Profiling Gene Transcription In Vivo Reveals Adipose Tissue as an Immediate Target of Tumor Necrosis Factor-α: Implications for Insulin Resistance

Male Wistar rats (Charles River) weighing 250–300 g were housed individually under controlled light (12/12 h) and temperature conditions and had free access to regular chow and water. All procedures were in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and approved by the Animal Subjects Committee of the University of California, San Diego.

Animals underwent surgery to implant an osmotic pump using standard sterile techniques. General anesthesia consisted of ketamine (50 mg/kg; Aveco, Fort Dodge, IA), xylazine (4.8 mg/kg; Lloyd, Shenandoah, IA), and acepromazine (1 mg/kg; Aveco) given intramuscularly. Through a small midline laparotomy, a mesenteric vein was isolated and catheterized (Micro-Renathane MRE-033, 0.033 OD and 0.014 ID; Braintree Scientific, Braintree, MA). An osmotic pump (Alzet; Alza, Palo Alto, CA) was connected to the catheter and placed inside the abdomen, and the abdominal wall and skin was sutured closed. The pumps delivered TNF-α or vehicle for 24 h (Alzet, Model 2001D; 34.6 mg/ml or 0.27 mg/h) or for 4 days (Alzet, Model 2ML1; 28.6 mg/ml or 0.27 mg/h). These infusion rates equated to ∼1 mg · kg⁻¹ · h⁻¹ for an average 275 kg per rat. TNF-α (Genzyme, Cambridge, MA) was dissolved in PBS containing 0.1% BSA.

Whole-body insulin sensitivity was assessed using the insulin tolerance test 24 h or 4 days following pump implantation. The rats were fasted for 5 h, and a basal blood sample (1 ml) was collected by nicking the tail-tip. The blood was immediately analyzed for glucose (5 ml) while the rest was centrifuged and the resultant plasma frozen for later determination of FFA, triglyceride, total cholesterol, ACRP30, and C-reactive protein. Insulin was then injected intraperitoneally at t = 0 (0.6 units/kg, Novolin R; Novo Nordisk, Copenhagen), and subsequent blood samples (5 ml) were collected at 15, 30, 60, and 90 min for glucose determination.

Blood glucose concentration was measured with a B Glucose Analyzer (HemoCue, Mission Viejo, CA). Plasma FFAs were determined using an Acyl-CoA oxidase-based colorimetric kit (NEFA C; Wako, Richmond, VA). Plasma triglyceride and total cholesterol levels were measured according to standard procedures. C-reactive protein was measured by the biochemistry laboratory at the University of California at San Diego Medical Center. Plasma ACRP30 was measured using a double-antibody radioimmunoassay kit (Mouse adiponectin; Linco, St. Charles, MO).

Total RNA was isolated from rat tissues, converted to biotin-labeled cRNA targets, hybridized to RG34A oligonucleotide arrays (Affymetrix, Santa Clara, CA), and scanned on Affymetrix scanners essentially as described previously (12). Each rat RG34A array contains oligonucleotide sequences representing ∼7,000 known genes and ∼1,400 expressed sequence tags (ESTs). We collected 12 data sets of steady-state mRNA levels in adipose tissue, liver, and skeletal muscle from rats infused with TNF-α for 1 or 4 days and from rats infused with saline for 1 or 4 days as matching controls. We used six rats for each experimental group; therefore, each data set included six independent scans for the specified experimental condition.

The average of P-calls (P-calls, according to Affymetrix algorithm) for adipose tissue, liver, and skeletal muscle arrays was 44.0 ± 0.9, 36.2 ± 0.9, and 30.5 ± 0.7% (mean ± SE, n = 24), respectively. We used A-calls and P-calls to define the expressed genes by requiring a minimum of four P-calls out of six measurements for each data set. An arbitrary value of 50 was allocated for genes whose expression levels were undetectable or scan readouts <50 to facilitate calculation of the fold changes compared with the control condition. Array measurements for all samples were normalized with arrays hybridized with cRNA targets prepared from the appropriate control rats by using the median of the hybridization signals of all genes with P-calls as a scaling factor.

P values were calculated for all genes by comparing the array measurements from saline-treated rats with those from TNF-α-treated rats, using the paired Student’s t test assuming unequal variances between the two groups. Fold changes were calculated by averaging the measurements of each data set and determining the ratio of the average values from the two data sets. The potential TNF-α-affected genes (known genes and ESTs) were determined by selecting those with a P value <0.05 and with an average of 1.5-fold or greater change in expression levels. We then performed hierarchical clustering (Fig. 2) to sort these genes according to their correlation with the control (saline infusion) versus TNF-α treatment, using Cluster and TreeView software (22). We assigned each of the TNF-α-affected known genes into functional categories according to the putative biological functions of the encoded protein, as determined by searches of public databases (Tables 3–5 and supplemental materials). The results of oligonucleotide microarrays were validated by semiquantitative RT-PCR analysis of mRNA levels of several known TNF-α-regulated genes (data not shown).

Fully differentiated 3T3-L1 adipocytes were incubated with 50 μmol/l [³H]oleic acid (oleic acid:BSA = 1:1) for 3 h at 37°C. Then the cells were washed exhaustively with PBS, and the radioactivity of the last wash was checked to be at the background level. The cells were incubated with fresh serum-free media containing 150 μmol/l BSA. At each indicated time point, an aliquot of media was taken and its associated radioactivity was determined by scintillation counting.

To determine whether TNF-α can acutely antagonize the actions of insulin in vivo, we first measured the plasma levels of several markers and/or potential mediators of insulin resistance after TNF-α treatment (Table 1). One day after TNF-α infusion, plasma concentrations of FFAs increased by 70% (P = 0.005, n = 6), with a concomitant 46% decrease in ACRP30 protein levels (P = 0.01, n = 6). Plasma glucose concentration was not changed 1 day after TNF-α treatment, and body weight and the levels of triglyceride, total cholesterol, and C-reactive protein were not altered.

Plasma FFA levels continued to increase by 150% 4 days after TNF-α infusion (P = 0.001, n = 6), whereas ACRP30 concentrations remained 51% that of the controls (P = 0.03, n = 6). In addition, the plasma concentration of triglyceride was increased 55% by a 4-day TNF-α infusion (P = 0.05, n = 6). No significant differences were observed in body weight, plasma levels of glucose, total cholesterol, or C-reactive protein 4 days after TNF-α treatment.

We next assessed the effect of TNF-α infusion on overall in vivo insulin sensitivity using the insulin tolerance test (Fig. 1). In rats infused with saline for 1 day, insulin injection caused a 17% decrease (P < 0.01, n = 5) in plasma glucose concentrations within 30 min and a 40% (P < 0.001, n = 5) and a 36% (P < 0.01, n = 5) decrease 60 and 90 min after insulin administration, respectively (Fig. 1A). In rats treated with TNF-α for 1 day, however, the drop in the plasma glucose levels in response to the injected insulin was significantly slower and smaller than that of control rats, indicating decreased insulin sensitivity (Fig. 1A). The hypoglycemic effects of insulin were further diminished 4 days after TNF-α infusion, indicating the development of significant insulin resistance (Fig. 1B), consistent with earlier reports (23). The kinetics of insulin’s glucose-lowering effects were not different in the 1-day and 4-day saline-infused groups, indicating unchanged insulin sensitivity. Thus, TNF-α infusion reduced systemic insulin sensitivity in a time-dependent manner, which is associated with a progressive elevation of plasma FFA concentrations and a significant drop in ACRP30 levels.

To determine whether gene transcription would be significantly altered during TNF-α treatment, we assessed the expression profiles of ∼7,000 known rat genes in major insulin-responsive tissues 1 and 4 days after TNF-α infusion. First, we filtered the normalized gene expression data to exclude genes showing minimal changes across the various groups of sample being analyzed, as described in research design and methods. Then, we performed hierarchical clustering to sort these genes by their correlation with the control versus TNF-α groups. Figure 2 shows that TNF-α evoked major and rapid changes in adipocyte gene expression and fewer changes in liver. There was only a moderate repressive effect on skeletal muscle gene expression. The TNF-α-affected genes form distinct clusters before and following the development of insulin resistance (Fig. 2), with clusters of genes in adipose tissue and the liver being most metabolically relevant (see below). Among them, the mRNA levels of 187 (2.7% of the 7,000 genes assayed) and 208 (3.0%) known adipocyte genes increased at least 1.5-fold (P < 0.05) after 1-day and 4-day TNF-α infusion, respectively. In parallel, the expression of 77 (1.1%) and 122 (1.7%) known adipocyte genes were repressed at least 1.5-fold (P < 0.05) by TNF-α after 1 and 4 days. In the liver, however, TNF-α induced 79 (1.1%) and 44 (0.6%) known genes and repressed 41 (0.6%) and 28 (0.4%) known genes at least 1.5-fold (P < 0.05) after 1- and 4-day infusions, respectively. In contrast, TNF-α infusion for up to 4 days did not significantly induce gene expression in skeletal muscle, but repressed the mRNA levels of 82 (1.2%) and 30 (0.4%) known genes at least 1.5-fold (P < 0.05) after 1- and 4-day infusions (Table 2).

Analysis of the tissue-specific responses to TNF-α infusion for up to 4 days revealed that TNF-α affected a unique subset of genes in each insulin-sensitive tissue, with only a small number of the 7,000 known genes tested regulated by TNF-α in the three different tissues (Fig. 3). Notably, the mRNA levels of 1.3% (91 known genes) of the adipocyte genes that are affected by 1-day TNF-α treatment remained altered after 4 days of TNF-α infusion. Noteworthy, only 0.3% (21 known genes) and 0.03% (2 known genes) of the 7,000 genes tested showed sustained changes in mRNA levels in liver and muscle, respectively, throughout the 4-day period of TNF-α infusion. These data demonstrate that TNF-α induced major and sustained changes in adipocyte gene expression and caused distinct early and late responses in liver and muscle, suggesting that the delayed responses in liver and muscle might result from effects secondary to TNF-α infusion. The entire gene lists as well as the full data sets for the 7,000 known genes (raw data) are available in an online appendix at http://diabetes.diabetesjournals.org.

To identify the potential correlation of changes in adipocyte gene expression with adipocyte energy metabolism and biological function, we further analyzed the TNF-α-affected adipocyte genes and assigned each gene into functional categories according to the biological functions of their encoded proteins. As shown in Table 3, TNF-α caused profound changes in adipocyte gene expression.

TNF-α repressed the mRNA levels of retinoid X receptor-γ (RXR-γ, −1.5-fold after 4 days) and PPAR-γ (−1.3-fold after 1 day), spot 14 (−1.6-fold after 1 day, −2.2-fold after 4 days), and CCAAT binding transcription factors NF1-A1 and NF1-X1 (−1.6-fold after 4 days), but induced nuclear factor (NF)-κB p105 (+1.3-fold after 1 day, +2.0-fold after 4 days), hypoxia inducible factor 1 (+2.6-fold after 1 day, +1.6-fold after 4 days), high-mobility group protein 2 (+1.6-fold after 1 day), and fos-related antigen (Fra, +1.5-fold after 1 day). While PPAR-γ, RXR-γ, and spot 14 (24) are essential for adipogenesis and lipogenesis, the accumulation of mRNAs encoding NF-κB, and Fra-1 inhibits adipocyte differentiation (25,26). Thus, TNF-α may negatively regulate adipocyte differentiation by reducing adipogenesis in vivo and restoring levels of preadipocyte transcription factors.

TNF-α repressed the expression of many genes involved in the uptake and storage of FFAs in adipocytes. Among them, the mRNA level of lipoprotein lipase was repressed by 1.4- and 1.9-fold (Fig. 4C) after 1 and 4 days of TNF-α infusion, respectively (Table 3). Since lipoprotein lipase is the major enzyme that hydrolyzes triglycerides in triglyceride-rich lipoproteins and initiates FFA uptake by adipose tissue, downregulation of lipoprotein lipase gene expression may result in an elevation of plasma triglyceride. The expression of many proteins involved in the utilization and storage of FFAs were also downregulated by TNF-α, such as genes encoding adipocyte fatty acid binding protein (−1.7-fold after 4 days), acyl CoA binding protein (−1.6-fold after 4 days), long-chain fatty acyl CoA synthase (−1.7-fold after 1 day, −2.3-fold after 4 days; Fig. 4D), diacylglycerol acyltransferase (−1.5-fold after 4 days), and perilipin A (−2-fold after 1 day) (Table 3). The overall changes in the mRNA levels of these proteins, while statistically significant, are relatively small, but their encoded proteins act in series in a common pathway of FFA metabolism and, therefore, the moderate changes of each protein aggregated together will have a significant effect, favoring fatty acid release while preventing fatty acid uptake by adipose tissue.

To ascertain whether TNF-α can directly stimulate FFA release from adipocytes, we treated 3T3-L1 adipocytes with TNF-α and determined the kinetics of TNF-α-stimulated FFA release. As shown in Fig. 5A, TNF-α did not acutely stimulate FFA release from adipocytes. Rather, TNF-α slightly increased FFA release after a lag period of 4 h (28%, P < 0.04), but resulted in a profound increase in FFA release after 8 h (2.4-fold, P < 0.001) and remained at 2.3-fold 24 h after TNF-α addition (P < 0.001, Fig. 5B). In contrast, isoproterenol (IPT), which rapidly activates hormone sensitive lipase, caused a 49% increase (P < 0.01) in FFA release as early as 10 min of incubation. The IPT-stimulated fatty acid release peaked at 30 min of incubation (3.0-fold, P < 0.04) and remained at 2.6-fold (P < 0.01), 2.2-fold (P < 0.001), and 1.7-fold (P < 0.01) that of control cells after 1, 2, and 4 h, respectively, of IPT treatment. The stimulatory effect of IPT diminished after 8 h of incubation (Figs. 5A and B), indicating a rapid, transient, and saturable nature of IPT-stimulated FFA release in adipocytes. Thus, the distinct time courses of TNF-α and IPT action suggest that different mechanisms are involved in stimulation of FFA release in fat cells.

In addition to affecting genes involved in FFA metabolism, TNF-α had profound effects on the expression of adipocyte genes involved in glucose uptake and utilization. One-day TNF-α infusion downregulated the mRNA levels of GLUT4 (−2-fold, Fig. 4E) and fatty acid synthase (−4.3-fold, Fig. 4B), as it did in 3T3-L1 adipocytes (12). In addition, since glucose provides the three-carbon backbone for triglyceride synthesis in adipose tissue, the lack of lipogenic substrates may also favor FFA release from adipocytes. Furthermore, oligonucleotide microarrays revealed that TNF-α had significant impact on the expression of many genes involved in energy metabolism, including glycolysis (Fig. 4F), the citric acid cycle, and the pentose phosphate pathway (Table 3). Taken together, the coordinate changes in gene expression induced by TNF-α may result in enhanced catabolism in adipose tissue and thus antagonize the anabolic effects of insulin.

The expression in adipocytes of several mRNAs whose overexpression has been reported in type 2 diabetes was also increased by TNF-α infusion, such as adrenomedullin, plasminogen activator inhibitor 1, plasminogen activator inhibitor 2, fibronectin, and ceruloplasmin. Thus, TNF-α may contribute to the induction of the molecular markers associated with obesity-linked type 2 diabetes. In addition, TNF-α upregulated the mRNA levels of several genes encoding adhesion molecules in adipose tissue (Table 3).

It has been increasingly recognized that adipose tissue is capable of producing a variety of chemokines and cytokines (27). Consistent with this notion, our microarray results indicate that many genes involved in inflammatory responses, as well as a number of chemokines, cytokines, and their receptors, and many proteins involved in cytokine production are strongly induced by TNF-α in adipose tissue. Examples of inflammatory mediators include 5-lipoxygenase activating protein (1.7-fold at 1 day, 1.9-fold at 4 days), prostacyclin synthase (1.5-fold at 1 day), phospholipase A2 precursor (1.7-fold at 4 days), and leukotriene A4 hydrolase (1.5-fold at 4 days) (online materials, http://diabetes.diabetesjournals.org). Genes involved in immune response include CXC chemokine LIX, CXC chemokine receptor, tumor growth factor (TGF)-β1, TNF-α converting enzyme, interleukin (IL)-1β converting enzyme, IL-1 receptor type II, and interferon-γ receptor (Table 3). Notably, many components of the complement pathway, including C1q, C3 (Fig. 4A), factor B, and properdin, were also induced by TNF-α infusion (Table 3). These secreted molecules can not only regulate adipocyte metabolism and function, but may contribute to the systemic immune response as well.

TNF-α induced rapid and significant changes in the expression of genes involved in energy metabolism and major physiological functions of the liver. As shown in Table 4, TNF-α repressed liver mRNA levels of several genes involved in glucose uptake and metabolism, such as GLUT2 (−1.8-fold after 1 and 4 days, Fig. 4K) and glycogen synthase (−1.7-fold after 1 day, −1.8-fold after 4 days; Fig. 4 L), and genes involved in fatty acid oxidation including carnitine palmitoyltransferase (CPT)-I (−1.6-fold after 1 day, −1.4-fold after 4 days), and CPT-II (−1.3-fold after 4 days). In addition, genes encoding transcription factors essential for hepatocyte function, such as D site albumin promoter binding protein (Table 4 and Fig. 4J), and genes involved in amino acid metabolism (online materials) are repressed by TNF-α. Meanwhile, TNF-α infusion induced many secreted proteins (Fig. 4G) and caused a significant increase in the mRNA levels of transcription factor SREBP/ADD1 (Fig. 4I), fatty acid translocase/CD36, as well as many genes involved in de novo cholesterol synthesis (Table 4).

Several genes involved in the de novo synthesis of fatty acids (fatty acid synthase, +1.8-fold, Fig. 4H) as well as genes involved in VLDL production (apolipoprotein B, +twofold) were induced by TNF-α, but only after 4 days of infusion, indicating that these effects of TNF-α may be secondary to its effects on adipose tissue or other peripheral tissues. Notably, the mRNA levels of enzymes involved in gluconeogenesis, such as phosphoenolpyruvate carboxykinase and pyruvate carboxylase, were not affected by TNF-α infusion up to 4 days. Thus, TNF-α- induced changes in hepatocyte gene expression appear to favor cholesterol and lipid synthesis and VLDL production, and reduce glucose uptake and conversion to glycogen in the liver.

In contrast to the effects in adipose tissue and the liver, TNF-α caused only moderate changes in gene expression in skeletal muscle (Table 5). We could not determine whether TNF-α represses the expression of essential insulin signaling molecules in muscle because the hybridization signals of the insulin receptor, IRS-1, 2, and 3, PI-3 kinase and AKT were too weak to be considered significant. On the other hand, the mRNA levels of many genes involved in the uptake and utilization of glucose and FFAs, such as GLUT4, hexokinase, glycogen phosphorylase, glucose-6-phosphatase, long-chain fatty acyl-CoA synthase, CPT-I, and CPT-II, were not affected by TNF-α infusion up to 4 days. Thus, TNF-α exhibits tissue-specific effects on gene transcription, and the decreased glucose uptake in skeletal muscle is likely due to covalent modifications that impair insulin signaling or other systemic mediators generated in response to TNF-α.

We report three major findings, which taken together describe how TNF-α alters energy metabolism in vivo as well as the tissue specificity of TNF-α actions. First, we demonstrated that TNF-α caused major changes in gene expression in adipose tissue and the liver, which may mediate, at least in part, the development of systemic insulin resistance. Second, we found that TNF-α did not cause changes in plasma glucose levels for up to 4 days of infusion. Rather, TNF-α administration for 1 day increased the plasma concentration of FFAs and reduced circulating levels of ACRP30, with significant insulin resistance and elevation of plasma triglyceride concentration occurring after 4 days of infusion. Third, we found that some of the major effects of TNF-α on gene transcription in liver occurred only after 4 days of TNF-α infusion, indicating that these changes are likely to be secondary.

The TNF-α-induced changes in adipocyte gene expression in vivo are broadly consistent with our previous observations in 3T3-L1 adipocytes (12) and favor FFA release. Indeed, TNF-α caused a profound increase in plasma levels of FFAs (Table 1) and stimulated lipolysis in cultured adipocytes with delayed kinetics compared with IPT (Fig. 5), indicating that TNF-α-induced changes in adipocyte gene expression have direct functional consequences in adipose tissue and on systemic energy homeostasis.

Since the balance between cellular triglyceride synthesis, FFA re-esterification, and triglyceride hydrolysis determines the amount of FFA released from adipocytes (28,29), aberrations of any of these cellular pathways may contribute to increased FFA production in adipose tissue. As shown in Table 3 and Figs. 4B-D, TNF-α significantly repressed many adipocyte-abundant genes, including lipoprotein lipase, long-chain fatty acyl-CoA synthetase, diacylglycerol acyltransferase, pyruvate carboxylase, transaldolase, and perilipin. These proteins act together in a common pathway of cellular triglyceride synthesis and storage, and the decreases in their expression levels when aggregated together will favor FFA release and inhibit FFA uptake and their conversion to triglyceride in adipocytes. Interestingly, a 4-day TNF-α infusion also downregulated the mRNA level of hormone-sensitive lipase (HSL, Table 3). However, the HSL-mediated triglyceride hydrolysis is mainly regulated by the phosphorylation-state of HSL rather than HSL protein mass and, thus, the repression of HSL mRNA levels may not contribute substantially to the amount of FFA released from fat cells. Alternatively, TNF-α may stimulate FFA release by inducing other members of the triglyceride hydrolase family. Table 3 shows that TNF-α increased the mRNA levels of lysosomal acid lipase (LAL) 1.7- and 2.2-fold after 1-day and 4-day infusions, respectively. LAL is a lysosomal enzyme that hydrolyses cholesteryl esters and triglycerides and, thus, may regulate cellular triglyceride content. Another attractive candidate is the triacylglycerol hydrolase (TGH) originally described by Lehner and colleagues (30–32). TGH was initially identified as a microsomal lipase that mobilizes cytoplasmic triglyceride stores in hepatocytes, and it is likely to play a significant role in adipocyte metabolism as well. However, we did not locate the TGH gene on the rat RG34A chip, and in this study we could not determine its expression level in fat cells and its regulation by TNF-α. Nevertheless, these triglyceride hydrolases may contribute, at least partially, to the regulation of FFA release from adipocytes, and further investigation is required to explore the role of the triglyceride hydrolase family in adipocyte biology and metabolism.

It is widely accepted that elevated plasma concentrations of FFAs can cause insulin resistance by impairing the ability of insulin to stimulate skeletal muscle glucose uptake and to inhibit hepatic glucose production. Although the mechanisms for FFA-induced insulin resistance are incompletely understood, they involve decreased function and/or activity of key insulin signaling molecules in muscle (33) and liver (34). Thus, elevated plasma concentrations of FFAs appear to mediate some of the delayed systemic effects of TNF-α to decrease insulin sensitivity.

Additionally, TNF-α may induce insulin resistance, in part, by decreasing the circulating levels of ACRP30. ACRP30 is a hormone synthesized exclusively in adipose tissue, and it is immediately downregulated by TNF-α infusion (Table 1). The mechanisms by which TNF-α represses plasma levels of ACRP30 may include inhibition of ACRP30 gene transcription, reduction of ACRP30 mRNA half-life, repression of ACRP30 protein synthesis, and enhancing ACRP30 protein degradation and clearance. Although ACRP30 mRNA was not included in the ∼7,000 known genes on the RG34A array, we (12) and others (13,14) have previously demonstrated that TNF-α represses the steady-state levels of ACRP30 mRNA in cultured adipocytes in a dose- and time-dependent manner. Moreover, ACRP30 mRNA and protein levels are significantly reduced in patients with type 2 diabetes. Thus, TNF-α-mediated downregulation of ACRP30 mRNA is likely to contribute, at least partially, to the decrease in circulating levels of ACRP30 in vivo as well. Although the signaling pathway, as well as the mechanisms of ACRP30 action remain uncertain, ACRP30 has been shown to sensitize liver to the actions of insulin (15,16) and reduce plasma FFA concentrations (35). Therefore, TNF-α may further antagonize the actions of insulin by repressing the plasma levels of ACRP30.

Leptin is another adipocyte-derived hormone that regulates systemic energy metabolism, food intake and adiposity, and insulin responsiveness (36–38). Several groups showed that TNF-α increased circulating levels of leptin (39,40) in both experimental animals and human subjects, and that TNF-α stimulation of leptin secretion in 3T3-L1 adipocytes is independent of protein synthesis (41). Our microarray data revealed that the mRNA levels of leptin were not altered in the adipose tissues from rats infused with TNF-α for up to 4 days, consistent with a previous report (41), supporting the notion that post-translational regulation of leptin production by TNF-α is likely to be involved. Further elucidation of the molecular mechanisms by which TNF-α regulates the secretion of adipocyte-derived hormones may advance our understanding of how TNF-α causes systemic insulin resistance and thereby facilitate the development of therapeutic interventions that will block these pathways in vivo.

TNF-α also caused distinct immediate and delayed changes in liver gene expression favoring FFA, triglyceride, and cholesterol synthesis (Table 3). At present, we could not determine whether TNF-α directly induces or represses the transcription of these genes in the liver, because the changes in gene expression could result from increased plasma FFA levels and/or decreased plasma ACRP30 protein content, or some other factors regulated by TNF-α.

As revealed by the insulin tolerance test, 1-day TNF-α treatment decreased insulin sensitivity, whereas TNF-α infusion for 4 days caused a significant decrease in overall in vivo insulin sensitivity (Fig. 1). The impaired glucose-lowering effect of injected insulin can be due to decreased insulin-stimulated glucose disposal, impaired inhibition of hepatic glucose production, or both. Since skeletal muscle is responsible for ∼80% of overall in vivo insulin-mediated glucose uptake, a significant defect in in vivo glucose uptake indicates the development of insulin resistance in muscle. Interestingly, infusion of TNF-α for 4 days did not repress muscle genes such as GLUT4 and hexokinase, which are essential for glucose uptake and utilization. Similarly, many liver genes involved in gluconeogenesis, including phosphoenolpyruvate carboxykinase and pyruvate carboxylase, were not affected by TNF-α infusion. Thus, our data suggest that impaired activity of the insulin signaling transduction pathways, rather than changes in gene expression, may play a significant role in the development of insulin resistance in muscle and liver after 4 days of TNF-α infusion. Furthermore, the defects in insulin signaling following exposure to TNF-α may result from the time-dependent mediators including FFA, ACRP30, or some other unidentified factors regulated by TNF-α.

In summary, we demonstrated that TNF-α induces rapid and major changes in adipocyte gene expression and that it is sufficient to trigger the development of systemic insulin resistance in vivo. Given that TNF-α is overexpressed in adipose tissues of obese individuals, either TNF-α-induced defects in insulin signal transduction in adipocytes or TNF-α-dependent changes in adipocyte gene expression could be essential to elicit the insulin resistance phenotype in humans. Inhibition of TNF-α-induced changes in gene expression in cultured adipocytes, using retroviral delivery of a dominant-negative inhibitor of NF-κB, could distinguish between these alternatives (12). Importantly, the obligatory role of NF-κB in TNF-α induction of insulin resistance could not have been predicted from various studies using gene knockout mice (9,10). Taken together, our study provides information to identify the direct target tissue and molecular mediators of TNF-α and provides a molecular basis for testing pharmacological interventions of insulin resistance in humans.

This work was supported in part by National Institutes of Health Grant R37DK47618 (to H.F.L.) and DK33651 (to J.M.O. and P.D.G.M.) and by grants from Affymetrix Inc., Millennium Pharmaceuticals Inc., and Bristol-Myers Squibb Company to the Genome Center at the Whitehead Institute. H.R. is supported by a Postdoctoral Fellowship for Physician Scientists from the Howard Hughes Medical Institute.

We thank Dr. Nir Hacohen (Whitehead Institute, Cambridge, MA) and Dr. Luk Van Parijs (MIT, Cambridge, MA) for reagents, helpful comments, and suggestions. We thank Dr. Fran Lewitter for help on computational analysis.