Crystal structure of TNFα complexed with a poxvirus MHC-related TNF binding protein

Abstract

The poxvirus 2L protein binds tumor necrosis factor-α (TNFα) to inhibit host antiviral and immune responses. The 2.8-Å 2L–TNFα structure reveals three symmetrically arranged 2L molecules per TNFα trimer. 2L resembles class I major histocompatibility complex (MHC) molecules but lacks a peptide-binding groove and β2-microglobulin light chain. Overlap between the 2L and host TNF receptor-binding sites on TNFα rationalizes 2L inhibition of TNFα–TNF receptor interactions and prevention of TNFα-induced immune responses.

Poxviruses express a wide range of secreted proteins that interfere with the vertebrate immune response¹. The pro-inflammatory cytokine TNFα is a common target for inhibition by secreted poxviral proteins, including virally encoded homologs of TNF receptors (TNFRs)^2,3. A class of secreted TNFα inhibitor that is not homologous to viral or host TNFRs was recently identified from Tanapox virus (TPV)⁴ and related viruses (Supplementary Fig. 1). The TPV 2L protein binds human TNFα with very high affinity (K_d = 43 pM), thereby effectively sequestering TNFα to prevent activation of TNF receptors and downstream antiviral effects such as TNF-induced apoptosis⁴.

Although 2L proteins are distantly related to extracellular regions of MHC class I molecules (∼13% sequence identity)⁴ (Supplementary Fig. 2), 2L from the Yaba-like disease virus (98% identical to TPV 2L; Supplementary Fig. 1) did not co-purify with β2-microglobulin (β2m, the class I MHC light chain) when expressed in insect cells (data not shown). We determined the structure of the 2L–TNFα complex to 2.8-Å resolution by molecular replacement (Supplementary Table 1 and Supplementary Methods; R_cryst = 23.6%, R_free = 26.6%). The structure revealed a central TNFα trimer bound by three symmetrically arranged 2L molecules, analogous to a three-fold symmetric host TNFR–TNF complex (TNFR1 bound to TNFβ (lymphotoxin α)⁵) (Fig. 1).

Figure 1.

Structure of 2L–TNFα and comparison with TNFR1–TNFβ. Disulfide bonds and ordered carbohydrates on 2L are shown as yellow sticks. Left, the 2L–TNFα complex shown as a ribbon diagram in two orientations: looking down the three-fold symmetry axis (above) and with the three-fold axis vertical (middle). Below left, a surface representation of TNFα from the complex structure, with the 2L-binding site highlighted in red at one of three interfaces on the TNFα trimer (TNFα subunits are different shades of blue). Right, The TNFR1–TNFβ structure⁵ shown as a ribbon diagram in analogous orientations (top and middle). Below right, a surface representation of TNFβ from the complex structure with one of the TNFR1-binding sites highlighted in red.

In common with classical class I MHC molecules and MHC homologs, a diverse family that includes host and viral proteins with immune and non-immune functions⁶, the 2L structure also includes an α1-α2 superdomain (an antiparallel β-sheet that is topped by two α-helices) and an α3 domain (an immunoglobulin constant region fold) (Supplementary Fig. 3a,b). Disulfide bonds are found in the expected positions for class I MHC molecules and homologs: one linking the α2 domain helix to the β-sheet floor of the α1-α2 platform and one joining the two β-sheets in the α3 domain. An additional disulfide bond near the C terminus joins residues separated by 3 residues (Supplementary Fig. 2).

Similar to many class I MHC homologs, 2L does not contain a noticeable groove between its α1 and α2 helices (groove surface area of 60 Ų) (Supplementary Fig. 3c and Supplementary Table 2), and there was no unexplained electron density in this region. In this regard, 2L is more similar to HFE and FcRn (groove surface areas of 415 Ų and 235 Ų, respectively)⁷, which do not present endogenous peptides, than it is to classical peptide-binding class I MHC molecules, the peptide-binding viral homolog UL18 or the lipid-presenting CD1d1 protein (groove surface areas of ∼760 Ų, 900 Ų and 1440 Ų, respectively)^7–9. Thus 2L is not expected to bind/present endogenous peptides or other antigens.

Groove closure in 2L is accomplished by displacement of the N-terminal half of each helix toward the opposite helix and insertion of large side chains, including Trp69 and Met140, which would prevent binding of a peptide antigen (Supplementary Fig. 3b). The 2L α1 helix is shorter than its MHC counterpart (Supplementary Fig. 2) because it includes irregular structure extending toward the α2 domain. The beginning of the 2L α1 domain helix is anchored by the insertion of Tyr64 into a hydrophobic pocket within the α1 domain. The irregular structure preceding the long helix positions His58 of the α1 domain almost over the α2 helix, where it interacts with Tyr87 of TNFα.

The 2L α3 domain includes an unusual large insertion between α3 strands C and D (residues 220–252) (Fig. 1 and Supplementary Figs. 2 and 3a,d). This insertion makes extensive contacts with the α1-α2 platform underside, resulting in an α3 domain and platform quaternary arrangement that differs markedly from β2m-binding MHC molecules and homologs. Whereas the distal end of the α3 domain in β2m-binding MHC molecules is positioned almost directly under the α1-α2 platform center, the 2L α3 domain is tilted ∼42° toward the α2 portion of the platform and also rotated ∼40° about its long axis as compared with its counterpart in HLA-A2 (Supplementary Fig. 3e).

The solvent-accessible surface area buried at the 2L platform–α3 interface (1,970 Ų) is comparable to the area buried at the HLA-A2 platform interface with both its α3 and β2m domains (2,080 Ų), but most of the HLA-A2 interface area results from platform interactions with β2m (1,520 Ų). The large interdomain area in 2L suggests that the α3 domain C-D loop, particularly residues 242–246, has a role analogous to β2m in other MHC molecules; that is, it establishes the relative orientation of α3 with respect to the α1-α2 platform (Supplementary Fig. 3d). In 2L, extensive contacts between the platform and the α3 insertion, interactions between α3 and the α2 domain β5-β6 loop and ionic interactions between the N-terminal amino group and acidic residues (Asp211 and Asp286) in proximal loops of the α3 domain presumably lock the position of α3 with respect to the α1-α2 platform. This effectively pre-organizes the TNFα-binding surface, which involves residues from all three 2L domains (Supplementary Fig. 3e).

Although the position of the 2L α3 domain does not overlap with β2m's location in MHC class I–like molecules, N-linked glycosylation at Asn6 under the 2L α1-α2 platform (Supplementary Fig. 3a) would prevent the canonical β2m interaction. The corresponding residue in the non–β2m-binding protein MIC-A (suggested to be the source of 2L proteins by horizontal gene transfer¹⁰) is also glycosylated¹¹. In addition, the altered position of the 2L α3 domain is not compatible with simultaneous binding of β2m to the α1-α2 platform and to α3 (Supplementary Fig. 3f). Thus, it seems that 2L lacks β2m to allow the α3 domain to be brought closer to the α2 helix to form a larger TNFα-binding site (Fig. 2).

Figure 2.

The 2L–TNFα interface. (a) Surface representations of 2L (left) and the 2L–TNFα complex (right). 2L is magenta and TNFα subunits are different shades of blue-green. Contact surfaces (≤4.0 Å) are highlighted in blue on 2L and red on TNFα. (b) Ribbon diagrams corresponding to the surfaces shown above. The C terminus of 2L is distant from the TNFα-binding site, rationalizing why truncations in this region did not affect binding to TNFα¹⁹. (c) Stereo close-up view of 2L–TNFα interface, with the α3 domain insertion highlighted in blue.

2L binds to TNFα using a large and highly complementary interface (Fig. 2a,b), rationalizing the picomolar affinity between 2L and TNFα⁴. The binding site spans nearly 40 Å, and the total buried surface area at the 2L–TNFα interface (2,060 Ų, of which 2L contributes 1,030 Ų and adjacent TNFα subunits contribute 710 Ų and 320 Ų) is larger than typical protein-protein interfaces (1,560–1,700 Ų)¹². The interface is also highly complementary, as assessed by the shape complementarity index (S_C)¹³ (S_C = 0.72 for 2L–TNFα,where a value of 0.0 indicates no complementarity and 1.0 indicates a perfect fit), compared with antibody-antigen interfaces (S_C = 0.64–0.68)¹³. The TNFR1–TNFβ interface is less complementary (S_C = 0.64), but buries a similar surface area (2,120 Ų, with 1,090 Ų from TNFR1 and 530 Ų and 510 Ų from adjacent TNFβ subunits).

2L inhibition of TNFα–TNFR interactions is rationalized by overlap between the 2L- and TNFR-binding sites on the TNF trimer (Fig. 1). Residues from all three 2L domains contribute to binding to TNFα at the shallow groove between adjacent TNFα subunits, burying loops D-E and A-A′ (also crucial for TNFα binding to TNFR1 and TNFR2 (ref. 14) and TNFβ binding to TNFR1 (ref. 5)) (Figs. 1 and 2 and Supplementary Table 3). Given the high (picomolar) affinity of the 2L–TNFα interaction, we assume that 3:3 2L–TNFα complexes are formed in vivo. However, formation of 3:3 2L–TNFα complexes may not be necessary for inhibition of signaling through TNFR because this kind of signaling requires multiple intact TNFR-binding sites on TNF, as demonstrated by inhibition of signaling when TNF exchanges subunits with dominant-negative TNF variants¹⁵.

The structure of TNFα shows few changes upon binding to 2L (Supplementary Fig. 4). The 2L–TNFα interface primarily involves TNFα D-E loop residues 86–87 and A-A′ loop residues 30–34 from the neighboring subunit interacting with 2L α2 helix residues 157–175; for example, the side chain of TNFα Tyr87 in the D-E loop, a crucial residue for binding both TNFRs¹⁴, contacts the kink in the 2L α2 domain helix and makes a hydrogen bond with His58 from the α1 helix (Fig. 2c). Some features of the TNF-binding sites for 2L and host TNFRs differ: G-H loop residues (Asp143 and Glu146) and residues in the E-F loop (part of the TNFα-binding site for TNFR1 (ref. 16)) do not contact 2L, and TNFα strand B′, which interacts with the insertion in the 2L α3 domain, is not part of the site for host TNFR binding.

Unlike host TNFRs, 2L does not bind to TNFβ or other TNF family members⁴. Superimposing TNFβ onto the 2L–TNFα structure shows no main chain steric clashes (data not shown), suggesting the specificity of 2L for TNFα is related to complementarity between acidic (2L Glu99 and Asp289) and basic (TNFα A′-A″ loop Arg31 and Arg32) residues, which are not present in TNFβ (Supplementary Fig. 5). Binding of TNFβ may be also prevented by different positions of the TNF D-E loops (residues 85–89 in TNFα and residues 106–110 in TNFβ), which differ by ∼2.5 Å (measured between the Cα atoms of analogous residues: TNFα Tyr87–TNFβ Tyr108).

Anti-TNFα antibodies and TNF receptors are used clinically to inhibit autoimmune inflammatory processes¹⁷. The 2L–TNFα structure reveals the molecular mechanism used by poxviruses to attenuate host immune responses and thus provides further insight into how TNFα may be modulated. The structure could facilitate efforts to develop 2L-based anti-TNFα pharmaceuticals; specifically, efforts to minimize the immunogenicity of this viral protein by removing T and B cell epitopes¹⁸ could be guided by identification of 2L residues involved in TNFα binding and maintaining its tertiary structure.