NMR structure determination of Ixolaris and factor X(a) interaction reveals a noncanonical mechanism of Kunitz inhibition

Hemostasis is initiated on vessel wall damage, which exposes tissue factor (TF) that activates blood coagulation and collagen that induces platelet aggregation.¹  Mechanistically, TF binds FVIIa with formation of catalytically active TF/FVIIa complex, which converts FIX to FIXa, or FX to FXa.^2,3  As a result of limited proteolysis reactions that occur at the cell membrane in the presence of cofactors, the coagulation cascade is markedly amplified, leading to intense thrombin generation and clot formation. The TF/FVIIa activity is under control of TF pathway inhibitor (TFPI), a potent physiological anticoagulant comprising 3 tandem Kunitz-type inhibitor domains designated K1-K3. TFPI has a peculiar mechanism of action in which the K2 interacts with the active site of FXa, which works as a scaffold for inhibition of the FVIIa/TF complex. The K1 assists in complex formation through binding to FVIIa catalytic site.^2,3 

Several studies indicate that the interaction of TFPI with FXa involves a number of other protein-protein interactions, in addition to the active site. The affinity of the separate K2 domain for FXa is about 3 orders of magnitude lower than that of TFPI.^4,5  In addition, cleavage of TFPI between K1 and K2 dramatically reduces the affinity of TFPI for FXa.⁶  Studies employing synthetic peptides or monoclonal antibodies have demonstrated that, in addition to K2, a number of residues in the K1 domain mediate the interaction of the TFPI with FXa.^7,8  Thus, it is proposed that K1 provides a high-affinity binding region for FXa that allows subsequent inhibition of the coagulation enzyme by the K2 domain. These results highlight the complexity of the reactions involving TFPI and emphasize the critical role of scaffold FXa in this process. Therefore, understanding the underlying molecular mechanisms of TFPI and related inhibitors may provide the rationale for designing anticoagulants targeting TF in a number of diseases, without major impairment of hemostasis.^9-11 

The TF pathway plays important roles in normal hemostasis, cardiovascular disease, and thrombosis. Coagulation activation by TF is implicated in cancer progression, cancer-associated thrombosis, and metastasis, although the molecular mechanism is not completely understood.^12,13  Overexpression of TF enhances tumor growth, whereas knockdown attenuates tumor expansion.^14-16 

In this respect, exogenous secretion such as saliva from blood-sucking arthropods contains an immense repertoire of molecules affecting vascular biology.^17,18  Ixolaris is a tick salivary anticoagulant containing 2 Kunitz-like domains that binds FXa or FX, which serve as scaffolds for inhibition of TF/FVIIa complex.¹⁹  In contrast to TFPI, Ixolaris does not interact with the active site cleft of FXa.¹⁹  The canonical mechanism of Kunitz inhibition, as observed for TFPI, resembles the formation of an enzyme-substrate complex and no interaction with the zymogen. In contrast, Ixolaris interacts with similar high affinity to FX and FXa^20,21  through (pro)exosites,²²  as scaffolds to inhibit TF-FVIIa complex (K_d, 0.67 and 0.25 nM, respectively).¹⁹ 

Additional studies revealed that Ixolaris prevents PAR2 cleavage specifically by human TF signaling complexes, indicating that it may also attenuate inflammatory tonus associated with TF expression.²³  Furthermore, in vivo experiments showed that Ixolaris prevents thrombosis in rats with a long half-life,²⁴  and blocks angiogenesis and tumor progression in a glioblastoma model.²⁵  More recently, treatment of chronically infected monkeys with simian immunodeficiency virus with Ixolaris was accompanied by inhibition of monocyte TF activity and a significant decrease in d-dimer and immune activation.²⁶  Altogether, these results suggest that Ixolaris represents a novel therapeutic option to attenuate coagulation and inflammation associated with thrombosis, cancer, and chronically infected patients with HIV.

In the present study, we describe the interaction of Ixolaris with FX. We show that Ixolaris employs both K1 and K2 domains to recognize the heparin-binding (pro)exosite (HBE) on FX. We determined the solution NMR structure and characterized the structural dynamics of the K1 and K2 domains of Ixolaris. Also, NMR data were used to construct a 3-dimensional structural model of the Ixolaris-FXa complex, revealing the noncanonical mechanism for FX binding, which occurs predominantly through the K2 domain. Taken together, our data provide new insight into the structural requirements for Ixolaris interaction with FX, which may contribute to our understanding of TF/VIIa inhibition.

¹³C/¹⁵N and ¹⁵N-labeled Ixolaris samples were overexpressed in Escherichia coli strain Rosetta-gami B (DE3) as a TRX-His₆ fusion protein and purified as described.²⁷  U-[¹⁵N,²H]-labeled Ixolaris was overexpressed M9 medium in ²H₂O containing 1 g/L ¹⁵NH₄Cl and supplemented with Celtone Base Powder [U-¹⁵N,²H] 1 g/L (Cambridge Isotopes Laboratories). The final sample conditions for NMR spectroscopy were 0.1 to 0.2 mM Ixolaris, 20 mM sodium phosphate buffer (pH 6.0), 0.02% NaN₃, 90/10% (vol/vol) H₂O/D₂O.

Human FX (58.8 kDa) was purchased from Enzyme research laboratories (catalog HFX 1010) as aliquots at concentration of 1.14 mg/mL in 20 mM Tris⋅HCl, 100 mM NaCl, 1 mM Benzamidine at pH 7.4. The FX was iteratively concentrated and buffer exchanged, using a Vivaspin Protein Concentrator centrifugal device (10 kDa molecular weight cutoff; GE). FX final sample concentration was 0.6 mM in 20 mM sodium phosphate buffer (pH 6.0), 0.02% NaN₃.

Codon optimized DNA of isolated Kunitz domains were synthesized and cloned into the expression vector pET32a, as described.²⁷  Both Kunitz domains K1 and K2 constructs were expressed in the E coli system according to the established protocol described for full-length Ixolaris. The purification of K1 and K2 domains included the combination of nickel affinity and reverse-phase chromatographies. Nonlabeled samples were dissolved in 20 mM sodium phosphate buffer at pH 6.0, 0.02% NaN₃.

All NMR experiments were recorded at 308 K, using Bruker DRX 600 equipped with a ¹H,¹⁵N,¹³C TXI cryoprobe or Bruker Avance III 800 spectrometers. A set of 3D triple-resonance spectra was collected for sequential backbone and side-chain resonance assignments (De Paula et al²⁸ ; BMRB: 27051). For obtaining nuclear Overhauser effect (NOE) connectivities, 3D [¹H-¹³C-¹H] and [¹H-¹⁵N-¹H] nuclear Overhauser spectroscopy (NOESY) spectra used for distance restrains were acquired with 120 ms mixing time. All spectra were processed using NMRpipe (Delaglio et al²⁹ ) and analyzed with CARA 1.8.4.

For assessing the backbone dynamics, relaxation experiments were recorded on 0.18 mM ¹⁵N-labeled Ixolaris at a field of 600 MHz. ¹⁵N R₁ and R₂ relaxation rates were measured from spectra with different relaxation delays: 20, 50, 100 (duplicate), 200, 250, 500, 750, and 1000 ms for R₁ and 48, 80 (duplicate), 112, 144, 176, 208, 240, 272, and 304 ms for R₂. The errors in the peak intensities were calculated from the duplicate experiments. The ¹H-¹⁵N heteronuclear NOEs were determined from the ratio of peak intensities with and without the saturation of the amide protons. Errors in the heteronuclear NOE values were calculated from the peak intensities and noise levels in the reference and saturated spectra. Relaxation data curve fitting was performed using CcpNmr analysis.³⁰ 

The CS-Rosetta/RASREC (resolution-adapted structural recombination) approach was used to generate the NMR structure of the Ixolaris. Briefly, assigned backbone chemical shifts (Cα, Cβ, N, and H) of the residues E13-S130 were used to predict backbone torsion angles (TALOS) and to generate a backbone fragment library representing the conformations of the protein.³¹  The N- and C-terminal residues were excluded, as they were identified as flexible according to the chemical shift-derived order parameter and heteronuclear NOE values. The Rosetta energy function was used to assemble and iteratively refine the structure.^31,32  Unambiguous assigned long-range NOEs (i, i > 4) were used to bias the RASREC CS-Rosetta search. The RASREC protocol was used to generate a large pool of initial seed structures, which are then iteratively recombined, refined, and scored.³³  The top 10 lowest-energy structures showing good Rosetta scores and agreement with the NOE and chemical shift data were selected as the final ensemble. The quality of structures was further checked by the Molprobity program.³⁴  Structure calculations were performed at the University of California, Santa Cruz Baker cluster with the kind assistance of David Flores and Santrupti Nerli.

For NMR binding experiments, FX (0.6 mM) was added progressively to a ¹⁵N Ixolaris sample (0.1 mM) in identical buffer conditions (20 mM sodium phosphate buffer at pH 6.0, 0.02% NaN₃). For titration of FX into ¹⁵N Ixolaris, the following FX:Ixolaris molar ratios were used: 0.2:1, 0.5:1, and 1:1, and a series of ¹H-¹⁵N HSQC spectra was collected. The pH at each step of the titration was kept constant. The chemical shift perturbations (CSPs) from the 2D ¹⁵N-¹H HSQC NMR experiments were used to identify the FX binding site, using the following equation: Δδ^NH = ([ΔH]2 + [Δ¹⁵N/10]²)^1/2, where ΔH is the difference between free and bound ¹H amide chemical shifts (ppm) and Δ¹⁵N is the difference between free and bound ¹⁵N chemical shifts (ppm). The change in peak intensity was determined by calculating the ratio of I_bound/I_free for each assigned resonance and normalized to the dilution factor after FX addition. Chemical shift perturbations and peak intensities were calculated using CcpNmr Analysis.³⁰ 

Molecular docking of the Ixolaris to human FXa (Protein Data Bank [PDB] accession code 1hcg) was performed using HADDOCK 2.1^35-37  to generate an interaction model for this protein complex. The calculations were performed using NMR chemical shift mapping as ambiguous interaction restraints to define the residues potentially involved in binding. The active residues for Ixolaris were defined as the residues that had chemical shift changes above the mean plus 2 standard deviations (W15, G54, E55, K73, Y94, D95, N97, E119), and the passive residues the ones having resonances intensity significantly attenuated (V16, S17, C18, L19, Y36, G41, H57, F58, N65, E66, S67, C68, W92, T96, F104, T105, N115, K123, E124, T125, K127) in the NMR titrations between Ixolaris and FX (Figure 3).

Active residues for FXa included R93, K96, R125, R165, K169, K236, and R240, and were defined from previous mutagenesis experiments.²¹  Passive residues were automatically defined by HADDOCK. Complex structures were sorted according to the intermolecular interaction energy (the sum of intermolecular van der Waals and electrostatic energies and restraint energies). The calculated models were clustered using a 7.5-Å interface root mean square deviation cutoff. HADDOCK clustered 146 of 200 calculated structures into 12 clusters. The top HADDOCK cluster with the lowest energy structure within the highest-score NMR-restrained were considered to be the most reliable Ixolaris-FXa complex.

Ixolaris is a monomeric protein in solution, and exhibits a well-resolved ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) NMR spectrum that is typically observed in compact folded proteins (supplemental Figure 1, available on the Blood Web site). The ¹H, ¹³C, and ¹⁵N resonance assignments and secondary structure information of Ixolaris based on ¹³C chemical shifts were determined previously (De Paula et al²⁸ ). Although they were stable and give good spectra, Ixolaris structure calculations using NMR data were a challenge and required a different strategy that applied NOE and chemical shift information.

Our first attempt was to calculate the structure with conventional methods, using dihedral angles, disulfide bonds, and all NOEs from the 3-dimensional [¹H-¹³C-¹H] and [¹H-¹⁵N-¹H] NOESY spectra. We used semiautomated NOE cross-peak assignments, using the UNIO’10 module^38-40  and the simulated annealing protocol with torsion angle molecular dynamics with CYANA version 2.1.⁴¹  However, we obtained nonconvergent structures that showed very large backbone root mean square deviation (RMSD) values resulting from the small number of unambiguous NOEs in the 3-dimensional [¹H-¹³C-¹H] NOESY spectra. Toward obtaining a more converged structural model of Ixolaris from our sparse NOE data (38 unambiguously assigned long-range NOEs), we turned to the RASREC CS-Rosetta. In the Rosetta methodology, a limited set of NMR data are sufficient to accurately determine solution structures of proteins up to 40 kDa.^32,33,42,43  We calculated 100 Rosetta structures for the core region of Ixolaris (residues 13-130) and selected the 10 lowest-energy structures for further analysis (Figure 1). The family of the 10 lowest-energy structures of Ixolaris is well defined, with an average backbone pairwise RMSD of 1.39 Å (Figure 1A) and corresponding RMSDs for the K1 and K2 domains in isolation of 0.50 and 0.40 Å, respectively (supplemental Figure 3). The final full-atom models have good packing and structural statistics, as assessed using MOLPROBITY, which satisfy all experimental NOE restraints (Table 1).

As previously predicted from sequence homology analysis, the structure of Ixolaris confirms the presence of 2 consecutive Kunitz domains. Each of the 2 domains of Ixolaris displays the characteristic fold of the Kunitz family of protease inhibitors, which is composed of a central antiparallel β sheet (residues 32-36, 43-45, 91-95, and 104-107) with 1 short α-helix at the C-terminal end (residues 61-68 and 119-125). In addition, a short α-helix (residues 13-18) is formed at the N-terminal of the K1 domain. These elements form the topological scaffold that is stabilized by the 5 canonical disulfide bonds involving Cys18-Cys68, Cys26-Cys51, Cys43-Cys64, Cys76-Cys126, and Cys101-Cys122. Analysis of the ¹³Cα and ¹³Cβ chemical shifts of the 10 cysteines confirms their oxidized state. Additional stabilization derives from a remarkably hydrophobic cluster formation in the K2 domain, formed by WYY. A triple aromatic residue cluster is highly conserved in Kunitz inhibitors (supplemental Figure 3). Ixolaris is an acidic protein (pI ∼4.6), and the surface electrostatic potential indicates that negative charges are densely distributed in the K1 and K2 domains (Figure 1C).

As seen in other Kunitz domain proteins, only ∼53% of the structure is engaged in secondary structure elements, which are connected by long loops (residues 46-60, 69-90, 108-118; Figure 1). In addition, the loop region spanning residues 71 to 75 is the peptide linker between the first and second Kunitz domains of Ixolaris. This linker contains 3 prolines and 1 residue (K73) mobile on the picosecond to nanosecond timescale, as evidenced by the reduced ¹⁵N-¹H NOE and R₂ values (Figure 2A). The most similar structure found by structural alignment software (PDB accession code efold) belongs to the Kunitz inhibitors, including Bikunin (PDB accession code 1BIK), APPI (Amyloid Precursor Protein Inhibitor; PDB accession code 1CA0), BPTI (Bovine Pancreatic Trypsin Inhibitor; PDB accession code 1MTN), and TFPI (Tissue Factor Pathway Inhibitor; PDB accession code 4BQD). All these structures displayed RMSD between 1.65 and 3.55 Å, with sequence identity ranging between 20% and 27%.

To gain insight into the behavior of Ixolaris in solution, we measured ¹⁵N relaxation parameters to probe dynamics on the fast (pico- to nanoseconds) and intermediate (micro- to milliseconds) timescale. We recorded R₁, R₂, and heteronuclear ¹⁵N-¹H NOE experiments at 600 MHz and 308 K (Figure 2). On the basis of the NMR relaxation data, Ixolaris backbone is well structured (with the exception of the N- and C-terminal residues), with an average tumbling rate of 6.6 ns, which is typical for the size of Ixolaris (∼16 kDa). The heteronuclear ¹⁵N-{¹H} NOE, which report specifically on backbone fluctuations on a pico- to nanosecond (ps-ns) timescale, show several residues of the K2 domain with significantly lower NOE values, indicating a higher degree of flexibility when compared with the K1 domain (Figure 2A).

In addition, our data indicate the presence of residues with motions in the intermediate timescale (higher R₂ values), indicative of conformational exchange in the K1; however, most notably, in the K2 domain. This suggests an evident distinction between the K1 and K2 dynamics. The R₂ values for several residues in the K2 domain (G83, Y94, D95, T96, T105, Y106, E119) were higher than the average, typical of internal motion on a micro- to millisecond timescale. It is important to mention that a few amide groups were not assigned (R3, F12, E45, Y82, V84, R86, I89, C101, T110, G111, N112, K113, and N114), which indicates local motion in the intermediate exchange regime. The majority of those residues are located at the K2 domain.

A summary of the Ixolaris dynamics is shown in Figure 2B, from which it is evident that the K2 domain is particularly dynamic, whereas the K1 domain keeps as a rigid platform supporting the conformational dynamic of the K2 domain.

The backbone assignment of free Ixolaris allowed us to probe conformational changes on binding to FX. The complex between Ixolaris and FX is 74.7 kDa, and is consequently difficult to be characterized by standard NMR methods. In agreement with previous studies,^21,22  Ixolaris formed a tight complex with FX under the NMR conditions. Detailed inspection of the [¹⁵N, ¹H] HSQC spectra of Ixolaris titrated with unlabeled FX at different ratios showed very little perturbation of chemical shifts, indicating that the binding process is intermediate to slow on the NMR timescale (Figure 3A). In addition, most amide resonances of Ixolaris exhibited an intensity reduction of ∼60% (low I_bound/I_free values), when the complex formation reached 50% from the stoichiometric titration (Figure 3C). This indicates that the line broadening is caused not only by the large complex formation but also by intermediate exchange from the interaction. Further addition (1:1 molar ratio) resulted in the significant line broadening, except for the flexible N- and C-terminal regions (supplemental Figure 4). As such, we evaluated the changes in peak positions and intensities in the presence of FX at a molar ratio of 1:0.5 (Ixolaris:FX). Analysis revealed 2 types of effects on the Ixolaris amide resonances: chemical shift perturbations (Figure 3B, pink bars), indicating a change in the local chemical environment, and conformational exchange-induced line broadening in the bound state (Figure 3C).

Mapping the FX-induced chemical shift perturbations onto the Ixolaris structure revealed a continuous and extensive interaction surface (Figure 3D). Relevant residues of K2 domain appear to contribute to the binding of FX (13 residues); however, only residues W92, Y94, F104, N115, and K123 are conserved in TFPI K2 domain (sequence alignment shown in supplemental Figure 3). In addition, the peak intensities of residues located in the H3-helix of K2 domain were also significantly attenuated in the bound form. Interestingly, the NMR data revealed effects on the K1 domain as well. These changes are highly localized, and cluster near the H2-helix (N65, E66, S67, C68, K73) at the interface with the K2 domain (Figure 3D). Notably, residues W15, V16, S17, C18, L19, G54, E55, H57, and F58, located at or near the H1-helix, also showed 2 standard deviation above average chemical shift perturbations and attenuated peak intensities, indicating that FX-binding-induced conformational changes are not restricted to the K2 domain. The perturbed residues on Ixolaris indicate that polar and hydrophobic interactions contribute for FX binding.

Remarkably, we found that the Ixolaris residues involved in FX association, specifically from the K2 domain, correlate with regions of increased dynamics (Figure 2), indicating a direct role for dynamics in driving the recognition of the inhibitors for the blood coagulation factor.

To derive a structural model for the Ixolaris/FXa complex binding site, we integrated the information that was obtained in the current work (Figure 4) with analysis obtained in a previous study showing that residues R93, K96, R125, R165, K169, K236, and R240 from FX(a) are critical for Ixolaris/FX(a) binding.²¹  To this end, the lowest energy conformation of Ixolaris and the crystal structure of human FXa were used as starting structures for the docking process. Ixolaris was docked to the solvent-exposed, hydrophilic FXa HBE,²⁰  using the software HADDOCK.^35-37  The lowest-energy structure obtained from the restrained docking of the Ixolaris-FXa complex (Figure 4A) was consistent with the NMR data, and the interface agrees with our experimental data. The Ixolaris-FXa interface buries a total of 1707 Ų of accessible surface area, with both domains contributing to a polar interaction to the hydrophilic pocket of FXa (Figure 4A; supplemental Figure 4). Ixolaris and FXa display pronounced electrostatic complementarity spanning the entire HBE, with Ixolaris presenting a negatively charged surface patch associating with the positively charged HBE of FXa (Figure 4B). This suggests that long-range electrostatic interactions may play an important role in attracting Ixolaris to FXa to establish the binary complex.

The observation that both K1 and K2 domains of the Ixolaris interact with human FX, we sought to characterize the relative contributions of the isolated domains toward the overall interaction. Using thermal stability shift assay, we measured the FX thermal stability on complex formation (Figure 4C; supplemental Figure 6). Ixolaris binding to FX results in a significant thermal stabilization of the protein: ΔT_m shift of 10°C. Interestingly, no stabilization was observed for both K1 and K2 single domains. In another sample, K1 and K2 single domains added simultaneously (up to 3:1 molar ratio) into FX also did not affect the T_m (supplemental Figure 6). Together, these results show that FX exhibits greater thermal stability on Ixolaris binding relative to association with single domains, and both K1 and K2 domains are necessary and must reside in the same polypeptide for binding to FX.

Additional insights into the possible noncanonical mechanism of Ixolaris are provided by the HADDOCK structure analysis. Close inspection of the specific interactions between Ixolaris and FXa involved salt bridges between residues E22, E55, K73, D95, and T96 of Ixolaris and residues R93, R240, E129, N166, and K169 of FXa, respectively (Figure 5A).

Most notably, E55 in the K1 domain makes a salt bridge with K236 and R240 of FXa (Figure 5B). Furthermore, structural superposition of FXa in its apo and Ixolaris-bound state shows that the R93 and residues H57 and D102 of the catalytic triad adopt different conformations (Figure 5C). The closest atom of Ixolaris is more than 15 Å from the active site Ser195 residue, indicating an allosteric switch in the catalytic site. No other significant changes in the FXa structure were observed, with the rest of the FXa main chain being very similar (RMSD of 0.46 Å for 210 aligned Cα atoms). Thus, the data support our hypothesis that the mechanism of allosteric modulation by Ixolaris is primarily driven by a change of the FXa 99-loop conformation (Figure 5C). The 99-loop modulates the protease activity by both hindering the substrate binding and keeping the proper arrangement of the catalytic triad. In addition, loop flexibility was observed in 2 trypsin-like serine proteases.^45,46 

In this study, we have employed solution NMR to directly evaluate the plasticity and conformational dynamics to propose a mechanistic blueprint for Ixolaris binding to scaffold FX, a step critical for inhibition of the TF/FVIIa complex. The basis of our mechanistic approach is the high-affinity Ixolaris-FXa encounter complex driven by long-range electrostatic attraction of Ixolaris to the FXa HBE, revealing a noncanonical mechanism, mainly conducted by the K2 domain. We therefore hypothesize that the negatively charged and dynamic region on K2 domain is the major binding site for the positively charged FXa HBE. In addition to the major contribution of the K2 domain, a secondary interface formed by the K1 domain is necessary for the FXa binding. More specifically, E22 of the K1 domain makes a salt bridge with R93 of the FXa, possibly revealing a conformational change at the 99-loop that could allosterically modulate the FXa catalytic site (Figure 5B). Notably, a previous study has demonstrated that R93 is critical for the recognition of the HBE by Ixolaris. In addition, the interaction of Ixolaris with R93-mutated FXa is by far the most severely impaired among all mutants tested.²¹  The presence of a secondary site in the complex possibly creates a loss of entropy in the binding that supports the interaction of the K2 domain and contributes to the stabilization of the complex. In addition, Ixolaris-FXa model highpoints a second key concerted structural event, in which the K1 domain could simultaneously interact with both FXa exosite and FVIIa/TF active site. Accordingly, prior in vitro studies have shown that Ixolaris inhibition of FVIIa catalytic activity is markedly reduced in T99 and Q143-mutated FVIIa.²³ 

To determine unique features of the Ixolaris in relation to other Kunitz domains proteins, we compared the structures of the K2 domain of Ixolaris, TFPI (1TFX), and Bikunin (1BIK) (supplemental Figure 7). The 3 domains showed high similarity with a backbone RMSD of 0.96 Å (Bikunin) and 1.2 Å (TFPI). The structural signature of the canonical Kunitz domains was observed in Ixolaris, despite the fact that 1 bridge is missing (supplemental Figure 3). The loops D81-I89 and I109-N115 without the canonical disulfide bridge differ only slightly from the arrangement in TFPI. In addition, we observed that the majority (56%) of the amide backbone chemical shifts for these loops were not assigned, which indicates local motions in the intermediate exchange regime.

These structural features in Ixolaris may affect the conformational mobility of its K2 domain, which merit further analysis in extended structural studies. To determine whether this observation is part of a unique mechanism of Ixolaris, it would be interesting to confirm the existence of similar conformational mobility in the related Kunitz structures by measuring their dynamics as well.

Ixolaris shares a similar anticoagulant mechanism with hematophagous nematode NAPc2 with respect to scaffold requirements for inhibition of TF/FVIIa complex. However, NAPc2 uses a different exosite in FXa to inhibit TF-FVIIa.⁴⁷  Structural analyses of the Ixolaris-bound FXa, NAPc2-bound FXa, and Apo FXa structures suggest that residues H57 and D102 of the catalytic triad exhibit conformational changes when bound to Ixolaris, whereas minor changes were observed for NAPc2-bound FXa (supplemental Figure 8). Given that NAPc2 blocks the active site of FVIIa, while locking in FXa in a signaling active conformation on the ternary TF-FVIIa-FXa complex, it seems reasonable to speculate that the small changes in the catalytic triad could render the FXa unable to be an active protease, perhaps explaining why Ixolaris blocks PAR2 cleavage and NAPc2 does not.²³ 

In contrast to a previous suggestion,⁴⁷  our NMR data showed that the C-terminal of Ixolaris is not involved in HBE binding. Slightly different results were obtained with NAP5 from Ancylostoma caninum. This inhibitor binds to both the catalytic site and the HBE of FXa.⁴⁸  Ixolaris is also distinct from the Kazal-type inhibitor Rhodniin, which employs its N- and C-terminus domains to interact simultaneously with the thrombin catalytic site and the anion-binding exosite, respectively.⁴⁹ 

Although Ixolaris shows a canonical Kunitz 3-dimensional structure, it exhibits a noncanonical mechanism of inhibition characterized by interaction outside the active site of FX, which operates as an efficient scaffold for formation of a tight quaternary inhibited FVIIa/TF/Ixolaris/FX(a) complex.¹⁹  This step is critical to understand how a binary Ixolaris/FX complex blocks FVIIa/TF. Accordingly, this study contributes to the development of specific inhibitors targeting nonconserved binding sites outside the active center. These domains are critical in the specificity and productive interaction of enzymes, cofactors, and zymogens, and are therefore potential targets for the inhibition of coagulation factors.

This work was supported by grants from Fundação Carlos Chagas Filho de Amparo a Pesquisa do Estado do Rio de Janeiro and Conselho Nacional de Desenvolvimento Científico e Tecnológico. The authors acknowledge the University of California, Santa Cruz Baker cluster for providing access to the computer resources. The authors also thank the facilities of National Center of Nuclear Magnetic Resonance (Federal University of Rio de Janeiro, Rio de Janeiro) for the NMR time.

Contribution: V.S.D.P., R.Q.M., and A.P.V. designed the research and wrote the manuscript. V.S.D.P. and A.P.V. performed and analyzed the NMR experiments and NMR data. V.S.D.P. generated constructs, prepared and purified isotopically labeled proteins; assigned NOEs; performed structure calculations with Cyana, RASREC CS-Rosetta, and HADDOCK; performed differential scanning fluorimetry; and analyzed and interpreted all data. V.S.D.P., I.M.B.F., R.Q.M., and A.P.V. provided reagents and analytical tools. N.G.S. and F.C.L.A. contributed with CS-Rosetta calculations. All authors discussed the results and contributed to writing the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Ana Paula Valente, Instituto de Bioquímica Médica, Centro Nacional de Ressonância Magnética Nuclear Jiri Jonas, Universidade Federal do Rio de Janeiro, Brazil; e-mail: valente.anap@gmail.com; and Viviane Silva de Paula, Department of Chemistry and Biochemistry, University of California, Santa Cruz, CA 95064; e-mail: vsilvade@ucsc.edu.