Vaccinia Virus Protein A49 Is an Unexpected Member of the B-cell Lymphoma (Bcl)-2 Protein Family

Background: Poxviruses contain several virulence factors with no identifiable cellular homologues. Results: Vaccinia virus immunomodulatory protein A49 adopts the same fold as cellular and viral Bcl-2 family proteins but does not share their function. Conclusion: Vaccinia virus Bcl-2-like proteins resemble each other structurally but have divergent functions. Significance: Viruses hijack protein scaffolds and evolve them to subvert host-cell immunity.

Vaccinia virus (VACV), 3 the vaccine used to eradicate smallpox, has a large dsDNA genome encoding ϳ200 ORFs (1). About half of these genes are not essential for virus replication but rather affect the virulence and host range of the virus by counteracting the host immune response to infection (2). Understanding how VACV modulates the host immune response can yield unexpected insights into cellular innate immunity (3) and is essential to fully exploit the promise of VACV as a safe vaccine vector (4).
VACV A49 is a virulence factor that inhibits the activation of the host transcription factor NF-B (5). NF-B activity is stimulated by extrinsic proinflammatory signals, such as interleukin 1 or tumor necrosis factor ␣, and by intracellular signals, such as the recognition of foreign cytoplasmic RNA by retinoic acidinducible gene I-like helicases (6). NF-B is usually retained, inactive, in the cytoplasm via an interaction with the inhibitor of B (IB␣). Upon proinflammatory stimulation, IB␣ is phosphorylated by IB kinases and subsequently ubiquitinated and degraded. This allows NF-B to translocate to the nucleus and promote transcription of cytokines and chemokines, including interferon ␤, that promote inflammation and an anti-viral state in surrounding cells (7). A49 inhibits the ubiquitination of IB␣ by binding ␤-TrCP, a component of the Skp1⅐Cullin1⅐F-box protein (SCF) ␤-TrCP ubiquitin E3 ligase complex. Mutational analysis suggested that it does so via an N-terminal peptide that binds the ␤-propeller domain of ␤-TrCP in an extended fashion, similar to the binding mode observed for ␤-catenin (8), and thereby blocks binding of IB␣ to the same site (5). The importance of inhibiting NF-B activation to the virus is underscored by the fact that VACV has at least 10 proteins that interrupt this pathway and deletion of any of these genes attenuates the virus (9). Inhibiting the degradation of IB␣ appears to be of particular importance for poxvirus virulence; ectromelia virus lacks an A49 homologue but it contains EVM005, an ankyrin family protein that contains an F-box domain and is thought to com-pete with the F-box region of ␤-TrCP for binding to Skp1, thus preventing IB␣ ubiquitination and degradation (10).
Previous studies have identified that many poxvirus immunomodulatory proteins, including 10 VACV proteins, share structural similarity with the Bcl-2 (B-cell lymphoma 2) family of cellular proteins despite lacking identifiable sequence similarity (11)(12)(13)(14)(15)(16). Cellular members of the Bcl-2 family generally share four Bcl-2 homology domains (BH1-BH4) in their primary sequence, and they are key regulators of apoptosis (17). Whereas poxvirus Bcl-2-like proteins F1, M11, and N1 all act to inhibit apoptosis, other poxvirus Bcl-2 family proteins have instead evolved to inhibit activation of the transcription factors NF-B or IRF3 (9,13). Of particular interest is VACV protein N1, which inhibits both apoptosis and NF-B via two distinct molecular surfaces; inhibition of apoptosis is mediated via a hydrophobic surface groove, whereas inhibition of NF-B is disrupted by mutating a surface on the opposite face of the protein that mediates N1 dimerization both in solution and in cells (18).
Homologues of A49 are found only in a subset of orthopoxviruses (5). Aside from a stretch of six amino acids at its N terminus, the 18.8-kDa A49 protein and its orthologues in other orthopoxviruses share no identifiable sequence identity with any protein of known structure or function. To further investigate the function of A49, we solved its structure to 1.8 Å resolution by x-ray crystallography. Unexpectedly, the structure showed that A49 possesses a Bcl-2-like fold despite not sharing sequence identity with known cellular or poxvirus Bcl-2 family proteins. Structure-based phylogenetic analysis shows that A49 is most closely related to other poxvirus Bcl-2-like proteins, consistent with its evolution from an ancestral poxvirus BCL2 family gene by means of gene duplication and diversification.
Protein Production and Characterization-N1 was expressed and purified as described previously (12). Full-length A49 and A49 ⌬12 were expressed in Escherichia coli Rosetta2-(DE3)pLysS (Novagen). Bacteria were grown in 2ϫ TY medium to an A 600 of 0.8 at 37°C and cooled to 22°C, and protein expression was induced by the addition of 0.2 mM isopropyl ␤-D-thiogalactopyranoside. After 16 h, cells were harvested by centrifugation at 5000 ϫ g for 15 min at 4°C, and the pellet was stored at Ϫ20°C until required.
Crystallization, Structure Solution, Refinement, and Analysis-All crystals were grown by sitting drop vapor diffusion (23) and snap-cryocooled by immersion in liquid nitrogen. Fulllength A49 (100 nl at 9.5 mg/ml) was mixed with 100 nl of reservoir solution and equilibrated at 21°C against 95-l reservoirs comprising 25% (w/v) PEG 3350, 0.2 M ammonium sulfate, and 0.1 M Tris, pH 9.5. Cryoprotection was achieved by quickly sweeping the crystal through a reservoir supplemented with 20% (v/v) glycerol. A49 ⌬12 (1 l at 25.0 mg/ml) was mixed with 1 l of reservoir solution and equilibrated at 20°C against 500-l reservoirs containing 0.1 M HEPES, pH 7.5, 1.6 M ammonium sulfate, and 1.5% (v/v) PEG 400. Crystals were cryoprotected by passage through 2 l of perfluoropolyether oil (Hampton Research) that had been overlaid onto the mother liquor. Because the presence of ammonium sulfate in the mother liquor prevented efficient heavy atom derivatization, ammonium sulfate was substituted for sodium malonate (24), and A49 ⌬12 crystals were grown by mixing 2 l of protein (20 -21 mg/ml) with 2 l of reservoir solution and equilibrating at 20°C against 500-l reservoirs containing 1.3-1.5 M sodium malonate, pH 6.8, 1% (v/v) PEG 400, and 4 -10% (v/v) glycerol, the best crystals being obtained when the reservoir was overlaid with 100 l of a 1:1 mixture of paraffin and silicone oil (Molecular Dimensions). After crystals had grown, 1 l of a 1:100 dilution of a saturated potassium dicyanoaurate(I) (KAu(CN) 2 ) solution was added to selected drops, and crystals were incubated for 6 days. Crystals were cryoprotected by brief immersion in 2.1 M sodium malonate, 1% (v/v) PEG 400, and 5% (v/v) glycerol.
Diffraction data were collected at 100 K on European Synchrotron Radiation Facility beamline ID14-2 and processed with HKL2000 (full-length A49) or on Diamond Light Source beamline I04-1 and processed with XDS and XSCALE (25), as implemented by xia2 (26) (A49 ⌬12). The structure of A49 ⌬12 was solved by single-wavelength anomalous dispersion analysis of a potassium dicyanoaurate(I) derivative using the autoSHARP (27) structure solution pipeline. Eight gold sites were identified with occupancy ranging from 0.88 to 0.18. autoSHARP implemented ARP/wARP (28) to build the initial model that was manually improved with COOT (29) and refined using Refmac5 (30). The resultant model was used to solve the structure of A49 ⌬12 grown in the sodium malonate condition by molecular replacement with MOLREP (31). This model was manually improved with COOT (29) and refined using Refmac5 (30) with one translation-libration-screw anisotropic displacement group per molecule and local non-crystallographic symmetry restraints. The structures of A49 ⌬12 crystallized in the ammonium sulfate condition and of full-length A49 were solved by molecular replacement with MOLREP (31) and PHASER (32), respectively, using the A49 ⌬12 sodium malonate structure as a search model and were refined using Refmac5 as detailed above. The stereochemistry of all structures was assessed and improved using COOT, the MolProbity Web server (33), and WHAT_CHECK (34).
A representative set of proteins with the Bcl-2 fold were selected with the assistance of the Pfam Web server (35). Where multiple structures were available for one protein, crystal structures determined at the highest resolution were selected. Where crystal structures had multiple molecules per asymmetric unit, LSQMAN (36) was used to select the molecule that could be superposed on all others with the lowest C␣ atom root mean square deviation. For NMR ensembles, the well ordered "core" of the most representative member of the ensemble was selected in consultation with the OLDERADO Web server (37). VACV protein F1, which exists as a domain-swapped dimer both in solution and in crystallo (15), was excluded from the analysis. The structural core of the Bcl-2 fold was identified by iterative determination of maximal equivalent substructures, structures were superposed onto this minimal Bcl-2 core, and the resultant superpositions were scored to generate a phylogenetic relationship based on structural equivalence using HSF as described previously (38). Clustering of equivalent substructures and the structure-based phylogenetic tree were visualized using Dendroscope (39). Sequence figures were generate with ALINE (40), molecular graphics were generated using PyMOL (Schroedinger LLC), and images were assembled using Inkscape.
To generate vA49-cTAP, the A49R ORF plus 213 bp upstream was amplified from VACV WR genomic DNA using oligonucleotides 5Ј-GTAGGTACCAACAAAAGGTATTACAAG-AAT-3Ј and 5Ј-GTAGCGGCCGCCAAATATCGTTCGCGG-ATATC-3Ј, containing KpnI and NotI restriction sites (underlined), respectively. The PCR fragment was cloned into a modified pUC13 plasmid upstream of a TAP tag comprising two copies of the streptavidin binding sequence (Strep-tag II) followed by one FLAG epitope (41). The modified pUC13 also contains E. coli guanylphosphoribosyl transferase fused in frame with enhanced GFP under the control of a VACV promoter (42). The right-flanking region of A49R was amplified using oligonucleotides 5Ј-GTACCGCGGAAATATTAAAAA-AAAATA-3Ј and 5Ј-TCTAGACGGATTTCTGTGTTCTCT-TTGAAG-3Ј, containing SacII and XbaI restriction sites (underlined), respectively, and was cloned downstream of the TAP tag to form pA49vTAP. vA49-cTAP was made by transient dominant selection (43) after infection of CV-1 cells with v⌬A49 (5) and transfection with pA49vTAP as described previously (5). VACV WR containing a C-terminally HA-tagged B14R (vB14-HA) was described previously (22).
Co-immunoprecipitation and Immunoblotting-HEK293T cells were transfected using the calcium phosphate method. Briefly, cells were seeded in 10-cm dishes to reach 50% density after 24 h, at which time the medium was refreshed and cells were allowed to rest for 20 min. DNA (5 g) was incubated with 50 l of 2.5 M CaCl 2 and sterile water to a final volume of 500 l for 20 min at room temperature. An equal volume of 280 mM NaCl, 10 mM KCl, 1.5 mM Na 2 HPO 4 , 12 mM glucose, 50 mM HEPES, pH 7.05, was added, and the mixture was incubated for a further 15 min. The mixture was dropped carefully onto the cells, and the dishes were swirled gently to distribute it evenly. After 24 h, the medium was removed, and cells were infected (or mock-infected) with the indicated viruses in fresh DMEM with 2% FBS plus penicillin and streptomycin.
To test binding of A49 to proapoptotic Bcl-2 proteins, cells were infected at a multiplicity of infection of 5 for 6 h before harvesting and washing twice with cold PBS. Cells were lysed with CHAPS buffer (20 mM Tris, pH 8, 137 mM NaCl, 2% (w/v) CHAPS, 1 mM EDTA, pH 8, and protease inhibitors (Roche Applied Science)). Lysates were treated with benzonase (Novagen) for 30 min, cleared, and then incubated with anti-FLAG M2 affinity resin (Sigma-Aldrich) for 16 h at 4°C. After four washes with CHAPS buffer, proteins were boiled off of the beads in 2ϫ SDS-PAGE loading buffer and analyzed by SDS-PAGE and immunoblotting.
To test for A49 self-association, cells were infected at a multiplicity of infection of 2 (multiplicity of infection of 1.5 for each virus when co-infecting) for 16 h before harvesting and washing with cold PBS. Cells were lysed with immunoprecipitation buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 10 mM CaCl 2 , 0.1% (v/v) Triton X-100, 10% (v/v) glycerol, and protease inhibitors (Roche Applied Science)). Cleared lysates were incubated with anti-FLAG M2 affinity resin for 16 h at 4°C. The resin was washed four times with immunoprecipitation buffer before proteins were boiled off of the beads in 2ϫ SDS-PAGE loading buffer and analyzed by SDS-PAGE and immunoblotting.

RESULTS
Crystal Structure Reveals That A49 Adopts the Bcl-2 Fold-Crystallization trials of full-length VACV A49 yielded just two crystals in a single crystallization condition after 4 months of equilibration. Diffraction data were recorded to 3.0 Å resolution (full-length A49; Table 1) but, because A49 does not share identifiable sequence similarity with any proteins of known structure, it was not possible to solve the structure by molecular replacement, and repeated attempts to reproduce these crystals were unsuccessful, frustrating attempts to solve the structure by experimental methods. Previous work had shown that the N-terminal 12 amino acids of A49 are essential for its interaction with ␤-TrCP and share similarity with the ␤-TrCP-binding sequence of ␤-catenin (5). The residues of ␤-catenin that bind ␤-TrCP have an extended conformation (8), suggesting that the equivalent residues of A49 may be poorly ordered and thus inhibiting crystallization. A truncated form of A49 lacking the N-terminal 12 amino acids (A49 ⌬12) was thus expressed and purified. It crystallized readily in two distinct conditions, and diffraction data were recorded to 1.8 -1.9 Å resolution (A49 ⌬12 ammonium sulfate and A49 ⌬12 sodium malonate; Table  1). The structure was solved by single wavelength anomalous dispersion analysis of a A49 ⌬12 crystal soaked with potassium dicyanoaurate(I). This initial model was used to solve the structures of full-length A49 and ⌬12 A49 in two distinct crystal forms by molecular replacement, which were refined as listed in Table 2.
The structure of full-length A49 closely resembles that of the truncated protein, superposing on A49 ⌬12 with 0.50 -0.67 Å root mean square deviation across 140 C␣ atoms. In the fulllength structure, residues 1-17 could not be modeled due to a lack of interpretable electron density N-terminal to residue 18. It is possible that these amino acids were removed by serendipitous in-drop proteolysis prior to crystallization. Residues 13-17 adopt three distinct conformations in the 10 A49 ⌬12 molecules observed in the two crystal forms (Fig. 1A, inset) and are disordered in one of the two molecules of A49 ⌬12 per asymmetric unit in crystals grown in the sodium malonate condition. The only other significant difference between the structures of full-length and ⌬12 A49 is the conformation of the region linking helices ␣2 and ␣3, helix ␣3 being one turn shorter in the full-length structure. In four of the eight molecules of A49 ⌬12 grown in the ammonium sulfate condition and both molecules of A49 ⌬12 grown in the sodium malonate condition, a single residue (Gly) of the C-terminal purification tag is observed; in the structure of full-length A49, the final two residues (161 and 162) and the purification tag were not ordered and could not be modeled.
Unlike M11, A49 Does Not Bind Effector BH3 Peptides-A characteristic feature of antiapoptotic Bcl-2-like proteins is their ability to bind amphipathic BH3 peptides via a hydrophobic surface groove delineated by helices ␣2, ␣3, ␣4, and ␣5 (17). MYXV M11 is a Bcl-2-like protein that inhibits host-cell apoptosis by binding the effector BH3 peptides of proapoptotic Bcl-2 family proteins Bax and Bak (14,48). Superposition of A49 onto the structure of M11 shows that the positions of heli-

FIGURE 1. A49 adopts a Bcl-2-like fold.
A, the structure of A49 ⌬12 is shown in two orthogonal views as a ribbon colored from blue (N terminus) to red (C terminus). The inset shows the alternative conformations of residues 13-17 observed, highlighting the mobility of this region. In crystals of A49 grown from full-length protein, no density was observed for these residues. B, sequence of VACV WR A49. The secondary structure is shown above the sequence, and residues thought to interact with the ␤-propeller domain of ␤-TrCP are in boldface type.
ces ␣2-␣5 are generally conserved between the two proteins ( Fig. 2A). A simple model of A49 bound to a BH3 peptide was generated by superposing A49 ⌬12 onto the structure of M11 in complex with the Bak BH3 peptide (14). As shown in Fig. 2, B and C, although a deep groove is evident on the surface of A49 flanked by helices ␣2-␣5, it is not compatible with the binding of BH3 peptides. The side chain of Bak residue Leu 78 would clash with the charged side chains of A49 residues Lys 97 and Asp 111 . Bak residue Leu 78 interacts with a hydrophobic patch on the M11 surface, is absolutely conserved across BH3 peptides (49) and is essential for strong interaction between BH3 peptides and Bcl-2 family proteins (50). More strikingly, the orientation of A49 helix ␣7 shortens the groove such that Bak residues 80 -90, including a conserved aspartic acid and hydrophobic residue (Asp 83 and Ile 85 in Bak, respectively), would clash with A49. Additionally, helix ␣4 is rotated in A49 relative to its orientation in M11 such that A49 residue Asp 92 would clash with Bak Met 71 . Co-immunoprecipitation experiments confirmed the inability of A49 to bind BH3 peptides of the proapoptotic proteins Bax and Bak (Fig. 2D). HA-tagged Bax and Bak are co-immunoprecipitated from HEK293T cells by an anti-FLAG affinity matrix when co-expressed with FLAG-tagged M11. These proteins are not co-immunoprecipitated from cells expressing nTAP-A49, where the A49 protein is Strep-tag II-and FLAGtagged. Concomitant infection by VACV lacking an endogenous A49R gene (v⌬A49) does not stimulate co-immunoprecipitation of Bax or Bak with A49, confirming that no other viral factor is required for their interaction. Myc-␤-TrCP is co-immunoprecipitated with nTAP-A49, confirming that nTAP-A49 is functional (5) and is not co-immunoprecipitated with FLAG-M11, confirming the specificity of the interaction between A49 and ␤-TrCP.
A49 Does Not Dimerize in Solution or in Cells-A conserved feature of the VACV Bcl-2-like proteins is their propensity to form dimers; all VACV Bcl-2 family proteins except for K7 crystallize as dimers (11-13, 15, 51, 52). Visual inspection of the structures of A49 ⌬12 and full-length A49 showed that, in all cases, two molecules of A49 self-associate via an interaction of helices ␣4 and ␣6 (the "4-6 face" ; Fig. 3A). In the structure of full-length A49 this interaction is formed by crystallographic 2-fold symmetry, whereas in the structures of A49 ⌬12 the interaction is formed by pseudo-2-fold non-crystallographic symmetry. The 4-6 face that mediates the interaction comprises a large patch of hydrophobic residues surrounded by a ring of charged and polar amino acids (Fig. 3, B and C). Whereas the self-association is predominantly hydrophobic in nature, in the structure of full-length A49 a symmetric salt bridge is formed at the periphery of the binding surface between the side chains of Arg 136 and Asp 92 . In the structures of A49 ⌬12 salt bridges Arg 136 -Asp 92 and Lys 103 -Asp 139 are formed, although these bonds are not symmetric and thus formed only once per pair of A49 molecules. In A49 ⌬12 hydrogen bonds are formed between the side chains of Asn 88 and Tyr 102 and the carbonyl oxygens of Thr 128 and Val 95 , respectively, and between the side chains of Arg 136 and Asn 96 , although again these interactions often occur only once per pair of A49 molecules. Analysis of the A49 structures using PDBePISA suggested that this 4-6 face plays an essential role in A49 self-association (CSS scores 0.5-1.0) (53). However, previous studies of VACV B14 showed that the self-association interface observed in crystallo was only of modest affinity in vitro and overlapped with the interface that mediates the interaction between B14 and IB kinase ␤ (IKK␤) in cells (54). We therefore sought to determine the oligomeric state of A49. SEC-MALS of full-length A49 and A49 ⌬12 produced in bacteria showed that both are monomeric across a  (14)) and VACV A49 (yellow ribbons) are shown, superposed, in two orthogonal views. Selected helices are labeled. B, the structure of the human Bak BH3 peptide (green ribbon and side chains) bound to M11 (blue molecular surface) is shown (PDB 2JBY (14)). C, model of A49 in complex with the human Bak BH3 peptide generated by superposing A49 onto the M11-Bak complex (14). Asterisks denote significant clashes. D, unlike M11, A49 does not co-immunoprecipitate with human Bak or Bax. HEK293T cells were transfected with Myc-␤-TrCP ( lanes 5 and 8), HA-Bak (lanes 1, 2, and 6), or HA- Bax (lanes 3, 4, and 7) and nTAP-A49 (lanes 1-5) or FLAG-M11 (lanes 6 -8). After 24 h, cells were infected with v⌬A49 at 5 pfu/cell (lanes 2 and 4) or mock-infected (lanes 1, 3, and 5-8). Cells were lysed 6 h after infection, and lysates were immunoprecipitated (IP) with an anti-FLAG matrix before immunoblotting using the antibodies specified. Tubulin served as a loading control, and D8, a VACV envelope protein, served as a positive marker of infection.
Relationship of A49 to Viral and Cellular Bcl-2 Family Proteins-Exhaustive searches of the NCBI nr (non-redundant) protein sequence database using PHMMER (55) and HHSENSER (56) yielded homologues of A49 only in orthopoxviruses and Yoka poxvirus. The genes encoding A49 homo-logues in ectromelia virus, taterapox virus, monkeypox virus, and camelpox virus are all severely truncated and predicted to not give rise to functional proteins. Among other orthopoxviruses A49 is well conserved, all sequences sharing Ͼ94% identity with A49 from VACV WR, and the residues of the N peptide predicted to interact with ␤-TrCP are absolutely conserved (5). The A49 homologue in Yoka poxvirus shares 33% identity with VACV WR A49. Yoka poxvirus A49 has 13 amino acids N-terminal to helix ␣0, but the sequence of this region differs significantly from that of VACV WR A49.
The three-dimensional structures of distantly related proteins are more highly conserved than their amino acid sequences (57). Given the high sequence divergence between A49 and other poxvirus Bcl-2-like proteins, we wondered whether A49 was structurally closer to poxvirus, herpesvirus, or cellular Bcl-2 family proteins. A representative set of 23 cellular, 5 herpesvirus, and 7 poxvirus Bcl-2 family protein structures, solved in the presence or absence of protein binding partners, was assembled and then aligned and clustered by iterative pairwise comparison of their structural and physicochemical properties as described previously (38). Briefly, pairwise comparison of structures was used to identify the set of equivalent residues that defines their common structural "core," and the two most closely related structures (or structural cores) were merged. This process was repeated until only one core was left, producing the hierarchical clustering of structural similarity shown in Fig. 4A and identifying the minimal set of amino acid residues that defines the Bcl-2 fold. This minimal core, comprising 56 residues, spans the majority of helices ␣1, ␣2, and ␣5 but also comprises elements of ␣3, ␣6 and ␣7 (Fig. 4B, inset). All structures were aligned upon this minimal core and all-pairs pairwise comparison of the structures was performed as described previously (38) to generate the phylogenetic tree shown in Fig.  4B. Strikingly, despite the lack of sequence conservation, the structures of poxvirus Bcl-2 family proteins, including A49, more closely resemble each other than they do cellular or herpesvirus Bcl-2-like proteins, consistent with previous analysis of the poxvirus proteins A52, B14, M11, and N1 (13). A49 lies closest on the tree to MYXV M11 and VACV N1. Both N1 and M11 bind BH3 peptides and inhibit apoptosis (12,18,48), although N1 does so only weakly compared with other VACV antiapoptotic proteins (58). The other VACV proteins that inhibit cellular innate immune responses (A52, K7, B14, and A46) all cluster together despite having a diverse range of cellular binding partners (12, 13, 59 -63). This is in contrast to cellular Bcl-2 proteins, where orthologues that share a function lie closer to each other on the tree than paralogues from the same species with different functions. Overall, the cellular proteins partition away from the virus proteins with the exception of Bid, which falls between the herpesvirus and poxvirus Bcl-2 proteins. Bid is unlike other cellular Bcl-2 family proteins in that it has only one Bcl-2 homology domain, BH3, and cleavage of Bid changes its conformation to promote apoptosis (64 -66). Because its structure is highly divergent from both cellular and viral Bcl-2 proteins (Fig. 4A), its position in the tree probably arises from "long branch attraction" (67) rather than greater structural similarity to viral Bcl-2 proteins.

DISCUSSION
We expressed recombinant VACV A49 and solved its structure to 1.8 Å resolution (Fig. 1). Surprisingly, A49 adopts a Bcl-  68), and Epstein-Barr virus (EBV)). Structures determined by x-ray crystallography are labeled in black, NMR models are shown in gray, and structures determined with BH3 peptides or C-terminal tails bound in the surface groove are italicized. Ribbon diagrams of representative structures for each protein are shown, color-ramped from blue (N terminus) to red (C terminus). The inset shows the 56 C␣ atoms that comprise the minimal core of the Bcl-2 fold, colored as above. Coordinate files containing the core C␣ atoms of the Bcl-2 fold plus superpositions of the 35 representative Bcl-2 family structures upon this core are supplied as supplemental material.
2-like fold, despite lacking identifiable sequence similarity with other members of the Bcl-2 family. The defining characteristic of cellular Bcl-2 family proteins is their involvement in the regulation of apoptosis. Intrinsic apoptosis is triggered when the proapoptotic Bcl-2 family effector proteins Bak and Bax oligomerize and permeabilize the outer mitochondrial membrane, leading to an irreversible caspase cascade and cell death (17). Antiapoptotic Bcl-2 proteins oppose apoptosis via a surface groove formed by helices ␣2-␣5. This groove binds to the exposed BH3 peptides of activated Bax or Bak, preventing their oligomerization. The groove also binds the BH3 peptides of BH3-only proteins like Bim, sequestering such peptides and thereby preventing them from binding to and activating Bax or Bak (17). MXYV protein M11 and VACV proteins N1 and F1 all adopt the Bcl-2 fold and inhibit apoptosis by binding BH3 peptides via a groove on their surface formed by helices ␣2-␣5 (12,14,15,18). The A49 structure shows that A49 lacks a surface groove compatible with binding BH3 peptides and is unable to bind the proapoptotic proteins Bax or Bak (Fig. 2). This indicates that unlike M11, its closest structural relative, A49 does not function to inhibit apoptosis by sequestering effector BH3 peptides.
A49 inhibits NF-B activation by inhibiting ubiquitination and subsequent degradation of IB␣ (5). A49 achieves this by sequestering the E3 ligase ␤-TrCP, preventing it from binding the phosphorylated form of IB␣. The interaction with ␤-TrCP requires A49 residues 6 -12 (5), which contain a double serine motif that is likely to be phosphorylated and bind the ␤-propeller domain of ␤-TrCP in an extended conformation similar to that observed in the complex of ␤-TrCP with ␤-catenin (5,8).
In the structure of full-length A49, we did not observe any interpretable electron density N-terminal to residue 18, nor did we observe "additional" electron density anywhere in the structure that could be interpreted as residues 1-17. Additionally, we observed that A49 crystallized much more readily upon removal of residues 1-12. These observations indicate that the N-terminal ␤-TrCP-binding residues of A49 lack intrinsic structure and are thus freely available to bind ␤-TrCP. However, we note that residues Ser 7 and Ser 12 of full-length A49 produced in E. coli are unlikely to have been phosphorylated, which may possibly influence the conformation of this region.
The HIV-1 immunomodulatory protein Vpu functions similarly to A49, binding ␤-TrCP and preventing degradation of IB␣ (68). However, Vpu also binds the HIV cell surface receptor CD4 and the restriction factor tetherin, promoting their ubiquitination and degradation by bringing them in close proximity to the SCF ␤-TrCP E3 ligase complex (69,70). Although a cellular target of A49-mediated proteasomal or lysosomal degradation has yet to be identified, inspection of the A49 crystal structures identified a surface of the protein that is a prime candidate for mediating such interactions. In all crystal forms presented here the two molecules of A49 self-associate via a symmetric interaction between helices ␣4 and ␣6 (the 4-6 face). This interaction is similar to the self-association observed in crystals of VACV A46 (51) but, unlike A46, A49 does not form dimers in solution or in cells (Fig. 3). Inspection of the 4-6 face reveals a hydrophobic surface cleft that would seem ideal for mediating interactions with cellular binding partners (Fig. 3).
Dimerization of VACV proteins in crystallo by surfaces that mediate binding to cellular partners has been observed before; the 1-6 face of B14 that mediates its reversible self-association in solution overlaps with its binding site for IKK␤ (54), mutations that abolish dimerization of N1 also disrupt its ability to inhibit NF-B activation (18), and a residue of A52 required for binding to TRAF6 lies partly buried within the dimerization interface (71), although in this case maintaining A52 dimerization seems to be required for efficient TRAF6-mediated stimulation of p38 and subsequent induction of IL-10 expression. It is tempting to speculate that A49 binds other cellular factors via the 4-6 face to promote their ␤-TrCP-mediated ubiquitination and degradation, although further experiments are required to probe this hypothesis.
The structure of A49 takes to 11 the number of VACV proteins that have been shown or predicted to share the Bcl-2 fold (Table 3). Although the bulk of poxvirus Bcl-2 proteins share weak but identifiable sequence similarity (16,72), A49 could not be identified as a Bcl-2 family protein based on sequence alone. Nonetheless, structure-based phylogenetic analysis shows A49 to be more closely related to poxvirus than herpesvirus or cellular Bcl-2 proteins (Fig. 4). This is consistent with poxvirus Bcl-2 family proteins having arisen from gene duplication and divergence following a single gene acquisition event, structural similarity having been conserved despite vast sequence divergence. The terminal regions of the poxvirus genome are highly variable, containing non-essential genes that act to determine host range and inhibit the host immune response (2). In addition to terminal transpositions, whereby genes from one end of the linear genome are duplicated at the other end (72-74), a recent study showed that poxviruses deploy "genomic accordions" when under selective pressure, their genomes rapidly expanding to incorporate multiple copies of genes near points of genomic instability (75). This expansion increases the probability that duplicated genes will acquire "advantageous" mutations, potentially conferring a divergent function upon the mutated gene. The host innate immune response places large DNA viruses under significant selective pressure (76), and poxvirus Bcl-2 family proteins all act to inhibit the cellular responses to infection. Inspection of the low frequency duplications observed in the Copenhagen strain of VACV (or strains derived therefrom) (75) show independent gene duplications encompassing several Bcl-2 family proteins: the duplicated VACV Copenhagen region spanning nucleo- . This is consistent with the gene duplication and differentiation events that gave rise to the 11 VACV Bcl-2 family immunomodulatory proteins having arisen as a result of ancestral poxviruses deploying their genomic accordions in response to selective pressure generated by adaptation in the host or changes in host range. Although A49 has high sequence divergence from the other poxvirus Bcl-2 proteins, its absence from Yata-, Lepri-, Sui-, Cervid-, or Capripoxviruses, all of which have multiple Bcl-2 like proteins, makes it a poor candidate for being the BCL2 gene acquired originally by the ancestral poxvirus.
In summary, we have shown that A49 is an unanticipated eleventh member of the VACV Bcl-2-like immunomodulatory protein family. A49 lacks a BH3 peptide binding groove and does not bind the proapoptotic proteins Bax and Bak. Although A49 self-associates via a hydrophobic 4-6 face in crystals, it does not self-associate in solution or in cells, suggesting that this 4-6 face may mediate binding to yet undetermined cellular partners. Conservation of the Bcl-2 fold by poxvirus proteins with highly divergent sequences is consistent with duplication and divergence of an ancestral gene encoding a Bcl-2 family protein.