Tumor Necrosis Factor-α Suppresses Adipocyte-Specific Genes and Activates Expression of Preadipocyte Genes in 3T3-L1 Adipocytes: Nuclear Factor-κB Activation by TNF-α Is Obligatory

3T3-L1 cells were purchased from American Type Culture Collection (Rockville, MD), maintained as fibroblasts, and differentiated into adipocytes, as previously described (25). TNF-α and insulin treatments were performed as described in figure legends.

The retroviral expression vector pMSCV-IRES-GFP (green fluorescent protein, pMIG), as well as the construct pMSCV-IκBα__DN-IRES-GFP (pIκBα-DN) encoding the dominant negative IκBα (a nondegradable mutant of IκBα) and GFP were gifts from L. Van Parijs (Massachusetts Institute of Technology, Cambridge, MA). To generate retroviruses, Phoenix packaging cells were transfected with 8 μg pMIG or pIκBα-DN plasmid cDNA using Polyfect (Qiagen, Chatsworth, CA). The supernatant containing the retroviruses was collected 48 h after transfection and used to infect 3T3-L1 preadipocytes in the presence of 4 μg/ml polybrene. Then, 48 h after infection, cells were selected for GFP expression by flow cytometry. Because both GFP and IκBα-DN are under control of the retroviral long terminal repeat (LTR) promoter, protein levels of GFP are therefore indicative of IκBα-DN expression. GFP-positive cells were amplified once and were sorted again by flow cytometry to select cells with the top 1% of GFP expression.

Blots containing 20 μg total RNA per lane isolated from indicated cells were hybridized with ³²P-labeled cDNAs encoding mouse plasminogen activator inhibitor-1 (PAI-1), adipocyte complement-related protein of 30 kDa (ACRP30), and GLUT4. The filters were exposed to an imager screen (Fuji Scientific), after which filters were stripped and reprobed with ³²P-labeled β-actin cDNA. All hybridization signals were scanned on a phosphoimager and normalized to β-actin signals.

Total RNA was isolated from indicated sources by using an RNeasy kit (Qiagen). cDNA was synthesized from equal amounts of total RNA by using the Superscript System (Life Technologies, Rockville, MD) with an oligo-dT primer containing T7 promoter sequences. Biotinylated cRNA was generated by in vitro transcription, purified by the RNeasy kit (Qiagen), and quantified by ultraviolet spectrometry. Hybridization to murine U74A or U74Av2 oligonucleotide arrays (Affymetrix, Santa Clara, CA) was performed according to vendor’s protocols and scanned on Affymetrix Gene-Array Scanners. Three independent experiments were performed to examine gene expression after 0, 4, and 24 h of TNF-α incubation. In addition, data were collected at multiple time points (0, 0.5, 1, and 2 h) after TNF-α stimulation of wild-type adipocytes or adipocytes expressing IκBα-DN to assess the functional involvement of NF-κB in the kinetics of TNF-α-induced changes in gene expression. Because extensive cell death occurred in IκBα-DN adipocytes after 2 h of TNF-α stimulation, we performed this experiment in the two cell types up to 2 h of TNF-α addition.

The quality of the hybridization on each array was first evaluated by the percent of P-calls (P-calls according to an Affymetrix algorithm) present on each array. Because nonspecific hybridization to mismatch oligonucleotides and severe truncation of mRNAs will substantially lower the percentage of P-calls, arrays with <25% of P-calls were discarded. The average of P-calls for all arrays used in this study was 40.1 ± 1.4% (means ± SE; n = 17). Analysis of differential gene expression was performed by using GeneSpring software (version 4.07; Silicon Genetics, Redwood City, CA). Array measurements for all samples were normalized with arrays hybridized with cRNA prepared from the appropriate control cells by using the median of the hybridization signals of all genes with P-calls as a scaling factor. An arbitrary value of 50 was allocated for genes whose expression levels were undetectable or scan readouts were below 50 to facilitate calculation of the fold changes compared with the control condition. Identification of genes up- or downregulated by TNF-α treatment was determined by selecting only those significantly (greater than threefold for upregulation and twofold for downregulation) and reproducibly increased or decreased in at least two independent experiments. We then classified each of the TNF-α-regulated genes into functional categories according to the putative biological functions of the encoded protein, as determined by searches of public databases. Gene expression profiles were confirmed by semi-quantitative RT-PCR analysis of 23 potentially important mRNAs regulated by TNF-α.

Total RNAs were isolated from 3T3-L1 adipocytes treated with TNF-α (1 nmol/l) for 0, 4, and 24 h. Each sample of first-strand cDNA was synthesized from 1 μg total RNA. Then, various gene-specific primers were used to amplify target genes using equal amounts of the first-strand cDNA as template. PCR reactions were performed for 30 and 35 cycles, unless otherwise indicated, to identify the linear range for PCR. PCR products were separated on a 1.5% agarose gel and photographed.

3T3-L1 adipocytes were incubated in growth medium with or without TNF-α (1 nmol/l) for various times. At the end of the incubation, cells were washed three times with ice-cold PBS (pH 7.4) and solubilized. The indicated cell lysates were separated by SDS-PAGE and electroblotted onto a nitrocellulose membrane (Amersham, Arlington Heights, IL). The filter was incubated for 1 h at room temperature with various primary antibodies (as indicated in the figure legends), washed, and incubated with the appropriate horseradish peroxidase-conjugated secondary antibody. Bound antibodies were detected using the enhanced chemiluminescence Western blotting analysis system (Amersham). Blots were stained with Ponceau S solution to visualize the amount of total protein in each lane.

Nuclear extracts were prepared from the indicated sources as described by Wiegmann et al. (33). The NF-κB-specific oligonucleotide was purchased from Santa Cruz (Santa Cruz, CA) and was end-labeled with [γ-³²P]ATP (Amersham) using polynucleotide kinase. Electrophoretic mobility shift assays were performed essentially as previously described (33). For the electrophoretic mobility supershift assay, the nuclear extract was preincubated with an appropriate amount of anti-RelA/p65 antibody (Santa Cruz), as suggested by the vendor, at room temperature for 20 min.

3T3-L1 adipocytes were grown in six-well plates, fixed with 4% paraformaldehyde, and stained with 0.2% Oil Red O for 10 min at room temperature. Cells were washed exhaustively with PBS and water and examined using an inverted microscope and photographed.

To determine the amount of TNF-α needed to elicit maximal changes in gene expression in 3T3-L1 adipocytes, we examined the dosage-dependent effect of TNF-α on mRNA levels of several adipocyte-specific genes. A 24-h treatment of 3T3-L1 adipocytes with insulin (100 nmol/l) caused a 20% increase in PAI-1 mRNA (Fig. 1A, lane 2, and Fig. 1C, top panel), but did not affect the level of mRNA encoding ACRP30, an adipocyte-derived hormone that decreases gluconeogenesis in liver, increases free fatty acid oxidation in skeletal muscle, and reduces body weight without affecting food intake (34,35) (Fig. 1B, lane 2, and Fig. 1C, bottom panel). In contrast, incubation of 3T3-L1 adipocytes with 0.04 nmol/l TNF-α for 24 h resulted in a 27% reduction in ACRP30 gene expression, with maximal suppression (49%) occurring at 1 nmol/l of TNF-α (Fig. 1B, lanes 3–6, and Fig. 1C, bottom panel). A 24-h exposure to TNF-α at a concentration as low as 0.04 nmol/l caused a 23% increase in PAI-1 mRNA, with maximal induction of PAI-1 mRNA occurring at 0.2 nmol/l of TNF-α (167% that of control) (Fig. 1A, lanes 3–6, and Fig. 1C, top panel). Thus, when using PAI-1 and ACRP30 as models, the maximal induction or suppression of gene expression in 3T3-L1 adipocytes by 24-h TNF-α treatment occurs at concentrations of 0.2–1.0 nmol/l.

Figure 2 shows that TNF-α-induced changes in gene expression in 3T3-L1 adipocytes can be detected by Northern blot analysis as early as 4 h. 3T3-L1 adipocytes were incubated with insulin, TNF-α, or both for various times, as indicated in the figure legend. Insulin alone did not significantly alter the mRNA levels of either ACRP30 or GLUT4 at any time point examined (Fig. 2A, lanes 2, 6, and 10, and Fig. 2B). In contrast, although 1-h TNF-α incubation had no detectable effect on ACRP30 and GLUT4 mRNA levels, 4 h exposure to TNF-α suppressed ACRP30 mRNA by 33% (Fig. 2A, lane 7, and Fig. 2B, top panel), and a 24-h TNF-α incubation reduced ACRP30 mRNA levels by 79% (Fig. 2A, lane 11, and Fig. 2B, top panel). Similarly, 4- and 24-h TNF-α treatment of 3T3-L1 adipocytes reduced GLUT4 mRNA by 47 and 78%, respectively (Fig. 2A, lane 7 and 11, and Fig. 2B, bottom panel). The addition of insulin (100 nmol/l) partially blocked the inhibitory effect of 24-h TNF-α incubation on GLUT4 and ACRP30 mRNA levels. As shown in Fig. 2A (lane 12) and Fig. 2B, insulin reduced TNF-α-triggered inhibition from 78 to 36% for GLUT4 mRNA, and from 79 to 59% for the ACRP30 transcript.

To determine whether adipocyte gene expression is extensively altered during TNF-α treatment, we performed oligonucleotide microarray analysis 4 and 24 h after TNF-α addition. By comparing TNF-α-treated 3T3-L1 adipocytes with control cells, we identified 59 and 130 known genes that were reproducibly upregulated by at least threefold after 4 and 24 h of TNF-α treatment, respectively. In addition, 78 known genes were downregulated at least twofold by 24 h exposure to TNF-α. The TNF-α responsive genes were then grouped according to the biological functions of their encoded proteins (Table 1).

To confirm that TNF-α indeed affected the levels of several potentially important mRNAs, we performed RT-PCR analysis. Specifically, we designed 23 pairs of different gene-specific primers and compared the amount of RT-PCR products amplified from each experimental condition after 30 and 35 PCR cycles. All of the primers yielded PCR products of the expected sizes. Although changes in the expression of 20 genes (87%) were confirmed by RT-PCR, changes in the expression of 3 genes (13%) were not. Figure 3 shows 18 representative genes. Among them, 12 genes were induced by 4 or 24 h of TNF-α treatment, including serine protease inhibitor 2-1 (Spi2-1), matrix metalloproteinase 3 (MMP3), interleukin 6 (IL-6), inducible IKK, Fas antigen, NF-κB p105, engrailed-1, Fos-related antigen-1 (Fra-1), lipopolysaccharide binding protein (LPSBP), high mobility group protein-1 isoform C (HMGP-1C), adrenomedullin (Adm), and B-cell leukemia/lymphoma-3 (Bcl-3). In contrast, mRNAs encoding CCAAT/enhancer binding protein-α (CEBP-α), glycerol-3-phosphate dehydrogenase, nuclear receptor 4A1, nuclear receptor retinoid X receptor-α (RXR-α), and hormone-sensitive lipase (HSL), were suppressed by TNF-α treatment of 3T3-L1 adipocytes (Fig. 3). Tubulin was used as an internal control for an mRNA whose level did not change. It is important to note that some changes in mRNA levels may be masked during the exponential amplification of PCR, and therefore RT-PCR is likely to yield false negative results when used to detect changes in gene expression, but unlikely to yield false positive results.

As judged by the microarray analysis, TNF-α incubation caused profound changes in gene expression in 3T3-L1 adipocytes. Several adipocyte-related transcription factors, including CEBP-α (−3.4-fold), RXR-α (−2.3-fold), and peroxisome proliferator-activated receptor-γ (PPAR-γ; −2.0 fold) were significantly downregulated by TNF-α (Table 1). In addition, TNF-α suppressed the expression of GLUT4 (−5.7-fold), glycogen synthase (−1.8-fold), and fatty acid synthase (−5.0 fold), which are essential for insulin-mediated glucose uptake and their subsequent conversion to glycogen or fatty acids. Notably, genes involved in the uptake of free fatty acids and conversion to triglyceride, such as lipoprotein lipase (−1.6-fold), long-chain fatty acyl-CoA synthetase (−2.8-fold), and diacylglycerol acyltransferase (−2.6-fold), were also downregulated by TNF-α. Moreover, a number of genes encoding adipocyte-abundant proteins, including HSL (−5.8-fold), glycerol phosphate dehydrogenase 1 (−5.2-fold), and pyruvate decarboxylase (−2.5-fold), as well as adipocyte-secreted factors such as angiotensinogen (−3.7-fold) and ACRP30 (−1.4-fold), were suppressed by TNF-α as well. This indicated that TNF-α induces insulin resistance in 3T3-L1 adipocytes at least in part through suppression of genes that are essential for the metabolic functions of adipocytes, including energy uptake and storage in response to insulin and protein secretion.

As opposed to the suppressive effects of TNF-α on adipocyte-abundant genes, TNF-α induced a number of genes that are normally enriched in preadipocytes, including CEBP-β (1.6-fold), Fra-1 (12-fold at 4 h, 5.8-fold at 24 h), GLUT1 (4.4-fold at 24 h), prostaglandin E synthase (3.2-fold at 4 h, 5.2-fold at 24 h), Spi-2 (30.5-fold at 4 h, 95.9-fold at 24 h), PAI-1 (11-fold at 4 h, 52-fold at 24 h), secreted/truncated form of vascular cell adhesion molecule 1 (VCAM-1; 9.4-fold at 4 h, 10.2-fold at 24 h), and fibronectin (2.2-fold at 24 h). Notably, many of the TNF-α-induced genes encode secreted proteins, such as PAI-1, secreted VCAM-1, fibronectin, tenascin C (7.4-fold at 24 h), adrenomedullin (7.3-fold at 4 h, 29.6-fold at 24 h), and ceruloplasmin (10-fold at 24 h), whose overexpressions are associated with type 2 diabetes. Thus, TNF-α may also contribute to at least some of the phenotypes associated with obesity and obesity-linked type 2 diabetes by modulating the endocrine function of the adipocytes, including selective induction of preadipocyte genes as well as suppression of adipocyte-secreted factors.

Another striking result is that a number of chemokines, cytokines and their receptors, and cytokine-induced proteins were immediately induced by TNF-α in 3T3-L1 adipocytes. For example, IL-6 mRNA was upregulated nearly 10-fold after 4 h of TNF-α incubation and 30-fold after 24 h of TNF-α incubation (Table 1 and Fig. 3). Many chemokines and cytokine-induced proteins were also strongly upregulated 4 and 24 h after TNF-α incubation (Table 1), indicating a direct induction by TNF-α.

TNF-α treatment of adipocytes induced or suppressed the expression of a number of transcription factors and their associated regulatory proteins. Among them, mRNAs that encode Bcl-3 and p105-NF-κB were significantly upregulated after 4 h exposure to TNF-α and remained 6.3-fold (p105-NF-κB) and 10.8-fold (Bcl-3) of control cells after 24 h of TNF-α incubation. P105-NF-κB encodes a precursor for both the p50-NF-κB transcription factor and the NF-κB repressor I-κBγ (5), whereas Bcl-3 is a DNA binding protein that has been implicated as a modulator of the activity of NF-κB (6–8). In the absence of TNF-α, NF-κB is sequestered in the cytoplasm by binding to I-κBs (9). Stimulation by TNF-α leads to degradation of I-κB proteins and the translocation of NF-κB to the nucleus. Importantly, the expression of inducible-IKK (IKKi), which mediates the rate-limiting step of phosphorylation and degradation of the NF-κB inhibitor (IκBs) (10), was also upregulated 8.4-fold after 24 h incubation with TNF-α. Increased IKKi expression or activity may enhance the rate of IκB degradation and the subsequent release and activation of NF-κB. In addition, TNF-α stimulation resulted in a 60% increase in the mRNA levels of CEBP-β, which can also dimerize with subunits of NF-κB and activate gene transcription (11). Thus, TNF-α rapidly induced the synthesis of several mRNAs encoding proteins implicated in NF-κB activation in 3T3-L1 adipocytes, including p105-NF-κB, IKKi, and other NF-κB-associated proteins such as Bcl-3 and CEBP-β (Table 1).

TNF-α also upregulated several growth factors and proteins associated with cell cycle progression, but downregulated proteins associated with cell cycle inhibition, including CEBP-α (Table 1) (18), cyclin-dependent kinase inhibitor 1C, cyclin-dependent kinase inhibitor 2C (p18), growth-arrest specific 1 (19,20), and growth-arrest and DNA-damage-inducible-γ gene (21). CEBP-α is an important cell growth and differentiation transcriptional regulator that is expressed in many tissues (22). In fully differentiated 3T3-L1 adipocytes, CEBP-α is highly induced, whereas CEBP-β is suppressed (23). CEBP-α not only maintains growth arrest in terminally differentiated 3T3-L1 adipocytes (24), but also regulates the expression of adipocyte-specific genes and energy metabolism (25). These coordinate changes in gene expression might cause 3T3-L1 adipocytes to reenter the cell cycle.

To assess suppression of gene expression as a potential mechanism for TNF-α-induced insulin resistance, we revisited the effects of TNF-α treatment on the levels of several proteins. TNF-α induced a progressive reduction in the levels of GLUT4, IR, and IRS-1 proteins (Fig. 4A and B), confirming results in a previous report (26). In addition, TNF-α suppressed protein levels of AKT, a key enzyme necessary for insulin-stimulated glucose uptake in adipocytes (Fig. 4C and D). Conversely, the level of GLUT1 protein increased 60% after TNF-α treatment for up to 48 h (Fig. 4A and B). Thus, our data clearly indicated that TNF-α modulates 3T3-L1 adipocyte function at the level of gene transcription, and that TNF-α-induced characteristic changes in gene expression provide a molecular basis for the development of insulin resistance.

To determine whether TNF-α affects gene transcription through activation of NF-κB, we first assessed NF-κB expression and its translocation into the nucleus during TNF-α stimulation. NFκB1/p105, RelB, RelA/p65, and c-Rel (a COOH-terminal truncated form of 53 kDa) were readily detectable in nuclear extracts of 3T3-L1 adipocytes (Fig. 5A, lane 1). The levels of NF-κB1/p105, Rel B, and c-Rel protein were not changed 15 or 60 min after TNF-α stimulation, and no proteolytic cleavage of NF-κB1/p105 occurred. Nuclear accumulation of p65, however, increased substantially 15 and 60 min after TNF-α addition (Fig. 5A, lanes 2 and 3).

We then used a gel mobility shift assay to confirm that TNF-α acutely activates NF-κB nuclear translocation, a prerequisite for its having a functional involvement in TNF-α action in 3T3-L1 adipocytes. After TNF-α stimulation, NF-κB RelA/p65 activation and nuclear translocation occurred within 15 min and lasted at least 60 min, consistent with results in other cell types (36) (Fig. 5B). This indicated that NF-κB is immediately involved in TNF-α action at the gene transcription level in 3T3-L1 adipocytes.

To ascertain whether NF-κB activation is essential to initiating changes in gene expression typical of insulin-resistant adipocytes, we established a 3T3-L1 cell line stably expressing IκBα-DN, a nondegradable mutant of NF-κB inhibitor IκB-α (37–39). 3T3-L1 cells expressing IκBα-DN differentiate normally into adipocytes (Fig. 6A), and respond to insulin to the same extent as do wild-type adipocytes, as measured by [³H]2-deoxy-glucose uptake (data not shown). We confirmed the function of the IκBα-DN protein in fully differentiated 3T3-L1 adipocytes by incubating it with TNF-α, and monitoring the phosphorylation and degradation of both the endogenous and the mutant IκB-α proteins. As shown in Fig. 6B, in wild-type adipocytes, TNF-α efficiently stimulated phosphorylation of endogenous IκB-α as early as 5 min, resulting in rapid and complete degradation of the wild-type IκB-α protein within 15 min. In contrast, no TNF-α-mediated phosphorylation and degradation of the IκBα-DN protein occurred in adipocytes expressing the mutant protein. After 60 min of TNF-α addition, IκB-α protein reaccumulated in the wild-type adipocytes, as activation of NF-κB also stimulates transcription of IκB-α. The absence of reaccumulation of endogenous IκB-α protein in cells expressing IκBα-DN again indicated the complete inhibition of NF-κB activation in these cells.

Extensive cell death occurred in 3T3-L1 adipocytes expressing IκBα-DN after 2 h of TNF-α addition, indicating a critical role of NF-κB activation in protecting against TNF-α-induced apoptosis. We collected total RNA from wild-type and IκBα-DN adipocytes stimulated for up to 2 h by TNF-α, and performed oligonucleotide microarray analysis of the global gene expression profiles.

We first compared the global gene expression profile in wild-type 3T3-L1 adipocytes and adipocytes expressing IκBα-DN. In the scatter plot shown in Fig. 7, the position of each gene on the plot is determined by its expression level in both wild-type adipocytes (x-axis) and IκBα-DN adipocytes (y-axis). Genes with identical or similar expression levels in both cell types cluster along the diagonal, whereas genes whose expression levels are higher or lower in wild-type adipocytes cluster below or above the diagonal line, respectively. Despite the similar global gene expression patterns in the two cell-types (Fig. 7), differences did exist in a subset of genes. We identified 200 genes (1.6% of 12,488 genes examined, including both known genes and ESTs) whose basal expression levels in wild-type adipocytes are at least twofold higher than those in IκBα-DN adipocytes, and 20 genes (0.16%, including both known genes and ESTs) whose expression levels are at least twofold lower than those in IκBα-DN adipocytes. Thus, basal NF-κB activity is required for the expression or suppression of a subset of genes in 3T3-L1 adipocytes. Because the differences in the basal gene expression did not affect adipocyte differentiation or the response to insulin, we did not further investigate their functional significance.

Next, we assessed the kinetics of TNF-α-induced changes in gene expression. In wild-type 3T3-L1 adipocytes, TNF-α suppressed the mRNA levels of 61, 193, and 264 genes (including known genes and ESTs) at least twofold at the 0.5-, 1-, and 2-h time points. In adipocytes expressing IκBα-DN, however, only 1.6% (1 of 61 genes), 0.5% (1 of 193), and 1.9% (5 of 264) of the genes that are normally suppressed by TNF-α were downregulated at the 0.5-, 1-, and 2-h time points, respectively (Fig. 8A). On the other hand, TNF-α induced 52, 33, and 39 genes (including known genes and ESTs) by at least twofold in 3T3-L1 adipocytes after 0.5-, 1-, and 2-h incubation, respectively. Notably, ∼27% (14 of 52), 39% (13 of 33), and 31% (12 of 39) of the genes that are normally induced by TNF-α were still upregulated by TNF-α in IκBα-DN adipocytes (Fig. 8B). Moreover, in IκBα-DN adipocytes, TNF-α induced and suppressed many genes that were not affected by TNF-α in wild-type adipocytes. In particular, TNF-α suppressed 35, 27, and 27 genes (known genes and ESTs) and induced 110, 214, and 266 genes (known genes and ESTs) in adipocytes expressing IκBα-DN protein at 0.5-, 1-, and 2-h time points, respectively. Thus, the absence of NF-κB activation abolished suppression of 98–99% of the genes normally suppressed by TNF-α and induction of 60–70% of the genes normally induced by TNF-α, and produced a distinct gene expression profile in response to TNF-α.

We then compared the kinetics of TNF-α treatment up to 2 h on the expression of several key adipocyte transcription factors, as well as genes involved in glucose and free fatty acid uptake and storage (Fig. 9). There was no significant difference in the basal expression levels of the genes we examined in wild-type and IκBα-DN adipocytes. By 2 h of TNF-α treatment of wild-type adipocytes, the mRNA levels of PPAR-γ, CEBP-α, GLUT4, 1-acyl-glycerol-3-phosphate acyltransferase, and diacylglycerol acyltransferase were decreased 50% or more. In contrast, their levels were unchanged or slightly increased after TNF-α treatment of IκBα-DN adipocytes. The decrease in mRNA levels was likely attributable to both inhibition of gene transcription and the short half-life of the mRNAs examined. The levels of abundant and stable mRNAs, such as those encoding ACRP30 and long-chain fatty acyl CoA synthase, were not affected at the 2-h time point, but were decreased after 24 h of TNF-α treatment (Fig. 1B and C; Table 1). Because of apoptosis of the IκBα-DN adipocytes after 2 h of TNF-α treatment, we could not determine whether TNF-α suppression of their expression depended on NF-κB.

Insulin resistance refers to a state in which physiological concentrations of insulin produce a less than normal response. Major metabolic consequences of insulin resistance include elevated plasma concentrations of circulating fuel molecules (glucose and free fatty acids) and in some cases, triglycerides and cholesterol as well. Although decreased receptor tyrosine kinase activity and impaired insulin receptor signal transduction contribute to the development of insulin resistance, these factors are insufficient to explain the complex phenotype. Insulin resistance in obesity and obesity-linked type 2 diabetes is often accompanied by significant changes in gene expression in major insulin-responsive tissues. Thus, it can be informative to identify mediators that elicit the insulin-resistance phenotype. TNF-α is increasingly recognized as a link between obesity and insulin resistance, and thus has been the focus of intensive study.

Here, we reported three major findings that, when taken together, describe one of the mechanisms by which TNF-α treatment causes insulin resistance in 3T3-L1 adipocytes. First, we found that TNF-α immediately alters the expression of many genes in 3T3-L1 adipocytes. It specifically downregulates many adipocyte-abundant genes that are critical for insulin responsiveness and selectively reactivates many preadipocyte genes. TNF-α also induces several transcription factors and genes implicated in NF-κB activation, as well as many genes involved in cell growth, proliferation, and inflammation. Second, we identified NF-κB as an obligatory mediator of the TNF-α-induced changes in gene expression that could lead to insulin resistance in 3T3-L1 adipocytes. Third, we found that NF-κB is also specifically required for protection against TNF-α-induced apoptosis in 3T3-L1 adipocytes. Thus, our findings implicate adipocyte NF-κB as a potential target for pharmaceutical interventions of the insulin resistance seen in obesity and obesity-linked type 2 diabetes.

TNF-α has direct effects on gene expression in 3T3-L1 adipocytes, as changes in mRNA levels can be seen as early as 0.5 h after TNF-α addition. Notably, TNF-α suppressed the expression of a number of transcription factors essential for the adipocyte phenotype, including PPAR-γ, RXR-α, CEBP-α, and Nrip-1. The PPAR-γ/RXR-α heterodimer induces many adipocyte-specific genes implicated in adipocyte differentiation and affects systemic glucose and lipid homeostasis (27). During 3T3-L1 adipocyte differentiation, CEBP-α is highly induced, whereas CEBP-β and CEBP-γ are significantly downregulated (23). A high CEBP-α/CEBP-β ratio is therefore a marker of adipocyte differentiation. CEBP-α acts synergistically with PPAR-γ and is critical for maintenance of the adipocyte phenotype as well as the expression of adipocyte-specific genes essential for insulin action (27). TNF-α potently induced the expression of CEBP-β (Table 1) and completely reversed the normal adipocyte CEBP-α/CEBP-β ratio. In addition, Nrip-1, a coactivator for many nuclear receptors, was also significantly suppressed. Downregulation of Nrip-1 mRNA might therefore potentiate the inhibitory effects of TNF-α on PPAR-γ, RXR-α, and CEBP-α actions.

Furthermore, TNF-α suppressed many adipocyte-abundant genes that encode proteins essential for insulin responsiveness (Table 1 and Fig. 4). Among them, IRS-1 and AKT are critical for insulin signal transduction, and lipoprotein lipase, long-chain fatty acyl CoA synthase, diacylglycerol acyltransferase, GLUT4, glycogen synthase, and fatty acid synthase are required for insulin-stimulated glucose and fatty acid uptake and their conversion to glycogen and triglyceride (28,29). Because the overall physiological effects of insulin on energy metabolism are to promote energy storage and inhibit energy mobilization, TNF-α may thus antagonize the action of insulin by shutting down the expression of genes that are essential for insulin responsiveness.

In addition to inhibiting adipocyte-specific gene expression, TNF-α also induced many growth factors and proteins associated with cell cycle progression, and in parallel downregulated cell cycle inhibitor proteins, including CEBP-α, GATA-6, cyclin-dependent kinase (CDK) inhibitor 1C, CDK inhibitor 2C (p18), growth-arrest specific 1, and growth-arrest and DNA-damage-inducible-γ gene. CEBP-α not only regulates the expression of adipocyte-specific genes and energy metabolism, but also inhibits the transcription of proteins necessary for cell cycle progression, such as cyclins or CDKs, and stimulates the expression of cell cycle inhibitors that inactivate specific CDKs (40). Although direct evidence for GATA-6 playing a role in the maintenance of adipocyte growth-arrest state is lacking, GATA-6 has been implicated in the regulation of cell proliferation and differentiation in other cell types. GATA-6 mRNA is downregulated in proliferating vascular smooth muscle cells, and forced expression of GATA-6 in vascular smooth muscle cells induces growth arrest that is associated with enhanced expression of CDK inhibitor p21 (41, 42). Thus, downregulation of mRNAs encoding CEBP-α and other growth arrest-inducing proteins, which is also enhanced by the induction of growth factors, might release cell cycle constraints and result in the activation of cell cycle and repression of adipocyte-specific genes in 3T3-L1 adipocytes. These changes, together with the suppression of adipocyte-specific proteins, may contribute to the loss of the adipocyte phenotype and insulin responsiveness seen after long-term TNF-α treatment of 3T3-L1 adipocytes.

Although TNF-α mRNA and protein levels are upregulated in adipose tissue and skeletal muscle in obese animal and human subjects (8,9,11,12), the plasma concentration of TNF-α is undetectably low (11,43). Thus, TNF-α is likely to act locally via an autocrine effect in adipose tissue and/or skeletal muscle, and may also act indirectly through mediators produced by these tissues in response to TNF-α.

Because TNF-α increases lipolysis in adipocytes (43–45), plasma free fatty acid is a potential candidate for a systemic mediator of TNF-α action. In support of this notion, it has been shown that infusion of TNF-α in rats increases plasma free fatty acid levels (6), and neutralization of TNF-α in Zucker rats has been associated with both increased insulin sensitivity and decreased plasma free fatty acid levels (5). Moreover, circulating free fatty acid and triglyceride levels are significantly lower in TNF-α knockout mice than their wild-type littermates (15). Excess plasma free fatty acids antagonize the actions of insulin in insulin-sensitive tissues (46) and may thus enhance the effects of TNF-α in the induction and aggravation of insulin resistance seen in obesity and obesity-linked type 2 diabetes.

Cytokines are other potential systemic mediators of TNF-α action. Table 1 lists a number of cytokines, cytokine-induced proteins, and chemokines induced by TNF-α. Among them, IL-6 and interferon-α have been implicated in the induction of insulin resistance in a number of pathological settings, including infection and cancer-induced cachexia (47–50). In addition, interferon-γ has been shown to induce IL-6 secretion and may thus enhance the metabolic effects of IL-6. Therefore, 3T3-L1 adipocytes produce several cytokines and chemokines in response to TNF-α that may potentially, at least in cell culture, contribute to the development of insulin resistance.

TNF-α also modulates protein secretion from adipocytes. Among the genes induced by TNF-α, secreted VCAM-1, fibronectin, tenascin C, ceruloplasmin, PAI-1, and adrenomedullin have been associated with type 2 diabetes or an excessive rate of cardiovascular diseases (51–53). In addition, TNF-α significantly suppressed mRNA levels of the gene encoding ACRP30, an adipocyte-derived hormone that inhibits gluconeogenesis in liver (34), enhances free fatty acid oxidation in skeletal muscle, and reduces body weight in mice (35). Because ACRP30 promotes both fatty acid and glucose utilization in skeletal muscle, downregulation of ACRP30 gene expression may mediate, at least in part, the effects of TNF-α in inducing insulin resistance in vivo. Hence, in addition to the direct effects on adipocyte gene expression, the metabolic effects of TNF-α may also be mediated and/or amplified by the systemic protein and lipid mediators produced by adipose tissue in response to TNF-α.

The blockade of NF-κB activation through the expression of IκBα-DN abolished the suppression of >98% of the genes normally suppressed by TNF-α, and induction of 60–70% of the genes normally induced by TNF-α. Strikingly, the absence of NF-κB activation significantly altered the global changes in gene expression in response to TNF-α, indicating NF-κB is a master regulator of gene expression and a key mediator of TNF-α action in 3T3-L1 adipocytes. Moreover, NF-κB is essential for protection against TNF-α-induced apoptosis in 3T3-L1 adipocytes. Whether NF-κB is essential for the initiation of changes in adipocyte gene expression and the induction of insulin resistance in vivo cannot be answered by our 3T3-L1 adipocyte model alone. Yet, 3T3-L1 adipocytes expressing IκBα-DN are useful as an initial step to define the molecular mediators in TNF-α-induced insulin resistance, and may provide a basis for pharmaceutical intervention in type 2 diabetes.

In summary, we conclude that TNF-α immediately causes changes in gene expression that provide a basis for the induction of insulin resistance in 3T3-L1 adipocytes, and that NF-κB activation is obligatory for TNF-α action. Further studies are required to identify the particular NF-κB subunits that mediate the suppression and/or induction of the adipocyte genes, as well as the molecular pathways mediating NF-κB activation in 3T3-L1 adipocytes.

This work was supported by National Institutes of Health Grant R37-DK-47618 (H.F.L.), and by grants from Affymetrix, Millennium Pharmaceuticals, and Bristol-Myers Squibb Company to the Genome Center at the Whitehead Institute. H.R. was supported by a postdoctoral fellowship for physician scientists from the Howard Hughes Medical Institute.

We thank C. Ladd for assistance with oligonucleotide array target preparation and scanning; A. Sirotkin for ACRP30 cDNA; Y. Sun for the PAI-1 probe; and J. Marszalek, M.K. Um, and L.J.S. Huang for helpful discussions and critical reading of the manuscript.