The Crystal Structure of the Active Domain of Anopheles Anti-platelet Protein, a Powerful Anti-coagulant, in Complex with an Antibody*

Background: Naturally occurring anticoagulant proteins provide models for new medications with highly desirable properties. Results: The crystal structure of the active region of mosquito protein AAPP has been solved. Conclusion: The mosquito protein AAPP uses a small turn region to block coagulation extremely effectively by binding collagen. Significance: New small molecule anti-coagulants may be developed with completely new mechanisms and none of the drawbacks of current treatments. Blood clotting is a vitally important process that must be carefully regulated to prevent blood loss on one hand and thrombosis on the other. Severe injury and hemophilia may be treated with pro-coagulants, whereas risk of obstructive clotting or embolism may be reduced with anti-coagulants. Anti-coagulants are an extremely important class of drug, one of the most widely used types of medication, but there remains a pressing need for novel treatments, however, as present drugs such as warfarin have significant drawbacks. Nature provides a number of examples of anti-coagulant proteins produced by blood-sucking animals, which may provide templates for the development of new small molecules with similar physiological effects. We have, therefore, studied an Anopheles anti-platelet protein from a malaria vector mosquito and report its crystal structure in complex with an antibody. Overall the protein is extremely sensitive to proteolysis, but the crystal structure reveals a stable domain built from two helices and a turn, which corresponds to the functional region. The antibody raised against Anopheles anti-platelet protein prevents it from binding collagen. Our work, therefore, opens new avenues to the development of both novel small molecule anti-clotting agents and anti-malarials.

Blood clotting is a vitally important process that must be carefully regulated to prevent blood loss on one hand and thrombosis on the other. Severe injury and hemophilia may be treated with pro-coagulants, whereas risk of obstructive clotting or embolism may be reduced with anti-coagulants. Anti-coagulants are an extremely important class of drug, one of the most widely used types of medication, but there remains a pressing need for novel treatments, however, as present drugs such as warfarin have significant drawbacks. Nature provides a number of examples of anti-coagulant proteins produced by blood-sucking animals, which may provide templates for the development of new small molecules with similar physiological effects. We have, therefore, studied an Anopheles anti-platelet protein from a malaria vector mosquito and report its crystal structure in complex with an antibody. Overall the protein is extremely sensitive to proteolysis, but the crystal structure reveals a stable domain built from two helices and a turn, which corresponds to the functional region. The antibody raised against Anopheles anti-platelet protein prevents it from binding collagen. Our work, therefore, opens new avenues to the development of both novel small molecule anti-clotting agents and anti-malarials.
For blood-sucking insects, leeches, or other animals, it is important to prevent the blood of the host clotting after puncture of the skin (1,2). These animals produce a number of factors in their saliva to block the action of host defense mechanisms and ensure blood flow (1,2). Anti-clotting factors from these parasitic animals are of medical interest, and leeches have been used historically for cleaning wounds (1,2). Because blood coagulation depends on numerous proteins, it is not surprising that different parasites have evolved anti-clotting factors with different targets (1). Bivalirudin, an oligopeptide analog of the leech anti-clotting factor hirudin, is a direct thrombin inhibitor (3), and desmoteplase, from the saliva of bats, is a plasminogen activator (4 -6). Both have undergone clinical trials. We have identified an abundant protein in the saliva of the female mosquito Anopheles stephensi, a known malaria vector. This protein, anopheline anti-platelet protein (AAPP), 5 is found to bind directly to type-I and type-III collagen and blocks platelet adhesion with an IC 50 of ϳ30 nM (7). Collagen is only exposed to circulating plasma proteins at sites of damage to blood vessels (8 -10). It causes platelets to aggregate through the action of the plasma protein von Willebrand factor, which initiates platelet tethering and blood clotting at sites of vascular injury. Activated platelets subsequently adhere directly to subendothelial collagen through glycoprotein 1b so that blocking exposed collagen prevents the initiation and progression of platelet aggregation. AAPP shows similarity to salivary gland proteins from other mosquito species, including aegyptin from Anopheles aegypti, which also binds collagen (11)(12)(13). These proteins are the first anti-coagulants found to have this mode of action. Earlier work by us has shown AAPP is able to inhibit blood aggregation without prolonging bleeding time, which holds out the promise of safer treatments for thrombotic disease (14). Other types of anti-platelet medication such as cyclooxygenase inhibitors have a risk of leading to excessive bleeding (15). Among present anticoagulants, warfarin in particular is known to be problematic, having a narrow dose window, complications due to interactions with food-derived molecules, and a very different pharmacological profile in different patients (16).
A molecular model of AAPP is highly desirable in order to understand the ability of the protein to block collagen and platelets from interacting. In particular it would be helpful for efforts to design small molecule drugs with the same mode of action as AAPP. Despite the ability to express and purify AAPP in significant amounts, crystallization has proved extremely challenging. Here we describe the use of an 8H7 Fab monoclonal antibody (mAb) fragment to stabilize AAPP sufficiently to allow well ordered crystals to be grown. The structure of the complex is described.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Purification of AAPP-Cloning and purification were essentially carried out as for the AAPP protein described previously (7,14,17). The gene sequence from A. stephensi mosquitoes was cloned into pET22 with a hexahistidine tag and tobacco etch virus cleavage site at the C terminus. The resulting expression plasmid was transformed into Escherichia coli BL21(DE3) strain, and cells were cultured at 15°C overnight after induction with 0.5 mM isopropyl 1-thio-␤-Dgalactopyranoside. The AAPP was purified by chromatography using nickel-nitrilotriacetic acid-agarose (Qiagen) followed by Q Sepharose (GE Healthcare). The histidine tag was removed by tobacco etch virus protease digestion after nickel-nitrilotriacetic acid chromatography, and the purified complex was then concentrated to 10 mg/ml by Centricon YM-3 (Millipore) for crystallization.
Production of Anti-AAPP Antibodies-A DNA fragment encoding AAPP exon 3-4 was excised from pET32-AAPP ex3-4 (17) by digested with NcoI and XhoI and cloned into NcoI/XhoI sites of the E. coli expression vector pET22-GEX6P2. The resulting expression plasmid, pET22-GEX6P2-AAPP ex3-4 , was transformed into E. coli BL21(DE3) strain, and cells were cultured at 37°C for 2 h after induction with 1 mM isopropyl 1-thio-␤-D-galactopyranoside. The AAPP ex3-4 was purified by chromatography using glutathione-Sepharose 4B (GE Healthcare). The GST tag was removed by PreScission Protease (GE Healthcare) digestion after GST chromatography. After immunization of BALB/c mice with the AAPP ex3-4 , the spleen cells were fused with P3X63Ag8.U1 myeloma cells (American Type Culture Collection, Manassas, VA) using an established procedure (19). Hybridoma lines were screened by enzyme-linked immunosorbent assay (ELISA) using the AAPP ex3-4 . Moreover, the ELISA-positive hybridoma lines were rescreened to obtain inhibitory monoclonal antibodies for AAPP-collagen interaction by AAPP binding assay described previously (7). Briefly, the AAPP ex3-4 was preincubated with each mAb, and the mixture was added to 96-well collagen-coated microtiter plates (Nunc, Rochester, NY). Binding of the AAPP ex3-4 to collagen was detected using the ExpressDetector nickel-HRP (KPL, Gaithersburg, MD), which can bind to the His tag at the C terminus of the AAPP ex3-4 . One of the inhibitory monoclonal antibodies, 8H7, was maintained in RPMI 1640 supplemented with 10% fetal calf serum. The 8H7 mAb was purified from ascites fluid using Protein G affinity column (GE Healthcare).
Preparation of 8H7 IgG and Fab-The 8H7 IgG mAb was purified using the Protein G affinity column (GE Healthcare) from the supernatant of cultured hybridoma cells expressing the murine mAb 8H7 IgG. After the filtration of the supernatant, the sample was loaded onto the column equilibrated with 20 mM potassium phosphate (pH 7.0) buffer. The mAb fraction was eluted with 100 mM glycine (pH 2.7). The eluate was neutralized immediately after elution 1 M Tris-HCl (pH 9.0) and dialyzed overnight against 20 mM potassium phosphate (pH 7.0). The 8H7 Fab fragment was prepared through limited digestion with immobilized papain (Thermo Scientific). The reaction was carried out in 20 mM potassium phosphate (pH 7.0) and 20 mM L-cysteine. 10 mg of IgG was added per 0.5 ml of immobilized papain and incubated for 6 h at 37°C. After the reaction the sample was separated with immobilized papain using the spin column by centrifugation at 4000 rpm for 15 min and dialyzed overnight against 20 mM Tris-HCl (pH 8.0). The

TABLE 1 Primers used in AAPP mutants
protein was then loaded onto Q-Sepharose (GE Healthcare) to remove undigested IgG and Fc. The 8H7 Fab was passed through Q-Sepharose (GE Healthcare) to remove minor proteins before loading onto Superdex 200 (GE Healthcare) equilibrated with the same buffer. Finally, 8H7 Fab was concentrated to 10 mg/ml by ultrafiltration using a Centriprep YM-30 (Millopore) for crystallization.
Cloning and Sequencing of the Variable Heavy and Light Chain Genes of 8H7 mAb-mRNA was extracted from 1 ϫ 10 7 hybridoma cell line 8H7 using the FastTrack 2.0 mRNA Isolation kit (Invitrogen), and first-strand cDNA was synthesized from the mRNA with reverse transcriptase using the Firststrand cDNA Synthesis kit (Novagen). The cDNA was used as a template for PCR amplification of the variable heavy and light chain genes of 8H7 mAb using Taq 2000 DNA polymerase (Stratagene) and the primer sets in the Mouse Ig-Prime kit (Novagen). The PCR fragments were cloned into pCR2.1 (Invitrogen), the variable heavy and light chain genes were sequenced, and the nucleotide sequence data have been deposited in GenBank TM database under the accession numbers AB903029 and AB903030.
Crystallization and Structure Determination-The AAPP-8H7 Fab complex was crystallized by vapor diffusion using the sitting drop method. Protein (20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 5 mM CaCl 2 ) and reservoir solution (0.1 M HEPES (pH 7.0), 15% PEG 20,000) were mixed in a 1:1 ratio then equilibrated against 1 ml of reservoir solution at 20°C. Crystals grew in space-group P2 1 2 1 2 1 , with a ϭ 93.8 Å, b ϭ 99.4 Å, c ϭ 166.0 Å and contained two molecules in the asymmetric unit. Diffraction data were collected at Ϫ180°C using crystal flash-frozen in crystallization buffer containing 18% (v/v) glycerol. Diffraction data were collected at 1.0 Å on beam line BL17A stations at the Photon Factory, Tsukuba, Japan using an ADSC Quantum 315 CCD detector. All data were processed and scaled using HKL2000 (20). The AAPP-8H7 Fab complex structure was solved by molecular replacement using Phaser (21) and the previously reported Fab structure (22) as a starting model. Two solutions of 8H7-Fab dimer were obtained with final TFZ (translation function Z-score) of 19.0. The model of 8H7-Fab was subjected to rigid-body refinement using PHENIX (23) giving an R-factor of 0.35. The resultant ͉2F o Ϫ F c ͉ and ͉F o Ϫ F c ͉ electron density maps clearly showed two AAPP molecules. The electron density was interpreted and traced using COOT (24), and the model was refined with PHENIX (23). Solvent molecules were placed at positions where spherical electron density peaks were found above 1.3 in the ͉2F o Ϫ F c ͉ map and above 3.0 in the ͉F o Ϫ F c ͉ map and where stereochemically reasonable hydrogen bonds could form. Structural evaluation of the final models of the AAPP-8H7 Fab complex using MolProbity (25) indicated that 98.9% of the residues are in the most favorable regions of the Ramachandran plot. A summary of the data collection and refinement statistics is given in Table  2. Figures were prepared with PyMOL (26) and LIGPLOT (27). Atomic coordinates and structure factors of the complex have been deposited in the Protein Data Bank under accession code 4OKV.
Collagen Binding Assay-The collagen binding assay was carried out as described previously (7,17). In brief, soluble collagen type-I was immobilized in 96-well enzyme immunoassay plates followed by blocking with blocking buffer (PBS containing 1% BSA). AAPP was serially diluted and incubated for 1 h with indicated doses of the 8H7 Fab and whole IgG. The proteins were incubated for 1 h at room temperature on the plates, and binding of AAPP to collagen was detected using anti-His Abs conjugated with HRP (Qiagen).
Pulldown Assay-GST (5.7 M), GST-fused AAPP ex3-4 (4.1 M), C3 (9.6 M), and 4A (4.6 M) were incubated with the 8H7 Fab (5.7 M) in a total volume of 50 l of PBS at room temperature for 30 min. The volume of PBS was raised to 500 l, and 30 l of glutathione Sepharose resin (GE Healthcare) was applied to the mixture and then incubated at room temperature for 1 h with rotating. Glutathione resin carrying the proteins was pelleted and washed 3 times with PBS. Proteins were eluted by boiling in 25 l of buffer with 2% 2-mercaptoethanol and loaded onto 12% SDS-PAGE gels.
Mass Spectrometry-Full-length and selenomethionine AAPP samples were analyzed by MALDI-TOF mass spectrometry (Bruker Daltonics). 2 l of the protein solution (30 M) was mixed with 0.5 l of the matrix solution (10 mg/ml 3,5-dimethoxy-4-hydroxycinnamic acid [sinapinic acid] in a 0.1% trifluoroacetic acid, 70% acetonitrile aqueous solution) on the MALDI sample target. The mixture was allowed to dry at room temperature before analysis. Each MALDI-TOF mass spectrum was acquired by 200 -250 laser shots. FlexAnalysis Version 2.0 (Bruker Daltonics) was used for data processing.

RESULTS AND DISCUSSION
Molecular Characteristics of AAPP-The cloning and expression of AAPP in E. coli with an N-terminal histidine tag has been described. Full-length AAPP can be readily expressed in a soluble form to a level of ϳ3 mg/liter of culture by overnight expression at 15°C. A simple procedure to remove the affinity tag followed by gel filtration yields samples that appear pure by gel electrophoresis. After purification the protein mass was measured by MALDI-TOF spectrometry. The experimental mass of 27302.7 Da agrees well with the predicted mass of 27292.1 Da (Fig. 1A). AAPP has a low complexity region of ϳ90 residues, rich in glycine and glutamic acid, from residue 60 to 150. Thousands of crystallization trials with freshly purified protein proved fruitless. A mAb was, therefore, raised against full-length AAPP, and the AAPP-8H7 Fab complex was subjected to crystallization trials. Thin, needle-shaped crystals  were obtained that diffracted to 1.8 Å resolution. Phases were obtained by molecular replacement of the previously reported Fab structure (22) as a starting model. This allowed a model of AAPP to be built from Tyr-202 to Glu-269, but no further ordered residues appear in the structure. To highlight other parts of the structure in the electron density map, a number of mutants were made changing individual leucine residues to methionine in order to place selenium atoms at selected points in the structure. Overall four leucine residues (162, 181, 184, 251) were mutated, and the selenomethionine protein was purified. In each case the mutant crystallized in space group P2 1 2 1 2 1 with similar cell parameters to the native crystal (Table 2). Diffraction data from these crystals, however, failed to reveal any selenium atoms outside the core region previously identified, indicating these residues are either absent or simply highly disordered.
After storage of the selenomethionine mutant samples at 4°C for 1 month, a repeat MALDI-TOF experiment showed complete loss of the full-length protein (Fig. 1B). The major peak was found to have a mass of 8128.7 Da, suggesting significant degradation. Testing samples at intermediate stages after storage of a few weeks yielded a greater range of sharply-defined peaks. Comparing the masses of these fragments revealed a succession of cleavage events removing single amino acid residues (Table 3). Reference to the sequence of AAPP showed the polypeptide fragments present in the sample (Fig. 1B). The low complexity region is found roughly 50 residues to the N terminus of the stable domain.
Overall Structure of AAPP-8H7 Fab Complex-There are two independent copies of the AAPP-8H7 Fab complex in the model ( Fig. 2A) that overlay closely (root mean square deviation ϭ 0.678). Residues from Tyr-202 to Arg-266 of both copies of AAPP are visible in the electron density map, and residue 201 is also modeled in one copy. The ordered residues form two equi-length helices, one severely kinked near one end, connected by a turn that brings the helices into close contact nearly anti-parallel to each other ( Fig. 2A). The two copies of AAPP in the asymmetric unit make a major contact through a classic knobs-in-holes interaction, as one helix pair lies against another. Although the Matthews coefficient (V M ) is 2.32 A 3 Da Ϫ1 , suggesting a solvent content of ϳ50%, large open cavities appear in the crystal lattice due to the highly helical AAPP holding apart the mAb molecules. The AAPP domain is stabilized by two disulfide bonds, one near each end of the helices (Fig. 2B). There are also salt-bridge interactions across the helices between Asp-215 and Arg-254 and between Glu-207 and Lys-265 ( Fig. 2A). Such a two-dimensional structure has little hydrophobic core in any meaningful sense, but there are also hydrophobic contacts such as between residues Leu-219, Leu-247, and Leu-251 (Fig. 2B). Without the disulfide bridges the structure shows little evidence of being stable. The turn region consists of residues Asp-233, Lys-234, Asn-235, Asn-236, and Pro-237 (Fig. 2C). Both asparagine residues form hydrogen bonds to main-chain atoms through their side chains. Asn-236 also bonds to the imidazole of His 238 (Fig. 3A). These interactions suggest the turn region is generally stable, which is supported by the temperature factors of atoms in these residues. The average temperature factors of all atoms in the two AAPP subunits are 29.7 and 32.7 Å 2 and are 16.1 and 18.6 Å 2 in the turn regions.
The 8H7 Fab structure shows the classical IgG domain structure of anti-parallel ␤-sheet sandwiches with an antigen binding pocket formed from loops on both the heavy and light chains (Fig. 2B). It interacts solely with the turn region of AAPP so that the helical region points away from the antibody (Fig. 2C  and Fig. 3A). Only AAPP residues from Glu-228 to Cys-239 make contact with the 8H7 Fab, but these turn residues make very close contact, including several salt bridges and hydrogen bonds (Fig. 3A). The SS bond formed between Cys-230 and Cys-239 also comes within van der Waals distance of Tyr-32 of the heavy chain (Fig. 3A). As well as the charge-charge interac-tions, such as between Asp-233 of AAPP and Arg 51 of the light chain, there are substantial apolar interactions such as Trp-33 of the heavy chain lying against the peptide bond formed by Asn-235 and Asn-236 (Fig. 3B). The surface area of AAPP buried by the 8H7 Fab is roughly 800 Å 2 , shared roughly equally between the heavy and light chains. This small interfacial area suggests that rigidity of the AAPP binding site contributes strongly to tight, specific binding (Fig. 3C).
Inhibition of Collagen Binding of AAPP by the 8H7 Fab-AAPP inhibits platelet aggregation via direct binding to collagen (7,14). Experiments were carried out to determine whether the 8H7 Fab can block this interaction. To this end, AAPP was preincubated with the 8H7 Fab or whole IgG, and the binding ability of AAPP to immobilized soluble collagen type I was assessed by a plate assay. Free AAPP effectively bound to soluble collagen in a concentration-dependent manner, whereas both 8H7 Fab and whole IgG significantly inhibited the interaction in a dose-dependent manner (Fig. 4, A and B). The EC 50 of collagen binding of free AAPP was 4.6 nM (95% CI, 4.3-5.0 nM). The 8H7 Fab at 1.0 M significantly reduced the effective   (Fig. 4C). The nonspecific mouse IgG did not significantly inhibit the interaction (Fig. 4D).
Involvement of a Stable Structure of AAPP with Collagen Binding-The aapp gene of A. stephensi includes four exons separated by three introns (17). The region of AAPP involved in collagen binding has been tested previously by expression of recombinant truncated genes carrying different combinations of the four exons making up the coding sequence (17). Exons 3 and 4, which encode residues 148 -204 and 205-269, respectively (Fig. 1B), were found to be absolutely required for collagen binding, and the full-length protein binds soluble collagen with similar affinity to AAPP ex3-4 (17). To locate the site of collagen interaction more accurately, several mutants of GSTfused AAPP ex3-4 were produced including a multiple mutant with Lys 232, Asp-233, Lys-234, and Asn-235 changed to alanines (termed "4A"), Cys-239 to alanine (termed "C3"), and both Cys-239 and Cys 261 to alanines (termed "C3/C4") (Fig.  5A). The pulldown assay demonstrated that the 8H7 Fab bound to wild-type AAPP ex3-4 but not to 4A or C3 (Fig. 5B), further confirming that the 8H7 Fab binds to the turn region between helices. The loss of interaction if one cysteine residue is mutated to alanine suggests that the tertiary structure of the loop region is also required for tight binding of AAPP to the 8H7 Fab (Fig. 5B).
The 4A mutant, missing key residues making close contact with the 8H7 Fab, was found to bind to collagen with very similar affinity to wild-type AAPP (Fig. 6A). The binding of C3 was markedly reduced (EC 50 ϭ 164.1 nM; 95% CI, 118.6 -227.1 nM) compared with the wild type (EC 50 ϭ 5.9 nM; 95% CI, 5.7-6.2 nM), and C3/C4 did not bind to collagen detectibly at 100 nM (Fig. 6A). The loss of collagen binding on mutating the cysteine residues seems likely to be due to destabilization of the protein fold. This is supported by the fact that the GST fusion peptide of AAPP 225-244 (corresponding to the helix-turn-helix region) does not bind collagen (Fig. 6B), indicating a stable structure is apparently required for this interaction.
Conclusions-Nearly all biological processes involve some sort of interaction between one protein and another, and protein-protein interactions have recently come strongly into focus as a target for drug design. Although it was previously believed that protein-protein interactions would prove extremely difficult to block with small molecule inhibitors, FIGURE 5. GST pulldown assay using AAPP ex3-4 mutants. A, schematic diagram illustrating the AAPP ex3-4 mutants. The several mutants of GST-fused AAPP ex3-4 were produced, including a multiple mutant with Lys-232, Asp-233, Lys-234, and Asn-235 changed to alanines (termed 4A), Cys-239 to alanine (termed C3), and Cys-261 to alanine (termed C4). B, GST, GST-AAPP ex3-4 (GST-ex3-4; wild type), GST-AAPP ex3-4 4A (GST-4A), and GST-AAPP ex3-4 C3 (GST-C3) were incubated with or without anti-AAPP 8H7 Fab before purification with glutathione resin (beads). The pelleted resin was washed 3 times, and then proteins were eluted and loaded in 12% SDS-PAGE. recent trends suggest that in fact this strategy for drug design can prove a very fruitful one (28). For example the binding of the von Willebrand factor to the glycoprotein Ib␣ receptor on blood platelets is a protein-protein interaction of major importance in the control of blood clotting. von Willebrand factor is a multimeric glycoprotein whose A3 domain binds to subendothelial collagen (29) and whose A1 domain binds to glycoprotein Ib␣ (30). The clinical importance of the latter interaction is underlined by multiple efforts to produce effective inhibitors. The nanobody ALX-0081 and the aptamer ARC1779 have been described that are active in vitro (31)(32)(33)(34) but unsuitable for oral administration as they are peptides. This has prompted work using fragment-based drug design and the discovery of several lead molecules. The work described here takes a different approach of using natural inhibitor proteins and mimicking their action through structure-based design of small molecule drugs. To this end we have solved the structure of the C-terminal region of AAPP, an effective anti-coagulant protein that functions by masking subendothelial collagen at sites of vascular injury, the principal trigger in platelet aggregation. AAPP prevents collagen types I and III from interacting with the platelet receptors and so prevents release of internal calcium, the second messenger that activates platelets (7). The mass spectrometry results show full-length AAPP to be highly unstable to proteolysis, even in highly purified form, but reveal a structural domain at the C terminus, which corresponds to the active region of the protein. This more stable region, although small and consisting essentially of only two ␣-helices, is tightly bound by a cognate antibody, which proved essential in obtaining crystals suitable for x-ray analysis.