The active form of tumor necrosis factor is a trimer.

Natural human and recombinant human and murine tumor necrosis factors (TNF) were fractionated by gel filtration chromatography on Sephadex G-75. The active form of TNF was identified by its inhibitory activity in receptor binding assays with HeLa cells and was eluted as a protein of Mr approximately 55,000. Radioiodinated human and murine TNF were fractionated by gel filtration into a major peak of Mr approximately 55,000, corresponding to a trimer, and a minor peak of Mr approximately 17,000, corresponding to a monomer. Binding assays showed that the timer was at least 8-fold more active than the monomer. The human TNF partially dissociated into monomers upon addition of the nonionic detergent Triton X-100. Isolated monomers showed low binding affinity (KD = 70 nM) and reduced cytotoxicity, whereas trimers showed high binding affinity (KD = 90 pM) and cytotoxicity. When 125I-TNF was bound to cells, no release of monomer was detectable, suggesting that the trimer could directly bind to cellular receptors without dissociating into subunits. Further evidence for such binding was obtained by cross-linking 125I-TNF trimers with bis[2-(succinimidooxycarbonyloxy)ethyl]sulfone. These trimers were bound to HeLa cells, could be dissociated from cellular receptors, and elicited a cytotoxic response. These results show that trimers, whether native or cross-linked, bind to receptors and are the biologically active form of TNF.

Numerous reports demonstrate that TNF interacts with cellular receptors (2, [12][13][14] and elicits cytotoxic (15; 16) or growth regulatory responses (17,181. It is not known, however, which form of TNF interacts with receptors. The goal of the present work was to examine whether hTNF and mTNF are under physiological conditions oligomers of defined size and to establish whether such oligomers directly bind to receptors. We report in this communication that TNF trimers bind to cellular receptors and elicit a cytotoxic response.

MATERIALS AND METHODS
Cytotoxicity Assay-HeLa S2 cells were grown in monolayer cultures in Dulbecco's medium supplemented with 10% heat-inactivated horse serum. For each assay, 4 X IO5 cells were resuspended in 0.2 ml of culture medium containing 5 pg/ml cycloheximide and the indicated concentrations of hTNF. After 18 h, the medium containing dead cells was removed and adherent cells were stained with 0.2% crystal violet in 2% ethanol (19). The dye was solubilized with 33% acetic acid, and the Ab,,, was measured with a Titertek Multiscan (Flow Laboratories). Cytotoxicity was expressed as a percentage of the Am of control cells that received cycloheximide alone.
Binding Assays-In competitive binding experiments, 0.3-0.5 ng of radioligand were incubated 5 h at 4 "C with 1 X lo6 cells in 0.15 ml of medium containing 5 mM MgClz and 40 mM HEPES, pH 7.5, as previously described (20). In experiments designed to recover cellhound radioligand, the binding assays were proportionately increased 10-fold to 1.5 ml. Following binding, the cells were centrifuged at 4 "C and washed twice with 1 ml of PBS. To dissociate TNF-receptor complexes, 5 ~l of 6 M GdnHC1, 0.1 M sodium phosphate, pH 7.5, were added to lo7 pelleted cells for 10 min at 4 "C. Lysed cells were diluted with 0.1 ml of 10 mM PBS, 0.1% BSA and centrifuged for 10 min at 15,500 X g. The supernatants were chromatographed on Sephadex G-75 columns. Bindability was determined by incubating radioligands with graded amounts of excess cellular receptors, and the results were expressed as the maximum percentage of counts added that were specifically hound to HeLa S2 cells at 4 "C. Zonal Centrifugation Studies-0.3 ng of 'T-hTNF and 0.5 mg of reference protein in 0.1 ml of PBS were layered on 5-20% linear sucrose gradients. The gradients were centrifuged for 24 h at 44,000 rpm in an SW 50 rotor at 5 "C, and fractions containing 4 drops were collected. Ovalbumin (Mr = 45,000) and BSA (M, = 66,000) were used as reference proteins.

RESULTS
Gel filtration on Sephadex G-75 was used to determine the M , value of native TNF and to establish whether iodination altered its size. Accordingly, the elution profiles of native hTNF and mTNF were compared with those of 'T-hTNF and T -m T N F (Fig. 1). Since small amounts of TNF were chromatographed in the presence of carrier protein, the TNF was localized by its inhibitory activity in receptor binding assays. Both recombinant TNF eluted as a single major peak (Fig. L4). The elution profile of natural hTNF (a gift of Dr. Walter Fiers, University of Ghent) was indistinguishable from that of recombinant hTNF (data not shown). Radioiodinated hTNF and mTNF eluted as a major peak (#I) in corresponding fractions, followed by a minor peak (#2) and free lZ5I ( and 17,000, respectively, corresponding to a trimer and a monomer. This interpretation was confirmed by zonal sedimentation studies using lZ5I-hTNF in isokinetic 5-20% sucrose gradients (21). The major peak sedimented as a globular protein of M , = 54,650 f 2,340 in three independent analyses. The TNF in the monomer peak appeared to compete poorly in the receptor binding assay, since it was not detected in the chromatogram shown in Fig. 1 A . To confirm this finding, lz5I-hTNF trimer and small amounts of monomer were isolated by gel filtration. 10,000 cpm of each fraction were tested for binding to HeLa cells; 1,180 f 90 cpm of trimer were bound in a standard assay, whereas only 140 -+ 20 cpm of monomer were specifically bound. In subsequent experiments, the biological activity of trimer and monomer was compared in cytotoxicity assays. Since monomer was recovered in relatively small amounts in physiological solutions, different procedures were tried to obtain sufficient quantities of this species. It was thus found by gel filtration analysis that '"1-hTNF trimer partially dissociated upon addition of low concentrations of the nonionic detergent Triton X-100. This dissociation was dependent on hTNF concentration, since the monomer/trimer ratio increased at low hTNF concentration (Fig. 2). Monomers were separated from trimers by gel filtration in a Sephadex G-75 column (0.7 x 24 cm) pre-equilibrated with either PBS, 0.1% BSA for binding assays or culture medium for cytotoxicity assays. Triton X-100 was not detected (22) in the trimer or monomer peak, but was eluted at a greater column volume than hTNF monomer. In separate experiments (data not presented), it was shown by gel filtration analysis that hTNF monomers (-1 ng/ml) prepared in this manner quantitatively reassociated to trimers when the concentration of hTNF was increased 500-fold by adding unlabeled hTNF. Therefore, hTNF trimers can be dissociated by the addition of Triton X-100 into monomers, which are relatively stable in dilute solutions but readily reassociate to trimers when the hTNF concentration is raised.
Pooled fractions of hTNF trimer and monomer were subsequently compared in competitive binding and cytotoxicity assays on an equal counts/min basis. The monomer fraction and monomer (0) fractions were isolated by gel filtration as described in the legend to Fig. 2 showed low binding activity and cytotoxicity compared to the trimer fraction. Binding of monomer was about 5.5-fold lower than that of trimer, as determined in competition binding assays (Fig. 3A). Scatchard plots of these data showed that the trimer was bound with a KD = 90 PM, whereas only a small component of the monomer fraction was bound with such high affinity (Fig. 3A, inset). Most of the monomer was bound with low affinity (KO = 70 nM). It seems unlikely that this binding has biological relevance, since 50% cytotoxicity of HeLa cells is observed with 2 pM hTNF (20). In parallel cytotoxicity assays, a monomer concentration 6-7-fold greater than that of trimer was needed to elicit the same biological response when tested at low concentrations (Fig. 3B). However, at the highest concentrations tested, the cytotoxicity of monomer was nearly equivalent to that of trimer. Rechromatography of the monomer fraction at the end of the incubation period showed the presence of about 10% trimer (data not shown), which could account for both the small component binding with high affinity and for the cytotoxicity at the highest concentrations tested. These results indicated that hTNF trimer binds with higher affinity to receptors and has greater cytotoxic activity than hTNF monomer.
In the following experiment, we examined whether hTNF dissociates into monomers upon binding to receptors. A binding assay was carried out a t 4 "C with low '*'I-hTNF concentration, and the supernatant obtained after spinning out the cells was analyzed by gel filtration. The trimer peak was reduced in proportion to the '"I-hTNF bound to the cells, but no increase in the monomer peak could be detected (Fig.  44). This result suggested that hTNF trimers could directly bind to receptors, but it could not be excluded that monomers or dimers were binding and that the subunits released were reassociating into trimers. Therefore, to demonstrate that trimers can bind to cells and have biological activity, the hTNF was cross-linked to prevent its dissociation.
The ""I-hTNF was reacted with the cross-linking reagent BSOCOES and compared to control "'I-hTNF by gel filtration chromatography, binding to HeLa cell receptors, and cytotoxicity assays. The cross-linked ' T -h T N F eluted with M , -55,000 even after treatment with 3 M GdnHCl (Fig. 4B). This demonstrated that cross-linking stabilized hTNF against dissociation. In contrast, 85% of the control l2'1-hTNF treated with 3 M GdnHCl eluted with an M , = 17,000 (Fig. 4C). This dissociation was in large part reversible, since after dialysis '"I-hTNF eluted as a trimer (Fig. 40). These experiments showed that hTNF cross-linked with BSOCOES is a trimer resistant to dissociation by relatively strong denaturing reagents, such as GdnHCl. However, drastic denaturing treatment of the cross-linked trimer, such as boiling in 1% sodium dodecyl sulfate under reducing conditions, resulted in partial dissociation into dimers and monomers, as judged by gel electrophoresis (Fig. 4, inset).
A binding assay carried out with cross-linked hTNF showed that it could bind to TNF receptors of HeLa cells. These cells were treated with 3 M GdnHCl to release bound BSOCOES-'2'I-hTNF, and the supernatant was analyzed by gel filtration. A single peak of radioactivity was present (Fig. 4E), demonstrating that the bound hTNF could be eluted from HeLa cell receptors as a trimer. In order to determine whether crosslinking or acylation of amino groups had altered its binding to cell receptors, BSOCOES-'"I-hTNF was compared with '251-hTNF in competitive binding assays (Fig. 5). Since BSOCOES-'2sI-hTNF had lower bindability (39% compared to 45% for 'T-hTNF), equivalent amounts of bindable radioligands were added. The binding of both ligands was inhibited in a parallel manner by unlabeled hTNF, but only half as much unlabeled hTNF was required for 50% competition of BSOCOES-12sI-hTNF. This indicated that chemical changes introduced by the cross-linking reagent resulted in partial loss of binding activity. However, binding was 90% specific for both ligands. In agreement with the loss of binding activity, the cytotoxicity of BSOCOES-""I-hTNF for HeLa cells was on average 4.5-fold less than that of control l2'I-hTNF. A similar 5-10-fold decrease in cytotoxicity of BSO-COES-""I-hTNF was observed in experiments with SK-MEL-109 melanoma cells (data not shown). These results with cross-linked T N F confirmed the findings with native T N F by showing that stable TNF trimers bind to cellular receptors and elicit a biological response.

DISCUSSION
Natural human TNF, recombinant hTNF and mTNF, and cross-linked "'I-hTNF coelute in gel filtration under nondenaturing conditions as a major peak with an apparent M , -55,000 ( Figs. 1 and 4). The formation of homotrimers from 17,100 monomers (1) gives a predicted M , = 51,300, which is in fairly good agreement with the M , value obtained from gel filtration or zonal sedimentation. Furthermore, after crosslinking I2'I-hTNF with BSOCOES, radioactive bands corresponding to trimers, dimers, and monomers are observed by gel electrophoresis. Cross-linking data for other oligomeric proteins similarly show that incompletely cross-linked homotrimers may be dissociated into monomers and dimers (23). Therefore, three lines of evidence indicate that natural and recombinant T N F exist predominantly as a trimer under physiological conditions. Gel filtration analyses show that monomers are present as a small component of radioiodinated T N F (Fig. 1). The isolated monomers are only 12% as active as trimers in binding assays. Furthermore, lZ5I-hTNF can be dissociated into monomers by several treatments, such as a short incubation at pH 3.0.' Several reports have indicated that the biological activity of T N F is pH-sensitive (6,9,11,24). Treatment with 3 M GdnHCl also dissociates T -h T N F (Fig. 4C). Of particular interest is the finding that low concentrations of the nonionic detergent Triton X-100 partially dissociate hTNF into monomers (Fig. 2), suggesting that weak hydrophobic interactions may be responsible for stabilizing the trimers (25). Dissociation by Triton X-100 can be used in combination with gel filtration chromatography to obtain monomer and trimer fractions for competitive binding and cytotoxicity assays.
These lZ5I-hTNF monomers show low receptor binding activity when compared to trimers. Monomer binding is characterized by high and low affinity components. The high affinity component is the same as that observed for the trimer, but rechromatography of monomer fractions shows the presence of some trimers. Therefore, small amounts of contaminating trimers may account for the high affinity binding component. In contrast, the trimer exhibits a single high affinity binding and greater cytotoxicity than monomer. At low concentrations, the cytotoxicity of the monomer fraction is 6-7-fold less than that of the trimer, but at higher concentrations monomer cytotoxicity becomes equivalent to that of trimer. This finding may be explained by the reassociation of monomers into active trimers. Other reports suggest that TNF monomers and oligomers are all active (1, 4, 12, 26).
The most direct evidence that hTNF trimer is biologically active comes from experiments wherein lZ5I-hTNF is crosslinked with BSOCOES. The cross-linked hTNF binds to receptors (Fig. 5 ) and is cytotoxic. Moreover, cross-linked hTNF trimers are recovered after binding to cells (Fig. 4E). In view of the remarkably low concentrations of hTNF that are biologically active in cytotoxicity assays (27), we are lead to speculate that these trimers may interact simultaneously or sequentially with more than one receptor. A possible result of such multiple interactions may be a heightened effective concentration at the cell surface (28). Furthermore, simultaneous binding to neighboring receptors might favor the interaction of the cytoplasmic domain of receptors (29) and either trigger or amplify the as-yet unknown signaling mechanism of the TNF receptor. In addition, dissociation of TNF into monomers a t low concentrations may have some physiological relevance in the action of this factor. Since the monomer appears to be less active than the trimer, this dissociation may limit some of the deleterious effects of T N F (30) at sites remote from those where it is produced in high amounts by macrophages (31).