Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Apr 9.
Published in final edited form as: J Phys Chem B. 2026 Mar 27;130(17):4528–4537. doi: 10.1021/acs.jpcb.6c00315

Probing Solution Dynamics of Tissue Factor Using Molecular Dynamics Simulations Guided by NMR Chemical Shifts

Muyun Lihan †,‡, Shashank Pant †,, Adedolapo M Ojoawo §,‖, Chad M Rienstra ⊥,#, Emad Tajkhorshid †
PMCID: PMC13058639  NIHMSID: NIHMS2162713  PMID: 41894586

Abstract

Structure and dynamics of proteins are key to understanding their roles in biological systems and provide a framework for rational development of novel therapeutics. Here, we combine NMR chemical shifts (CSs), X-ray crystal structures, and molecular dynamics (MD) simulations to characterize the extracellular domain of human tissue factor, i.e., soluble tissue factor (sTF), a protein that is involved in the initiation of the blood clotting process by forming a complex with the coagulation factor VIIa (fVIIa). Starting with the X-ray structures, solution NMR CSs were incorporated as restraints in CS-guided MD simulations to obtain structures in agreement with the NMR solution data of the protein. Our results reveal a dynamic ensemble of configurations in a loop that is key to sTF interaction with fVIIa. Key residues have been identified in the fVIIa-binding loop with divergent backbone and/or side-chain configurations to account for the loop dynamics. We demonstrate that the resulting structural ensemble from incorporation of solution NMR CSs provides a better description of sTF dynamics in solution. The integrated approach used in this study can be applied to provide a better molecular guide for therapeutics that specifically target sTF.

Graphical Abstract

graphic file with name nihms-2162713-f0001.jpg

Introduction

Knowledge of a protein’s structure and dynamics is key to fully understanding its role in biological processes. X-ray crystallography, cryo-electron microscopy (cryo-EM), and nuclear magnetic resonance (NMR) spectroscopy are most commonly used to obtain high-resolution structural information of biomolecules at the atomic level. X-ray crystallography is a well-established method and accounts for ~90% of the structures deposited in the RCSB PDB database (Protein Data Bank, 2025). When a good crystal can be obtained, X-ray crystallography is advantageous because it can provide structural information at atomic resolution regardless of the size. A major bottleneck, however, is the growth of high-quality crystals. While many X-ray structures have been reported for a variety of systems, crystallization remains challenging when working with structurally flexible regions, intrinsically disordered proteins, or membrane-embedded proteins.1,2 Another drawback of crystallography, as the name suggests, relates to the fact that the macromolecule of interest is confined/forced to/into a crystal environment, where it may or may not be able to undergo its natural motion(s), which are sometimes necessary for the biological function.

NMR, on the other hand, is able to provide atomic-level information for biomolecules, both in solution and in solid state.3–6 Solid-state NMR (ssNMR) has been useful in characterizing membrane proteins,7–9 and insoluble aggregates such as fibrils.10–12 However, structure determination often requires complete or near-complete resonance assignments, as well as a large amount of unambiguous distance restraints. Measurements and analysis of NMR restraints used for structural calculations are usually labor-intensive and time-consuming.1 X-ray crystallography provides information mainly about the global conformation of a biomolecule, while NMR is best used to characterize local effects on the conformation.13–15 Thus, NMR and X-ray crystallography are considered complementary sources of structural information.1,16 In the presence of sub-optimal NMR and/or crystallography data, the combination of NMR and X-ray diffraction data may be useful to obtain more precise structural models of proteins.2

NMR reports on the positions of each atom in a molecule primarily in terms of chemical shifts (CSs).17 CSs are highly sensitive probes that are readily observable in NMR both in solution and in solid state. They report on the local environment of nuclei, thereby providing insights into the structure and dynamics of a molecule.18

A number of studies have exemplified the power of combining NMR and X-ray data in an integrative approach for structure calculation and refinement.19–21 The notion of using only CSs as the source of experimental information for NMR structure determination has been adopted by computational approaches to obtain native protein structures.22–24 These studies aim to overcome the need for acquiring and analyzing distance restraints typically obtained from nuclear Overhauser effects (NOEs) in solution NMR and from dipolar couplings in ssNMR. Computational approaches including CamShift,25 CHESIRE,19 or CS-Rosetta26 have demonstrated that backbone CSs may be sufficient to obtain reliable structures of proteins of up to 130 residues. Integration of these computational approaches with X-ray structures would provide a more efficient tool for structure determination and/or assist in identifying structural differences of proteins in different environments.

CamShift,25 the method used in this study, expresses the conformational dependencies of backbone CSs as a polynomial function of interatomic distances. In this method, the CS calculations are fast and readily differentiable with respect to atomic positions of a protein, an advantage over other semi-empirical methods for calculating CSs.27–29 This allows CSs to be incorporated directly as restraints in the conformational search carried out, e.g., in MD simulations.

Here, we employ an integrative CS-guided MD approach using the CamShift method25 on the extracellular soluble domain of human tissue factor (TF) in an attempt to refine the poorly resolved regions in the X-ray structures and provide a better representation of the protein in solution.

TF plays an important role both in normal hemostasis and in the pathogenesis of blood coagulation disorders such as thrombosis and hemophilia.30 TF is a membrane-anchored protein with 263 amino acids, consisting of an extracellular soluble domain (sTF, residues 1–219), a transmembrane domain (residues 220–242), and a cytoplasmic domain (residues 243–263). TF initiates blood coagulation by forming a complex with its substrate, factor VIIa (fVIIa), through its fVIIa-binding loop (residues 80–92). The extracellular domain, sTF, has been characterized by a few X-ray structures.31–33 However, these structures miss clear electron densities in some functionally relevant regions of sTF, mostly in the disordered loops and β-turns that interact with fVIIa or the membrane. These loop residues are important for sTF interactions with substrate and membrane according to mutagenesis studies.31,34 Detailed structural and dynamic information on sTF in solution, at the atomic scale, is needed for understanding the nature of interactions with its binding partners during the blood coagulation cascade.

Materials and Methods

NMR experiments

Protein Expression and Purification.

Uniformly-[13C, 15N]-labeled sTF was expressed and purified as described in detail in a previous study.35 Briefly, T7 Express Escherichia coli competent cells (New England BioLabs, Inc., Ipswich, MA) were transformed with the plasmid pJH677, which expresses sTF (amino acids 2–219) with a C-terminal 6×His-tag. Transformed cells were grown on a Luria Broth (LB) agar plate containing kanamycin. A starter culture was grown in modified Studier media MDG until the OD600 reached about 7. The cells were then harvested and resuspended in modified Studier MBG medium containing 13C, 15N BioExpress (Cambridge Isotopes Laboratories, Inc., Andover, MA). Expression of sTF was induced with 100 μM isopropyl β-D-1-thiogalactopyranoside (IPTG). Cells were harvested around 15 h later by centrifugation. sTF was then released by osmotic shock, and cell debris was pelleted by centrifugation. The supernatant was incubated with Q-Sepharose® Fast Flow ion exchange resin to remove E. coli proteins, and the supernatant was collected via vacuum filtration (Sigma-Aldrich, St. Louis, MO, USA). Ni2+ affinity chromatography was then used to purify sTF from the supernatant (GE Healthcare, Piscataway, NJ, USA). The protein yield was about 80 mg/L growth medium.

NMR data collection

The NMR samples of [13C, 15N]-labeled G90A (0.2 mM) and S85T (0.2 mM) sTF mutants were prepared in a solvent containing 10% D2O (v/v), 50mM sodium phosphate buffer, pH 6.5, 50mM NaCl and contained 0.2mM [13C, 15N]-labeled sTF in 50 mM phosphate buffer, pH 6.5, 50 mM NaCl and 0.1% (w/v) 2,2-dimethyl2-silapentanesulfonic acid (DSS) as internal standard. All NMR experiments were performed at 308K on Varian VNMRS 800 MHz (Agilent Technologies, Walnut Creek, CA), Bruker Avance III 600 MHz, and Bruker Avance III 900 MHz spectrometers. A 1H-15N TROSY36 was also collected on 15N labeled wild-type (WT) sTF for comparison with mutants on a Varian VNMRS 750 MHz. Backbone resonance assignments of mutants (1H, 15N, 13Cα, Hα, and 13CO) were performed using double and non-uniformly sampled (NUS) triple resonance experiments of 1H-15N TROSY,36 HAcacoNH, TROSY-based 3D HNCA, HNCOCA, HNCO,19,37 3D NOESY 15N-HSQC, 3D HCCH TOCSY. All data sets were processed using NMRPipe with zero filling, Lorentzian-to-Gaussian apodization and/or sine bells to optimize resolution.38 The 1H, 15N, 13C CSs were referenced to the internal standard, DSS, using the empirical coefficients (http://www.bmrb.wisc.edu/ref_info/cshift.shtml). Peak picking and assignment were performed using NMRFAM-Sparky.39 Secondary structure analysis was performed using TALOS+.40 The 1H-15N TROSY signals were collected after each triple-resonance experiment to monitor changes in samples that may be due to aggregation. The 1H and 15N CSs of G90A and S85T sTF mutants were compared to WT by the weighted averaged CSs (Δδ) obtained using the equation: Δδ = [(ΔH)2 + (0.14 × ΔN)2]1/2. A cut-off value of 0.038 ppm (two standard deviations above the average value) was used in the CS perturbation analysis.

Molecular dynamics (MD) protocols

The NMR structure calculation protocols involved MD simulations to obtain a low-energy structure that satisfied as much experimental restraints as possible. In this study, the X-ray structure of sTF was refined by performing MD simulations which incorporated NMR CSs as additional biases. The CS-guided MD simulations employed CamShift25 which converts NMR CSs as bias potentials that depend on interatomic distances, as implemented in PLUMED.41 All MD simulations were carried out using NAMD42,43 with the CHARMM36 force field44 and the TIP3P water model.45

Equilibrium MD simulations

We started from three available sTF crystal structures retrieved from Protein Data Bank, to have diverse initial conditions. The three crystal structures included an apo structure (PDB: 1BOY), one in complex with fVIIa (PDB: 6R2W), and an engineered mutant with the fVIIa-binding loop missing (PDB: 2HFT), covering three distinct configurations of the fVIIa-binding loop. The engineered mutant (PDB: 2HFT) was mutated back to WT with the missing loops constructed into the model using CHARMM-GUI.46 Two sTF mutant systems, G90A and S85T, were also generated by mutating the apo structure, 1BOY.

For each system, a C-terminal carboxylate cap, an N-terminal ammonium cap, and hydrogen atoms were added using the PSFGEN plugin of the visual molecular dynamics (VMD).47 The initial protonation states of titratable residues were predicted using the PROPKA webserver.48 Next, the entire protein was solvated with TIP3P water using the SOLVATE plugin of VMD. The solvated system was then neutralized with Na+ and Cl− ions (0.15 M NaCl) using the AUTOINONIZE plugin.

The constructed system was then subjected to energy minimization for 1,000 steps followed by a 5-ns simulation with Cα atoms harmonically restrained to their initial positions with a force constant of k = 1.0 kcal/mol/Å2 and a 5-ns restraint-free simulation. Long-range electrostatic forces were calculated using the particle mesh Ewald (PME) method49 with a grid spacing of 1.0 Å. Langevin dynamics and Langevin piston Nosé-Hoover methods were used to maintain constant temperature at 310K and constant pressure at 1 atm,50 respectively. The SETTLE algorithm was used to constrain all bonded hydrogens.51 All the non-bonded forces were calculated with a cutoff of 12 Å and a switching distance of 10 Å. A time-step of 2 fs was used with an update of vdW interactions every 2 fs and electrostatics ones every 4 fs.

CS-guided MD simulations

Starting with the equilibrated systems, CS-guided MD simulations were performed using solution NMR CSs derived from experimental data as the target to guide the atomic coordinates and refine the sTF structures. The solution NMR CSs of WT sTF52 were taken from BioMagResBank53 (BMRB code: 16,83852), and the shifts of the two sTF mutants were taken from the experiments described above. At each step of the CS-guided simulations, CSs for Cα, Cβ, C, H, and N were calculated for the instantaneous conformation using CamShift and the difference between the predicted and measured CS was used to create a linear biasing potential centered at 0 with a slope of 2. Each system was simulated for 8 ns to achieve convergence.

Validation

In order to validate our CS-guided simulations, additional validation of the solution structure was conducted using SHIFTX2 CS prediction54 to verify the reliability of the final refined structure. SHIFTX2 provides better accuracy than CamShift in terms of CS prediction due to its usage of ensemble machine learning techniques.54 Using SHIFTX2, the backbone and side-chain carbon CSs were predicted with the temperature set to 310K and the pH set to 7.0. The SHIFTY option was turned off to only consider effects from structural changes when making CS prediction. The root mean squared error (RMSE) was calculated using the predicted values and experimentally obtained NMR CSs as

RMSE=1N∑iδi,MD−δi,NMR2 (1)

where δi,MD − δi,NMR is the difference between the predicted and experimental shifts for residue i, and N is the total number of residues.

Results and discussion

Solution NMR CS-guided refinement of sTF

We first carried out CS-guided MD simulations using solution NMR backbone CSs (Cα, Cβ, CO, and N) (BMRB code: 1683852) on three distinct crystal structures. The purpose of using three different initial structures was to eliminate systematic bias from different crystallization conditions. The three selected x-ray structures included one in complex with fVIIa (PDB: 6R2W), one apo conformation (PDB: 1BOY), and one with missing loops modeled (PDB: 2HFT) (Fig. 1).

Figure 1: Comparison of sTF structures in CS-guided MD simulations.

Figure 1:

Open in a new tab

Top: three initial crystal structures of sTF with distinct fVIIa-binding loops (highlighted in the boxes); 6R2W (blue): in complex with fVIIa (shown in white); 1BOY (red): an apo structure; and 2HFT (green): with the missing loop modeled. Bottom: Snapshots taken from CS-guided MD trajectories from 0 (white) to 8 ns (colored) at 1-ns intervals.

These three different structures represent a frequent scenario in which one uses multiple crystal structures to refine NMR structures. Due to the different crystallization conditions, the three structures show distinct configurations, particularly for the fVIIa-binding loop, although they share a common overall fold. One of the goals of CS-guided MD simulations was to refine the fVIIa-binding loop based on the solution NMR backbone CSs so that we could obtain the loop configuration in the solution state.

From CS-guided MD, we observed that the overall fold of sTF remained almost identical to the initial conditions but the fVIIa-binding loop displayed a wide spectrum of configurations in all three systems (Fig. 1). Throughout the CS-guided MD, the β-sheet parts of sTF stayed very close to the initial crystal structures with only slight fluctuation, whereas the fVIIa-binding loop deviated from the initial loop configuration and never seemed to converge to one conformation.

In order to test whether our CS-guided MD results captured solution conformations consistent with solution NMR CSs, we then calculated all Cα CSs using SHIFTX254 for structures taken from the CS-guided MD trajectories. We calculated the RMSE of Cα CSs both for the entire protein and for the fVIIa-binding loop only (Fig. 2). Compared to the initial conformations, the sTF structures were improved significantly both globally and locally.

Figure 2: Improvement of the overall sTF structure and the fVIIa-binding loop by CS-guided MD simulations.

Figure 2:

Open in a new tab

RMSE was calculated every 0.01 ns and averaged over a 1-ns smoothing window to account for structure ensembles. In comparison to the initial structures (dashed lines), RMSE values in the 6R2W (blue) and 1BOY (red) systems were significantly reduced throughout CS-guided MD simulations (solid lines) both for the entire structure and for the fVIIa binding loop, whereas RMSE in 2HFT (green) showed only slight improvement.

Except for 2HFT, in which we modeled the fVIIa-binding loop, the other two systems adopted loop configurations in much better agreement with solution NMR data. It was expected that 6R2W, in which sTF is in complex with fVIIa, would adopt a different loop configuration that is conformationally more restrained owing to its interactions with fVIIa. The apo conformation in 1BOY was likely due to crystal packing contact and thus showed a loop configuration not captured in solution. Overall, these results supported the notion that the fVIIa-binding loop of sTF preferred to adopt configurations distinct from those resolved in the crystal structures and that our CS-guided MD provided refined loop configurations in better agreement with experimental solution data.

CS-guided MD captures a dynamic ensemble of fVIIa-binding loop

To further examine whether the fVIIa-binding loop configurations obtained from CS-guided MD simulations represent sTF structures in solution, we analyzed CS differences at the residue level.

As shown in Fig. 3, the improvement of each CS was almost instant after the start of the CS-guided MD simulations, which was also reflected in the loop RMSE discussed previously (Fig. 2). Improved CSs for each residue could be observed when we compared the CSs at the start and the end of the CS-guided MD simulations (Table S1). We note that since CS is an ensemble parameter, using only one static snapshot of the structure cannot capture its average behavior. However, with either the averaged RMSE or the averaged shift difference calculated per snapshot, we observed improved CSs and reduced CS difference. The small CS difference throughout the CS-guided MD simulations between the experimental data and the predicted SHIFTX2 values indicated that the variable loop configurations among all three systems showed convergent performance and could be gathered to represent the sTF structure ensemble in solution.

Figure 3: Convergence of CSs in the fVIIa-binding loop.

Figure 3:

Open in a new tab

The difference in the Cα CSs for each residue in the fVIIa-binding loop was calculated using CSs predicted by SHIFTX2 and experimental NMR data for structures taken from the MD simulations every 0.01 ns. Plots show converged agreement with solution NMR data in the CS-guided MD simulations among all three systems (6R2W labeled as complex; 1BOY labeled as apo; 2HFT labeled as modeled).

Next, we analyzed the fVIIa-binding loop configurations by their Cα root mean squared deviation (RMSD) to 6R2W and 1BOY, respectively, and used the two RMSDs to define a configuration space for the loop (Fig. 4). Interestingly, despite the small RMSE observed previously, the fVIIa-binding loop showed a large variation of conformations with RMSDs up to 15 Å. An overlap of the loop’s configurations between the three simulations can also be observed. Notably, the loop modeled in 2HFT exhibited the largest overlap with either the loop in the complex (6R2W) or with the one in the apo form (1BOY). We speculate that the overlap originates from the free fluctuations and movement of the loop in these simulations: on one hand, during the modeling of the missing loops in 2HFT, the constructed fVIIa-binding loop was free of artificial contacts such as crystal contacts or the complex contact. On the other hand, when simulated in solution with CS restraints, the loops were able to explore more native configurations which could resemble the constructed loop in 2HFT. In general, the ensemble dynamics of the fVIIa-binding loop in CS-guided MD provide a more accurate description of sTF in solution.

Figure 4: FVIIa-binding loop configuration space.

Figure 4:

Open in a new tab

Left: Plot shows Cα RMSD of the fVIIa-binding loop with reference to 6R2W and 1BOY in the CS-guided MD simulations. Right: The fVIIa-binding loop configuration space defined by the Cα RMSD with reference to 6R2W or 1BOY. Plot shows overlapping distributions in the configuration space of the fVIIa-binding loop in the CS-guided MD simulations started from distinct structures (6R2W shown in blue; 1BOY in red; 2HFT in green). The Cα RMSD of the fVIIa-binding loop (residues 80–92) was calculated following alignment of the entire protein, such that the reported RMSD values reflect deviations of the loop relative to the global protein.

NMR results on fVIIa-binding loop mutants, S85T and G90A

To validate the fVIIa-binding loop configurations and the dynamic ensemble described above, we performed NMR experiments on two mutants of sTF, S85T and G90A, in which perturbations are introduced to different parts of this loop. Using the enhanced manual protein CS assignment (Transfer and Simulate Assignment tool in NMRFAM-Sparky39), a first round of assignments was performed with the aid of the available assignments of WT sTF from BMRB.52 Manual assignment with the aid of 3D NOESY 15N-HSQC and HCCH TOSCY spectra was further used to resolve ambiguity and to assign the Hα CSs as well as residues perturbed due to the mutations. The CS dispersion in the 1H-15N TROSY of the mutants compared to the WT are very close, indicating similar folds in the proteins (Fig. S1). Since the mutants were not deuterated, obtaining Cβ information was challenging for a protein of this size.

Residues I22, S16, L23, K28, C49, F50, and I63, which had missing assignments in WT sTF52 were unambiguously assigned in both mutants. The 1H, 15N of residues L59, T60; CO of V36, K46, T70, A73, R74, G81, and E105 were incorrectly assigned in WT and have been re-assigned unambiguously in both mutants. Majority of unassigned signals are either prolines or next to proline residues. Unassigned non-proline residues are located in terminal regions of the protein. Overall, backbone assignments were obtained for 96.7% of 15N and 1HN atoms, 98.7% of Cα atoms, 90.1% of C atoms, and 87% of Hα atoms.

All of the 1H and 15N CSs observed in WT sTF are present in G90A and S85T mutants. Secondary structure analysis using TALOS+40 suggests that the general fold of the mutants are similar to the WT crystal structure, 1BOY (Fig. S2). However, some differences in predicted secondary structure are observed in the loop regions where G90A or S85T mutations are located. Residues N95 to F100 were predicted to have loop conformations contrary to the β-sheet conformation observed in the crystal structure (Fig. S2). We identified 14 residues (A8, A9, Y10, N31, A73, A80, E84, S85, A89, E91, Y94, N96, E99, and A191) that showed significant deviations from the cutoff value (Fig. 5) when 1H and 15N CSs of residues in G90A and WT sTF are compared, based on their weighted average CS difference.

Figure 5: Chemical Shift (CS) comparison between WT and S85T/G90A mutants.

Figure 5:

Open in a new tab

CS differences (Δδ) were calculated using the equation: Δδ = [(ΔH)2 + (0.14 × ΔN)2]1/2. A cut-off of 0.023 ppm (1 standard deviation from the average CS differences) was used. Residues with significant deviation in the fVIIa-binding loop are labeled. Top: S85T−WT. Bottom: G90A−WT. Asterisks indicate residues not included in the analysis due to lack of assignment in either the WT or the mutant.

Some of these residues near the mutation site in the loop P79-P92 showed marked CS differences, which are likely due to changes in conformation attributed to the glycine to alanine substitution at position 90. In the S85T mutant, we identified 13 residues (A8, A9, W14, V33, Y34, A73, E84, T86, G87, N96, E99, Q190, and A191) that show notable deviations from the cutoff value (Fig. 5) when 1H and 15N CSs were compared to WT sTF. The residues that showed distinct CS differences in S85T are also near the mutation sites (except V33, Y34, E99, and residues in the N & C-terminal ends of sTF), thus are most likely due to the conformational variation attributed to the serine to threonine substitution at position 85. To summarize, both mutants showed CS differences for residues near the mutation site, which could result directly from the environment neighboring them and also indirectly from the overall change in the loop dynamics.

Dynamic hotspots

Using CS-guided MD simulations, we refined the solution NMR structures of WT and the two mutants, S85T and G90A, using their respective CSs. To understand how each residue in the fVIIa-binding loop contributes to the ensemble dynamics, we carried out dihedral angle analysis for all backbone ϕ and ψ angles, as well as the side chain dihedral angles χ1 and χ2 for N82, E84, and E91 in the loop (Fig. 6). Interestingly, although the overall loop displayed large variations in its configurations, only a few residues showed divergent dihedral angle distributions.

Figure 6: Dihedral angle analysis of the fVIIa-binding loop.

Figure 6:

Open in a new tab

Plots show probability density function (solid lines) of backbone dihedral angles, i.e., ϕ and ψ, or side chain dihedral angles, i.e., χ1 and χ2, for each residue in the fVIIa-binding loop. The two mutants, S85T and G90A, are shown with dihedral angles of their respective mutated residue, i.e., T85 and A90. The values in the initial structures are shown as dashed vertical lines. Dynamic hotspot residues, e.g., N82, S85, G87, S88, G90, and E91 can be identified with divergent dihedral angle distributions among the three WT and two mutant systems (6R2W in blue; 1BOY in red; 2HFT in green; S85T in cyan; and G90A in yellow).

Among the loop residues, the backbone dihedral angles of A80, G81, V83, E84, T86, A89, and P92 and the side-chain dihedral angles of E84 showed almost identical distributions among the three WT and two mutant systems. Residue A80 and P92 showed converged distributions to the angles observed in 1BOY, likely because both residues are at the end of the loop and connected to the rest of sTF. CS-guided MD started with 6R2W, however, showed an outlier distribution for the backbone dihedral angles of V83, E84, T86, and A89, suggesting that the loop could be trapped in an energy minimum close to the crystal structure 6R2W when in complex with fVIIa. The outlier behavior of G81 backbone dihedral angles for the 2HFT system could be explained by the inherent flexibility of glycine residues, which also led to the persistence of its CS differences in the modeled loop shown in Fig. 3.

In contrast, residues N82, S85, G87, S88, G90, and E91 showed divergent distributions among WT and mutant systems. The ϕ angle of N82 is mostly centered around the apo conformations, but simulations started with 6R2W showed a broader distribution. The ψ angle of N82 showed an outlier distribution from 2HFT and S85T, which could indicate a local minimum. The varied backbone dihedral angles of N82 could also be coupled to the divergent χ angle of N82, suggesting high flexibility of the N82 side chain. Residue S85 can also be identified as a flexible residue as indicated by its high variations in backbone dihedral angle distributions. The mutant S85T showed a backbone configuration more similar to 6R2W, but the ϕ angle of T85 displayed a double-peak distribution. In Fig. 3, V83 showed the persistence of CS differences in the modeled loop (2HFT), but CS-guided MD started with 2HFT here did not show a much different distribution for its backbone dihedral angles. This can likely be attributed to its location between a few dynamic hotspot residues in the loop (e.g., N82 and S85), whose divergent dihedral angle distributions exert conflicting local structural influences on V83. This coupling between neighboring residues limits the ability of the CS restraints to fully converge V83. In addition to the hotspot involving N82 and S85, the bond between residues G87 and S88 could be another flexible hinge, as evidenced by the divergent distributions for the ψ angle of G87 and the ϕ angle of S88. Residue G90 showed inherent backbone flexibility as a glycine residue, and the distribution of A90 backbone dihedral angles became much more concentrated in the G90A mutant. The last residue E91, although showing a convergent backbone configuration, manifested its side chain flexibility across three populations of both χ1 and χ2 angles. To conclude, we believe that residues N82, S85, G87, S88, G90, and E91 act as dynamic hotspots responsible for the ensemble dynamics observed in the sTF fVIIa-binding loop.

Concluding remarks

Capturing protein structures under physiological conditions is essential for understanding its function and for guiding structure-based drug design. Here, we developed solution representations of sTF structure by incorporating solution NMR CSs in CS-guided MD simulations to refine the starting X-ray structures. We tested three different initial conditions, i.e., crystal structures, that represent common scenarios in practice for NMR structure refinement. From our CS-guided MD simulations, we were able to generate an ensemble of sTF structures that were in better agreement with NMR CSs than initial crystal structures. The ensemble structures mimic a more realistic physiological condition that is not affected by artifacts present in a crystal environment such as crystal packing effects. We also note that NMR CS is a time-averaged property and should be calculated using structure ensembles rather than a single static structure. By CS-guided MD, the snapshots we generated from trajectories will provide a more accurate description of the sTF dynamics in solution.

The CS differences between the refined NMR structures and the crystal structures reveal that the sTF fVIIa-binding loop shows significant backbone changes in solution. Despite better agreement with NMR CSs, the sTF fVIIa-binding loop displays a dynamic ensemble of configurations indicating its intrinsic flexibility. This flexibility could be critical in the recognition and interaction with fVIIa and thus important for sTF function. Compared with crystal structures, the loop ensembles captured in the solution NMR structure demonstrate a less biased view of loop configurations that are more suitable for therapeutic design targeting the sTF-fVIIa binding interface. Using NMR CSs from sTF mutants, we have identified a number of key hotspot residues in the fVIIa-binding loop that could account for the intrinsic flexibility observed in the loop. One potential application of our results would be modulating the sTF loop dynamics by designing small molecules that bind specifically to the identified, dynamic hotspot residues. One question we have not addressed in this study is how these dynamic hotspots function in sTF interaction with fVIIa. While in this study we initialized CS-guided simulations from three distinct crystal structures, disparate starting configurations can in principle be drawn from long, unrestrained MD simulations when multiple crystal forms are unavailable, thereby maximizing sampling of the CS-compatible conformational space. A follow-up study would derive the sTF-fVIIa complex structure using CS-guided MD approach with NMR CSs of the complex.

Supplementary Material

SI revised

Table S1: Cα CSs of the fVIIa-binding loop residues predicted by SHIFTX2 for initial crystal structures and final MD frames compared to experimental values. Figure S1: Overlay of 2D 1H-15N TROSY HSQC spectra of WT-sTF, G90A, and S85T mutants. Figure S2: Secondary structure predictions by TALOS+ and dihedral angles for G90A, S85T, and the sTF crystal structure (1BOY).

Acknowledgments

This study was supported in whole or in part by the National Institutes of Health (NIH) through grants P41-GM104601 (E.T.), R24-GM145965 (E.T.), and R01-GM123455 (E.T. and C.R.). It is subject to the NIH Public Access Policy. Through acceptance of this federal funding, NIH has been given a right to make this manuscript publicly available in PubMed Central upon the Official Date of Publication, as defined by NIH. Computational simulations were conducted using resources provided by the National Science Foundation Supercomputing Centers (ACCESS grant number MCA06N060), and Delta advanced computing and data resource which is supported by the National Science Foundation (award OAC 2005572) and the State of Illinois. This study made use of the National Magnetic Resonance Facility at Madison (NMRFAM), which is supported by NIH grant R24-GM141526, and NMRbox: National Center for Biomolecular NMR Data Processing and Analysis, a Biomedical Technology Research Resource (BTRR), which is supported by NIH grant P41-GM111135 (NIGMS).

Footnotes

Competing interests

The authors declare no competing interests during the conduct of this research. Subsequent to this research, S.P. has become an employee of Eli Lilly & Co. is a shareholder of stock in Eli Lilly & Co.; however, those activities are separate and distinct from this research report.

References

  • (1).Feng W; Pan L; Zhang M Combination of NMR Spectroscopy and X-ray Crystallography Offers Unique Advantages for Elucidation of the Structural Basis of Protein Complex Assembly. Science China Life Sciences 2011, 54, 101–111. [DOI] [PubMed] [Google Scholar]
  • (2).Schirò A; Carlon A; Parigi G; Murshudov G; Calderone V; Ravera E; Luchinat C On the Complementarity of X-ray and NMR Data. Journal of Structural Biology 2020, 4, 100019. [Google Scholar]
  • (3).McDermott AE Structural and Dynamic Studies of Proteins by Solid-state NMR Spectroscopy: Rapid Movement Forward. Current Opinion in Structural Biology 2004, 14, 554–561. [DOI] [PubMed] [Google Scholar]
  • (4).Wagner G An Account of NMR in Structural Biology. Nature Structural Biology 1997, 4, 841–844. [PubMed] [Google Scholar]
  • (5).Wüthrich K Protein Structure Determination in Solution by NMR Spectroscopy. Journal of Biological Chemistry 1990, 265, 22059–22062. [PubMed] [Google Scholar]
  • (6).Wütrich K NMR of Proteins and Nuclear Acids; Wiley-Interscience: New York, 1986. [Google Scholar]
  • (7).Gayen A; Banigan JR; Traaseth NJ Ligand-Induced Conformational Changes of the Multidrug Resistance Transporter EmrE Probed by Oriented Solid-State NMR Spectroscopy. Angewandte Chemie – International Edition in English 2013, 52, 10321–10324. [Google Scholar]
  • (8).McDermott A Structure and Dynamics of Membrane Proteins by Magic Angle Spinning Solid-state NMR. Annual Review of Biophysics 2009, 38, 385–403. [Google Scholar]
  • (9).Schneider R; Ader C; Lange A; Giller K; Hornig S; Pongs O; Becker S; Baldus M Solid-state NMR Spectroscopy Applied to a Chimeric Potassium Channel in Lipid Bilayers. Journal of the American Chemical Society 2008, 130, 7427–7435. [DOI] [PubMed] [Google Scholar]
  • (10).Niu Z; Zhao W; Zhang Z; Xiao F; Tang X; Yang J The Molecular Structure of Alzheimer β-Amyloid Fibrils Formed in the Presence of Phospholipid Vesicles. Angewandte Chemie – International Edition in English 2014, 126, 9448–9451. [Google Scholar]
  • (11).Tuttle MD; Comellas G; Nieuwkoop AJ; Covell DJ; Berthold DA; Kloepper KD; Courtney JM; Kim JK; Barclay AM; Kendall A; Wan W; Stubbs G; Schwieters CD; Lee VMY; George JM; Rienstra CM Solid-state NMR Structure of a Pathogenic Fibril of Full-length Human α-synuclein. Nature Structural & Molecular Biology 2016, 23, 409–415. [Google Scholar]
  • (12).Tycko R Molecular Structure of Amyloid Fibrils: Insights From Solid-state NMR. Quarterly reviews of biophysics 2006, 39, 1–55. [DOI] [PubMed] [Google Scholar]
  • (13).Carlon A; Ravera E; Parigi G; Murshudov GN; Luchinat C Joint X-ray/NMR Structure Refinement of Multidomain/multisubunit Systems. Journal of Biomolecular NMR 2019, 73, 265–278. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (14).Soubias O; Pant S; Heinrich F; Zhang Y, R. NS; Li J; Jian X; Yohe ME; Randazzo PA; Lösche M; Tajkhorshid E; Byrd RA Membrane Surface Recognition by the ASAP1 PH Domain and Consequences for Interactions with the small GTPase Arf1. Science Advances 2020, 6. [Google Scholar]
  • (15).Zhang Y; Soubias O; Pant S; Heinrich F; Vogel A; Li J; Li Y; Clifton LA; Daum S; Bacia K; et al. Myr-Arf1 conformational flexibility at the membrane surface sheds light on the interactions with ArfGAP ASAP1. Nature communications 2023, 14, 7570. [Google Scholar]
  • (16).Yee AA; Savchenko A; Ignachenko A; Lukin J; Xu X; Skarina T; Evdokimova E; Liu CS; Semesi A; Guido V; Edwards AM; Arrowsmith CH NMR and X-ray Crystallography, Complementary Tools in Structural Proteomics of Small Proteins. Journal of the American Chemical Society 2005, 127, 16512–16517. [DOI] [PubMed] [Google Scholar]
  • (17).Marion D An Introduction to Biological NMR Spectroscopy. Molecular & Cellular Proteomics 2013, 12, 3006–3025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (18).Nerli S; McShan AC; Sgourakis NG Chemical Shift-based Methods in NMR Structure Determination. Progress in nuclear magnetic resonance spectroscopy 2018, 106, 1–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (19).Cavalli A; Salvatella X; Dobson CM; Vendruscolo M Protein Structure Determination From NMR Chemical Shifts. Proceedings of the National Academy of Sciences, USA 2007, 104, 9615–9620. [Google Scholar]
  • (20).Gong H; Shen Y; Rose GD Building Native Protein Conformation From NMR Backbone Chemical Shifts Using Monte Carlo Fragment Assembly. Protein Science 2007, 16, 1515–1521. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (21).Robustelli P; Cavalli A; Dobson CM; Vendruscolo M; Salvatella X Folding of Small Proteins by Monte Carlo Simulations with Chemical Shift Restraints Without the Use of Molecular Fragment Replacement or Structural Homology. Journal of Physical Chemistry B 2009, 113, 7890–7896. [DOI] [PubMed] [Google Scholar]
  • (22).Berjanskii M; Tang P; Liang J; Cruz JA; Zhou J; Zhou Y; Bassett E; MacDonell C; Lu P; Lin G; Wishart DS GeNMR: a Web Server for Rapid NMR-based Protein Structure Determination. Nucleic Acids Research 2009, 37, W670–W677. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (23).Perilla JR; Zhao G; Lu M; Ning J; Hou G; Byeon I-JL; Gronenborn AM; Polenova T; Zhang P CryoEM Structure Refinement by Integrating NMR Chemical Shifts with Molecular Dynamics Simulations. Journal of Physical Chemistry B 2017, 121, 3853–3863. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (24).Shen Y; Lange O; Delaglio F; Rossi P; Aramini JM; Liu G; Eletsky A; Wu Y; Singarapu KK; Lemak A; et al. Consistent Blind Protein Structure Generation From NMR Chemical Shift Data. Proceedings of the National Academy of Sciences, USA 2008, 105, 4685–4690. [Google Scholar]
  • (25).Kohlhoff KJ; Robustelli P; Cavalli A; Salvatella X; Vendruscolo M Fast and Accurate Predictions of Protein NMR Chemical Shifts From Interatomic Distances. Journal of the American Chemical Society 2009, 131, 13894–13895. [DOI] [PubMed] [Google Scholar]
  • (26).Shen Y; Vernon R; Baker D; Bax A De Novo Protein Structure Generation From Incomplete Chemical Shift Assignments. Journal of Biomolecular NMR 2009, 43, 63–78. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (27).Meiler J PROSHIFT: Protein Chemical Shift Prediction Using Artificial Neural Networks. Journal of Biomolecular NMR 2003, 26, 25–37. [DOI] [PubMed] [Google Scholar]
  • (28).Neal S; Nip AM; Zhang H; Wishart DS Rapid and Accurate Calculation of Protein 1H, 13C and 15N Chemical Shifts. Journal of Biomolecular NMR 2003, 26, 215–240. [DOI] [PubMed] [Google Scholar]
  • (29).Shen Y; Bax A SPARTA+: a Modest Improvement in Empirical NMR Chemical Shift Prediction by Means of an Artificial Neural Network. Journal of Biomolecular NMR 2010, 48, 13–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (30).Steffel J; Lüscher TF; Tanner FC Tissue factor in cardiovascular diseases: molecular mechanisms and clinical implications. Circulation 2006, 113, 722–731. [DOI] [PubMed] [Google Scholar]
  • (31).Banner DW; D’Arcy A; Chène C; Winkler FK; Guha A; Konigsberg WH; Nemerson Y The Crystal Structure of the Complex of Blood Coagulation Factor VIIa with Soluble Tissue Factor. Nature 1996, 380, 41–46. [DOI] [PubMed] [Google Scholar]
  • (32).Muller YA; Ultsch MH; de Vos AM The Crystal Structure of the Extracellular Domain of Human Tissue Factor Refined to 1.7 Å Resolution. Journal of Molecular Biology 1996, 256, 144–159. [DOI] [PubMed] [Google Scholar]
  • (33).Harlos K; Martin DMA; O’Brien DP; Jones EY; Stuart DI; Polikarpov I; Miller A; Tuddenham EGD; Boys CWG Crystal Structure of the Extracellular Region of Human Tissue Factor. Nature 1994, 370, 662–666. [DOI] [PubMed] [Google Scholar]
  • (34).Gajsiewicz JM; Nuzzio KM; Rienstra CM; Morrissey JH Tissue Factor Residues That Modulate Magnesium-dependent Rate Enhancements of the Tissue Factor/factor VIIa Complex. Biochemistry 2015, 54, 4665–4671. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (35).Nuzzio KM; Watt ED; Boettcher JM; Gajsiewicz JM; Morrissey JH; Rienstra CM High-resolution NMR studies of human tissue factor. PLoS One 2016, 11, e0163206. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (36).Pervushin KV; Wider G; Wüthrich K Single transition-to-single transition polarization transfer (ST2-PT) in [15N, 1H]-TROSY. Journal of Biomolecular NMR 1998, 12, 345–348. [DOI] [PubMed] [Google Scholar]
  • (37).Loria JP; Rance M; Palmer AG A TROSY CPMG sequence for characterizing chemical exchange in large proteins. Journal of Biomolecular NMR 1999, 15, 151–155. [DOI] [PubMed] [Google Scholar]
  • (38).Delaglio F; Grzesiek S; Vuister GW; Zhu G; Pfeifer J; Bax A NMRPipe: a multidimensional spectral processing system based on UNIX pipes. Journal of Biomolecular NMR 1995, 6, 277–293. [DOI] [PubMed] [Google Scholar]
  • (39).Lee W; Tonelli M; Markley JL NMRFAM-SPARKY: enhanced software for biomolecular NMR spectroscopy. Bioinformatics 2015, 31, 1325–1327. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (40).Shen Y; Delaglio F; Cornilescu G; Bax A TALOS+: a hybrid method for predicting protein backbone torsion angles from NMR chemical shifts. Journal of Biomolecular NMR 2009, 44, 213–223. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (41).Bonomi M; Branduardi D; Bussi G; Camilloni C; Provasi D; Raiteri P; Donadio D; Marinelli F; Pietrucci F; Broglia RA; et al. PLUMED: A Portable Plugin for Free-energy Calculations with Molecular Dynamics. Computer Physics Communications 2009, 180, 1961–1972. [Google Scholar]
  • (42).Phillips JC; Braun R; Wang W; Gumbart J; Tajkhorshid E; Villa E; Chipot C; Skeel RD; Kale L; Schulten K Scalable molecular dynamics with NAMD. Journal of Computational Chemistry 2005, 26, 1781–1802. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (43).Phillips JC; Hardy DJ; Maia JDC; Stone JE; Ribeiro JV; Bernardi RC; Buch R; Fiorin G; Hénin J; Jiang W; McGreevy R; Melo MCR; Radak B; Skeel RD; Singharoy A; Wang Y; Roux B; Aksimentiev A; Luthey-Schulten Z; Kalé LV; Schulten K; Chipot C; Tajkhorshid E Scalable molecular dynamics on CPU and GPU architectures with NAMD. Journal of Chemical Physics 2020, 153, 044130. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (44).Huang J; MacKerell AD CHARMM36 All-atom Additive Protein Force Field: Validation Based on Comparison to NMR Data. Journal of Computational Chemistry 2013, 34, 2135–2145. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (45).Mahoney M; Jorgensen W A Five-site Model for Liquid Water and the Reproduction of the Density Anomaly by Rigid, Nonpolarizable Potential Functions. Journal of Chemical Physics 2000, 112, 8910. [Google Scholar]
  • (46).Jo S; Kim T; Iyer VG; Im W CHARMM-GUI: a Web-based Graphical User Interface for CHARMM. Journal of Computational Chemistry 2008, 29, 1859–1865. [DOI] [PubMed] [Google Scholar]
  • (47).Humphrey W; Dalke A; Schulten K VMD – Visual Molecular Dynamics. Journal of Molecular Graphics 1996, 14, 33–38. [DOI] [PubMed] [Google Scholar]
  • (48).Bas DC; Rogers DM; Jensen JH Very Fast Prediction and Rationalization of pK(a) Values for Protein-ligand Complexes. PROTEINS: Structure, Function, and Bioinformatics 2008, 73, 765–783. [Google Scholar]
  • (49).Darden T; York D; Pedersen LG Particle Mesh Ewald: An N·log(N) Method for Ewald Sums in Large Systems. Journal of Chemical Physics 1993, 98, 10089–10092. [Google Scholar]
  • (50).Martyna GJ; Tobias DJ; Klein ML Constant Pressure Molecular Dynamics Algorithms. Journal of Chemical Physics 1994, 101, 4177–4189. [Google Scholar]
  • (51).Miyamoto S; Kollman PA Settle: an Analytical Version of the SHAKE and RATTLE Algorithm for Rigid Water Molecules. Journal of Computational Chemistry 1992, 13, 952–962. [Google Scholar]
  • (52).Boettcher JM; Clay MC; LaHood BJ; Morrissey JH; Rienstra CM Backbone 1H, 13C and 15N Resonance Assignments of the Extracellular Domain of Tissue Factor. Biomolecular NMR Assignments 2010, 4, 183–185. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (53).Ulrich EL; Akutsu H; Doreleijers JF; Harano Y; Ioannidis YE; Lin J; Livny M; Mading S; Maziuk D; Miller Z; et al. BioMagResBank. Nucleic acids research 2007, 36, D402–D408. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • (54).Han B; Liu Y; Ginzinger SW; Wishart DS SHIFTX2: Significantly Improved Protein Chemical Shift Prediction. Journal of Biomolecular NMR 2011, 50, 43. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

SI revised
NIHMS2162713-supplement-SI_revised.pdf (1.2MB, pdf)

RESOURCES