Expression Profiling Identifies Genes That Continue to Respond to Insulin in Adipocytes Made Insulin-resistant by Treatment with Tumor Necrosis Factor-α*

We have employed microarray technology using RNA from normal 3T3-L1 adipocytes and from 3T3-L1 adipocytes made insulin-resistant by treatment with tumor necrosis factor-α to identify a new class of insulin-responsive genes. These genes continued to respond normally to insulin even though the adipocytes themselves were metabolically insulin-resistant, i.e. they displayed a significantly decreased rate of insulin-stimulated glucose uptake. Approximately 12,000 genes/expressed sequence tags (ESTs) were screened. Of these, 40 genes/ESTs were identified that became insulin-resistant as expected (e.g. Socs-3, junB, and matrix metalloproteinase-11). However, 61 genes/ESTs continued to respond normally to insulin. Although some of these genes were previously shown to be regulated by insulin (e.g. Glut-1 and β3-adrenergic receptor), other novel insulin-sensitive genes were also identified (e.g. Egr-1, epiregulin, Fra-1, and ABCA1). Real-time reverse transcription-PCR analysis confirmed the expression patterns of several of the differentially expressed genes. One gene that remained insulin-sensitive in the insulin-resistant adipocytes is the transcription factor Egr-1. Using an antisense strategy, we show that tissue factor and macrophage colony-stimulating factor, two cardiovascular risk factors, are downstream EGR-1 target genes in the adipocyte. Taken together, these data support the hypothesis that some signaling pathways remain insulin-sensitive in metabolically insulin-resistant adipocytes. These pathways may promote abnormal gene expression in hyperinsulinemic states like obesity and type II diabetes and thus may contribute to pathologies associated with these conditions.

Insulin resistance and obesity represent rapidly growing major health problems worldwide and are hallmarks for type II diabetes (1). A strong correlation exists between insulin resistance and abdominal visceral fat accumulation (2), and a substantial body of evidence points to the existence of genetic determinants of fat mass (3)(4)(5)(6). However, despite recent advancements in these fields, the molecular changes that contribute to these conditions and to the subsequent development of type II diabetes remain enigmatic.
Studies of cultured adipocytes and of mouse models of obesity have begun to shed light on this problem. For example, the adipocyte can no longer be viewed as a passive participant in the development of obesity and is now considered to be a metabolically active secretory cell that plays an important role in the regulation of energy balance and body composition (7). During the development of obesity and type II diabetes, these cells increase in size and exhibit modified metabolic properties, and their sensitivity/response to hormones may change. Experimental and clinical observations support a connection between adipocyte-derived factors and the association of insulin resistance with an increased risk for cardiovascular events (8).
Insulin itself is a hormone that induces a cascade of intracellular events in the adipocyte (9). For example, upon binding to the ␣-subunits of its receptor at the cell surface, insulin activates the intrinsic tyrosine kinase activity of the ␤-subunits of the receptor. This interaction leads to phosphorylation of intracellular proteins, including insulin receptor substrate-1-4, GAB-1, and Cbl. The insulin signals subsequently diverge through different pathways that appear to control distinct functions such as glucose transport, glucose/lipid metabolism, cell growth, protein synthesis, and gene expression (10). Although the exact molecular mechanisms that govern these pathways are not known, some key molecules have been identified. For example, numerous studies have confirmed a role for phosphatidylinositol (PI) 1 3-kinase and AKT/protein kinase B in metabolic signaling; and recently, it was reported that TC10 is important for glucose transport and metabolism (11). However, the effects of insulin on gene expression often occur independently of PI 3-kinase activity and may instead require activation of the Ras-Raf-MAPK pathway, especially in the insulin-resistant (IR) state (12)(13)(14).
By definition, insulin resistance describes an impaired biological responsiveness to insulin (15). It is frequently used to describe the impairment of insulin-stimulated glucose uptake by muscle and adipocytes. However, the degree to which some of the other actions of insulin (e.g. gene expression) are normal or resistant in this state is not clear. If some genes do not become resistant to insulin, then it is possible that the hyperinsulinemia that frequently accompanies obesity and insulin resistance may promote the abnormal expression of these genes in insulin-responsive tissues. This hypothesis is supported by previous observations showing that IR adipocytes and mice remained sensitive to insulin in terms of the expression of plasminogen activator inhibitor-1 (PAI-1) (12), sterol regulatory element-binding protein-1c (16), and monocyte chemoattractant protein-1 (MCP-1) (17). Identification of the genes that continue to respond to insulin in IR adipocytes, in concert with studies of the properties of these genes and the signaling pathways that regulate them, may provide novel insights into the molecular mechanisms that control abnormal gene expression in obesity and type II diabetes.
To begin to define this gene family, we screened ϳ12,000 transcripts by performing expression profiling using microarray technology. By comparing the gene expression profiles of insulin-treated normal (N) 3T3-L1 adipocytes and metabolically IR 3T3-L1 adipocytes prepared by treatment with tumor necrosis factor-␣ (TNF-␣) (12), we have identified 61 genes/ expressed sequence tags (ESTs) that respond normally to insulin in IR adipocytes. One of these genes is the transcription factor Egr-1. Antisense studies demonstrated that activation of Egr-1 promoted the induction of tissue factor and macrophage colony-stimulating factor (MCSF) mRNAs by insulin.

EXPERIMENTAL PROCEDURES
Cell Culture-Murine 3T3-L1 fibroblasts (American Type Culture Collection, Manassas, VA) were cultured in vitro and differentiated into fully mature adipocytes as described (12,18). IR cells were obtained by incubating the differentiated adipocytes for 3 days in the presence of 3 ng/ml TNF-␣ (R&D Systems, Minneapolis, MN) with fresh media changes each day as described (12,17,19). Adipocytes treated in parallel but without TNF-␣ were considered N adipocytes. In both cases, after overnight incubation in serum-free medium supplemented with 0.2% bovine serum albumin, insulin-stimulated 2-[ 3 H]deoxyglucose transport was determined as described (12). For experiments with insulin, the cells were incubated with or without 1000 nM bovine insulin (Sigma) for 3 h in fresh serum-free medium. Under the experimental conditions used in this study, this concentration of insulin appears to signal mainly via the insulin receptor (12,17). After completion of each experiment, the cells were lysed using TRIzol reagent (Invitrogen), and the extracts were stored at Ϫ80°C until further processed. For protein extraction, cells were washed three times with phosphate-buffered saline and extracted using 1% SDS.
RNA Isolation-Total RNA was isolated using TRIzol reagent together with a subsequent cleanup RNeasy protocol (QIAGEN Inc., Valencia, CA) following the instructions of the manufacturers.
Gene Expression Profiling-For each experiment, total RNA was isolated and pooled from duplicate culture dishes and reverse-transcribed into cDNA, which, in turn, was used as template for the generation of biotin-labeled cRNA. The cRNA was fragmented and hybridized to Affymetrix GeneChips® U74A v2 (Affymetrix, Santa Clara, CA) at the DNA Array Core Facility of the Scripps Research Institute according to the Affymetrix GeneChip Expression Analysis manual. However, it should be noted that the first experiment with IR cells was conducted using the original version U74A arrays. The hybridized chips were stained with streptavidin-phycoerythrin conjugate and then scanned using an Agilent GeneArray scanner. Quantitative analysis of hybridization patterns and intensities was performed automatically by Affymetrix software as described (20,21), and the resulting data were analyzed using Affymetrix Microarray Suite Version 5.0 software. To ensure reproducibility of the microarray results, the experiments were repeated three times (twice using the U74A v2 chip and once using the U74A chip) each for N and IR 3T3-L1 adipocytes over a time period of 9 months using cells at the same passage for each experiment.
Each GeneChip® was evaluated using "single array analysis" as described by Affymetrix. The RNA levels were quantitatively estimated by "signal," which serves as an indicator of the level of expression of a transcript, and by "detection call," which is obtained using an algorithm based on signal intensity and quality. With the detection call, a transcript is classified as "absent," "marginal," or "present." Genes/ESTs (from hereon referred to only as genes) with altered expression after insulin treatment were identified by two criteria using "comparison analysis" as follows. First, a "change call" (CC) was obtained for each comparison. With the CC, a gene is classified as in-creased (I), marginally increased (MI), no change (NC), marginally decreased (MD), or decreased (D). A scoring system for the CC was established, with a CC of I ϭ 1, MI ϭ 0.5, NC ϭ 0, MD ϭ Ϫ0.5, and D ϭ Ϫ1. The CC score was calculated as the sum of the CC for each of the three individual experiments with N and IR cells, respectively. Second, the "-fold change" (FC) in expression between insulin-treated cells and control cells was calculated using the signal log ratio and formulas provided by Affymetrix.
The selection criteria for increased expression between control and insulin-treated cells for each condition (N and IR cells) for any given gene were as follows. First, a detection call of present for the gene in the insulin-treated cells was required in all three replicate experiments. Second, the CC score for the gene had to be Ն2.5, i.e. the independent replicates for a given gene all had to show a consistent change of expression. Finally, the FC had to be Ͼ3 for the gene in either N or IR cells. The selection criteria for decreased expression between control and insulin-treated cells were a detection call of present in the control cells in all three experiments, a CC score of Ϫ2.5 or less, and an FC of less than Ϫ3. Similar selection criteria, sometimes referred to as "directional consistency," have been used before and have proven successful for accurate identification of differentially expressed genes using Affymetrix GeneChips® (22)(23)(24). This analysis generated two classes of genes: 1) genes that remained insulin-sensitive and that were regulated by insulin in both N and IR cells with no significant difference in FC between the two conditions and 2) genes that became insulin-resistant and that were regulated by insulin in N adipocytes, but were not regulated or were significantly less regulated in IR cells.
Real-time Reverse Transcription (RT)-PCR-Total RNA was reversetranscribed, and the resulting cDNA was analyzed by real-time PCR as described (17) using gene-specific primer sets (see Supplemental Material). Real-time PCRs were performed with SYBR Green I chemistry in an iCycler (Bio-Rad). The authenticity of the PCR products was verified by melting curve analysis and agarose gel electrophoresis. cDNA quantities were normalized to 18 S rRNA quantities measured in the same samples.
Antisense Morpholino Oligonucleotide (MO) Inhibition of Egr-1 Translation-Fully differentiated 3T3-L1 adipocytes were treated with Egr-1-specific antisense MO (Gene Tools, LLC, Philomath, OR) at a final concentration of 1.4 M following the Special Delivery protocol indicated by the manufacturer. The sequence of the Egr-1 antisense MO was 5Ј-CAGCGAGCTGGAGAACTGATGTTGG. A control antisense MO composed of a random sequence (Gene Tools, LLC) was also used at the same concentration. MO-treated cells were incubated with or without 1000 nM insulin for 3 h at 37°C, and total cell protein and RNA were prepared as described above.
Data and Statistical Analysis-Microarray data obtained from Affymetrix Microarray Suite Version 5.0 software were imported into Microsoft Excel 2000 for further analysis. All other calculations were done using GraphPAD Prism Version 3.02 software. The data are expressed as the mean Ϯ S.E. Statistical significance between two groups was determined using Student's t test.

Chronic TNF-␣ Treatment of 3T3-L1 Adipocytes Leads to
Metabolic Insulin Resistance in Vitro-To obtain IR adipocytes, fully differentiated 3T3-L1 adipocytes were treated for 3 days with low concentrations of TNF-␣ (3 ng/ml), a treatment that previously had been shown to significantly reduce the rate of insulin-stimulated glucose uptake without inducing morphological changes in mature adipocytes (12,17,19,26). In this regard, the TNF-␣-treated cells did not display general signs of dedifferentiation because the expression levels of several adipocyte markers (e.g. adipsin, fatty-acid synthase, and lipoprotein lipase) were not altered (data not shown). To verify that the cells used in these experiments were IR, we measured insulin-stimulated 2-[ 3 H]deoxyglucose transport. As illus-trated in Table I, TNF-␣ pretreatment of the adipocytes decreased the rate of insulin-stimulated glucose uptake by ϳ60%. Microarray analysis of the RNA prepared from these cells also indicated that there were defects in IR cells regarding expression of genes that control insulin-stimulated glucose uptake. For example, expression of the glucose transporters GLUT-4 and GLUT-1 was significantly decreased in IR cells compared with N cells (40 and 50%, respectively) ( Table I). The expression of hexokinase II, an enzyme that catalyzes the first step in intracellular glucose metabolism, also appeared to be decreased by ϳ30%, reaching borderline significance. Finally, although the low expression levels of the PI 3-kinase p85 ␤-subunit, of insulin receptor substrate-2, and of protein kinase C⑀ in N cells were close to the detection limit of the microarrays, they were given a present call. However, these genes were not detected in the IR cells and thus were given an absent call in these cells. The absent call makes it impossible to determine quantitative changes in the expression levels of these genes after treatment with TNF-␣. However, it is anticipated that the changes are in the same range as those for Glut-4, hexokinase II, and Glut-1. Taken together, these data support previous observations showing that TNF-␣-treated adipocytes become metabolically IR (19,27).
Global Gene Expression Profiling of Insulin-treated N and IR Adipocytes-Experiments were performed to examine global changes in the gene expression profiles of insulin-treated N and IR adipocytes. Initially, the specificity of the microarray analysis was investigated. To estimate the false-positive rate, comparison analysis was performed using the data from the three independent replicates of the untreated control N 3T3-L1 adipocytes. In theory, there should be no differentially expressed genes in these comparisons. However, minor experimental errors and variations between RNA samples may introduce falsepositive results that would seem to indicate differentially expressed genes. The false-positive rate was 0.72% using the directional consistency criterion as described under "Experimental Procedures." Further filtering out of genes that displayed an FC of Ͻ2 (increase or decrease) decreased the falsepositive rate to 0.42%, and increasing the stringency to an FC of Ͻ3 (increase or decrease) resulted in a false-positive rate of only 0.13%. This high stringency ensured confidence in the reproducibility of the results and was therefore subsequently used to identify differentially expressed genes in insulintreated 3T3-L1 adipocytes. Once the analysis system was optimized to minimize false positives, the microarrays were screened for differentially expressed genes (Tables II-V). The differential expression of the various genes was qualitatively evaluated using the detection call and the CC score (see "Experimental Procedures"). The quantitative changes in gene expression are illustrated using the parameter FC, which represents the relative change in the abundance of a transcript after insulin treatment in relation to its untreated control. To minimize any bias in the interpretation of the data, we used statistical tools to assign differentially expressed genes to different classes (insulin-sensitive and insulin-resistant). Because the power of statistics increases as the number of replicates are increased, the analysis of only a few replicates that display relatively high variation may lead to the misleading conclusion that there is no difference between two treatments. To draw attention to this potential limitation, we listed the genes in Tables II-V in order of decreasing confidence in the results. The majority of the genes (i.e. those in the top 80% or so in each table) were classified with a high level of confidence. However, a few genes displayed rather large quantitative variation in response to insulin between experiments, and these are listed in the bottom of the tables. Thus, although the statistical analysis indicated no significant difference in the response of these genes to insulin in N and IR adipocytes, these data should be interpreted with some caution. Further validation experiments are needed to classify these genes with certainty. Table II summarizes the results for the genes that were up-regulated in response to insulin in both N and IR cells and for which there was no statistical difference between the FCs. The CC score was Ն2.5 for both the N and IR cells, indicating that these genes showed a reproducible response to insulin in all six experiments. It should be noted that, prior to insulin treatment, the basal level of expression of some genes (e.g. MCP-1) (17) was different in the N and IR cells. This difference was expected because the IR cells were previously treated with TNF-␣. Nevertheless, the genes in the two cell types responded in a similar way to the subsequent insulin treatment. The 38 genes in this category are ranked based on the average FC IR / FC N ratio. Analysis of the expression profiles revealed that insulin continued to induce expression (ϳ3-5-fold) of several immediate-early response genes (e.g. thrombospondin-1, Stra13, and Pip92), including a few transcription factors (Egr-1, Fra-1, and Egr-2) that were up-regulated by ϳ6 -20fold. Interestingly, some of the genes induced by insulin were not previously identified as insulin-responsive genes, including epiregulin and heme oxygenase-1 (ϳ90-and ϳ5-fold induction, respectively). The expression of the chemokines MCP-3 and MCP-1 also remained insulin-sensitive in IR adipocytes, and they were up-regulated by ϳ5-6-fold after insulin treatment. The potential role of MCP-1 in insulin resistance and obesity has been further investigated in a separate report (17). Table III lists the genes that passed the criteria for downregulated genes in response to insulin in N and IR cells. For these genes, the CC score was Ϫ2.5 or less for both N and IR cells, and there was no statistically significant difference in the FC between the two conditions. The 23 genes in this category are ranked in the same way as in Table II. Interestingly, insulin down-regulated the expression of the reverse cholesterol transporter ABCA1 (ATP-binding cassette transporter A1) by ϳ2-5-fold in N and IR adipocytes. A few receptors were also detected in this group (␤ 3 -adrenergic receptor, thyroid hormone receptor, and prostaglandin F receptor), and these were down-regulated by ϳ3-6-fold after insulin treatment. resistance in 3T3-L1 adipocytes Differentiated 3T3-L1 adipocytes were treated without (N) or with (IR) 3 ng/ml TNF-␣ for 3 days. Cells were washed twice with Dulbecco's modified Eagle's medium and incubated overnight in serum-free medium containing 0.2% bovine serum albumin. Insulin-stimulated glucose uptake was then determined as described under ''Experimental Procedures,'' and the data are expressed as the percentage of insulinstimulated uptake in normal cells set to 100%. Changes in expression of genes previously reported to be involved in insulin-stimulated glucose uptake and metabolism were monitored using Affymetrix U74A v2 microarrays. The quantitative changes in gene expression were illustrated using the signal parameter, which is proportional to the abundance of a transcript. Data are expressed as the mean Ϯ S.E. from three independent experiments. ND, not detected; IRS-2, insulin receptor substrate-2; PKC⑀ , protein kinase C⑀. Taken together, the data in Tables II and III support the original hypothesis that a number of genes respond to exogenous insulin in a similar manner in both N and IR adipocytes. These genes represent the novel gene family we set out to define. Tables IV and V list those genes that responded to insulin in N cells, but had an absent or significantly blunted response in IR cells, i.e. they are insulin-resistant genes. Table IV shows the 21 genes that passed the criteria for up-regulated genes (CC score of Ն2.5) in response to insulin in N cells, but displayed an absent or significantly blunted response in IR cells. The genes are listed starting with the genes that displayed the least similarity in the response to insulin in N versus IR cells. Among these are genes involved in intracellular signaling (Socs-3 and adenylate kinase-3/4) and gene regulation (junB and NgfiA-binding protein-2) and a chemokine (Gro-1). Table V shows the 19 genes that passed the criteria for down-regulated genes (CC score of Ϫ2.5 or less) in response to insulin in N cells, but displayed no or significantly less responsiveness in IR cells. The genes are listed in the same way as in Table IV. A variety of genes with diverse functions were identified in this group, including retinoic acid receptor-␥, transforming growth factor-␤ type I receptor, and matrix metalloproteinase-11. These genes were down-regulated by ϳ3-11-fold in N cells after insulin treatment. The identification of such IR genes not only emphasizes the unique behavior of the genes that remain in-sulin-sensitive, but also supports the usefulness of the TNF-␣ model of insulin resistance.
Confirmatory Real-time RT-PCR Analysis-To verify and validate the results obtained by microarray analysis, we also performed real-time RT-PCR analysis of some of the differentially expressed genes (Fig. 1). We selected both insulin-sensitive (epiregulin, Egr-1, Pip92, Glut-1, Fra-1, PAI-1, ABCA1, and ␤ 3 -adrenergic receptor) and insulin-resistant (Socs-3, junB, and matrix metalloproteinase-11) genes for subsequent real-time RT-PCR analysis. In addition to the genes identified in Fig. 1, we also confirmed the expression pattern of MCP-1 by real-time RT-PCR (17). Overall, the expression patterns obtained by real-time RT-PCR reflected the results obtained from GeneChip® analysis. In a few cases (e.g. Egr-1 and Fra-1), the FC in response to insulin treatment was lower using the Gene-Chips® compared with real-time RT-PCR. This behavior is, however, not unusual and has been previously observed, especially for transcripts that are highly regulated (28,29). Among the 12 genes (including MCP-1) tested and validated using real-time RT-PCR, PAI-1 was the only gene that was classified differently by RT-PCR analysis (i.e. insulin-responsive in IR adipocytes) (12) versus microarray analysis (insulin-resistant).
Role of Egr-1 in Insulin-induced Gene Transcription-Insulin has acute and delayed effects on gene expression and may act directly on some genes or indirectly on others by activating The genes indicated represent those that were unrecognized in the first experiment with IR cells by probe sets subsequently defined by Affymetrix to be faulty, but that were considered increased in the other five experiments with IR and N cells.
transcription factors, which, in turn, promote or inhibit gene transcription. Experiments were performed to begin to investigate potential downstream consequences of insulin on gene expression in this model of insulin resistance. Egr-1 was selected for further analysis not only because it remained insulinsensitive in the IR adipocytes, but also because it is a potent transcription factor in itself, one that regulates genes previously implicated in insulin resistance, obesity, and associated syndromes (e.g. thrombosis) (30,31). To explore the potential role of this transcription factor in mediating insulin-induced gene transcription, we used antisense MO to block Egr-1 translation. Initial time course experiments showed that insulin induced maximal Egr-1 mRNA expression at 1 h (data not shown). To allow time for EGR-1 protein synthesis and subsequent target gene activation, we selected the 3-h time point to investigate the effect of Egr-1 blockade on potential down-

TABLE III
Genes that are down-regulated by insulin and continue to respond to insulin in IR adipocytes N and IR 3T3-L1 adipocytes were incubated with or without 1000 nM insulin for 3 h. Differentially expressed genes were identified using Affymetrix U74A v2 microarrays as described under ''Experimental Procedures.'' Shown are the genes that were down-regulated in response to insulin treatment. The basal gene expression levels may differ between N and IR cells for certain genes. Nevertheless, there is not a statistically significant difference in the FC between N and IR cells. a The genes indicated represent those that were unrecognized in the first experiment with IR cells by probe sets subsequently defined by Affymetrix to be faulty, but that were considered decreased in the other five experiments with IR and N cells.

TABLE IV Genes that are up-regulated by insulin and demonstrate a blunted response to insulin in IR adipocytes
N and IR 3T3-L1 adipocytes were incubated with or without 1000 nM insulin for 3 h. Differentially expressed genes were identified using Affymetrix U74A v2 microarrays as described under ''Experimental Procedures.'' Shown are the genes that were up-regulated in response to insulin treatment and for which there is a significant difference in the FC between N and IR cells.  stream target genes. Control experiments demonstrated that treatment of 3T3-L1 adipocytes with Egr-1 antisense MO decreased the magnitude of insulin-induced expression of EGR-1 protein as measured by Western blotting of total cell lysates ( Fig. 2A). Tissue factor is one of the few target genes for EGR-1 detected in vivo (32), and it was previously reported that insulin induces tissue factor expression in 3T3-L1 adipocytes (33). The level of tissue factor mRNA after insulin treatment of 3T3-L1 adipocytes was analyzed using real-time RT-PCR. Fig.  2B shows that the insulin-induced expression of tissue factor was reduced to approximately base-line levels in cells pretreated with Egr-1 MO. Analysis of the mRNA levels of MCSF, another gene with EGR-1-binding sites within its promoter (34), showed that the Egr-1 antisense MO also blocked insulininduced MSCF expression by 40% (Fig. 2C). On the other hand, insulin-induced Fra-1 (an immediate-early response gene) mRNA expression was not affected by the MO treatment (Fig.  2D). Similar experiments were performed using IR 3T3-L1 adipocytes, and similar results were obtained (data not shown). Taken together, these results indicate that, at least in vitro, the insulin-induced expression of tissue factor and MCSF in differentiated adipocytes is mediated via EGR-1.

DISCUSSION
Insulin resistance is a condition that usually develops prior to clinically diagnosed type II diabetes (15,35). The resistance is characterized by an impaired ability of insulin to stimulate translocation of GLUT-4 from intracellular sites to the plasma membrane, with a subsequent decrease in cellular glucose uptake by adipocytes and skeletal muscle. Besides its critical role as a regulator of glucose homeostasis, insulin has a number of other biological effects, including the regulation of gene expression (9). Increasing evidence suggests that there may be signaling pathways that remain insulin-sensitive in IR states (12-14, 16, 17). If true, this notion may have significant biological consequences because obesity and insulin resistance are closely associated with hyperinsulinemia (15). Thus, the high levels of circulating insulin in these conditions may stimulate abnormal gene expression in insulin target tissues such as adipose, liver, and muscle.
The aim of this study was to test this hypothesis and to define more precisely the novel family of genes that continue to respond to insulin in metabolically IR adipocytes. To do so, mature 3T3-L1 adipocytes were treated with TNF-␣, a treat-TABLE V Genes that are down-regulated by insulin and demonstrate a blunted response to insulin in IR adipocytes N and IR 3T3-L1 adipocytes were incubated with or without 1000 nM insulin for 3 h. Differentially expressed genes were identified using Affymetrix U74A v2 microarrays as described under ''Experimental Procedures.'' Shown are the genes that were down-regulated in response to insulin treatment and for which there is a significant difference in the FC between N and IR cells. TGF-␤ , transforming growth factor-␤ ; MMP-11, matrix metalloproteinase-11.
Gene name Accession no.
a Not calculated since the CC score was greater than Ϫ2. b The genes indicated represent those that were unrecognized in the first experiment with IR cells by probe sets subsequently defined by Affymetrix to be faulty, but that were considered decreased in the other five experiments with IR and N cells.

FIG. 1. Effect of insulin on gene expression in vitro.
N and IR 3T3-L1 adipocytes were incubated without or with 1000 nM insulin for 3 h. Changes in relative mRNA levels were then determined in the N (open bars) and IR (closed bars) adipocytes using real-time RT-PCR. In each case, the data were normalized to the expression level of 18 S rRNA and are expressed as the relative mRNA level compared with the average expression level in N and IR cells, respectively, incubated without insulin (ϭ1). The error bars represent the S.E. (n ϭ 3). Statistical analysis revealed that, with the exception of junB expression in insulintreated IR cells, all changes in gene expression in insulin-treated cells were significantly different from those in control cells (p Ͻ 0.05). ␤3-AR, ␤ 3 -adrenergic receptor; MMP-11, matrix metalloproteinase-11. ment known to induce metabolic insulin resistance in vitro (17,19,26) and in vivo (27) and then analyzed using high density microarrays. Although the exact mechanisms by which TNF-␣ induces insulin resistance in adipocytes are not completely known, it appears that down-regulation of GLUT-4 expression and reduced kinase function of the insulin receptor may be important (36). Preliminary studies using this model demonstrated that TNF-␣ treatment resulted in a decreased rate of insulin-stimulated glucose uptake in these cells and in decreased expression of a number of genes, including those encoding GLUT-4, GLUT-1, PI 3-kinase, and protein kinase C⑀ (Table I). These genes were previously characterized as important players in insulin-stimulated glucose transport in adipocytes (9). Taken together, these observations provide compelling evidence that the cells we analyzed were indeed metabolically IR.
Based on previous studies from our laboratory (12, 17), we designed the experimental conditions to result in a robust response to insulin in a relatively short amount of time, i.e. adipocytes were treated with 1000 nM insulin for 3 h, thus allowing us to study the early effects of insulin on gene transcription. We also chose to analyze one time point repeatedly, instead of performing time course analysis. There are obvious advantages and disadvantages with both approaches. For example, several replicates of one time point improve the precision and confidence in the output data, but it may be more difficult to observe regulation of whole pathways or to define other consecutive events. One possible complication of using the relatively high concentrations of insulin in this study is the difficulty in excluding the contribution of signaling through the insulin-like growth factor-1 receptor. In this regard, it was recently reported that a substantial number of genes are regulated in a similar way through both the insulin receptor and the insulin-like growth factor-1 receptor in 3T3-L1 cells (37). This observation indicates that the dissection of signaling pathways has to be performed using sensitive techniques (other than microarrays) on a single gene by gene basis. In previous studies, we demonstrated gene-specific regulation using lower concentrations of insulin (10 nM), and we performed experiments using antibodies that block the insulin-like growth factor-1 receptor (12,17). These observations suggest that the insulin-like growth factor-1 receptor plays a minor role in the regulation of these genes. In our present study, we extended these observations and detected changes in Egr-1 expression using real-time RT-PCR at low insulin concentrations (data not shown). We believe that the microarray studies presented here should be used as a source from which selected single genes can be studied in more detail using a variety of techniques. Such studies will help to elucidate the specific signaling pathways that govern the expression of these genes and the biological significance of their altered expression in response to insulin.
Using these approaches, we identified a subset of genes that continued to respond normally to insulin despite the fact that the cells displayed significantly blunted insulin-stimulated glucose uptake (i.e. metabolic insulin resistance) (Tables I-III). Some of the insulin-responsive genes were previously shown to be regulated by insulin in mature 3T3 adipocytes, validating our experimental approach. These genes include PAI-1 (12), Glut-1 (38), Socs-3 (39), and the ␤ 3 -adrenergic receptor (40). In most instances, real-time RT-PCR confirmed the expression patterns observed by microarray analysis (Fig. 1). These observations lend support to our experimental approach and data analysis. However, there was one discrepancy between the data obtained by microarray analysis and real-time RT-PCR, and this concerned the expression of PAI-1. By microarray analysis, PAI-1 was classified as an IR gene, whereas RT-PCR analysis showed that the response to insulin was similar in N and IR adipocytes. These differences are difficult to reconcile. The target for the PCR amplification was nucleotides 809 -1348 in the MUSPAI1 locus. We considered the possibility that the primers used for the RT-PCR studies were problematic and therefore performed additional real-time RT-PCRs using other primer sets that amplified different regions of the PAI-1 mRNA (positions 144 -336 and 2389 -2547 in MUSPAI1). The results were similar to those presented in Fig. 1 and showed that PAI-1 remained insulin-sensitive in IR adipocytes. Thus, insulin stimulated PAI-1 mRNA levels by 14 Ϯ 2-fold in N cells and by 13 Ϯ 1-fold in IR cells when amplifying positions 144 -336. The corresponding FCs obtained when amplifying positions 2389 -2547 were 5.4 Ϯ 0.4 and 6.5 Ϯ 0.8 for N and IR adipocytes, respectively. Because the different PCRs generated similar results and based on previous work (12), we conclude that PAI-1 remains insulin-sensitive and was incorrectly classified as an IR gene in the microarray analysis. Despite this discrepancy, the fact that 11 (including MCP-1) (17) of the 12 genes analyzed (92%) behaved similarly in both assays indicates the reliability of the microarray approach.
Several of the genes identified in our microarray analysis as insulin-inducible genes were previously identified as immediate-early genes that were induced by adipogenic stimuli in 3T3-L1 preadipocytes (e.g. thrombospondin-1, Stra13, Egr-1, Pip92, and junB) (41). The former study was performed using a mixture of fetal bovine serum, methylisobutylxanthine, dexamethasone, and insulin to stimulate the cells. Our results indicate that similar effects on this subset of genes may be obtained using insulin alone to stimulate mature adipocytes. In addition, it was reported that CHO-K1 cells respond to insulin by increasing their abundance of mRNA for EGR-1, EGR-2, FRA-1, JunB, and MCP-1 (42), consistent with our results.
It is tempting to speculate that some of the novel genes identified as insulin-sensitive genes may play a role in the development of the insulin resistance syndrome. In this regard, we recently reported that MCP-1 is overexpressed in obesity, that it remains insulin-sensitive both in vitro and in vivo, and that it can induce metabolic insulin resistance and adipocyte dedifferentiation in vitro (17). Furthermore, thrombospondin-1, a secreted multifunctional extracellular matrix glycoprotein, is overexpressed in visceral adipose tissue of OLETF rats (43), and it was increased by ϳ4-fold in response to insulin in N and IR adipocytes (Table II). Thrombospondin-1 displays inhibitory activities against angiogenesis and may promote platelet aggregation (44), but its potential role in obesity and insulin resistance remains to be established. Another gene, Fra-1, a component of the AP-1 transcription factor complex, was recently reported to be induced in 3T3-L1 cells by a peroxisome proliferator-activated receptor-␥ ligand during adipocyte differentiation (45). Moreover, FRA-1 and JunD appear to form functional heterodimers in differentiated 3T3-L1 adipocytes. We observed an ϳ20-fold induction of Fra-1 expression in response to insulin by microarray analysis (Table II) and an ϳ90-fold induction using real-time RT-PCR (Fig. 1). In cultured 3T3-F442A adipocytes, insulin also activates AP-1 (46). The consensus AP-1 site (TGAg/cTCA) is present in a number of cellular promoters, and AP-1 has been implicated in a variety of processes in different cell types, including proliferation, apoptosis, differentiation, and growth arrest (47). Each of these processes contributes to the development and remodeling of adipose tissue.
We also identified another 23 genes that remained insulinsensitive, but their expression was repressed by insulin (Table  III). One of these genes, ABCA1, is a peripheral cholesterol efflux transporter that plays a significant role in reverse cholesterol transport and is a major determinant of high density lipoprotein serum concentration (48). Interestingly, the expression of the ␤ 3 -adrenergic receptor was also repressed by ϳ6-fold by insulin in N and IR adipocytes (Table III). Although this effect of insulin on N adipocytes was observed previously (40), the observation that ␤ 3 -adrenergic receptor expression remains insulin-sensitive in IR adipocytes was not reported. In vivo experiments showed that administration of insulin to ob/ob mice significantly reduced the expression of the ␤ 3 -adrenergic receptor in white adipose tissue (data not shown), indicating that the ␤ 3 -adrenergic receptor also remains insulin-sensitive in vivo in these metabolically IR animals. These observations may help to explain why ␤ 3 -adrenergic receptors are downregulated in obese hyperinsulinemic adipose tissue compared with the wild type (49).
Because of its broad effects on gene expression (30), the observation that the pro-inflammatory transcription factor EGR-1 remained insulin-sensitive in IR adipocytes was of interest not only in the context of these studies of abnormal gene expression, but also in the larger context of the known association of inflammation with insulin resistance and obesity (31). EGR-1 is a zinc finger transcription factor that belongs to a gene family that also includes EGR-2, -3, and -4 and WT1 (30). A variety of stimuli, including growth factors, cytokines, hypoxia, and physical forces, induce the expression of EGR-1 in cells. Once expressed, EGR-1 interacts with a consensus GCrich region in the DNA and transcriptionally activates a number of target genes, at least in vitro (30). However, despite these in vitro results, few target genes have been identified for EGR-1 in vivo. One such gene is tissue factor (32), a regulatory component of the coagulation system and a cardiovascular risk gene (50). Although there is no published link between EGR-1 and gene expression in adipocytes, several of the reported downstream target genes for EGR-1 are also expressed by the adipocytes. Importantly, in some cases, overexpression of these genes in adipose tissue is associated with obesity and insulin resistance (51,52). In fact, tissue factor is overexpressed in obesity (33). Thus, it is possible that EGR-1 could provide a link between the altered expression of this and other genes in obesity and the accompanying hyperinsulinemia. To assess the potential contribution of EGR-1 to adipocyte gene expression, we performed experiments using antisense MO to block EGR-1 translation. Fig. 2 shows that tissue factor was induced by insulin and that the Egr-1 antisense MO blocked this induction. Thus, tissue factor is as an EGR-1 target gene. MCSF may be another EGR-1-responsive gene. MCSF stimulates adipose tissue growth and adipocyte hyperplasia (53) and, based on its promoter sequence, was predicted to be an EGR-1 target gene (34). Fig. 2 also shows that insulin-induced MCSF expression was mediated, at least in part, by EGR-1.
Finally, our studies also revealed a number of genes that have not been reported previously to be regulated by insulin in adipocytes, including epiregulin, heme oxygenase-1, Gro-1, and matrix metalloproteinase-11. Further studies are necessary to evaluate the molecular basis for their regulation by insulin and to determine their role in adipocyte physiology/pathophysiology in vitro and in vivo. In addition, several ESTs were identified and classified as "insulin-sensitive" (Tables II and III) and "insulin-resistant" (Tables IV and V) based on their response to insulin in N and IR adipocytes. It will be interesting to further characterize these ESTs and to evaluate their functionality and potential contribution to conditions related to insulin resistance, obesity, and type II diabetes. Although very powerful, one limitation of the microarray technology is that it measures only mRNA levels. Novel proteomics technologies provide an attractive complement to the microarrays and make it possible to rapidly screen large numbers of potential targets also at the protein level.
In conclusion, our results show that 3T3-L1 adipocytes made insulin-resistant by treatment with TNF-␣ not only displayed reduced sensitivity to insulin as measured by insulin-stimulated glucose uptake (i.e. metabolic insulin resistance), but also could still respond normally to exogenous insulin in terms of the expression of a subset of genes. To establish the broader applicability of these in vitro observations, the results must be verified using other in vitro models of insulin resistance (e.g. non-esterified fatty acids, dexamethasone, glucosamine, etc.) and by in vivo experiments using obese insulin-resistant rodents during hyperinsulinemic/euglycemic clamp conditions. However, even without this information, one can speculate that the biological and pathological effects of this "selective insulin resistance" may be clinically important, especially in conditions characterized by hyperinsulinemia such as obesity, type II diabetes, and polycystic ovary syndrome. Knowledge about dysregulated gene expression, together with detailed studies on the relevant signaling pathways, may provide novel opportunities for rational drug development to delay and prevent the onset of these conditions and to reduce their clinical cost repercussions.