Crystal structure of baculovirus P35 reveals a novel conformational change in the reactive site loop after caspase cleavage.

Baculovirus P35 is a universal suppressor of apoptosis that stoichiometrically inhibits cellular caspases in a novel cleavage-dependent mechanism. Upon caspase cleavage at Asp-87, the 10- and 25-kDa cleavage products of P35 remain tightly associated with the inhibited caspase. Mutations in the alpha-helix of the reactive site loop preceding the cleavage site abrogate caspase inhibition and antiapoptotic activity. Substitution of Pro for Val-71, which is located in the middle of this alpha-helix, produces a protein that is cleaved at the requisite Asp-87 but does not remain bound to the caspase. This loss-of-function mutation provided the opportunity to structurally analyze the conformational changes of the P35 reactive site loop after caspase cleavage. We report here the 2.7 A resolution crystal structure of V71P-mutated P35 after cleavage by human caspase-3. The structure reveals a large movement in the carboxyl-terminal side of the reactive site loop that swings down and forms a new beta-strand that augments an existing beta-sheet. Additionally, the hydrophobic amino terminus releases and extends away from the protein core. Similar movements occur when P35 forms an inhibitory complex with human caspase-8. These findings suggest that the alpha-helix mutation may alter the sequential steps or kinetics of the conformational changes required for inhibition, thereby causing P35 loss of function.

Apoptosis or programmed cell death is an active process of cellular self-destruction essential in normal development, tissue homeostasis, and defense against foreign pathogens, including viruses (1,2). Disruptions in the apoptotic program are associated with diseases such as AIDS, Alzheimer's disease, cancer, and those caused by viruses (3). There are a multitude of signals that can trigger apoptosis. However, all signals culminate in the activation of a well-conserved family of aspartylspecific cysteine proteases called caspases (4,5). Caspases are expressed as inactive proenzymes that are proteolytically activated via autoactivation, transactivation, or cleavage by other caspases (6,7). The activation of caspases represents the decisive commitment to apoptotic death, suggesting that these enzymes are important targets for antiapoptotic drugs (7)(8)(9).
During viral infection, host cells often respond by undergoing apoptosis as an innate defense mechanism. Consequently, viruses have evolved diverse antiapoptotic genes to prevent the premature cell death of host cells and thereby promote virus replication (1). Baculovirus P35 is particularly interesting because not only can P35 prevent virus-induced apoptosis in insect host cells but it can also prevent cell death in phylogenetically diverse organisms when produced ectopically (10 -13). The ability of P35 to act as a general apoptotic suppressor is correlated with the stoichiometric inhibition of the cellular caspases through cleavage of P35 and formation of a stable P35-caspase complex, which precludes subsequent caspase protease activity (14 -16).
Previously, we reported the crystal structure of the active uncleaved P35 at 2.2 Å resolution (17). The crystal structure provided insight into P35's multistep mechanism of caspase inhibition. The most remarkable feature of the P35 structure is a large loop domain (residues 60 -98) that protrudes above the central ␤-sheet core. The loop contains the caspase recognition site, 84-DQMD-87 (P 4 -P 1 residues). The apex of the reactive site loop (RSL) 1 consists of the caspase cleavage site, Asp-87-Gly-88, which is solvent-exposed and fully accessible to the caspase target. The loop is maintained and stabilized by the single amphipathic ␣-helix (␣1) that traverses and interacts with the top of the central ␤-sheet. Distortion of the ␣-helix by substituting a proline residue for valine 71 or by replacing hydrophobic residues of the ␣-helix with charged residues caused loss of caspase inhibition (17,18). Thus, a distinct conformation of the RSL as mediated through ␣1's interaction with P35's ␤-sheet core is necessary for the inhibition of caspases.
Caspases recognize and cleave V71P-mutated P35 at Asp-87, indicating that the uncleaved mutant structure is comparable to that of wild-type P35. However the mutant P35 fails to form a stable caspase-P35 complex and therefore does not inhibit caspase activity or apoptosis (17). Thus, caspase inhibition by P35 involves post-cleavage events (17). To determine the molecular mechanism by which these subsequent steps inhibit caspase and reveal insights on the post-cleavage conformational changes in the RSL, we have determined the threedimensional crystal structure of the nonfunctional P35 V71P * This work was funded by National Institutes of Health Grants GM56774 (to A. J. F.) and AI40482 (to P. D. F.) and the W. M. Keck Foundation Center for Structural Biology at University of California, Davis. 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.
The mutant after cleavage by its cognate caspase. Comparison of the cleaved mutant P35 with the uncleaved wild-type model revealed two large conformational changes that provide valuable clues in identifying functionally important P35 domains. These changes also revealed a novel mechanism for cleavagedependent protease inhibition.
Recently, the crystal structure of P35 complexed with human caspase-8 has been determined (19). This structure reveals the presence of a thioester covalent adduct between the requisite Asp-87 of P35 and the caspase active site cysteine. Additionally, the amino terminus of P35 swings out and interacts with the caspase active site. Surprisingly, the structure of V71P loss-of-function P35 mutant reported here is strikingly similar to that seen in wild-type protein when complexed with caspase. Our findings suggest that the V71P mutant may alter the kinetics or sequential order of movements in P35 needed for caspase inhibition.
Crystallization and Data Collection-Caspase-3-cleaved V71P P35 mutant was crystallized using hanging-drop vapor diffusion at 25°C in 100 -400 mM sodium chloride and 9 -13% polyethylene glycol (M r 20,000), buffered at pH 6.0 with 100 mM 4-morpholineethanesulfonic acid. Crystals of the trigonal space group P3 2 21 appeared in 2-3 days and grew to a size of 200 ϫ 200 ϫ 50 m. The unit cell parameters are as follows: a ϭ b ϭ 75.81 Å, and c ϭ 120.69 Å. There is 1 molecule/ asymmetric unit, giving rise to a solvent content of 62.1% (V M is 3.27 Å 3 /Da) (21). These crystals diffract to 4.0 Å at home source and 3.1 Å at the synchrotron source.
Crystals used for data collection were cryoprotected using 30% ethylene glycol or 30% 2-methyl-2,4-pentanediol in 11% polyethylene glycol (M r 20,000), 100 mM sodium chloride, and 100 mM 4-morpholineethanesulfonic acid, pH 6.0. The crystal was immediately mounted in a loop and frozen in a nitrogen stream at Ϫ170°C. Data were collected on the MAR345 image plate system at the Stanford Synchrotron Radiation Laboratory. The 3.1 Å resolution data set from the trigonal crystal was collected on Beamline 9-1, whereas the 2.7 Å resolution data set from the orthorhombic crystal was collected on Beamline 7-1 at the Stanford Synchrotron Radiation Laboratory. Both data sets were processed with DENZO and SCALEPACK (22). Table I lists data collection and processing statistics.
Phase Determination-The structure of the caspase-3-cleaved V71P P35 mutant from the trigonal P3 2 21 crystal was determined by molecular replacement method with the CCP4 program AMoRe (23). The wild-type P35 structure was used as a search model (17). Data between 10 and 4 Å resolution were used in the rotation search, yielding two equivalent peaks that corresponded to the two possible choices of origin in the P3 2 21 space group. The rotation solutions were then applied in a translation search, resulting in a distinct solution with an R-factor of 48.6% and a correlation coefficient of 0.35. After least squares fastrigid-body refinement (AMoRe) of the wild-type P35 model using data between 20 and 4.0 Å resolution, the R-factor and correlation coefficient improved slightly to 47.81% and 0.441, respectively.
Model Building and Refinement-The initial electron density map obtained from molecular replacement solution of the trigonal P3 2 21 crystal data set was readily interpretable. However, residues corresponding to the reactive site loop (62-97) were disordered and removed before the first cycle of CNS refinement (24). Model building was carried out using the molecular graphics program O (25). After the first round of least squares refinement with CNS, the conventional R-factor dropped to 37.52%, and the R free was 41.91% (5% of data set) for the recorded data between 30 and 3.1 Å resolution (26). The new density map allowed tracing of additional residues in the reactive site loop. After iteratively subjecting the model to several rounds of simulated annealing and group temperature factor refinement with CNS and manual rebuilding of the electron density map, the final conventional R-factor is 24.0% for 95% of all recorded data, and the final R free value is 29.5% for 5% of all recorded data. The structure is characterized by a high overall B factor (B ave ϭ 75.6 Å 2 ), consistent with the Wilson plot statistics that predict an overall B factor of ϳ70 Å 2 . The high temperature factor is attributed to the limited crystal contacts made by the trigonal monomer with monomers in the adjacent asymmetric units. Despite the high B factor, none of the 274 residues modeled plot in the disallowed region of the Ramachandran plot, and 98% of the residues lie in the most favored or additionally allowed regions as defined by the program PROCHECK (27).
The 3.1 Å resolution structure from the trigonal P3 2 21 crystal was used as a search model for the molecular replacement solution of the 2.7 Å resolution data set from the orthorhombic P2 1 2 1 2 1 crystal (see above). The rotation search using data between 20 and 4 Å resolution did not yield any one peak that was significantly larger than the rest of the solutions. The top 20 rotation solutions were applied in the translation search for the first molecule, resulting in a distinct solution with a marginal R-factor and correlation coefficient of 52.2% and 0.241, respectively. The translation search for the second molecule in the asymmetric unit yielded a distinct but poor solution with an R-factor and correlation coefficient of 53.6% and 0.309, respectively. The translation search for the third molecule was performed and yielded a solution with an R-factor and correlation coefficient of 38.4% and 0.626, respectively. After the first round of refinement with CNS of the three molecules, the R-factor and R free were 26.64% and 31.79%, respectively. The model was subjected to several iterative rounds of simulated annealing refinement with CNS (24), applying noncrystallographic restraints between the three subunits in the asymmetric unit (which was relaxed in the later stages), and manual model building with O (25). The final R-factor is 20.1% for 95% of all recorded data, and the final R free is 25.7% for 5% of all recorded data. The three monomers in the asymmetric unit have a much lower overall B factor (B ave ϭ 36 Å 2 ) compared with the trigonal model. This is attributed to the tighter crystal packing observed in the orthorhombic model. All of the 882 residues plot in the most favored or b R-factor and R-free ϭ ͚ʈF obs ͉ Ϫ ͉F calc ʈ/͚͉F obs ͉ ϫ 100 for 95% of recorded data (R-factor) or 5% data (R-free). Numbers in parentheses represent highest resolution shell.

Structure of Cleaved Baculovirus P35
additionally allowed regions of the Ramachandran plot as defined by the program PROCHECK (27).

RESULTS
The caspase-cleaved P35 V71P mutant crystallized in two different space groups, P3 2 21 and P2 1 2 1 2 1 , which diffract xrays to 3.2Å and 2.7Å resolution, respectively. The structure solutions were determined by molecular replacement methods using the uncleaved wild-type P35 as a search model. The solution of the trigonal crystal form consists of one molecule in the crystallographic asymmetric unit, whereas the orthorhombic solution consists of three subunits in the asymmetric unit. The main core of the wild-type P35 structure, the eightstranded ␤-sheet, the ␣-helix that traverses the ␤-sheet, the cellulose-binding domain-like region, and the large hairpin loop and the helix turn helix region, are conserved in the cleaved V71P mutant (Fig. 1). The main-chain rms deviation between these regions in the wild-type P35, minus the RSL (residues 87-101), and the corresponding regions of the cleaved V71P is 2.226 Å.
The first striking feature of the caspase-cleaved V71P mutant structure is observed in the carboxyl-terminal half of the reactive site loop (residues 88 -98). These residues fold back and rearrange to form a new ␤-strand, ␤EЈ (Figs. 1 and 2). The new strand, ␤EЈ, augments the outside of a ␤-sheet formed by strands ␤D, ␤E, and ␤F. This conformational change is observed in both trigonal and orthorhombic crystal forms. The main chain of ␤EЈ hydrogen bonds with ␤E main chain atoms, whereas its side chains interact with side chains of ␣2 and main chain and side chain atoms of ␤DЈ. Fig. 2 shows a representative 2Fo-Fc electron density map revealing the new strand ␤EЈ. The first residue observed in the electron density map for the carboxyl-terminal half of the reactive site loop in the cleaved structure is Ser-92, whereas residues 88 -91 are disordered.
The swinging down of the carboxyl-terminal half of the re-active site loop results in a major relocation of its residues (Fig.  1). One interesting rearrangement involves the hydrophobic residues Ile-93, Tyr-95, and Phe-96. In the wild-type structure, residues, Ile-93, Tyr-95, and Phe-96 are completely solventexposed in the carboxyl-terminal part of the RSL (Fig. 3A). After cleavage by caspase and the rearrangement of the carboxyl-terminal half of the RSL in the V71P mutant, these residues become partially buried in a hydrophobic patch (Fig.  3B). Ile-93 and Tyr-95 reside on the same side of the new ␤EЈ and pack against Ile-118, whereas Phe-96, on the opposite side of ␤EЈ, packs in the hydrophobic patch outlined by Ile-53, Val-56, and Val-103. In contrast to the large movements observed in the carboxylterminal portion of the RSL, the region immediately preceding the scissile bond, Lys-80-Asp-87, moves very little. The hydrophobic interactions of Tyr-82 with side chains Tyr-260 and Trp-262 (in ␤L) seen in the wild-type structure are conserved in the cleaved model. Of the caspase recognition sequence 84-DQMD2G-88, electron density unambiguously defines main chain and side chain atoms up to Gln-85, whereas only main chain atoms of Met-86 and Asp-87 are observed. Asp-87 shifts toward the ␤L-␤K hairpin loop region (residues 252-257), which itself shifts away from the RSL (Fig. 1).
The second remarkable difference observed between the wild-type structure and the cleaved V71P mutant structure is located at the protein's amino terminus. The short aminoterminal strand ␤A (residues 2-5), which is buried in the hydrophobic core of the wild-type structure, releases and is fully solvent-exposed in the cleaved structure (Fig. 1). In the trigonal crystal form, residues 1-5 are disordered, whereas the main chain atoms of residues 6 -10 are observed to extend out and interact with a symmetry-related molecule; in the orthorhombic crystal form, residues 1-11 are disordered. The swinging out of the amino terminus creates a cavity in the cleaved model where ␤A was inserted. In the orthorhombic crystal form, wa- The long arrow represents the conformational change that occurs after cleavage of the V71P P35 mutant by caspase. The amino terminus, which is disordered in the orthorhombic crystal form, is extended and interacts with a symmetry-related molecule in the trigonal crystal form. Very small deviations are seen in the P35 ␤-sheet core. All protein structure figures were generated using the program Bobscript (28) or Molscript (29) and rendered with Raster3D (30).
ter molecules and a DTT molecule, which covalently modifies Cys-137, occupy this cavity. DTT is used in the reaction buffer (10 mM) and in the storage buffer (1 mM).
Most of the more conservative changes seen between the cleaved structure and the wild-type structure are localized in loops of the amino-terminal cellulose-binding domain-like region. Residues 97-101 in the carboxyl-terminal portion of the reactive site loop hydrogen bond with residues 163-166 in the loop between strands ␤F and ␤G. These residues in turn interact with residues 35-44 in the ␤C-␤D loop, which in turn interacts with residues 8 -12 in the ␤A-␤B loop. Thus, the carboxyl-terminal segment of the reactive site loop acts as a restraint on these loops, keeping them in a relatively fixed orientation. The cleavage of the reactive loop and its subsequent rearrangement remove the restraint and network of interactions in the wild-type structure that hold these loops in place. As a result, the loops become more flexible, and large movements are observed (Fig. 4). The c␣ of Gly-162 in the ␤F-␤G loop shifts 10 Å past the vacated RSL restraint (Fig. 4). The displacement in the ␤F-␤G loop pulls the ␤C-␤D loop, which in turn pulls the ␤A-␤B loop. The conformational change in the ␤A-␤B loop destabilizes the amino terminus (strand ␤A), causing it to disengage and swing up and away and become solvent-exposed in the cleaved structure.
The amphipathic helix ␣1, which initiates the reactive site loop (residues 60 -98) in the wild-type structure, is slightly kinked in the cleaved structure as a direct result of the Pro mutation at Val-71. The bend in the helix shifts the residues before Pro-71 (residues 54 -70) with only minimal movement in the main chain of residues after Pro-71 (residues 71-84) (Figs. 1 and 5). This distortion in helix ␣1 is stabilized by a new salt bridge between Asp-69 and Lys-73 within the helix. Despite the distortion in helix ␣1, the interactions between the hydrophobic residues of helix ␣1 and the complementary hydrophobic residues of the ␤-sheet main core are still intact. In addition to the bending of helix ␣1, an additional turn is observed at the amino-terminal portion of helix ␣1. This could account for the loop between strand ␤D and helix ␣1 in the wild-type structure packing closer to strand ␤G and forming a new ␤-strand, ␤DЈ (residues 55-58) (Fig. 3). Thus, distortion of ␣1 may have caused a destabilization in the conformation of the reactive site loop, which could explain why the Val-71 to Pro substitution fails to stably interact with caspase after cleavage. In fact, disruption in the conformation of helix ␣1 by Ala-Ser insertional mutagenesis at residue 74 or alteration of the size of the RSL by Ala-Ser insertion at residue 83 also results in a loss of antiapoptotic activity (16).
Lastly, another disparity of note between the wild-type and mutant structures is observed in the side chain of Tyr-101. In the wild-type structure, Tyr-101 points toward the core of the protein and is buried in the hydrophobic pocket defined by Leu-155, Leu-167, Cys-153, and Met-45, and its phenolic oxygen hydrogen bonds with the side chain of Asn-32. In the cleaved mutant structure, Tyr-101 swings out 180°and points toward the surface of the protein, with its phenolic oxygen exposed to the solvent (Fig. 3). This results in a movement of 12.6 Å in the phenolic oxygen. This large movement is directly due to the movement on the carboxyl-terminal portion of the RSL.
Recently the crystal structure of wild-type P35 complexed with human caspase-8 was reported (19). The structure reveals that P35 is trapped as a thioester covalent adduct between Asp-87 (P 1 ) of P35 and the nucleophile cysteine residue (Cys-360) of caspase. During normal substrate cleavage, a water molecule would complete the hydrolysis of the thioester intermediate. However, the amino terminus of wild-type P35 is also released after cleavage and interacts with the active site histidine residue, preventing access or activation of a water molecule (19). Interestingly, all conformational changes seen in P35 after cleavage by caspase-8 and subsequent formation of a covalent inhibitory complex are also observed in the caspase-3-cleaved V71P P35 mutant. The rms deviation between the 284 equivalent ␣-carbons of the cleaved P35 in the complex structure and the caspase-cleaved mutant is only 1.58 Å (Fig.  5). The largest differences are observed in the disposition of the amino terminus (residues 6 -12) and the caspase-recognition sequence (residues 84 -87). These differences are due to crystal contacts in the uncomplexed structure, but the gross overall movements are similar. Otherwise, the two structures are virtually identical. One intriguing similarity observed in both structures is the kink in helix ␣1. The distortion in helix ␣1 in the cleaved mutant P35 is directly attributable to the valine to proline substitution at central residue 71, yet a similar distortion is seen in the wild-type helix after cleavage with caspase The density is clearly defined for both main chain and side chain residues of ␤EЈ. (Fig. 5). Similar dispositions are seen in the ␤C-␤D and ␤F-␤G loops in both P35 crystal structures that are likely the result of the movement of the reactive site loop restraint described above. Additionally, the large movement seen in the Tyr-101 side chain is also observed in both wild-type and mutant structures.

DISCUSSION
The stoichiometric inhibition of caspases by P35 is a multistep process that involves cleavage of the reactive site loop and a subsequent formation of a very stable complex (15)(16)(17)(18)(19). The structure of wild-type P35 shows a solvent-exposed reactive site loop that projects from the main core of the protein. At the apex of the loop is the caspase recognition site 84-DQMD-87; thus it is easily accessible to the target caspase (17). Whereas cleavage at Asp-87 (P 1 ) is required for antiapoptotic activity, it is not sufficient for stable association and inhibition of caspases. Thus, a post-cleavage conformational change occurs to stabilize the interaction of P35 with caspases and accounts for its high affinity binding. The three-dimensional structure of a caspase-cleaved P35 V71P mutant reveals two significant conformational changes. The carboxyl-terminal segment of the reactive site loop folds back down to the side of the protein and forms a new ␤-strand (Fig. 1). This conformational change is The hydrophobic residues Ile-93, Tyr-95, and Phe-96, which are solvent-exposed in the wild-type structure (A), become partially buried in the cleaved mutant structure (B). Additionally, Tyr-101, which is completely buried in the wild-type precleaved structure, flips out 180°in the cleaved V71P mutant structure. Also, the new interactions formed between residues Arg-58, Lys-94, Asp-98, and His-100 of the two structures are shown. characterized by the stabilization of hydrophobic residues (Ile-93, Tyr-95, and Phe-96) that are solvent-exposed in the wildtype precleaved structure (Fig. 3). The post-cleavage environment for these hydrophobic residues is provided by the regions that already exist in the precleaved form (i.e. the hydrophobic pocket for Phe-96) or by conformational changes involving other regions in the protein (i.e. shifts ␣1-loop-␣2 domain). This RSL movement indirectly causes the second large conforma- FIG. 4. Stereo superposition between the P35 wild-type (green) and V71P caspase-cleaved mutant (red) revealing the conformational changes observed in the ␤A-␤B, ␤C-␤D, and ␤F-␤G loops. View is looking from the top of P35, rotated approximately 90°about the horizontal toward the viewer from Fig. 1. The disposition of the loops is highlighted in dark green (wild-type) and red (V71P). The carboxylterminal portion of the reactive site loop interacts with and fixes the ␤F-␤G loop in the precleaved wild-type structure. In the V71P mutant, after cleavage, the carboxyl-terminal portion of the reactive site loop relocates and no longer restrains the ␤F-␤G loop, which pulls the ␤C-␤D loop with it, which in turn pulls the ␤A-␤B loop, causing the amino terminus to unpack from the protein's core. Part of the ␤C-␤D loop undergoes dramatic rearrangement and forms one turn of a 3 10 helix.
FIG. 5. Stereo superposition between the wild-type uncleaved P35 (black), V71P caspase-cleaved mutant (magenta), and cleaved wildtype P35 (cyan) as complexed with caspase-8 (Protein Data Bank coordinates 1I4E). A, front view; B, top view (90°rotation from A). The superposition reveals that the post-cleavage conformational changes observed in both cleaved forms of P35 are virtually identical (rms deviation is 1.58 Å). A bend in helix ␣1 (residues 60 -79) is observed in both nonfunctional V71P mutant (magenta) and wild-type (cyan) cleaved structures. Additionally, the dispositions of the ␤C-␤D loop (34 -42) and ␤F-␤G (156 -165) loop are very similar in both cleaved structures but differ greatly from uncleaved wild-type P35 (black). tional change: release of the P35 amino terminus from the protein core.
Similar conformational changes are also observed in wildtype P35 after cleavage and subsequent formation of the inhibitory complex with caspase-8 (19). The kink in helix ␣1 and the considerable movements seen in the carboxyl-terminal portion of the reactive site loop and the release of the amino terminus are observed in both cleaved wild-type P35 (complexed with caspase-8) and V71P-mutated P35 crystal structures. Additionally, the rearrangements in the ␤C-␤D and ␤F-␤G loops are also observed in both P35 crystal structures. However, given the similar post-cleavage conformational changes seen in the wild-type and loss-of-function mutant, the P35 V71P mutant does not form an inhibitory complex with caspase (17). This suggests the possibility that the premature distortion of helix ␣1 before cleavage by exchanging Val-71 with Pro may alter the sequential steps or kinetics of protein movements after caspase cleavage that are required for caspase inhibition. In the P35caspase-8 crystal structure, the amino terminus releases and interacts with the active site, possibly blocking the thioester hydrolysis through solvent exclusion or reorientation of active site residues. The kinetics of the amino-terminal release may be detained in the V71P mutant, thus allowing a water molecule to finish hydrolysis of the thioester between Asp-87 and the active site cysteine, and thereby allowing the V71P mutant to disengage from the enzyme. This would allow for caspase turnover and explain the loss of caspase inhibition by V71Pmutated P35. Therefore, our findings propose that the kinetics or order of P35's multistep mechanism is essential for caspase inhibition, which needs further examination.