Transmembrane Tumor Necrosis Factor (TNF)-α Inhibits Adipocyte Differentiation by Selectively Activating TNF Receptor 1*

Tumor necrosis factor α (TNFα) is a potent cytokine with multiple biological activities and exists in two forms as follows: a 17-kDa soluble form that is a cleaved product of the 26-kDa transmembrane form (mTNFα). It has been suggested that the transmembrane form of TNFα is mainly responsible for localized responses via cell-cell contact. Here, we have examined the activities of transmembrane TNFα in cultured adipocytes. A non-cleavable transmembrane form of TNFα (mTNFΔ1–9K11E) was expressed in several preadipocyte cell lines using retroviral gene transfer. In wild type preadipocytes carrying both TNF receptors, expression of mTNFΔ1–9K11E resulted in inhibition of the differentiation program. The extent of this varied depending on the nature and strength of the adipogenic stimuli. The TNF receptor responsible for this function was determined by expressing mTNFΔ1–9K11E in preadipocyte cell lines lacking either TNF receptor 1 (TNFR1), 2 (TNFR2), or both. In order to confirm the results in the same cellular background, TNF receptors were also reconstituted in the cell lines lacking corresponding receptors. These experiments demonstrated that TNFR1 was necessary and sufficient for mediating mTNFΔ1–9K11E-induced inhibition of adipogenesis and that this action was similar to that of soluble TNFα. In conclusion, our results indicate that mTNFΔ1–9K11E is biologically active in cultured adipocytes and can alter the adipogenic program of these cells by selectively activating TNFR1. This may have physiological implications where local TNFα actions are thought to be generated at sites such as adipose tissue.

Tumor necrosis factor ␣ (TNF␣) is a potent cytokine with multiple biological activities and exists in two forms as follows: a 17-kDa soluble form that is a cleaved product of the 26-kDa transmembrane form (mTNF␣). It has been suggested that the transmembrane form of TNF␣ is mainly responsible for localized responses via cell-cell contact. Here, we have examined the activities of transmembrane TNF␣ in cultured adipocytes. A noncleavable transmembrane form of TNF␣ (mTNF⌬1-9K11E) was expressed in several preadipocyte cell lines using retroviral gene transfer. In wild type preadipocytes carrying both TNF receptors, expression of mTNF⌬1-9K11E resulted in inhibition of the differentiation program. The extent of this varied depending on the nature and strength of the adipogenic stimuli. The TNF receptor responsible for this function was determined by expressing mTNF⌬1-9K11E in preadipocyte cell lines lacking either TNF receptor 1 (TNFR1), 2 (TNFR2), or both. In order to confirm the results in the same cellular background, TNF receptors were also reconstituted in the cell lines lacking corresponding receptors. These experiments demonstrated that TNFR1 was necessary and sufficient for mediating mTNF⌬1-9K11E-induced inhibition of adipogenesis and that this action was similar to that of soluble TNF␣. In conclusion, our results indicate that mTNF⌬1-9K11E is biologically active in cultured adipocytes and can alter the adipogenic program of these cells by selectively activating TNFR1. This may have physiological implications where local TNF␣ actions are thought to be generated at sites such as adipose tissue.
Originally identified as a mediator of necrosis of certain tumor cells, tumor necrosis factor ␣ (TNF␣) 1 has now been shown to have a wide array of biological activities (1,2). It is also implicated in the pathogenesis of several diseases such as septic shock, rheumatoid arthritis, autoimmune disorders, and insulin resistance (1,3). TNF␣ is primarily produced by activated macrophages and lymphocytes but is also expressed in endothelial cells and other cell types including adipocytes (1,2,4). It exists in two forms as follows: a 17-kDa soluble form (sTNF␣) that is cleaved from the 26-kDa transmembrane protein (mTNF␣) at the cell surface by TNF␣-converting enzyme (5,6). Although the majority of TNF␣-induced responses has been attributed to sTNF␣, a few studies have shown that mTNF␣ is also biologically active and capable of mediating similar responses including apoptosis, proliferation, B cell activation, and some inflammatory responses (7,8). Furthermore, mTNF␣ has been implicated in some disease states such as experimental hepatitis where serum sTNF␣ levels were found to be within the normal range (9), indicating the relevance of localized TNF␣ responses. The existence of two different forms of TNF␣ makes its physiology more complicated. Furthermore, the fact that mTNF␣ relies on cell contact-dependent signaling may render the actions of mTNF␣ cell type-specific in vivo. In support of this, mTNF␣ has been reported to trigger inflammatory responses in astrocytes but not in neurons, whereas sTNF␣ can induce similar effects in both cell types (10).
The biological functions of both mTNF␣ and sTNF␣ can be signaled by two distinct TNF receptors: TNFR1 (55 kDa) and TNFR2 (75 kDa). The lack of homology in intracellular domains of two TNF receptors indicates that they can mediate distinct biological activities. Indeed, whereas a broad array of cellular responses has been attributed to TNFR1, many other effects are mediated by TNFR2 (11,12). These two receptors can also act in concert under many circumstances (9,13). The role of TNFR1 and TNFR2 in mediating the actions of sTNF␣ and the downstream signaling mechanisms has been studied extensively. In contrast, little information is available regarding the pathways and mechanisms utilized by mTNF␣. Some early studies have demonstrated that transmembrane TNF␣ is superior to sTNF␣ in activating TNFR2 (7,12). However, subsequent reports have indicated that transmembrane TNF␣ can signal through both receptors depending on the cellular context (8). Other studies have used TNFR-deficient mice to demonstrate that both receptors were required as in the case of experimental hepatitis (9) and arthritis (14), whereas TNFR2 alone is sufficient to mediate the effects of transmembrane TNF in experimental cerebral malaria (15).
Soluble TNF␣ plays an important role in regulation of energy metabolism. It has profound effects on adipocytes, including mobilization of triglycerides and inhibition of insulin action (2,3). In adipocytes, it can regulate the expression of several genes (4) and modulate the secretion of free fatty acids and leptin which play active roles in systemic energy balance (16,17). Recent studies demonstrated that TNF␣ is a candidate mediator of insulin resistance in obesity. The expression level of TNF␣ in adipose tissue is elevated in a variety of rodent obesity models (4) and also in obese humans (18,19). Soluble TNF␣ has been shown to inhibit insulin action in cultured adipocytes (20) and other cell types (21,22) as well as in whole animals (23)(24)(25). Several studies on various models of rodent obesity demonstrated increased insulin sensitivity upon genetic loss of TNF␣ function (26 -28), although one recent report could not demonstrate this in TNFR Ϫ/Ϫ R2 Ϫ/Ϫ mice with dietary obesity (29). Similar to genetic studies, pharmacological blocking of TNF activity also results in significant reversal of insulin resistance in obesity (4,30).
Despite strong evidence of a role for TNF␣ in obesity-related insulin resistance, circulating levels of sTNF␣ appear to be very low or undetectable (3). It is therefore possible that obesity might be one example where TNF␣ action is localized to the site(s) of production, such as adipose tissue. Thus, mTNF␣ may be a potential candidate mediator of such local events. However, the effects of mTNF␣ on adipocyte biology and energy metabolism remain unknown.
In this study, we have examined the effects of transmembrane TNF␣ on cultured 3T3-F442A adipocytes and determined the TNF receptor responsible for its functions by using TNFR1 Ϫ/Ϫ , TNFR2 Ϫ/Ϫ , and TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ preadipocyte cell lines developed in our laboratory. 2 These studies demonstrate that transmembrane TNF␣ is indeed biologically active in cultured adipocytes and that it alters the differentiation program of adipocytes by selectively activating TNFR1.

EXPERIMENTAL PROCEDURES
Cells and Reagents-TNFR1 Ϫ/Ϫ , TNFR2 Ϫ/Ϫ , and TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ fibroblast cell lines were established from day 16 -17 mouse embryos with targeted mutations in the corresponding TNFR(s) using the classic 3T3 protocol. Multiple fibroblast cell lines were established for each genotype and tested for their capacity to differentiate into adipocytes. For each genotype, one cell line with the highest rate of differentiation was selected and used for the experiments. 3T3-F442A, TNFR1 Ϫ/Ϫ , TNFR2 Ϫ/Ϫ , and TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ preadipocytes were grown in Dulbecco's modified Eagle's medium (DMEM, Life Technologies, Inc.) supplemented with 10% bovine calf serum (HyClone). Infected cells were maintained in the same medium in the presence of appropriate selection drugs. For differentiation, cells were seeded at 1.5 ϫ 10 5 per well on 6-well plates in DMEM supplemented with 10% cosmic calf serum (HyClone). Cells were grown to confluency and exposed to adipogenic reagents for 3 to 4 days, followed by culturing for 4 more days in medium containing insulin only. Recombinant murine soluble TNF␣ (Genzyme, MA) treatments were started at confluency and continued throughout the experiments with a new dose applied every 2 days at the indicated concentrations. Cells were then either stained with oil red O for microscopy or processed for RNA collection. Unless otherwise indicated, insulin was used at a concentration of 5 g/ml, dexamethasone at 1 M, isobutylmethylxanthine at 0.5 mM, and BRL49653 at 1 M. The polyclonal rabbit anti-murine TNF antibody was purchased from Genzyme (Cambridge, MA). The fluorescein-conjugated anti-rabbit IgG was purchased from Jackson ImmunoResearch (West Grove, PA). The polyclonal rabbit anti-human aP2 antibody was provided by Dr. Rex Parker (Bristol-Myers Squibb Co.). The polyclonal rabbit anti-rat Na,K-ATPase antibody was provided by Dr. Lorraine Santy (Harvard University). The polyclonal rabbit anti-murine ACRP30 antibody was provided by Dr. Philip Scherer (Albert Einstein College of Medicine, New York).
Cell Fractionation and Immunoblot Analysis-Fully differentiated adipocytes (1 ϫ 10 7 ) were collected from 10-cm dishes in the presence of breaking buffer (500 mM KCl, 250 mM sucrose, 25 mM Tris⅐HCl, pH 8.0, 2 mM EGTA, 5 mM EDTA, 2 g/ml aprotinin, 0.5 g/ml leupeptin, 2 M pepstatin, and 200 g/ml Pefabloc). After homogenization for 1 min at the speed of 25,000 rpm per min with a Polytron (model PT3000, Brinkmann), cell lysates were centrifuged at 4,700 ϫ g (Megafuge 3.0R, Heraeus Instruments) for 100 min, and the pellet containing plasma membranes was collected. The supernatant was further centrifuged at 450,000 ϫ g (L8-M ultracentrifuge, Beckman) for 2 h to precipitate the remaining subcellular membranes and harvest cytosolic material. The pellets were extracted in lysis buffer (1% Triton X-100, 50 mM Hepes, 100 mM sodium pyrophosphate, 100 mM sodium fluoride, 10 mM EDTA, 2 g/ml aprotinin, 0.5 g/ml leupeptin, 2 M pepstatin, and 200 g/ml Pefabloc) and centrifuged at 14,000 rpm in a microcentrifuge. The supernatants, which contain solubilized membrane proteins, were collected, and equal amounts of protein from each fraction were used for immunoblot analysis. To analyze secretion of sTNF␣, 48-h conditioned medium (10 ml) was collected from each cell line and concentrated to a final volume of 1 ml. The polyclonal rabbit anti-murine TNF antibody was used for immunoprecipitation, as described previously (26). Immunoblots were performed using polyclonal rabbit antibodies against human insulin receptor and aP2, mouse ACRP30, and rat Na,K-ATPase, respectively.
Quantitation of Differentiation-Cells were stained with oil red O (Sigma) for visualizing the lipid droplets and hematoxylin (Sigma) for nuclei according to conventional methods. The percentage of differentiation was calculated as number of cells containing visible lipid droplets divided by the total number of nuclei per microscopic field, under 400-fold magnification. Three representative fields were counted for each sample, and the mean Ϯ S.D. was used for comparisons.
Vector Construction-The cDNA of the noncleavable murine membrane TNF␣ (mTNF⌬1-9K11E) was provided by Dr. Els Decoster and Dr. Walter Fiers (Gent University, K. L. Ledeganckstraat, Belgium) in vector pSV235 (7). The coding region was amplified by polymerase chain reaction (5Ј primer, CTAGATCTCCCTCCAGAAAAGACA, and 3Ј primer, GGATCCAGAGTAAAGGGTCAGAGTG) and cloned into the PCRII vector (Invitrogen). The integrity of the PCR product was confirmed by sequencing. The coding region was then excised with XbaI/ BamHI digestion followed by Klenow fill-in. The 0.76-kilobase pair mTNF⌬1-9K11E cDNA fragment was cloned in sense orientation into the SnaBI site of the retroviral vector, pBabe-hygro, which contains the hygromycin B resistance gene (32). The cDNA of TNFR1 was obtained by performing reverse transcription-PCR (5Ј primer, TGCGAGGTCCT-GGAGGACC, and 3Ј primer, AAGGTTGTGGGTGTGGCTTTAT) using mouse spleen cDNA (strain C57BL/6). The final 1.37-kilobase pair PCR product was cloned into PCRII vector. One point mutation was detected by sequencing and corrected by site-directed mutagenesis based on the published sequence (33). The coding region of TNFR1 was excised with NaeI/EcoRI digestion and cloned into the SnaBI/EcoRI site of the retroviral vector, pBabe-puro, which contains the puromycin resistance gene (32). The cDNA of murine TNFR2 was obtained from Immunex (34). The coding sequence was excised with BamHI (followed by Klenow fill-in)/EcoRI digestion and cloned into the SalI (followed by Klenow fill-in)/EcoRI site of the retroviral vector, pBabe-bleo, which contains the bleomycin resistance gene (32). All the expression constructs were sequenced to confirm the integrity and correct orientation of the cloned cDNAs.
Transfection and Infection-Packaging of the viral particles was achieved by transfecting the expression plasmids into Bosc 23 cells, a human kidney cell line (35), with the Cell Phect calcium phosphate coprecipitation kit (Amersham Pharmacia Biotech). Forty-eight hours post-transfection, supernatant from packaging cells was collected and filtered through sterile 0.45-m syringe filters. Twenty-four hours before infection, recipient cells were seeded at a density of 2 ϫ 10 5 per 75 cm 2 . For infection, recipient cells were incubated with viral supernatant plus fresh DMEM (3:1) containing a final concentration of 4 g/ml Polybrene (Sigma). After a 24-h incubation, cells were fed with fresh DMEM and allowed to grow to 80% confluency in 2-3 days. Cells were re-seeded at a density of 6 ϫ 10 5 per 75 cm 2 for selection, and corresponding antibiotics were added the following day. Puromycin (Sigma) selection was completed in 3-4 days, while hygromycin B (Sigma) and zeocin (a derivative of bleomycin, Invitrogen) lasted 1 week and 1 month, respectively. Unless otherwise indicated, cells were maintained in appropriate antibiotics throughout experiments to maintain stable expression levels of mTNF⌬1-9K11E, TNFR1, and TNFR2.
Total RNA Preparation and Northern Blotting-RNA samples were extracted according to the guanidinium thiocyanate method (4). Following denaturation, RNAs were loaded on a 1% agarose gel containing 3% formaldehyde (4). After electrophoresis, RNAs were transferred to a biotran membrane (ICN), UV cross-linked, and baked at 80°C for 1 h. Hybridization with [␣-32 P]dCTP (NEN Life Science Products)-labeled cDNA probes and subsequent washings were done as described previously (4). Northern blots were quantitated by using NIH image program, and 18 S rRNA was used for loading adjustment.
Immunofluorescence-Cells were grown on coverslips in 6-well plates. After being rinsed 5 times with modified phosphate-buffered saline (PBS containing 1 mM MgCl 2 and 0.1 mM CaCl 2 ), cells were fixed in 3% paraformaldehyde. Following 5 min incubation in PBS containing 50 mM NH 4 Cl, cells were rinsed 3 times with PBS and 2 times with PBS containing 0.5% bovine serum albumin. This was followed by a 45-min incubation at room temperature in PBS containing 0.5% bovine serum albumin and 1:500 dilution of a rabbit anti-murine TNF antibody (Genzyme, MA). Cells were then washed 5 times with PBS and 2 times with PBS containing 0.5% bovine serum albumin. After a 30-min incubation with fluorescein-conjugated anti-rabbit IgG (Jackson ImmunoResearch), cells were washed 8 times with PBS, once with water and mounted with fluoromount-G (Southern Biotechnology Associates, Inc.). Photographs were taken under fluorescence microscope as described previously (36).
NF-B Activation-The mTNF⌬1-9K11E expression construct or control vector was cotransfected with a NF-B promoter-driven luciferase reporter gene (provided by Dr. Christopher K. Glass, University of California, San Diego) using LipofectAMINE-plus kit (Life Technologies, Inc.) The luciferase activity was determined by a luminometer and corrected for transfection efficiency as assessed by ␤-galactosidase assays.

Expression of a Non-cleavable Transmembrane Form of TNF␣ in Preadipocyte Cell
Lines-To study the potential effects of transmembrane TNF␣ in cultured adipocytes, we have ectopically expressed a non-cleavable transmembrane form of murine TNF␣ in several preadipocyte cell lines. These include the 3T3-F442A cells and the newly developed preadipocyte cell lines deficient in TNFR1, TNFR2, or both. 2 The non-cleavable form of TNF␣ (mTNF⌬1-9K11E) has been generated by deleting amino acids 1-9 and mutating residue 11 from Lys to Glu in the N-terminal part of mature TNF␣ and previously characterized in L929, CT6, PC60-R55/R75, and U937 cells (8). In this study, we chose a retroviral expression system to express constitutively the non-cleavable mutant (Fig. 1A) because of its high integration efficiency. This system also allowed expression in a large number of cells and prevented the common problem of clonal variability. The exogenous TNF␣ message was readily identified in infected cells since it is larger than the endogenous one produced by LPS-stimulated Raw264.7 macrophages (Fig. 1B). Comparable levels of mTNF⌬1-9K11E expression were demonstrated by Northern blot analysis in all stably infected preadipocyte cell lines (Fig. 1B). Messenger RNA levels of both TNF receptors were also determined in all cell types to demonstrate the presence and potential regulation of the relevant TNF receptors. The TNFR2 mRNA level in 3T3-F442A cells expressing mTNF⌬1-9K11E was substantially higher than those expressing vector alone (Fig. 1B, lanes  3 and 4). Expression of mTNF⌬1-9K11E in TNFR1 Ϫ/Ϫ cells did not affect TNFR2 mRNA level (Fig. 1B, lanes 5 and 6). This indicates that either the regulation of TNFR2 mRNA expression by mTNF⌬1-9K11E requires the presence of TNFR1, or base-line TNFR2 expression in these cells is already at maximum levels. Endogenous TNFR1 mRNA levels were unaffected by the presence of mTNF⌬1-9K11E.
Proper expression and localization of mTNF⌬1-9K11E as a membrane protein was determined by both indirect immunofluorescence in intact cells and immunoblot analysis of cellular fractions. As shown in Fig. 1C, expression of mTNF⌬1-9K11E protein on TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ cells could be visualized by the use of a polyclonal rabbit anti-murine TNF␣ antibody (left panel) followed by a fluorescein-conjugated anti-rabbit IgG. The peripheral distribution of fluorescence is consistent with plasma membrane-associated localization. To confirm this further, we performed immunoblot analysis of cellular fractions of adipocytes (Fig. 1D, top panel) which showed that mTNF⌬1-9K11E was predominantly detected in a high density membrane-containing fraction. This fraction is enriched with plasma membranes as confirmed by the detection of Na,K-ATPase (Fig. 1D, middle panel) which is a commonly used plasma membrane marker (37). Insulin receptor was also detected in this fraction (data not shown). The mTNF⌬1-9K11E protein was expressed as a 25-kDa protein compared with the wild type 26-kDa mTNF␣. In both mTNF⌬1-9K11E and wtTNF␣-expressing cells, two additional smaller molecular weight bands have been consistently detected in the membrane fractions. These are likely to be products of alternative initiation sites as previously reported (38). Transmembrane TNF⌬1-9K11E protein level in total cell extracts from different cell lines was also similar as determined by immunoblotting (data not shown). The protein expression level of the non-cleavable mutant in our system is estimated to be 5% of the endogenous counterpart produced by LPS-stimulated macrophages quantitated by densitometry scanning of immunoblots. No mTNF⌬1-9K11E protein was detected in cytosol. This fraction is indeed enriched with cytosolic proteins as confirmed by immunoblotting for cytosolic adipocyte fatty acid-binding protein, aP2 (Fig.  1D, bottom panel). Finally, we also examined whether in our experimental system, the mTNF⌬1-9K11E could be aberrantly cleaved at other sites to yield sTNF␣ products. Immunoblot analysis of concentrated conditioned media did not reveal any detectable TNF␣ immunoreactivity from vector and mTNF⌬1-9K11E-expressing cells but did show the presence of sTNF␣ from wtTNF␣-expressing cells (Fig. 1E, top panel). ACRP30/ AdipoQ (adipocyte complement-related protein of 30 kDa) (39,41), a secreted protein exclusively made in adipocytes, was used as a control for proper protein secretion to the conditioned media from these cells (Fig. 1E, right panel). These data demonstrated that mTNF⌬1-9K11E was expressed on the cell surface and did not produce detectable sTNF␣. These results are essentially identical to those observed in lymphocytes using the same construct (8).
Effects of mTNF⌬1-9K11E on the Differentiation of Adipocytes-The 3T3-F442A preadipocyte cell line is commonly used as an experimental model for adipocyte differentiation in vitro (40,42). These cells express both TNF receptors but do not produce detectable levels of endogenous TNF␣ (Fig. 1B). In these cells, we first examined whether ectopic expression of mTNF⌬1-9K11E could lead to alterations in the differentiation process. Four different permissive conditions for adipocyte differentiation were used to compare control 3T3-F442A cells to those expressing mTNF⌬1-9K11E. The induction conditions used were as follows: (a) insulin alone (5 g/ml); (b) insulin, dexamethasone (1 M), and isobutylmethylxanthine (0.5 mM); (c) insulin and BRL49653 (1 M), an activator for the adipogenic transcription factor peroxisome proliferator-activated receptor ␥ (PPAR␥); and (d) a mixture containing all four of the above reagents. After the induction of differentiation with these reagents for 3 days, cells were maintained in 5 g/ml insulin for 4 more days. Experiments were then stopped for morphological and molecular comparison of vector-infected control cells with those expressing mTNF⌬1-9K11E. In the absence of any adipogenic stimuli, 6 Ϯ 0.6% of control 3T3-F442A cells spontaneously differentiated into adipocytes, whereas no morphological sign of differentiation was detectable in mTNF⌬1-9K11E-expressing cells (Fig. 2A). The use of insulin as the only inducer resulted in the differentiation of 37 Ϯ 2.9% control cells, but this effect was completely blocked by the presence of mTNF⌬1-9K11E ( Fig. 2A). When a mixture of insulin, dexamethasone, and isobutylmethylxanthine was used, all control F442A cells differentiated. However, only 36 Ϯ 2.1% of mTNF⌬1-9K11E-expressing cells differentiated into fat cells under this condition. The addition of BRL49653, a thiazolidinedione compound which acts as a high affinity ligand for PPAR␥, significantly reduced the effect of mTNF⌬1-9K11E on adipocyte differentiation. The effect of BRL49653 was incomplete (18 Ϯ 3.7% differentiation) when used with insulin alone but complete when used in combination with insulin, dexamethasone, and isobutylmethylxanthine. To compare these effects of mTNF⌬1-9K11E to sTNF␣, 3T3-F442A cells were also treated with recombinant murine soluble TNF␣. At a concentration of 1 ng/ml, sTNF␣ generated inhibitory effects similar to those observed with the mTNF⌬1-9K11E. The percentages of differentiated cells under the four different induction conditions were 0% (insulin), 26 Ϯ 5.7% (insulin and BRL49653), 41 Ϯ 6.3% (insulin, dexamethasone, and isobutylmethylxanthine), and 99 Ϯ 3.5% (combination of all reagents). At a concentration of 10 ng/ml, sTNF␣ completely blocked differentiation under all conditions (data not shown). During differentiation, experiments with both mTNF⌬1-9K11E and soluble TNF␣, cells were closely monitored every day under microscope, and no obvious cytotoxic effect was observed. At the end of differentiation, trypan blue uptake was performed to examine cell viability, and no difference was observed between vector-and mTNF⌬1-9K11E-infected cells. The morphological changes that occur during adipocyte differentiation are accompanied by changes in expression patterns of fat-specific genes, most of which are involved in creating and maintaining the adipocyte phenotype (42). Therefore, expression of a panel of adipose-specific genes can be used to serve as molecular indicators of the state of differentiation. In order to evaluate the effects of mTNF⌬1-9K11E on adipocyte differentiation at the molecular level, we next examined the mRNA levels of four genes that are expressed in a differentiationdependent manner in adipocytes. These were the adipogenic transcription factor PPAR␥, the adipocyte fatty acid-binding protein aP2, the serine protease adipsin, and the insulin-dependent glucose transporter 4 (Glut4). Northern blot analyses were consistent with the morphological changes described above. The expression levels of all of these genes were also significantly decreased in mTNF⌬1-9K11E-expressing cells compared with the controls. Fig. 2B (left panel) shows the comparison of the expression levels of PPAR␥, Glut4, adipsin, and aP2 in uninfected, vector-infected, and mTNF⌬1-9K11Eexpressing cells. The presence of mTNF⌬1-9K11E completely inhibited expression of these genes in cells induced to differentiate with insulin alone (lanes 7-9), whereas no obvious difference was observed between uninfected and vector-infected F442A cells. Addition of BRL49653 or dexamethasone and isobutylmethylxanthine dramatically increased the expression of all of these fat-specific genes in uninfected and vector-infected cells. This also partially antagonized the effect of mTNF⌬1-9K11E on adipocyte differentiation (Fig. 2B, lanes  10 -14). The combination of all adipogenic reagents completely prevented the effect of mTNF⌬1-9K11E, and no difference in gene expression could be detected between vector-infected and mTNF⌬1-9K11E-expressing cells under this condition (Fig.  2B, lanes 15 and 16). However, when cells were induced to differentiate in the presence of continuous selection pressure, the inhibitory effect of mTNF⌬1-9K11E could still be observed even when the strongest induction mixture was applied (Fig.  2C). This was consistent with a higher expression level of mTNF⌬1-9K11E in the presence of continuous selection (Fig.  2C, bottom panel). For these reasons, continuous antibiotic selection was used throughout the remaining experiments described below.
We also examined the effect of soluble TNF␣ on adipocyte differentiation in these cellular systems. Recombinant murine TNF␣ was applied every 2 days at two doses of 1 and 10 ng/ml, respectively. Treatment of TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ or TNFR1 Ϫ/Ϫ cells with sTNF␣ did not affect differentiation at both concentrations used in experiments (data not shown). In contrast, in the presence of 1 ng/ml sTNF␣, differentiation was evident in only 53.3 Ϯ 6.4% of the TNFR2 Ϫ/Ϫ cells. At the concentration of 10 ng/ml, sTNF␣ almost completely inhibited adipocyte differentiation (Ͻ1% differentiation). These data indicate that both sTNF␣ and mTNF␣ can inhibit adipocyte differentiation through TNFR1, and the extent of inhibition is dose-or expression level-dependent, respectively. mTNF⌬1-9K11E Inhibits Adipocyte Differentiation through TNFR1 Alone-To confirm the role of TNFR1 in mediating the anti-adipogenic effect of mTNF⌬1-9K11E within the same cellular background, we next introduced intact TNFR1 back into the TNFR1 Ϫ/Ϫ preadipocytes expressing mTNF⌬1-9K11E. Since these cells are already resistant to neomycin and hygromycin B, the retroviral vector containing puromycin resistance gene was used for exogenous TNFR1 expression. Four stablyinfected TNFR1 Ϫ/Ϫ cell lines were established, expressing 1) vectors with hygromycin B and puromycin resistance genes, 2) vector with hygromycin B resistance gene and TNFR1, 3) mTNF⌬1-9K11E and vector with puromycin resistance gene, and 4) mTNF⌬1-9K11E and TNFR1. In cells expressing both mTNF⌬1-9K11E and TNFR1, exogenous TNFR1 expression levels were significantly lower than those expressing only TNFR1 (Fig. 4). Since direct regulation of the exogenous gene is not expected, it is likely that overexpression of high levels of both mTNF⌬1-9K11E and TNFR1 simultaneously is cytotoxic, so only cells with low TNFR1 expression levels survived the selection protocol. After the initial selection period, no cytotoxicity was observed when cells were kept growing in appropriate antibiotics. All cell lines were also tested for differentiation. As shown in Fig. 4, mTNF⌬1-9K11E-induced inhibition of adipogenesis was only observed when both mTNF⌬1-9K11E and TNFR1 were expressed simultaneously in TNFR1 Ϫ/Ϫ cells.

DISCUSSION
The wide array of biological actions of TNF␣ is regulated at many levels. The presence of both transmembrane and secreted forms of functional TNF␣ ligands adds a spatial mode of control to TNF␣ actions. As the local actions of this molecule are recognized in both physiological and pathological states, understanding the biology of its cell surface-associated form becomes more critical. However, in contrast to the well characterized function and signaling of sTNF␣, information regarding mTNF␣ is still limited.
The actions of sTNF␣ on adipocytes have also been extensively investigated. Numerous reports have shown that sTNF␣ has strong negative effects on adipocyte differentiation (43)(44)(45)(46). Furthermore, the use of human TNF␣ in cultured murine adipocytes has indicated that TNFR1 can mediate this effect (43,44). On the other hand, the biological activities and signaling mechanisms of mTNF␣ have not yet been determined in this respect. In this study, we have generated several preadipocyte cell lines stably expressing a non-cleavable form of TNF␣ (mTNF⌬1-9K11E) to examine its actions in adipocytes. These studies demonstrate that mTNF⌬1-9K11E is biologically active in several independent preadipocyte cell lines and can induce marked alterations in the adipocyte differentiation program. In wild type cells carrying both functional TNF receptors, expression of mTNF⌬1-9K11E led to significant inhibition of terminal differentiation into adipocytes. The biological activities generated by mTNF⌬1-9K11E in our experimental system are unlikely to be supra-physiological since the protein expression level of the noncleavable mutant is about 5% of the endogenous counterpart produced by LPS-stimulated macrophages.
Our parallel experiments with sTNF␣ generated results consistent with previous reports and demonstrated that adipocyte differentiation can be altered similarly by both mTNF⌬1- 9K11E and sTNF␣ under the experimental conditions used in this study. In addition, the observation that the use of BRL49653 can reverse the anti-adipogenic effects of mTNF⌬1-9K11E is also consistent with previous reports, which demonstrated that similar compounds could block the inhibitory effects of sTNF␣ on adipocyte differentiation (47,48). The extent of inhibition of differentiation by both mTNF⌬1-9K11E and sTNF␣ was dependent on the amount of ligand and on the strength of the induction conditions. Under the experimental conditions where mTNF⌬1-9K11E expression is stable, it is possible that stronger induction conditions could deliver stronger or multiple adipogenic signals, which cannot be completely blocked by a constant level of mTNF⌬1-9K11E. Consistent with this, we observed that high levels of both sTNF␣ (10 ng/ml) and mTNF⌬1-9K11E (obtained by continuous selection, Fig. 2C) can still block adipocyte differentiation even under the most permissive condition. Therefore, it is reasonable to assume that there is a balancing point between adipogenic and anti-adipogenic signals, the outcome of differentiation will be determined by the relative strength of signals from either side.
In this study, we have also determined the TNF receptor responsible for mediating this action of mTNF⌬1-9K11E by utilizing newly developed preadipocyte cell lines deficient in TNFR1, TNFR2, or both receptors. These cells can differentiate into adipocytes with high efficiency and thus provided a valuable experimental system to study signaling through each TNF receptor. Expression of mTNF⌬1-9K11E in TNFR1 Ϫ/Ϫ , TNFR2 Ϫ/Ϫ , and TNFR1 Ϫ/Ϫ R2 Ϫ/Ϫ preadipocyte cell lines demonstrated that TNFR1 is necessary for the inhibitory action of mTNF⌬1-9K11E on adipocyte differentiation. By introducing TNFR1 or TNFR2 back into the TNFR1 Ϫ/Ϫ or TNFR2 Ϫ/Ϫ cell lines, respectively, we have further demonstrated that mTNF⌬1-9K11E selectively acts through TNFR1 and signaling through this receptor is necessary and sufficient for its effects. This latter experiment also allowed us to use the identical cellular background to examine the possibility of cooperative signaling between TNFR1 and TNFR2. Reconstitution of TNFR1 in TNFR1 Ϫ/Ϫ preadipocytes restored the inhibitory effect of mTNF⌬1-9K11E on differentiation. In contrast, overexpression of TNFR2 in TNFR2 Ϫ/Ϫ preadipocytes did not enhance this inhibitory action of mTNF⌬1-9K11E mediated by TNFR1. Therefore, we conclude that mTNF⌬1-9K11E inhibits adipocyte differentiation by selectively activating TNFR1.
The fact that mTNF⌬1-9K11E does not utilize TNFR2 to inhibit adipocyte differentiation is somewhat unexpected since other studies have shown that it engages TNFR2 (15) or both TNFR1 and TNFR2 (9,14). TNFR2 is expressed in both preadipocytes and adipocytes and is elevated in obesity (4,49). In human adipocytes, it plays a complementary role in sTNF␣mediated inhibition of insulin receptor signaling through TNFR1 (50). However, it is not involved in inhibition of adipogenesis by either transmembrane or soluble TNF␣. Taken together, these observations suggest that mTNF␣ can engage both TNFR1 and TNFR2, but receptor selectivity or biological outcome may be dependent upon the cell type or the underlying pathophysiology or possibly the expression level of the trans- Finally, earlier studies have shown that TNFR1 can mediate other sTNF␣ functions in adipocytes, in addition to signaling the anti-adipogenic activity. These include modulation of leptin production (51,52) and inhibition of insulin signaling in murine adipocytes (31). Further studies will be necessary to elucidate whether mTNF␣ exerts similar actions, and selective usage of TNFR1 is a general mechanism in adipocytes.
The fact that TNF␣ is biologically active in adipocytes when retained on the cell surface makes it a candidate mediator of other local TNF␣-induced responses. It is tempting to speculate that this might potentially be relevant in disease states involving adipose mass and/or altered local levels of tissue TNF␣, such as obesity, or lipodystrophies. It is also possible that mTNF␣ produced in the stroma-vascular component of adipose tissue, from either preadipocytes or macrophages, could also affect adipocytes through cell-cell contact. Thus, if mTNF␣ is active in mediating other effects that influence the metabolism and/or number of adipocytes, it might also have a strong impact on systemic energy metabolism. However, the limitations of the applicability of the studies in cultured cells to whole animals are obvious. Further in vivo studies, including gain of function transgenic mice, are needed to address the role of this form of TNF␣ in regulating adipocyte biology locally under physiological or pathological conditions.