Thrombin inhibition by the serpins


  • J. A. Huntington

    Corresponding author
    1. Department of Haematology, University of Cambridge, Cambridge Institute for Medical Research, Cambridge, UK
    • Correspondence: James A. Huntington, Department of Haematology, University of Cambridge, Cambridge Institute for Medical Research, Wellcome Trust/MRC Building, Hills Road, Cambridge CB2 0XY, UK.

      Tel.: +44 1223 763230; fax: +44 1223 336827.


    Search for more papers by this author


Thrombin is the central protease in the blood coagulation network. It has multiple substrates and cofactors, and it appears that four serpins are responsible for inhibiting the thrombin produced in haemostasis and thrombosis. Structural studies conducted over the last 10 years have resolved how thrombin recognises these serpins with the aid of cofactors. Although antithrombin (AT), protein C inhibitor (PCI), heparin cofactor II (HCII) and protease nexin-1 (PN1) all share a common fold and mechanism of protease inhibition, they have evolved radically different mechanisms for cofactor-assisted thrombin recognition. This is likely to be due to the varied environments in which thrombin is found. In this review, I discuss the unusual structural features of thrombin that are involved in substrate and cofactor recognition, the serpin mechanism of protease inhibition and the fate of thrombin in the complex, and how the four thrombin-specific serpins exploit the special features of thrombin to accelerate complex formation.

Thrombin function, structure, interactions and localisation

Befitting its central and pleiotropic role in haemostasis, thrombin is a unique and complex enzyme [1, 2]. It is formed at the end of the coagulation cascade of proteolytic events by cleavage of two bonds, one of which (Arg320–Ile321, 15–16 in chymotrypsinongen numbering, used throughout for thrombin [3]) converts the zymogen to an active protease (meizothrombin), and the other (Arg271–Thr272) releases thrombin from its membrane anchoring pro-domains. This is believed to occur on the activated platelet where coagulation factors (f) have assembled by binding to the negatively charged, phosphatidylserine-rich cell surface [4]. Several vitamin K-dependent factors, including fVII, fX, fIX, and prothrombin (fII) will bind via their gammacarboxyglutamic acid containing Gla domains, and cofactors fV and fVIII bind to the platelet surface via their C-domains. Unactivated platelets and fibrinogen also localise to the site of the nascent clot. The environment in which newly formed thrombin finds itself is therefore rich with substrates. Its uniqueness partially derives from the fact that thrombin is the only protease generated in blood coagulation that has lost its membrane anchoring Gla domain and can freely diffuse to encounter substrates bound to other cells and in the interstitial space. Thus, thrombin activates fV, fVIII and fXI to effect an increase in its own production, activates platelets by cleavage of PARs (protease-activated receptors) and cleaves fibrinogen to form the fibrin network. It further stabilises the clot by activating the fibrin-crosslinker fXIII and by inhibiting fibrinolysis through activation of the metalloprotease TAFI (thrombin-activatable fibrinolysis inhibitor). When thrombin diffuses to the edge of the clot and interacts with thrombomodulin on the intact endothelium its activity is reversed. It is no longer able to interact with and cleave any of its procoagulant substrates, and instead activates protein C to shut off the Xase (fVIIIa–fIXa) and prothrombinase (fVa–fXa) complexes, and thus attenuate its formation. Thrombin is therefore unique amongst the coagulation proteases due to the removal of its pro-domains to allow free diffusion throughout the forming clot so that it can encounter its multiple substrates in several environments.

How thrombin carries out these multiple functions is reasonably well understood on a structural level [3]. It is composed of a single chymotrypsin-family (S1 in MEROPS nomenclature [5]) serine protease domain (the heavy chain) covalently linked via a disulphide bond to a short remnant of the prodomain (Fig. 1A). The light chain has no known function, although a recent study suggested that it acts as a tethered ligand to stabilise the conformation of the protease domain [6]. Thrombin is generally (and should be) shown in the standard orientation (as in Fig. 1A,B), with its active site facing so that a substrate runs from left-to-right (N-terminal to C-terminal). The active site is located between two β-barrels, with Asp102 and His57 donated by the N-terminal β-barrel (upper) and Ser195 and the oxyanion hole (Gly193 and Ser195) provided by the C-terminal β-barrel (lower). The C-terminal β-barrel also forms the primary specificity pocket, S1, that accommodates a P1 Arg side chain of all thrombin substrates (traditional subsite (S) and substrate (P) numbering scheme of Schechter and Berger [7] is used in this manuscript). Thrombin is longer than chymotrypsin, containing ten, so-called ‘insertion loops’ in its heavy chain, the two largest of which form the top and bottom lips of the active site cleft. The 60-loop has an insertion of nine residues, and due to the presence of two prolines and hydrophobic residues (Tyr, Trp, and Phe) is rather rigid and is thought of as a hinged lid. The 147-loop (also known as the 149-, the ‘autolysis-’ and γ-loop) is five residues longer than that of chymotrypsin, is flexible, and protrudes from the bottom of the active site cleft. The presence of these two loops restricts access to the active site of thrombin (the depth of the active site cleft is evident in Fig. 1C,D). Substrates must therefore be a part of long unstructured loops, or the 60 and γ-loops will likely make contact with the body of the substrate protein. Thus, the relatively inaccessible active site cleft of thrombin helps to restrict its activity, and also, by providing further interaction sites, to confer potential for improved specificity. This is particularly relevant for recognition of thrombin by serpins.

Figure 1.

Thrombin structure and properties. (A) A cartoon representation of thrombin (S195A thrombin taken from 1JMO) is shown in stereo, with its heavy chain coloured from its N-to-C terminus (blue to red) and the light chain in grey. The catalytic triad is shown as sticks and the oxyanion hole nitrogen atoms are blue balls. Important features are labelled, including anion binding exosites I and II (ABE1 and 2) and the 60- and γ-loops. The Na+ ion is a magenta ball. (B) Electrostatic surface representation of thrombin (same structure and orientation as in (A) with the RCL of HCII in place (P4–P3′) to indicate the active site cleft. (C) ABE1 is seen by rotating thrombin by 45°to the left. (D) ABE2 can only be seen side-on with a 65° rotation to the right. The depth of the active site cleft is evident.

Colocalisation is an important theme in thrombin substrate recognition [2], and since thrombin has lost the ability to bind to membranes it relies on cofactors to help tether it to cell surfaces. The cofactors, by definition must bind away from the active site. Thrombin possesses two such exosites that mediate all cofactor interactions, and contribute to the recognition of nearly all substrates: anion-binding exosites I and II (ABE1 and ABE2). In the standard orientation, ABE1 is visible as a basic (blue) patch just beyond where the C-terminal portion (P′ side) of the substrate loop exits the active site cleft (Fig. 1B). A 45° rotation about the y-axis shows the extent of ABE1 (Fig. 1C). What is not evident from the figure is the hydrophobic nature of the site, which contributes greatly to the binding energy of ABE1-binding ligands. ABE2 is located on the opposite side of thrombin, around the back in the standard orientation, so that a −65° rotation from the standard orientation is required to get a good view of it (Fig. 1D). ABE2 is considered to be more basic than ABE1, and hydrophobic contacts play a limited role in substrate and cofactor interactions [8]. ABE1 was originally called the fibrinogen-recognition exosite, since it forms an obligate interaction for fibrinogen cleavage, and ABE2 was known as the heparin-binding exosite, since it binds heparin and other highly acidic glycosaminoglycans (GAGs).

Crystal structures have been solved for all of the thrombin-cofactor interactions. Although fibrinogen is a substrate, the product of thrombin cleavage, fibrin, is a cofactor for the activation of fXIII. The main interaction is between ABE1 and the fibrinogen E domain (Fig. 2A). For fibrinogen this interaction (two thrombins per E domain [9]) presents the cleavage sites on the A and B chain into the active site cleft of thrombin. It is expected that much of the thrombin formed will remain bound to fibrin in the mature clot [10]. A splice variant of the γ-chain leads to an extended, highly acidic C-terminus present in about 15% of fibrinogen molecules [11]. This fibrinogen γ′ region binds to ABEII [12] with high affinity (Fig. 2B), and may contribute to the sequestration of thrombin within the fibrin clot. The high affinity receptor for thrombin on platelets also binds via ABEII [13, 14] (Fig. 2C). The C-terminal portion of GpIbα is highly acidic and contains sulphated tyrosine residues (a hallmark of thrombin exosite binding sequences). Although some reports suggest the involvement of ABE1 in binding to GpIbα [15], the biochemical data support a cofactor function for GpIbα in cleavage of PAR1, which itself requires ABE1 [16]. Thus, a fraction of the thrombin formed during the initiation and growth of a thrombus will be bound to the components of the clot, fibrin and platelets, via ABE1 and ABE2. Thrombin which diffuses to the intact endothelium at the edge of the clot will encounter two cofactors with anticoagulant functions. Thrombomodulin (TM) is a type-1 transmembrane protein composed of several domains, including six EGF-like domains [17]. The two C-terminal EGF domains (5 and 6) bind to ABE1 of thrombin with high affinity (Fig. 2D) and effectively prevent interaction with any procoagulant substrate [18]. The fourth EGF domain binds to PC and helps present its activation loop to the active site of thrombin for cleavage. Activated protein C (APC) is an anticoagulant protease that efficiently shuts down the coagulation response by inactivating fVa and fVIIIa with the help of its cofactor, protein S. A fraction of TM contains a chondroitin sulphate (CS) moiety in an unstructured region following EGF6 that can improve affinity of TM for thrombin by the addition of an ABE2 interaction. Although there is no structure of CS bound to thrombin, it is anticipated to resemble that of heparin [19] (Fig. 2E), which makes several ionic and hydrogen-bonding interactions with ABE2. The other cofactor resident on the intact endothelium is heparan sulphate containing glycoproteins. Heparan sulphate is similar to heparin and is expected to bind in the same way. It similarly serves as an anticoagulant cofactor by improving the rate of thrombin inhibition by serpins [20, 21].

Figure 2.

Thrombin interactions with relevant cofactors. Thrombin is oriented and displayed as in Fig. 1C,D, depending on whether the cofactor binds to ABE1 (right) or ABE2 (left). (A) The central E-domain of fibrin is shown as a cartoon. (B) The γ′ peptide is shown as sticks. (C) GpIbα is shown as cartoon, with the C-terminal acidic region shown as sticks. (D) The EGF5 and 6 domains are shown in cartoon representation. (E) Heparin is shown as sticks.

The two exosites of thrombin, as well as its active site, are not available in the zymogen prothrombin. ABE2 is occupied by the kringle 2 domain of its own profragment [22], and the active site and ABE1 are unstructured and therefore unable to recognise substrates and cofactors [23]. Thus, prothrombin does not circulate bound to its substrate fibrinogen or line the intact vasculature by interacting with TM. It is a zymogen with respect to its ability to bind to any substrate or cofactor, and thus free to circulate at a high concentration to provide a reservoir of potential thrombin generation to any site in the vascular tree. Interestingly, when thrombin is formed it still retains some features of the zymogen, with a flexible active site and ABE1 [6, 24]. Binding of thrombin to the ever-present Na+ (see Fig. 1A) helps to form the active site, including the S1 pocket and the oxyanion hole, but is not a regulatory cofactor, as has been argued by some [25, 26]. Rather, Na+ should be considered part of the structure of thrombin, like any internal hydrogen bond or van der Waals contact, similar to the high-affinity Ca2+ binding sites in fVIIa, fIXa, fXa and APC, amongst others.

Serpin structure and mechanism

There are multiple families of serine endopeptidase inhibitors in man, but only members of the serpin family inhibit thrombin. In fact, of the 36 serpins in the human genome, four (11%) are known physiological inhibitors of thrombin: antithrombin (AT), heparin cofactor II (HCII), protein C inhibitor (PCI) and protease nexin-1 (PN1). This speaks to the importance of thrombin in many tissues, and to the manifold cofactors to which thrombin can bind. From the viewpoint of haemostasis, thrombin will be found bound to the clot via interactions with fibrin and platelets, on the edge of the clot bound to TM and glycosaminoglycans, and importantly it is also found free in the lumen of the vessel. One serpin is evidently not enough to recognise and inhibit thrombin in these varied environments.

The mechanism by which serpins inhibit serine proteases is very well characterised, and several reviews have described it in detail [27-29]. Serpins are composed of three β-sheets and around nine α-helices, and in their native state, a flexible 20-residue RCL protrudes away from the body of the serpin as ‘bait’ for proteolytic attack. The standard orientation of a serpin has the RCL on top and the main sheet (β-sheet A) facing forward. The mechanism, general to all inhibitory serpins, is easily summarised in a three-panel figure (Fig. 3) showing the protease approaching the native or activated serpin, the two proteins docked in a productive Michaelis complex (RCL accommodated in the active site cleft to align the P1–P1′ bond with the catalytic machinery), and the final complex with the two proteins linked by an ester bond between Ser195 of the protease and the P1 residue of the serpin. The final complex is trapped as the acyl-enzyme intermediate by the conformational changes in the serpin (RCL insertion into sheet A) and in the protease (distension of the catalytic loop and disruption of the oxyanion hole) [30].

Figure 3.

Serpin structure and mechanism of protease inhibition. The common serpin mechanism can be described minimally by three panels. Left shows the protease (top, cyan) and the serpin (bottom, grey with RCL in yellow and P1–P1′ as sticks, and β-sheet A in red) before interacting with one another. The central panel depicts a typical Michaelis complex. The right panel shows the final serpin-protease complex, with the protease translocated to the opposite pole of the serpin and covalently linked by an ester bond (acylenzyme intermediate). The protease is coloured according to B-factor, with orange and red indicating disorder. About 40% of the protease is missing in the structure due to the reversal of the zymogen activation mechanism.

Each step confers potential for regulation. Formation of the Michaelis complex requires the diffusion of the two proteins in three-dimensional space and a productive collision. This is clearly too important to leave to chance, and so directional diffusion is conferred by ‘electrostatic steering’ [31], either through direct attraction of complementary charged surfaces on the protease and the serpin, or indirectly by the influence of a cofactor on the surface charge. In addition, GAGs can influence the dimensionality of the diffusion, either by localising protease and serpin to the same cell (two-dimensional diffusion), or by binding to the same linear GAG chain (one-dimensional diffusion). This first step is of critical importance for the serpins, because of the irreversible nature of their inhibition mechanism. Essentially all encounters lead to inhibition, so that the first step is only reversible in principle [32]. The protease is effectively inhibited upon formation of the Michaelis complex. It is not know how long-lived the Michaelis complex is, or indeed what triggers the rapid incorporation of the RCL into sheet A, and this is likely to depend on the serpin-protease pair. However, the P1′ residue is likely to stay in position for a while after cleavage, especially if there are many P′ and exosite interaction. It may therefore be appropriate to consider serpins as transient canonical serine protease inhibitors, similar to aprotinin and other ‘reversible’ inhibitors.

Eventually, though, the RCL will insert into sheet-A and the protease will be flung to the opposite pole of the serpin. It is unclear how rapidly this occurs, but it should be sufficiently fast to prevent deacylation. This conformational change results in a hyper-stable serpin, and the ‘plucking’ of the tethered catalytic loop out of the active site. Ser195 moves by 4 Å and the carboxylic acid group of the side chain of Asp194 is pulled away from the activation pocket and its salt-bridge with the N-terminus (Ile16). This event reverses the zymogen-to-protease conformational change and re-disorders the zymogen activation domain, resulting in a highly destabilised protease. Consequently, the protease becomes susceptible to proteolysis in regions previously protease resistant [30], and structural features important for ligand binding become dysfunctional [33-35]. This is particularly important for thrombin, since much of it is associated to cofactors on cells and to fibrin when it is encountered by serpins. We and others have shown that the two anion binding exosites of thrombin become disrupted when complexed by serpins, and dissociate rapidly from any bound ligand [34-36]. Inhibition of thrombin by plasma serpins AT, HCII and PCI releases it from its cofactor and allows the complex to circulate freely until it encounters its clearance receptor LRP (low-density lipoprotein receptor related protein) in the liver [37, 38]. We have also identified the cryptic LRP binding site in thrombin that is revealed by its complexation by serpins (unpublished data). It is not a new conformation per se, but an unfolding of a key cationic region that results in the high-affinity interaction with LRP and the subsequent clearance of the thrombin-serpin complex.

Thrombin-serpin Michaelis complexes

We have solved the crystal structures of the Michaelis complexes between thrombin and AT [39], HCII [40], PCI [41] and PN1 [42]. The only other structure of a complex with thrombin was solved by the Gettins group, and involved AT [43]. Each crystal contained a single copy of the complex in the asymmetric unit, except for PN1, which contained two. That makes six structures of serpin-thrombin Michaelis complexes. When the serpin component is superimposed, a surprisingly wide range of thrombin binding modes is evident (Fig. 4A). Similarly, when thrombin is superimposed, the serpins clearly take different approaches to accessing the active site cleft (Fig. 4B). It is an absolute requirement of a productive Michaelis complex that P1–P1′ bond is properly oriented in the active site, but the rest of the contacts reveal a surprising degree of freedom (Fig. 4C). This is likely due to the differences in composition of the RCLs, but may also be a function of the exosite contacts. It is generally considered that interactions involving RCL residues from P4 to P3′ constitute active site contacts [7], and all other regions of the serpin that engage a protease constitute exosite contacts. Due to the depth of the active site cleft of thrombin (discussed above) it is not surprising that all serpin-thrombin Michaelis complexes engage the 60- and γ-loops, and that exosite contacts are therefore a necessary part of serpin recognition by thrombin. A calculation of the solvent-inaccessible (buried) surface for each complex (using the program PISA [44]), with and without the P4–P3′ contribution, gives an idea of the relative contributions of the RCL and the exosites (Table 1). The particulars of each serpin's mechanism of thrombin recognition, and the role of cofactors is discussed below.

Table 1. Contact interfaces for serpin-thrombin Michaelis complexes
 Interface residuesBuried surface area (Å2)
1TB6 (AT-T-H)523012131455266812021466
1SR5 (AT-T-H)40228018871688762926
3B9F (PCI-T-H)4217920115320731209864
1JMO (HCII-T)593916951793348811162372
4DY7 (PN1-T) 126147258411566822744
4DY7 (PN1-T) 240259681107207510291046
Figure 4.

Superposition of crystal structures of serpin-thrombin recognition (Michaelis) complexes. (A) All six complexes are shown with the serpin components superimposed (serpin coloured according to secondary structure, with helices in red, sheets in yellow and loops in green). The heavy chain of thrombin is in cyan and the light chain in magenta. (B) Same depiction as in A, but with thrombin superimposed. (C) RCL binding modes for productive Michaelis complexes shows how only P1–P1′ positions are conserved (thrombin surface in grey, AT from 1TB6 in blue, AT from 1SR5 in cyan, PCI in yellow and PN1 in red).


AT is the principal thrombin inhibitor in the blood. It circulates at 2.5 μm, and reduction in levels is strongly associated with thrombosis [45], but it is not strictly ‘specific’ for thrombin since it inhibits both fIXa and fXa. However, inhibition of thrombin by AT is probably the principal mechanism of removal of the majority of the thrombin produced during thrombosis and the normal haemostatic response. The rate constant for thrombin inhibition is approximately 104 m−1 s−1, giving thrombin a half-life of about 30 s the vessel lumen. In the presence of heparin the rate is increased by about 1000-times due to the template provided by the linear GAG on which AT and thrombin interact. Heparin chains of 18 monosaccharides and larger are capable of forming a ‘bridged’ ternary complex, with heparin binding to AT via its specific pentasaccharide and thrombin binding towards the non-reducing end. This presumably also happens on the edge of the clot where the intact endothelium is lined with heparan sulphate.

Two structures of AT in complex with thrombin were published back-to-back in 2004 [39, 43]. Both employed a synthetic hexadecasaccharide composed of the AT-specific pentasaccharide, a 7-unit unsulphated linker and 4-unit sulphated thrombin-binding region on the non-reducing end. The structure by Dementiev et al. utilised human plasma-derived α-AT (glycoform with N-linked carbohydrate at Asn135) and anhydrothrombin, produced from human thrombin. The structure by Li and colleagues used a recombinant β-glycoform (S137A) of human AT that contained an extra disulphide bond to prevent conversion to the latent form (V317C/T401C), and recombinant S195A thrombin. Crystallisation conditions and the crystal symmetry and dimensions were remarkably similar for the two groups, but the structures were different in several important ways. In the 2.5 Å structure by Li, thrombin is oriented towards the front of AT with its heparin binding site aligned with that of AT (Fig. 5A). Every atom of the hexadecasaccharide was modelled into high-quality electron density, including the non-sulphated linker. In the Dementiev structure, thrombin is translated 11 Å towards the ‘back’ of AT, and the heparin binding site is rotated by 46°, also towards the back of AT (Fig. 5B). They also observe no electron density for the linker region, and only find one unsulphated monsaccharide bound to thrombin. Close analysis of the electron density ascribed to this monosaccharide reveals that it is actually two glycerol molecules, consistent with previous crystal structures of thrombin (e.g. 1JOU [46]). It was argued that the binding of heparin to ABE2 is non-specific and therefore non-directional in nature, however, the lack of density for the linker or for any sulphated saccharides in the known heparin binding region of thrombin, suggests that the complex in the crystal is not actually bridged by the synthetic heparin. Indeed, it would require at least two more linker monosaccharides to allow bridging of the Dementiev complex, providing modes of binding observed in the Li structure are preserved. It is therefore necessary to conclude that the structure of the AT-thrombin complex solved by Dementiev (1SR4) is not bridged by the synthetic hexadecasaccharide. However, the position of thrombin relative to AT is consistent with the observation that an 18mer heparin chain is minimal to accelerate thrombin inhibition by AT. It can also be argued that the forward position of AT observed in the Li complex is actually induced by the short length of the synthetic heparin, and possibly contributed to by the mutations in AT. The Li structure is more intimate than the Dementiev structure (Fig. 5 and Table 1), however, burying 2668 Å2 compared to 1688 Å2. The forward location of thrombin in the Li structure is also more consistent with the insensitivity of AT to the pentasaccharide-induced conformational change (hinge-region expulsion). One further possibility is suggested by the structures of PN1 in complex with thrombin (discussed later); perhaps both structures are sampled, and both are on the pathway to full inhibition, with the Dementiev structure occurring first and rotation and translation of thrombin to the Li structure optimises contacts and sets the stage for rapid RCL insertion.

Figure 5.

Individual serpin-thrombin complexes with contact regions. Left panel is the complex coloured as in Fig. 4 (heparin shown as sticks). The central panel is a cartoon representation of heavy chain of thrombin in the standard orientation, with contact residues coloured red. The right panel is the top surface of the serpin, with contact residues in red. A–F are AT from 1TBG, AT from 1SR5, PCI, HCII, PN1 complex 1, and PN1 complex 2. The asterisk indicates that heparin was in the crystal but a bridge between the serpin and thrombin was not observed in electron density.

Protein C inhibitor

PCI is an unusual serpin in several ways [47, 48]. It is expressed in many cell types and is present in many tissues. It has an extended RCL, with three extra residues on the P′ region. It has dual specificity with a P1 Arg directing it towards trypsin-like proteases and a P2 Phe that can be used as a P1 by chymotrypsin-like proteases. The bulky P2 Phe is also difficult to accommodate in the S2 pocket of thrombin. PCI is activated by heparin towards thrombin and APC, but the presence of heparin abrogates inhibition of tissue kallikrein. In the blood PCI can inhibit the fVIIa–TF complex, fXa, thrombin, the thrombin-TM complex and APC, making it both an anti- and a pro-coagulant serpin. It can bind to hydrophobic ligands, including retinoids and phospholipids. It is internalised by cells and vesicles in a receptor-independent fashion. Its heparin binding region is along helix H, not on helix D as all other known heparin binding serpins. Finally, PCI is not expressed in the liver of mice, meaning that PCI is not present in the mouse blood, thereby making it difficult/impossible to use mouse models to determine its role in haemostasis in humans. Knocking out the mouse PCI gene leads only to male infertility due to a defect in spermatogenesis. Although there is no human disease associated with PCI deficiency, PCI-APC complex can be detected in the plasma of people who have suffered a thrombotic event, and increased circulating levels are associated with thrombosis and myocardial infarction. PCI inhibits thrombin quite well in the absence of heparin, with a second-order rate constant of about 2 × 104 m−1 s−1. Heparin accelerates the rate by only about 10-fold by a template mechanism. Its concentration in blood plasma is only about 90 nm, giving a t½ for thrombin of about 6 ½ min in the absence of heparin and 38 s in the presence of heparin. Thrombin inhibition by PCI is also accelerated by TM (2.6 × 10m−1 s−1), giving a t½ of 3 s. It is therefore possible that a principal function of PCI in the circulation is to inhibit thrombin bound to TM.

There is no structure of PCI bound to APC or the thrombin-TM complex, but we have solved the structures of cleaved [49], native [50] and heparin-bound PCI [51], and the PCI-thrombin-heparin Michaelis complex to 1.6 Å resolution [41] (Fig. 5C). The structure provides some insight into some of the unusual features of PCI described above. Thrombin makes an intimate complex with PCI, burying a total surface area of 2073 Å2, but most of the contact is contributed by the RCL (1209 Å2), leaving only 864 Å2 for exosites (Fig. 5C, centre and right panels). The thrombin contacts are quite similar to those seen in the non-bridged AT-thrombin-heparin complex (1SR5, Fig. 5B, centre), with the γ-loop devoid of contacts with the serpin. Relative to its position in the heparin-bridged complex with AT (1TBG, Fig. 5A), thrombin has translated by 17 Å and rotated by 77°. This orientation aligns ABE2 with the heparin-binding H-helix of PCI and suggests that a linear 14-monosaccharide heparin chain would be minimal to bridge the two molecules. This unusual position of thrombin is allowed by the extension of the P′ side of the RCL of PCI, which effectively holds thrombin at arms-length and allows the alignment of the heparin binding sites. The P2 Phe is accommodated only by the opening of the S2 pocket, achieved by the displacement of the 60-loop, and it is the 60-loop that makes the major exosite contact with PCI.

Heparin cofactor II

HCII is also an unusual and strange serpin [52]. It is the only thrombin substrate or inhibitor that doesn't have an Arg at the P1 position. Its P1 Leu residue would suggest activity against chymotrypsin-type serine proteases, and indeed HCII is an inhibitor of chymotrypsin and neutrophil cathepsin G. Cathepsin G and another protease in neutrophils, elastase, cleave HCII at the N-terminus (residue 66), releasing a chemoattractant peptide. It is possible that HCII cleavage can serve as a signal of tissue damage. However, the main function of HCII is to inhibit thrombin. It circulates at 1.2 μm in a conformation that inhibits thrombin very slowly indeed (600 m−1 s−1, t½ of 16 min). Thrombin specificity is conferred by a 90-residue N-terminal extension that has two acidic stretches containing sulphated Tyr residues and resembling the C-terminus of hirudin. Heparin, heparan sulphate and dermatan sulphate can all activate HCII 1000-fold by releasing the tail from a cryptic position in the native serpin to an exposed position in the GAG-bound state. HCII is not as important a thrombin inhibitor as AT, since its deficiency does not lead to thrombosis [53, 54]. However, it appears to provide an important back-up system since increased levels protect against restenosis after placement of coronary stents [55]. Studies using HCII knockout mice also suggest a role in protection from altherosclerosis/arterial thrombosis in a dermatan sulphate-dependent manner [56].

In 2002 we published the structures of native and thrombin-bound HCII [40]. The native structure surprisingly resembled that of native AT, with a partially inserted RCL that effectively reduced its flexibility and accessibility to proteolytic attack. This was an interesting find because the inactivity of the native state was already ensured by the sequestration of the acidic N-terminal tail and the composition of the P1 residue (Leu instead of Arg). Unfortunately the position of the N-terminal tail could not be resolved in the crystal structure of native HCII. A small helical portion, corresponding to residues Leu61–Asp71, was observed, but in a position clearly different than that sampled in solution. It was anticipated that the acidic tail would interact with the heparin binding site (similar to that of AT, along helix D), and that its binding would displace the tail in a purely competitive fashion [57]. The conservation of the native fold and the heparin binding site suggested a similar allosteric mechanism of activation to AT. This was further verified by biochemical studies [58]. The details of the mechanism of tail release depend on determining the precise mode of interaction between the tail and the body of HCII, but it is clear that HCII undergoes the same large scale conformational change upon GAG binding as AT. We have unpublished data supporting the sequestration of the tail in the vicinity of the heparin binding site, so a combined partial direct displacement and conformational change is likely to release the tail for interaction with thrombin.

This hypothesis was strengthened by the structure of HCII in complex with S195A thrombin, which showed an expelled RCL and the elongation of the hinge region to allow the docking of thrombin to the back-left of HCII (25.6 Å translation and 120° rotation relative to 1 TB6). The complex could not be formed without the expulsion and extension of the hinge region of the RCL, and so necessitated the AT-like conformational change. As for other thrombin-serpin Michaelis complexes, the heparin binding ABE2 was aligned with that of HCII so that a linear GAG could bridge the complex, but only if the GAG bound in a perpendicular fashion to helix D. Only a portion of the tail could be resolved in electron density (residues [54-62]), but this region contains the two hirudin-like sequences that confer thrombin specificity (56EDDDY*LD62 and 69EDDDY*ID75). In a manner similar to hirugen, the acidic region seemed to play a secondary role in the interaction with thrombin, with only two ionic interactions (Asp70 and Asp72 binding to Lys110 of thrombin). The bulk of the contact was hydrophobic, and involved a short helical segment between the two acidic stretches (63LEKIFS68). The contact surface area between thrombin and HCII is extensive, covering almost 3500 Å2, with 2/3rd provided by the exosite (Fig. 5D and Table 1).

Protease Nexin-1

PN1 is the most efficient thrombin inhibitor of all the serpins, with a second-order rate constant of about 1 × 106 m−1 s−1 in the absence of heparin and 1 × 10m−1 s−1 in the presence of heparin [59]. Unlike all other heparin-binding serpins, PN1 is not produced in the liver and does not circulate in its free form. Rather, PN1 is produced by macrophages, platelets and smooth muscle cells, and when secreted it remains associated to the surface of cell that made it. Cell surface association is through an unusually tight interaction with GAGs of various sorts. Its affinity to heparin is in the low nM range and is entirely insensitive to the formation of the final serpin-protease complex. So, instead of being released from the cell surface for clearance of the complex by the liver, PN1-thrombin complexes are cleared by the cells on which they reside (again through LRP binding) [60]. The residence of PN1 on cell surfaces has been proposed to set a threshold for cell signalling by thrombin, including platelet activation by PAR-1 cleavage [61]. Consistent with this hypothesis, recent studies have demonstrated an important anticoagulant effect of platelet PN1 [62].

The question of why PN1 is such an efficient inhibitor of thrombin was partially answered by the recent publication of structures of PN1 bound to heparin, and the PN1-thrombin complex [42]. PN1 was seen to have a highly basic patch along helix D, consistent with it providing a conformationally insensitive heparin binding site, and another basic area on and around the RCL. The crystal structure of the PN1-thrombin complex contained two copies in the asymmetric unit, and we were surprised, and initially disappointed to find thrombin in two very different orientations. In one (complex 1), thrombin was tilted towards the front right of the molecule (Fig. 5E) in an orientation similar to that seen for the ternary AT-thrombin-heparin complex (only 10 Å translation and 15° rotation relative to 1 TB6). The heparin binding sites are therefore aligned and only a short distance away, so that a heparin chain of 17 monosaccharide units would be sufficient to bridge thrombin and PN1. This is what would be expected for a heparin-dependent Michaelis complex. However, although the P1 Arg of PN1 is buried in the S1 pocket, for some reason, the P1′ Ser residue is not in the S1′ pocket, but rather abruptly exits the active site cleft. The P1–P1′ bond is therefore not positioned in a manner that would permit nucleophilic attack by the catalytic Ser, nor is the P1 main chain oxygen atom in position to H-bond with the components of the oxyanion hole. Complex 1 is thus able to be bridged by heparin, but is unable to support proteolysis of the RCL, so cannot form the final serpin-protease complex.

The second complex in the asymmetric unit is also highly unusual (Fig. 5F). Thrombin has shifted position relative to PN1 by a small translation of 8 Å and a large rotation of 108°. As a result, the heparin binding ABE2 of thrombin has moved around to the back of PN1, so that bridging by a linear heparin molecule would no longer be supported. However, in complex 2 the RCL is fully engaged with the active site of thrombin, and the catalytic machinery is properly aligned with the scissile P1–P1′ bond. The difference between correct engagement of the P′ side in complex 2 vs. complex 1 is clear from the central panel of Fig. 5. For all complexes, save complex 1 (Fig. 5E), there are contacts on the S′ side of thrombin. Both complexes have substantial buried surface area, but complex 2 has more extensive RCL and exosite interactions. Another surprising finding (perhaps the most surprising) was that the thrombin component of complex 2 was highly disordered relative to complex 1, or indeed any other serpin-thrombin Michaelis complex. Disorder was manifested by poor electron density throughout, and high B-factors (average Cα B-factor for thrombin in complex 1 was 40 Å2, compared to 64 Å2 in complex 2). The highest B-factors were found in ABE2. We found no electron density for critical residues Arg93 and Arg126, and the remaining ABE2 residues (91, 101, 233, 236, and 240) had average Cα B-factors of 88 Å2 (vs. 51 Å2 for the same ABE2 in complex 1). ABE2 thus appears to be ordered and aligned in complex 1, but disordered and out of alignment in complex 2. So, in a single crystal we found two complexes, each proficient in one respect and deficient in another, and hypothesized that both are on the pathway to formation of the final complex. Complex 1 is the initial encounter complex whose rate of formation is accelerated by directional diffusion along heparin, guided by electrostatics. The S1 pocket of thrombin is occupied and therefore thrombin is effectively inhibited. This is similar to the binding of anti-thrombin proteins from blood-sucking organisms, where electrostatically-driven exosite contacts and non-productive active site engagement is the basis of inhibition. PN1, however, does not stop there. Rotation of thrombin's ABE2 towards the back of PN1 dissociates thrombin from heparin, and simultaneously, productive engagement of the RCL or exosite contacts somehow disorders the heparin binding site to prevent its re-association to heparin. Such a mechanism would ensure directionality of the reaction from bridged to Michaelis to final complex. We hypothesise that this unusual mechanism is necessitated by the extremely tight binding of PN1 for GAGs, and the insensitivity of the interaction to conformational change. HCII and PCI bind to GAGs weekly, so can dissociate easily upon complex formation for clearance. AT loses 1000-fold of its affinity for its specific pentasaccharide upon insertion of the RCL, so dissociates from heparin as soon as RCL insertion begins. On the other hand, PN1 stays bound to GAGs during the transition to the final complex, and the bridging of thrombin would predictably slow RCL insertion sufficiently to allow deacylation to occur.


Perhaps it's unsurprising that a protease with as many substrates as thrombin will also have multiple endogenous inhibitors. Although all of these thrombin inhibitors share a common core serpin fold, the mechanisms by which they recognise thrombin with the aid of cofactors differ radically from one another. Functional and crystallographic studies conducted over many years have combined to paint a rather complete picture of how the special features of thrombin are exploited by serpins and the role that cofactors play.


Funding for the Huntington Lab is provided by the British Heart Foundation.

Disclosure of Conflicts of Interest

The author states that he has no conflict of interest.