Identification of Tumor Necrosis Factor (TNF) Amino Acids Crucial for Binding to the Murine p75 TNF Receptor and Construction of Receptor-selective Mutants*

The bioactivity of tumor necrosis factor (TNF) is mediated by two TNF receptors (TNF-Rs), more particu-larly TNF-RI and TNF-RII. Although human TNF (hTNF) and murine TNF (mTNF) are very homologous, hTNF binds only to mTNF-RI. By measuring the binding of a panel of mTNF/hTNF chimeras to both mTNF-R, we pin-pointed the TNF region that mediates the interaction with mTNF-RII. Using site-specific mutagenesis, we identified amino acids 71–73 and 89 as the main inter-acting residues. Mutein hTNF-S71D/T72Y/H73 (cid:1) /T89E interacts with both types of mTNF-R and is active in CT6 cell proliferation assays mediated by mTNF-RII. Mutein mTNF-D71S/Y72T/ (cid:1) 73H/E89T binds to mTNF-RI only and is no longer active on CT6 cells. However, the L929s a template. Sequences derived from this structure with the desired substitutions in positions 71, 72, 89, and 102 were used as target sequences. The constructed models were re-fined with GROMOS96 (21). To that end, modeled muteins were opti- mized by applying 200 cycles of the steepest descent method followed by 300 cycles of the conjugate gradient method. Finally stereochemical analysis of the models was achieved using WHAT IF (22).

Tumor necrosis factor (TNF) 1 is a pleiotropic cytokine with a wide range of biological activities including cytotoxicity, immune cell proliferation, and mediation of inflammatory responses (1)(2)(3). It exerts both direct and indirect antitumor effects on a number of tumors in vivo. However, its therapeutic application in the treatment of tumors is severely hampered by pathogenic side effects such as hypotension and liver toxicity (4). The multiple biological effects of TNF are mediated by two cell surface TNF receptors (TNF-Rs), namely TNF-RI and TNF-RII (5,6). These receptors bind in the groove regions between TNF subunits; hence one trimeric TNF molecule binds three receptor molecules (7). Clustering of surface-bound receptors initiates a signaling cascade in the cell. Moreover, receptorspecific muteins of hTNF have been obtained by mutating amino acids located at the intersubunit grooves (8 -10).
Although murine TNF (mTNF) and human TNF (hTNF) are 79% identical at the amino acid level, hTNF interacts with mTNF-RI but not with mTNF-RII. Due to the high homology, chimeric TNF genes can be obtained by exchanging homologous regions between the hTNF and mTNF genes. Bacterial expression of such in-frame chimeric genes results in chimeric TNF subunits that are able to trimerize into bioactive molecules. These chimeric proteins are subsequently used to localize the epitopes for species-specific TNF monoclonal antibodies (11).
We created and characterized different mTNF/hTNF chimeras to identify the region(s) responsible for interaction with mTNF-RII. Subsequently we obtained receptor-specific muteins by replacing murine amino acids in this region by their human counterparts and vice versa. This resulted in a mTNF-RI-specific mutein of mTNF and a hTNF mutein also binding efficiently to mTNF-RII.

MATERIALS AND METHODS
Cytokines and Antisera-Recombinant mTNF and hTNF were produced in our laboratory and had a specific biological activity of 2.1 ϫ 10 8 and 6.7 ϫ 10 7 units/mg, respectively. TNF muteins were purified to near homogeneity as described previously (11). The extracellular domains of mTNF-RI and mTNF-RII were expressed in the baculovirus/ Sf9 system and purified as follows. The cell supernatant was subjected to ammonium sulfate precipitation (20 -80% cut). The pellet was dissolved in 50 mM Tris-HCl (pH 8.5) and passed over a Q-Sepharose column (Amersham Pharmacia Biotech). The flow-through with soluble receptors was dialyzed against 50 mM sodium acetate (pH 5.3), applied on an SP-Sepharose column (Amersham Pharmacia Biotech), and eluted with a linear NaCl gradient. Peak fractions had a purity of Ͼ90% as measured by SDS-polyacrylamide gel electrophoresis. Expression and purification of bivalent chimeric proteins formed with the Fc region of human IgG1 and the ligand-binding domains of hTNF-RI (hTNF-RI/ Fc) or hTNF-RII (hTNF-RII/Fc) were described previously (12,13).
Plasmids and Proteins-mTNF and hTNF chimeric plasmids were obtained by exchanging restriction fragments between previously constructed pPLcHTNF, pATTrpMTNF, pPLcMHTNF-11, and pPLcM-HTNF-22 (11). The TNF coding information of these plasmids comprises a unique 5Ј NcoI and 3Ј HindIII restriction site as well as a unique HincII site at amino acids 89-90 (see Fig. 1). pATTrpMHTNF-23 was created by ligation of the fragments NcoI/pPLcMHTNF-22/HincII and HincII/pAT-TrpMTNF/HindIII into pATTrpMTNF. pATTrpMHTNF-24 was con-* This work was supported in part by the Fonds voor Wetenschappelijk Onderzoek-Vlaanderen and the Interuniversitaire Attractiepolen and by United States Public Health Service, National Institutes of Health Grant CA69381. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
CT6 Proliferation Assays-CT6 cells were cultured in RPMI 1640 medium supplemented with 10% fetal calf serum, antibiotics, 2 mM L-glutamine, 1 mM sodium pyruvate, and 50 M 2-mercaptoethanol (complete RPMI) containing 10% EL4 thymoma supernatant. Growth induction on CT6 was determined as described previously (18). Briefly, cells were grown for 3 days in complete RPMI plus 5% EL4 supernatant, and they were subsequently seeded in 96-well microtiter plates at 50,000 cells/well. After a 1-h preincubation, a serial dilution of TNF was added to the cells. After incubation for 24 h, the assay plates were pulsed with 0.5 Ci of [ 3 H]thymidine/well and further incubated for 5 h. The cells were harvested, and the incorporated label was measured.
Gel Filtration of Labeled TNF-Purified TNF proteins were radiolabeled using Iodogen iodination agent according to the manufacturer's instructions (Pierce Chemicals). Labeled proteins were separated from unincorporated radioactivity on a G-25 column (PD10, Amersham Pharmacia Biotech) and had a specific radioactivity of 10 -50 Ci/g. Gel filtration was performed on a Sephacryl S-100 column (Amersham Pharmacia Biotech) equilibrated and eluted in phosphate-buffered saline containing 0.02% bovine serum albumin and 0.02% NaN 3 at a flow rate of 0.4 ml/min. Fractions of 0.3 ml were collected and tested for radioactivity in a ␥-counter. Gel filtration chromatography was performed at 4°C.
Direct Receptor Binding Assay-TNF binding to soluble TNF receptors was measured in a BIAcore apparatus (Biacore, Uppsala, Sweden). A flow of HEPES-buffered saline (10 mM HEPES, 3.4 mM EDTA, and 150 mM NaCl, pH 7.4) passing over the sensor surfaces was maintained at 10 l/min. The carboxylated dextran matrix of the sensor-chip flow cell was activated by the injection of 50 l of a solution containing 0.2 M N-ethyl-NЈ-(3-dimethylaminopropyl)carbodiimide and 0.05 M N-hydroxysuccinimide in water. Then 50 l of 10 mM sodium acetate buffer (pH 4.0) as such or containing 25 g/ml purified soluble murine TNF-R (smTNF-R) I, smTNF-RII, hTNF-RI/Fc, or hTNF-RII/Fc were passed over one flow cell. The remaining binding sites were blocked by injection of 50 l of 1 M ethanolamine (pH 8.5). The surface plasmon resonance signal after immobilization of smTNF-RI or smTNF-RII generated 800 or 870 response units (see Fig. 2, A and B) and 1654 or 1381 response units (Fig. 2, C-F). Immobilization of hTNF-RI/Fc (Fig. 2G) or hTNF-RII/Fc generated 5500 and 5650 response units, respectively. To characterize TNF binding, a TNF solution was passed over differently coated sensor-chip flow cells (association phase) followed by HEPESbuffered saline (dissociation phase). The changes in response between control cells and cells coated with smTNF-R were measured and represented in sensorgrams.
Homology Modeling-The theoretical three-dimensional structure of mTNF muteins was constructed by homology modeling using the computer program Swiss-Model (19). The crystal structure of mTNF (20) was obtained from a protein data library (www.rcsb.org/pdb, PDB accession code 2TNF) and served as a template. Sequences derived from this structure with the desired substitutions in positions 71, 72, 89, and 102 were used as target sequences. The constructed models were refined with GROMOS96 (21). To that end, modeled muteins were optimized by applying 200 cycles of the steepest descent method followed by 300 cycles of the conjugate gradient method. Finally stereochemical analysis of the models was achieved using WHAT IF (22).

RESULTS
Design and Receptor Specificity of TNF Muteins-Since hTNF and mTNF are very homologous, mTNF/hTNF in-frame chimeras may form bioactive homotrimeric molecules. Because hTNF binds to mTNF-RI only, such chimeras were used to identify the part of the mTNF protein required for binding to mTNF-RII. The interaction of wild-type (wt) and chimeric TNF proteins with both types of mTNF-R was characterized in a direct in vitro binding assay. Recombinant smTNF-Rs were covalently bound to the surface of a CM5 chip, after which a TNF solution was passed over the receptor-coated surface. The resulting sensorgrams clearly showed that mTNF interacts with both types of mTNF-R and that it dissociates much faster from mTNF-RII than from mTNF-RI. Chimeric proteins with a human sequence between amino acids 50 and 90, viz. MHTNF-22, MHTNF-23, and MHTNF-16, interact with mTNF-RI only, albeit with different efficiency. On the other hand, chimeras with a corresponding murine sequence, viz. MHTNF-11, MHTNF-24, and MHTNF-15, bind to both types of mTNF-R (Fig. 2, A, B, E, and F). We also measured the interaction of these mTNF/hTNF chimeric proteins with both types of hTNF-R/Fc proteins. As mTNF, just like hTNF, interacted with both types of hTNF-R, the chimeric TNF proteins tested also interacted with both hTNF-R/Fc proteins and could not be used to delineate TNF regions important for interaction with a specific hTNF-R (Fig. 2, G and H).
In a next step, a series of mTNF muteins was constructed in which one or more of the 10 amino acids, differing between mTNF and hTNF in the region 50 -90 (Fig. 1), had been replaced. The receptor interaction of the resulting muteins was analyzed in vitro. mTNF-D71S/Y72T/⌬73H and mTNF-E89T still interacted with both types of mTNF-R. Mutein mTNF-D71S/Y72T/⌬73H/E89T, which combines both substitutions, did, however, no longer bind to mTNF-RII. The latter mutein, whose amino acids 71-73 and 89 had been replaced, still efficiently interacted with mTNF-RI (Fig. 2, C and D). Mutein hTNF-S71D/T72Y/H73⌬/T89E, on the other hand, was able to bind not only to mTNF-RI but also to mTNF-RII (Fig. 2, E and  F). The results show that amino acids 71-73 and 89 are important for the interaction of TNF with smTNF-RII. Although these muteins were constructed to study the interaction of TNF with different mTNF-Rs, we also tested whether these mutations influenced the interaction with hTNF-R. mTNF-D71S/ Y72T/⌬73H/E89T and hTNF-S71D/T72Y/H73⌬/T89E interacted as efficiently as wt mTNF and hTNF, respectively, with hTNF-R/Fc proteins (Fig. 2, G and H).
Biological Activity of TNF Muteins-To test the biological effect of the mutations, the activity of the muteins in L929s cytotoxicity and CT6 proliferation assays was measured. On the basis of activity of agonistic receptor-specific antibodies, cytotoxicity to L929s cells is known to depend on mTNF-RI triggering, and proliferation of CT6 cells is known to depend on mTNF-RII triggering (23). Like hTNF, mTNF-D71S/Y72T/ ⌬73H/E89T did not induce CT6 proliferation. Mutein hTNF-S71D/T72Y/H73⌬/T89E, on the other hand, was biologically active on CT6 cells (Table I). In L929s assays, the bioactivity of hTNF-S71D/T72Y/H73⌬/T89E was more pronounced than hTNF; surprisingly, mTNF-D71S/Y72T/⌬73H/E89T was 100fold less active than mTNF. As such a difference was not in agreement with receptor binding results in vitro, the trimer stability of the different TNF molecules was investigated.
Trimer Stability of Muteins as Measured by Gel Filtration-As bioactive trimeric TNF molecules have been shown to dissociate into inactive monomers during incubation at low concentrations (24), the stability of mTNF and mutein mTNF-D71S/Y72T/⌬73H/E89T was compared. 125 I-Labeled proteins (0.4 nM) were incubated at 26°C for 16 h in Dulbecco's modified Eagle's medium (ϩ10% fetal calf serum) as such or in the presence of 4 nM unlabeled mutein, after which they were subjected to size-exclusion chromatography. After incubation, the fraction in the monomeric peak was much greater for mTNF-D71S/Y72T/⌬73H/E89T than for mTNF (Fig. 3, A and  B). Mutein mTNF-D71S/Y72T/⌬73H behaved in the same way as mTNF-D71S/Y72T/⌬73H/E89T after preincubation and subsequent gel filtration (data not shown). Based on these results, it may be concluded that the faster dissociation of mTNF-D71S/ Y72T/⌬73H/E89T negatively affects its bioactivity in L929s assays where subnanomolar amounts of TNF are used.
Design of a Stable mTNF-RI-specific Mutein-Additional substitution of amino acids in the region 50 -90 in mTNF-D71S/Y72T/⌬73H/E89T did not enhance the cytotoxic activity to L929s cells (data not shown). The three-dimensional structure of mTNF (20) shows that amino acids 102-104, which differ in mTNF and hTNF, are located close to amino acids 71-73 (Fig. 4). Hence mutein mTNF-D71S/Y72T/⌬73H/E89T/ P102Q was created and tested. The receptor specificity of the latter mutein and mTNF-D71S/Y72T/⌬73H/E89T was similar in direct in vitro receptor binding assays (Fig. 2). As expected, both muteins were inactive in mTNF-RII-mediated CT6 assays. In mTNF-RI-mediated L929s cytotoxicity assays, however, mTNF-D71S/Y72T/⌬73H/E89T/P102Q was 25-fold more active than mTNF-D71S/Y72T/⌬73H/E89T and had a specific activity almost comparable with that of hTNF (Table I). To determine whether this increase in cytotoxic activity was due to increased trimer stability, the TNF mutein was labeled with 125 I and subsequently subjected to size-exclusion chromatography. Mutein mTNF-D71S/Y72T/⌬73H/E89T/P102Q appeared to be much more stable than mTNF-D71S/Y72T/⌬73H/E89T and to behave as wt mTNF (Fig. 3C). DISCUSSION At present, the mechanism of toxicity observed after in vivo administration of TNF-RI-specific muteins remains unclear. In the murine model, where hTNF is a specific agonist for mTNF-RI, the LD 50 dose of hTNF is about 50 times higher than that of mTNF, but it still has an antitumor activity that is comparable with that of mTNF (25). In monkeys, administration of hTNF or a TNF-RI-specific mutein of hTNF induced similar changes in important physiological parameters (blood pressure and liver or kidney functions) when measured over a relatively short time interval (26). The pharmacokinetics, however, are very different; the mutein remains in circulation for a much longer time. Hence pathophysiological studies of TNF in rela- tionship with its action on the two different receptors are quite complex. As tools for such studies we developed receptor-specific muteins both in the human and the murine system.
Based on the high homology between hTNF and mTNF and the fact that hTNF does not bind to mTNF-RII, we obtained a panel of mTNF/hTNF chimeras with different mTNF-R binding characteristics. By comparing their in vitro receptor binding, we concluded that the TNF region between amino acids 50 and 90 is important for the interaction between TNF and mTNF-RII. It may be noted that these chimeric TNF proteins still interacted with both hTNF-Rs because mTNF (just like hTNF) interacted with both types of hTNF-R. Using muteins created by site-specific mutagenesis, it became clear that the replacement of amino acids 71-73 and 89 in mTNF with their human homologues (D71S/Y72T/⌬73H and E89T) was sufficient to eliminate the interaction of this mutein with mTNF-RII as measured in vitro. As expected, mutein mTNF-D71S/Y72T/ ⌬73H/E89T was not active in mTNF-RII-mediated CT6 proliferation assays. In addition, mutein hTNF-S71D/T72Y/H73⌬/ T89E, which contains corresponding murine substitutions, interacted with mTNF-RII in vitro and clearly induced proliferation of murine CT6 cells. The results indicate that these amino acids are directly involved in the binding between TNF and mTNF-RII. However, mTNF-RI-mediated cytotoxic activity to L929s cells of mutein mTNF-D71S/Y72T/⌬73H/E89T was 100-fold lower than that of wt mTNF, although this mutein still efficiently interacted with mTNF-RI in vitro. As biologically active trimeric TNF molecules have been shown to dissociate into inactive monomers after incubation at subnanomolar concentrations (27), we investigated the trimer stability of wt and mutant TNF. Size-exclusion chromatography of labeled protein indicated that mutein mTNF-D71S/Y72T/⌬73H/E89T is more prone to dissociation than mTNF during incubation at low concentrations. This trimer instability probably influences the bioactivity in L929s cytotoxicity assays, which are very sensitive and involve an 18-h incubation. The instability, however, does not affect the assessment of in vitro receptor binding using BIAcore as this takes only a few minutes at TNF concentrations of 200 nM. By substituting other amino acids, differing between hTNF and mTNF, and by testing the bioactivity of resulting muteins, mutein mTNF-D71S/Y72T/⌬73H/E89T/ P102Q was obtained. This mutein has in vitro receptor binding characteristics similar to those of mTNF-D71S/Y72T/⌬73H/ E89T but is as stable as wt mTNF after incubation at low concentrations. Furthermore, mutein mTNF-D71S/Y72T/ ⌬73H/E89T/P102Q, which is inactive in CT6 proliferation assays, is 25-fold more active than mTNF-D71S/Y72T/⌬73H/ E89T in L929s cytotoxicity assays.
The structure of a TNF subunit folds into a ␤-sandwich, comprising almost entirely antiparallel ␤-strands, and forms a bell-shaped, rigid, trimeric molecule containing two other subunits. Although the structure of hTNF and mTNF is very similar (20,28), the amino acids 71-73, 89, and 102 are all located in flexible, surface-exposed loops that differ strongly (loops 64 -76, 84 -91, and 99 -112). Moreover, the amino acids 71-73, 89, and 102 are present near the intersubunit grooves of the TNF trimer (Fig. 4). These intersubunit grooves are the main site for interaction between TNF and TNF-R as clear from mutagenesis studies (8) and crystallography of human lymphotoxin bound to hTNF-RI (7). Furthermore, amino acids 32 and 86 of hTNF, which are crucial for binding to hTNF-RI, are also located near these intersubunit grooves (8 -10).
To rationalize the effect of amino acid substitutions at positions 71-73 and 102 on the trimer stability, theoretical threedimensional structures were constructed for two mTNF muteins using the crystal structure of mTNF as a template (Fig.   TABLE I Bioactivity of TNF muteins in different murine bioassays L929s cells were treated in the presence of 1 g/ml actinomycin D with a serial dilution of different TNF solutions. After an 18-h incubation, surviving cells were determined by staining with 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazoliumbromide. One L929s unit/ml TNF is the concentration of TNF at which 50% of the cells survive. CT6 cells were treated with TNF for 24 h and subsequently treated for 5 h with 3 H-labeled thymidine (incorporated label was determined). One CT6 unit/ml is the concentration of TNF needed to reach 30% of maximal growth induction obtained with mTNF.  (16 h, 26°C) at 20 ng/ml in Dulbecco's modified Eagle's medium/fetal calf serum as such (q) or in the presence of the same unlabeled TNF or mutein at 200 ng/ml (E). Trimeric and monomeric TNF molecules were separated by size-exclusion chromatography, after which the radioactivity in each fraction was determined. 4). Substituting Asp 71 -Tyr 73 with Ser-Thr-His did not affect the internal hydrogen bonds in the predicted model but resulted in the creation of a hydrophilic patch protruding into the aqueous environment. As both Ser and Thr may participate in the formation of hydrogen bonds, we assume increased possibilities to interact with surrounding water molecules, possibly resulting in destabilization of the loop. According to our predicted model, substitution of Pro 102 with Gln leads to the formation of a hydrogen bond between Gln 102 NE2 and Glu 110 OE1, which may enhance the stability of the loop (Fig. 4). Since a Cys 69 -Cys 101 bridge anchors loop 99 -112 with loop 64 -76, this substitution may affect the top of the molecule and hence influence the overall trimer stability.