Molecular basis of factor IXa recognition by heparin-activated antithrombin revealed by a 1.7-Å structure of the ternary complex

Abstract

Factor (f) IXa is a critical enzyme for the formation of stable blood clots, and its deficiency results in hemophilia. The enzyme functions at the confluence of the intrinsic and extrinsic pathways by binding to fVIIIa and rapidly generating fXa. In spite of its importance, little is known about how fIXa recognizes its cofactor, its substrate, or its only known inhibitor, antithrombin (AT). However, it is clear that fIXa requires extensive exosite interactions to present substrates for efficient cleavage. Here we describe the 1.7-Å crystal structure of fIXa in its recognition (Michaelis) complex with heparin-activated AT. It represents the highest resolution structure of both proteins and allows us to address several outstanding issues. The structure reveals why the heparin-induced conformational change in AT is required to permit simultaneous active-site and exosite interactions with fIXa and the nature of these interactions. The reactive center loop of AT has evolved to specifically inhibit fIXa, with a P2 Gly so as not to clash with Tyr99 on fIXa, a P4 Ile to fit snugly into the S4 pocket, and a C-terminal extension to exploit a unique wall-like feature of the active-site cleft. Arg150 is at the center of the exosite interface, interacting with AT residues on β-sheet C. A surprising crystal contact is observed between the heparin pentasaccharide and fIXa, revealing a plausible mode of binding that would allow longer heparin chains to bridge the complex.

Blood coagulation (hemostasis) is traditionally described as two separate cascades of proteolytic activation events, the so-called intrinsic and extrinsic pathways (1, 2) (for a recent review, see ref. 3). The extrinsic pathway is initiated by tissue damage that exposes tissue factor (TF) and subendothelial matrix proteins to the blood. Platelets adhere to collagen, and circulating factor (f) VIIa binds to and is activated by TF. The fVIIa-TF complex (extrinsic Xase) activates fX, and fXa in the presence of fVa produces thrombin. Just enough thrombin is generated at this stage for the formation of an initial clot through platelet activation; however, unless much more thrombin is produced in short order, the clot will not persist and bleeding will result. The second stage of hemostasis utilizes components of the intrinsic pathway, factors VIIIa and IXa, and deficiency of these critical proteins is the cause of hemophilia A and B, respectively. The fIXa-fVIIIa complex is known as intrinsic Xase and forms on the surface of activated platelets to efficiently produce fXa, resulting in a burst of thrombin formation.

Regulatory proteins guard against overgrowth or dissemination of the clot, and all proteases generated in the cascade must eventually be inhibited. Factors VIIa and Xa are inhibited during the initiation phase by tissue factor pathway inhibitor (4), a triple Kunitz-domain canonical inhibitor, but the principal inhibitor of the coagulation proteases is the aptly named member of the serpin family of inhibitors, antithrombin (AT, previously known as ATIII) (5–7). AT circulates at a high concentration (2.3 μM) in a low activity state that is permissive of the initiation of hemostasis. This form is likely to be responsible for inhibition of thrombin and fXa in the luminal space, because rates of inhibition are on the order of 10³–10⁴ M^-1 s^-1. However, on the surface of the intact endothelium, AT interacts with heparan sulfate and its rate of protease inhibition is accelerated by ∼1000-fold. High-affinity binding of AT to heparan sulfate depends on the presence of a pentasaccharide sequence containing a critical 3O sulfate on a glucosamine residue (8). This sequence is commonly found in preparations of heparin and accounts for its anticoagulant properties. The effect of heparin on thrombin inhibition is entirely due to a template mechanism, where AT and thrombin bind the same heparin chain to accelerate diffusion and stabilize the encounter (Michaelis) complex to allow proteolysis to begin. In contrast, fIXa and fXa inhibition is significantly accelerated by the pentasaccharide alone due to conformational changes in AT that allow the formation of exosite contacts (so-called “allosteric activation”) (9). This conformational change has been described in detail elsewhere (10, 11), but functionally it amounts to the expulsion of the N-terminal portion (hinge region) of the reactive center loop (RCL) from β-sheet A to effectively increase its length. Factors IXa and Xa have similar heparin-binding sites to the well-characterized exosite II of thrombin (12, 13), and, therefore, long heparin chains can provide additional acceleration through the template effect. Crystal structures of the heparin-activated AT Michaelis complexes with thrombin and fXa have been solved (14, 15), illustrating why thrombin inhibition is not allosterically activated and why fXa recognition requires the pentasaccharide-induced conformational change in AT.

Factor IXa is an unusually poor protease, cleaving peptide substrates extremely slowly (< 100 M^-1 s^-1) in the absence of high concentrations of ethylene glycol (16), but the reasons behind this are only partly understood (17). It appears that fIXa has fully converted from the zymogen state upon cleavage activation, in contrast to fVIIa (18), because binding to its cofactor fVIIIa increases the rate of cleavage of its natural substrate fX by one million times but has a negligible effect on peptide substrates. It thus appears that fIXa is stably folded into a low activity state, and in the absence of exosite interactions rates of hydrolysis cannot reach physiologically relevant levels. Consistent with this, AT inhibition of fIXa is entirely dependent on exosite contacts (19). Defining these exosite interactions is therefore critical to understanding the requirements for fIXa activity and for determining how AT, the sole physiological inhibitor of fIXa, recognizes this important target.

Results and Discussion

Overall Structure.

As indicated by the data processing and refinement statistics given in Table 1, the structure of the AT-(S195A)fIXa-pentasaccharide complex is of high quality and resolution. Previous structures of pentasaccharide-bound AT were at resolutions between 2.6 and 3.3 Å (11, 15, 20, 21), and the three deposited structures of fIXa [porcine (22, 23) and human (24)] were also of low resolution (3.0, 2.9, and 2.8 Å). We are thus able to address (in later sections) certain issues that depend on high resolution, such as metal coordination and ligand binding, but can also discuss in detail the interactions between AT and fIXa that confer pentasaccharide-dependent specificity. A single complex was found in the asymmetric unit (Fig. 1), with AT binding the high-affinity pentasaccharide in the normal fashion and fIXa interacting with the RCL of AT as if poised to cleave the P1-P1′ bond [nomenclature of Schechter and Berger (25)]. The position of fIXa relative to AT is similar to that seen in the previous structure of the Michaelis complex between activated AT and fXa (15), with the engagement of similar exosites on AT and fIXa serving to stabilize the complex. The common exosite explains the absolute dependence on RCL expulsion and extension for pentasaccharide activation of fIXa and fXa inhibition by AT (19), because it would not be possible to simultaneously engage the exosite and active site without the extra length afforded by hinge region expulsion. Although the active-site and exosite contacts are similar for fIXa and fXa, fIXa is rotated by ∼40° relative to AT with the exosite contact with Arg150 (discussed later) as the pivot point (Fig. S1A). The heavy chain of fIXa is essentially identical to what was observed in previous structures and appears not to change conformation in response to its interaction with AT (rmsd = 0.65 Å for 223 C^α atoms). The second EGF domain is highly flexible in the structure with average B factors ∼2-fold higher than for the catalytic domain. The conformation of AT is also very similar to that of the original pentasaccharide-activated structure (1E03), with a C^α rmsd of 1.80 Å for 392 C^α atoms and 0.75 Å when the N terminus and the RCL are excluded (residues 45–378 and 401–431 compared). One segment of this region that does differ significantly is the AT exosite (strands 3 and 4 of sheet C), which is seen to shift by up to 2 Å (Fig. S1B), in apparent accommodation of fIXa.

Table 1.

Data processing, refinement, and model (3KCG)

Crystal
Space group	P212121
Cell dimensions (Å)	a = 78.78, b = 88.44, c = 147.23
(°)	α = β = γ = 90
Solvent content (%)	60.8
Data processing statistics
Wavelength (Å)	0.98 (Diamond, beam line I02)
Resolution (Å)	58.83–1.70	1.79–1.70
Total reflections	588,145	59,108
Unique reflections	112,588	16,392
Multiplicity	5.2	3.6
〈I/σ(I)〉	13.0	2.2
Completeness (%)	99.3	100
Rmerge	0.069	0.580
Model
Number of atoms modeled:
Protein	5,492
Idraparinux	100
Water	667
Carbohydrate	87
MPD	16
Calcium ions	1
Average B factor (Å2)	33.3
Refinement statistics	38.05–1.70 Å	1.81–1.70 Å
Reflections in working/free set	106,876/5670	17,733/951
R factor/R-free (%)	20.8/23.0	31.6/34.3
rms deviation from ideality
bonds (Å)/angles (°)	0.005/1.3

Fig. 1.

Stereo view of the AT-pentasaccharide-fIXa ternary complex. The Michaelis complex is shown in the standard serpin orientation, with the protease (catalytic domain in orange and EGF2 in purple) attacking from the top. The majority of AT is colored gray, with the heparin-binding helix D in cyan, β-sheet A in red, and the RCL in yellow. The P1 side chain (Arg393) and the pentasaccharide (green) are represented by sticks, and disulfide bonds are also shown.

Tyr99 and the Active-Site Cleft.

The active-site cleft of fIXa has some unusual features that partly explain its poor catalytic activity. Of particular relevance is the position of the side chain of residue Tyr99 that protrudes into the S2 pocket in the so-called “relaxed” state, precluding the binding of any P2 residue except Gly (24). In order to bind its natural substrate (fX), which contains a Thr at the P2 position, the Tyr side chain must swing out of the way to adopt the “strained” conformation. The movement of Tyr99 required to form the S2 pocket effectively blocks the S4 pocket, so full engagement of a P4 side chain requires a Gly at the P2 position (Fig. S2). Cleavage of fX relies on the binding of fVIIIa, which is thought to help populate the stressed state to allow binding of the activation peptide containing a P2 Thr (24). In the absence of cofactor, AT must bind to the relaxed conformation and accordingly has a Gly at the P2 position. As shown in Fig. 2A and B, the N-terminal portion of the RCL fits neatly into the active-site cleft, exploiting the only available features, the S1 and S4 subsites. The S4 pocket is formed by the side chains of Tyr99, Phe174, and Trp215 and snugly fits the P4 Ile side chain (Fig. 2B). The S4 pocket extends farther and may be able to accommodate larger hydrophobic side chains such as Trp, but this interaction would necessitate some change in the main chain conformation. This pocket has been successfully targeted in drug design to obtain specific inhibitors of fIXa (23).

Fig. 2.

Stereo views of the interactions between the RCL of AT and the active-site cleft of fIXa. (A) The catalytic domain of fIXa is shown in the standard orientation as a green ribbon and a semitransparent surface colored according to electrostatic potential (red for negative and blue for positive). The RCL of AT from P6-P6′ is shown as sticks (yellow from P6-P3′ and cyan for the “knuckle” region P4′-P6′), with surrounding electron density contoured at 1σ (blue). (B) A close-up of the N-terminal interactions is shown, with P6-P1 residues represented by sticks and semitransparent spheres (fIXa represented as above). The side chain of Ile390 at the P4 position is snugly accommodated in the S4 pocket formed by Tyr99, Phe174, and Trp215 (side chains shown as sticks). (C) A close-up of the P′-side interactions is shown, with P4′-P6′ residues colored cyan with semitransparent spheres. Factor IXa residues that interact with this knuckle region are shown as green sticks, and some intra- and intermolecular contacts are shown as broken red lines. Of particular importance is the interaction between P5′ Asn with Glu60 and Lys36 of fIXa.

The P′ Wall and Knuckle.

The fIXa-AT structure is the first showing fIXa bound to a substrate (serpins are suicide-substrate inhibitors) and is therefore the first to show any interactions on the P′ side. For fIXa, the inherent trade-off between formation of the S2 or the S4 pocket potentially allows the P′ side to play an important role in determining substrate specificity. In the deposited structures of porcine and human fIXa, the 36 and 60 loops form a wall on the S′ side (Fig. S2). These short, rigid loops protrude out of the active-site cleft and are capped by a salt bridge between Lys36 and Glu60. This wall would predictably force a 90° exit from the active-site cleft at P4′ but also provides a feature that could potentially extend favorable interactions C-terminal to P4′. AT has a unique three-residue insert of Pro-Asn-Arg from P4′ to P6′ that forms a tight β-turn in native AT (26), and here it is seen making extensive contacts with the P′ wall of fIXa (Fig. 2C). The Pro397 residue makes the required 90° turn to initiate an exit from the active-site cleft and is stabilized by a stacking interaction with the side chain of Phe41. The side chain of Asn398 interposes between the previously salt-bridged Lys36 and Glu60, forming compensatory hydrogen bonds. The last insertion residue, Arg399, lies between the body of AT and fIXa and mediates several hydrophobic and water-mediated contacts. Interestingly, the 395–399 residues in other structures of AT (native, etc.) form the same highly hydrogen-bonded β-turn conformation (resembling a knuckle), indicating that the conformation of this region is stable and preformed in AT and merely slots up against the P′ wall.

Exosite Contacts.

For serpin-protease Michaelis complexes, exosites have been defined in various ways, but due to the extensive P′ side interactions we have chosen to exclude everything involving the RCL from this distinction. Thus, contacts involving residues P6–P6′ are considered RCL contacts (1,623-Ų buried surface), and those with any other part of AT are exosite contacts (921-Ų buried surface). The exosite interactions observed in this structure for both AT and fIXa are consistent with previous mutagenesis data (27–29) and are detailed in Tables S1 and S2. The exosite of AT is the contiguous loop 232–255, which forms strands 3 and 4 of β-sheet C and contacts the autolysis loop residues 143–153 and Glu74 in fIXa (Fig. 3). The principal exosite interaction involves Arg150 in the center of the fIXa contact site. This residue is effectively buried in the interface, accounting for about 48% of the total exosite surface area on fIXa. The exosite contacts observed previously with fXa are very similar to those seen here, in spite of the 40° rotation of the proteases relative to AT. This similarity is presumably due to the fact that the pivot point is the Arg150 residue itself. The rotation appears to be caused by the differences in the S′ side of the active-site cleft (Fig. S3).

Fig. 3.

Exosite contacts between AT and fIXa. (A) The surface of fIXa is oriented and colored as before, with the AT residues mediating the exosite interaction (232–255) shown as yellow sticks. A stereo close-up of the contact region is shown on the right, with hydrogen bonds, salt bridges, and water-mediated contacts depicted with broken black lines. The residues of principal importance are labeled. (B) The same exosite contact region as in (A) but with AT in its surface representation and fIXa residues 74 and 143–153 shown as green sticks. A stereo close-up is given to the right, with interactions and labels as before.

Heparin Contacts.

One of the advantages of a high-resolution structure is that it allows unequivocal placement of residues and waters that interact with the high-affinity pentasaccharide. The result is a more complete description of how AT interacts with the pentasaccharide, with waters, main chain atoms, hydrophobics, and the N terminus playing a greater role than previously appreciated (Tables S3 and S4 and Fig. S4). We also observed a surprising heparin interaction in the form of a crystallographic contact between the pentasaccharide and fIXa. The heparin-binding site of fIXa has been inferred from sequence homology to other heparin-binding proteases (exosite II), such as thrombin and fXa, and the contribution of several basic residues was determined by mutagenesis (12). We find that the pentasaccharide binds in the known heparin-binding site of fIXa, utilizing some of the previously implicated residues (Fig. 4A) but in an orientation orthogonal to that observed in the crystal structure of thrombin bound to a heparin octasaccharide (30) (Fig. 4B). This orientation is unlikely to be the favored binding mode in solution because it is on the periphery of the basic exosite II; however, there is no evident reason why this binding mode and not another became the crystal contact. An interesting feature of the observed binding mode is the alignment of the pentasaccharide bound to fIXa with that bound to AT, so that an extension of ∼14 saccharide units between the two sites would be expected to be minimal to bridge the complex (Fig. 4C). Details of the fIXa–pentasaccharide interaction are given in Fig. 4A and Tables S5 and S6 and include salt bridges with known exosite II residues and intimate contacts with several His and Asn residues. Because the binding of heparin to fIXa is likely to be nonspecific, mutagenesis of basic residues will normally have an effect on heparin Sepharose elution or rate of inhibition by AT, even if they only affect long-range electrostatics and do not directly participate in binding. Our structure may not represent the favored heparin-binding mode of fIXa in solution, but it shows a binding mode that is consistent with bridging. It is distinctly possible that fIXa utilizes one binding orientation for diffusion along heparin and another once exosites are engaged to form the proper bridged Michaelis complex.

Fig. 4.

The pentasaccharide-fIXa crystal contact. (A) A stereo view of the surface of fIXa colored according to electrostatics reveals a large basic (blue) patch that interacts with a crystallographically related pentasaccharide molecule (green sticks). Interactions are indicated with broken red lines, with waters shown as red balls. Important residues are labeled. (B) Factor IXa and the pentasaccharide are in the same orientation and colors as above, and for comparison the heparin-binding mode of thrombin is shown by the white sticks. The orientation of heparin bound to thrombin is orthogonal to that observed for fIXa. (C) The observed binding mode of the pentasaccharide on fIXa (cyan) is perfectly aligned and oriented with the pentasaccharide bound to AT (green), which permits the modeling of the bridged Michaelis complex by simply inserting a linear 14-mer heparin chain (yellow) between the two pentasaccharides.

Metal Ions.

Factor IXa contains a high-affinity Ca²⁺-binding site in the 70s loop, and it is also believed, on the basis of sequence homology and some biochemical data, that fIXa coordinates Na⁺ in a manner analogous to thrombin (31). The first structure of porcine factor IXa was obtained in the absence of Ca²⁺, so no information on its coordination was provided. The structure of human fIXa (EGF2 and catalytic domain) was obtained in the presence of Ca²⁺, but the poor resolution resulted in improper octahedral geometry, with distances and angles deviating greatly from accepted values (32). At 1.7 Å we find octahedral coordination with monodentate carboxylates from glutamates 70, 77, and 80, main chain O binding from Asn72 and Glu75, and a single water molecule (Fig. S5A). All distances are below 2.5 Å (average 2.3 ± 0.1 Å), and the angles are close to 90° (average 90 ± 8°), as expected for octahedral geometry. Although the details of Ca²⁺ coordination are different from previous structures, even at low resolution it is difficult to miss a Ca²⁺ atom. Identification of a Na⁺-binding site, however, requires high resolution. Our structure is of very high quality to 1.7 Å with over 670 waters. Because our crystals were grown and cryoprotected in the presence of over 100 mM NaCl, we expected to observe Na⁺ coordinated to fIXa on the basis of a reported Kd in the range of 30 mM (with the Ca²⁺ and substrate bound) (31). However, manual evaluation of waters around the proposed Na⁺-binding site (the 220 loop) did not result in any candidate Na⁺ atom (Fig. S5B). We therefore utilized the program WASP (33) to determine if Na⁺ might be coordinated to an unexpected site, but the results were negative. Exhaustive analysis has shown that with structures of resolution better than 1.8 Å it is unlikely to make a wrong decision with respect to ion coordination (34). In spite of the fact that our structure is at 1.7-Å resolution and the putative Na⁺-binding site is highly ordered with many associated water molecules, we observe no possible Na⁺ ion. This observation is consistent with a recent study showing that the biochemical and functional properties of fIXa do not change in the presence of Na⁺, provided Ca²⁺ is bound (35). Taken together, these data suggest that fIXa is not a Na⁺-binding protease.

Conclusions

The high-resolution structure of the Michaelis complex between pentasaccharide-activated AT and fIXa presented here resolves many of the questions concerning the regulation of fIXa. The overarching questions relate to the poor catalytic activity of fIXa. It is unlikely that fIXa has incompletely converted from the zymogen state to the active state upon cleavage of the Arg15-Val16 bond, as seen for fVIIa (18) and perhaps thrombin (36), because binding to fVIIIa does not increase the rate of hydrolysis of small substrates (16). Factor IXa is thus a fully converted active protease with a conformation that precludes rapid cleavage of small substrates. Improved cleavage of fX requires the engagement of exosites on fIXa and fVIIIa, which help to present the activation loop to fIXa and stabilize the Michaelis complex (37), consistent with the observed improvement of both the k_cat and the K_m terms (17). Our structure shows that AT is a good “substrate” for a similar reason; it forms a stable exosite contact that presents the RCL favorably to the active site of fIXa and maintains it in position until proteolysis commences. However, AT has also evolved to exploit the unique active-site features of fIXa, including the S2–S4 trade-off and the P′ wall, in order to ensure successful completion of the reaction. In conclusion, this structure explains how AT selectively inhibits fIXa and how complex formation is dependent on heparin binding. Factor IXa inhibition by AT is exquisitely sensitive to the presence of the pentasaccharide and may therefore play a larger than expected anticoagulant role when low molecular weight heparins are given therapeutically. This structure also surprisingly reveals how fIXa interacts with heparin. Heparin binding to fIXa has been shown to interfere with the formation of the intrinsic tenase complex (38), and therefore the observed interaction may provide a starting point for the design of exosite-directed fIXa inhibitors.

Materials and Methods

Protein Expression and Purification.

Recombinant human AT (β-glycoform S137A) was expressed in BHK cells and purified as described previously (39). Recombinant human fIXa (EGF2/protease domain) was expressed with small modifications to the method of Hopfner and colleagues (40). Briefly, fIX cDNA (a kind gift from J. McVey, MRC Clinical Sciences Centre, London) corresponding to residues 103–431 was cloned into PET -23(+) expression vector (Novagen), and the S195A mutation was made by site-directed mutagenesis (Stratagene). S195A fIX was expressed in BL21 Star (DE3) cells, and refolding of the inclusion bodies was essentially as previously described (40). Refolded fIX was purified by heparin Sepharose and Q Sepharose chromatography (GE Healthcare). S195A fIX was activated with snake venom protease RVV-X (Haematologic Technologies) prior to repurification by Q Sepharose chromatography.

Crystallization, Data Collection, and Refinement.

Proteins were concentrated and buffer exchanged into 20 mM Tris, 150 mM NaCl, 2 mM CaCl₂, pH 7.4 prior to complex formation in the presence of an equimolar concentration of high-affinity pentasaccharide [idraparinux, a kind gift from M. Petitou (Sanofi Synthelabo, France)]. Crystals grew in 2 days in hanging drops consisting of 2 μL complex (159 μM) and 1 μL precipitant (0.25 M ammonium acetate, 19.5% PEG 3350). After cryoprotecting the crystals in 25% 2-methyl-2,4-pentanediol (MPD), 7% PEG 3350, 8.8 mM ammonium acetate, 100 mM NaCl, 13 mM Tris, pH 7.4, 1.3 mM CaCl₂, crystals were flash-cooled in a stream of nitrogen vapor at 100 K prior to data collection at station I02 at the Diamond Light Source (Didcot, UK). Data were processed by using Mosflm, Scala, and Truncate (41), and the structure was solved by molecular replacement with Phaser (42), by using 1RFN and 1E03 as search models. Refinement was carried out by using CNS (version 1.2) (43), and XtalView (44) was used for model building. Data processing and refinement statistics are given in Table 1. The geometry was validated by using the MolProbity server (45), which found 96.6% of the amino acids in the favored Ramachandran regions. Figures were made by using Pymol (46), and the interaction area was calculated by using AREAIMOL (part of the CCP4 suite). The catalytic domain of fIXa is numbered according to chymotrypsin, with insertion loops indicated by sequential letters. Coordinates and structure factors are deposited in the Protein Data Bank under PDBID code 3KCG.