Impaired Mitochondrial Biogenesis in Adipose Tissue in Acquired Obesity

Adipocytes are important contributors to energy balance and metabolic homeostasis. Within these highly dynamic cells, mitochondria are at the center of energy metabolism, using carbohydrates, lipids, and proteins to produce ATP and metabolites for growth, as well as contributing to adipocyte differentiation and maturation (1). Mitochondria possess their own multicopy genome, a 16.6-kb circular mitochondrial DNA (mtDNA) that encodes two ribosomal RNAs (12S and 16S), 22 transfer RNAs, and 13 polypeptides (2). These proteins are the core catalytic components of electron transport chain complexes I, III, and IV, and ATP synthase (3). Together, these complexes form the oxidative phosphorylation (OXPHOS) system in the inner mitochondrial membrane. mtDNA-encoded proteins are translated on mitochondrial ribosomes that, together with OXPHOS complexes, are the only cellular entities encoded by both nuclear and mitochondrial genomes. However, ∼1,500 other nuclear-encoded proteins—8% of all nuclear genes—encode mitochondrial targeted proteins, emphasizing the central role of mitochondria in cellular metabolism (4).

Evidence from various study systems suggests that mitochondrial function and biogenesis are compromised in subcutaneous adipose tissue (SAT) in type 2 diabetes (5), morbid obesity (6), and insulin resistance (7). Recently, a reduction of total oxygen consumption rates, but not mtDNA content, was shown to occur in visceral and subcutaneous adipocytes of obese adults (8), as well as after β-adrenergic stimulation (9), suggesting a decrease in mitochondrial oxidative activity. We have previously shown in SAT of monozygotic (MZ) weight-discordant twin pairs that mtDNA depletion together with the downregulation of mitochondrial branched-chain amino acid (BCAA) catabolism characterize obesity, insulin resistance, fatty liver (10), and poor fitness (11). However, we still lack a thorough understanding of how mitochondrial biogenesis and function in human adipose tissue changes in obesity, at which stage this process begins, and whether it contributes to the early development of metabolic disturbances. Moreover, it is not known whether the possible link between mitochondrial dysfunction and obesity-related metabolic disorders is genetic or acquired.

MZ twins discordant for obesity are, despite their different phenotype, completely matched for genetic variants, age, and sex. This allows the research of acquired obesity without interference of genetic and familial factors that confound studies comparing groups of unrelated lean and obese individuals. Here, we focus on rare weight-discordant, healthy, young adult MZ twins to investigate in detail the mitochondrial pathways in SAT and their association with whole-body metabolism in acquired obesity.

The current study included 26 rare, healthy MZ pairs discordant for obesity (within-pair difference, ΔBMI ≥3 kg/m², n = 9 males, n = 17 females, mean age 29.9 ± 0.6 years), identified from two population-based twin cohorts, FinnTwin16 (n = 2,839 pairs) and FinnTwin12 (n = 2,578 pairs) (12). In addition, 14 MZ twin pairs concordant for BMI (ΔBMI <3 kg/m², n = 9 males, n = 5 females, mean age 31.6 ± 0.6 years) were included as control pairs and to provide a wider range of BMI values for transcriptomics analyses. A detailed description of the twin material has been published previously (13,14). Written informed consent was obtained from all participants. The study was approved by the Ethical Committee of the Helsinki University Central Hospital.

Weight and height were measured for the calculation of BMI. Body composition was determined by dual-energy X-ray absorptiometry (Lunar Prodigy software version 8.8; GE Healthcare, Madison, WI) (15), the amount of SAT and visceral adipose tissue (VAT) by MRI, and liver fat by proton magnetic resonance spectroscopy with a 1.5 Tesla MRI imager (Avanto; Siemens, Erlangen, Germany) (13). Physical activity was assessed by the Baecke questionnaire and its three subcompartments (sport, leisure time, and work indexes). Parental weight, height, and age were asked at the twin age of 16 years.

The concentrations of fasting plasma glucose, leptin, adiponectin, and adipsin, and serum hs-CRP were measured, and HOMA insulin resistance and Matsuda insulin sensitivity indexes were calculated, as previously described (16).

Surgical biopsy samples of abdominal SAT under the umbilicus were obtained under local anesthesia and were snap-frozen in liquid nitrogen. Based on the availability of sample material, the frozen SAT specimens were used for transcriptomics analyses (all 26 discordant and 14 concordant twin pairs), determination of the amount of mtDNA, mtDNA-encoded transcript levels (15 discordant twin pairs), and an OXPHOS protein analysis (seven discordant twin pairs). Twenty-four discordant and 11 concordant twin pairs were available for DNA methylation analyses. Clinical characteristics of the selected and unselected twin pairs were similar. Part of the fresh SAT sample was digested with collagenase, and was used for the measurement and calculation of adipocyte volume and number in all 26 discordant and 14 concordant twin pairs (16). The discordant twin pairs were divided into two groups with regard to hyperplastic and hypertrophic obesity (16).

Total RNA was extracted from SAT by the RNeasy Lipid Tissue Mini Kit (Qiagen) and treated with DNase I (Qiagen). Transcriptomics analyses were performed using the Affymetrix U133 Plus 2.0 array and were validated as in the study by Naukkarinen et al. (14). Preprocessing of the expression data was performed using BioConductor and the GC-RMA algorithm. We first analyzed the differentially expressed transcripts genome wide between the obese and the lean co-twins in the discordant pairs by paired moderated t tests (17), and then determined whether they were found in MitoCarta (4), an online atlas of 1,013 human proteins with mitochondrial localization. Forty-nine of the 1,013 MitoCarta protein transcripts were not detectable by the Affymetrix probes, including the 13 mtDNA-encoded OXPHOS transcripts. Identified mitochondrial-related transcripts were further subjected to ingenuity pathway analysis (IPA; Qiagen, Redwood City, CA [www.qiagen.com/ingenuity]) to reveal the pathways and the upstream regulators of the transcripts. Top 10 significant pathways were selected for further examination and for the calculation of overall pathway activity (the mean centroid; Table 3) (10). In addition, the mean centroids were calculated for the expression of the mitochondrial ribosomal protein (MRP) small subunit (MRPS; n = 30 subunits) and the mitochondrial ribosomal protein large subunit (MRPL; n = 50 subunits) (Supplementary Table 4).

DNA methylation values of all probes mapping to the differentially expressed MitoCarta genes and peroxisome proliferator–activated receptor γ coactivator (PGC)-1α were picked from the Illumina HumanMethylation BeadChip. Differential methylation analysis of the cytosine guanine dinucleotide (CpG) sites on these genes was performed by using paired moderated t tests (18) in the discordant pairs, and relationships between methylation in significant CpG sites and the expression of their associated genes by Spearman correlations in the individual twins.

The amount of mtDNA, extracted by the standard phenol/chloroform and ethanol precipitation method, was determined by quantitative real-time PCR (qPCR) of the mitochondrial cytochrome B (CYTB) gene normalized to the nuclear amyloid β precursor protein (APP) gene (10). The qPCR was performed with 25 ng of total DNA using SYBR Green PCR Master Mix (iQ Custom SYBR Green Supermix; Bio-Rad) according to the manufacturer's instructions. The qPCR data were analyzed by the comparative “ΔΔCt method” (with Bio-Rad CFX Manager version 1.6 software). The primer sequences are available in Supplementary Table 7.

Total extracted RNA (250 ng) was used to generate cDNA with the Maxima First Strand cDNA Synthesis Kit for quantitative RT-PCR (qRT-PCR) (Thermo Scientific) according to the manufacturer's instructions. Expression levels of mtDNA-encoded 12S rRNA (MT-RNR1), 16S rRNA (MT-RNR2), and mRNAs (MT-COX1, MT-ND5, and MT-CYTB), normalized with nuclear 18S rRNA, were analyzed in 1:100 cDNA dilutions by qRT-PCR with the iQ Real-Time PCR Detection System (Bio-Rad) and the comparative ΔΔCt method.

The expression of the upstream regulator PGC-1α was validated in a 1:30 dilution with 18S rRNA as a housekeeping gene. The PGC-1α probes were targeted to 3′ untranslated region of the longest mRNA isoform (ENSG00000109819) that is not shared with other protein coding PGC-1α isoforms. The primer sequences are available in Supplementary Table 7.

SAT lysates were obtained by homogenization in ice-cold 1% n-dodecyl β-d-maltoside in PBS with protease inhibitors (Roche). Quantitative Western blot analysis was performed with 20 µg of total SAT lysates for measurement of the levels of the following OXPHOS proteins: complex I NDUFA9 (MS111), complex II 70-kDa subunit (MS204), complex III core 2 subunit (MS304), complex IV subunit I (MS404), and complex V subunit α (MS507), as well as porin (Abcam), β-tubulin (Cell Signaling Technology), and actin (Santa Cruz Biotechnology). Antibodies were diluted in 1% BSA/Tris-buffered saline and 0.1% Tween 20 (TBST) at 1:2,000 (CI), 1:5,000 (CII), 1:2,500 (CIII), 1:500 (CIV), and 1:1,000 (porin, tubulin), and incubated at 4°C overnight. Secondary HRP-conjugated anti-mouse or anti-rabbit antibodies (1:10,000; Molecular Probes) were incubated with the membranes in 5% milk/TBST. The proteins were quantified with Image Laboratory software version 3.0.

Statistical analyses were performed using Stata statistical software (release 12.0; Stata Corporation, College Station, TX). Differences between co-twins were calculated by paired Wilcoxon signed rank tests. Correlations between mitochondrial and metabolic variables in all concordant and discordant twin individuals were calculated by using Pearson correlations, corrected for the clustered sampling of co-twins by survey methods (19) and adjusted for multiple comparisons by the number of principal components. The variables in Table 5 produced one principal component, which explained 88% of the variance. Therefore, the adjusted significance level for analyses in Table 5 was P < 0.05 divided by 1, that is, P < 0.05. Logarithmic corrections were used for non-normally distributed variables.

Characteristics of the twins are presented in Table 1. The obese (BMI 31.2 ± 1.0 kg/m²) and the lean (25.3 ± 0.9 kg/m²) co-twins of the discordant pairs (n = 26) had a mean difference of 18 ± 0.5 kg (P < 0.001) in body weight. The obese co-twins had significantly more SAT, VAT, and liver fat; larger adipocytes; higher plasma leptin levels; and lower plasma adiponectin levels; and were more insulin resistant than the lean co-twins. The concordant co-twins (n = 14) had similar body composition and metabolic measures (Table 1). As no differences in mitochondrial measures were found within the concordant pairs, results are presented only for the discordant pairs.

We first investigated the within-pair differences in the expression of mitochondrial proteins in SAT (n = 26) by linking the results from the transcriptomics data with those of MitoCarta. Among the 2,108 significantly differentially expressed genes, 222 genes were listed in MitoCarta (Supplementary Table 1). Most of these transcripts (187 of 222 transcripts, 84%) were downregulated in SAT of the obese co-twins compared with the lean co-twins. The top significantly upregulated and downregulated genes are shown in Table 2.

To investigate which functional entities the 222 mitochondrial transcripts represented, we subjected the transcripts to IPA. Among the most significantly different and downregulated pathways in the obese co-twins compared with the lean co-twins were the following key mitochondrial functions: OXPHOS (P < 0.0001), BCAA degradation (P < 0.0001), ketogenesis/ketolysis (P < 0.0001), the tricarboxylic acid cycle (TCA; P < 0.0001), glutaryl-CoA degradation (P < 0.0001), and fatty acid β-oxidation (FAO; P < 0.0001) (Table 3 and Supplementary Table 2).

Next we analyzed whether the most significant pathways were associated with obesity-related pathology by correlating the mean centroids of the pathways to measures of adiposity (SAT, VAT, and liver fat, adipocyte volume), insulin sensitivity (Matsuda index), insulin resistance (HOMA index), inflammation (adipsin, hs-CRP), leptin, and adiponectin. After multiple testing episodes, the mean centroid values of OXPHOS and TCA pathways correlated with SAT, VAT, liver fat, adipocyte volume, Matsuda and HOMA indexes, adipsin, and hs-CRP (Table 4). The TCA also correlated with adiponectin. Glutaryl-CoA degradation, ketogenesis, and tryptophan degradation pathways correlated with all of the selected clinical variables except adiponectin. BCAA degradation (valine, isoleucine, and leucine degradation pathways), FAO, and ketolysis correlated with all clinical variables.

We performed IPA upstream regulator analysis to identify the transcriptional regulators that can explain the observed gene expression changes and pathways between the co-twins (n = 26) (20). We identified FOXO1 (P < 0.0001), PGC-1α (PPARGC1A, P < 0.0001), PGC-1β (PPARGC1B, P < 0.0001), and lysine methyltransferase 2D (KMT2D, P = 0.027) (Table 5 and Supplementary Table 3) as the most significant upstream transcription factors regulating the 222 transcripts. Based on the predicted z scores calculated in IPA, the biological activation state of all of these genes was lower in the obese co-twins compared with the lean co-twins. In our transcriptomics data, the expression level of PGC-1α was highly downregulated in the obese co-twins (lean 51.6 ± 5.1 arbitrary units [AU] vs. obese 34.1 ± 4.9 AU; arbitrary Affymetrix units, adjusted P < 0.001). The reduction of PGC-1α expression was verified with qRT-PCR (P = 0.004; Fig. 1). The expression of other upstream regulators was not different between the co-twins, but the expression levels of their target genes were reduced, suggesting post-transcriptional regulation (Table 5).

Next, we investigated whether any of the 222 differentially expressed genes were also differentially methylated within the twin pairs, thus providing a potential molecular mechanism underlying the expression differences. Altogether, 130 CpG sites within 74 genes showed differential methylation (P < 0.05) within the discordant pairs (n = 24; Supplementary Table 5). Most of these sites (96 of 130 sites, 74%) were hypermethylated in the obese co-twins compared with the lean co-twins. When investigating the CpG sites in relation to CpG density, 89% (116 of 130 sites) were outside any CpG islands. Approximately half (68 of 130 sites, 52%) of the differentially methylated CpG sites were located in the gene bodies, and 42% (55 of 130 sites) were located in promoters. There was no difference between the promoter- and gene body–associated CpG sites in regard to their methylation status (hypermethylation vs. hypomethylation). We then explored whether there are any significant associations between the differentially expressed and methylated genes. Of the 74 genes, 40 showed significant correlation between expression and methylation (r = −0.24 to −0.70 and 0.25 to 0.44, P < 0.05) (Supplementary Table 6). There was in general an inverse relationship between expression and DNA methylation, and only 12 genes showed positive correlations.

Furthermore, we explored whether the methylation levels differed between the co-twins for the PGC-1α gene, the only upstream regulator that was differentially expressed between the co-twins. We found hypermethylation in the obese co-twins in two CpG sites within the gene body of PGC-1α, but no differences in other regions of the gene (Supplementary Table 5). Methylation in one of the CpG sites correlated significantly with PGC-1α expression (r = −0.31, P = 0.01) (Supplementary Table 6).

To further investigate the effect of obesity on mitochondrial amount and biogenesis in SAT, we measured the amount of mtDNA and the expression of mtDNA-encoded genes (12S rRNA, 16S rRNA, cytochrome c oxidase subunit 1 [COX1], NADH dehydrogenase subunit 5 [ND5], and CYTB, n = 15 discordant pairs). The mtDNA amount was reduced by ∼20% (P = 0.031; Fig. 2A) in SAT of the obese co-twins compared with the lean co-twins. Significant negative correlations were observed between the mtDNA amount and HOMA (insulin resistance) and adipsin (inflammation) (Table 4).

Furthermore, mitochondrial 12S (MT-RNR1, P = 0.0064) and 16S (MT-RNR2, P = 0.0090) rRNAs, as well as mRNAs COX1 (P = 0.0064, CIV subunit), ND5 (P = 0.027, CI subunit), and CYTB (P = 0.0015, CIII subunit; Fig. 2B), were decreased in SAT of the obese co-twins compared with their lean counterparts. We also studied the expression of 30 nuclear transcripts for the MRPS and 50 nuclear transcripts for the MRPL of the mitochondrial ribosome using the Affymetrix data (n = 26 discordant pairs). In line with 12S and 16S rRNA expression, the average value of the total expression (a mean centroid) of both MRPS (lean 0.03 ± 0.09 AU vs. obese −0.31 ± 0.1 AU, P < 0.001) and MRPL (lean −0.04 ± 0.1 AU vs. obese −0.26 ± 0.1 AU, P = 0.0034) was lower in the obese compared with the lean co-twins (Fig. 2C).

Finally, we measured mitochondrial content and the level of OXPHOS protein subunits in SAT lysates of seven discordant pairs by Western blot. Mitochondrial mass per cell (porin, a mitochondrial outer membrane protein compared with a cytoskeletal protein β-tubulin) trended downward in the obese co-twins (P = 0.09; Fig. 2D and E). Mitochondrial OXPHOS protein subunits of CIII, CIV, and CV were reduced in obesity: the level of CIII-core 2 (CIII subunit, P = 0.018), MT-CO1 (CIV subunit, P = 0.018), and CV-α (CV subunit, P = 0.028), compared with the levels of β-tubulin, were all lower in the obese co-twins than in the lean co-twins (Fig. 2E). As the different complex subunits typically follow the amounts of full holocomplexes, these results support deficiency of CIII, CIV, and CV. Fully nuclear-encoded CII did not differ between the co-twins. Furthermore, we analyzed whether the OXPHOS levels were reduced per mitochondria in obesity. Thus, we normalized the OXPHOS protein signals against the mitochondrial protein porin. The amount of CI (P = 0.063) trended downward, and complexes CIII (P = 0.018) and CIV (P = 0.028) were reduced in the obese co-twins (Fig. 2E). Overall, the protein results suggest a trend of lower mitochondrial mass (porin) and a decrease of OXPHOS levels per cell and per mitochondria as a downstream effect of the reduced mtDNA amount (Fig. 3).

Because the mitochondrial pathways had significant negative correlations with adipocyte volume (P < 0.01, Table 4), we analyzed whether the mitochondrial measures differed in discordant pairs with hyperplastic versus hypertrophic pattern of adipocyte morphology. In the discordant pairs where the obese co-twins had a solely hypertrophic pattern (larger and fewer adipocytes than their co-twins), the obese co-twins had a significantly reduced amount of mtDNA compared with their lean co-twins (lean 1.05 ± 0.7 vs. obese 0.82 ± 0.07, P = 0.028), but no other differences in the mitochondrial measures were found between the two groups.

The within-pair differences in all measures of adiposity and metabolism (Table 1), as well as those in mitochondrial measures, were similar in males and females. The mean age at the onset of obesity discordance was 22.6 ± 0.6 years. Consequently, the length of obesity discordance was on average 6.3 ± 0.3 years. Of the total 44 parents of the discordant pairs for whom data were available, 15.9% were obese (4 of 23 of the mothers and 3 of 21 of the fathers), including one twin pair where both parents were obese. The parents were examined when the twins were 16 years of age. The age of the mothers was 42.2 ± 0.7 years, and that of the fathers, 45.4 ± 0.8 years. The total as well as sport, leisure time, and work physical activity levels of the obese and the lean co-twins were similar (Table 1), and did not correlate with mitochondrial measures.

We performed a comprehensive set of experiments to test a hypothesis that mitochondrial biogenesis is downregulated in human SAT in obesity. We used a unique study design: rare MZ weight-discordant twin pairs, which enabled us to control for sex, age, and genetic background between the lean and the obese groups—a major advantage for conclusions regarding a multifactorial trait such as obesity. We demonstrate a widespread reduction in the expression of both mitochondrial- and nuclear-encoded mitochondrial genes leading to downregulation of mitochondrial biogenesis in SAT of the obese twins compared with their leaner co-twins. Many of the differentially expressed transcripts were also differentially methylated within the twin pairs. Reduction of the mitochondrial oxidative metabolism in SAT correlated with whole-body insulin resistance and inflammation. Our findings provide evidence of a substantial insufficiency of mitochondrial biogenesis in SAT in acquired obesity that already exists prior to the onset of clinical metabolic complications.

The OXPHOS system is unique because it is encoded by both the nuclear and the mitochondrial genome. We showed that mtDNA, its transcripts, OXPHOS subunits, as well as nuclear genes encoding OXPHOS proteins are reduced in the obese co-twins, suggesting a major downregulation in mitochondrial oxidative ATP production and catabolic functions in SAT in acquired obesity. A similar reduction of mtDNA in SAT in obesity has been found in previous studies by us (10) and others (21), but the generalized downregulation of mitochondrial proteins and metabolism has not been reported previously. After mtDNA transcription, mitochondria translate their mRNAs on mitochondrial ribosomes that comprise two mtDNA-encoded rRNAs and >80 nuclear-encoded proteins (22). We report an extensive reduction in the transcript levels of rRNAs and MRPs, emphasizing the downregulation of the translation machinery of both the nuclear and mitochondrial expression systems. These findings are in agreement with previous findings in two mouse models of obesity and type 2 diabetes, which showed a low abundance of MRP transcripts and mitochondrial biogenesis in inguinal adipose tissue (23), as well as in a rat model for late-onset obesity, in which a truncation of MRPS26 was associated with the accumulation of visceral fat and insulin sensitivity (24). Finally, the OXPHOS complexes, partially encoded by mtDNA, were downregulated without a significant decrease of mitochondrial mass (porin/tubulin), showing that the mitochondrial OXPHOS levels within the cell and within the mitochondria are reduced in the SAT of the obese co-twins.

Pathway analysis of nuclear-encoded mitochondrial transcripts revealed changes in mitochondrial oxidative pathways, including FAO, TCA, ketogenesis, ketolysis, and BCAA degradation pathways in the obese co-twins. We found that a downregulation in oxidative metabolic pathways in SAT inversely correlated with several clinical parameters. Mitochondrial pathways correlated negatively with all adiposity measures, insulin resistance, and inflammation and correlated positively with adiponectin levels. We also observed that the larger the adipocyte volume, the more severe was the decrease in mitochondrial oxidative metabolic pathways. Recently, it has been proposed that a decrease in mitochondrial oxidative metabolism measured by oxygen consumption rate in isolated adipocytes is caused by obesity rather than the adipocyte size (8). Our study design controls for genetic factors between the obese and the lean groups and suggests that the hypertrophic but not the hyperplastic pattern of obesity relates to decreased mtDNA amount. Thus, our findings are in line with previous studies (16,25–27) reporting associations between hypertrophy and metabolic abnormalities.

Although mitochondrial function and the development of type 2 diabetes are suggested to be linked (28), previous studies on mitochondrial biogenesis in relation to insulin sensitivity in adipocytes and adipose tissue in human subjects are surprisingly few. In one study (29), the exposure of 3T3-L1 mouse preadipocytes to high levels of glucose and free fatty acids resulted in decreased mitochondrial size and membrane potential and in the downregulation of PGC-1α, indicating mitochondrial dysfunction. In another study (30), high-fat feeding of rats resulted in impaired glucose homeostasis and decreased PGC-1α expression, mtDNA content, and CI and CIV protein levels in epididymal adipose tissue. These studies supported a relationship between abundant glucose or fatty acids, decreased mitochondrial function, and impaired glucose tolerance, whereas studies on mouse adipocytes did not confirm a causal link between reduced mitochondrial biogenesis and impaired whole-body glucose homeostasis (31). Our study, combining molecular analysis of mtDNA metabolism with mitochondrial pathway analyses, suggests that impaired mitochondrial biogenesis in SAT in humans is related to insulin resistance and to other mild metabolic alterations already present before the clinical diagnosis of type 2 diabetes or other obesity-related diseases.

PGC-1α was the only predicted upstream regulator whose expression was different between the co-twins. PGC-1α is a transcription coactivator and a master regulator of mitochondrial biogenesis that controls and coordinates the expression of nuclear and mitochondrial genes in white adipocytes (32–35). Low PGC-1α expression in SAT has also been found to correlate with increased BMI (21) and to be associated with morbid obesity (6), type 2 diabetes (5), and insulin resistance (7). PGC-1α–responsive OXPHOS genes showed reduced expression in VAT in type 2 diabetes (36). Fat-specific PGC-1α knockout mice displayed a significant reduction in transcripts of OXPHOS, FAO, and TCA genes in inguinal SAT, and developed glucose intolerance and insulin resistance when challenged with a high-fat diet, demonstrating the necessity of oxidative metabolism in fat for whole-body metabolism (37). FOXO1 was the first predicted upstream regulator of our 222 transcripts. The expression of FOXO1 has been shown (38) to be downregulated in VAT compared with SAT of obese and nonobese individuals, and its protein levels reduced in a dose-dependent manner in 3T3-L1 adipocytes treated with free fatty acids as well as in intraperitoneal adipose tissue of db/db mice (39). We did not find significantly changed FOXO1 transcript levels, but FOXO1 and also PGC-1α are post-translationally regulated (40,41), and therefore the transcript levels do not fully reflect the gene activity. We show that three CpG sites in the body of PGC-1α were hypermethylated in the obese co-twins. This hypermethylation correlated negatively with gene expression, suggesting that DNA methylation may reduce the transcription of the gene. Previous studies (42) have reported that compared with the lean controls, obese patients with type 2 diabetes have a PGC-1α promotor hypermethylation in SAT. Our data also show that 74 of the 222 differentially expressed genes targeting mitochondria are differentially methylated between the co-twins, and in general inversely correlated with expression. This poses an interesting possibility that epigenetic modifications in certain nuclear genes associate with the decrease in mitochondrial biogenesis in SAT in acquired obesity. In obesity-discordant MZ twins, such an effect is acquired, and independent of the genetic background.

Healthy young adult MZ twins discordant for obesity are extremely rare, as they share both genes and family. We succeeded in finding 26 highly discordant pairs—mainly in their third decade of life, with relatively recent onset of weight discordance—from 10 birth cohorts of Finnish twins. Because the twins are matched on multiple factors, including genes, our study has increased power over a study comparing unrelated obese and lean subjects. Our study is cross-sectional and therefore does not allow definitive conclusions about causality. However, genetic variants and shared exposures/experiences can be excluded as confounders of the observed associations. All analyses in this study were performed with whole SAT biopsy samples that include stroma-vascular fraction cells, the contribution of which cannot be distinguished from that of the adipocytes.

In summary, we observed strong evidence for an association between acquired obesity and downregulation of mitochondrial biogenesis in SAT. Our findings suggest that obesity is related to suppressed mitochondrial oxidative activity in SAT. Importantly, we showed that the downregulation of mitochondrial biogenesis and oxidative metabolic pathways in SAT is a phenomenon in obesity associated with a range of mild metabolic alterations, insulin resistance, and low-grade inflammation. The genes targeting mitochondria may be epigenetically regulated. These data suggest that obesity-related disease development may be halted by improving mitochondrial activity in SAT.

Acknowledgments. The authors thank the twins for their invaluable contributions to the study. The authors also thank Barbara Every, of BioMedical Editor, for English language editing, and Eija Pirinen for her valuable comments.

Funding. The study was supported by Helsinki University Hospital Research Funds and by grants from the Novo Nordisk Foundation, the Diabetes Research Foundation, the Jalmari and Rauha Ahokas Foundation, the Finnish Foundation for Cardiovascular Research (for K.H.P.), the Academy of Finland (A.S.; grants 265240 and 263278 for J.K.; grants 272376 and 266286 for K.H.P.), the Orion Foundation (for S.H.), the University of Helsinki, the European Research Council, and the Sigrid Jusélius Foundation (for A.S.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.H. performed and designed the research, collected and analyzed the data, and wrote the manuscript. J.B. performed and designed the research, analyzed the data, and wrote the manuscript. M.M. performed the transcriptomics data analysis. R.K., J.K., and A.R. participated in the revision of the work. M.O. wrote the epigenetic analyses. K.I. performed the epigenetic analyses. A.H., J.L., and N.L. participated in the imaging of the twins. K.V. and E.M. participated in analyzing adiponectin and leptin. A.S. designed the research, supervised the work, and participated in the revision and discussion of the results. K.H.P. designed the research, supervised the work, collected the data, and participated in the revision and discussion of the results. K.H.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.