Molecular dynamics simulations to the bidirectional adhesion signaling pathway of integrin αVβ3

The bidirectional force transmission process of integrin through the cell membrane is still not well understood. Several possible mechanisms have been discussed in literature on the basis of experimental data, and in this study, we investigate these mechanisms by free and steered molecular dynamics simulations. For the first time, constant velocity pulling on the complete integrin molecule inside a dipalmitoyl‐phosphatidylcholine membrane is conducted. From the results, the most likely mechanism for inside‐out and outside‐in signaling is the switchblade model with further separation of the transmembrane helices.


| INTRODUCTION
Integrin is a membrane protein of crucial importance in the adhesion complex of cells and transmits forces bidirectionally through the membrane. The two monomers integrin α V and β 3 of the heterodimer consist of three main regions each-the extracellular domain outside the cell, the transmembrane part, which perforates the cell membrane, and the cytoplastic tails inside the cell. Integrins with several possible combinations of αand β-chains are known, 1 and we focus here on α V β 3 . Generally, the transport of adhesion signals by integrin from outside the cell to the inside and vice versa is not well understood. Therefore, in this study, the conformational changes and signal transmission for the outside-in activation resulting from an external force stimulus are investigated. By molecular mechanics, the response of only the extracellular domain to external forces has been traced. 2 Here, we develop a model for the full integrin structure and observed the change of the tertiary structure resulting from an applied force to the extracellular domain. The inside-out activation is studied by binding talin to closely connected and separated cytoplastic tail conformations and analyzing whether the presumably active and inactive transmembrane domain conformations give rise to switchblade movements in the extracellular domain.

| Outside-in activation mechanism
Outside-in activation is either mediated by binding of adhesion activators or by applying an external force to the cell. In the extracellular matrix, integrin molecules bind to fibronectin, collagen and more generally to the whole structural extracellular network, and an external force is not likely to act directly on a specific single integrin molecule, 3 but is transmitted simultaneously to several receptors. We mimic this by applying a weaker external force directly to the integrin head, to which the structural proteins are bound. Three possible activation pathways were discussed in literature for the outside-in activation so far: Model 1: The simplest mechanism does not provide any substantial atomic displacements in the extracellular domain [4][5][6][7] (Figure 1, model 1).
The forces only trigger slight shifts of some atomic groups being sufficient for enhancing the binding affinity of talin to the cytoplastic tail.

Model 2:
The switchblade mechanism is more widely accepted, where the integrin extends by unbending of the knees near the head domain 8,9 (Figure 1, model 2).

Model 3:
The third mechanism presumes a hinge like motion as described in (model 2) as a first step. In a second step, talin binds to only one of the two cytoplastic tails and the transmembrane helices are separated by a force resulting from the motion of the actin cytoskeleton which is connected to talin [10][11][12] (Figure 1, model 3).

| Inside-out activation mechanism
The adhesion of integrin to other proteins in the extracellular matrix can also be triggered from inside the cell by binding of talin to the cytoplastic tails. For this inside-out activation, four models have been postulated that are related to the outside-in activation models: Model 1b: The deadbolt model assumes that the β-tail domain acts as a lock for the adhesion site in the βA domain. Interactions of talin with the cytoplastic tails can then induce slight shifts in the β-tail domain that make the adhesion site accessible. 7,13,14 Model 4: In the fourth model, talin binds to one or both cytoplastic tails and initiates the activation by pulling them together until they interact with each other or disrupting them. 15,16 This leads to a rearrangement of atomic groups in the transmembrane and the extracellular domain and ultimately integrin opens in a switchblade-like motion (Figure 1, model 4).

Model 5:
The fifth possible mechanism is that talin binds specifically to the cytoplastic tail of the β-chain only. Then, the actin cytoskeleton connects to talin and pulls the transmembrane helical domains apart, which initiates the opening of integrin 17 (Figure 1, model 5).

Model 6:
The sixth model is a combination of (model 4) and (model 5). In a first step, integrin is unbent as in (model 4) and in a second step the transmembrane helices are separated as in (model 5) 11,12,18 ( Figure 1, model 6). Dihedral angle principle component analysis of  simulated conformations   Protein conformational states are commonly characterized by reaction   coordinates such as radius of gyration, number of hydrogen bonds, principle component analysis (PCA), end-to-end distance, solvent accessible surface area, and root mean square displacements, and often two coordinates have to be combined for specifically denoting conformations. In this study, dihedral PCA (dPCA) 19 is used to identify the conformation, describing the protein with internal backbone dihedrals φ i and ψ i , instead of using atomic coordinates as in standard PCA. This transformation greatly reduces the degrees of freedom and is not affected by translational or rotational overall motions of the protein.  21,22 After homology modeling, α V was aligned to the α IIb chain in the α IIb β 3 template and the resulting three-dimensional structure of the α V β 3 unit was finally connected to the extracellular domain by hand. Further short amino acid sequences, for which no secondary structures exist, were added by hand as random coils. This is justified by the assumption that these missing sequences were not resolved due to a high structural flexibility. Since the template α IIb of the transmembrane domain is shorter than the α V chain itself, the remaining part of α V on the cytoplastic tail, 1027 to 1048, is also modeled as coil.

|
Glycosylations and phosphorylations were applied according to UniProt (P06756 and P05106, cf. Figure S1). 23 The partially glycosylated protein was solvated in explicit water and the negative charge of the biomolecule was compensated by 44 sodium counter ions. Six calcium ions were added: Five ions are located in the Tigh and β-propeller domains (1 and 4, respectively) providing structural stability. 24 The sixth Ca 2+ binds to the metal-ion-dependent-adhesion side (MIDAS) in the βA-domain. Experiments have shown that this calcium does not activate integrin, especially if the ion sites adjacent to MIDAS and the synergistic metal ion-binding side are unoccupied. 24 Therefore, integrin is not activated by ions and we assume that only responses to the external force are observed. In a later step, the integrin was imbedded into a dipalmitoyl-phosphatidylcholine (DPPC) membrane. 25 F I G U R E 1 Schematic drawing of integrin structures according to different activation mechanisms (models 1-6) investigated in this study. The extracellular (red), transmembrane ( green), and (blue) cytoplastic domains are indicated. Talin molecules are represented as orange circles

| Molecular dynamics simulation
All simulations were calculated with the software package GROMACS 5.1. 26,27 Standard amino acids of the protein were represented within the AMBER99SB-ildn force field 28,29 and the FFPTM force field was applied to phosphothreonine (PTR, 767) and phosphotyrosins (773 and 785). 30 The parameters for glycosylated N-linked asparagine as well as the carbohydrates ( Figure S1) were taken from Glycam, 31 and water was represented by the TIP3P model. 32 Periodic boundary conditions were applied with the cell sizes according to Table 1. All bond lengths were constrained with the LINCS algorithm. 33 Coulomb interactions were treated by the particle mesh Ewald method 34  The cell sizes were adjusted by applying two independent Parrinello-Rahman piston barostats 36 separating the xy-plane from the z direction, both with coupling time constants of 12 ps. This permitted to maintain the membrane with little stress curvature while adjusting the water density essentially by the cell height. The membrane itself was built by multiplying an equilibrated DPPC 25 membrane unit cell in the xydirection and solvating the system with water. After minimization, the system was equilibrated for 0.4 and 0.6 ns in an NVT and NPT ensemble, respectively. The minimized structure of integrin was integrated into the system using the GROMACS tool "gmx insert-molecules," which placed the molecule into the membrane and removed overlapping DPPC and water molecules. The resulting artificial gap in the membrane was closed during a followed 5 ns long equilibration. This initial system was then used for the constant velocity pulling and interaction studies by either adding additional water in the case of vertical pulling or enlarging and equilibrating the membrane for the horizontal pulling (Table 1).
In preparation for the vertical pulling, a 102.5 ns long simulation in a NPT ensemble was conducted and different conformations at the times of 2.5, 52.5, and 102.5 ns were chosen as starting structures (Table 1). Thereby, we assume that the structure of the extracellular domain after 2.5 ns is still related to the initial X-ray structure and the later two conformations correspond to the equilibrium ensemble ( Figure S2).
For the constant velocity pulling, a harmonic potential was applied with a force constant of 50 kJ/mol/nm 2 (0.08 N/m) between the C a atoms of the integrin head and the transmembrane domain with reference distances constantly moving with speeds of 2 or 0.1 nm/ns in vertical or horizontal directions, respectively. The vertical pulling speed was previously optimized by Reference 2 on the same X-ray structure of the extracellular integrin domain. The very small force constant in combination with moderate pulling speeds resulted in a smooth and soft opening of integrin. After pulling, the systems relaxed for several 100 ns during NPT runs (Table 1). For consecutive relaxation simulations exceeding 1 μs, hydrogen mass repartitioning to 4 a.m.u. 37 was applied, and the time step was increased to 7 fs.
The reverse process, the closing of integrin, was accelerated by pulling the extracellular and transmembrane domains reducing their distance. Again, a force constant of 0.08 N/m and a constant velocity of 2 nm/ns were applied for pulling the two domains relative to each other.
In the horizontal pulling simulations, the high viscosity of the membrane, as compared to water, was accounted for by drastically reducing the pulling speed from 2 to 0.1 nm/ns, thereby reducing the viscous friction on the integrin transmembrane domain. In fact, each of the two helices in the transmembrane domain is pulled by 0.05 nm/ns away from its starting point, increasing their mutual distance by 0.1 nm/ns. This velocity of 5 × 10 7 nm/s is still magnitudes T A B L E 1 Simulation details. Columns contain from left to right the cell volume C S , the numbers of membrane building units N M and of water molecules N W , the times of equilibration t E and of pulling or morphing, t Pu or t M , respectively, and the relaxation time t R for each simulation run. The brackets refer to the configurations at the end of each stage as plotted in Figures 2 and 5

| Dihedral PCA
The application of dPCA to peptides has been described in Reference 19.
In Reference 41, we showed that the conformations of small to medium sized peptides are unambiguously described by the first two main components of dPCA. Still the use of a larger number of main components might enhance the precision of the conformational assignment, but for clarity, we limit the evaluation to the first two.

| Probabilities of interactions
The interaction probabilities were calculated with a homemade program, which counts contacts between two amino acid residues on two different molecules in a trajectory and normalizes the results with respect to the number of frames similar to the method described in

| Torsion angle
In the inactive state, torsion angles between 25 and 35 are found (Table S1), which is consistent with the observation for the similar integrin α IIb β 3 . 43,44 The vertically pulling in each of the three simulations resulted in a rapid decrease of the torsion angle between the transmembrane domains to less than 20 ( Figure 5, top row, symbol I and Table S1). This event can occur directly during the pulling ( Figure 5, left) or up to several 100 ns later ( Figure 5, right). Then, 10 to 100 ns after this event, the angle recovers to around 20 to 24 , F I G U R E 3 Force-distance curves for three vertical (A-C) and one horizontal (D) pulling simulations. The initial and final states of these curves are A1-4 and B1-4, respectively (cf. Figure 2, right column) and only after long relaxation times it further increased to the initial 25 to 35 ( Figure 5, right). We assume that this recovery indicates the start of spontaneous deactivation.

| Tail-to-tail distance
During the initial relaxation, the two tails did not approach (Figure 5, bottom). Even after the long equilibration time of 100 ns, no contact between the two cytoplastic tails of α V and β 3 was observed and their separation was about 1.5 nm in the inactive state.
After observing the decrease of the torsion angle between the transmembrane domains to less than 20 , the two tails approached within a delay of about 70 ns and reached a stably interconnected configuration ( Figure 5, bottom, symbol II). This suggests that the closing of the cytoplasmic tails outside the cell is not incidental but correlated to the pulling of the ectoplasmic domain inside. We assume that unbending is a collective process involving rather complex motions beyond the change of the knee torsional angles only, and that the reduction of the tail-to-tail distances and torsional angles is correlated as parts of this motion. The final minimal end-to-end distance of 0.2 nm between the tails is determined by one to three unspecific hydrogen bonds. The most common case was the interaction of Arg1026 and Asn1042 with the phosphorylated amino acids Thr779 and Tyr785, but a short antiparallel sheet conformation with Thr777 <-> Glu1044 and Glu775 <-> Asn1046 was also observed.
We assign the close interconnection of the cytoplastic tails to the activation of integrin. The time for this activation event of integrin increases with increasing initial equilibration time. Presumably, this indicates that the initial X-ray structure is stressed and already in a semiactivated state. As integrin is a large protein, the relaxation from this initial stress takes several 100 ns ( Figure 5, bottom row, Figure S2).

| Deactivation as reverse process to activation
We tried to reverse the activation process and transfer the active state C as attained in the vertical pulling simulations back into the initial inactive conformation A in two different ways: 1. First, the deactivation was enforced by steered molecular dynamics after the vertical pulling simulation 1 ( Figure 5, top row, left). The head group was pulled downward toward the F I G U R E 4 Naming for the integrin regions and domains. The metal-ion-dependent-adhesion side is located in the βA domain. The colors correspond to the secondary structure elements sheet (yellow), helix (magenta), 3-10-helix (blue), turn (cyan), and disordered (white) [Color figure can be viewed at wileyonlinelibrary.com] F I G U R E 5 The torsion angle (top) between the main axes of the transmembrane helices between α V and β 3 and the end-to-end (bottom) distance between the cytoplastic tails during the three vertical pulling simulations (from left to right: 1 to 3, with relaxation times 2.5, 52.5, and 102.5 ns, respectively) are plotted. The respective conformations are indicated (cf. Figure 2, right column). The symbols I, II, and III denote the drop of the torsion angle between the transmembrane helices, the approach of the cytoplastic tails and the increase of the torsion angle after long relaxation, respectively [Color figure can be viewed at wileyonlinelibrary.com] transmembrane helices, but the inactive form could not be regained (Figure 2, top row).
2. To achieve spontaneous deactivation, the third vertical pulling simulation was continued by free molecular dynamics for about 1 μs. A trend toward the deactivated conformation A was revealed by the dPCA analysis ( Figure 2, third row). The increase in the torsion angle of the transmembrane helices at around 0.9 μs (Figure 5, right, symbol III) could be a further indication of the start of deactivation.
This indicates that the activated state is extraordinarily stable and the reverse process is very slow. We expect that the spontaneous deactivation will take a very long time and exceed the 1 μs simulation time by at least an order of magnitude. It is feasible from the cell point-of-view that the activation is a rapid response to the stimulus, but that the deactivation is much slower keeping the cell responsive for a significant time after the end of the stimulation. 7 3.2 | Inside-out activation

| Horizontal pulling
The pathway of inside-out activation was analyzed by a horizontal pulling simulation. Here, the integrin displaces inside the highly viscous membrane. Pulling in a direction in the membrane plane ("horizontally") results in a completely different force distance curve ( Figure 5D), and the contributions to the force are probably very different: 1. We assume that at the start the protein has to overcome a high energy barrier, since the lipid network is rather rigid and the intermolecular Coulomb and van der Waals interactions between the two helices in the transmembrane domain are strong.
2. The transmembrane helices then are pulled through regions of medium and high viscosity leading to a strong fluctuation in the plot.
The average membrane resistance itself is about 180 pN as seen in the force distance curves after the initial high barrier (Figure 3). This corresponds to a lipid viscosity of 0.12 Pa/s for a spherical shape with 1.6 nm radius. Experimentally, lipid viscosities are between 0.03 and 1.5 Pa/s, 45 which is 30 to 1500 times higher than in water and could give rise to a force fluctuation between 45 and 2300 pN. We ascribe the noisy trace in Figure 5D to this effect.
The total force is much higher than during vertical pulling ( Figures 2D and 3D) and acts on a smaller protein fragment. This high stress led to the destruction of the secondary structure of the F I G U R E 6 A, Interaction maps between amino acid species on integrin vs talin for the active (left) and inactive (right) cytoplastic tails conformations. Black and yellow colors indicate low and high interaction probabilities, respectively. The colors indicate that binding of talin to the inactive integrin is by roughly a factor of three more frequent than to the active molecule. B, Interaction probabilities between specified residues on integrin tails α V (residues 1017-1048) and

| Talin binding behavior
In the living organism, the forces, which were artificially imposed here to the system by steered molecular dynamics, may arise from binding of talin inside the cell to the integrin cytoplastic tails, initiating the unbending and activation of the extracellular domain. We analyzed the binding of the talin domains F2 and F3 to these tails using the starting configurations of inactive (A1) and active (C3) integrin, respectively. In the inactive state both cytoplastic tails interact with the F2 and F3 domain of talin ( Figure 6B,C), mainly with the PTR, threonine, glutamate and the C-terminal residues ( Figure 6A). In the active state, integrin shows a weaker, but more specific interaction, and mainly the F2 domain of talin binds to the β 3 -tail of integrin.
This binding pattern suggests the following inside-out mechanism: Binding of talin to both tails initiates unbending by the switchblade motion of the ectodomain. After the activation, talin switches to a specific binding to the β 3 cytoplastic domain, still being connected to the cytoskeleton. Its natural motion displaces talin and forces a separation of the transmembrane helices, keeping integrin in the activated state of model 6.

| Reaction of the extracellular domain to activation and deactivation
For the signal transfer from the cell inside to its outside, it is important that the activation of the cytoplastic tails transforms the extracellular domain conformation, probably by initiating a switchblade motion. We observed this reaction by morphing the transmembrane and cytoplastic domains within several ns either from the respective structures in inactive A5 to active C1 or active A6 to inactive A3 (Figure 7). During the simulation time, no major structural changes or motions were observed within the extracellular domain, but dPCA revealed trends for both processes toward the respective state of the extracellular domain ( Figure 7).

| CONCLUSION
In this study, the question of the mechanism for integrin outside-in and inside-out signaling was tackled by molecular dynamics simulations on the full integrin structure. Several mechanisms that had been F I G U R E 7 As Figure 2, but for inside-out activation (top) and deactivation (bottom). (Left) Dihedral angle principle component analysis (dPCA) plots and (right) the corresponding conformations for the peaks A5 to D6. A5 and A6 are identical to A1 and C3, respectively, while for the conformations B5 and B6, only the cytoplastic tails and transmembrane domains are identical to those of the structures C1 and A3, respectively (cf. Figure 2). The extracellular matrix domain has evolved freely during the morphing simulation [Color figure can be viewed at wileyonlinelibrary.com] described earlier [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18] were examined for their feasibility (Figure 1). The best description of the observations provides a switchblade motion followed by a separation of the transmembrane helices (Figure 1, model 6). Based on molecular dynamics simulations, the same mechanism was suggested recently as the most likely one for the similar integrin α IIb β 3 . 3 For the outside-in signaling, we demonstrated that by forcing this switchblade motion in the extracellular domain, the torsional angle between the transmembrane helices changes enforcing the contact between the previously separated cytoplastic tails ( Figure 5).
The same mechanism seems to apply for the inside-out signaling. Talin binds to both cytoplastic tails of the inactive integrin and presumably initiates an unbending of the extracellular domain ( Figure 6). This then reduces the binding affinity of talin especially to α V and make the protein interact with the β 3 cytoplastic tail of integrin only ( Figure 6B), leading presumably to a separation of the transmembrane helices with the above explained mechanism.
Although the study provides promising results, they are based on a few assumptions and limitations. The integrin model was put together by hand from an extracellular X-ray structure, a homology model of the transmembrane domain and several assumed random coil structures of missing sequences. Since the transmembrane domain is entirely in α-helix conformation, the homology guess seems accurate. The remaining added sequences on the other hand might have a distinct secondary structure, which they may not have adopted during the equilibration. Also, the observables obtained during the pulling depend majorly on the used parameters. In this study, we considered the resulting conformation after the pulling (Figure 2, structures B1-B4) as independent from these parameters. Finally, even though the dPCA pictures indicate a lengthy deactivation process after the integrin activation (Figures 2 and 7), it is not clear if this process will spontaneously complete during a long enough simulation.
Nevertheless, the presented study provides valuable insights and a plausible explanation for the integrin activation process and hopefully encourages future experimental studies for verifying the proposed mechanism.