Downregulation of Electron Transport Chain Genes in Visceral Adipose Tissue in Type 2 Diabetes Independent of Obesity and Possibly Involving Tumor Necrosis Factor-α

The Affymetrix study group (dataset 1) comprised nonobese subjects, obese otherwise healthy subjects, and obese subjects with newly diagnosed type 2 diabetes. Nonobese subjects (n = 8) were aged 47 ± 10 years, with BMI 24 ± 3 kg/m², waist circumference 86 ± 9 cm, calculated body fat (22) 31 ± 6%, homeostasis model assessment of insulin resistance (HOMA_IR) (23) 1.7 ± 0.6, fasting plasma glucose 5.0 ± 0.2 mmol/l, serum insulin 8.0 ± 2.6 mU/l, plasma cholesterol 5.8 ± 1.4 mmol/l, plasma HDL cholesterol 1.3 ± 0.4 mmol/l, and plasma triglyceride levels 1.7 ± 1.0 mmol/l. Obese otherwise healthy subjects (n = 8) were aged 35 ± 9 years, with BMI 44 ± 2 kg/m², waist circumference 123 ± 9 cm, calculated body fat 60 ± 4%, HOMA_IR 3.9 ± 1.6, fasting plasma glucose 5.4 ± 0.3 mmol/l, serum insulin 16.0 ± 5.8 mU/l, plasma cholesterol 5.0 ± 0.5 mmol/l, plasma HDL cholesterol 1.3 ± 0.4 mmol/l, and plasma triglyceride levels 1.6 ± 0.6 mmol/l. The obese subjects with newly diagnosed type 2 diabetes (n = 8) were aged 38 ± 12 years, with BMI 43 ± 3 kg/m², waist circumference 128 ± 9 cm, calculated body fat 58 ± 5%, HOMA_IR 6.3 ± 4.2, fasting plasma glucose 6.0 ± 0.7 mmol/l, serum insulin 23.4 ± 14.6 mU/l, plasma cholesterol 5.4 ± 0.8 mmol/l, plasma HDL cholesterol 1.2 ± 0.2 mmol/l, and plasma triglyceride levels 2.1 ± 1.2 mmol/l.

The nonobese subjects were scheduled for gall bladder surgery and were otherwise healthy. The obese subjects were scheduled for gastric banding because of obesity. No subject had evidence of general inflammation, i.e., elevated C-reactive protein, before the operation. Before surgery all obese subjects were given an oral glucose load. All healthy obese subjects had normal glucose tolerance, and all obese diabetic subjects had pathological oral glucose tolerance (plasma glucose at 2 h of <7.7 mmol/l and >11.0 mmol/l, respectively). All of these women were free of medication.

It was only possible to measure a limited number of individual mRNAs in dataset 1. Visceral mRNA expression of TNF-α, estrogen-related receptor-α (ESRRA), and GA-binding transcription factor α-subunit (GABPA) and β-subunit (GABPB) were therefore quantified in dataset 2, which comprised nonobese subjects and obese subjects with type 2 diabetes of various duration. The nonobese subjects (n = 10 women) were aged 43 ± 12 years, with BMI 23 ± 2 kg/m², waist circumference 76 ± 7 cm, calculated body fat 29 ± 8%, HOMA_IR 1.5 ± 0.7, fasting plasma glucose 4.8 ± 0.5 mmol/l, and serum insulin 7.6 ± 2.3 mU/l. Obese subjects with type 2 diabetes of various duration (n = 5 women and 5 men) were aged 42 ± 6 years, with BMI 41 ± 4 kg/m², waist circumference 122 ± 22 cm, calculated body fat 46 ± 12%, HOMA_IR 5.1 ± 4.7, fasting plasma glucose 5.8 ± 1.4 mmol/l, and serum insulin 17.9 ± 10.3 mU/l. Plasma lipids were not available for dataset 2. The subjects were recruited as for dataset 1, except that the type 2 diabetes diagnosis was of various duration and patients were on antidiabetic treatment.

The subjects were operated in the morning after an overnight fast. In the morning before surgery, a venous blood sample was obtained for determination of fasting circulating levels of insulin, lipids, and glucose by the routine chemistry laboratory of the hospital. Only saline was given intravenously until adipose tissue was removed, which was done at the beginning of surgery. From each subject a piece of abdominal subcutaneous fat (∼1 g) from the surgical incision and a piece of visceral adipose tissue (∼1 g) were obtained during endoscopic abdominal surgery.

In experimental studies, abdominal subcutaneous or omental adipose tissue samples were obtained from healthy women who were not selected on the basis of BMI or age and who were undergoing abdominal liposuction for cosmetic reasons or gastric banding because of obesity. Adipose tissue was brought to the laboratory in saline. Pieces (1–2 × 500 mg) were immediately frozen in liquid nitrogen and kept at −70°C. The committee on ethics of the hospital approved the study. It was explained in detail to each participant, and her/his consent was obtained.

Abdominal subcutaneous adipose tissue pieces obtained during liposuction were collagenase treated as previously described for isolation of adipocytes (24). Preadipocytes were isolated from the stroma vascular fraction of adipose tissue and differentiated into adipocytes under serum-free conditions exactly as described (25). The cells were seeded out in 12-well plates at a density of 50,000 cells/cm² and incubated for 48 h with TNF-α (100 ng/ml; Sigma, St. Louis, MO) before RNA isolation. Control cells were incubated with medium alone. Previous studies have shown that the maximum effects of TNF-α on intracellular signaling and lipolysis in this type of fat cell preparation are obtained with 100 ng/ml of the cytokine (26). Furthermore, at the time of study, the cell culture was essentially free from contamination cells, such as macrophages, endothelial cells, and fibroblasts, according to methodological experiments.

Human preadipocytes were obtained from the subcutaneous region (n = 3) and from the omental region (n = 1) and cultured in the absence or presence of TNF-α as described above. Thereafter, the medium was removed and fatty acid oxidation determined exactly as described before (27). In brief, the cells were incubated for 3 h in Dulbecco’s modified Eagle’s medium without glucose, 50 mmol/l HEPES, pH 7.8, 1% fatty acid–free BSA, 2 mmol/l l-carnitine, 50 μmol/l palmitate, and 118 nmol/l [¹⁴C]palmitate (850 μCi/μmol; Amersham Biosciences, Uppsala, Sweden). Medium was transferred to a Carbosorb E flask (PerkinElmer Life Sciences, Courtaboeuf, France). ¹⁴CO₂ was liberated by acidification with 5 mol/l HCl and collected overnight on Carbosorb. ¹⁴CO₂ was measured by scintillation counting. The acid-soluble fraction of the medium containing ¹⁴C-labeled oxidation metabolites was measured by scintillation counting after 1-butanol extraction of palmitate. Data are expressed as ¹⁴CO₂ in Carbosorb plus the acid-soluble fraction and referred to as total palmitate oxidation. As a control, fatty acid oxidation was either inhibited with 50 μmol/l etomoxir (from Wolfgang Langhans, Swiss Federal Institute of Technology, Zurich, Switzerland) or stimulated with 10 μmol/l m-chlorocarbonylcyanide phenylhydrazone (Sigma) during the 3-h incubation.

Total RNA was extracted from adipose tissue, isolated adipocytes, and differentiated preadipocytes, using the RNeasy mini kit (Qiagen, Hilden, Germany). RNA samples for real-time quantitative PCR were treated with RNase-free DNase (Qiagen, Hilden, Germany). The RNA concentration was determined by spectrophotometer. High-quality RNA was confirmed, using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).

We used 5 μg of adipose tissue total RNA per subject in a standard protocol from Affymetrix (Santa Clara, CA) to obtain biotinylated cRNA. In vitro transcription reaction was performed using a BioArray high-yield RNA transcript labeling kit (Enzo Diagnostics, Farmingdale, NY). Labeled cRNA was purified with RNeasy Mini Kit spin columns (Qiagen, Valencia, CA), quantified spectrophotometrically, and fragmented in buffer according to the Affymetrix protocol. Test-3 arrays (Affymetrix Baracaldo, Vizcaya, Spain) were hybridized to check the cRNA integrity. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) 3′-to-5′ probeset hybridization ratios were between 3 and 6, with random variation between the samples in the six adipose tissue groups. There were three outliers with GAPDH 3′-to-5′ ratios between 7 and 8; one control, one healthy obese, and one type 2 diabetes sample. This material was next hybridized to the U133A and U133B arrays from Affymetrix.

For dataset 1, 500 ng RNA from each adipose tissue sample was reverse transcribed to cDNA, using TaqMan reverse transcription reagents (N8080234; Applied Biosystems) and random hexamer primers. Only six nonobese control subjects from dataset 1 were used in these experiments because of a lack of RNA. Subsequently, specific mRNAs were quantified, using TaqMan gene expression assays (Applied Biosystems) for COX5B (Hs00426948_m1), NDUFB4 (Hs00853558_g1), DAP13 (Hs00276572_m1), ATP5C1 (Hs00270906_m1), ATP5F1 (Hs00829774_s1), NDUFB2 (Hs00190006_m1), SDHC (Hs00818427_m1), COX15 (Hs00189238_m1), NDUFC1 (Hs00159587_m1), SLC25A4 (Hs00154037_m1), 18S rRNA (4308329), GAPDH (4310884E), LDL receptor–related protein 10 (LRP10) (Hs 00204094_ml), TNF-α (Hs00174128_m1), peroxisome proliferator–activated receptor-γ (PPAR-γ) coactivator 1 (PGC1)-α (Hs00173304_m1) as well as PGC1β (Hs00370186_m1), and Taqman Universal Master Mix (4318157; Applied Biosystems) according to the manufacturer’s instructions. Expression of target gene was normalized to the GAPDH, 18S, and LRP10 genes, using the geNorm method (available online at http://www.gene-quantification.de/hkg.html#genorm).

For dataset 2, 1 μg RNA was reverse transcribed to cDNA using the Omniscript RT kit (Qiagen, Hilden, Germany) and random hexamer primers (Invitrogen, Tåstrup, Denmark). Subsequently, specific mRNAs were quantified using TaqMan gene expression assays (Applied Biosystems) for ESRRA (Hs00607062_gH), GABPA (Hs00745591_s1), GABPB (Hs00242573_m1), TNF-α (Hs00174128_m1), 18S rRNA (4308329), and Taqman Universal Master Mix (4318157) according to the manufacturer’s instructions. Expression of the target genes were normalized to the 18S internal control, using the formula 2^{(Ct target gene calibrator –Ct target gene sample)}/2^{(Ct 18S calibrator –Ct 18S sample)}, where the calibrator is a random sample and Ct is threshold cycle.

From each RNA sample, 1 μg was reverse transcribed to cDNA, using the Omniscript RT kit (Qiagen, Hilden, Germany) and oligo dT primers (Invitrogen). Specific mRNAs were quantified, using TaqMan gene expression assays (Applied Biosystems) for ATP5F1 (Hs01076982_g1), COX5B (Hs00426950_g1), DAP13 (Hs00984333_m1), NDUFB4 (Hs01071485_gH), NDUFB9 (Hs00601381_mH), ESRRA (Hs00607062_gH), GABPA (Hs00745591_s1), GABPB (Hs00242573_m1), PGC1α (Hs00173304_m1), and PGC1β (Hs00370186_m1) as well as Taqman Universal Master Mix (4318157; Applied Biosystems), according to the manufacturer’s instructions. GAPDH was quantified, using SYBR Green–based real-time quantitative PCR with primers 5′-CACATGGCCTCCAAGGAGTAAG-3′ and 5′-CCAGCAGTGAGGGTCTCTCT-3′. Dissociation curves and agarose gel electrophoresis were used to check for a single product. The confirmation and experimental real-time quantitative PCR studies were performed in different laboratories, which explain the differences in PCR reagents. Expression of target gene was normalized to the GAPDH internal control, using the formula 2^{(Ct target gene calibrator –Ct target gene sample)}/2^{(Ct internal control calibrator –Ct internal control sample)}.

After probing and scanning the 48 chip sets, using the standard protocols from Affymetrix, signal intensities were calculated in Microarray Analysis Suite 5.0 (Affymetrix, Santa Clara, CA) with default parameter settings for the statistical algorithm. Using the all-probeset option, average intensities in each sample were scaled to a target signal of 250, enabling multiple array comparisons. The scaling factor was between 1.69 and 5.66 (median 3.23). Distinct algorithms were used to evaluate (i.e., present or absent) absolute calls for each transcript.

Differences in expression of individual genes between nonobese, obese healthy, and type 2 diabetic women, respectively, were analyzed using significance analysis of microarrays (SAM) (28), which adjusts for the multiple comparisons caused by the presence of thousands of genes on the microarrays. Due to limitations in the size of datasets that SAM can handle, chips U133A and U133B were analyzed separately. Furthermore, to exclude noise from genes represented by low-quality probesets or not expressed in adipose tissue, we included in the SAM only genes displaying significant expression on more than six chips according to the criteria in Microarray Analysis Suite 5.0 in at least one of the analyzed groups. In each analysis, the false-discovery rate was set to 5%. Although this threshold will produce some false-positive genes, we used it because we planned to use the results from SAM for pathway analysis. In pathway analysis, it has been shown to be relevant to include information about genes that are regulated by type 2 diabetes, although this regulation may not be significant at the single-gene level (16,17). A total of 1,000 permutations were used in the numerical analysis, which led to highly reproducible results of the analysis.

Changes in expression of biological pathways in visceral fat between nonobese and type 2 diabetic women were analyzed using MAPPFinder version 2.0 and gene set enrichment analysis (GSEA) (29,30). The microarray genes were assigned to gene ontology terms and GenMAPP pathways (31,32). Subsequently, MAPPFinder 2.0 was used to rank pathways and terms based on differences in gene expression between nonobese and type 2 diabetic women (29). Briefly, MAPPFinder assigns to each analyzed pathway, or gene set, a z score that is based on the percentage of genes in each pathway or set that meets a user-defined criterion for change in expression. Criterion for change was significant difference in expression according to SAM. Up- and downregulated genes were analyzed separately. To obtain a more comprehensive pathway analysis, we also used GSEA (available online at www.broad.mit.edu/GSEA), including information about whether gene expression was present or absent (30).

Data are the means ± SD. For comparison of data between three or more groups, we used ANOVA. Otherwise, Student’s t test was used for analysis of clinical phenotypes as well as analysis of expression of individual genes. When necessary, the analyzed phenotype was log-transformed to become normally distributed. Unpaired t tests were used for clinical samples, and paired t tests were used for experimental samples. One-sided analysis was used for confirmation. Otherwise, we used two-sided tests.

In study sample 1, the difference in age between the lean and the two obese groups was of borderline significance (P = 0.064). The healthy obese and type 2 diabetic women had similar mean BMI (43–44 kg/m²), but the type 2 diabetic subjects were more insulin resistant (i.e., HOMA_IR 6.3 ± 4.2 vs. 3.9 ± 1.6 in healthy obese subjects). The difference in HOMA_IR between the three groups was significant (P = 0.01). There was no significant difference in waist circumference between type 2 diabetic and healthy obese women, indicating that there was no major difference in fat distribution. All groups had mean fasting plasma glucose within the normal range, which makes major dearrangements in glucose homeostasis among the type 2 diabetic women unlikely.

In dataset 2, the nonobese subjects were older and age matched to the obese type 2 diabetic subjects. Furthermore, the obese type 2 diabetic subset included five (50%) men. Otherwise, the phenotypes, and phenotypic differences between type 2 diabetic and nonobese subjects, were similar to dataset 1.

The overall number of expressed probesets was similar in both fat depots and in all three clinical groups. The number of expressed probesets per group, defined as probesets displaying significant expression on more than six of eight chips according to Microarray Analysis Suite 5.0, ranged between 10,489 and 12,547. Next, gene expression was compared between nonobese, obese, and type 2 diabetic women. Subcutaneous and visceral adipose tissue samples were analyzed separately. In each comparison, a larger number of genes displayed significant differences in expression between groups in visceral than in subcutaneous adipose tissue (Table 1). The largest number of differentially expressed genes were observed in the nonobese versus type 2 diabetic groups (1,072 and 2,133 genes in subcutaneous and visceral adipose tissue, respectively).

The large number of genes in visceral adipose tissue differing in expression between nonobese and type 2 diabetic women encouraged subsequent analysis and ranking of pathways differing in expression between these groups, using MAPPFinder and GSEA. In the analysis of genes downregulated in type 2 diabetic subjects, GenMAPP pathway ribosomal proteins, electron transport, and translation factors obtained the highest score (Table 2). In gene ontology, mitochondrion and ribosomal terms also obtained high scores (Table 2). Only one pathway was upregulated in type 2 diabetes (i.e., adjusted P value <0.05): mRNA processing reactome. We next performed GSEA on the same dataset. Two pathways regulated by type 2 diabetes overlapped between both analyses: electron transport chain and Krebs–tricarboxylic acid cycle (results not shown). These pathways obtained marginally significant nominal P values in GSEA (P = 0.10). The other MAPPFinder pathways did not overlap with the results in GSEA.

In subsequent analysis we focused on the electron transport chain. In visceral adipose tissue of type 2 diabetic versus nonobese women, 28 of 91 genes in this pathway were downregulated according to SAM (Table 3). No gene was upregulated. No electron transport chain gene was downregulated in visceral fat of healthy obese women. Nine genes were downregulated in subcutaneous fat of type 2 diabetic women versus nonobese control subjects (results not shown). Only three of these genes over-lapped with those regulated in visceral fat (Table 3). A total of 6 of the electron transport chain genes downregulated in visceral fat of type 2 diabetic women overlapped with the genes in the oxidative phosphorylation–coregulated set previously reported by Mootha et al. (16), and 11 genes overlapped with the oxidative phosphorylation genes shown by Patti et al. (17) to be downregulated in muscle of insulin-resistant subjects.

We quantified expression of five electron transport genes randomly selected among those downregulated in type 2 diabetic versus control subjects according to SAM, using real-time quantitative PCR, and we quantified five electron transport genes with unchanged expression on the microarrays. The results were consistent between microarrays and real-time quantitative PCR (Table 4). The genes with unchanged expression in type 2 diabetes on the microarrays displayed a mean fold change of 1. The genes downregulated in type 2 diabetes displayed a fold change of <0.75–0.90. Two genes were significantly downregulated (NDUFB4, ATP5C1), two displayed marginally significant P values (0.06 and 0.10; COX5B, ATP5F1), and one gene was not significantly downregulated (DAP13). Of five tested electron transport genes, four genes (ATP5F1, COX5B, NDUFB4, and NDUFB9) were expressed at similar levels in isolated adipocytes and in adipose tissue pieces (Fig. 1A). The level of DAP13 mRNA in isolated adipocytes was 15% of that in adipose tissue fragments (Fig. 1A).

With neither the arrays nor real-time quantitative PCR did we observe in visceral fat of type 2 diabetic women a significant downregulation of the transcription factors and coregulators nuclear respiratory factor 1 (NRF1), PGC1α, PGC1β, ESRRA, GABPA, and GABPB, which are implicated in regulation of electron transport chain gene expression (17,33) (Table 4). Surprisingly, expression of PGC1β, GABPA, and GABPB was increased in type 2 diabetic subjects (Table 4).

Treatment of differentiated human preadipocytes with TNF-α downregulated the expression of all of the five tested electron transport chain genes (ATP5F1, COX5B, NDUFB4, NDUFB9, and DAP13) by 20–40% (P < 0.05) (Fig. 1B). PGC1β, ESRRA, GABPA, and GABPB, which have been implicated in regulation of electron transport chain gene expression, were not downregulated by TNF-α, according to real-time quantitative PCR (Fig. 1C). PGC1α was not detected with real-time quantitative PCR in preadipocytes. There was no effect of TNF-α on the expression of the reporter gene GAPDH (mean Ct for GAPDH 20.5 and 20.6 in untreated and treated samples, respectively).

Three fatty acid oxidation experiments with subcutaneous preadipocytes and one with omental preadipocytes were performed. Because results were the same, data were combined and presented in Fig. 2. As expected, etomoxir significantly inhibited and m-chlorocarbonylcyanide phenylhydrazone significantly stimulated basal palmitate oxidation. TNF-α at a maximum effective concentration inhibited basal palmitate oxidation by 30% (P = 0.017).

TNF-α mRNA was expressed at significantly higher levels in visceral fat than in subcutaneous fat (3.4 ± 1.9 vs. 1.8 ± 1.0 arbitrary units [AU], P = 0.0002). In visceral fat the mean expression in nonobese, healthy obese, and type 2 diabetic women were 3.0, 4.5, and 3.6 AU, respectively. In subcutaneous fat the corresponding mean expression levels were 1.7, 1.9, and 2.5 AU, respectively. In dataset 2, TNF-α expression in visceral fat was significantly higher in the type 2 diabetic subjects compared with nonobese control subjects (1.2 vs. 0.4 AU, P = 0.04).

In this study, we demonstrated that type 2 diabetes is associated with a major and region-specific decreased expression in visceral adipose tissue of genes belonging to the mitochondria electron transport chain. Furthermore, the expression of electron transport chain genes was markedly downregulated by TNF-α treatment of preadipocytes obtained from healthy subjects.

In our samples, individual electron transport genes were downregulated ∼20% in visceral fat from type 2 diabetic women compared with nonobese control subjects. Furthermore, our analysis was limited to gene expression. Because of the laparoscopic procedure, it was not possible to obtain sufficient additional adipose tissue for direct studies of metabolic pathways, such as oxygen consumption. However, several lines of evidence support the notion that the small differences in expression of electron transport genes observed in this study are biologically relevant. First, human type 2 diabetes has been associated with defects in oxidative phosphorylation gene expression and fatty acid oxidation, although up to now this has primarily been discussed in the context of insulin resistance in liver and muscle (16,17,34,35). Second, in our study the downregulation of electron transport chain genes was specific for type 2 diabetes and visceral fat. No clear pattern was observed among healthy obese women or in subcutaneous fat. The organ-specificity of downregulated expression of electron transport genes is further supported by the observation that PGC1α-responsive genes for mitochondria oxidative phosphorylation are upregulated in liver of patients with type 2 diabetes (36).

The gene expression profile of adipose tissue in obese humans has been extensively studied. However, the impact of type 2 diabetes has not been examined, and we are the first to report a comparison of gene expression profiles in human subcutaneous and visceral fat of nonobese, obese nondiabetic, and type 2 diabetic subjects. A previous published gene expression profiling on human visceral adipose tissue described the prominent impact of morbid obesity on glycolysis enzymes, oxysterol biosynthesis and signaling, ATP-binding transporters, and solute carriers (37). The absence of a clear impact on electron transport chain genes in that study supports downregulated expression of these genes being directly associated with type 2 diabetes, rather than obesity in general. In mice, weight gain has been associated with downregulation of oxidative phosphorylation genes in visceral fat (38). To what extent this precedes development of type 2 diabetes in mice is unknown.

The importance of the PPAR-γ transcription complex for mitochondria oxidative activity is established. The PPAR-γ coactivators PGC1α and -β are important regulators of mitochondria gene expression and are implicated in the orchestration of the decreased mitochondria oxidative activity and downregulation of oxidative phosphorylation genes observed in muscle of insulin-resistant humans (16,17,35). In addition, thiazolidinediones induce mitochondria biogenesis and upregulate β-oxidation and electron transport genes in subcutaneous fat of mice and type 2 diabetic humans (19,20). Surprisingly, genes that are implicated in the regulation of oxidative phosphorylation gene expression in muscle (16,17,33) were either not significantly altered in our study (PGC1α, NRF1A, and ESRRA) or displayed increased expression in adipose tissue of type 2 diabetic subjects (PGC1β, GABPA, and GABPB). To our knowledge, the expression of the latter genes in human visceral fat has not been investigated before. Increased expression may hypothetically represent an organ-specific mechanism to counteract type 2 diabetes.

As mentioned previously, TNF-α plays an important role in the development of insulin resistance and adipose inflammation. Because we observed no clear correlation among PGC1α and PGC1β expression, obesity, and type 2 diabetes in human visceral fat, we investigated the possible role of TNF-α for downregulation of oxidative phosphorylation genes in fat cells. Methodological studies revealed that four of five examined genes were expressed in the same order of magnitude in isolated fat cells as in adipose tissue pieces (i.e., adipocytes plus stroma cells). It has been reported that obese humans express twofold more TNF-α mRNA and protein in fat tissue relative to the nonobese control subjects (12,39). In our samples from dataset 1 that were used for microarray analysis, TNF-α expression was modestly upregulated (20%) in visceral fat from type 2 diabetic women compared with nonobese control subjects. This difference was not significant. Because not all nonobese control subjects were used in this analysis due to lack of RNA, we repeated the analysis in dataset 2. In these samples, TNF-α expression was significantly upregulated by 180% in type 2 diabetic subjects. In our experimental system, five electron transport genes that were downregulated by type 2 diabetes were also downregulated in differentiated preadipocytes by treatment with TNF-α. Animal and human models have indicated that TNF-α plays an important role in mediating the insulin resistance of obesity through overexpression in fat tissue (12,39,40). This effect was originally thought to be caused by the inflammatory and lipolytic effects of the cytokine. This study suggests an additional role, namely inhibiting oxidative phosphorylation in visceral adipose tissue. A role of TNF-α in regulating electron transport chain gene expression and consequently fatty acid oxidation in human adipocytes is further demonstrated by our findings that TNF-α can inhibit basal fatty acid oxidation in differentiated human preadipocytes by at least 30%.

In conclusion, this study sheds new light on the importance of visceral adipose tissue for development of type 2 diabetes. Oxidative phosphorylation gene pathways, in particular the electron transport chain, are the most downregulated genes in this adipose region in type 2 diabetes, which at least in part involves local TNF-α action and is independent of obesity. The changes in oxidative phosphorylation pathways are in part specific for the adipose organ, compared with muscle and liver, and might be a factor behind insulin resistance in visceral fat of type 2 diabetic subjects.

This study was supported by grants from the Swedish Research Council, the Swedish Diabetes Association, the Novo Nordic Foundation, the Swedish Heart and Lung Foundation, the European Foundation for the Study of Diabetes, the European Commission (HEPADIP [Hepatic and Adipose Tissue in the Metabolic Syndrome] Contract LSHM-CT-2005-018734), the King Gustaf V and Queen Victoria Foundation, the Karolinska Institutet, Tore Nilsson’s Foundation, Thuring’s Foundation, and AFA Forsakring.

The skilful technical assistance of nurses Britt-Marie Leijonhufvud and Katarina Hertel is greatly appreciated.