f 5.7 6 0.3 6.6 6 1.3f 5.6 6 0.19 5.7 6 0.23mmol/mol 41.0 6 4.4 58.0 6 19.7c 37.0 6 3.3 51.0 6 12.0f 39.0 6 3.3 49.0 6 14.0f 38.0 6 2.1 39.0 6 2.5

Data are shown as mean 6 SD. Among the discordant twins, the nondiabetic co-twin had NGT in 4 pairs and impaired glucosetolerance in 10 pairs. Glucose tolerance was determined according to the 1999 World Health Organization criteria. Discordant twinpairs were recruited through the Swedish (18) (9 pairs) and Danish (5 pairs) twin registries. a.u.; arbitrary units. aIn discordant twins, 11%of the T2D subjects received insulin treatment, 33% received oral agents (metformin, sulfonylurea, or rosiglitazone), and 33% receivedboth insulin and oral agents. bPatients with diabetes in case-control cohort 1 were on average 12 years younger than the discordanttwins. None of the subjects received insulin treatment and 50% received oral agents (metformin, sulfonylurea, or sitagliptin). cP , 0.05,T2D vs. nondiabetic/NGT subjects. dFat percentage was measured using bioimpedance. eData of self-reported physical fitness wasretrieved from a questionnaire (levels 1–5 [1, poor physical fitness; 5, very high physical fitness]). fP , 0.001, T2D vs. nondiabetic/NGTsubjects.

diabetes.diabetesjournals.org Nilsson and Associates 2963

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

Clinical ExaminationAfter an overnight fast, participants visited clinics for met-abolic characterization, and subcutaneous adipose tissue bi-opsy samples were obtained under local anesthesia, frozenimmediately in liquid nitrogen, and stored at 280°C. Glu-cose tolerance was measured by a 75-g oral glucose toler-ance test.

RNA and DNA Extractions From Adipose TissueRNA was extracted using the miRNeasy kit followed by theRNeasy MinElute Cleanup kit (Qiagen, Hilden, Germany)for twins or by TRIzol (Invitrogen, Carlsbad, CA) for non-twins. DNA was extracted using the DNeasy Blood andTissue kit (Qiagen). DNA/RNA quantity and purity weredetermined by spectrophotometry (NanoDrop Technolo-gies, Wilmington, DE), and RNA integrity was determinedusing the Experion system (Bio-Rad, Hercules, CA).

Expression ArraysmRNA expression was analyzed in adipose tissue fromdiscordant twins using GeneChip Human Gene 1.0 STarrays (Affymetrix, Santa Clara, CA) according to themanufacturer’s recommendations. We computed RobustMultichip Average expression measures using the Oligopackage in Bioconductor (22).

Gene Set Enrichment AnalysisWe applied gene set enrichment analysis (GSEA) (23) toexpression array data using KEGG (Kyoto Encyclopedia ofGenes and Genomes) pathways. All probes correspondingto transcripts were used and ranked according to thet statistics in a paired t test. GSEA was run with thehighest occurrence for genes with multiple probes andconsidered pathways with 1–500 transcripts.

DNA Methylation ArraysDNA methylation was analyzed in adipose tissue of dis-cordant twins, case-control cohort 2, and the twin cohort forheritability estimates using Infinium HumanMethylation450BeadChips (Illumina). This array contains 485,577 probes,which cover 21,231 (99%) RefSeq genes (24). DNA wastreated with bisulfite using the EZ DNA Methylation kit(Zymo Research, Orange, CA). Analysis was performed fol-lowing the Infinium HD Assay Methylation protocol (Illu-mina). To avoid a batch effect, co-twins and matched pairswere hybridized on the same chip. The bioinformatics anal-yses were performed as previously described (9). Probeswith a mean detection P value of .0.01, non-cytosine gua-nine dinucleotide (CpG) probes, Y-chromosome probes, and65 single nucleotide polymorphism (SNP) probes were re-moved. b-Values, ranging from 0 to 100% methylation,were used to report the outcome.

Validation of Expression Array DatacDNA was produced using the QuantiTect Reverse Tran-scription Kit (Qiagen). mRNA levels were detected withthe 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 HPRTexpression (4326321E; Applied Biosystems). NormFinder(25) software demonstrated stable HPRT expression.

Mitochondrial DNA ContentAssays (Hs02596879_g1, Hs02596860_s1 and Hs02596867_s1; Applied Biosystems) targeting ND6, RNR2 and CYTBwere used to analyze mitochondrial DNA (mtDNA) con-tent with the ViiA 7 system (Applied Biosystems). Sam-ples were analyzed in duplicate and normalized to thequantity of the nuclear gene RNAseP (4316831; AppliedBiosystems).

StatisticsComparisons between discordant twins and individuals incase-control cohort 2 are based on paired Wilcoxonstatistics. To account for multiple testing, we appliedfalse discovery rate (FDR) analyses. Regression analysiswas used to relate expression and methylation in discor-dant 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 geno-typing data were performed with the PennCNV softwaretool (26). PennCNV implements a hidden Markov modelthat integrates signal intensity data and SNP allelic ratiodistribution from the Illumina HumanOmniExpress chip.Heritability and common environmental influences (c2)were estimated using DF analyses with double-entry twindata (27).

RESULTS

Clinical CharacteristicsThe characteristics of subjects included in the study co-horts are shown in Table 1. Fasting glucose (f-glucose)levels, glucose tolerance, and HbA1c levels differed signif-icantly between T2D and nondiabetic subjects. In discor-dant twins, BMI and fat percentage were significantlyincreased in subjects with diabetes, but the physical fit-ness level did not differ between co-twins. In twins usedfor heritability estimates, there were no differences inage, BMI, glucose levels, or HbA1c levels between MZand same-sex DZ twins.

Differential Expression in MZ Twins Discordant for T2DExpression data were obtained from the adipose tissue of12 twin pairs discordant for T2D. We first tested whethersets of related genes were altered in diabetic versusnondiabetic twins. The GSEA yielded 14 significant genesets with downregulated expression and 17 gene sets withupregulated expression in adipose tissue from diabeticversus nondiabetic co-twins (q , 0.05; Table 2). Down-regulated gene sets include pathways involved in oxidativephosphorylation and amino acid, carbohydrate, and lipid

2964 Epigenetics in T2D Discordant Twins Diabetes Volume 63, September 2014

Tab

le2—Signifi

cantgene

setswith

differentialexp

ressionin

adipose

tissueofdiab

eticco

mpared

with

nond

iabetic

co-tw

ins(G

SEAanalysis

with

q<0.05)

Gene

setnam

ePathw

aygroup

Regulated

/totalES

Pvalue

qvalue

Dow

nregulatedgene

setsHSA00280

valine,leucine,

andisoleucine

degrad

ationAmino

acidmetab

olism26/44

20.67

,0.001

,0.001

HSA00640

prop

anoatemetab

olismCarb

ohydrate

metab

olism21/33

20.72

,0.001

,0.001

HSA00020

citratecycle

Carb

ohydrate

metab

olism21/27

20.68

,0.001

,0.001

HSA01040

polyunsaturated

fattyacid

biosynthesis

Lipid

metab

olism10/13

20.79

,0.001

0.001HSA00650

butanoate

metab

olismCarb

ohydrate

metab

olism19/45

20.56

,0.001

0.001HSA00190

oxidative

phosp

horylationEnergy

metab

olism52/112

20.48

,0.001

0.001HSA00071

fattyacid

metab

olismLip

idmetab

olism19/44

20.55

,0.001

0.003HSA00620

pyruvate

metab

olismCarb

ohydrate

metab

olism19/41

20.53

0.0020.007

HSA03022

basaltranscrip

tionfactors

Transcription

17/3220.52

0.0040.03

HSA00072

synthesisand

degrad

ationof

ketonebod

iesLip

idmetab

olism4/9

20.73

,0.001

0.03HSA00720

reductive

carboxylate

cycleEnergy

metab

olism9/11

20.68

0.010.03

HSA00310

lysinedegrad

ationAmino

acidmetab

olism18/47

20.46

0.0020.03

HSA00252

alanineand

aspartate

metab

olismAmino

acidmetab

olism13/33

20.50

0.0020.03

HSA00061

fattyacid

biosynthesis

Lipid

metab

olism4/6

20.79

0.0080.045

Upregulated

genesets

HSA01032

glycanstructures

degrad

ationGlycan

biosynthesis

andmetab

olism16/29

0.71,0.001

,0.001

HSA00603

glycosphingolipid

biosynthesis

globoseries

Glycan

biosynthesis

andmetab

olism7/14

0.88,0.001

,0.001

HSA04610

complem

entand

coagulationcascades

Immune

system32/68

0.55,0.001

,0.001

HSA00511

N-glycan

degrad

ationGlycan

biosynthesis

andmetab

olism10/15

0.79,0.001

,0.001

HSA04640

hematop

oieticcelllineage

Immune

system43/86

0.48,0.001

0.001HSA04612

antigenprocessing

andpresentation

Immune

system28/78

0.48,0.001

0.002HSA00604

glycosphingolipid

biosynthesis

ganglioseriesGlycan

biosynthesis

andmetab

olism5/16

0.69,0.001

0.004HSA00532

glycosaminoglycan

biosynthesis

Glycan

biosynthesis

andmetab

olism9/17

0.67,0.001

0.006HSA00531

glycosaminoglycan

degrad

ationGlycan

biosynthesis

andmetab

olism8/17

0.660.004

0.007HSA04514

celladhesion

molecules

Signaling

molecules

andinteraction

58/1300.42

,0.001

0.008HSA01030

glycanstructures

biosynthesis

1Glycan

biosynthesis

andmetab

olism42/109

0.42,0.001

0.02HSA03010

ribosom

eTranslation

27/670.45

,0.001

0.02HSA04060

cytokinecytokine

receptor

interactionSignaling

molecules

andinteraction

94/2440.37

,0.001

0.02HSA04940

type1diab

etesmellitus

End

ocrineand

metab

olicdiseases

22/410.49

,0.001

0.02HSA00600

sphingolip

idmetab

olismLip

idmetab

olism13/36

0.510.002

0.02HSA04512

ECM

receptor

interactionSignaling

molecules

andinteraction

41/850.43

,0.001

0.02HSA04620

Toll-likerecep

torsignaling

pathw

ayIm

mune

system31/98

0.39,0.001

0.03

ES,enrichm

entscore

(theprim

aryoutcom

eof

GSEA);Regulated

/total,num

ber

ofdow

nregulatedor

upregulated

genesout

ofthe

totalnum

ber

ofgenes

included

inthe

pathw

ay;ECM,

extracellularmatrix.


metabolism; and upregulated gene sets include pathwaysinvolved in inflammation and glycan metabolism. Genescontributing to the enrichment for significant gene setsare presented in Supplementary Tables 1 and 2. More-over, selected gene sets representing metabolism and im-mune system pathways are shown in Fig. 1A.

We next tested whether the expression of individualgenes was altered in adipose tissue from diabetic versusnondiabetic co-twins. After FDR correction, 197 individ-ual genes were found to be differentially expressed indiabetic versus nondiabetic co-twins (q , 0.15; Supple-mentary Table 3). One hundred sixteen of these genes

Figure 1—Differential mRNA expression and mtDNA content in adipose tissue from subjects with T2D compared with nondiabetic sub-jects. A: Genes contributing to the significant enrichment scores of GSEA for gene sets involved in oxidative phosphorylation (OXPHOS)divided by the different complexes, pyruvate metabolism, and Toll-like receptor signaling pathway in adipose tissue of diabetic vs. non-diabetic co-twins. B–E: Technical and biological validation of six metabolically relevant genes selected from among the ones with thelargest expression differences in adipose tissue of diabetic vs. nondiabetic co-twins. B: Expression data obtained from the array. C:Technical validation of the expression data from the array with qPCR. D: Biological validation of expression differences found in the MZtwins discordant for T2D in adipose tissue from 120 unrelated subjects with NGT or T2D (case-control cohort 1). E: Comparison of mRNAexpression levels in adipose tissue from nonobese (BMI <30 kg/m2, n = 40) and obese (BMI>30 kg/m2, n = 30) NGT subjects. F: ReducedmtDNA content in adipose tissue from diabetic compared with nondiabetic co-twins. Three mitochondrial genes, CYB, ND6, and RNR2,were quantified by qPCR and normalized to the quantity of the nuclear gene RNAseP. Data are shown as the mean 6 SD. *P < 0.05compared with nondiabetic/NGT. a.u., arbitrary units. For Affymetrix expression data, while log2 intensity values after robust multiarrayaverage data processing and normalization were used for all statistical analyses, unlogged expression values are plotted in the figures.





were upregulated, and 81 genes were downregulated indiabetic twins. The expression differences of these genesrange from 5% to 43% (Supplementary Table 3). A post hocpower calculation to detect expression differences of $5%is presented in Supplementary Table 4. Furthermore, geneswith the largest expression differences between diabeticand nondiabetic co-twins (P , 0.05) are shown in Table3. Six genes were technically validated by quantitative real-time PCR (qPCR) in the adipose tissue of twins with arraydata. We observed significant correlations between the twomethods for each gene (r = 0.76–0.97, P , 0.0001; Sup-plementary Fig. 1), and there were significant differencesbetween diabetic and nondiabetic co-twins with fold differ-ences of the same magnitude for qPCR compared witharray data (Fig. 1B and C).

We proceeded to investigate whether genes previouslylinked to obesity and T2D in published genome-wideassociation studies (GWASs) (P , 1.0 3 1025 [www.genome.gov/gwastudies, accessed 22 August 2012]) weredifferently expressed in adipose tissue from twins discor-dant for T2D. These genes are associated with these traitsthrough SNPs, which may be located in intergenic regions,and they do not need to have functional evidence as causalgenes for obesity or T2D. We found 27 of 144 obesitygenes with differential expression in adipose tissue fromdiscordant twins (e.g., IRS1 and VEGFA, P , 0.05; Supple-mentary Table 5). This was more than expected (P = 0.02,x2 test). Among the 87 T2D candidate genes, 8 were dif-ferentially expressed in discordant twins (e.g., PPARG andGLIS3, P , 0.05; Supplementary Table 6). This was notmore than expected (P = 0.4, x2 test). As some candidategenes are associated with both obesity and T2D, the dif-ferentially expressed candidate genes correspond to 33unique genes.

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

Biological Validation of mRNA ExpressionTo biologically validate our findings in discordant twins, weanalyzed the expression of six selected genes in adiposetissue from unrelated subjects with NGT or T2D (case-control cohort 1). The selected genes have known functionsrelated to fat metabolism (ELOVL6 and FADS1) (30,31),glucose metabolism (GYS2) (32), or inflammation (SPP1[OPN], CCL18, and IL1RN) (33–35); and were selectedfrom the genes contributing to the enrichment scores ofGSEA and/or from the list of most downregulated andupregulated genes in discordant twins. All six genes couldbe validated, and showed significant expression differences

between unrelated patients with diabetes and nondiabeticsubjects in the same direction as the discordant twins (Fig.1B and D). This was also true when excluding obese sub-jects (BMI.30 kg/m2) and only comparing nonobese NGTwith 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 expres-sion of these six genes by comparing nonobese and obeseNGT subjects. Their characteristics are presented in Sup-plementary Table 9. ELOVL6 and GYS2 had significantlylower expression, and SPP1 and IL1RN had significantlyhigher expression in obese versus nonobese NGT subjects(Fig. 1E). Additionally, the expression of all genes corre-lated significantly with BMI and/or glucose levels in NGTsubjects (Supplementary Table 10).

mtDNA Content in Discordant TwinsSince the expression of genes representing oxidativephosphorylation is decreased in adipose tissue of T2Dtwins (Table 2 and Fig. 1A), we quantified mtDNA contentto follow up these results. The mtDNA content was re-duced in adipose tissue from diabetic versus nondiabeticco-twins (Fig. 1F), which is in line with our expressionresults.

CNVs in Discordant TwinsCNVs may contribute to phenotypic variation in twins(36). We therefore analyzed CNVs in MZ twins discor-dant for T2D and filtered CNVs present in two or moreindividuals of the same status. We identified six differ-ent CNVs, including three that were present only in di-abetic twins and three present only in nondiabetic twins(Table 4).

Global DNA Methylation in MZ Twins Discordant forT2DWe next studied the global DNA methylation pattern withthe Infinium HumanMethylation450 BeadChip in adiposetissue from 14 twin pairs discordant for T2D. After qualitycontrol (QC) and filtering, methylation data were obtainedfor 480,403 CpG sites. To evaluate the global humanmethylome in adipose tissue, we calculated the averagelevel of methylation for all sites divided into groups basedon 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 averageDNA methylation for these regions between diabetic andnondiabetic co-twins. While the average methylationlevel was high within the gene body, 39 untranslated region(UTR), and intergenic regions, it was low in TSS1500,TSS200, 59 UTR, and the first exon (Fig. 2A). Moreover,the methylation level was low within CpG islands and in-termediate within shores, whereas shelves and open seashowed the highest methylation levels (Fig. 2B).

To visualize the overall relationship between DNAmethylation profiles of individuals, both within and betweentwin pairs, hierarchical cluster analysis was applied to themethylation data using all CpG sites that passed QC and






http://www.genome.gov/gwastudies











Tab

le3—

Individua

lgen

eswiththelarges

tmRNA

express

iondifferen

cesin

adipose

tiss

uebetwee

ndiabetic

andno

ndiabetic

co-twins(P

<0.05

)

Gen

esy

mbol

Gen

edes

criptio

nNon

-T2D

(mea

n6

SD)

T2D

(mea

n6

SD)

Differen

ce(%

)P

value

qva

lue

Gen

eswith

lower

expressionin

diabetic

compared

with

nond

iabe

ticco

-twins

ELO

VL6

ELO

VLfattyac

idelon

gase

626

5.36

328.1

102.86

31.5

261

.30.01

0.23

GYS2

glyc

ogen

syntha

se2

216.46

206.9

104.76

64.6

251

.60.02

0.25

FADS1

fattyac

iddes

aturas

e1

1,01

0.16

607.5

529.96

223.6

247

.50.00

50.19

C12

orf39

chromos

ome12

open

read

ingfram

e39

734.36

735.8

392.56

449.8

246

.50.00

50.19

SAA1

serum

amyloidA1

132.86

93.1

80.3

662

.3239

.50.00

20.17

STO

X1

storkh

eadbox

120

7.86

89.0

129.76

56.8

237

.60.01

0.21

CASQ2

calseq

uestrin

226

2.16

142.4

165.16

97.3

237

.00.00

20.17

AGPAT9

1-ac

ylglyc

erol-3-pho

spha

teO-acy

ltran

sferas

e9

146.76

97.5

94.5

627

.1235

.60.01

0.21

FADS2

fattyac

iddes

aturas

e2

722.36

316.9

465.46

136.2

235

.60.01

0.23

B4G

ALT

6UDP-G

al:betaG

lcNAcbeta1,4-

galactos

yltran

sferas

e,polyp

eptid

e6

377.06

300.7

247.66

196.6

234

.30.01

0.22

Gen

eswith

high

erex

pressionin

diabetic

compared

with

nond

iabe

ticco

-twins

SPP1(OPN)

secreted

phos

pho

protein

1(osteo

pon

tin)

500.86

399.0

1,10

2.66

643.1

120.2

0.01

0.21

TM4S

F19

tran

smem

brane

4Lsixfamily

mem

ber

1925

3.76

189.3

485.16

220.1

91.2

0.01

0.22

MMP9

matrix

metallopep

tidas

e9(gelatinas

eB,92

kDage

latin

ase,

92kD

atype

IVco

llage

nase

)29

3.56

189.5

557.36

255.3

89.9

0.00

20.17

CCL1

8ch

emok

ine(C-C

motif)

ligan

d18

178.56

180.4

335.36

201.1

87.8

0.00

50.19

PRG4

proteo

glyc

an4

567.46

370.6

1,06

2.46

694.0

87.2

0.00

50.19

IL1R

Ninterle

ukin

1rece

ptoran

tago

nist

134.66

77.1

244.86

112.0

81.9

0.00

10.15

PLA

2G7

phos

pho

lipas

eA2,

grou

pVII(platelet-ac

tivatingfactor

acetylhy

drolase

)23

3.96

195.4

413.26

166.2

76.7

0.01

0.23

MSR1

mac

ropha

gesc

aven

gerrece

ptor1

266.36

182.2

447.76

172.0

68.1

0.00

70.21

VSIG4

V-set

andim

mun

oglobu

lindom

ainco

ntaining

440

4.06

255.2

668.86

296.4

65.5

0.00

90.22

LGI2

leuc

ine-ric

hrepea

tLG

Ifamily,mem

ber

298

.06

35.5

139.96

38.3

42.8

0.00

050.14

ELO

VL,

elon

gatio

nof

long

-cha

infattyac

ids;

UDP,uridinedipho

spha

te.


filtering. Twelve of 14 twin pairs clustered together, sug-gesting a strong role for underlying genetic or commonenvironmental factors in determining the epigenetic profilein adipose tissue of discordant twin pairs (Fig. 2C). Afterremoving CpG sites directly affected by SNPs (37), 11 of 14twin pairs still clustered together.

Differential DNA Methylation of Individual Sites inPatients With Diabetes Versus Nondiabetic SubjectsWe further tested whether any individual sites exhibitdifferential methylation in the discordant twins. We found23,470 (;5%) of 480,403 analyzed CpG sites with differ-ential methylation in adipose tissue from diabetic versusnondiabetic twins at P , 0.05 (Supplementary Table 11).Of those, 11,494 sites had increased methylation, and11,976 sites had decreased methylation in diabetic twins.Moreover, 3,670 sites had an absolute difference in meth-ylation of $3%. However, none of these differencesremained significant after FDR correction, and the numberof differences (23,470) is not more than that expected bychance. A post hoc power calculation to detect absolutemethylation differences of $3% is presented in Supple-mentary Table 4.

These data suggest either that there are limited dif-ferences in methylation between patients with diabetesand nondiabetic subjects or that DNA methylation inhuman adipose tissue is highly heritable. We thereforeproceeded to test whether we could identify methylationdifferences in adipose tissue from unrelated subjectswith T2D compared with nondiabetic subjects, matchedfor age and sex (case-control cohort 2; Table 1). After QCand filtering, the Infinium array generated methylationdata for 481,086 sites. We found 71,074 sites (;15%)with differential DNA methylation at P , 0.05 in adiposetissue from case-control cohort 2. This is almost threetimes more than that expected by chance (P , 0.0001,x2 test). After FDR analysis, 15,627 sites representing7,046 unique genes were found to be differentially meth-ylated in patients with diabetes versus nondiabetic sub-jects (q , 0.15; Supplementary Table 12). The absolutedifferences in DNA methylation between patients withdiabetes and nondiabetic subjects are illustrated in Fig.2D and E. While 6,754 sites showed increased methyla-tion, 8,873 sites showed decreased methylation inpatients with diabetes (Fig. 2D and E). Moreover, CpGsites with differential methylation were over-representedamong sites with 20–70% methylation (Fig. 2F and G),suggesting that sites with an intermediate degree ofmethylation are more likely to change. We analyzed thedistribution of differentially methylated sites in patientswith diabetes versus nondiabetic subjects at q , 0.15(Supplementary Table 12), based either on their relation togene regions (Fig. 2H) or their relation to CpG islands (Fig.2I). Differentially methylated sites were over-represented inthe gene body and open sea, and under-represented inTSS1500, TSS200, and CpG islands compared with theprobe distribution on the array (Fig. 2H and I). Moreover,

Tab

le4—

Iden

tifica

tionofCNVsin

MZtw

inpairs

disco

rdan

tforT2D

Chr

Startpos

ition

End

positio

nSNPswith

inCNV(n)CNVleng

th(bp)Num

ber

ofco

pies

StartSNP

End

SNP

Close

stge

neDistanc

eto

clos

estge

ne(bp)Tw

instatus

1155

93444

055

94945

85

15,019

1rs43

0040

8rs46

1605

8OR5J

20

T2D

1155

94332

255

95677

36

13,452

1rs17

5331

14rs17

1501

47OR5J

20

T2D

1368

89795

868

93524

96

37,292

3rs95

9930

1rs13

8017

1LINC00

550

500,16

8Non

diabetic

1368

91880

668

94284

76

24,042

3rs79

9440

0rs95

8121

LINC00

550

492,57

0Non

diabetic

257

85083

657

86289

15

12,056

1rs82

0795

rs26

9560

5VRK2

410,88

6T2

D

257

85083

657

86289

15

12,056

1rs82

0795

rs26

9560

5VRK2

410,88

6T2

D

719

03085

319

03690

98

6,05

71

rs27

1734

4rs77

7673

5HDAC9

0T2

D

719

03419

119

03690

97

2,71

91

rs77

8883

3rs77

7673

5HDAC9

0T2

D

814

499

210

314

501

835

46

26,252

3rs65

5840

6rs11

7868

96PLE

C0

Non

diabetic

814

500

718

714

505

085

611

43,670

3rs11

1363

36rs28

3646

64MIR66

1,PLE

C0

Non

diabetic

9670

113

0670

582

45

4,69

51

rs82

0502

rs47

4223

9KDM4C

15,039

Non

diabetic

9670

113

0670

582

45

4,69

51

rs82

0502

rs47

4223

9KDM4C

15,039

Non

diabetic

9670

113

0670

582

45

4,69

51

rs82

0502

rs47

4223

9KDM4C

15,039

Non

diabetic

Chr,c

hrom

osom

e.CNVswereca

lledifthey

hadat

leas

tfive

SNPs,

weredetec

tedin

onlyon

eof

thetw

insin

apa

ir,an

dwerese

enat

leas

tin

twoor

morediabetic

orno

ndiabetic

individua

ls.

DNAan

dge

notypedatafortheCNVan

alys

iswereav

ailable

from

19MZtw

inpa

irsdisco

rdan

tforT2

D.







Figure 2—Impact of T2D on DNA methylation in human adipose tissue. Global DNA methylation in human adipose tissue of nondiabeticand T2D co-twins is shown for the nearest gene regions (A) and CpG island regions (B). Global DNA methylation is calculated as theaverage DNA methylation based on all CpG sites in each annotated region on the Infinium HumanMethylation450 BeadChip and is shownas the mean 6 SD. TSS, proximal promoter, defined as 200 or 1,500 bp upstream of the transcription start site. Shore, flanking region ofCpG islands (0–2,000 bp); Shelf, regions flanking island shores (2,000–4,000 bp from the CpG island). N, northern; S, southern. C: Clusterdendogram to visualize the overall relationship between DNA methylation profiles of individuals in the discordant twin cohort. Contr,nondiabetic subject; F, female; M, male. The cluster dendogram was generated using a hierarchical cluster analysis in R (http://stat


http://stat.ethz.ch/

based on the Illumina annotations, differentially methyl-ated sites were over-represented in enhancer regions(31% of significant sites compared with 21% of analyzedsites, P , 0.0001, x2 test).

To gain biological relevance of methylation differencesin adipose tissue from unrelated patients with diabetesversus nondiabetic subjects, we performed a KEGG path-way analysis of the 7,046 genes that exhibit differentialmethylation at q , 0.15 (Supplementary Table 12) usingWebGestalt. Pathways involved in inflammation, glycanmetabolism, cancer, Wnt signaling, and mitogen-activatedprotein kinase signaling were among those significantlyenriched (Fig. 2J and Supplementary Table 13).

We also examined whether candidate genes for T2Dand/or obesity, previously identified by GWAS (P , 1.0 31025 [www.genome.gov/gwastudies]), exhibit differentialmethylation in adipose tissue of unrelated patients withdiabetes versus nondiabetic subjects. We found differen-tial methylation in 123 sites (q , 0.15) representing 50T2D candidate genes including IRS1, PPARG, KCNQ1, andTCF7L2 (Supplementary Table 14 and Fig. 2K). This ismore than expected (P = 0.002, x2 test). Moreover, 127sites representing 65 candidate genes for obesity weredifferentially methylated in case-control cohort 2 (q ,0.15, P = 0.5, x2 test; Supplementary Table 15).

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

We proceeded to test whether any of the sites thatexhibit differential methylation in case-control cohort 2(q , 0.15; Supplementary Table 12) also exhibit differen-tial methylation in adipose tissue of the discordant twins atP , 0.05 (Supplementary Table 11). Of 15,627 differen-tially methylated CpG sites in case-control cohort 2, 1,410

sites were also differentially methylated in the same direc-tion in the discordant twins (P , 0.05; SupplementaryTable 12). This overlap is almost twice as much as expected(P , 0.0001, x2 test).

Based on the potential problem with cross-reactiveInfinium probes (37), we tested whether any of the probesthat detected differential methylation in our cohorts cross-react to alternative genomic locations (SupplementaryTables 11 and 12). Importantly, only 25 and 10 possiblycross-reactive probes, respectively, have a perfect match toanother location in the genome in the discordant twins andcase-control cohort 2. Additionally, we found fewer SNPprobes than expected (Supplementary Table 12; P ,0.0001, x2 test).

Expression in Case-Control Cohort 2We next analyzed the expression of two genes, S100A4(encoding S100 calcium-binding protein A4) and SLC37A2(encoding glucose-6-phosphate transporter), in adipose tis-sue 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 methylationin both the discordant twins (P , 0.05; Fig. 3B and D) andcase-control cohort 2 (q , 0.15; Fig. 3B and D). In agree-ment with our expression data in discordant twins, we alsofound increased expression of S100A4 and SLC37A2 inpatients with diabetes in case-control cohort 2 (Fig. 3Aand C).

Heritability of DNA Methylation in Human Adipose TissueTo further investigate the genetic contribution to globalmethylation variability, we studied genome-wide DNAmethylation in adipose tissue from 10 MZ and 10 same-sex DZ twin pairs with NGT (Table 1). We also included 20unrelated individuals with NGT, pairwise matched for ageand sex, in this analysis. The genome-wide methylationdata correlated more strongly in MZ versus DZ twin pairs,while the weakest correlation was found in unrelated indi-viduals (Fig. 4A), supporting a genetic impact on adiposetissue DNA methylation. When we estimated correlationsfor methylation including only 15,627 sites differentiallymethylated in case-control cohort 2, correlations were

.ethz.ch/R-manual/R-patched/library/stats/html/hclust.html). The absolute differences in DNA methylation of 15,627 individual sites, in-cluding 6,754 sites with increased (D) and 8,873 sites with decreased (E) DNA methylation in 28 diabetic subjects compared with 28nondiabetic unrelated subjects in case-control cohort 2. The degree of DNA methylation in adipose tissue of the 28 nondiabetic subjects isshown for the 6,754 CpG sites with increased DNA methylation (F) and the 8,873 CpG sites with decreased DNA methylation (G) in 28diabetic compared with 28 nondiabetic unrelated subjects in comparison with the degree of methylation of all analyzed CpG sites using theInfinium HumanMethylation450 BeadChip. Distributions of individual sites that exhibit differential DNA methylation in adipose tissue from28 diabetic compared with 28 nondiabetic unrelated subjects in relation to nearest gene regions (H) and CpG island regions (I). The dis-tribution of the significant sites compared with the distribution of all analyzed sites on the Infinium HumanMethylation450 BeadChip issignificantly different (P values) from what is expected by chance based on x2 tests (Pchi2). J: Significantly enriched KEGG pathways (FDR-adjusted P values<0.05) of genes that exhibit differential methylation in adipose tissue from 28 diabetic vs. 28 nondiabetic unrelated subjects.ECM, extracellular matrix; MAPK, mitogen-activated protein kinase; VEGF, vascular endothelial growth factor; GnRH, gonadotropin-releasinghormone; RI, receptor 1. K: Differential DNA methylation of IRS1, PPARG, KCNQ1, and TCF7L2 in adipose tissue from 28 diabetic vs. 28nondiabetic unrelated subjects. The three most significant sites for each gene are presented. Data are shown as the mean 6 SD. *q< 0.15compared with nondiabetic subjects.















weaker but still strongest in MZ twins (Fig. 4B). We nextestimated heritability values (27) for DNA methylation inadipose tissue of 10 MZ and 10 DZ twin pairs with NGT.While heritability was 0.18 when using the average degreeof methylation of all analyzed sites on the array, heritabilitywas 0.28 when only the 15,627 differentially methylatedsites (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 bySNPs (37) from this analysis, heritability values were 0.18and 0.28, respectively. The influence of shared environmen-tal factors on the methylation pattern was low (c2 = 0.07for all sites; c2 = 0.08 for 15,627 differentially methylatedsites). This is also in agreement with the findings ofGrundberg et al. (38).

Associations Between DNA Methylation andExpressionEpigenetic modifications have been associated with tran-scriptional regulation (8,16,39). Therefore, we relatedmRNA expression with DNA methylation for the 197 dif-ferentially expressed genes in Supplementary Table 3.DNA methylation of 266 sites, corresponding to 103genes, were significantly associated with expression inthe discordant twins at q , 0.15 (Supplementary Table18). Additionally, among the genes with largest expres-sion differences between diabetic and nondiabetic co-twins (Table 2), there were significant associations betweenmethylation and expression for 20 sites corresponding to11 genes (Supplementary Table 19). Among differentiallyexpressed obesity and T2D candidate genes (Supplemen-tary Tables 4 and 5), there were significant associationsbetween methylation and expression for 65 sites corre-sponding to 17 genes (Supplementary Table 20). Represen-tative associations between methylation and expression areshown in Fig. 3E and F. DNA methylation is known toregulate expression in different ways, depending on thelocation of CpG sites (39). Hence, we examined the genomiclocation of CpG sites where methylation correlated withexpression (Fig. 3G–J). Methylation sites showing inverserelationships with expression were over-represented inpromoter regions, whereas methylation sites showing pos-itive relationships with expression were over-represented ingene bodies and 39 UTRs (Fig. 3G and H), which is inagreement with the current literature (39). We next exam-ined whether the expression of genes that may contributeto the regulation of DNA methylation was altered in thediscordant twins. While TET1 expression was decreased inpatients with diabetes (q , 0.15), numerous other candi-date genes were only nominally regulated (SupplementaryTable 21).

Replication of Methylation DataWe finally tested whether we could replicate findings froma previous study (13) in which methylation was analyzedwith the Illumina 27k array. After correction for multipletesting, Ribel-Madsen et al. (13) identified two CpG siteswith methylation differences of .3% in adipose tissue of

five twin pairs discordant for T2D. These sites were alsodifferentially methylated in the discordant twins includedin our study (Supplementary Fig. 2).

DISCUSSION

Despite the awareness of lifestyle factors in modulatingthe risk of T2D, the molecular mechanisms behind suchacquired effects remain largely unknown. In this study, wecapitalized on the strengths of a twin study design topresent for the first time both genome-wide mRNA andDNA methylation profiles in adipose tissue from MZ twinpairs discordant for T2D. There was a significant enrich-ment of gene expression disruption to pathways relevantin oxidative phosphorylation; and carbohydrate, aminoacid, and lipid metabolism in diabetic co-twins. In addition to decreased expression of numerous genes withkey functions in mitochondria, diabetic twins also dis-played reduced mtDNA content. Disturbed mitochon-drial metabolism in adipose tissue has been suggestedto shift lipid storage into other tissues, such as skeletalmuscle, liver, and pancreas, resulting in severe insulinresistance (40).

Metabolic disorders related to obesity-associated in-sulin resistance have been characterized by an increasedinflux of inflammatory cells into adipose tissue (41). Here,we found increased expression of four gene sets relatedto the immune system in adipose tissue from diabetictwins. The most over-expressed gene in adipose tissueof diabetic co-twins was SPP1 encoding the inflammatorycytokine osteopontin, which recruits macrophages intoadipose tissue and stimulates T-cell proliferation duringinflammation (34). Mice lacking osteopontin are pro-tected against developing insulin resistance despite diet-induced obesity (42). Recent human studies (43) suggesta link between gastric inhibitory polypeptide and osteo-pontin in adipose tissue and insulin resistance. Interest-ingly, SPP1 was over-expressed in adipose tissue of theobese co-twin in healthy MZ twin pairs discordant forBMI (44), suggesting that this defect is present beforethe development of overt T2D. This was further sup-ported by our biological validation experiments whereSPP1 expression was increased in adipose tissue fromobese versus nonobese NGT individuals.

Furthermore, GYS2, ELOVL6, and FADS1 were themost downregulated genes in adipose tissue from diabe-tic twins. Our follow-up analysis in an independent case-control cohort confirmed the decreased expression of thesegenes in patients with T2D. GYS2 encodes the rate-limitingenzyme 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 glycerol3-phosphate, which is required for esterification (or re-esterification) of fatty acids into triglycerides. ELOVL6encodes a key enzyme involved in the elongation oflong-chain fatty acids (31), and FADS1 encodes an enzymethat introduces double bonds into growing fatty acidchains (30). Interestingly, recent GWASs (45–47) have












identified SNPs near FADS1 that are associated with con-ditions of altered metabolism, including f-glucose, dys-lipidemia, and circulating sphingolipid levels. Our findingsof decreased GYS2, ELOVL6, and FADS1 expression in ad-ipose tissue from patients with diabetes could potentially

explain the reduced glucose uptake and impaired ability tostore lipids in the adipose tissue of these individuals.

Unsupervised clustering proposed a large geneticcontribution to the methylation variability as the affectedtwin from the pair discordant for T2D was epigenetically

Figure 3—Associations between DNA methylation and gene expression in human adipose tissue. S100A4 mRNA expression (A) and DNAmethylation (B) differ significantly between diabetic compared with nondiabetic subjects both in the discordant twins and in case-controlcohort 2. SLC37A2 mRNA expression (C) and DNA methylation (D) differ significantly between diabetic compared with nondiabeticsubjects both in the discordant twins and in case-control cohort 2. Data are shown as mean6 SD. E: mRNA expression of CTSZ correlatesnegatively with DNA methylation of cg18563860 located in TSS1500 of the gene, in adipose tissue from T2D discordant twins. F: mRNAexpression of CHPT1 correlates positively with DNA methylation of cg15068132 located in gene body, in adipose tissue from T2Ddiscordant twins. *P < 0.05 compared with nondiabetic/NGT. a.u., arbitrary units. Distribution of individual CpG sites located in or neargenes that exhibit differential expression in T2D discordant twins and that show an inverse (G) or positive (H) relationship between DNAmethylation and mRNA expression in the discordant twins in relation to their functional genome distribution. Distribution of individual CpGsites located in or near genes that exhibit differential expression in T2D discordant twins, and that show an inverse (I) or positive (J)relationship between DNA methylation and mRNA expression in the discordant twins in relation to CpG island regions. N, northern; S,southern. The distribution of the significant sites compared with the distribution of all analyzed sites on the Infinium HumanMethylation450BeadChip is significantly different (P value) from what is expected by chance based on x2 tests (Pchi2).


“closer” to his/her unaffected co-twin than to the otherdiabetic twins. In further support of a large genetic contri-bution, the methylation data correlated strongest withinhealthy MZ twin pairs compared with DZ twins andunrelated individuals. Our heritability estimates also sup-port genetic effects. This is in line with a previous study(48) in which we showed that T2D SNPs affect DNAmethylation in human pancreatic islets. Others havealso proposed strong genetic effects on the DNA methyl-ation pattern (13,38,49,50). However, 15,627 CpG sitesin/near ;30% of all genes exhibited differential methyla-tion in adipose tissue from a case-control cohort of un-related individuals, supporting a key role for epigeneticmodifications in T2D patients. Moreover, 1,410 of these15,627 sites were also differentially methylated in thediscordant twins and ;50% of the differentially expressedgenes in discordant twins were associated with DNAmethylation. These results support the idea that DNAmethylation affects the phenotype in adipose tissue fromdiscordant twins too. Additional epigenetic mechanismssuch as histone modifications or microRNAs may also playa role in the twins. However, this remains to be investi-gated in future studies. The identification of CNVs in thediscordant twins provides an additional potential expla-nation for phenotypic variation within MZ twin pairs.

Protein glycosylation is a common post-translationalmodification. Seven significantly upregulated gene sets indiabetic twins belong to glycan biosynthesis and metab-olism. Interestingly, we also found enrichment of differ-entially methylated genes related to glycan metabolismin the pathway analysis in adipose tissue from unrelatedpatients with diabetes versus nondiabetic subjects. N-glycosylation is important for efficient trafficking ofGLUT4 to its proper compartments in adipocytes (51), sug-gesting one mechanism by which dysregulation of thesegenes could contribute to T2D.

MZ twin pairs discordant for T2D are rare, and ourrelatively small sample size increases the risk of type IIerrors. More genes may have been identified in a largercohort. However, we were able to validate eight of eight ofour most differentially expressed genes in unrelated case-control cohorts, suggesting that our results represent truefindings with biological relevance in people other than twins.It is difficult to draw conclusions about causality for differ-entially methylated and expressed genes in case-controlcohorts. The ideal study design would be to longitudinallyassess methylation and expression changes in adipose tissueduring an individual’s transition into disease. However,genes analyzed in our replication cohort already had differ-ential expression in obese versus nonobese NGT subjects.Additionally, BMI and/or glucose levels correlated with ex-pression and/or methylation in nondiabetic subjects. Obe-sity increases the risk for T2D, and, hence, it is possible thataltered expression and/or methylation of these genes maycontribute to the development of T2D.

In conclusion, our study adds to the understanding ofmolecular mechanisms that contribute to T2D. Decreasedexpression of genes involved in energy metabolism andincreased expression of inflammatory genes in adiposetissue accompany T2D. Importantly, our expressionresults from twins discordant for T2D could be replicatedin case-control cohorts of unrelated subjects, showing theimportance of our findings to the general population.Finally, our study highlights the importance of epigeneticsin T2D as adipose tissue from unrelated subjects with thedisease exhibit genome-wide methylation changes.

Acknowledgments. The authors thank Ylva Wessman and Targ Elgzyri(Scania University Hospital, Malmö, Sweden) for skilled technical assistance andcollection of clinical material. The authors also thank Maria Sterner (Departmentof Clinical Sciences, Lund University, Malmö, Sweden) for analysis of genotype

Figure 4—Genetic variation and DNA methylation in human adipose tissue. Pairwise Spearman correlations of DNA methylation levelsbased on all analyzed CpG sites on the array (A) or restricted to 15,627 individual CpG sites with q < 0.15 when comparing patients withdiabetes and nondiabetic subjects in case-control cohort 2 (B) in 10 nondiabetic MZ twin pairs, 10 nondiabetic same-sex DZ twin pairs, and10 pairs of nondiabetic unrelated individuals matched for age and sex.


data. The authors also thank the Swegene Center for Integrative Biology at LundUniversity for analyzing global mRNA expression and DNA methylation.Funding. This work was supported by grants from the Swedish ResearchCouncil (Dnr 521-2010-2745 and Dnr 523-2010-1061), the Swedish MedicalResearch Council (K2008-55X-15358-04-3), the Danish Council for StrategicResearch, 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 equip-ment grants from the Knut and Alice Wallenberg Foundation (2009-0243), theLundberg Foundation (grant no. 359), Avtal om Läkarutbildning och Forskning,the Novo Nordisk Foundation, the Tore Nilsson Foundation, the Syskonen Svens-son Foundation, the Diabetes Foundation, Kungliga Fysiografiska Sällskapet iLund, the European Foundation for the Study of Diabetes/Lilly Foundation, theSöderberg Foundation, and the Påhlsson Foundation. The group of Swedish twinswas recruited from the Swedish Twin Registry, which is supported by grants fromthe Swedish Department of Higher Education and the Swedish Research Council.The Centre of Inflammation and Metabolism is supported by a grant from theDanish National Research Foundation (DNRF55), and the Centre for PhysicalActivity Research is supported by a grant from TrygFonden.Duality of Interest. M.K.S. is currently employed by Amgen AB. No otherpotential conflicts of interest relevant to this article were reported.Author Contributions. E.N. researched the data and wrote the manu-script. P.A.J., M.P., M.K.S., P.P., B.K.P., and C.S. collected the clinical data andreviewed the manuscript. A.P., P.V., P.A., J.F., and T.R. analyzed the dataand reviewed the manuscript. R.R.-M. and N.L.P. contributed to the study designand reviewed the manuscript. A.V. designed the study, collected the clinicaldata, and reviewed the manuscript. C.L. designed the study and wrote andreviewed the manuscript. E.N. and C.L. are guarantors of this work and, assuch, had full access to all the data in the study and take responsibility forthe integrity of the data and the accuracy of the data analysis.

References1. Temelkova-Kurktschiev T, Stefanov T. Lifestyle and genetics in obesity andtype 2 diabetes. Exp Clin Endocrinol Diabetes 2012;120:1–62. Eriksson JW, Smith U, Waagstein F, Wysocki M, Jansson PA. Glucoseturnover and adipose tissue lipolysis are insulin-resistant in healthy relatives oftype 2 diabetes patients: is cellular insulin resistance a secondary phenomenon?Diabetes 1999;48:1572–15783. Hardy OT, Czech MP, Corvera S. What causes the insulin resistance un-derlying obesity? Curr Opin Endocrinol Diabetes Obes 2012;19:81–874. Ling C, Groop L. Epigenetics: a molecular link between environmentalfactors and type 2 diabetes. Diabetes 2009;58:2718–27255. Fraga MF, Ballestar E, Paz MF, et al. Epigenetic differences arise duringthe lifetime of monozygotic twins. Proc Natl Acad Sci U S A 2005;102:10604–106096. Rönn T, Poulsen P, Hansson O, et al. Age influences DNA methylation andgene expression of COX7A1 in human skeletal muscle. Diabetologia 2008;51:1159–11687. Jacobsen SC, Brøns C, Bork-Jensen J, et al. Effects of short-term high-fatoverfeeding on genome-wide DNA methylation in the skeletal muscle of healthyyoung men. Diabetologia 2012;55:3341–33498. Nitert MD, Dayeh T, Volkov P, et al. Impact of an exercise intervention onDNA methylation in skeletal muscle from first-degree relatives of patients withtype 2 diabetes. Diabetes 2012;61:3322–33329. Rönn T, Volkov P, Davegårdh C, et al. A six months exercise interventioninfluences the genome-wide DNA methylation pattern in human adipose tissue.PLoS Genet 2013;9:e100357210. Jacobsen SC, Gillberg L, Bork-Jensen J, et al. Young men with low birth-weight exhibit decreased plasticity of genome-wide muscle DNA methylation byhigh-fat overfeeding. Diabetologia. 26 February 2014 [Epub ahead of print]11. Barrès R, Osler ME, Yan J, et al. Non-CpG methylation of the PGC-1alpha pro-moter through DNMT3B controls mitochondrial density. Cell Metab 2009;10:189–198

12. Hall E, Dayeh T, Kirkpatrick CL, Wollheim CB, Dekker Nitert M, Ling C. DNAmethylation of the glucagon-like peptide 1 receptor (GLP1R) in human pancreaticislets. BMC Med Genet 2013;14:7613. Ribel-Madsen R, Fraga MF, Jacobsen S, et al. Genome-wide analysis ofDNA methylation differences in muscle and fat from monozygotic twins discor-dant for type 2 diabetes. PLoS One 2012;7:e5130214. Volkmar M, Dedeurwaerder S, Cunha DA, et al. DNA methylation profilingidentifies epigenetic dysregulation in pancreatic islets from type 2 diabetic pa-tients. EMBO J 2012;31:1405–142615. Yang BT, Dayeh TA, Kirkpatrick CL, et al. Insulin promoter DNA methylation

correlates negatively with insulin gene expression and positively with HbA(1c)levels in human pancreatic islets. Diabetologia 2011;54:360–36716. Yang BT, Dayeh TA, Volkov PA, et al. Increased DNA methylation and de-

creased expression of PDX-1 in pancreatic islets from patients with type 2 di-abetes. Mol Endocrinol 2012;26:1203–121217. Dayeh T, Volkov P, Salö S, et al. Genome-wide DNA methylation analysis of

human pancreatic islets from type 2 diabetic and non-diabetic donors identifiescandidate genes that influence insulin secretion. PLoS Genet 2014;10:e100416018. Lichtenstein P, Sullivan PF, Cnattingius S, et al. The Swedish Twin Registry

in the third millennium: an update. Twin Res Hum Genet 2006;9:875–88219. Pedersen M, Pedersen KK, Bruunsgaard H, et al. Cognitive functions inmiddle aged individuals are related to metabolic disturbances and aerobic ca-

pacity: a cross-sectional study. PLoS One 2012;7:e5113220. Poulsen P, Grunnet LG, Pilgaard K, et al. Increased risk of type 2 diabetes inelderly twins. Diabetes 2009;58:1350–135521. Ribel-Madsen R, Friedrichsen M, Vaag A, Poulsen P. Retinol-binding protein

4 in twins: regulatory mechanisms and impact of circulating and tissue ex-pression levels on insulin secretion and action. Diabetes 2009;58:54–6022. Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison of normali-

zation methods for high density oligonucleotide array data based on variance andbias. Bioinformatics 2003;19:185–19323. Subramanian A, Tamayo P, Mootha VK, et al. Gene set enrichment analysis:

a knowledge-based approach for interpreting genome-wide expression profiles.Proc Natl Acad Sci U S A 2005;102:15545–1555024. Bibikova M, Barnes B, Tsan C, et al. High density DNA methylation array

with single CpG site resolution. Genomics 2011;98:288–29525. Andersen CL, Jensen JL, Ørntoft TF. Normalization of real-time quantitativereverse transcription-PCR data: a model-based variance estimation approach to

identify genes suited for normalization, applied to bladder and colon cancer datasets. Cancer Res 2004;64:5245–525026. Wang K, Li M, Hadley D, et al. PennCNV: an integrated hidden Markov

model designed for high-resolution copy number variation detection in whole-genome SNP genotyping data. Genome Res 2007;17:1665–167427. Kohler HP, Rodgers JL. DF-analyses of heritability with double-entry twin data:asymptotic standard errors and efficient estimation. Behav Genet 2001;31:179–19128. Zambelli F, Pesole G, Pavesi G. Pscan: finding over-represented tran-scription factor binding site motifs in sequences from co-regulated or co-expressed genes. Nucleic Acids Res 2009;37(Web Server issue):W247–W25229. Mathelier A, Zhao X, Zhang AW, et al. JASPAR 2014: an extensively ex-panded and updated open-access database of transcription factor binding pro-files. Nucleic Acids Res 2014;42(Database issue):D142–D14730. Marquardt A, Stöhr H, White K, Weber BH. cDNA cloning, genomic structure,and chromosomal localization of three members of the human fatty acid de-saturase family. Genomics 2000;66:175–18331. Matsuzaka T, Shimano H. Elovl6: a new player in fatty acid metabolism andinsulin sensitivity. J Mol Med (Berl) 2009;87:379–38432. Mandard S, Stienstra R, Escher P, et al. Glycogen synthase 2 is a novel

target gene of peroxisome proliferator-activated receptors. Cell Mol Life Sci 2007;64:1145–115733. Perrier S, Darakhshan F, Hajduch E. IL-1 receptor antagonist in metabolicdiseases: Dr Jekyll or Mr Hyde? FEBS Lett 2006;580:6289–6294


34. Mazzali M, Kipari T, Ophascharoensuk V, Wesson JA, Johnson R, Hughes J.Osteopontin—a molecule for all seasons. QJM 2002;95:3–1335. Schutyser E, Richmond A, Van Damme J. Involvement of CC chemokine ligand18 (CCL18) in normal and pathological processes. J Leukoc Biol 2005;78:14–2636. Bruder CE, Piotrowski A, Gijsbers AA, et al. Phenotypically concordant anddiscordant monozygotic twins display different DNA copy-number-variationprofiles. Am J Hum Genet 2008;82:763–77137. Chen YA, Lemire M, Choufani S, et al. Discovery of cross-reactive probesand polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray.Epigenetics 2013;8:203–20938. Grundberg E, Meduri E, Sandling JK, et al.; Multiple Tissue Human Ex-pression Resource Consortium. Global analysis of DNA methylation variation inadipose tissue from twins reveals links to disease-associated variants in distalregulatory elements. Am J Hum Genet 2013;93:876–89039. Jones PA. Functions of DNA methylation: islands, start sites, gene bodiesand beyond. Nat Rev Genet 2012;13:484–49240. Heilbronn L, Smith SR, Ravussin E. Failure of fat cell proliferation, mito-chondrial function and fat oxidation results in ectopic fat storage, insulin re-sistance and type II diabetes mellitus. Int J Obes Relat Metab Disord 2004;28(Suppl. 4):S12–S2141. Hotamisligil GS. Inflammation and metabolic disorders. Nature 2006;444:860–86742. Nomiyama T, Perez-Tilve D, Ogawa D, et al. Osteopontin mediates obesity-induced adipose tissue macrophage infiltration and insulin resistance in mice.J Clin Invest 2007;117:2877–288843. Ahlqvist E, Osmark P, Kuulasmaa T, et al. Link between GIP and osteopontinin adipose tissue and insulin resistance. Diabetes 2013;62:2088–2094

44. Pietiläinen KH, Naukkarinen J, Rissanen A, et al. Global transcript profiles offat in monozygotic twins discordant for BMI: pathways behind acquired obesity.PLoS Med 2008;5:e5145. Dupuis J, Langenberg C, Prokopenko I, et al.; DIAGRAM Consortium; GIANTConsortium; Global BPgen Consortium; Anders Hamsten on behalf of ProcardisConsortium; MAGIC investigators. New genetic loci implicated in fasting glucosehomeostasis and their impact on type 2 diabetes risk. Nat Genet 2010;42:105–11646. Lemaitre RN, Tanaka T, Tang W, et al. Genetic loci associated with plasmaphospholipid n-3 fatty acids: a meta-analysis of genome-wide association studiesfrom the CHARGE Consortium. PLoS Genet 2011;7:e100219347. Wu JH, Lemaitre RN, Manichaikul A, et al. Genome-wide association studyidentifies novel loci associated with concentrations of four plasma phospholipidfatty acids in the de novo lipogenesis pathway: results from the Cohorts for Heartand Aging Research in Genomic Epidemiology (CHARGE) consortium. Circ Car-diovasc Genet 2013;6:171–18348. Dayeh TA, Olsson AH, Volkov P, Almgren P, Rönn T, Ling C. Identificationof CpG-SNPs associated with type 2 diabetes and differential DNA methylationin human pancreatic islets. Diabetologia 2013;56:1036–104649. Kaminsky ZA, Tang T, Wang SC, et al. DNA methylation profiles in mono-zygotic and dizygotic twins. Nat Genet 2009;41:240–24550. Ollikainen M, Smith KR, Joo EJ, et al. DNA methylation analysis of multipletissues from newborn twins reveals both genetic and intrauterine components tovariation in the human neonatal epigenome. Hum Mol Genet 2010;19:4176–418851. Zaarour N, Berenguer M, Le Marchand-Brustel Y, Govers R. Deciphering therole of GLUT4 N-glycosylation in adipocyte and muscle cell models. Biochem J2012;445:265–273


altered dna methylation and differential expressionofgenes...

Documents