Research ArticleBiochemistry

Solution of the Structure of the TNF-TNFR2 Complex

See allHide authors and affiliations

Science Signaling  16 Nov 2010:
Vol. 3, Issue 148, pp. ra83
DOI: 10.1126/scisignal.2000954


Tumor necrosis factor (TNF) is an inflammatory cytokine that has important roles in various immune responses, which are mediated through its two receptors, TNF receptor 1 (TNFR1) and TNFR2. Antibody-based therapy against TNF is used clinically to treat several chronic autoimmune diseases; however, such treatment sometimes results in serious side effects, which are thought to be caused by the blocking of signals from both TNFRs. Therefore, knowledge of the structural basis for the recognition of TNF by each receptor would be invaluable in designing TNFR-selective drugs. Here, we solved the 3.0 angstrom resolution structure of the TNF-TNFR2 complex, which provided insight into the molecular recognition of TNF by TNFR2. Comparison to the known TNFR1 structure highlighted several differences between the ligand-binding interfaces of the two receptors. Additionally, we also demonstrated that TNF-TNFR2 formed aggregates on the surface of cells, which may be required for signal initiation. These results may contribute to the design of therapeutics for autoimmune diseases.


Tumor necrosis factor (TNF) is an immunity-modulating cytokine that is required for defense against infectious diseases and carcinogenesis (1). Excess amounts of TNF, however, cause various autoimmune diseases, such as rheumatoid arthritis (RA), Crohn’s disease, and ulcerative colitis (24). TNF activates signals through its two receptors [the type I TNF receptor (TNFR1) and TNFR2], and these molecules are well-known targets in therapies against autoimmune diseases (1, 5). Currently, TNF neutralization therapies through the use of a soluble TNFR2-Fc chimera (etanercept), a mouse-human chimera monoclonal antibody against TNF (infliximab), or a fully humanized monoclonal antibody against TNF (adalimumab) have proven to be effective treatments for RA (6, 7). Unfortunately, however, a block of TNF-mediated host defense often increases the risk of bacterial or viral infection (8, 9) or of the development of lymphoma (10). Thus, a thorough understanding of the function of the TNF-TNFR complex is important for the design of optimal therapies against the various TNF-related autoimmune diseases.

TNFR1 and TNFR2 activate distinct cell signaling pathways. TNFR1 is ubiquitously expressed, whereas TNFR2 is found mostly on certain populations of immune cells. In general, TNFR1 is largely associated with the apoptotic activity of TNF, whereas TNFR2 is involved in T cell survival (11). Thus, both proteins must be fully analyzed to better understand the function of TNF. Previous studies with animal models of diseases such as arthritis and hepatitis demonstrated the predominant role of TNFR1 in the pathogenesis and exacerbation of inflammation (12, 13). Nonetheless, TNFR2 is crucial for antigen-stimulated activation and proliferation of T cells (1416), which is essential for cell-mediated immunity to pathogens. In addition, transmembrane TNF (tmTNF), the prime activating ligand of TNFR2 (17), is sufficient to control infection by Mycobacterium tuberculosis (18, 19), indicating the importance of TNFR2 in this type of bacterial infection. Other reports showed that TNFR2 is important in the function of regulatory T cells (20), suggesting a role for tmTNF-TNFR2 signaling in anti-inflammatory effects. On the basis of these studies, the specific blocking of TNFR1 signaling appears to be a promising approach to minimize the side effects that are associated with “anti-TNF” therapy (5, 11, 21, 22). Thus, it is highly desirable to establish a structural basis for the differences between TNFR1 and TNFR2.

One structural characteristic shared by most members of the TNFR superfamily is that they contain from about two to four cysteine-rich domains (CRDs) (5). The first structure of a TNFR superfamily member to be determined was the crystal structure of the lymphotoxin-α (LT-α)–TNFR1 complex (23). The authors of that study suggested that the protein fold is characterized by multiple disulfide linkages in the CRD and that these bonds are important in stabilizing the structure of the TNFR. Moreover, a trimer of LT-α binds to three TNFR1 monomers through CRD2 and CRD3 in TNFR1, suggesting that trimerization of TNFRs is directly related to their signaling. The structural similarity between TNF and LT-α suggests that TNF should be able to form a complex with TNFR1 that resembles that of LT-α–TNFR1 (2325). More recently, the structures of complexes of other TNF-TNFR superfamily proteins have been solved, including TNF-related apoptosis-inducing ligand (TRAIL)–death receptor 5 (DR5) (2628) and CD134 antigen (OX40) ligand (OX40L)–OX40 (29). These reports suggest that the structural features that were described for the LT-α–TNFR1 complex are common to other TNF-TNFR superfamily members. Moreover, a study revealed the crystal structure of TNF in a complex with a viral TNF inhibitor (poxvirus 2L protein) (30) that is important for viral escape from TNF-mediated immunity (31). Therefore, determination of the structure of TNFR2 and of its role in immunity against pathogens would be useful in understanding the details of basic TNF functions.

Another important characteristic of the TNFR superfamily is that many of these proteins exist as preassembled oligomers on the cell surface [for example, Fas (Apo-1/CD95), TNFR1, and TNFR2] (32, 33). This ligand-independent assembly of TNFR oligomers is mediated by the preligand assembly domain (PLAD), which resides within the N-terminal CRD (CRD1) of the TNFRs and is not directly involved in binding to ligand (33). PLAD-mediated, ligand-independent assembly has also been reported for TRAIL receptors and viral homologs of TNFR (34, 35). Ligand-independent assembly of receptors was reported for other cytokine members, such as the interleukin-17 receptor (36), which indicates that this phenomenon is important for various immune responses. Indeed, the PLAD of TNFRs is critical for TNF-mediated responses (33), and soluble PLAD can effectively prevent TNFR signaling and potently inhibit inflammatory arthritis (37). These results suggest that PLAD-mediated receptor assembly is essential for TNFR signaling. However, in the crystal structures of other TNF-TNFR superfamily complexes, such as TNFR1, DR5, and OX40, individual PLADs are disassociated (23, 2629). This apparent contradiction regarding PLAD-mediated receptor assembly must be resolved to understand the molecular basis for TNFR-mediated signal initiation. We reasoned that the behavior of the TNF-TNFR complex on the surface of live cells needed to be investigated to understand the molecular basis of the ligand-receptor interactions.

Here, we determined the first crystal structure of the TNF-TNFR2 complex. With these data, we analyzed the structural basis for how TNF can bind to two divergent receptors (TNFR1 and TNFR2) of the same superfamily. This finding contributes to an understanding of the differences between TNFR1 and TNFR2, which may be useful for rationalizing selectivity data and for generating hypotheses to design future TNFR-selective drugs (11, 21). Finally, we discuss the signal initiation mechanism of the TNFR superfamily by analyzing the behavior of the preassembled TNFR2 and TNF-TNFR2 complexes on the surface of a live cell. This is the first report to describe the structural details of a TNF-TNFR2 complex in a crystal and in cells. These findings contribute to our knowledge of members of the TNF superfamily and may provide a new focus for investigation of the signaling machinery of cell surface receptors.


Determination of the structure of the TNF-TNFR2 complex

In a previous report, we obtained a crystal of the TNF-TNFR2 complex belonging to the space group P212121, as well as preliminary x-ray diffraction data to 2.95 Å (38). Here, we solved the structure of the TNF-TNFR2 complex by molecular replacement with the crystal structure of another TNF mutant that we described previously [Protein Data Bank (PDB) code 2e7a] (39). Diffraction data sets at 3.0 Å were used in refinements, and the final R-factor was 21.3% (with a free R-factor of 28.1%) (Table 1). Although we used mutTNF Lys (−) (a lysine-deficient mutant of human TNF that is fully active) (40) as the TNF molecule in the TNF-TNFR2 complex (38), it was confirmed that mutTNF Lys (−) in the TNF-TNFR2 complex retained almost the same structure as that of wild-type TNF [root mean square deviations (RMSDs) of 0.94 Å for 420 Cα atoms].

Table 1

Refinement statistics of the TNF-TNFR2 crystal. Ramachandran statistics indicate the fraction of residues in the favored, allowed, and outlier region, respectively, of the Ramachandran diagram as defined by the RAMPAGE program (65).

View this table:

We found that the asymmetric unit in the crystal contained two copies of the TNF-TNFR2 complex, which formed an interlocking dimer (two trimers of TNF and six monomers of TNFR2) (Fig. 1, A and B). In this interlocking dimer, CRD2-CRD4 interactions were mainly observed between opposite TNFR2 molecules (Fig. 1, C and D). We observed a potential intermolecular hydrogen bond between Asp81 in CRD2 and the amido NH group near Thr151 in CRD4 (Fig. 1, E and F), but each TNFR2 mainly interacted through van der Waals contacts. The buried surface area of this interface was relatively extensive (~4300 Å2 for every two copies of the complex), in contrast to ~2500 Å2 in the high-affinity TNF-TNFR2 binding interface. However, according to analytical gel-filtration experiments, the purified TNF-TNFR2 complex in aqueous solution is 110 kD (38), which suggests that the TNF-TNFR2 complex contains one trimer of TNF (51 kD) and three monomers of TNFR2 (19 kD each) in aqueous solution. Moreover, the position of the C terminus of TNFR2 suggested that this interlocking dimer would be difficult to form on the cell surface (Fig. 1C). Therefore, we suggested that the formation of such interlocking dimers was a result of crystal packing.

Fig. 1

TNF-TNFR2 complex in the asymmetric unit. Two TNF-TNFR2 complexes are observed in the asymmetric unit (consisting of two TNF trimers and six TNFR2 monomers). TNFR2 molecules from different complexes interact with each other in the crystal. TNF molecules are shown in green and orange; TNFR2 molecules are shown in blue and red. (A and B) Side view (A) and top view (B) of the complexes. (C and D) Side view (C) and top view (D) of the TNFR2-TNFR2 interaction in the crystal. N, N terminus; C, C terminus. (E and F) Details of the TNFR2-TNFR2 interfaces. Close contacts that are suggestive of potential hydrogen bonds are shown as yellow dashed lines.

Nonetheless, previous studies showed that mutation of Met174 of TNFR2 to Arg, which is referred to as the “M196R polymorphism,” is associated with the presence of soluble TNFR2 in the serum (41) and autoimmune diseases, such as systemic lupus erythematosus (42, 43). The crystal packing of TNFR2 showed that Met174 was located near Arg77 of other TNFR2 molecules (Fig. 1, E and F), which suggests that the mutant Arg174 residue influences the interaction between the CRD2 of one TNFR2 molecule and the CRD4 of another TNFR2 molecule in the crystal. Although previous gel-filtration analysis suggested that this interlocking dimer was formed by crystal packing, the report on the mutation of Met174 in TNFR2 suggests that the interlocking dimer might form only under the specific condition in which TNFR2 is soluble.

TNFR2 structure

The structures of the extracellular domains of members of the TNFR superfamily are composed of CRDs that typically contain six cysteine residues that form three disulfide bonds (23). TNFR1 and TNFR2 contain four CRDs, termed CRD1 through CRD4 (Fig. 2A). CRD1 (also known as PLAD) is essential for forming the TNFR self-complex on the cell surface (33). CRD2 and CRD3 are known as TNF-binding domains (23), whereas the function of CRD4 remains unclear.

Fig. 2

Basic structure and folding of TNFR2 and TNFR1. (A) Structure of the extracellular domain of TNFR2 in blue (PDB ID 3alq) and TNFR1 in red (PDB ID 1ext). Disulfide linkages are shown as green spheres. For a comparison of the basic structures, we superimposed a crystal structure of unbound TNFR1 (CRD1 to CRD3 of PDB ID 1ext) (55) onto the structure of TNFR2 (CRD1 to CRD3) with the SUPERPOSE program (67) in CCP4i. (B) Alignment of the amino acid sequences of TNFR1 and TNFR2. A1, A2, B1, B2, and C2 are the names of the module structures. Cysteine residues are highlighted in green. Differences between the structures of TNFR1 and TNFR2 (regions 1 and 2) are highlighted in orange. The amino acid sequence alignment was performed with the ClustalW program (68). Abbreviations for the amino acids are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr. (C) The A1 and A3 modules. There are different modules in the CRD3 regions of TNFR1 and TNFR2. (D) Ligand-binding interface of TNFR2. (E) Receptor binding interface of TNFR1. To compare the binding interfaces, we superimposed a crystal structure of TNFR1 complexed with LT-α (CRD1 to CRD3 of PDB ID 1tnr) onto the structure of TNFR2 (CRD1 to CRD3) with the SUPERPOSE program. The side chain of Glu109 is missing in the structure of TNFR1 (PDB ID 1tnr). Structural differences between TNFR1 and TNFR2 (regions 1 and 2) are also highlighted by orange dashed circles.

Through comparison of the structures of TNFR2 and TNFR1, together with alignment of their corresponding sequences, we found that CRD1 and CRD2 were topologically and structurally similar in both receptors (Fig. 2B). These CRDs contained the modules A1 and B2, which are typically observed in conventional members of the TNFR superfamily (44), such as the structurally determined TNFR1 (23), DR5 (2628), and OX40 (29) proteins. The A1 module contains a single disulfide bond, whereas the B2 module contains two disulfides, in a consensus sequence pattern. Of note, and different from the CRD3 of TNFR1, the CRD3 of TNFR2 contained the A2 module that is observed in a certain type of TNFR superfamily members, such as the CD30, CD40, and 4-1BB proteins (44). Because these TNFR superfamily members have not been structurally identified yet, our structure of the TNF-TNFR2 complex is the first to show the structure of the A2 module. The A2 module in TNFR2 contained two disulfides that were linked in a 1-4, 2-3 topology (Fig. 2C). The 2-3 disulfide (between Cys104 and Cys112) contributed to deflect a β-turn motif that was near the ligand-binding region (Fig. 2D), similar to the structure predicted from the A2 module in a viral protein homologous to TNFR2 (45). As a result of this disulfide bond, a gap between CRD2 and CRD3 in TNFR2 was buried in the structure (region 1 in Fig. 2, D and E). The location of Arg77 in the CRD2 of TNFR1, which is thought to be essential for binding to TNF (23), was compensated by Arg113 of the TNFR2 CRD3 in the TNF-TNFR2 structure. This structural difference between TNFR1 and TNFR2 was thought to depend on the diversity of their modules.

We observed another structural difference between TNFR1 and TNFR2 in region 2 (Fig. 2, D and E). A loop structure in TNFR1, which is reported to constitute the ligand-binding region (23), consisted of five residues in TNFR1 (Arg77 to Gly81) but only three residues in TNFR2 (Ser79 to Asp81) (Fig. 2B). The shorter loop structure of TNFR2 was further from the molecular surface of TNF ligand compared with that of TNFR1 as described below.

The TNF-TNFR2 complex

We found that TNF formed a central homotrimer around which three TNFR2 molecules were bound, similar to the known structures of other TNF superfamily members, including LT-α–TNFR1 (23), TRAIL-DR5 (2628), and OX40L-OX40 (29) (Fig. 3, A and B). The structure of the TNF-TNFR2 complex revealed that the TNFR1-binding region of TNF overlapped with its TNFR2-binding region, as previously predicted (46). One TNFR2 molecule interacted with two TNF molecules, as is the case for the LT-α–TNFR1 complex. The core of the interface between TNFR2 and TNF was separated into two regions, termed regions 3 and 4 (Fig. 3, B to D). Region 3 consisted of the A1 module of CRD2, whereas region 4, which was near regions 1 and 2, consisted of the B2 module of CRD2 and the A2 module of CRD3. Comparison of the electrostatic surface potentials of TNFR1 and TNFR2 showed that they were different despite sharing the same binding partner (Fig. 3, C and D). In the interface of TNFR2 region 3, acidic amino acid residues (Asp54, Glu57, and Glu70) were clustered, forming a more negatively charged surface than that of TNFR1. Moreover, in region 4 of TNFR2, the molecular surface was different from that of TNFR1, possibly as a result of diversity in the modules present (Fig. 2B). This structural feature contributed to the exposure of basic amino acids (Arg77, Lys108, and Arg133) at the binding interface, which generated a positively charged surface on TNFR2.

Fig. 3

Electrostatic surface potentials on the binding interfaces of TNFR2 and TNFR1. (A and B) Top view (A) and side view (B) of TNF-TNFR2 complexes (PDB ID 3alq). The TNF trimer is in green and the TNFR2 monomer is in blue. The binding regions (regions 3 and 4) are highlighted by orange, dashed circles. (C) Electrostatic surface potential of TNFR2. (D) Electrostatic surface potential of TNFR1 (PDB ID 1tnr). Each electrostatic surface potential was calculated with the GRASP program (69). Electrostatic charges (red indicates a negative charge, blue indicates a positive charge) are shown between ±8kBT, where kB is the Boltzmann constant and T is the absolute temperature.

Although a cobalt ion (Co2+) from the crystallization reagent was observed at all six interfaces in the asymmetric unit (Fig. 4A), Arg31 of TNF appeared to interact with the negatively charged region 3 on the surface of TNFR2 (Fig. 4B). By contrast, our model of the TNF-TNFR1 complex, constructed from the structure of the LT-α–TNFR1 complex, indicated that the interaction between Arg31 and the TNFR1 might be weak, because of the neutral charge of region 3 (Fig. 4C). Indeed, a previously reported R31D mutant of TNF displays a marked loss of affinity for TNFR2 while retaining affinity for TNFR1 (47). Therefore, these results suggest that this electrostatic interaction between Arg31 of TNF and TNFR is more important for the TNF-TNFR2 complex than for that of TNF and TNFR1. We found that Arg32 of TNF was located in a position that enabled a potential hydrogen bond to be formed with Ser73 of TNFR2 and that it seemed to interact with Ser72 of TNFR1 in the same way. Therefore, we suggest that Arg32 of TNF contributes equally to the binding of TNF to TNFR1 and TNFR2.

Fig. 4

Difference in the mode of binding of TNF to TNFR1 and TNFR2. Details of the ligand-receptor binding interfaces of TNF-TNFR are shown. (A and D) 2FobsFcalc map of the TNF-TNFR2 complex contoured at 1.0 σ. (B and E) The TNF-TNFR2 complex. The predicted interaction between R31 of TNF and the acidic surface of TNFR2 (consisting of D54, E57, and E70) is shown as a green arrow. (C and F) The TNF-TNFR1 model complex. To construct the TNF-TNFR1 model complex, we superimposed the LT-α portion of the LT-α–TNFR1 complex (PDB ID 1tnr) (23) onto the TNF portion of the TNF-TNFR2 structure. TNF is in green; TNFR1 is in red; TNFR2 is in blue; the 2FobsFcalc map is represented by the pink mesh. Close contacts that are suggestive of potential hydrogen bonds are represented by yellow dashed lines.

Detail of region 4 showed that Arg113 and Arg77 of TNFR2 formed close contacts to Asp143, Gln149, and Glu23 of TNF, potentially through the formation of hydrogen bonds. Thus, Arg113 and Arg77 of TNFR2 were important residues for binding to TNF (Fig. 4, E and F). This result is also supported from previous analysis of point mutations of TNF (46, 48, 49). Meanwhile, Arg77 of TNFR1 appeared to interact with Asp143 and Gln149 of TNF (Fig. 4F). We suggest that Arg113 of TNFR2 and Arg77 of TNFR1 might have similar roles in binding to TNF (Fig. 4, E and F). This difference regarding arginine residues between TNFR1 and TNFR2 was also indicated earlier (Fig. 2, D and E). The origin of this difference in interaction was a result of the unique A2 module and the dynamic structural diversity close to regions 1, 2, and 4.

Usefulness of the molecular pocket formulation for the design of TNFR1-selective inhibitors

Differences in the composition of the respective modules together with the diversity in the length of the main chain near the binding interface constitute the basic structural elements that distinguish TNFR1 from TNFR2. These structural considerations could form the basis of the design of receptor-specific drugs. Previous mutational analysis showed that region 1 of TNFR1 and TNFR2 (Fig. 2, D and E) is essential for the interaction with the loop of TNF (amino acid residues 143 to 149) (46, 48, 49). In this ligand-binding area of TNFR2, the turn motif of CRD3 (Ser107 to Cys112) fit to its CRD2 in the presence of the disulfide bond between Cys104 and Cys112 (Figs. 2D and 5A). By contrast, there was a space between the turn motif of the CRD3 of TNFR1 and the β strand of its CRD2, which resulted in the formation of a molecular pocket specifically on the surface of TNFR1 (Fig. 5B). These observations suggest that this region of TNFR1 constitutes a promising target for the structure-based development of TNFR1-selective drugs.

Fig. 5

TNF-TNFR complexes contain a molecular pocket. Difference in the basic structures of TNFR1 and TNFR2. (A and C) The TNF-TNFR2 complex. (B and D) The TNF-TNFR1 model complex, which was constructed as described in Fig. 4. TNF is in green; TNFR1 is in red; TNFR2 is in blue. The β strands of CRD2 and CRD3 are indicated by white text. The side chain of Glu109 is missing in the structure of TNFR1 (PDB ID 1tnr). We observed that a distinct molecular pocket was formed in (B) and (C), which is highlighted by an orange dashed circle.

Another point of interest was observed in region 2. A number of amino acid residues contained in the loop structure of region 2 differ between TNFR1 and TNFR2 (Fig. 2B). In the shorter loop of TNFR2 (residues Ser79 to Asp81), there was a space between TNF and the receptor (Fig. 5C). By contrast, the longer loop structure of TNFR1 (residues Arg77 to Gly81) was predicted to bind to TNF across a wide surface area by van der Waals contacts (Fig. 5D). This interaction is also observed in the structure of the LT-α–TNFR1 complex (23). These observations suggest that the loop motif in TNFR1 could be a focal point for creating new drugs that specifically inhibit the interaction between TNF and TNFR1.

A TNF-TNFR2 complex on the cell surface

Previous studies have confirmed that some members of the TNFR superfamily form a self-complex through their CRD1 (PLAD) regions at the cell surface, which also suggests that stimulation by ligand of these assemblies is necessary for efficient signaling (32, 33). In the crystal structure of the TNF-TNFR2 complex, however, the CRD1 regions were separated from each other by >30 Å, which is too far to enable an interaction to occur. This phenomenon is also observed in other TNF-TNFR complexes (23, 2629). These apparently contradictory observations suggest that the binding of the TNF ligand induces a dynamic behavior in the TNFR self-complex that may trigger signal initiation.

To understand the state of TNFR2 at the cell surface, we transfected human embryonic kidney (HEK) 293T cells with plasmids encoding hemagglutinin (HA)–tagged wild-type TNFR2 (HA-wtTNFR2), TNFR2 lacking its PLAD (HA-TNFR2ΔPLAD), or TNFR2 lacking its intracellular domain (HA-TNFR2ΔCD). To identify self-complexes formed through PLAD-mediated interactions, we used the thiol-cleavable, membrane-impermeant, chemical cross-linker 3,3′-dithiobis(sulfosuccinimidyl propionate) (DTSSP), as described in a previous report (33). We performed Western blotting analysis of purified membrane fractions with antibody against HA to detect cross-linked, HA-tagged TNFR2 molecules. We observed a band corresponding to monomeric TNFR2 with a molecular mass of 65 kD (Fig. 6A, lanes 1 to 4). We also detected self-complexes of TNFR2 in the absence of TNF with molecular sizes about two or three times greater (130 or 195 kD) (Fig. 6A, lane 1), consistent with previous reports (33). Furthermore, analysis with high–molecular mass markers and a low-density polyacrylamide gel enabled us to identify a band of TNFR2 with a molecular mass of >1000 kD in samples treated with TNF (Fig. 6A, lane 2). These complexes were reduced to monomeric proteins by cleaving the cross-linker with dithiothreitol (DTT) (Fig. 6A, lanes 3 and 4). A similar experiment with cells transfected with plasmid encoding HA-TNFR2ΔPLAD revealed that the formation of TNFR2 self-complexes in the presence and absence of TNF was inhibited by the deletion of PLAD (Fig. 6A, lanes 5 and 6). In addition, treatment with TNF did not induce a shift in the band corresponding to HA-TNFR2ΔPLAD (Fig. 6A, lane 6), suggesting that most of the TNF did not bind to TNFR2 without PLAD at the cell surface. A previous report showed that deletion of the PLAD from TNFR1 markedly decreases the amount of TNF that binds to cell surface TNFR1 (33). Thus, our result suggests that the binding of TNF to TNFR2 is also disrupted by the deletion of the PLAD from TNFR2. In experiments with cells transfected with plasmid encoding HA-TNFR2ΔCD, we showed that this mutant TNFR could still form self-complexes (Fig. 6A, lane 9); however, the band corresponding to HA-TNFR2ΔCD did not shift upon treatment with TNF (Fig. 6A, lane 10). These results suggest that self-complexes of TNFR2 are formed through its PLAD on the surface of cells and that the stimulation of TNF was important for the formation of high–molecular mass aggregates of TNFR2.

Fig. 6

Formation of TNF-TNFR2 aggregates on the cell surface. (A and B) TNF-TNFR2 complexes in the plasma membrane of transfected HEK 293T cells were detected by Western blotting analysis with antibodies against (A) the HA epitope and (B) TNF. This result was confirmed by three independent experiments for each group. Predicted molecular masses of related molecules are as follows: HA-wtTNFR2 monomer, 65 kD; HA-wtTNFR2 dimer, 130 kD; HA-wtTNFR2 trimer, 195 kD; TNF monomer, 17 kD; TNF trimer, 51 kD; HA-TNFR2ΔPLAD monomer, 60 kD; HA-TNFR2ΔCD monomer, 25 kD.

We also analyzed the same samples by Western blotting with an antibody against TNF (Fig. 6B). We observed high–molecular mass, TNF-specific bands of >150 kD in samples containing HA-wtTNFR2 and TNF (Fig. 6B, lane 14), but saw only monomeric TNF (17-kD band) under reducing conditions (Fig. 6B, lane 16). This result suggests that TNF molecules were contained in the aggregates of TNFR2 that we observed earlier (Fig. 6A, lane 2). In similar experiments with cells containing HA-TNFR2ΔPLAD, we did not observe TNF-specific bands in any group (Fig. 6B, lanes 17 to 20), indicating that TNF bound rarely to TNFR2ΔPLAD, as was predicted from our earlier results (Fig. 6A). These findings suggest that TNF bound to the PLAD-dependent self-complex of TNFR2 and that the PLAD was a key domain in forming a TNF-TNFR2 aggregate on the cell surface. We could not observe TNF within the self-complex of HA-TNFR2ΔCD (Fig. 6, A and B), indicating that the intracellular domain of TNFR2 also played an important role in forming the TNF-TNFR2 aggregate on the cell surface.


Here, we described the first crystal structure of the TNF-TNFR2 complex at a resolution of 3.0 Å. TNF formed a central homotrimer around which were bound three TNFR2 molecules. This overall arrangement was similar to those of other members of the TNF superfamily, including LT-α–TNFR1 (23), TRAIL-DR5 (2628), and OX40L-OX40 (29) (Fig. 3). However, our determination of the crystal structure of TNFR2 revealed subtle differences with that of TNFR1. The basic structure of TNFR2 differed from that of TNFR1 mainly as a result of variation in the configuration of the folding module. These structural differences altered the mode of ligand recognition of each receptor (Figs. 4 and 5). The results contribute to our understanding of how TNF is able to discriminate between the common binding areas of these two different receptors. In addition, we have already created many mutant TNFs that exhibit different receptor selectivities (39, 46, 50, 51). The structures of these mutants and TNF-TNFR complexes are potentially useful to analyze “consensus” information that is essential for the TNF-TNFR interaction (52). Such information will be useful in the future for enhancing accuracy in the rational design of new drugs, such as TNFR-selective inhibitors.

We also revealed the formation of an aggregate of TNFR2 on the cell surface (Fig. 6). These aggregates contained both TNF and TNFR2, which indicated that an aggregate of TNF-TNFR2 complexes (>1000 kD) was present on the cell surface. This result was observed in cells transfected to express TNFR2 at the surface and needs to be confirmed by another experimental method in primary cells in the future. However, the importance of such TNF-TNFR2 aggregates in the initiation of TNFR signaling can also be predicted from a previous report on TNF heterotrimers that contained inactive, mutant TNF molecules that acted as dominant-negative TNF because of their lack of trivalent binding potency (53).

The HA-TNFR2ΔCD mutant formed a self-complex, but not an aggregate of TNF and TNFR2. Because the structure of the intracellular domain of TNFR2 is still unknown, we are unable to discuss the implications of these findings in detail. Nonetheless, we can speculate that the arrangement of the intracellular domain of TNFR2 might be important for the formation of aggregates of TNF and TNFR2 on the cell surface. Our deletion experiment with HA-TNFR2ΔPLAD indicated that the TNFR2 self-complex forms through PLAD-PLAD interactions, resulting in generation of the TNF-TNFR2 aggregate; however, despite the possibility of the formation of the TNF-TNFR2 complex through PLAD-mediated interactions, we observed that the PLADs of each TNFR2 were dissociated in the crystal structure (Fig. 3). To resolve this apparent contradiction in TNF-mediated signal initiation, we used a structure-based hypothesis based on information concerning our observation of TNF-TNFR2 aggregates of >1000 kD and the crystal structure of TNFR2.

Regarding our prediction of the configuration of the TNFR on the cell surface, previous studies showed that two types of ligand-free TNFR1 proteins form dimers in crystal structures, which are termed parallel dimers at pH 7.5 (54) and antiparallel dimers at pH 3.7 (55). The TNFR1 parallel dimer forms through PLAD-PLAD interactions in the crystal, and it has been speculated that such a dimer may also be formed on the cell surface (54); however, our data (Fig. 6) and another report (33) suggest that TNFR1 and TNFR2 form a self-complex as homodimers or homotrimers on the cell surface. Given that the stoichiometry of the TNFR2 self-complex is unclear, these findings imply two possible models for the complex (dimer and trimer models) (Fig. 7). In these models, two or three TNF trimers bind around a central dimer or trimer of TNFR2 molecules, respectively. Other self-complexes of TNFR2 are subsequently recruited to bind around the TNFs, generating an aggregate of TNF-TNFR2 in a two-dimensional network on the cell surface. TNF-TNFR2 networks in the dimer and trimer models would maintain six- and threefold symmetries, respectively. The crystal structure of the TNF-TNFR2 complex suggests that these arrangements of complexes appear to be structurally realistic in both models. This structural feasibility will strengthen a predicted two-dimensional network model described previously (54, 56). Finally, expansion of the network may influence the arrangement of the intracellular domains of TNFR2, thereby possibly inducing the recruitment of intracellular molecules, such as TNFR-associated factor 2 (TRAF2) in TNFR2 signaling.

Fig. 7

Structural feasibility of a two-dimensional network model for the initiation of signals through TNFR2. (A and B) Top views of (A) the dimer model and (B) the trimer model of a two-dimensional network of TNFR2. TNFR2 molecules can interact with each other through PLAD-PLAD interactions (deep blue) at the cell surface. TNF trimers can bind around the self-complex of TNFR2. The binding of TNF to TNFR2 may link it to other TNFR2 self-complexes through alternative binding sites, which could result in the formation of a two-dimensional TNF-TNFR2 network. TNF-TNFR2 networks in the dimer and trimer models would maintain six- and threefold symmetry, respectively. These arrangements of TNF-TNFR2 complexes appear to be structurally realistic in both models. The extracellular domain of TNFR2 is in blue and TNF is in green.

With respect to TRAF2, we can speculate about its intracellular behavior after the formation of the TNF-TNFR2 aggregate. TRAF2 is essential for TNFR2-mediated signaling. Indeed, the structure of a complex of the C-terminal domain of TRAF2 and a peptide from the intracellular domain of TNFR2 has been reported (57). The C-terminal domain of TRAF2 forms a trimer that binds to the intracellular domains of three TNFR2 molecules (57); however, there is no structural difference between the peptide-bound form and the unbound form. Therefore, it was unclear how TRAF2 transduces a signal to downstream molecules. The crystal structure of the N-terminal domain of TRAF6, which is homologous to TRAF2, has been solved (58). The structure shows that the N-terminal domains of TRAF6 are complexed to each other, suggesting that this interaction forms part of its signaling mechanism. Together, these findings suggest that the TNF-TNFR2 network might organize the intracellular domains of TNFR2 and induce the recruitment of TRAF2, which would result in a TRAF2-TRAF2 intermolecular interaction that is needed for signal transduction. Although this structure-based hypothesis needs to be confirmed by other experiments (for example, role of the intracellular domain and the use of a non–cross-linking methodology such as the direct observation of cell surface complexes with an electron microscope), we suggest that it might provide a new direction for solving the enigma concerning the mechanism of signal initiation of TNFR superfamily members.

The TNF-TNFR2 structures revealed in this report show the diversity in the molecular basis of TNF-TNFR recognition and provide a better understanding of the mechanism of signal initiation by members of the TNFR superfamily. We hope to develop the next generation of therapeutics with an approach based on our structural data, such as new drugs that can selectively inhibit the interaction between one type of TNF-TNFR complex or the formation of one specific type of TNF-TNFR aggregate.

Materials and Methods

Data collection and refinement

The complex of amino acid residues 1 to 157 of human TNF (which corresponds to residues 77 to 233 in UniProt P01375) and residues 11 to 183 in human soluble TNFR2 (which corresponds to residues 33 to 205 in UniProt P20333) were prepared as previously described (38). The human TNF used in this experiment was mutTNF Lys (−), a lysine-deficient, hexamutant TNF (K11M, K65S, K90P, K98R, K112N, and K128P) with full bioactivity (40). This TNF molecule was expressed as inclusion bodies in Escherichia coli and refolded as described previously (38, 40). Recombinant human soluble TNFR2 was purchased from PeproTech Inc. (catalog no. 310-12). This TNFR2 molecule was also expressed in E. coli. Crystallization and x-ray diffraction experiments were performed as previously described (38). Diffraction data were collected at SPring-8 in Harima and Photon Factory in Tsukuba, Japan. The data were indexed, integrated, and scaled with HKL2000 software (59). The data set used for structural analysis was collected in BL41XU of SPring-8. Molecular replacement was performed by the MOLREP program (60) in CCP4i (61) with the structure of the TNF mutant described in our previous report (39) (PDB code 2e7a) as a search model. The model from molecular replacement was refined with crystallography and nuclear magnetic resonance (NMR) system (CNS) software (62). The Cα chains of TNFR2 molecules were manually traced on the basis of the structure of TNFR1 (PDB code 1tnr) with the Coot program (63) in CCP4i. The final structure was refined by the PHENIX program (64), and validation of the final model was performed with the RAMPAGE program (65) in CCP4i. Data collection statistics (at a resolution of 2.95 Å) were described previously (38), and the final structure was refined at a resolution of 3.0 Å. Refinement statistics are given in Table 1. The diffraction data set has a poor Rmerge value of 0.18, as reported in a previous paper (38). This might arise from high and anisotropic mosaicity of the crystals. The gap (6.7%) between the values of R and Rfree is slightly larger than the 5% that is accepted as no overfitting, which might result from flexible loops with poor electron densities. However, because almost all of the electron densities were interpretable (as shown in Fig. 4, A and D), the overall structure is of sufficient quality to characterize the TNFR2 structure and reveal the recognition mechanism between TNF and TNFR2. All molecular graphics were rendered by PyMOL program (66).

Plasmid construction

Plasmids encoding TNFR2 were constructed with the help of a previous report (33). Briefly, the leader sequence and the first 10 amino acid residues of full-length, wild-type TNFR2 (wtTNFR2: residues 1 to 32 in UniProt P20333) were connected to the HA epitope tag (YPYDVPDYA) at its C terminus to generate an HA tag fused to the N terminus of TNFR2. Complementary DNAs (cDNAs) of HA-wtTNFR2 (encoding residues 1 to 32–HA–residues 33 to 461 in UniProt P20333), HA-TNFR2ΔPLAD (encoding residues 1 to 32–HA–residues 77 to 461 in UniProt P20333), and HA-TNFR2ΔCD (encoding residues 1 to 32–HA–residues 33 to 287 in UniProt P20333) were amplified and directly cloned into pcDN3.1D/V5-His-TOPO vectors (Invitrogen Corp.). Primers 1 (5′-CACCATGGCGCCCGTCGCCGTCTGGGCCGCGCTGGCCGTCGGACTGGAGCTCT GGGCTGCGGCGCACGCCTTGCCCGCCCAGGTGGCATTTACACCCTACTACCCCTATGATGTGCCAGACTACGCCG CCCCGGAGCCCGGGAGCACATGC-3′) and 3 (5′-TTAACTGGGCTTCATCCCAGCATC-3′) were used to amplify HA-wtTNFR2, whereas primers 2 (5′-CACCATGGCGCCCGTCGCCGTCTGGGCCGCGCTGGCCGTCGGACTGGAGCTCTGGGCTGCGGCGC ACGCCTTGCCCGCCCAGGTGGCATTTACACCCTACGCCCTACCCCTATGAGGTGCCAGACTATCCTGTGA GGACAGCACATACACC-3′) and 3 were used to amplify HA-TNFR2ΔPLAD cDNA. Primers 1 and 4 (5′-TCACACCTGGGTCATGATGACACAGTT-3′) were used for the amplification of HA-TNFR2ΔCD.

Expression and cross-linking of TNFR2 at the cell surface

HEK 293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Wako Pure Chemical Industries Ltd.), containing 10% fetal bovine serum at 37°C under 5% CO2. Cells were transfected with plasmids encoding HA-wtTNFR2, HA-TNFR2ΔPLAD, or HA-TNFR2ΔCD with Lipofectamine LTX and Plus reagent (Invitrogen). After incubation for 6 hours at 37°C under 5% CO2, TNFR2 proteins were expressed on the surface of the HEK 293T cells. For cells treated with TNF, recombinant human TNF (R&D Systems Inc.) was added to the cells at a final concentration of 5 μg/ml, and the cells were then incubated at 4°C for 1 hour. Cells were scraped from the culture dish and incubated in 1 mM DTSSP (Thermo Fisher Scientific Inc.) for 30 min at room temperature to cross-link cell surface TNFR2 complexes. The cross-linking reaction was terminated by the addition of 20 mM tris-HCl (pH 7.4). Membrane proteins, which contained cell surface TNFR2 complexes, were purified with the Plasma Membrane Protein Extraction Kit (BioVision).

Western blotting analysis

Purified membrane proteins were mixed with an equal volume of Laemmli sample buffer (Bio-Rad Laboratories Inc.) with or without 50 mM DTT (Thermo Fisher Scientific). The samples were subjected to SDS–polyacrylamide gel electrophoresis (SDS-PAGE) with a 3 to 10% gradient polyacrylamide gel (ATTO Corp.). Proteins were then transferred onto a polyvinylidene difluoride (PVDF) membrane (GE Healthcare Bio-Sciences Corp.). We used antibody against the HA epitope (Abcam Inc.) and horseradish peroxidase (HRP)–conjugated antibody against mouse immunoglobulin G (IgG) (GE Healthcare Bio-Sciences) to detect HA-wtTNFR2, HA-TNFR2ΔPLAD, and HA-TNFR2ΔCD proteins. To detect TNF, we used antibody against human TNF (Genzyme Corp.) and HRP-conjugated antibody against mouse IgG. Specific bands were visualized with the ECL Plus reagent (GE Healthcare Bio-Sciences). To estimate the molecular mass of large complexes, we used high–molecular weight markers (NativeMark and HiMark HMW standard; Invitrogen).


Acknowledgments: We thank T. Mayumi for his advice about this research. Funding: This study was supported in part by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan and from the Japan Society for the Promotion of Science. This study was also supported in part by Health Labour Sciences Research Grants from the Ministry of Health, Labor and Welfare of Japan; Health Sciences Research Grants for Research on Publicly Essential Drugs and Medical Devices from the Japan Health Sciences Foundation; and The Nagai Foundation Tokyo. Author contributions: Y.M., S.T., and Y.T. designed the research; Y.M., T.N., and M.Y. performed the research; Y.M., T.N., and Y. Yamagata analyzed the data; Y. Yoshioka and S.N. contributed new reagents; and Y.M., Y. Yamagata, and Y.T. wrote the paper. Competing interests: The authors declare that they have no competing interests. Accession numbers: Coordinates and structure factors have been deposited in the PDB with the accession number 3alq.

References and Notes

Stay Connected to Science Signaling

Navigate This Article