A combined structural dynamics approach identifies a putative switch in factor VIIa employed by tissue factor to initiate blood coagulation



Coagulation factor VIIa (FVIIa) requires tissue factor (TF) to attain full catalytic competency and to initiate blood coagulation. In this study, the mechanism by which TF allosterically activates FVIIa is investigated by a structural dynamics approach that combines molecular dynamics (MD) simulations and hydrogen/deuterium exchange (HX) mass spectrometry on free and TF-bound FVIIa. The differences in conformational dynamics from MD simulations are shown to be confined to regions of FVIIa observed to undergo structural stabilization as judged by HX experiments, especially implicating activation loop 3 (residues 365–374{216–225}) of the so-called activation domain and the 170-loop (residues 313–322{170A–175}) succeeding the TF-binding helix. The latter finding is corroborated by experiments demonstrating rapid deglycosylation of Asn322 in free FVIIa by PNGase F but almost complete protection in the presence of TF or an active-site inhibitor. Based on MD simulations, a key switch of the TF-induced structural changes is identified as the interacting pair Leu305{163} and Phe374{225} in FVIIa, whose mutual conformations are guided by the presence of TF and observed to be closely linked to the structural stability of activation loop 3. Altogether, our findings strongly support an allosteric activation mechanism initiated by the stabilization of the Leu305{163}/Phe374{225} pair, which, in turn, stabilizes activation loop 3 and the S1 and S3 substrate pockets, the activation pocket, and N-terminal insertion.

Abbreviations: FVII(a), coagulation factor VII(a); FX(a), coagulation factor X(a); SPR, surface plasmon resonance; sTF, soluble tissue factor (residues 1–219); TF, tissue factor; HX, hydrogen exchange; MD, molecular dynamics; SMD, steered molecular dynamics; RMSD, root mean square deviation.

Binding of factor VIIa (FVIIa) to its membrane-bound receptor tissue factor (TF) renders it biologically active and capable of initiating blood coagulation (Davie et al. 1991). FVIIa consists of a trypsin-like protease domain and an N-terminal light chain composed of three noncatalytic domains that tether FVIIa to TF and to the membrane surface (Banner et al. 1996). Parts of the light chain and the protease domain form an extended interface with the N-terminal domain of TF. While contributing little to the total binding energy of the complex (Persson 1997), interactions involving the protease domain promote a series of structural changes in FVIIa that greatly enhance its catalytic efficiency.

Zymogen FVII is converted to the activated form by proteolytic cleavage of the Arg152{15}–Ile153{16}11 bond in the protease domain (Ruf and Dickinson 1998). Insertion of the newly formed N terminus of Ile153{16} into the so-called activation pocket to form a salt bridge with Asp343{194} occurs spontaneously in most trypsin-like proteases and is required for correct definition of the active-site region involving an ordering of the activation loops, maturation of the S1 pocket, and generation of the oxyanion hole (Fehlhammer et al. 1977; Huber and Bode 1978). The propensity of the N terminus in free FVIIa to become buried, however, is poor, resulting in zymogen-like latent features of the protein, and FVIIa first realizes its full catalytic potential upon binding to TF, which promotes N-terminal insertion (Higashi et al. 1994). Three regions of the protease domain seem to be important for the function and allosteric regulation of FVIIa. These are the TF-binding region, the active-site region, and the macromolecular substrate exosite, which also comprises the activation pocket (Ruf and Dickinson 1998; Persson et al. 2001b, 2004; Petrovan and Ruf 2002). The three-dimensional structures of the protease domain of free (Kemball-Cook et al. 1999; Pike et al. 1999; Sichler et al. 2002) and TF-bound (Banner et al. 1996; Zhang et al. 1999) FVIIa are very similar, presumably because of the presence of an active-site inhibitor during the crystallization, and offer very few clues as to the allosteric activation pathway. In one crystal structure of free FVIIa (Sichler et al. 2002), the cocrystallized active-site inhibitor was diffused out of the crystal lattice, revealing a virtually unperturbed S1 pocket, presumably because of constraints exerted by the crystal packing. Recently, structures of TF-bound FVIIa in complex with other small inhibitors (Bajaj et al. 2006) displayed an oxyanion hole that is not preformed as compared to structures with a covalently bound inhibitor. All structures, however, have the N terminus buried and engaged in the salt bridge to Asp343{194}.

Zymogen FVII was crystallized in complex with an exosite binding inhibitory peptide with the structure revealing a new registration of two β-strands, resulting in large differences in the TF-binding region as well as in the macromolecular substrate exosite region when compared to previous structures of FVIIa (Eigenbrot et al. 2001). However, subsequent analysis of the crystal packing showed that intermolecular contacts at the site of the reregistration make it unlikely that the β-sheet organization plays a role in activation (Perera and Pedersen 2005). This was further substantiated by mutagenesis (Olsen et al. 2004) and recently by HX experiments on zymogen FVII in solution (Rand et al. 2006). Detailed insights into the structure of noninhibited, free FVIIa are needed to reveal the differences at the single-residue level between the unbound or latent forms and the enzymatically active TF-bound form of FVIIa.

The functionality of an increasing number of proteins appears to be critically linked to internal structural dynamics. The task of monitoring this delicate yet important property inherent to proteins requires new approaches. Recently, theoretical methods, based on molecular dynamics (MD) and steered molecular dynamics (SMD) simulations, have been applied to explore fundamental aspects of molecular mechanisms that could not be answered or addressed experimentally (Isralewitz et al. 2001; Jensen et al. 2002; Gullingsrud and Schulten 2003). On the other hand, a promising experimental technique for studying protein structural dynamics in solution is the use of amide hydrogen exchange (HX) monitored by mass spectrometry. Pioneering work by Linderstrøm-Lang and coworkers and later on by Englander established the theoretical background for correlating structural dynamics of proteins (i.e., local structural stability) to observed HX rates (Hvidt and Nielsen 1966; Englander et al. 1996). HX monitored by mass spectrometry enables sensitive detection of protein structural dynamics at medium resolution (Hoofnagle et al. 2003; Wales and Engen 2006). Recently, we used this method to compare the structural dynamics of zymogen FVII and FVIIa and to pinpoint regions throughout the FVIIa structure that were affected upon TF binding (Rand et al. 2006). Results showed that the solution structures of zymogen FVII and FVIIa are highly similar (identical HX rates) and that FVIIa apparently retains the zymogen structure despite endoproteolysis at Arg152–Ile153{15–16}. The allosteric activation induced by TF was shown to comprise a dramatic stabilization of the protease domain, in particular involving the TF-binding helix (residues 306–312{164–170}), the 170-loop (residues 313–322{170A-175}), the 94-shunt (residues 231–239{91–99}), activation loop 3 (residues 365–374{216–225}), activation loop 1 (residues 285–294{142–152}), β-strands A2 (residues 279–283{136–140}), and B2 (residues 298–303{156–161}) of the activation pocket, the N-terminal tail (residues 153–169{16–32}), and the Ca2+-binding loop (residues 208–220{68–80}), spanning across one face of the protease domain (Fig. 1A). However, the level of resolution of the HX data did not allow for a precise identification of individual residues involved in the allosteric activation mechanism.

Figure Figure 1..

(A) The twin β-barrel structure of the protease domain of FVIIa from two orientations. The following regions are shown in color: (red) the TF-binding helix and the 170-loop; (yellow) the catalytic triad; (black) the covalently bound inhibitor FFR-cmk; (green) activation loops 1–3; and (blue)the N-terminal tail. The activation loops and the N-terminal tail constitute the so-called activation domain. The β-strands A2 and B2, the Ca2+-binding loop, as well as the 94-shunt are indicated. The direction from which TF interacts with FVIIa is denoted by an arrow. (B) TF-responsive regions of the FVIIa protease domain. Regions include the TF interaction region, active-site region, and the cavity for N-terminal insertion (activation pocket). The coloring scheme from A is applied with the TF-binding helix (residues 306–312{164–170}) and the adjacent 170-loop (residues 313–322{170A-175}) shown in red and activation loop 3 (residues 365–374{216–225}) in green. Activation loop 2 (residues 331–342{184A-193}) is omitted for clarity. The active-site inhibitor FFRcmk is present to illustrate the active-site region but was not included in any of the simulations. The numbers in subscript denote the chymotrypsin numbering. The structure of FVIIa was obtained from the 1DAN PDB entry (Banner et al. 1996) and is shown from the orientation in subsequent figures.

Simulated docking of protein interfaces using constraints from HX data has been performed in order to study subunit interfaces of protein kinase A at the single-residue level (Anand et al. 2003). In this study, we use a combination of MD and SMD simulations verified by HX at medium resolution to explore, at the single-residue level, the structural dynamics and potential allosteric pathways inherent to FVIIa activation. The MD simulations address dynamics in the nanosecond time scale, while HX detects dynamics at a much broader time scale ranging from milliseconds to hours. The MD and SMD simulations were set up in two systems: free FVIIa and TF-bound FVIIa. The SMD approach was applied to induce conformational changes pertinent to the latent structure of FVIIa by slowly pulling the N terminus out of the activation pocket. Based on analyses of the resulting trajectories from MD and SMD simulations, the differences in conformation and dynamics between free and TF-bound FVIIa are shown to be confined to regions of FVIIa also observed to undergo structural stabilization as judged by HX (Rand et al. 2006) and specific biochemical experiments. In particular, activation loop 3 and the 170-loop that succeeds the TF-binding helix are shown to be less dynamic upon TF binding. Further analyses identify the key switch of this structural change to be the interacting pair Leu305{163} and Phe374{225} in FVIIa whose mutual conformations are guided by the presence of TF.


Molecular dynamics simulations

To identify allosteric pathways, we aimed at detecting conformational differences in the protease domain between free and TF-bound FVIIa by means of MD. Simulations were performed using the catalytically competent form of FVIIa as it appears in the crystal structure with soluble TF (sTF) (Banner et al. 1996) as an initial model. Bound calcium ions and water molecules observed in the X-ray structure were preserved. The protein was solvated in a periodically truncated octahedron, and to obtain a physiological ionic strength of ∼0.15, sodium and chloride ions were added, resulting in system sizes of ∼45,000 atoms. All simulations were run for at least 20 nsec. Analysis of the resulting trajectories focused on segments comprising the loops of the so-called activation domain referred to here as activation loop 1 (residues 285–294{142–152}), activation loop 2 (residues 331–342{184A-193}), and activation loop 3 (residues 365–374{216–225}), which are bordered by the TF-binding region, the active-site region, and the macromolecular substrate exosite, as allosteric pathways were anticipated to pass through these segments (Fig. 1).

In order to compare the dynamics of free and TF-bound FVIIa, isotropic fluctuations were calculated from average structures within each nanosecond interval. As shown in Figure 2, fluctuations of the main-chain Cα atoms in both simulations are found to vary according to the amplitudes of vibration derived from the crystallographic B-factors of the FVIIa–sTF complex. The dynamic fluctuations of free FVIIa, however, are much more pronounced than those of TF-bound FVIIa, particularly in the activation loops (Fig. 3A,C), in good agreement with previous HX results showing a stabilization of β-strands A2 and B2 and activation loops 1 and 3 upon TF binding (Rand et al. 2006). In the simulations, the uncomplexed structure has been generated by removing sTF from the structure of the sTF–FVIIa complex (entry code: 1DAN) (Banner et al. 1996). Hence, one may suspect that the observed behavior is merely due to perturbation of an energy-minimized crystal structure. Therefore, we have also performed simulations for 20 nsec on the structure of free FVIIa by Sichler and coworkers (entry code: 1KLI) (Sichler et al. 2002). The simulations of free FVIIa based on either the 1DAN structure with TF removed or the 1KLI structure gives similar results (data not shown). In both cases, the C-terminal β-barrel of the protease domain that contains the three activation loops (see Fig. 1A) is quite dynamic compared to the neighboring N-terminal β-barrel. In both simulations, the RMSDs of this region of FVIIa fluctuate at 1.5 Å. In contrast, the C-terminal β-barrel is less dynamic in the simulation of the sTF–FVIIa complex (based on the 1DAN structure), with RMSDs comparable to those of the N-terminal β-barrel (fluctuating at 1.1 Å).

Figure Figure 2..

Conformational flexibility of FVIIa during MD simulation. Isotropic fluctuations, Δ, of backbone atoms in the protease domain of free (red dots) and TF-complexed FVIIa (black lines) are plotted versus residue number. The isotropic fluctuations were calculated from average structures within each nanosecond interval. The activation loops are specified (black arrows). The mean amplitude of vibrations, Ui, was calculated from crystallographic B-factors, Bi, from the relation Ui = Bi½/(2 × 2½π) and is plotted in green. B-factors were taken from 1DAN.PDB (Banner et al. 1996).

Figure Figure 3..

Overview of results from MD and SMD simulations. RMSDs versus time (relative to the initial structures) for the protease domain of (A,B) free FVIIa and (C,D) TF-bound FVIIa. RMSDs in black are calculated for the protease domain, while the RMSDs in red are calculated exclusively for the activation loops 1 (residues 285–294{142–152}), 2 (residues 331–342{184A-193}), and 3 (residues 365–374{216–225}). In A and C, the N terminus is in the activation pocket, while it has been extracted from the pocket in B and D during the first 1.5 nsec of the simulation using the SMD approach, described in Materials and Methods. Snapshots (from 10 nsec to 12 nsec in intervals of 0.1 nsec) of structures along trajectories for free FVIIa from (E) MD and (F) SMD: (yellow) N terminus; (purple) activation loops; and (blue) the 170-loop.

In order to induce conformational changes pertinent to the zymogen-like structure of FVIIa, the structural dynamics of FVIIa in the enzymatically competent conformation were challenged by pulling out the N terminus from the activation pocket of free and TF-bound FVIIa. N-terminal extraction was initiated after 8 nsec of MD and lasted 1.5 nsec. RMSD variations of the protease domain and the activation loops in free FVIIa during and after the extraction are shown in Figure 3B and reveal that the major changes are confined to the activation loops, which are seen to rapidly experience structural changes compared to the entire protease domain. In contrast, the presence of TF considerably stabilizes the activation loops (the RMSDs of the activation loops are comparable to the RMSDs of the protease domain as a whole) and diminishes the effect of the pulling event (Fig. 3D). The changes in dynamics along the trajectories in MD and SMD simulations are further illustrated in Figure 3, E and F, respectively, where snapshots of Cα traces have been overlaid. Particularly, the dynamics of the 170-loop and the activation loops become more mobile upon extraction of the N terminus.

It has previously been shown that the region in the vicinity of Met306{164} is involved in the allosteric regulation of FVIIa (see Fig. 1B; Dickinson et al. 1996; Persson et al. 2001c). The 170-loop succeeding the TF-binding helix is five amino acids longer in FVIIa than in any other member of the chymotrypsin family and constitutes one rim bordering the aryl binding pocket in the active-site region comprising the S3 and S4 subsites. In order to explore the dynamics of this region, the distances from Pro321{170I} and Asn322{175} in the 170-loop to Trp364{215} (a highly conserved residue anchored in the core of FVIIa) were monitored as a function of time in the MD and SMD simulations. In thrombin, Trp364{215} is known to play an important role in allostery (Johnson et al. 2005) but in the present simulations, Trp364{215} remains in a stable conformation. In the case of free FVIIa with the N terminus residing in the activation pocket, distances from Trp364{215} to Pro321{170I} as well as to Asn322{175} are found to vary dramatically (Fig. 4A). In particular, the distance from Trp364{215} to Asn322{175} fluctuates between 5 Å and 9 Å, suggesting a highly dynamic state of the 170-loop. A comparison with the corresponding simulations for TF-bound FVIIa show that the distances are virtually constant, indicating that the presence of TF stabilizes the structure in the vicinity of Asn322{175} (Fig. 4C). Distance variations following extraction of the N terminus during SMD simulations of free and TF-bound FVIIa are shown in Figure 4B and D, respectively. Interestingly, the region close to Pro321{170I} in free FVIIa responds promptly to the alterations in the activation pocket induced by N-terminal extraction as seen from the abrupt change in the distance from Pro321{170I} to Trp364{215} (Fig. 4B). The corresponding simulations for TF-bound FVIIa show that the apparent stabilizing effect bestowed by TF causes a delayed induction of conformational changes in the 170-loop, with distances between Pro321{170I} or Asn322{175} and Trp364{215} remaining constant for 6 and 4.5 nsec, respectively, after initiated extraction of the N terminus (Fig. 4D). While such a short time span is biologically irrelevant, the results of the SMD experiment do reflect a stabilizing effect of TF on FVIIa structure.

Figure Figure 4..

Structural dynamics of the 170-loop monitored by distances between side-chain atoms in the protease domain versus time for (A,B) free FVIIa and (C,D) FVIIa complexed to TF. The simulated structures were heated for 0.5 nsec prior to time t = 0. (Black) Distance from Trp364{215}CH2 to Pro321{170I}CG; (red) Trp364{215}CH2 to Asn322{175}; (green) Arg315{170C}N to Gly372{223}CO; and (blue) Ile153{16}N to Asp343{194}CG. The latter distance is measured to follow the position of the N terminus. A and C show the distances when the N terminus is residing in the activation pocket during the simulations, while in B and D, the N terminus is extracted during the first 1.5 nsec of simulations using the SMD approach. After the 1.5 nsec of SMD simulation, the simulations are continued in a conventional MD approach. (E) HX curves, fitted to a triexponential model by linear least squares regression as described in Materials and Methods, are shown of peptide 314–325{170B-178}, comprising the 170-loop in the presence (red triangles) or absence (blue triangles) of TF. (F) Representative HPLC traces of free FVIIa at 0, 18, and 90 min. HC/LC and HC*/LC* represent glycosylated and deglycosylated heavy chain/light chain, respectively. (G) Time courses for protease domain deglycosylation of 3 μM FVIIa by 50 U/μL PNGase F at pH 7.0 and 30°C were recorded for the free form (circles), in the presence of 6 μM sTF (squares), or with a covalent EGR-cmk inhibitor occupying the active-site (triangles).

Allosteric pathways identified by simulations

In order to explore possible conformational pathways relaying the allosteric signal from the TF-binding interface to the active-site region, the conformation of individual side chains in the vicinity of the TF-binding interface were monitored during the MD and SMD simulations. Interestingly, a considerable structural rearrangement is observed for the interacting side chains of Leu305{163} and Phe374{225} as demonstrated from plots of side chain χ1 and χ2 torsion angles for the two residues in free (Fig. 5A,B) and TF-bound FVIIa (Fig. 5C,D).

Figure Figure 5..

Dynamic interplay between the TF-binding region and activation loop 3. Side-chain torsion angle χ2 plotted versus χ1 every 0.5 psec for Leu305{163} and Phe374{225} of (A,B) free FVIIa and (C,D) FVIIa in complex with TF. (Black) Data extracted from MD simulations; (red) data from the SMD simulations. Asterisks indicate the χ1, χ2 angles as observed in the X-ray structure of the FVIIa–sTF complex. In A, a side chain in the MD simulations rotates after 3 nsec and 7 nsec of simulation (black arrows) to attain new conformations, while in B, a rotation of the side chain occurs after 7 nsec of simulation (black arrow). In C, an intermittent rotation occurs during MD simulations after 14.5 nsec (black arrows) that is persistent for 1 nsec. During the SMD simulation, a switch to another side-chain orientation is seen after 16 nsec and is persisting for 0.5 nsec. It should be noted that the dense cloud of data points at (−77, 180) in A and C corresponds to a rotated conformation of the isopropyl of Leu305{163} that places its methyl group in close contact with the phenyl ring of Phe374{225} as observed in the X-ray structure. In E and F, snapshots are shown of the conformations of Leu305{163} and Phe374{225} in the MD simulation after 6 nsec and 14 nsec, respectively (side chains in blue and activation loops 2 and 3 in gray). The conformation in the X-ray structure of the FVIIa–sTF complex is overlaid for comparison (side chains in purple and activation loops 2 and 3 in green). (G) HX curves, fitted to a triexponential model by linear least squares regression as described in Materials and Methods, are shown of peptides 361–369{212–221A} and 371–377{222–228} comprising activation loop 3. Data points and the HX curves of FVIIa and FVIIa–TF are displayed in diamonds with blue and red curves, respectively. The fitted parameters of peptide 371–377{222–228} are Nfast = 0.3, kfast = 20 sec−1, Nintermediate = 1.3, kintermediate = 0.1 sec−1, Nslow = 2, kslow = 5 × 10−4 sec−1 for free FVIIa; and Nfast = 0.4, kfast = 20 sec−1, Nintermediate = 1.2, kintermediate = 3 × 10−2 sec−1, Nslow = 6, kslow = 1.4 × 10−4 sec−1 for TF-bound FVIIa. Fitted parameters of peptide 361–369{212–221A} are identical for both FVIIa and TF-bound FVIIa and are not shown for clarity.

During the course of the MD simulation, the side chain of Leu305{163} is observed to pass through two distinct conformational transitions at 3 and 7 nsec, respectively, characterized by torsion angles (χ1, χ2) centered at (−75, −65), and (−160, 70) (Fig. 5A, black dots). The initial conformation is close to that observed in the X-ray structure of FVIIa–sTF (indicated by an asterisk in Fig. 5A) with torsion angles at (−77, 108). The transition at 3 nsec consists of an ∼180° rotation of Leu305's isopropyl group, resulting in a virtually identical isopropyl interaction with Phe374.

Similarly, the side chain of Phe374{225} is found to undergo a structural transition from (−60, −90) to (−150, −100) after 7 nsec (Fig. 5B, black dots). The concerted motion of the two side chains at 7 nsec involves rotations of the isopropyl group of Leu305{163} and the phenyl ring of Phe374{225}. Interestingly, the new orientation adopted by Phe374{225} induces a conformational change in activation loop 3 (Fig. 5E,F), resulting in breakage of the hydrogen bond between Gly372{223}NH and Ser333{185}CO. The two side chains thus seem to constitute a conformational switch relaying structural changes at the FVIIa–TF interface to the activation loops via the impact of Phe374{225} on activation loop 3. Of note is the direct link to the activation pocket via the interaction between Ala369{221A} and Val154{17} as well as to the S1 subsite via Cys368{220} and its disulfide partner Cys340{191}.

During the SMD simulation of free FVIIa, the conformations of Leu305{163} and Phe374{225} remain relatively constant and clustered around (−160, 70) and (−150, −100), respectively. (SMD was initiated after 8 nsec of MD simulation, and the side chains were already rotated at this time.)

Plots of the side-chain torsion angles of Leu305{163} and Phe374{225} in TF-bound FVIIa are shown in Figure 5C and D. Apart from a few transient structural fluctuations of <1 nsec in duration, Leu305{163} and Phe374{225} both remain in their initial conformations during MD and SMD simulations. Thus, the presence of TF stabilizes their side-chain conformations even when the N terminus is extracted from the activation pocket.

The simulated allosteric pathways of FVIIa are supported by HX data

We recently presented HX data on FVIIa in its free and TF-bound states and demonstrated detection of TF-induced allosteric effects throughout the structure (Rand et al. 2006). A dramatic reduction in HX rates following TF binding was observed for peptides of β-strand A2 and B2, forming part of the cavity accommodating the N terminus, and for the peptide comprising the N-terminal tail. In the X-ray structures of FVIIa, the amide nitrogen of Val299{157} in peptide 297–302{155–160} of β-strand B2 makes a hydrogen bond to the carbonyl of Lys157{20} in the N-terminal peptide of the heavy chain. Increased HX rates in B2 and the N-terminal tail in FVIIa as opposed to TF-bound FVIIa may well be a result of a breaking or solvent exposure of this hydrogen bond. Interestingly, the SMD simulations show a rotation around the main-chain φ,ψ angles of Gly156{19}, resulting in solvent exposure of the Val299{157}–Lys157{20} hydrogen bond upon N-terminal extraction. Likewise, the hydrogen bond between the carboxylate of Asp343{194} (which also interacts with the amine of the N-terminal Ile153{16}) and the Gly285{142}NH in the peptide 281–287{138–144} of β-strand A2, which is present in catalytically competent FVIIa, is broken when the N terminus is extracted from the activation pocket during simulations. The neutralization of Asp343{194} (see Materials and Methods) may, however, influence this last observation.

We also observed reduced HX rates for the TF-binding helix/170-loop (peptide 306–325{164–178}) in the active conformation of FVIIa as a consequence of TF binding or active-site inhibition (Rand et al. 2006). In order to specifically verify the conformational changes in the 170-loop observed in the simulations, we analyzed the HX of peptide 314–325{170B–178}, encompassing the 170-loop. As shown in Figure 4E, peptide 314–325{170B–178} displays significantly reduced HX upon TF binding, specifically in the fast kinetic components. These data suggest that an unstructured, fully solvent-exposed amide (Mandell et al. 1998) in the 170-loop of FVIIa forms an intramolecular interaction upon TF binding that is likely to stabilize this segment of the 170-loop. A similar stabilization of the 170-loop was observed in the simulations.

In the previous study, peptides of activation loop 3 displayed decreased HX in the active conformation following TF binding (Rand et al. 2006). These effects were localized to residues 370–377{221–228}, involving both fast and slower kinetic components. The present simulations suggest an interplay between the TF-binding helix and activation loop 3 via the interacting pair Leu305{163} and Phe374{225} in FVIIa during transmission of the allosteric signal. In order to investigate this further, we analyzed the HX data on two smaller peptides of activation loop 3 (Fig. 5G, peptide 361–369{212–221A} and peptide 371–377{222–228}). Residues 373–377{224–228} are responsible for the reduction in HX of the slow kinetic components, observed in peptide 371–377{222–228}.22 Fast kinetic components are, however, affected in peptide 361–377{212–228} (Rand et al. 2006), and thus, by subtraction, these effects can be localized to residues 370–372{221–223}. This observation suggests that an amide hydrogen in this short segment is unstructured and fully solvent exposed in free FVIIa but engages in intramolecular interactions in TF-bound FVIIa. Inspection of FVIIa structures shows that the amide of Gly372{223} participates in a hydrogen bond to the carbonyl of Ser333{185}. The amide of Val371{222} is solvent exposed, while that of Thr370{221} participates in hydrogen-bonding to an internal water network. This suggests that the hydrogen bond from Gly372{223} to Ser333{185} is the one responsible for the altered HX of the fast kinetic component. Importantly, the hydrogen bond between Ser333{185} and Gly372{223} was observed to be disrupted during rotation of Phe374{225} during MD simulations.

The affected slower kinetic components of peptide 371–377{222–228} can be localized to the segment spanning residues 373–377{224–228}. This segment includes Phe374{225}, which is engaged in a critical interaction with Leu305{163}, according to MD simulations (see Fig. 1B). The observed stabilization of this segment of activation loop 3 provides further experimental support for the implication of this region as a mediator of an allosteric response.

In the FVIIa–TF crystal structure, the carbonyl of Gly372{223} generates a hydrogen bond to main-chain nitrogen of Arg315{170C} (see Fig. 1). Among the structures of the trypsin-like enzymes, this hydrogen bond is exclusively observed in FVIIa. Interestingly, HX data from overlapping peptides (sublocalization via peptides 312–325{170–175} and 314–325{170B-175}) of the TF-binding helix/170-loop showed increased HX for free FVIIa relative to TF-bound FVIIa in the segment consisting of Ser314{170B}–Arg315{170C} (Rand et al. 2006). This could be due to the loss of a hydrogen bond between Arg315{170C} and Gly372{223} in free FVIIa as shown in Figure 1B.

Deglycosylation of Asn322{175}: Probing the dynamics of the 170-loop

FVIIa carries two N-linked glycans at Asn145 and Asn322{175} in the light chain and protease domain, respectively. Both sites can be deglycosylated by peptide N-glycanase F (PNGase F) under native conditions. Interestingly, while Asn322{175} is a much poorer substrate, we found that the rate of deglycosylation is affected by inhibitor docking in the active site, suggesting sensitivity to conformational changes in the protease domain. In order to investigate this further, we devised a biochemical experiment that uses the deglycosylation kinetics of Asn322{175} to probe the structural dynamics of the 170-loop in FVIIa. In the experiment, we measure the rate of protease domain deglycosylation under conditions known to affect the equilibrium between free and active forms of FVIIa. The reaction is followed by reversed-phase HPLC, allowing separation and quantification of glycosylated and deglycosylated protease domain species (Fig. 4F). Under the conditions chosen, free FVIIa is completely deglycosylated within 100 min (Fig. 4G). However, with a covalent L-EGR-chloromethyl ketone (EGR-cmk) inhibitor occupying the active site or when bound to sTF, almost complete protection of Asn322{175} is observed (Fig. 4G). In all cases, light chain deglycosylation had gone to completion within the first 2 min as detected by reducing SDS-PAGE analysis, excluding the possibility that sTF itself or residual EGR-cmk in the protein preparations had interfered with PNGase F function (data not shown). The data correlate with results from both MD simulations and HX experiments, as the reduced deglycosylation of FVIIa complexed to TF or an inhibitor indicates an allosterically induced stabilization of the 170-loop in the active form of FVIIa.


MD simulations are rapidly becoming a valuable tool for investigating mechanistic and functional properties of proteins and protein systems (Karplus and Kuriyan 2005). Recently, SMD simulations have yielded qualitative insights into the dynamic and kinetic processes of ligand–receptor interactions as well as of conformational changes in biomolecules (Isralewitz et al. 2001; Jensen et al. 2001). Such simulations allow direct observations at atomic resolution of protein dynamics and function in solution that are difficult to access by experimental methods. However, as the applications of molecular dynamics simulations grow, so does the need for experimental verification. Monitoring of HX rates represents a sensitive means of investigating the structural dynamics of proteins in solution. HX detected by mass spectrometry provides information on the structural dynamics of proteins at medium resolution (five to eight residues on average), and as such it can be used as an experimental supplement to molecular dynamics simulations that provide details of structural dynamics at the atomic level and nanosecond time scale.

The hypothesized activation mechanism

Substantial evidence from various biochemical experiments has established the existence of a link between the TF-binding region, the active site, and the activation domain. An allosteric signal from the TF-binding site to the active site is substantiated by enhanced (>20-fold) amidolytic activity and affinity for active-site inhibitors upon TF binding, and a signal to the activation domain by burial of the N terminus in the activation pocket (Ruf and Dickinson 1998). Covalent active-site modification of FVIIa also increases its affinity for TF, suggesting a two-way allosteric pathway and that the same conformation of FVIIa is preferred for TF binding and inhibitor interaction (Dickinson and Ruf 1997; Sørensen et al. 1997).

In this study, we use MD and SMD simulations to obtain information at the single-residue level on the mechanism of activation of FVIIa by TF using medium resolution HX data as an experimental reference frame. Further support comes from a panel of indirect biochemical experiments addressing specific allosteric effects in FVIIa upon TF binding. In good agreement with available HX data (Rand et al. 2006), MD simulations of free and TF-bound FVIIa show that the uncomplexed structure is more dynamic with increased flexibility particularly in the activation loops. Apparently, the global stabilizing effect of N-terminal insertion on FVIIa structure can be simulated largely as it occurs in solution monitored by HX. However, according to HX data, activation loop 2 does not exhibit increased mobility as predicted by simulation. This discrepancy might be due to the relative time scales of experimental and simulated dynamics, as the latter only accounts for nanosecond dynamics and some fluctuations of protein structure might not have sufficient time to equilibrate.

SMD simulations of N-terminal extraction had a drastic effect on the FVIIa structure. The activation loops and the 170-loop became more dynamic (see Fig. 3). Furthermore, the average distance between Trp364{215} and Pro321{170I} of the 170-loop increased, suggesting an increased mobility of the 170-loop and the TF-binding helix relative to the rest of the protease domain (see Fig. 4). Analysis of trajectories of free FVIIa and TF-bound FVIIa from MD simulations revealed that TF also stabilizes the conformation of the interacting pair Leu305{163} and Phe374{225} in FVIIa (see Fig. 5). In free FVIIa, the side chains undergo conformational changes that appear to be correlated to the changes in dynamics of Asn322{175} and Pro321 in the 170-loop. Importantly, these observations are supported by decreased HX of peptides of the TF-binding helix, the 170-loop, and activation loop 3 in TF-bound FVIIa. Furthermore, the changed flexibility of the 170-loop could be further verified by deglycosylation experiments that showed a reduced deglycosylation rate of FVIIa complexed to TF or an active-site inhibitor compared to that of free FVIIa. (The EGR-cmk inhibitor was chosen as it does not physically interact with the 170-loop.) In accordance with the above observations, replacement of the 170-loop in FVIIa with that of trypsin, resulting in a truncation of the 170 loop, has been shown to cause a dramatic increase in amidolytic and proteolytic activity even in the absence of sTF (Soejima et al. 2002).

The SMD simulations showed that the impact on the dynamics of the FVIIa structure upon N-terminus extraction is delayed when FVIIa is complexed to TF. This suggests that TF binding not only facilitates N-terminal insertion but also affects the conformational mobility of regions of FVIIa even in the absence of N-terminus insertion. This observation correlates with HX data on zymogen FVII, which showed that certain regions of zymogen FVII were stabilized by TF binding even in the absence of N-terminal insertion, including the 170-loop (Rand et al. 2006).

When FVIIa and TF-bound FVIIa are challenged by extraction of the N terminus from the activation pocket, TF is still capable of stabilizing the conformations of Leu305{163} and Phe374{225}. Interestingly, the FVIIaL305V and FVIIaL305I analogs had (threefold) increased amidolytic and proteolytic activity in the absence of TF and normal activity in the presence of TF (Persson et al. 2001a). In contrast, the FVIIa F374P analog (Petrovan and Ruf 2000; Persson et al. 2001a) had a slightly improved amidolytic activity (1.4-fold) in the absence of TF, while the activity in the presence of TF was reduced to one-third of normal, hinting at the important role of TF in stabilizing the conformations of Leu305{163} and Phe374{225}.

Previously it was pointed out that the combination of Leu and Phe in positions 305{163} and 374{225} is unique for FVIIa (Persson et al. 2001a). The corresponding positions are rather occupied by pairs of Leu and Pro or by Val and Phe/Tyr in other members of the chymotrypsin family. Particularly, the side chain in position 374{225} is known to play a role in allosteric activation of other proteases, for example, thrombin. The recently solved structure of sodium-free human thrombin (Johnson et al. 2005) reveals that the sodium binding loop, corresponding to residues 368–374{220–225} in FVIIa (a major part of activation loop 3), shifts 5 Å compared to the catalytically competent structure with a bound sodium ion. The movement of the loop affects the conformation of another loop containing the active-site Ser via the disulfide bond corresponding to Cys340{191}–Cys368{220} in FVIIa, resulting in blocking of the S1 and S2 pockets.

Based on these observations and the results presented in this study, we propose the interacting pair in FVIIa, Leu305{163} and Phe374{225}, as a central mediator of TF-induced allostery. TF-induced stabilization of this pair, in turn, stabilizes activation loop 3, the 170-loop, and the S1 and S3 substrate subsites as well as the activation pocket facilitating N-terminal insertion. Hence, activation loop 3 constitutes a direct link to the N terminus via the interaction between Ala369{221A} and Val154{17} as well as to the S1 subsite via Cys368{220} and its disulfide bridge partner Cys340{191}.

Materials and Methods


The recombinant soluble ectodomain of human tissue factor (sTF) comprising residues 1–219 was refolded and purified from Escherichia coli inclusion bodies as described (Freskgård et al. 1996). The concentration of sTF was determined by absorption at 280 nm using a molar extinction coefficient of 32,750 M−1 cm−1. Purification of FVIIa from the culture medium of stably transfected baby hamster kidney (BHK) cells was performed as described (Persson et al. 2001c). Recombinant Chryseobacterium menigosepticum peptide N-glycanase F (PNGase F) supplied in a glycerol-free buffer at a concentration 500 U/μL was obtained from New England Biolabs.

Simulated systems

The two structures analyzed by molecular dynamics were those of free FVIIa and TF-bound FVIIa. The models were based on the crystal structure of the FVIIa–TF complex obtained from PDB entry code 1DAN (Banner et al. 1996) in which the cocrystallized, covalently bound active-site inhibitor had been removed to prevent stabilization of the protein structures by the inhibitor. In this structure the first five amino acids, the last nine amino acids, and the cleaved loop 85–89 of TF are missing, while in FVIIa the last 11 amino acids of the light chain are missing. No attempts have been made to model the missing parts to prevent perturbation of the X-ray data. A final energy minimization was conducted using 250 steps of conjugated gradient (CONJ) minimization in CHARMm (Accelrys Inc.) using the CHARMM22 force field.

Molecular dynamics

Molecular dynamics simulations were performed using the program NAMD 2.5 (Gullingsrud and Schulten 2003) and the CHARMM22 force field for proteins (McKerell Jr. et al. 1998). The nine calcium ions embedded in the X-ray structure were preserved. The proteins were solvated (with TIP3 water molecules) in a periodic truncated octahedron with box boundaries at least 6 Å from any given protein atom. Water molecules were treated as rigid. In order to obtain a physiological ionic strength of ∼0.15, sodium and chloride ions were added. The resulting systems were composed of ∼12,000 water molecules, ∼40 sodium ions, ∼35 chloride ions, and nine calcium ions. In total, each system consisted of ∼45,000 atoms. A cutoff of 12 Å (switching function starting at 10 Å) for van der Waals interactions was applied. The particle mesh Ewald (PME) method was used to compute long-range electrostatic forces in all simulations (Essman et al. 1995). An integration time step of 1 fsec was used. Langevin dynamics was applied to enforce constant temperature (T = 300 K) conditions. The Langevin damping coefficient was set to 5 psec−1. The pressure was maintained at 1 atm using the hybrid Nose-Hoover Langevin piston method with a decay period of 100 psec and a damping time constant of 50 psec (Martyna et al. 1994; Feller et al. 1995). Each system was equilibrated in the constant number, pressure, and temperature ensemble (NpT), and the resulting models were used for further molecular dynamics and steered molecular dynamics simulations (see the following section). The systems were simulated for at least 20 nsec.

Steered molecular dynamics simulations

After 8 nsec of MD simulations, the resulting models were in parallel to MD simulations challenged by SMD simulations. The N-terminal amine of Ile153{16}, NH3+, was extracted from the activation cavity by applying a constant velocity pulling method (Isralewitz et al. 2001). In this type of simulation, the so-called SMD atom is attached to a dummy atom via a virtual spring. The dummy atom is moved at constant velocity, and the force between both is measured according to F = k[vt − (rr0) · n]n, where k is the spring constant, v is the speed of the dummy atom, t is time, r is the position of the SMD atom, r0 is the initial position of the SMD atom, and n is the pulling direction. A pulling potential was derived from the force equation and incorporated into the force field. In the present simulations, the spring constant was 2.16 kcal/mol·Å2 corresponding to 159 pN/Å, the velocity was 0.005 Å/psec, and the pulling direction was selected to be in the direction of the vector from the Cβ atom of Ile153{16} to the Cβ atom of Leu135, with the SMD atom chosen as the amine nitrogen of Ile153{16}. This direction is optimal in the sense that it allows the SMD atom to move out of the cavity without major collisions with surrounding atoms. In order to prevent the protein from translating during an SMD simulation, certain atoms are often restrained by harmonic constraints that may introduce artifacts during the dynamics. In the present study, however, the salt bridge between the amine of Ile153{16} and the carboxylate of Asp343{194} was neutralized by scaling the charges of the amine and carboxylate to zero. (Although the charges are scaled, the hydrogen-bonding capacity still persists because of the quite strong dipoles.) Simulation experiments were performed to explore the impact of SMD simulations on the atomic interactions at the interface between TF and FVIIa. When the salt bridge was intact (i.e., side chains charged), a translation of the complex in the pulling direction was observed, which influenced the interface. If the side chains were neutralized, we observed low impact on the interface. Hence, the charge of the amine nitrogen of the N-terminal Ile153{16} was −0.46, the charge of each of its hydrogens were 0.10, the CA atom's charge was 0.06, and that of its hydrogen was 0.10, the charges of CG, OD1, and OD2 of Asp343{194} were 0.62, −0.26, and −0.26, respectively, while the charges of the CB atom and that of its hydrogens were −0.28 and 0.09, respectively. In this way, the neighborhood of the activation pocket was only marginally perturbed during the pulling.

Analysis tools

In order to monitor structural changes of the proteins, their root-mean-square deviation (RMSD) was computed using CHARMm and coordinates saved every 0.5 psec. The initial conformations were used as reference points, and only positions of the protein backbone atoms were compared. Various distances between atoms were also calculated using CHARMm. Likewise, hydrogen bonds were calculated in CHARMm using standard criteria for hydrogen bonds (Kabsch and Sander 1983). Further molecular analyses and figure preparations were performed within the framework of the molecular modeling package Quanta. Plots were prepared using xmgrace (http://plasma-gate.weizmann.ac.il/Grace).

HX reactions of FVIIa monitored by mass spectrometry

HX reactions and mass analysis were performed as described earlier (Rand et al. 2006).

Several overlapping peptides in specific regions of FVIIa were identified and analyzed in this study to localize individual exchanging amide hydrogens in the corresponding peptide segments. Localization of exchanging amides in both N- and C-terminally overlapping peptides was performed by subtraction, assuming complete back-exchange of the N-terminal amide hydrogen in each peptide. HX time-course plots of FVIIa peptides were fitted by nonlinear least squares regression to a triexponential equation as described (Rand et al. 2006).

Deglycosylation of FVIIa

PNGase F-catalyzed cleavage of the N-glycosidic linkage to Asn322{175} in the protease domain was monitored by HPLC for FVIIa in the free form, in complex with sTF, and with a covalently bound GluGlyArg-chloromethyl ketone (EGR-cmk) inhibitor in the active site. Active-site-inhibited FVIIa was prepared by overnight incubation at room temperature in the presence of a 20-fold molar excess of EGR-cmk (Bachem) in 50 mM HEPES, 100 mM NaCl, 10 mM CaCl2, 0.01% Tween 80 (pH 7.0). Excess inhibitor was subsequently removed by gel filtration of 50-μL aliquots on Pro-Spin CS-800 columns (Princeton Separations) into the same buffer containing 0.01% Tween 80. Protein concentrations were determined by absorbance measurements on a NanoDrop ND-1000 spectrophotometer at 280 nm using an extinction coefficient of 61,900 M−1 cm−1. Based on differences in the retention times of free and active-site labeled protease domain by reversed-phase HPLC (see below), it could be concluded that the inhibition reaction had gone to near completion. This was further supported by a residual amidolytic activity of <1% as measured by hydrolysis of the chromogenic substrate S-2288 (Chromogenix).

The deglycosylation reactions were carried out at 30°C in a total volume of 100 μL of buffer (50 mM HEPES, 100 mM NaCl, 10 mM CaCl2, 0.01% Tween 80 at pH 7.0) containing 3 μM active or EGR-cmk-inhibited FVIIa with or without 6 μM sTF. Reactions were initiated by addition of PNGase F to a final concentration of 50 U/μL. Aliquots (15 μL) were taken out at timed intervals and acid-quenched by eightfold dilution into a solution consisting of 70 mM tris(2-carboxyethyl)phosphine (TCEP) and 10% formic acid and then placed on ice. Prior to HPLC analysis, all samples were heated to 70°C for 10 min to ensure complete disulfide reduction and separation of heavy and light chains.

The glycosylated and unglycosylated forms of the protease domain were separated and quantified by reversed-phase HPLC on a C4 column (Vydac, 300 Å, particle size 5 μm, 4.6 × 250 mm) maintained at 30°C. Mobile phases consisted of 0.1% TFA in water (solvent A) and 0.085% TFA in acetonitrile (solvent B). Following injection of 25 μL of sample, the system was run isocratically at 25% solvent B for 4 min followed by a linear gradient from 25%–46% B for 10 min during which the light chain, sTF, and PNGase F eluted. Finally, a linear gradient from 46% to 52% B for 21 min was applied to separate protease domain species. The flow rate was 1.5 mL/min, and peaks were detected by fluorescence using excitation and emission wavelengths of 280 and 348 nm, respectively. The relative amounts of glycosylated and unglycosylated protease domain were determined by peak integration using Millenium32 v4.00 software (Waters). For those reactions where significant protease domain deglycosylation had taken place, time courses were fitted to a monoexponential function to obtain apparent rate constants for the reactions.


We thank Anette Østergaard for expert technical assistance.


  1. 1

    Chymotrypsin numbering is denoted in subscript.

  2. 2

    The N-terminal amide has intrinsic exchange rates several orders of magnitude higher than amides of the rest of the peptide and is subsequently back-exchanged before analysis even under quench conditions (Bai et al. 1993).