As shown in Table 1, the SVPLA₂s from Viperidae snake CBc and MtxII interact with FXa with very high affinity and constitute group VS (<K_d^app> 0.6 – 2 nM). These enzymes strongly inhibit the formation of the prothrombinase complex (IC₅₀ 1–3 nM). Under identical conditions, the <K_d ^app> value determined for hsPLA₂ is 14 nM. CBa₂, AtxA and CbII have high affinity values and anticoagulant potency, and form group S (20–50 nM). A third group of SVPLA₂, group M, is formed by Vbb_, VRV-PLVIII, and bAhp. In this group, the affinity with FXa is smaller (<K_d^app> 400 – 830 nM) and the IC₅₀ ranges from 90 to 130 nM. AGTX (the neutral PLA₂ from A. halys pallas (blomhoffii)), reported previously as weakly anticoagulant in the presence of PL38, does not interact with FXa and does not inhibit prothrombinase activity in the absence of PL. Neither CbI nor Catx inhibit prothrombinase activity in the absence of PL. These three SVPLA₂s constitute group NB. Last, reduced and carboxymethylated CBc does not interact with FXa. Furthermore, we found a positive linear correlation between <K_d ^app> and IC₅₀ (R = 0.995) for the three groups of SVPLA₂ that interact with FXa. Thus, the strongly anticoagulant SVPLA₂s bind with high affinity to FXa, whereas the less efficient anticoagulant SVPLA₂ possess low affinity for FXa. Also, basic SVPLA₂s (8.35<pI<9.10) bound to FXa and inhibited prothrombinase activity, whereas those with acidic pIs (4.60–5.43) did not (Table 1).

It is interesting to note that the isoenzymes CBc and CBa₂, which differ by 8 amino acids (His1Ser, Ile18Val, Arg34Gln, Pro74Arg, Glu92Lys, Tyr115Asn, Gly116Glu, Gly128Glu) and associate with the acidic subunit CA to form two pharmacologically distinct classes of crotoxin complexes, present differences in toxicity, enzymatic activities and stability6768. These two isoforms bind to FXa with different kinetics (Table 1). The average rate of dissociation constant <k_off> for the FXa-CBa₂ complex is about two orders of magnitude greater than that of the FXa-CBc complex, implying a more stable FXa-CBc complex (<K_d ^app> 0.6 nM). Consequently, CBc strongly inhibits the prothrombinase complex (IC₅₀ 0.7 nM), whereas the inhibition by CBa₂ is much weaker (IC₅₀ 41 nM). Indeed, we had observed in the past that the difference in stability between crotoxin isoforms was due only to the CB subunit 5568.

We also investigated by SPR the possibility of the formation of a ternary CA-CB-FXa complex. On one hand, CA binds to immobilized CB 69, whereas FXa does not interact with immobilized CB23. An anti-CA monoclonal antibody (mAb) A-73.13 70 was covalently attached to the chip, as previously described 71; CTX was then captured via this mAb before the injection of FXa (Fig. 1). FXa bound to CTX, as seen in the rise of the resonance signal. Fig. 1 also shows that after the injection of a specific anti-CB mAb (B32.13), the signal increased further, indicating the presence of CB on the chip and showing that the CB-FXa complex is stable and remained attached to CA.

In order to identify those residues or sequence patterns that differ between members of the SVPLA₂s, we obtained Weblogo plots of the four groups VS (CBc, MtxII), S (CbII, AtxA, CBa₂), M (VRV-PLVIII, bAhp, Vbb) and NB (AGTX, Catx, CbI). Thereafter, we performed comparisons of all the sequences among themselves and of the four groups against each other.

Considering only segments of three or more residues, we observed three conserved regions for AGTX, Catx and CbI of the NB group (Tyr25-Gly30, Thr41-Gly53 and Cys96-Asp99; Fig. 2A), and six for the CBc-MtxII pair of the VS group (Tyr25-Cys27, Gly33-Gly35, Pro37-Cys45, Cys50-Tyr52, Cys96-Asp99 and Cys105-Arg107; Fig. 2B). A striking difference between the two groups is the presence of the acidic residue Glu at position 128 in the C-terminal region of all SVPLA₂s of the NB group, as opposed to a Lys or a Gly residue in the VS group. For the other groups, only CBa₂ and Vbb contain a Glu at this position. With respect to the S group, the M group contains a conserved basic residue at position 115. Interestingly, the M group contains a conserved Lys in position 132 that is absent in the VS and NB groups. Group NB contains acidic side chains in the first strand of the β-wing and the adjacent β-turn (residues 75–82). For the VS and S groups, the tendency is towards a net balance of positive charges in the β-wing.

The NB group Catx presents three Glu and one Asp residue in the β-wing (residues 74–90); CbI presents two Asp residues. The β-wing of CBc of group VS contains two Lys residues; that of MtxII two Lys, one Asp and one Glu residues. For the S and M groups the tendency is towards net positive charges, except for bAhp, for which the total charge is neutral.

We obtained 3D homology models of CBc, CBa₂, CbI, CbII, AtxA and Vbb. The models show the canonical structural features of the PLA₂ template molecules: a N-terminal α-helix (helix A), a short helix (helix B), a Ca²⁺-binding loop, a long α-helix (C), a loop preceding an anti-parallel two-stranded sheet (β-wing), a long α-helix (D), anti-parallel to helix C, and a C-terminal extended fragment. All seven disulfide-bridges are at their expected positions. Fig. 3 shows the molecular model of AtxA in the canonical ("front") face orientation, i.e., presenting the catalytic hydrophobic channel.

Fig. 4A shows the MEP calculated at the molecular surface of the modeled SVPLA₂s in the canonical view. Fig. 4B shows the MEP on the "back" face. Qualitatively, we observe that the MEP is positive in the front face of strong and mild anticoagulants (CBc, CBa₂, CbII, and AtxA); approximately neutral for the weak anticoagulant Vbb, and negative for CbI (Fig. 4A). The back face is largely devoid of positive potential, except for Vbb (Fig. 4B). This difference in MEP leads to an electrostatic asymmetry. We observe the same trend for the crystallized SVPLA₂ (MtxII, VRV-PLVIII, bAhp, AGTX, Catx and hsPLA₂; not shown).

The output generated by the docking program PatchDock showed many complexes in contact with one of the binding site regions. Only the AtxA-FXa candidate complex (rank number 7) showed the best compatibility with the binding site derived from the mutagenesis data39 (in this complex, position 150, which shows a Glu -> Arg sequence difference between the purchased FXa and the crystal structure (PDB entry 2BOH) is neither at the light/heavy chains interface nor at the AtxA/FXa interface). Several residues belonging to the two regions identified by the mutagenesis of AtxA and defining part of the binding site (the "front" strand of the β-wing and the C-terminal fragment) are at the interface in this complex. In it, the relative orientation of FXa positions the N-terminus of the EGF-like 2 domain of the light chain of FXa towards the β-wing of AtxA, allowing the 1-domain (not visible in the X-ray structure) to enter in interaction with the β-wing and thus contact the remaining residues of the front edge of the β-wing. Fig. 5 shows the ribbon representation of the final complex between AtxA and the light and heavy chains of FXa. Fig. 6 shows the same complex in the form of a solvent-accessible surface. In both representations, the SVPLA₂ is in the "classical" orientation, i.e. that of Fig. 4A. Overall, we observe just small conformational changes after complex formation.

For the FXa strong-binding SVPLA₂s CBc, MtxII, CbII and CBa₂, the complexes that ranked 18, 10, 20 and 17, respectively, showed the same binding mode as the AtxA-FXa complex, i.e., the same relative orientation of FXa with respect to the SVPLA₂. As measured by the FIT function of Pymol, these complexes showed Cα RMSD's with respect to the reference AtxA-FXa complex of 3.4, 2.2, 5.1 and 4.8 Å, respectively.

The percentage of surface residues at the interface of the complexes is of 24–36% for the five SVPLA₂s and of 20–32% for FXa. The interface area for the five SVPLA₂s is in the 860–1700 Ų range and 915–1470 Ų for FXa. The interface area of the complexes varies from 1610 to 1930 Ų. The values obtained for the solvent accessible area buried in the five characterized complexes fall between the small and the large interfaces categories in a study of 362 protein-protein interfaces 72. Fig. 7 shows the mapping of the SVPLA₂-FXa interaction surface of the complexes between the five SVPLA₂s and FXa. The shadowed residues in that figure are those at the interface of the AtxA-FXa complex satisfying the mutagenesis data, and at the interface of the complexes of the other four SVPLA₂s that show the same complexation mode as the AtxA-FXa complex. The consensus residues that participate to FXa binding are: solvent-exposed parts of helix A (positions 2, 3, 7) and helix B (positions 18, 19); positions 16, 23 and 24; a part of the Ca²⁺ loop (positions 31–34); a part of the 69-loop (between helix C and the β-wing; positions 53, 59, 60, 69 and 70); and the C-terminal segment (positions 118, 119, 121–124, 129–131, 133). By taking into account the type of atom-atom contacts as defined by Sobolev et al.73, 36.5% of the contacts are hydrophobic at the AtxA-FXa light chain interface, as compared to 26.4% at the AtxA-FXa heavy chain interface. Overall, 29.7% of the contacts are hydrophobic at the AtxA-FXa interface. These interface regions are identified separately by the protein-protein interface prediction server PPI-Pred. For example, for AtxA, the highest score PPI-Pred patch, which represents the most probable protein-protein binding site, includes helices A and B, the Ca²⁺-loop, part of the 69-loop, two residues from the front strand of the β-wing, and the four C-terminal residues (results not shown). Fig. 8 shows the regions mapped by docking simulations in the modeled structure of complexed AtxA.

The first line of Fig. 9A and 9B show the amino acid sequences of the parts of the heavy and light chains of the crystal structure of FXa (PDB 2BOH), detected at the surface respectively. The heavy chain (Fig. 9A) includes the serine proteinase catalytic domain of FXa (catalytic-site residues in bold), and the light chain (Fig. 9B) includes the EGF-like 2 domain. The following lines in Fig. 9A and 9B show the FXa residues at the interface of the docked FXa-SVPLA₂ complexes (residues appearing three or more times). The regions of FXa at the interface with CBc, MtxII, CbII, AtxA and CBa₂ in the resulting docked complexes include, in the light chain (Fig. 9B), the N-terminal region of the EGF-like 2 domain: Arg86-Ser90 and Cys100-Glu102, Gln104, Asn105. In the heavy chain (Fig. 9A), they include the following regions. Region I: Arg93; Phe101 (from the 99-loop). Region II: Arg125; Asp126; Glu129-Ser130; Thr134. Region III: Tyr162; Asp164-Asn166; Lys169; Leu170. Region IV: Gln178 and Asn179 (from the 174-loop). Region V: Lys230, Thr232, Ala233, Phe234 and Lys236 (all residues from the C-terminal helix). Fig. 10 shows these regions in the 3D model of complexed FXa. Some of these residues belong to loops surrounding the catalytic-site cleft.

We show in Fig. 11A and 11B the intermolecular residue contacts for the AtxA-FXa interface for which the contact area is equal to or greater than 10 Ų. We observe from this map that the heavy chain of FXa interacts with the following regions of the PLA₂: a portion of the Ca²⁺ loop, the helix C-β-wing loop, and the C-terminal segment (including Lys127, which has an effect in the binding to FXa) (Fig. 11A). The light chain of FXa establishes intermolecular contacts with the remaining regions: helices A and B, and the β-wing (Arg77) of AtxA (Fig. 11B). From the CSU analysis of interfaces, it follows that the contacts between SVPLA₂ and FXa are of diverse nature -aromatic-aromatic, hydrophobic-hydrophobic, salt bridges and H-bonds. There are also destabilizing hydrophobic-hydrophilic interactions. Few of the contacts take place between same-nature residues, like Lys127 from AtxA and Arg93 from FXa's heavy chain (Fig. 11A). These contacts are between non-polar atoms or between non-polar and polar atoms of side chains at the surface of the molecules, where the presence of the aqueous solvent can buffer the electrostatic interactions.

The classification of the anticoagulant potency (strong, weak, non-anticoagulant) of a SVPLA₂ depends on the anticoagulant assay used and on whether this assay leads to establishing the mechanism of action. Usually, the recalcification time assay, one of the simplest assays, is used. This method is very sensitive to the lipid levels in the plasma but does not show whether the mechanism is PL-independent or not. Our results deal with the interaction of anticoagulant PLA₂s with FXa through the non-catalytic, PL-independent mechanism of action and are to be compared to studies performed under the same conditions. As mentioned before, only three PLA₂s -CM-IV of N. nigricolis, AtxA of V. ammodytes ammodytes, and hsPLA₂- have been studied under these conditions. In this work, we complete these studies by testing several other SVPLA₂ from Viperidae venom. Our results clearly show that CBc and MtxII interact with FXa with very high affinity and strongly inhibit the formation of the prothrombinase complex. CbII, AtxA and CBa₂ possess good affinity for FXa and good anticoagulant potency. VRV-PLVIII, bAhp and Vbb possess weak affinity for FXa and show weaker inhibition of prothrombinase activity. Some of these enzymes (CB, AtxA) also inhibit prothrombinase activity in the presence of PL (not shown). AGTX, the CA subunit of CTX, Catx and CbI do not interact with FXa and do not inhibit prothrombinase activity in the absence of PL. Therefore, AGTX, described previously as a weakly anticoagulant PLA₂ 38, and VRV-PLVIII, described previously as strongly anticoagulant 74, appear to inhibit blood coagulation through different mechanisms.

hsPLA₂ contains an unusually large number of prominent cationic patches on its molecular surface, some of which lie on the putative IBS 14, in contrast to bovine pancreatic PLA₂ and the SVPLA₂ from Agkistrodon p. piscivorus, which display only a limited number of such patches 10. Given the charged nature of the residues critical for binding hsPLA₂ and AtxA to FXa, it is clear that electrostatic interactions play a role in the binding. Indeed, the electrostatic asymmetry showed by the MEP calculations must be enhanced by the presence of the essential Ca²⁺ cofactor ion 77 and may be at the origin of the increased affinity of hsPLA₂ for FXa in the presence of Ca²⁺ 50. Thus, long distance electrostatic forces operating at the molecular surface are important and may optimally orient the molecules before binding to FXa. However, electrostatic interactions might not exclusively drive binding to FXa by the Viperidae SVPLA₂s, as in hsPLA₂. Indeed, in addition to basic residues, hydrophobic and aromatic residues play also key roles in optimizing the interaction between SVPLA₂s and FXa, given that a ring of hydrophobic residues surrounds the opening to the catalytic site cavity. Thus, for AtxA, many hydrophobic residues (Leu2, 3, 10, 18 and 58; Pro17, 59, 68 and 121; Val31), most of which are located in the N-terminal region, and several aromatic residues (Phe24 and 123; Tyr52) are part of the surface presented to FXa (Fig. 7). Lastly, even though a Phe24Ala mutation in AtxA was not found to change the FXa-binding kinetic parameters 39, we emphasize the need to probe other residues belonging to the hydrophobic ring.

On our results, the catalytic site of FXa is free and not involved in the interface, in agreement with the conservation of the serine-proteinase catalytic activity after binding by SVPLA₂ 51. Our AtxA-FXa complex shows that one region of the light chain of FXa is involved in the binding to the SVPLA₂ and that the N-terminus of the EGF-like 2-domain is pointing south (Fig. 5). Therefore, the EGF-like 1-domain, not visible in the crystal structure of FXa, can interact with the front edge of the β-wing of the PLA₂. The Gla domain of FXa (bound to the N-terminus of the EGF-like 1-domain), is well beyond reach in space and not in contact with the PLA₂. These two features are in agreement with experimental and biochemical data which support the conclusion that the Gla domain is needed rather for insertion into the PL membrane 51, and with a model of the entire FXa. Indeed, based on the crystal structure of the Gla domain of bovine prothrombin and the NMR coordinates of the bovine FX EGF1 domain, Bajaj and coworkers 2 proposed a model structure for the entire FXa molecule, based upon the crystal structure of porcine FIXa. The EGF domains in the model are oriented south and thus capable of establishing contacts with the β-wing of the PLA₂.

Based on the ability of synthetic peptides from FXa to inhibit FXa-induced clotting, a number of authors have reported the FVa binding sites on FXa. The last line of Fig. 9A shows the experimentally reported residues of the heavy chain of FXa incontrovertibly involved in the binding to FVa. These residues are: FXa(His83-Lys96) 78; FXa(Arg165, Lys169) 798081; the 185–189 loop, i.e., FXa(Lys186) 82; FXa(Val231-Thr244) 83; and FXa(Arg240) 81. Three of four of the segments of the catalytic domain of FXa identified to interact with FVa overlap with those identified by us to interact with the Viperidae SVPLA₂s, namely the segment about Lys90, the 162-loop, and the C-terminal part, about Lys237 (Fig. 9A). This suggests that several residues are shared by both, the SVPLA₂-FXa and the FVa-FXa binding interfaces. Moreover, the experimentally identified FXa heavy chain residues involved in binding to FVa are located on the same 3D face of FXa, as in our SVPLA₂-FXa complexes.

Arni and coworkers have recently reported the crystal structures of human Gla domainless FXa complexed with two small anticoagulant proteins from a hematophagous nematode 8485. The determined exosite from those complexes involves residues from one of the strands of the N-terminal seven-stranded β-barrel (strand β6, residues 80–93) and from the short C-terminal α-helix (residues 233–243) of the catalytic subunit of FXa. Several of these residues fall in regions I and V mapped in our complexes with the SVPLA₂s (Fig. 9A and 10).

By combining the residues defining the IBS 1314 with the site-directed mutagenesis experiments for probing the basic residues of hsPLA₂ involved in binding to FXa51, we deduce that IBS residues Arg7 and Lys10 bind to FXa.

By homology with hsPLA₂, the presumed IBS amino acid residues for AtxA are Leu3, Leu18, Phe24, Lys74 and Tyr113. On one hand, Lys74, experimentally found to be critical for binding of AtxA to FXa 39, belongs to the IBS. On the other hand, our simulations indicate that IBS residues Leu3, Leu18 and Phe24 are in contact with FXa. In conclusion, several IBS residues are part of the PL-independent anticoagulant site and participate in formation of the complex with FXa.

The Viperidae SVPLA₂s studied here are multifunctional proteins, raising the possibility of overlapping or multiple pharmacological sites distinct from the catalytic site86. One team52 has suggested that the neurotoxic site of group II neurotoxic enzymes overlaps with the anticoagulant region. Thus, from the mutants used to test the anticoagulant potency 39 and the neurotoxic function 87 of AtxA, it appears that several residues in the C-terminus are clearly shared by both functions. The overlapping of pharmacological sites is most easily understood in terms of the small size of this family of proteins.

We determined the anticoagulant potency of the SVPLA₂s by measuring in vitro the inhibition of prothrombinase activity (IC₅₀). We used an in vitro biological test in which the prothrombinase complex was reconstituted at 37°C from purified human factors FVa and FXa in the presence of Ca²⁺, but without the addition of PL39. The purified prothrombinase components (FVa 10 nM, FXa 10 nM, and different concentrations of SVPLA₂: 0, 10, 20, 50, 100, 200, 500 nM) were incubated 5 min at 37°C in Tris-buffered saline (0.05 M Tris/HCl, 0.1 M NaCl, 0.5% BSA, 5 mM CaCl₂, pH 7.4). The reaction was then started with 110 nM prothrombin. We measured activated prothrombin activity every 20 min, as described previously 50, with small modifications. We determined the IC₅₀ value, which corresponds to 50% inhibition of thrombin generation for the different SVPLA₂s in the absence of PL.

We analyzed of the interaction between the SVPLA₂s from the Viperidae family, and FXa by SPR using a BIACORE^® 2000 system (Biacore AB, Uppsala, Sweden). The running and dilution buffer in all experiments was Hepes (HBS; 10 mM Hepes, 150 mM NaCl, 5 mM CaCl₂, 0.005% surfactant P20, pH 7.4). The experiments were conducted at 37°C. Human FXa was covalently coupled via primary amino groups on a CM5 sensor chip surface according to Prijatelj et al.39. One independent flow cell of the same sensor chip was used as a control flow cell and was subjected to a "blank immobilization," i.e., with no FXa added. We found the SPR signal for immobilized FXa on three different flow cells to be 1500 RU, 3000 RU and 5840 RU (1 RU corresponds to 1 pg/mm² of immobilized protein). We injected PLA₂ samples (0, 0.25, 0.5, 1, 2, 4, and 8 μg/ml) at 37°C with a flow rate of 20 μl/min on independent runs on the control and assay flow cells and their binding was monitored. Between each injection, we regenerated surfaces with twice 5 μL of 1 M NaCl. The apparent equilibrium constant, <K_d ^app> = <k_off>/<k_on>, the average dissociation rate constant <k_off> and the average association rate constant <k_on> were calculated using Biacore's BIAevaluation 3 software. The kinetic models used to fit the data included the Langmuir association, heterogeneous analyte and conformational change. Only the first one showed the lowest closeness-of-fit value (χ²). In addition, the other models resulted in affinities much lower (μM) than expected (nM).

The sequence databases and identifiers are the following: CBc (also known as CB1), Uniprot P62022; CBa₂ (also known as CB2), P24027; AGTX, P14421; AtxA, P00626; MtxII, P24605; Vbb, P31854; CbI, gi 1345182 57; CbII, gi 1345181 57; CA, isoform CA₂ 56(obtained by post-translational modification of ProCA, P08878).

For sequence alignments, we used the LALIGN program 9192, which finds multiple matching sub-segments in two sequences and shows the local sequence alignments. For representing sequences and their alignments, we used Weblogo, a web-based application designed to generate sequence logos 9394.

Given the high sequence homology between the SVPLA₂s and diverse PLA₂s whose crystal structures are available, we applied homology modeling to generate 3D model structures of the SVPLA₂s using the Biopolymer and Homology modules of the insightII software package (Accelrys Inc. San Diego, CA, USA). We retrieved the PLA₂s of the template proteins of known 3D structure from the Protein Data Bank95. The template proteins are: AGTX, the neutral PLA₂ from Agkistrodon halys pallas96 (PDB 1A2A); VRV-PLVIII, the PLA₂ from Daboia russelli pulchella97 (PDB 1FB2); MtxII, myotoxin II, a K49-PLA₂ from Bothrops asper. 98 (PDB 1CLP); the PLA₂ from Crotalus atrox99 (PDB 1PP2); the PLA₂ from Vipera russelli russelli52 (PDB 1VIP) and bAhp, the basic PLA₂ from Agkistrodon halys pallas 100 (PDB 1JIA).

After building the structurally conserved regions (SCR) and the structurally variable regions (SVR), we replaced the side chains of the template protein by the target protein's side chains in a predetermined library conformation. We then performed a rotamer search assignment in order to avoid atomic steric clashes and optimize the inter-residue energies. We included a structurally conserved catalytic water molecule during the modeling. We did not model the Ca²⁺ ion at the Ca²⁺ binding site, since it is absent in several X-ray structures. We subjected the obtained models to an overall internal energy minimization using the CFF91 force field. The protonation state of ionizable side chains and of the N- and C-termini was set for pH 7. Atomic partial charges were those of the CFF91 field. We used a distance-dependent dielectric function of 4r. We applied the cell multipole summation method for van der Waals and coulomb interactions. We used none of the cross terms of the force field. We applied the same conditions to the amino acid residues in the crystal structures used. We checked the stereo chemical quality of the models with the Struct_Check program, and the correctness of the folding with the Profiles3D Verify functionality (self-compatibility scores, insightII). The amino acid residue numbering is based on that of Renetseder et al.15.

The MEP is generated by the combined presence of all partial charges residing on the atoms as a function of their positions. The potential was calculated with the DelPhi 2.5 program in insightII. The grid resolution was of 0.833 Å/grid point. The interior and exterior dielectric constants were 2 and 80, respectively. The value of the ionic strength was 0.145; the probe radius was 1.4 Å and the ionic radius 2.00 Å. The treatment of the grid points at the boundary used a full Coulomb approximation. The values of the potential are given in kT/e units at T = 298 K.

We generated the molecular complexes with the PatchDock rigid body molecular docking procedure101. Given two molecules, the surfaces of the molecules to dock are divided by PatchDock into patches according to the surface shape. These patches correspond to patterns that visually distinguish between puzzle pieces. We used a clustering RMSD criterion of 4.0 Å. The output lists the rank of the complex and its approximate interface area. We selected PatchDock for the docking simulations since it allows a priori focusing on the vicinity of potential binding sites. In other words, it is possible to upload a receptor-binding site and a ligand-binding site. We took the SVPLA₂s as the "receptor" molecules and FXa as the "ligand" molecule.

Thus, we generated molecular complexes only for those SVPLA₂s for which we find experimental evidence of biophysical interaction, i.e., in which the binding affinity between the PLA₂ and FXa, as measured by SPR, was high. In addition, we used the available mutagenesis data for AtxA39 to filter the docked complexes so that AtxA residues that show a decrease in the binding affinity for FXa are at the interface of the complex and define a binding site. The residues defining the binding site on AtxA are Arg72, Lys74, His76 and Arg77 (front edge of the β-wing), and Arg118, Lys127, Lys128 and Lys132 (C-terminal fragment)39. We defined no ligand-binding site for FXa.

From the candidate complexes generated by PatchDock for AtxA, only one showed a binding mode compatible with the ensemble of the mutagenesis data -many complexes showed binding to either the β-wing or the C-terminal fragment of the SVPLA₂ or other regions of the AtxA. Thereafter, we selected those complexes for the other SVPLA₂s that showed the same binding mode as AtxA and whose Cα RMSDs with respect to the AtxA-FXa complex were minimal. After generation of the complexes, we used the "move apart" option of PatchDock, which separates (by 1.6 Å) the receptor and ligand subunits in order to eliminate steric hindrances at the interface. We further improved the fitting of the complex by applying firstly the SCWRL3 side chain modeling procedure 102, in which we froze all disulfide bonds, as well as the side chains of heavy chain residues Asp70 and Glu80, which chelate the Ca²⁺ ion in FXa. After the rotamer search, we applied additional energy minimizations in order to reach a minimal internal energy conformation. The entire approach assumes that no drastic conformational changes occur during complexation.

On another hand, we submitted the SVPLA₂ models to the Protein-Protein Interface Prediction (PPI-Pred) server 26103 to predict their binding sites. PPI-Pred predicts protein-protein binding sites using a combination of surface patch analysis and a support vector machine trained on 180 proteins involved in both obligate and non-obligate interactions.

The interface between the two polypeptide chains of each of the complexes was characterized with the Protein interfaces, surfaces and assemblies service PISA 104105. The interface contacts were obtained through a contact map analysis and characterized with the SPACE bioinformatics tools CMA (Contact Map Analysis) and CSU (Contacts of Structural Units) 106107. We show only interface contacts for which the contact area is equal to or greater than 10 Ų.