Differential Epigenomic and Transcriptomic Responses in Subcutaneous Adipose Tissue Between Low and High Responders to Caloric Restriction1–3

Differential epigenomic and transcriptomic responses in subcutaneous adipose tissue between low and high responders to caloric restriction1–3

Luigi Bouchard, Re´mi Rabasa-Lhoret, May Faraj, Marie-E`ve Lavoie, Jonathan Mill, Louis Pe´russe, and Marie-Claude Vohl

ABSTRACT weight loss responses to caloric restriction show considerable Background: Caloric restriction is recommended for the treatment interindividual variability (7). Studies of genetically identical of obesity, but it is generally characterized by large interindividual monozygotic twins have been particularly useful in disentangling variability in responses. The factors affecting the magnitude of the role of environmental and heritable factors in determining the weight loss remain poorly understood. Epigenetic factors (ie, heri- degree of weight loss. It has been shown that within-pair changes table but reversible changes to genomic function that regulate gene in body fat variability after a caloric deficit is significantly lower expression independently of DNA sequence) may explain some of than between-pair variability, which suggests that genetic factors the interindividual variability seen in weight-loss responses. have an important influence on an individual’s response to caloric Objective: The objective was to determine whether epigenetics and deficit (8, 9). However, the concordance between twin pairs was gene expression changes may play a role in weight-loss responsiveness. not complete, which suggests that environmental factors or other Design: Overweight/obese postmenopausal women were recruited DNA sequence–independent mechanisms may be involved. for a standard 6-mo caloric restriction intervention. Abdominal sub- It has been suggested that monozygotic twin discordance for cutaneous adipose tissue biopsy samples were collected before (n = complex traits such as body weight could be accounted for by 14) and after (n = 14) intervention, and the epigenomic and tran- epigenetic factors (10, 11). Epigenetics refers to the heritable, but scriptomic profiles of the high and low responders to dieting, on the reversible, regulation of various genomic functions, including basis of changes in percentage body fat, were compared by using gene transcription, that are mediated principally through changes microarray analysis. in DNA methylation and chromatin structure (12). The epigenetic Results: Significant DNA methylation differences at 35 loci were regulation of cellular functions is a normal and essential process found between the high and low responders before dieting, with 3 in cell development and differentiation, and epigenetic factors are regions showing differential methylation after intervention. Some of subjected to reprogramming in response to both stochastic and these regions contained genes known to be involved in weight con- environmental stimuli (12). Such changes can be mitotically trol and insulin secretion, whereas others were localized in known imprinted genomic regions. Differences in gene expression profiles were observed only after dieting, with 644 genes being differen- 1 From the Nutraceuticals and Functional Foods Institute (LB, LP, and M- tially expressed between the 2 groups. These included genes likely CV), the Department of Preventive Medicine (LP), and the Department of to be involved in metabolic pathways related to angiogenesis and Food Science and Nutrition (M-CV), Universite´ Laval, Laval, Canada; the cerebellar long-term depression. Department of Medicine, Universite´ de Montreál, Chicoutimi Hospital, Sa- Conclusions: These data show that both DNA methylation and gene guenay, Canada (LB); the Department of Nutrition (RR-L, MF, and M-EL), expression are responsive to caloric restriction and provide new the Montreal Diabetes Research Centre (RR-L and MF), Universite´ de Mon- treál, Montreal, Canada; and the Institute of Psychiatry, SGDP Research insights about the molecular pathways involved in body weight loss Centre, King’s College London, London, United Kingdom (JM). as well as methylation regulation during adulthood. Am J Clin 2 Supported by the Canadian Institute of Health Research through the Nutr 2010;91:309–20. MONET project (Montreal-Ottawa New Emerging Team; OHN-63279 and MOP62976) and the Quebec New Emerging Team (OHN 63276). LB was INTRODUCTION funded by the Laval University Merck Frosst/Canadian Institute of Health Research Chair in Obesity and the Heart and Stroke Foundation of Canada/ In 2005, 1.1 billion adults and 10% of children were over- Sanofi-Aventis research fellowship awards. RRL was supported by the Fonds weight or obese worldwide (1). Obesity is defined as an excessive de la Recherche en Sante´ du Que´bec and held the chair for clinical research accumulation of fat resulting from a long-term imbalance be- J-A de Se`ve at IRCM (Montreal Institute for Clinical Research). MF was the tween energy intake and expenditure (2). It has a dramatic effect recipient of the Canadian Institute of Health Research New Investigator on an individual’s health, with musculoskeletal, pulmonary, and Award. M-EL was supported by a scholarship from the Fonds de la Re- ´ ´ psychosocial-related problems, and is associated with an in- cherche en Sante du Quebec. 3 Address correspondence to M-C Vohl, Lipid Research Center, 2705 creased risk of morbidity and mortality attributable to cardio- Laurier Boulevard, (TR93), Que´bec City, PQ, Canada G1V 4G2. E-mail: vascular diseases, diabetes, and certain forms of cancer (1, 3). [email protected]. Interestingly, only a moderate loss of initial body weight Received May 18, 2009. Accepted for publication November 2, 2009. provides significant metabolic improvements (4–6). However, First published online November 25, 2009; doi: 10.3945/ajcn.2009.28085.

Am J Clin Nutr 2010;91:309–20. Printed in USA. Ó 2010 American Society for Nutrition 309 310 BOUCHARD ET AL stable and enduring, producing long-term changes to gene ex- adipose tissue biopsy collections were preceded by a 4-wk pression, but can also be short-lived and rapidly reversed (13). weight-stability period (within 62 kg), verified on a weekly Indeed, such dynamic epigenetic changes have the potential to basis at our research unit. Also, the subjects were instructed not offer a mechanism by which cellular metabolism can be rapidly to exercise and to eat a high-carbohydrate diet for the 3 d before regulated independently of long-term, irreversible evolutionary the biopsy procedure. Fasting baseline and postintervention mutagenesis. Cytosine methylation (Cmet), occurring at position subcutaneous adipose tissue biopsy samples were obtained from 5 of the cytosine pyrimidine ring in CpG dinucleotides, is the the periumbilical level at both sides of the body by needle bi- best understood epigenetic modification. The methylation of opsy under local anesthesia (20 mg xylocaine/mL) (18–20). CpG sites disrupts the binding of transcription factors and at- One hundred thirty-seven women were recruited and com- tracts methyl-binding proteins that are associated with gene si- pleted the 6-mo weight-loss program. Baseline and post- lencing and chromatin compaction (14). Because CpG intervention adipose tissue biopsy samples were available for 29 dinucleotide methylation is associated with the regulation of of these women. Consent for biopsy was an optional part of the gene expression, altered DNA methylation could explain in- larger study. There was no other criteria to be met to be included terindividual phenotypic differences (15). in the biopsy subsample. Fourteen women were further selected We report here the results of the first comprehensive analysis of for the current study based on their response to the caloric re- epigenomic and transcriptomic responses in subcutaneous adi- striction. Women who lost 3% of their body fat were consid- pose tissue after a caloric restriction intervention in overweight ered “high responders,” whereas those who lost ,3% of their and obese women. We found that although both DNA methyl- body fat were considered “low responders” to the caloric re- ation and gene expression differences existed between the high striction. Low (n = 7) and high (n = 7) responders were matched and low responders to dietary restriction after the intervention, for baseline age, BMI, percentage body fat, resting blood only epigenetic differences exist before dieting. These data pressure, fasting blood lipids, glucose and insulin concen- suggest that the epigenetic profile has the potential to differentiate trations, and changes in fat-free mass (Tables 1 and 2). between good and poor responders to caloric restriction. Anthropometric and metabolic measurements SUBJECTS AND METHODS Standardized procedures were used to measure body weight, height, waist girth, resting blood pressure, and blood lipid, Study population and experimental design glucose, and insulin concentrations, as previously described (16, The sample used in this study was a subsample of a larger 17). Total energy expenditure was measured by using the doubly weight-loss study (n = 137) aimed at exploring the effect of labeled water technique (21), and resting metabolic rate was a caloric restriction intervention on body composition, energy measured by indirect calorimetry (16, 17). Total fat mass and fat- expenditure, insulin sensitivity, and metabolic, inflammatory, free mass were measured by dual-energy X-ray absorptiometry hormonal, and psychosocial profiles in overweight and obese (16, 17). postmenopausal women. The recruitment for the larger study began in May 2003. The substudy presented here explored the DNA and RNA extraction epigenomic and transcriptomic responses of 14 overweight and obese postmenopausal women to a caloric restriction in- Genomic DNA was extracted by using the Qiagen DNeasy tervention. All individuals provided written informed consent Blood and Tissue DNA purification kit (Qiagen, Mississauga, before their inclusion in the study, which was approved by the Canada). Total RNA was prepared from ’50 mg of subcutaneous Universite´ de Montreál ethics committee. adipose tissue biopsy samples and was concentrated with the The inclusion and exclusion criteria of the weight-loss study Qiagen RNeasy Lipid Tissue Minikit and Qiagen RNeasy were presented in detail previously (16, 17). Briefly, sedentary MinElute Cleanup Kit (Qiagen). The concentration (5–10 lg/50 mg overweight and obese postmenopausal women were included if fat tissue) and integrity of purified total RNA was verified by they had been weight stable for 3 mo before the study and were using an Agilent 2100 bioanalyzer with the RNA 6000 Nano not taking medications known to affect cardiovascular function LabChip kit (Agilent Technologies, Palo Alto, CA). and/or metabolism. The subjects were nonsmokers, had a low-to- , moderate alcohol consumption ( 2 drinks/d), and were free of Microarrays diabetes and, uncontrolled thyroid, inflammatory, cardiovascular, or pituitary diseases. Epigenomic profiling The aim of the medically supervised weight-loss program was Our epigenomic profiling of abdominal subcutaneous adipose to reduce body weight by 10% over 6 mo. To determine the levels tissue used Human CpG-island 15K arrays obtained from the of caloric restriction, 500–800 kcal were subtracted from the University Health Network (Toronto, Canada). This array con- baseline resting metabolic rate (determined by indirect calo- tains 14,923 CpG-island clones derived from the Human CpG- rimetry) and then multiplied by a physical activity factor of 1.4, island 8.1K array (22) and a set of 6800 additional CpG-island which corresponds to a sedentary state. Dietary prescriptions loci. Clones with internal repeat sequences, with multiple or ranged from 1100 to 1800 kcal/d. The macronutrient composition absent BLAST hits, mapped on mitochondrial chromosome or of the diets was standardized: 55%, 30%, and 15% of energy those with .20% overlapping sequences (duplicates) were intake from carbohydrates, fat, and protein, respectively (16). To excluded. Clones from Arabidopsis and the Stratagene’s reduce the acute effect of weight changes on the outcomes SpotReport Alien cDNA Array Validation System were also measured, baseline and postintervention measurements as well as spotted on the array and served as controls. MOLECULAR PATHWAYS INVOLVED IN BODY WEIGHT LOSS 311

TABLE 1 Baseline anthropometric and metabolic characteristics of the low and high responders1 Low responders (n = 7) High responders (n =7)

Mean 6 SD Range Mean 6 SD Range

Age (y) 57.71 6 4.41 53.04–64.24 57.80 6 5.30 52.80–68.57 BMI (kg/m2) 32.25 6 4.18 26.53–38.03 29.54 6 2.78 26.13–34.57 Fat mass (kg) 38.21 6 7.06 28.04–48.84 34.40 6 6.64 23.97–44.70 Body fat (%) 45.91 6 3.61 40.30–50.30 45.40 6 4.30 39.80–50.50 Fat-free mass (kg) 44.82 6 7.10 35.73–57.22 41.06 6 5.31 34.77–48.67 Fat-free mass (%) 54.09 6 3.61 49.70–59.70 54.60 6 4.30 49.50–60.20 Waist girth (cm) 105.96 6 15.36 90.50–136.00 97.04 6 7.87 86.50–105.75 DBP (mm Hg) 72.43 6 6.27 61.00–79.00 78.57 6 7.46 69.00–91.00 SBP (mm Hg) 115.43 6 16.22 95.00–146.00 126.71 6 19.73 104.00–157.00 Total cholesterol (mmol/L) 5.71 6 0.50 4.76–6.27 5.50 6 0.78 4.58–6.60 HDL cholesterol (mmol/L) 1.59 6 0.30 1.13–2.04 1.44 6 0.11 1.28–1.58 LDL cholesterol (mmol/L) 3.49 6 0.40 2.93–4.08 3.24 6 0.61 2.60–4.09 Triglycerides (mmol/L) 1.37 6 0.45 0.49–1.76 1.81 6 0.77 0.90–3.01 Glucose (mmol/L) 5.22 6 0.95 3.93–6.57 5.13 6 0.48 4.40–5.67 Insulin (mmol/L)2 14.30 6 5.14 5.98–21.55 13.32 6 5.83 8.57–21.29 Total energy expenditure (kcal/d) 2625 6 511 2057–3301 2317 6 337 1832–2760 1 DBP, diastolic blood pressure; SBP, systolic blood pressure. None of the differences between groups were statistically signiﬁcant, P , 0.05 (unpaired t test). 2 For the high responders, only 6 samples were available for fasting insulin measurements.

The unmethylated fraction of genomic DNA was enriched for chain reaction (PCR) with U-CG1B primer in conditions such epigenomic profiling, as previously reported (23, 24). Briefly, 500 ng that short fragments (,1.5 kb), and thus unmethylated DNA, genomic DNA was digested with the restriction enzyme HpaII was preferentially amplified. For each PCR reaction, 80 ng di- (New England Biolabs, Ipswich, MA). The DNA fragments gested genomic DNA were added to a reaction mixture (final were then ligated with DNA adaptors (annealing products of volume = 100 lL) containing 6.25 units Taq DNA polymerase U-CG1A 5#-CGTGGAGACTGACTACCAGAT-3# and UCG1B (New England Biolabs), 1 · thermopol buffer (New England 5#-AGTTACATCTGGTAGTCAGTCTCCA-3#)followedbyfurther Biolabs), 2.5 mmol MgCl2/L, 125 lmol of each dNTP and 1.5 digestion with McrBC restriction enzyme (New England Biol- lmol U-CG1b primer. PCRs were run in triplicate as follow: abs). HpaII cleaves DNA only when the restriction site is un- initial cycle at 72°C for 5 min and 95°C for 1 min, 25 cycles at methylated, whereas McrBC cleaves methylated fragments only. 95°C for 40 s, 68°C for 2 min, 72°C for 20 s, and a final ex- Therefore, both methylation-sensitive restriction enzymes con- tension at 72°C for 5 min. The pooled triplicates provided tributed to enrich the unmethylated fraction of genomic DNA. ’10 lg enriched DNA for labeling and hybridization. We com- The remaining fragments were then amplified by polymerase bined 2.5 lg of each DNA sample to produce a common

TABLE 2 Anthropometric and metabolic changes after the caloric restriction intervention1 Low responders High responders

Mean 6 SD Range Mean 6 SD Range

Change in BMI (kg/m2) 20.91 6 0.92 22.03 to 0.74 22.78 6 1.512 25.32 to 21.15 Change in fat mass (kg) 21.81 6 1.84 23.69 to 0.77 27.34 6 3.053 212.98 to 23.87 Change in percentage body fat (%) 21.07 6 1.19 22.70 to 1.00 26.10 6 2.864 210.90 to 23.00 Change in fat-free mass (kg) 20.55 6 1.07 21.95 to 1.09 0.26 6 2.27 23.21 to 3.49 Change in percentage fat-free mass (%) 1.07 6 1.19 21.00 to 2.70 6.10 6 2.864 3.00 to 10.90 Change in waist girth (cm) 24.46 6 3.44 28.25 to 2.00 27.79 6 5.15 217.00 to 21.50 Change in DBP (mm Hg) 0.29 6 7.54 216.00 to 6.00 24.86 6 8.63 222.00 to 4.00 Change in SBP (mm Hg) 4.71 6 9.07 27.00 to 17.00 25.57 6 9.11 220.00 to 5.00 Change in total cholesterol (mmol/L) 20.07 6 0.52 21.08 to 0.31 20.12 6 0.58 21.21 to 0.41 Change in HDL cholesterol (mmol/L) 20.10 6 0.22 20.30 to 0.34 20.01 6 0.10 20.21 to 0.09 Change in LDL cholesterol (mmol/L) 20.03 6 0.55 20.93 to 0.48 0.03 6 0.69 20.98 to 0.98 Change in triglycerides (mmol/L) 0.13 6 0.41 20.28 to 0.95 20.29 6 0.62 21.14 to 0.68 Change in glucose (mmol/L) 0.03 6 0.56 21.10 to 0.53 20.16 6 0.32 20.50 to 0.43 Change in insulin (mmol/L)5 21.89 6 2.79 25.45 to 1.29 21.93 6 4.59 29.53 to 4.00 1 DBP, diastolic blood pressure; SBP, systolic blood pressure. 2–4 Significantly different from low responders (unpaired t test): 2P 0.05, 3P 0.01, 4P 0.00. 5 For the high responders, only 6 samples were available for fasting insulin measurements. 312 BOUCHARD ET AL reference pool that was used to standardize between hybrid- Expression values were normalized by using the Robust Multi- izations. All samples were labeled with amino-allyl-dUTP, cou- array Average algorithm (26), as implemented in FlexArray pled to alexa dyes (alexa 647 for experimental samples and alexa software (version 1.1) (27). 555 for the reference pool), following the Bioprime Plus Array CGH Genomic labeling system protocol. The labeled probes were Affymetrix HG U133 plus 2.0 GeneChip microarray results then purified and hybridized (via incubation for 16 h at 42°C) on validation Human CpG-island 15K arrays by using the Advalytix Slide- Complementary DNA (cDNA) was generated from 40 ng total Booster (Advalytix, Munich, Germany) and AdvaHybc hybrid- RNA by using a random primer hexamer following the manu- ization solution. The arrays were scanned on an Agilent facturer’s protocol for Superscript II (Invitrogen, Carlsbad, CA). G2565BA scanner, quantified with ArrayVision v.8.0 (Imaging Primers and TaqMan probes overlapping the first and second Research), and the data were normalized with MIDAS software exons of each of the selected genes were obtained from Applied v.2.19 by using a subgrid intensity-based method. All 28 micro- Biosystems (Table 3). Each sample was analyzed in duplicate arrays passed initial quality control measures and were suitable and calibrated to LRP10 and GAPDH genes (endogenous con- for analysis. trol; LRP10: Hs00204094_m1; GAPDH: Hs99999905_m1). Relative quantification estimations were performed on an Ap- CpG-island 15K microarray results validation plied Biosystems 7500 Real Time PCR System as recom- The Sequenom EpiTYPER (Sequenom Inc, San Diego, CA) mended by the manufacturer (Applied Biosystems, Foster City, approach has been used to determine base-specific cytosine CA), and the DDCT calculation method was used to evaluate methylation levels of loci showing differential methylation levels the mean fold expression differences (MFED) between the 2 using microarray (25). Briefly, the EpiTYPER assay combines groups (28). DNA sodium bisulfite conversion chemistry, PCR amplification of target sequences, and base-specific cleavage. The cleavage products are quantitatively analyzed by MALDI-TOF mass Statistics spectrometry. The presence of methylated cytosine within the Sample characteristics original DNA sequence determines the cleavage pattern. The Baseline anthropometric and metabolic characteristics and EpiDESIGNER software has been used for PCR primer selection. changes in these characteristics in the low and high responders to The loci showing the strongest methylation differences in low- caloric restriction were tested by using an unpaired t test. Cor- and high-methylation groups (DNASE1L2 and TRIM14) were relations were tested by using Spearman rank correlation co- selected for validation as well as the 2 regions with 2 proximal efficients; t tests and correlations were performed by using SAS probe sets with methylation differences (chr7p15 and chr10q26). software, version 9.1.3 (SAS Institute, Cary, NC). The PCR primers were as follows: DNASE1L2-2f-GTTTAGT- AGTGTTTTGGGAGTTTGT and DNASE1L2-2r-CCTACCC- ACCACACCTATTAATCTA, DNASE1L2-6f-GGGTTTTTTTT- Microarray-data analysis ATTTTTTAGGAAAG and DNASE1L2-6r-ACCACTTAAA- Significance Analysis of Microarray (SAM) (29) was used to AACCTCACTACTCCTC; TRIM14-4f-TTTTTGGGGTATTTT- determine significant differences in DNA methylation (via CpG- TGGTTTTTA and TRIM14-4r-CTTCCCATTTCTAATAAAA- island arrays) and expression levels (via expression arrays) be- CCACCT, TRIM14-8f-ATGTTTGGGTTGGTTTTTTTAATTT and tween the low- and high-responder groups before and after TRIM14-8r-CCCATCATCAAAACTACAATTTTCT; HOXA6-3f- dieting. A modified t test was applied to each probe set (log2 TGGTTTTTAGAAGTTTTTGTTTTTTTG and HOXA6-3r-CCA- transformed), and the raw P values were converted to the false ATCTCCTACCTAAACTAAACCC ; OAT-3f-TGGAATTGGTTT- discovery rate (FDR, q values) to correct for multiple testing TATGTATAGGAGG and OAT-3r-AAAACACCAAATAACTCC- according to Benjamini and Hochberg (30). The same procedure CTACCTT, OAT-4f-TGGAATTGGTTTTATGTATAGGAGG and was applied to analyze the CpG-island validation results. For OAT-4r-CAACCAAATTAATAATCAAAACACCA. gene expression arrays, the significance threshold was set to detect at least a 1.20-fold change, with an FDR value 0.05. Transcriptomic profiling Because the number of differentially methylated genes identified was limited on the basis of the same criterion used for the Affymetrix HG U133 plus 2.0 GeneChip microarrays con- methylation experiment, a minimum fold-change of 1.15-fold taining .47,000 transcripts were used to determine the tran- and an FDR ,0.10 were used instead to establish significant scriptome of subcutaneous adipose tissue by using the standard differences between the groups. In Table 3, Table 4, and Table manufacturer’s protocol (Affymetrix). Briefly, RNA was reverse 5 and in supplementary Tables 1 and 2 (see “Supplemental data” transcribed into cDNA, and in vitro transcription was performed in the online issue), the low-responder group was the reference to generate biotin-labeled cRNA. Biotin-labeled cRNA was then for fold-change computation. hybridized to microarrays and stained with streptavidin phyco- erythrin. Arrays were scanned on a GeneArray scanner. One microarray (high-responder group, postcaloric restriction sam- Power estimates for CpG-island and expression microarray ple) was discarded because of a low signal intensity likely re- experiments sulting from poor RNA quality. In total, analyses were We assessed the power of the present study sample in SAM performed by using data from 27 microarrays (low-responder using a test based on the permutation analysis of real data (31). group: 7 arrays before and 7 after caloric restriction; high- This analysis estimates the SD of each gene and the overall null responder group: 6 arrays before and 7 after caloric restriction). distribution of the genes. The results provide accurate estimates MOLECULAR PATHWAYS INVOLVED IN BODY WEIGHT LOSS 313

TABLE 3 Signiﬁcant differentially methylated loci (by false discovery rate) between low and high responders to the caloric restriction intervention1 No. of samples with consistent Gene Gene Fold differences in the Gene symbol Probe set symbol (5#) symbol (3#) Localization change2 q Value HWL group

Before weight loss: hypomethylated (n =3) RAB3C UHNhscpg0019055 FLJ33641 PDE4D chr5q12 21.28 ,0.001 6 HYPK UHNhscpg0007880 SERINC4 MFAP1 chr15q15 21.23 ,0.001 6 DNASE1L2 UHNhscpg0028273 E4F1 DCI chr16p16 21.32 ,0.001 7 Before weight loss: hypermethylated (n = 32) KCNA3 UHNhscpg0027651 KCNA2 CD53 chr1p13 1.24 0.0602 7 LHX8 UHNhscpg0007328 C1orf171 SLC44A5 chr1p31 1.22 ,0.001 6 UHNhscpg0028140 AJ301580 FAF1 chr1p33 1.18 0.0802 6 UHNhscpg0022257 ZNF691 SLC2A1 chr1p34 1.17 0.0621 6 UHNhscpg0022498 MAP1LC3C PLD5 chr1q43 1.17 0.0845 6 RTKN UHNhscpg0024207 WDR54 ZNHIT4 chr2p13 1.17 0.0860 6 WDR5B UHNhscpg0024950 C3orf28 KPNA1 chr3q21 1.16 0.0802 6 CYTL1 UHNhscpg0025147 MSX1 STK32B chr4p16 1.17 0.0621 6 PRDM8 UHNhscpg0018526 ANTXR2 FGF5 chr4q21 1.40 0.0602 6 C6orf120 UHNhscpg0028283 THBS2 PHF10 chr6q27 1.18 0.0602 6 UHNhscpg0014466 HOXA6 HOXA7 chr7p15 1.24 ,0.001 6 UHNhscpg0013650 FKBP14 PLEKHA8 chr7p15 1.15 0.0602 6 CTTNBP2 UHNhscpg0004673 CFTR LSM8 chr7q31 1.21 0.0621 5 UHNhscpg0014893 KIAA1539 UNC13B chr9p13 1.16 0.0705 6 SLC1A1 UHNhscpg0025968 GLIS3 C9orf68 chr9p24 1.24 0.0602 6 UHNhscpg0026046 C9orf36 MGC21881 chr9q21 1.37 0.0892 5 TRIM14 UHNhscpg0026076 NANS CORO2A chr9q22 1.44 0.0507 6 UHNhscpg0017442 ZFP37 SLC31A2 chr9q32 1.40 0.0705 7 UHNhscpg0022618 GALNAC4S-6 OAT chr10q26 1.27 ,0.001 7 UHNhscpg0022695 SPRN BC038300 chr10q26 1.18 0.0602 6 TRIM3 UHNhscpg0022714 HPX ARFIP2 chr11p15 1.17 0.0507 6 CLPB UHNhscpg0026822 CLPB PDE2A chr11q13 1.22 0.0705 5 UHNhscpg0019513 KIRREL3 ETS1 chr11q24 1.18 0.0602 7 ITGA7 UHNhscpg0018627 METTL7B BLOC1S1 chr12q13 1.21 0.0892 5 AMDHD1 UHNhscpg0023031 CCDC38 HAL chr12q22 1.17 0.0802 7 BX161511 UHNhscpg0005471 RPS29 RPL36AL chr14q22 1.23 0.0892 5 UHNhscpg0023555 RASGRF1 TMED3 chr15q24 1.30 0.0892 4 NDRG4 UHNhscpg0027656 FLJ13912 NDRG4 chr16q21 1.32 0.0602 6 UHNhscpg0023806 FOXL1 FBXO31 chr16q24 1.38 0.0802 7 CYB5A UHNhscpg0027654 C18orf55 CR749350 chr18q23 1.23 0.0602 7 LYL1 UHNhscpg0004606 NFIX TRMT1 chr19p13 1.16 0.0621 5 UHNhscpg0018857 C20orf74 C20orf19 chr20p11 1.21 ,0.001 6 After weight loss: hypomethylated (n =0) None After weight loss: hypermethylated (n =3) PLCL4 UHNhscpg0022169 PEX10 PANK4 chr1p36 1.18 ,0.001 7 PRDM8 UHNhscpg0018526 ANTXR2 FGF5 chr4q21 1.38 ,0.001 6 UHNhscpg0025297 MGC13034 BTF3 chr5q13 1.38 ,0.001 6 1 Genes in the vicinity of those identified in the table [gene symbol, gene symbol (5#), and gene symbol (3#) columns] may also have been affected by the identified DNA methylation changes. HWL, high-weight-loss group. A modified t test was applied as implemented in Significance Analysis of Microarray to each probe set (log2 transformed), and the raw P values were converted to the false discovery rate (q values) to correct for multiple testing according to Benjamini and Hochberg (30). 2 The low responders were the reference group for fold-change computation. The change reflects the mean methylation levels computed before and after weight-loss treatment separately for each locus.

of the power (1-FDR) according to the number of samples. This As estimated by the method proposed by Tibshirani (31), the approach has the advantage that it is based on a real data set and power to detect differences in DNA methylation level was .90% takes into account gene correlation. The mean difference in DNA before caloric restriction and .99% after caloric restriction (14 methylation or mRNA levels used for power computation was arrays each). For the expression arrays, the power to identify 1.4-fold, which we consider to be a reasonable estimate based on genes with mean expression changes of 1.4-fold was the observed mean differences from the list of signiﬁcant genes .60% before dieting (13 arrays) and .80% after dieting (14 obtained from SAM. arrays). 314 BOUCHARD ET AL

TABLE 4 Bisulfite-treated DNA analyses of 4 probe set sequences identified by using CpG-island microarray1 No. of CpG CpG dinucleotide with MFED Sequenom Probe set (gene) MFED array (HpaII sites) tested significant MFED (FDR) R2 (P value)

UHNhscpg0028273 (DNASE1L2) 21.32 77 (3) CpG 15 21.71 (0) 0.71 (0.005) CpG 16 21.71 (0) 0.71 (0.005) CpG 17 21.71 (0) 0.71 (0.005) UHNhscpg0022618 (OAT) 1.27 21 (9) CpG 21 21.18 (0) 20.74 (0.003) UHNhscpg0026076 (TRIM14) 1.44 48 (3) None — — UHNhscpg0014466 (HOXA6) 1.24 25 (6) None — — 1 MFED, mean-fold expression difference; FDR, false discovery rate. 2 Spearman correlation coefﬁcient between CpG island microarray and Sequenom (Sequenom Inc, San Diego, CA) results.

Pathway analysis were relatively hypermethylated in the high-responder group. Pathway analyses allowed us to determine whether genes For 2 regions (chromosomes 7p15 and 10q26), supportive evi- found to be differentially expressed belong to predefined net- dence for differential methylation was provided by 2 adjacent , works more than expected by chance alone and help to add probe sets located within a range of 3 Mb. Interestingly, these structure to the vast amount of data generated by microarrays. 2 regions, in addition to a region nominated on chromosome The Database for Annotation, Visualization, and Integrated 1p36, were previously identified as human imprinted loci (33). Discovery (DAVID; http://david.abcc.ncifcrf.gov/home.jsp), The most biologically relevant genes identified were the po- a web-based program (32), allowed us to classify our differen- tassium channel, voltage-gated, shaker-related subfamily, tially expressed genes into shared biological categories or member 3 (KCNA3; 1.24-fold), the glis family zinc finger pro- chromosomal localization. DAVID computes Fisher’s exact tein 3 (GLIS3; 1.24-fold), the V-ets avian erythroblastosis virus P values and their resulting FDR (q value). E26 oncogene homolog 1 (ETS; 1.18-fold), the nuclear factor I/X (NFIX; 1.16-fold), and insulinoma-associated 1 (INSM1; 1.21-fold; directly flanking the C20orf74). Other genes located RESULTS in the vicinity (’0.5 Mb apart) of the differentially methylated loci were also of early interest. Of these, corticotropin-releasing Effects of caloric restriction hormone receptor 2 (CHRH2; 1.15-fold) was located at 0.6 Mb At baseline, the anthropometric and metabolic characteristics of the PLEKHA8 gene (chromosome 7p15); enoyl-CoA hydra- were not significantly different between the low and high res- tase, short-chain, 1, mitochondrial (ECHS1; 1.18-fold) was ponders to caloric restriction (Table 1). By design, the high found at 0.05 Mb of the SPRN gene (chromosome 10q26); and responders had stronger overall changes in adiposity than did the cholecystokinin B receptor (CCKBR; 1.17-fold) was located at low responders (Table 2). On average, the 2 groups lost 1.81 61.84 0.2 Mb of the HPX gene (chromosome 11p15). and 7.34 63.05 kg (P , 0.01) of fat mass, respectively, whereas changes in fat-free mass were not significantly different between Epigenomic profile adaptation to caloric restriction groups. Accordingly, changes in percentage body fat showed very comparable trends, as compared with changes in fat mass. Probes near 3 loci (chromosomes 1p36, 4q21, and 5q13) were In accordance with differences in fat mass loss between the found to be differentially methylated after caloric restriction groups, leptin gene expression was down-regulated by 1.24-fold (Table 3). For each of these loci, postintervention DNA meth- (FDR = 0.0875) in the high-fat-loss group compared with the ylation levels were higher in the high-responder group than in the nonresponsive women. No correlation between initial anthro- low-responder group. Given the postulated role of ectodermal- pometric and metabolic characteristics and changes in body neural cortex 1 gene (ENC1) in obesity (34) and our observation composition was observed (percentage body fat; Spearman that it was also differentially expressed in the present study (see correlation coefficient = 0.30, P = 0.30). below), the most attractive locus found to be differentially As presented in Table 2, body fat loss in response to caloric methylated after caloric restriction was located at 5q13 (1.38- fold)—the genomic region to which ENC1 belongs. ENC1 was restriction was very heterogeneous among the participants, ’ ranging from a gain of 0.77 kg to a loss of 12.98 kg. There was no found at 1 Mb from the probe set UHNhscpg0025297 on significant difference between the 2 groups in the changes in chromosome 5q13 (Table 3). blood pressure, lipid profile, and glucose and insulin values after The probe set UHNhscpg0018526, located on chromosome the intervention. 4q21 (located in the vicinity of PRDM8), was the only one to demonstrate differential methylation both before and after intervention. Epigenomic profiling (CpG-island profiling) Adipose tissue epigenomic profiling before dieting CpG-island microarray results validation Before caloric restriction, 35 loci were differentially meth- We further investigated DNA methylation across several loci ylated (FDR , 0.10) between the groups (Figure 1 and Table 3). nominated from our epigenomic microarray analyses. A total of Three loci were relatively hypomethylated in the high-responder 4 CpG dinucleotides within 2 loci showed significant methylation group compared with the low-responder group, whereas 32 loci differences by using the bisulfite-based Sequenom EpiTYPER MOLECULAR PATHWAYS INVOLVED IN BODY WEIGHT LOSS 315

TABLE 5 After-treatment signiﬁcant differentially regulated genes (by false discovery rate) with 1.5-fold expression differences between low and high responders to the caloric restriction intervention1

Gene symbol UniGene ID Fold change Probe set Chromosomal location

PSPH Hs.512656 24.25 205048_s_at chr7p15.2-p15.1 PHLDA2 Hs.154036 22.40 209803_s_at chr11p15.5 SFRP4 Hs.658169 22.07 204052_s_at chr7p14.1 CPEB4 Hs.127126 22.03 224828_at chr5q21 hCG_2565 Hs.144151 22.03 237339_at chr10p11.21 CLEC2B Hs.85201 22.01 1556209_at chr12p13-p12 ERBB4 Hs.390729 21.98 214053_at chr2q33.3-q34 TNMD Hs.132957 21.98 220065_at chrXq21.33-q23 IGHM NA 21.97 209374_s_at chr14q32.33 ADAMTS18 Hs.188746 21.94 230040_at chr16q23 NA Hs.670282 21.94 237356_at chr16q23.1 LHCGR Hs.468490 21.87 207240_s_at chr2p21 EGFL6 Hs.12844 21.85 219454_at chrXp22 SFRP4 Hs.658169 21.80 204051_s_at chr7p14.1 C9orf19 Hs.493819 21.78 225604_s_at chr9p13-p12 CCND2 Hs.376071 21.77 200952_s_at chr12p13 PPP2R1B Hs.584790 21.77 222351_at chr11q23.2 PTGER3 Hs.445000 21.75 210834_s_at chr1p31.2 USP36 Hs.464243 21.74 220370_s_at chr17q25.3 PEMT Hs.287717 21.73 207621_s_at chr17p11.2 ANKRD30A Hs.373787 21.71 223864_at chr10p11.21 GCNT2 Hs.519884 21.69 230788_at chr6p24.2 WASL Hs.143728 21.68 205809_s_at chr7q31.3 PALLD Hs.151220 21.67 200907_s_at chr4q32.3 CBL Hs.504096 21.67 229010_at chr11q23.3 PALLD Hs.151220 21.66 200906_s_at chr4q32.3 NA Hs.658612 21.66 238752_at chr6p22.2 KRT5 NA 21.66 201820_at chr12q12-q13 NA Hs.650577 21.63 230064_at NA PCDH17 Hs.106511 21.63 205656_at chr13q21.1 NA Hs.585479 21.63 236297_at chr10p12.31 TCF4 Hs.644653 21.62 212382_at chr18q21.1 QPRT Hs.513484 21.62 204044_at chr16p11.2 ENC1 Hs.104925 21.60 201340_s_at chr5q12-q13.3 NA Hs.134650 21.60 238755_at chr11p15.2 ABCC6 Hs.643018 21.60 214033_at chr16p13.1 C9orf19 Hs.493819 21.60 225602_at chr9p13-p12 CNKSR3 Hs.16064 21.59 227481_at chr6q25.2 NA Hs.660596 21.58 222877_at chr2q33.3 IQGAP1 Hs.430551 21.58 213446_s_at chr15q26.1 GPLD1 Hs.591810 21.57 206265_s_at chr6p22.3-p22.2 KCMF1 Hs.654968 21.57 222471_s_at chr2p11.2 CLCN4 Hs.495674 21.57 214769_at chrXp22.3 TACSTD2 Hs.23582 21.56 202286_s_at chr1p32-p31 OSBPL11 Hs.477440 21.55 222586_s_at chr3q21 HIST1H1C Hs.7644 21.55 209398_at chr6p21.3 NBN Hs.492208 21.55 217299_s_at chr8q21 NA Hs.586365 21.54 228987_at chr8q24.21 GPLD1 Hs.591810 21.54 215554_at chr6p22.3-p22.2 PDLIM4 Hs.424312 21.54 211564_s_at chr5q31.1 GPLD1 Hs.591810 21.53 206264_at chr6p22.3-p22.2 NDUFS1 Hs.471207 21.53 236356_at chr2q33-q34 RASGRF2 Hs.162129 21.53 228109_at chr5q13 TXN Hs.435136 21.53 216609_at chr9q31 NA Hs.363526 21.52 235696_at chr8p12 DHX9 Hs.74578 21.52 212107_s_at chr1q25 HIST2H2A Hs.530461 21.52 214290_s_at chr1q21.2 NA NA 21.51 240467_at NA PKP1 Hs.497350 21.51 221854_at chr1q32 (Continued) 316 BOUCHARD ET AL

TABLE 5 (Continued)

Gene symbol UniGene ID Fold change Probe set Chromosomal location

SYNC1 Hs.696281 21.51 221276_s_at chr1p34.3-p33 STK38L Hs.184523 21.51 212565_at chr12p11.23 TMEM20 Hs.632085 21.51 239265_at chr10q23.33 TMEM182 Hs.436203 21.50 238867_at chr2q12.1 GPR62 Hs.232213 1.50 1554559_at chr3p21.1 NPHP3 Hs.511991 1.50 235432_at chr3q22.1 NA Hs.673033 1.52 1557459_at chr11q23.1 CTSLL3 NA 1.52 1563445_x_at chr9q22.1 BCAP29 Hs.303787 1.54 241640_at chr7q22-q31 BCL2L11 Hs.469658 1.56 1553096_s_at chr2q13 LOC38983 Hs.632605 1.62 227715_at NA HIF3A Hs.420830 1.64 1555318_at chr19q13.32 NA NA 1.82 244181_at chr5q13.1 CETP Hs.89538 2.40 206210_s_at chr16q21 1 NA, not available; ID, identification. A modified t test was applied as implemented in Significance Analysis of Microarray to

each probe set (log2 transformed), and the raw P values were converted to the false discovery rate (q values) to correct for multiple testing according to Benjamini and Hochberg (30). The low responders were the reference group for fold-change computation. analysis approach. Three of these CpG were located in the sponsive (24.25-fold). Only 5 other genes showed MFEDs .2, DNASE1L2 probe set (CpG 15, 16, and 17; P = 0.002; FDR = 0%). and these were all found to be up-regulated: PHLDA2 (chro- DNA methylation across all 3 CpG sites was highly correlated mosome 11p15.5), SFRP4 (chromosome 7p14.1), and 3 other with one other and positively correlated with the CpG-island hypothetical genes. Other genes of interest based on biological array results (Spearman r = 0.71, P = 0.005). Another CpG function were those encoding phospholipase A2 group 6 dinucleotide with significant methylation differences between (PLA2G6, 1.29-fold), phospholipase C beta 1 and beta 2 the groups was located within the OAT probe set (CpG 21: P = (PLCB1, 1.35-fold; PLCB2, 1.29-fold), retinoic acid receptor 0.002 and FDR = 0%). The results correlated significantly with gamma (RARg, 1.27-fold), SH2Badaptor protein 2 (SH2B2, the microarray results (Spearman r = 20.74, P = 0.003) (Table 1.39-fold), nitric oxide synthase 1 (NOS1, 1.23-fold), sex hor- 4). The regions tested with bisulfite sequencing are somewhat mone–binding globulin (SHBG, 1.29-fold), angiopoietin 2 arbitrary, and it is likely that the specific critical CpG sites (ANGPT2, 21.41-fold), tumor necrosis factor receptor–associated causing differential enrichment (and thus microarray signal in- factor 3 (TRAF3, 21.27-fold), oxysterol binding protein-like 11 tensity) were not included in all the amplicons tested. However, (OSBPL11, 21.55-fold), glucocorticoid receptor DNA binding the observation that differential methylation was confirmed factor 1 (GRLF1, 21.32-fold), and protein inhibitor of activated within the sequence of 2 out of 4 probe sets tested suggests that STAT, 2 (PIAS2, 21.35-fold). the microarray data reflect true differences. See Supplementary Interestingly, 5 differentially expressed loci were also found to Figure 1 (under “Supplemental data” in the online issue) for be differentially methylated. These loci were located on chro- Spearman correlation coefficients between CpG-island results mosomes 3q21 (KPNA1), 5q13 (ENC1), 6q27 (C6orf120), 11p15 and corresponding HpaII-validated restriction sites. (HPX), and 15q15 (HYPK). For these loci, DNA methylation and gene expression were correlated for most of the probe sets (Table 6). Gene expression profiling Adipose tissue transcriptome before and after dieting Expression microarray results validation None of the tested genes showed differential adipose tissue A total of 11 genes showing 1.5-fold mRNA level differ- gene expression between the groups before caloric restriction. ences were validated by real-time PCR (Table 7). Seven of these However, after caloric restriction, a total of 334 transcripts were up-regulated (1.2-fold to 2.39-fold), whereas 342 transcripts were down-regulated (21.2-fold to 24.25-fold) in the high-responder group as compared with the low-responder group (see supplementary Tables 1 and 2 under “Supplemental data” in the online issue). The differentially expressed genes with 1.5-fold expression differences are shown in Table 5. These transcripts corre- spond to 327 and 317 differentially expressed genes in each group, respectively. The gene encoding cholesteryl ester transfer protein (CETP; NM_000078; chromosome 16q21) demonstrated the highest increase in mRNA levels (2.40-fold), and, among down-regulated genes, that encoding phosphoserine phosphatase FIGURE 1. Overview of the study design and of both methylation and (PSPH; NM_004577; chromosome 7p15) was the most re- expression microarray results. MOLECULAR PATHWAYS INVOLVED IN BODY WEIGHT LOSS 317

TABLE 6 Spearman correlation analysis between DNA methylation and expression levels for probe sets with signiﬁcant differences with both approaches Expression levels Gene Localization Methylation after intervention RPvalue

KPNA11 3q21 Increased at baseline Down-regulated 20.588 0.035 ENC1 5q13 Increased after treatment Down-regulated 20.689 0.007 C6orf120 6q27 Increased at baseline Down-regulated 20.720 0.006 HPX1 11p15 Increased at baseline Up-regulated 0.604 0.029 HYPK 15q15 Decreased at baseline Down-regulated 0.720 0.006 1 Two probe sets with significant expression level differences. genes showed comparable MFEDs, with a strong correlation belong to any of the tested KEGG metabolic pathways. The between microarray and real-time PCR results. TXN also differentially expressed genes falling into the protein transport showed very similar MFEDs with both methods, but the corre- category were as follows: TMEM48, EXOC2, RAB9A, C18orf55, lation between microarray and real-time PCR results was SRP19, PEX13, CHMP2B, TIMM17A, TLOC1, KPNA1, SCFD2, modest. Overall, most of the genes tested were validated by RT- RAB2A, SDAD1, PTPN11, NVTF2, RAB1A, RAB3IP, VPS37A, PCR, which suggests that our microarray data represent true and PAMC1. Those expressed in blood vessel development and expression changes. vasculature development were as follows: PDGFA, LAMA4, ANGPT1, ANGPT2, C9orf105, and CHD7. Finally, up-regulated Pathway analysis genes were more likely to be located on chromosomes 3q21 (P = DAVID software (32) was used to identify Gene Ontology 0.007, FDR = 0.12) and 22q11.23 (P = 0.01, FDR = 0.17). These [GO, biological processes (BP)] terms, KEGG metabolic path- genes were as follows: ABTB1, IFT122, and MYLK for the 3q21 ways, and chromosomal localizations in which a significant region and C22orf15, DERL3, GGH, HS.648268, and proportion of differentially expressed genes could be found. We 203877_AT for the latter region. report here only the most significant findings (P 0.05 and FDR 0.20). For up-regulated transcripts, none of the GO-BP terms reached this significance level. However, the cerebellar long- DISCUSSION term depression pathway was identified as the most significant The aim of this study was to document genome-wide dynamic KEGG term (P = 0.02, FDR = 0.19). The differentially ex- adaptations to caloric restriction in terms of cytosine methylation pressed genes falling into this category were as follows: PLCB1, and transcriptional activity occurring in adipose tissue. Our PLCB2, PLA2GVI, GNAO1, NOS1, and GUCY1B2. For the hypothesis was that failure to respond satisfactorily to caloric down-regulated genes, protein transport, blood vessel de- restriction in terms of fat mass loss could be accounted for by velopment, and vasculature development GO-BP terms were epigenomic and/or transcriptomic differences. At baseline, we identified (P = 0.002, FDR = 0.04; P = 0.006, FDR = 0.10; and found that despite various loci being differentially methylated P = 0.006, FDR = 0.10, respectively), but these genes did not between low and high responders to caloric restriction, none of

TABLE 7 Spearman correlations between real-time polymerase chain reaction (RT-PCR) and microarray measurements of selected adipose tissue gene transcripts1 Genes Probe set RT-PCR assay MFED array MFED RT-PCR RPvalue

PSPH 205048_s_at Hs00190154_m1 0.24 0.98 20.07 0.81 SFRP4 204052_s_at Hs00180066_m1 0.48 0.60 0.92 ,0.0001 SFRP4 204051_s_at 0.56 0.60 0.94 ,0.0001 PALLD 200907_s_at Hs00363101_m1 0.60 0.72 0.79 0.0008 PALLD 200906_s_at 0.60 0.72 0.77 0.001 PALLD 200897_s_at 0.68 0.72 0.74 0.003 ABCC6 214033_at Hs01081201_m1 0.63 0.75 0.77 0.001 GPLD1 206265_s_at Hs00412832_m1 0.64 0.73 0.73 0.003 GPLD1 215554_at 0.65 0.73 0.91 ,0.0001 GPLD1 206264_at 0.65 0.73 0.90 ,0.0001 NDUFS1 236356_at Hs00192297_m1 0.65 0.92 0.10 0.73 TXN 216609_at Hs00828652_m1 0.66 0.75 0.46 0.099 BCL2L11 1553096_s_at Hs00197982_m1 1.56 1.00 0.37 0.19 BCL2L11 1553088_a_at 1.36 1.00 0.03 0.91 OSBPL11 222586_s_at Hs00224361_m1 0.64 0.90 0.65 0.01 PCDH17 205656_at Hs00205457_m1 0.61 0.64 0.64 0.01 CETP 206210_s_at Hs00163942_m1 2.40 2.25 0.95 ,0.0001 1 MFED, mean-fold expression difference. 318 BOUCHARD ET AL the 47,000 transcripts tested was differentially expressed between restriction may be made easier (48). Thus, in our study, the the groups. Recently, in a study designed to verify whether ad- greater body fat loss in the high-responders group may be at- ipose tissue gene expression profiling can differentiate and tributed, at least in part, to the increased adipose tissue CETP predict dietary responders, Mutch et al (35) also reported no expression. At the molecular level, no evidence of differential significant differences between the low and high dietary res- DNA methylation was found for both probes located in the vi- ponders (at FDR 5%). Their conclusion was that tran- cinity of the CETP gene. Some polymorphisms have been scriptomic data were, at best, only a weak predictor of weight shown to affect CETP plasma concentrations (49, 50) and loss after caloric restriction. Our results and conclusions agree weight loss (51), but our study was not designed to test this with those reported by Mutch et al. specific hypothesis. Further studies are therefore needed to ex- However, the observation of adipose tissue DNA methylation plore the potential dysregulation of CETP cis and trans regu- differences characterizing each group at baseline suggests that latory elements in weight loss. the epigenomic profile could represent a more promising ap- Whereas the transcriptome is probably of limited use in dif- proach to differentiate low from high responders to caloric re- ferentiating dietary responders before an intervention, gene ex- striction. Of the gene loci showing differential DNA methylation, pression profiling after weight loss is more likely to be useful in 2 general categories emerged. KCNA3 and NFIX are represen- identifying genes and metabolic pathways that have to be acti- tative of the first category and are associated with weight con- vated or repressed to achieve a better fat mass loss. Our pathway trol. Knockout (KO) mice for both genes weigh significantly less analysis results suggest that down regulation of genes involved in than wild-type animals and are resistant to weight gain (36, 37). angiogenesis characterizes high responders to caloric restriction. INSM1, GLIS3, and CCKBR are representative of the second Given that angiogenesis dysregulation has been associated with category and are associated with diabetes and/or insulin secre- obesity and its metabolic complications (52), achieving down- tion. INSM1-KO mice have impaired pancreatic b cell de- regulation of these genes in the adipose tissue of obese indi- velopment and therefore insulin production and secretion (38). viduals could offer a better prognosis in terms of fat mass loss Defects in GLIS3 production are thought to be responsible for (53). On the other hand, the analyses showed that some of the up- permanent neonatal diabetes and congenital hypothyroidism in regulated genes in the high-responders group were associated human patients (39). Cholecystokinin (CCK) is a brain and gut with cerebellar long-term depression pathway. In the cerebellum, satiety peptide secreted in response to a meal that has the ca- this pathway is associated with memory and learning, and its pacity to stimulate insulin secretion through the activation of its long-term depression has been associated with expression receptors, CCKAR and CCKBR (40). Finally, the differentially changes for many genes (54). The brain-derived neurotrophic methylated region on chromosome 11p15 is located only 4.5 Mb factor (BDNF) is one of these genes and has been associated from the IGF2-H19 imprinted gene cluster, known to be asso- with obesity and weight control in numerous studies (55). ciated with growth regulation (41, 42). The insulin-like growth However, it was not found to be differentially expressed in this factor 2 gene (IGF2) is normally paternally expressed and has study, and the role of this pathway in human adipose tissue re- been associated with eating disorders. Interestingly, a recent mains to be determined. Finally, more up-regulated genes than study showed altered DNA methylation across IGF2 in in- would be expected by chance alone were localized on chro- dividuals exposed to prenatal famine and low calorific intake mosomes 3q21 and 22q11, which suggests that long-range DNA (43). Overall, even if only a small number of gene loci were modifications may have impaired gene expression at these loci. found differentially methylated, this approach identified several Interestingly, differential methylation levels were also found for strong candidate genes and proved to be potentially useful to probes in the chromosome 3q21 region. However, it is not differentiate low from high dietary responders. possible to determine whether the chromosome 3q21 locus After caloric restriction, only 3 chromosomal regions were signal corresponds to a larger genomic region than the one differentially methylated between the groups. This strongly identified by the probe set contained on the microarrays used in suggests that epigenetic markers are responsive to dieting. this study. However, there is no obvious explanation for the modest number In this study, DNA methylation and gene expression were of genes that were differentially methylated, but it is possible that correlated for a limited number of genes. Our data thus concur qualitative improvements in the diet, independently of fat mass with findings from the pilot Human Epigenome Project, which loss, may have made both groups more comparable in terms of suggest that the relation between the promoter methylation and methylation patterns. In contrast, adipose tissue gene expression transcription is not necessarily straightforward. Whereas there profiles appear to be strongly responsive to caloric restriction was a clear correlation between DNA methylation and gene with a total of 644 genes showing differential transcript levels expression for some loci, this was not the case for most of the after the intervention, compared with none at baseline. The genes (56). DNA methylation has several other genomic func- cholesterol ester transfer protein (CETP; chromosome 16q21) is tions in addition to regulating gene expression, including pro- 1 of the top 3 genes based on MFEDs between the groups and is tecting the genome against retroviral elements and playing a key of biological relevance to obesity. CETP is well known for its role in DNA replication, which should be taken into account role in mediating the transfer of cholesterol ester (CE) and tri- when interpreting the results. glycerides among plasma lipoproteins (44–46) but has recently A limitation of our study was that only a fraction of the genome been shown to be associated with lipid metabolism and storage could be interrogated using restriction enzymes. Indeed, DNA in adipose tissue (47). The results of a recent study suggest that methylation at CpG dinucleotides can be assessed only if they are met met higher CETP concentrations in adipocytes facilitate lipid located within McrBC [PuC (N40–3000)PuC ]orHpaII (un- transport to cellular sites, where access to hydrolytic enzymes is methylated CCGG) restriction sites. Therefore, it is plausible improved and therefore their utilization in situation of caloric that the remaining fraction of the genome harbors additional MOLECULAR PATHWAYS INVOLVED IN BODY WEIGHT LOSS 319 regions showing differential methylation levels. Nevertheless, female obese identical twins. Int J Obes Relat Metab Disord 2000;24: this approach has proven to be effective (23) and more sensitive 1051–7. 10. Petronis A. Human morbid genetics revisited: relevance of epigenetics. than other strategies used to identify small DNA methylation Trends Genet 2001;17:142–6. differences (57), such as those expected (44, 58). Moreover, it 11. Petronis A. Epigenetics and twins: three variations on the theme. Trends has been shown that the methylation levels of neighboring cy- Genet 2006;22:347–50. tosines correlate with one another, expanding the theoretical 12. Henikoff S, Matzke MA. Exploring and explaining epigenetic effects. Trends Genet 1997;13:293–5. genome coverage of this approach. Finally, although the current 13. Kangaspeska S, Stride B, Metivier R, et al. Transient cyclical methyl- sample size provided acceptable statistical power, the number of ation of promoter DNA. Nature 2008;452:112–5. samples used was still limited. Accordingly, these results need to 14. Bird A. DNA methylation patterns and epigenetic memory. Genes Dev be validated in a larger study. 2002;16:6–21. In summary, our results showed significant epigenomic and 15. Ling C, Del Guerra S, Lupi R, et al. Epigenetic regulation of PPARGC1A in human type 2 diabetic islets and effect on insulin secretion. Dia- transcriptomic differences in adipose tissue between the high and betologia 2008;51:615–22. low responders to caloric restriction. Furthermore, our data 16. Faraj M, Messier L, Bastard JP, et al. Apolipoprotein B: a predictor of showed that the epigenetic and transcriptomic profiles differ- inflammatory status in postmenopausal overweight and obese women. entiating these groups were responsive to the intervention. Diabetologia 2006;49:1637–46. 17. Karelis AD, Faraj M, Bastard JP, et al. The metabolically healthy but Whereas the results suggest that DNA methylation profiling obese individual presents a favorable inflammation profile. J Clin En- could be used to predict good responders to dieting, gene ex- docrinol Metab 2005;90:4145–50. pression profiling may enable the identification of genes and 18. Faraj M, Beauregard G, Loizon E, et al. Insulin regulation of gene ex- metabolic pathways that have the potential to improve fat mass pression and concentrations of white adipose tissue-derived proteins in vivo in healthy men: relation to adiponutrin. J Endocrinol 2006;191:427–35. loss. Understanding adipose tissue molecular adaptation to ca- 19. Faraj M, Beauregard G, Tardif A, et al. Regulation of leptin, adiponectin loric restriction may provide an opportunity to understand the and acylation-stimulating protein by hyperinsulinaemia and hyper- mechanisms put in place to protect the body’s energy reserves glycaemia in vivo in healthy lean young men. Diabetes Metab 2008;34: and may thus help to tailor obesity prevention and treatment 334–42. 20. Meugnier E, Faraj M, Rome S, et al. Acute hyperglycemia induces interventions. a global downregulation of gene expression in adipose tissue and skel- We express our gratitude to Zachary Kaminsky for his highly helpful ad- etal muscle of healthy subjects. Diabetes 2007;56:992–9. 21. St-Onge M, Mignault D, Allison DB, Rabasa-Lhoret R. Evaluation of vice while preparing DNA samples for epigenomic harm. We also acknowl- a portable device to measure daily energy expenditure in free-living edge the contribution of the Gene Quantification core laboratory of the Centre adults. Am J Clin Nutr 2007;85:742–9. de Geńomique de Que´bec, the Microarray Centre at the University Health 22. Heisler LE, Torti D, Boutros PC, et al. CpG Island microarray probe Network in Toronto (CpG-island arrays), the McGill University/Geńome sequences derived from a physical library are representative of CpG Que´bec Innovation Center (expression arrays), and Bioserve Inc (Benjamin Islands annotated on the human genome. Nucleic Acids Res 2005;33: Winterroth and Rama Modali). 2952–61. The authors’ responsibilities were as follows—LB, RR-L, MF, LP, and M- 23. Mill J, Tang T, Kaminsky Z, et al. Epigenomic profiling reveals CV: designed the experiment; LB, MF, and M-EL: collected the data; LB: DNA-methylation changes associated with major psychosis. Am J Hum performed the analyses and wrote the manuscript; and RR-L, MF, M-EL, Genet 2008;82:696–711. JM, LP, and M-CV: provided significant advice regarding the analyses and 24. Schumacher A, Kapranov P, Kaminsky Z, et al. Microarray-based DNA methylation profiling: technology and applications. Nucleic Acids Res interpretation of the data. All authors reviewed the manuscript. None of 2006;34:528–42. the authors had a financial or personal interest in a company or an organization 25. Ehrich M, Nelson MR, Stanssens P, et al. Quantitative high-throughput that could benefit directly from this research. analysis of DNA methylation patterns by base-specific cleavage and mass spectrometry. Proc Natl Acad Sci USA 2005;102:15785–90. 26. Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison of normalization methods for high density oligonucleotide array data based REFERENCES on variance and bias. Bioinformatics 2003;19:185–93. 1. Haslam DW, James WP. Obesity. Lancet 2005;366:1197–209. 27. Blazejczyk M, Miron M, Nadon R. FlexArray: a statistical data analysis 2. Flatt JP. Dietary fat, carbohydrate balance, and weight maintenance: software for gene expression microarrays. Montreal, Canada: Genome effects of exercise. Am J Clin Nutr 1987;45:296–306. Quebec, 2007. 3. Rodgers A, Ezzati M, Vander Hoorn S, Lopez AD, Lin RB, Murray CJ. 28. Livak KJ, Schmittgen TD. Analysis of relative gene expression data Distribution of major health risks: findings from the Global Burden of using real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Disease study. PLoS Med 2004;1:e27. Methods 2001;25:402–8. 4. National Institutes of Health. Clinical guidelines on the identification, 29. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays evaluation, and treatment of overweight and obesity in adults—the ev- applied to the ionizing radiation response. Proc Natl Acad Sci USA idence report. Obes Res 1998;6(suppl 2):51S–209S. 2001;98:5116–21. 5. Katz DL, O’Connell M, Yeh MC, et al. Public health strategies for 30. Benjamini Y, Hochberg Y. Controling the false discovery rate: a pratical and preventing and controlling overweight and obesity in school and powerfull approach to multiple testing. J R Stat Soc B 1995;57:289–300. worksite settings: a report on recommendations of the Task Force on 31. Tibshirani R. A simple method for assessing sample sizes in microarray Community Preventive Services. MMWR Recomm Rep 2005;54:1–12. experiments. BMC Bioinformatics 2006;7:106. 6. Klein S, Burke LE, Bray GA, et al. Clinical implications of obesity with 32. Dennis G Jr, Sherman BT, Hosack DA, et al. DAVID: Database for specific focus on cardiovascular disease: a statement for professionals Annotation, Visualization, and Integrated Discovery. Genome Biol from the American Heart Association Council on Nutrition, Physical 2003;4:3. Activity, and Metabolism: endorsed by the American College of Car- 33. Luedi PP, Hartemink AJ, Jirtle RL. Genome-wide prediction of im- diology Foundation. Circulation 2004;110:2952–67. printed murine genes. Genome Res 2005;15:875–84. 7. Hainer V, Zamrazilova H, Spalova J, et al. Role of hereditary factors in 34. Zhao L, Gregoire F, Sul HS. Transient induction of ENC-1, a Kelch-related weight loss and its maintenance. Physiol Res 2008;57:S1–15. actin-binding protein, is required for adipocyte differentiation. J Biol Chem 8. Bouchard C, Tremblay A, Despre´s JP, et al. The response to exercise 2000;275:16845–50. with constant energy intake in identical twins. Obes Res 1994;2:400–10. 35. Mutch DM, Temanni MR, Henegar C, et al. Adipose gene expression 9. Hainer V, Stunkard AJ, Kunesova M, Parizkova J, Stich V, Allison DB. prior to weight loss can differentiate and weakly predict dietary res- Intrapair resemblance in very low calorie diet-induced weight loss in ponders. PLoS One 2007;2:e1344. 320 BOUCHARD ET AL

36. Driller K, Pagenstecher A, Uhl M, et al. Nuclear factor I X deficiency 47. Izem L, Morton RE. Possible role for intracellular cholesteryl ester causes brain malformation and severe skeletal defects. Mol Cell Biol transfer protein in adipocyte lipid metabolism and storage. J Biol Chem 2007;27:3855–67. 2007;282:21856–65. 37. Xu J, Koni PA, Wang P, et al. The voltage-gated potassium channel 48. Zhou H, Li Z, Hojjati MR, et al. Adipose tissue-specific CETP ex- Kv1.3 regulates energy homeostasis and body weight. Hum Mol Genet pression in mice: impact on plasma lipoprotein metabolism. J Lipid Res 2003;12:551–9. 2006;47:2011–9. 38. Gierl MS, Karoulias N, Wende H, Strehle M, Birchmeier C. The 49. Lottenberg AM, Nunes VS, Nakandakare ER, et al. The human cho- zinc-finger factor Insm1 (IA-1) is essential for the development of lesteryl ester transfer protein I405V polymorphism is associated with pancreatic beta cells and intestinal endocrine cells. Genes Dev 2006;20: plasma cholesterol concentration and its reduction by dietary phytosterol 2465–78. esters. J Nutr 2003;133:1800–5. 39. Taha D, Barbar M, Kanaan H, Williamson Balfe J. Neonatal diabetes 50. Sandhofer A, Tatarczyk T, Laimer M, et al. The Taq1B-variant in the mellitus, congenital hypothyroidism, hepatic fibrosis, polycystic kid- cholesteryl ester-transfer protein gene and the risk of metabolic syn- neys, and congenital glaucoma: a new autosomal recessive syndrome? drome. Obesity (Silver Spring) 2008;16:919–22. Am J Med Genet A 2003;122A:269–73. 51. Ruanõ G, Windemuth A, Kocherla M, et al. Physiogenomic analysis of 40. Ahren B, Holst JJ, Efendic S. Antidiabetogenic action of cholecysto- weight loss induced by dietary carbohydrate restriction. Nutr Metab kinin-8 in type 2 diabetes. J Clin Endocrinol Metab 2000;85:1043–8. (Lond) 2006;3:20. 41. Smith AC, Choufani S, Ferreira JC, Weksberg R. Growth regulation, 52. Costa C, Incio J, Soares R. Angiogenesis and chronic inflammation: imprinted genes, and chromosome 11p15.5. Pediatr Res 2007;61:43R–7R. cause or consequence? Angiogenesis 2007;10:149–66. 42. Waterland RA, Lin JR, Smith CA, Jirtle RL. Post-weaning diet affects 53. Rupnick MA, Panigrahy D, Zhang CY, et al. Adipose tissue mass can be genomic imprinting at the insulin-like growth factor 2 (Igf2) locus. Hum regulated through the vasculature. Proc Natl Acad Sci USA 2002;99: Mol Genet 2006;15:705–16. 10730–5. 43. Heijmans BT, Tobi EW, Stein AD, et al. Persistent epigenetic differences 54. Vigot R. [Cerebellar long-term depression: a mechanism for learning associated with prenatal exposure to famine in humans. Proc Natl Acad and memory.] Med Sci (Paris) 2003;19:437–41 (in French). Sci USA 2008;105:17046–9. 55. Lebrun B, Bariohay B, Moyse E, Jean A. Brain-derived neurotrophic 44. Hesler CB, Swenson TL, Tall AR. Purification and characterization of factor (BDNF) and food intake regulation: a minireview. Auton Neu- a human plasma cholesteryl ester transfer protein. J Biol Chem 1987; rosci 2006;126-127:30–8. 262:2275–82. 56. Eckhardt F, Lewin J, Cortese R, et al. DNA methylation profiling of 45. Hesler CB, Tall AR, Swenson TL, Weech PK, Marcel YL, Milne RW. human chromosomes 6, 20 and 22. Nat Genet 2006;38:1378–85. Monoclonal antibodies to the Mr 74,000 cholesteryl ester transfer pro- 57. Irizarry RA, Ladd-Acosta C, Carvalho B, et al. Comprehensive high- tein neutralize all of the cholesteryl ester and triglyceride transfer ac- throughput arrays for relative methylation (CHARM). Genome Res tivities in human plasma. J Biol Chem 1988;263:5020–3. 2008;18:780–90. 46. Morton RE, Zilversmit DB. Purification and characterization of lipid 58. Tobi EW, Lumey LH, Talens RP, et al. DNA methylation differences transfer protein(s) from human lipoprotein-deficient plasma. J Lipid Res after exposure to prenatal famine are common and timing- and sex- 1982;23:1058–67. specific. Hum Mol Genet 2009;18:4046–53.