Integrin Activation Dynamics between the RGD-binding Site and the Headpiece Hinge*

Integrins form mechanical links between the extracellular matrix and the cytoskeleton. Although integrin activation is known to be regulated by an allosteric conformational change, which can be induced from the extracellular or intracellular end of the molecule, little is known regarding the sequence of structural events by which signals propagate between distant sites. Here, we reveal with molecular dynamics simulations of the FnIII10-bound αVβ3 integrin headpiece how the binding pocket and interdomain βA/hybrid domain hinge on the distal end of the βA domain are allosterically linked via a hydrophobic T-junction between the middle of the α1 helix and top of the α7 helix. The key results of this study are: 1) that this T-junction is induced by ligand binding and hinge opening, and thus displays bidirectionality; 2) that formation of this junction can be accelerated by ligand-mediated force; and 3) how formation of this junction is inhibited by Ca2+ in place of Mg2+ at the site adjacent to the metal ion-dependent adhesion site (“ADMIDAS”). Together with recent experimental evidence that integrin complexes can form catch bonds (i.e. become strengthened under force), as well as earlier evidence that Ca2+ at the ADMIDAS results in lower binding affinity, these simulations provide a common structural model for the dynamic process by which integrins become activated.

Integrins anchor cells to the extracellular matrix. They are transmembrane heterodimers, composed of non-covalently bound ␣ and ␤ subunits that associate to form the extracellular, ligand-binding head, two multidomain "legs," two single-pass transmembrane helices, and two short cytoplasmic tails (Fig.  1A). All known integrin heterodimers contain the ␤A domain (also called I-like or ␤I domain), located at the extracellular end of the ␤-subunit. The top of the ␤A domain contains three metal ion binding sites, termed the "ligand-induced metal binding site" ("LIMBS"), 3 the "metal ion-dependent adhesion site" ("MIDAS"), and the "adjacent to the MIDAS" ("ADMIDAS").
The LIMBS has been called the synergistic metal binding site instead, since it was found to contain a metal ion in a crystal structure of the unliganded ␣ IIb ␤ 3 integrin (44).
Integrin-mediated adhesion often occurs under tensile forces such as fluid flow or myosin-mediated contractions that cells exert to sample the rigidity of their surroundings. Thus, to enable mechanosensing, integrins cannot be constitutively active. Rather, integrin activation is regulated by long-range conformational changes that can originate from the cytoplasmic or extracellular end of the integrin molecule (1). For example, ligand binding has been shown to induce the activating conformational change that leads to hinge opening between the ␤A and hybrid domains in the integrin headpiece (2)(3)(4). Vice versa, events originating from the cytoplasmic region of the molecule have been shown to switch the extracellular binding site to the high affinity state (5). This bidirectional reciprocity is typical for allosterically regulated proteins (6).
Although many factors that regulate the allosteric pathway of integrin activation are known, such as ligand binding (7), divalent cations (8), and mechanical force (9,10), how structural alterations propagate from one end of the molecule to the other remains poorly understood. For instance, x-ray crystallographic structures of the unliganded ␣ v ␤ 3 (11) and liganded ␣ IIb ␤ 3 (12) integrins provide snapshots of the closed and open headpiece hinge conformations, respectively. These differences highlight the structural events that accompany activation. Yet little is known regarding the sequence by which these events occur. In another example, comparison between the liganded ␣ v ␤ 3 (11) and liganded ␣ IIb ␤ 3 (12) integrin crystal structures reveals near identical configurations of the ligand-bound ␤ 3 integrin binding pocket, differing only in the backbone hydrogen bonding of the ␤1-␣1 loop (see Fig. 1E) (12,13). However, these structures display a ϳ62°difference in the hinge angle between the ␤A and hybrid domains, a conformational change that has been shown to govern binding affinity (2)(3)(4).
With no experimental techniques currently available to obtain dynamic, atomic level insights into the integrin activation pathway, we used molecular dynamics (MD) simulations and steered molecular dynamics (SMD) simulations to derive these insights computationally. Here, we show how a key structural event at a location halfway between the RGD-binding pocket and the headpiece hinge region directly regulates hinge opening. This event is the formation of a T-junction between hydrophobic residues at the middle of the ␣1 helix and the top of the ␤6 strand and C-terminal ␣7 helix. This hydrophobic contact was originally identified in the description of the first open-hinge crystal structure of an integrin headpiece, that of the ␣ IIb ␤ 3 integrin (12). Here, we show that this event immedi-ately precedes ligand-induced hinge opening. Vice versa, inhibition of this T-junction prevents hinge opening, and we show how Ca 2ϩ at the ADMIDAS conducts this inhibitory role. In addition, we show that simulated mechanical force or "manual" opening of the hinge accelerates formation of this junction. Together with recent experimental data (10), these findings lay the groundwork for a high-resolution model of how the lifetime of the RGD-␣ v ␤ 3 integrin complex can be lengthened under force, the hallmark of a "catch" bond. Differences between the ␤A domain model of ADMIDAS-regulated, T-junction formation described here and the pull-spring model for the structurally homologous ␣A integrin domain proposed elsewhere (14) are discussed.

EXPERIMENTAL PROCEDURES
System Setup-The crystal structures of the ␣ v ␤ 3 integrin in complex with an RGD-containing mimetic ligand (Protein Data Bank entry code 1L5G, 3.2 Å resolution (13)) and FNIII 10 from the crystal structure of the FNIII 7-10 tetramer (PDB entry code 1FNF, 2.0-Å resolution (15)) were used to build the FNIII 10 -␣ v ␤ 3 starting structure with the program VMD (16) as described previously (17). To reduce the integrin complex to a size that can be simulated on a feasible time scale, we used only the ␤A and hybrid domains of the ␤ 3 subunit and the ␤-propeller domain of the ␣ v subunit. The starting structure was solvated in a 116 Å ϫ 115 Å ϫ 122 Å box of explicit TIP3 (18) water molecules, resulting in 153,570 atoms (Fig. 1B). For comparison, the corresponding domains from the unliganded ␣ v ␤ 3 integrin crystal structure (11) were also solvated in a box of explicit water molecules, resulting in 88,417 atoms. Seven cation binding sites are resolved in the crystal structure of the headpiece, three in the MIDAS, LIMBS, and ADMIDAS binding pocket motifs and four in the solvent-exposed ␤ hairpin loops at the bottom of the ␣-subunit domain. These were occupied by Mn 2ϩ in the crystal structure and by Mg 2ϩ or Ca 2ϩ in our simulations.
An additional complex was built in which the ␤A/hybrid domain hinge was opened "manually." To do this, the FNIII 10 -␣ v ␤ 3 integrin complex was aligned after 1 ns of equilibration with the same complex after 7 ns of equilibration, at which point the hinge angle had increased by ϳ22°. Alignment was done via the 6 ␤-strands of the ␤A domain. Once aligned, the computational coordinates of the (closed hinge) hybrid domain at 1 ns were replaced by those of the (open hinge) hybrid domain at 7 ns. In this fashion, the hinge angle of the FNIII 10 -␣ v ␤ 3 integrin complex, after 1 ns of equilibration, was increased manually by ϳ22°.
Simulation Procedures and Parameters-All MD simulations were carried out with the program NAMD (19) using the CHARMM27 force fields (20). For a detailed description of the simulation protocol, see Refs. 17 and 21. All complexes remained stable during equilibrations, exhibiting an overall C␣ root mean square deviation of less than 2.0 Å in the integrin head. The integrin head is comprised of the ␤-propeller domain from the ␣ v subunit and the ␤A domain from the ␤ 3 subunit.
To probe the relative kinetics of force-induced hinge angle opening, external forces were applied by SMD protocols at constant forces of 100, 150, 250, 500, 700, 800, and 900 pN (22). To examine the mechanical response of the complex in various pulling geometries, we fixed either one or both of the C-terminal residues of the integrin headpiece (i.e. the C␣ atoms of either ␣ V -Arg 438 and/or ␤ 3 -Asp 434 ). In every case, force was applied to the C␣ atom at the C terminus of FnIII 10 (Thr 93 ). Generally, the direction of the stretching force was along the vector pointing from the fixed atom to the pulling atom. When the C-terminal ends of both the ␣ and ␤ subunits were fixed, the force vector was directed from the midpoint of the two fixed atoms to the pulling atom in the FnIII 10 domain.
Simulations were conducted on the Gonzales cluster at ETH Zürich and on the Cray XT3 Palu supercomputer at the Swiss National Computing Centre, Manno, Switzerland. A 1-ns simulation required about 24 h on 64 Dalco nodes with 2.4 GHz AMD Opteron processors, or 10 h on 64 dual core AMD Opteron nodes with 2.6 GHz processors, respectively. All structure alignments were done in VMD (22) via the backbone atoms of the 6 ␤-strands of the ␤A domain. Relative hinge angle movement was measured using a script called Hingefind (23) and the frequency of attacks by free water molecules was measured as described previously (21). Customized scripts to measure the secondary structure, the center of mass, or atom-atom distances over specific structural regions were written in Tcl within VMD. Figures were rendered using VMD.

RESULTS
Our MD and SMD approaches take advantage of the existence of the semi-equilibrated, liganded ␣ v ␤ 3 integrin crystal structure, which was obtained by soaking RGD-peptides into preformed, unliganded integrin crystals (13). Previously, our MD equilibrations of this structure revealed that the closed ␤A/hybrid domain hinge opens on the nanosecond time scale when the constraints of the bent leg domains are lifted and the RGD-ligand is replaced with the 10th type III module of fibronectin (FnIII 10 , colored yellow in all of our figures) (17).
Here, the FnIII 10 -␣ v ␤ 3 integrin headpiece complex was further investigated in 25 separate simulations ranging from 3 to 7 ns each. These simulations consist of 8 MD equilibrations and 17 SMD simulations. Because Mg 2ϩ is known to regulate integrin activation (8) and is present in the integrin under physiological conditions, all integrin metal ion-binding sites were occupied with Mg 2ϩ unless replaced by Ca 2ϩ , where specified, for comparison. The studies presented here, totaling 127 ns of simulation time, were then considered together with the 107 ns of simulation time described earlier (17).

Spontaneous Opening of the Hinge Immediately Follows Formation of a Hydrophobic T-junction between the ␣1 and ␣7
Helices-The liganded ␣ v ␤ 3 (11) and liganded ␣ IIb ␤ 3 (12) integrin crystal structures display a ␤A/hybrid domain hinge angle difference of ϳ62°(12, 13), a conformational change that has been linked to the switch from the inactive to the active state (2)(3)(4). However, comparison between these structures reveals near-identical configurations of the ligand-bound ␤A domain binding pocket, differing only in the backbone hydrogen bonding of the ␤1-␣1 loop. In contrast, considerable variation is found between the ␤A domains of these structures in the region surrounding the middle of the ␣1 helix and the top of the C-terminal ␣7 helix (Fig. 1, shown overlaid in E and shown separately  (13). Integrin ␣and ␤-subunits are blue and gray, respectively. Domains not resolved in the crystal structure were added in pink. B, the headpiece complex was simulated with explicit water molecules. C, the region of the ␣1/␣7 helix junction in the integrin headpiece is shown in the box. D, the ␤A and hybrid domains from the liganded ␣ v ␤ 3 integrin crystal structure (13), in black, are aligned with the same domains from the unliganded ␣ v ␤ 3 integrin crystal structure (11), shown in color (red for the ␤6-␣7 loop and ␣7 helix, green for the ␤1-␣1 loop and ␣1 helix, and transparent purple for everything else). E, the ␤A and hybrid domains from the liganded ␣ v ␤ 3 integrin crystal structure (13), in black, are aligned here with the same domains from the liganded ␣ IIb ␤ 3 integrin structure (12), shown in color. Where the hinge is closed in both structures (D), the greatest structural change is in the binding pocket. Where the difference in the hinge angle is ϳ62°(E), the region of greatest change is that of the ␣1/␣7 T-junction. F-H, the three integrin crystal structures are shown separately. MD-derived snapshots (I and J) are depicted beneath the starting structures to which they correspond (F and G, respectively). The ␤1-␣1 loop/␣1 helix and ␣7 helix/hybrid domain regions are shown in green and red, respectively. Residues of the ␤6-␣7 loop and ␣7 helix shown space-filling in red are Leu 343 and Ile 344 , located at the top of the ␣7 helix, Val 247 , Ile 307 , and Ala 309 , located on underlying ␤-strands, and Leu 333 , located where the ␤6-strand becomes the ␤6-␣7 loop. Shown in space-filling in green are Leu 134 and Leu 138 in the ␣1 helix. Black arrows identify the region where the ␤1-␣1 loop meets the ␣1 helix, and the region where the ␣1 helix comes in contact with the ␤6 strand and ␣7 helix during T-junction formation.
in G and H). Indeed, a novel hydrophobic packing in this region surrounding the middle of the ␣1 helix and the top of the C-terminal ␣7 helix and ␤6 strand was first identified in the description of the liganded ␣ IIb ␤ 3 integrin crystal structure (12) (Fig.  1H). Here, we report that the transition to this hydrophobic packing arrangement led to hinge opening in the RGD-occupied ␣ v ␤ 3 integrin headpiece. When Leu 134 , at the middle of the ␣1 helix, formed a hydrophobic junction with residues surrounding the top of the ␣7 helix, the hinge increased by more than 10°over a time frame of 1 to 2 ns. This junction is evident by a tight packing between the middle of the ␣1 helix and the top of the ␤6-strand, as shown in Fig. 1, H and J (black arrows). Residues of the ␣1 helix and ␤6 strand/␣7 helix are colored green and red, respectively.
To identify T-junction formation, we measured the decrease in C ␤ -atom distance between the middle of the ␣1 helix (Leu 134 ) and the top of the ␤6-strand, where it becomes the ␤6-␣7 loop (Leu 333 ). This distance is traced in black in Figs. 2, 4, and 6. Hinge opening is also displayed in supplemental Videos 1 and 2.
When the corresponding headpiece domains from the unliganded ␣ V ␤ 3 integrin structure were equilibrated, the hinge remained closed and the ␣1/␣7 helix junction did not form. This is shown in the snapshot in Fig. 1I and the traces in Fig. 2C. These findings were reproduced in repeat simulations of the FnIII 10 -bound and unliganded integrin headpiece domains (supplemental Fig. S1).
The difference in the ␣1/␣7 helix junction region between the liganded ␣ v ␤ 3 and liganded ␣ IIb ␤ 3 structures displayed in Fig. 1E contrasts the close alignment of this region between the liganded and unliganded ␣ v ␤ 3 integrin structures displayed in Fig. 1D. In both the liganded and unliganded ␣ v ␤ 3 integrin structures, the headpiece hinge is closed and the ␣1/␣7 helix junction is not formed (Fig. 1, shown overlaid in D and shown separately in F and G) (11,13). Thus, the link between ␣1/␣7 T-junction formation and hinge opening is consistent with crystallographic data.
Characteristic Structural Alterations That Accompany Formation of the ␣1/␣7 Helix Junction-In the absence of the ␣1/␣7 T-junction, a break exists in the ␣1 helix. During equilibrations of the unliganded headpiece domains, this break in the ␣1 helix was maintained at Asn 133 (Fig. 3B). In contrast, with the junction in place, the ␤1-␣1 loop and ␣1 helix tended to join into one continuous helix structure as shown in Fig. 1, H and J, and traced over time for residues at the center of the helix (132 to 135) in Fig. 3A. To quantify the fluctuations of the ␣1-helix conformation, we added the number of residues in the ␤1-␣1 loop and ␣1 helix that join in the ␣-helix conformation over time (residues 120 to 146). This tally is traced in gray in Figs. 2, 4, and 6. Another feature of ␣1/␣7 T-junction formation is an increase in the distance between the ␤1-␣1 loop and the ␤6-␣7 loop, as shown in the pink traces in Figs. 2, 4, and 6.
Bidirectionality of the ␣1/␣7 T-Junction/Hinge Opening Pathway-Experiments have shown that structural factors distal to the binding site can transition the integrin headpiece to the high affinity state by opening the ␤A/hybrid domain hinge (2,4,24). Thus, we asked if opening the hinge manually would induce the ␣1/␣7 T-junction to form. The hinge was opened manually by replacing only the hybrid domain coordinates after 1 ns of equilibration (at which time the hinge had not yet opened), with the hybrid domain coordinates after a hinge increase of ϳ22°. In this fashion, a starting structure was obtained in which the ␣1/␣7 helix junction was not in place and the hinge was increased by ϳ22°. When this complex was equilibrated in three separate simulations of 4 ns each, the ␣1/␣7 T-junction formed within the first 40 ps in each case. As expected for a bidirectional allosteric process junction formation was accompanied by a decrease in the average Leu 333 -Leu 134 C ␤ -atom distance (Fig. 2B, black trace), an increase in the separation of the ␤1-␣1 and ␤6-␣7 loops (Fig. 2B, pink trace) and reduced fluctuations in the ␣1 helix conformation (Fig. 2B, gray line).
How Ca 2ϩ in Place of Mg 2ϩ at the ADMIDAS Inhibits Formation of the ␣1/␣7 T-Junction-Previously, we described 4 separate FnIII 10 -␣ v ␤ 3 integrin equilibrations in which small structural perturbations near the binding pocket prevented hinge opening on the nanosecond time frame (17). Here, we report that in each of these cases, hinge opening was deterred by a common mechanism: inhibition of ␣1/␣7 T-junction formation. In one case, the structural perturbation was the replacement of Mg 2ϩ with Ca 2ϩ at the LIMBS and ADMIDAS in the binding pocket, which was found to promote an outwards break between the ␤1-␣1 loop and ␣1-helix (17). Now, we report that in two separate repeat equilibrations of the Ca 2ϩoccupied complex, this outward break consistently prevailed (Fig. 3C) and the T-junction between the ␣1 and ␣7 helices did not form. Because Ca 2ϩ has long been known to inhibit integrin activation (8), we asked how Ca 2ϩ in place of Mg 2ϩ promotes the split between the ␤1-␣1 loop and ␣1-helix and thus abrogates T-junction formation.
Bidentate coordination is known to be a characteristic feature of Ca 2ϩ ions, relative to Mg 2ϩ (25), and this distinction was consistently displayed in all of our MD and SMD simulations (17). While the preferred number of atoms coordinating Mn 2ϩ or Mg 2ϩ in integrin crystal structures is six, the preferred coordination number of the Ca 2ϩ ion occupying the ADMIDAS is seven (11,12). Three of the four residues that directly coordinate the ion occupying the ADMIDAS are located along the ␤1-␣1 loop. These are Ser 123 , Asp 126 , and Asp 127 . When Ca 2ϩ occupies the ADMIDAS instead of Mn 2ϩ or Mg 2ϩ the seventh atom that joins the ADMIDAS coordination sphere in ␤ 3 -integrin crystal structures is the second carboxylate oxygen from Asp 126 , located where the bottom of the ␤1-␣1 loop joins the top of the ␣1-helix (Fig. 4A) (12,13). To investigate the influence of this Ca 2ϩ -specific, bidentate ADMIDAS coordination, we turned off the charge on the second Asp 126 oxygen. In this fashion, we produced a complex in which the ADMIDAS displayed an "Mg 2ϩ or Mn 2ϩ mimetic" configuration while occupied by Ca 2ϩ . This configuration is characterized by a single ADMIDAS-coordinating oxygen from each of the Asp 126 and Asp 127 ␤1-␣1 loop residues (Fig. 4B), as is the case when Mg 2ϩ or Mn 2ϩ occupy the ADMIDAS. Hereafter, we shall refer to this as the "Ca 2ϩ -specific Asp 126 -charge off" complex.
In three of four separate equilibrations of the Ca 2ϩ -specific Asp 126 -charge off complex (2 ns each), the 6-fold coordination of the ADMIDAS ion was maintained. In two of these simula-tions, the ␣1/␣7 T-junction formed. In both of these cases, the most prevalent conformation of the ␤1-␣1 loop and ␣1-helix switched from that of the split conformation that is typical of Ca 2ϩ occupation (Fig. 4A) to the continuous helix conformation that we otherwise found in Mg 2ϩ -occupied complexes (Fig. 4B). In the third simulation, a hydrogen bond between the , and the distance between the ␤1-␣1 loop and ␤6-␣7 loop (pink). More specifically, the orange trace in each panel displays the relative movement of the ␤A and hybrid domains, the gray trace in each panel displays the number of residues across the ␤1-␣1 loop and ␣1 helix (residues 120 to 146) that form an ␣ helix over time, the black trace in each panel is the C ␤ -atom distance between Leu 134 of the ␣1 helix and Leu 333 on the ␤6 strand, and the pink trace in each panel is the distance between the centers of mass of the ␤1-␣1 loop (residues 122 to 127) and the ␤6-␣7 loop (residues 333 to 340). To obtain the number of residues across the ␤1-␣1 loop and ␣1 helix that form an ␣ helix over time (gray trace), each of the 26 residues is assigned either a "1" or "0," corresponding to whether the residue was participating in an ␣ helix (1) or not (0). Then these numbers were summed and traced over time. The maximum value of this measurement was 24, because Leu 120 and Ser 121 , located at the top of the ␤1-strand, do not ever join the ␣ helix.  DECEMBER 25, 2009 • VOLUME 284 • NUMBER 52 top of the ␣1-helix (at Trp 129 ) and the ␤6-␣7 loop (at Ser 338 ) remained formed, which caused the split between the ␤1-␣1 loop and ␣1-helix to remain in place. In the fourth simulation, 7-fold coordination of the Ca 2ϩ -occupied ADMIDAS was restored when the second carboxylate oxygen from Asp 127 , instead of Asp 126 , joined the Ca 2ϩ coordination sphere. Strikingly, in this case the split between the ␤1-␣1 loop and ␣1-helix became exaggerated, relative to the split configuration that is typical of "wild-type" Ca 2ϩoccupied complexes (supplemental Fig. S2).

Integrin Activation Dynamics
Together with experimental evidence linking Ca 2ϩ at the ADMI-DAS with integrin activation inhibition (26) and mutations of the ADMIDAS with changes in the headpiece hinge angle (27), these simulations suggest that the structural basis for activation inhibition by Ca 2ϩ is as follows: Ca 2ϩ at the ADMIDAS promotes bidentate coordination by a ␤1-␣1 loop Asp, which in turn favors the outwards split between the ␤1-␣1 loop and ␣1-helix, which in turn hinders ␣1/␣7 helix junction formation, which obstructs hinge opening (Fig. 5).
Force Accelerates Formation of the ␣1/␣7 Helix T-Junction-Previously, we used SMD simulations to show that tensile force accelerates the opening of the integrin headpiece hinge (17). Here, we asked if force also accelerates the formation of the T-junction between the ␣1 and ␣7 helices. In nine separate SMD simulations, force was applied to the Mg 2ϩ -occupied integrin complex after 1 ns of equilibration, at which point the ␣1/␣7 T-junction was not yet formed and the hinge was still closed. Forces of 100 to 900 pN were applied via the C terminus of the FnIII 10 module. Forces in the hundreds of piconewtons were chosen to observe structural changes over a computationally feasible time frame. To vary the direction of strain, force was applied with the C terminus of only the hybrid domain fixed (at 100, 150, and 250 pN) or with the C termini of both the ␣ subunit ␤-propeller domain and the hybrid domain fixed (at 500, 700, 800, and 900 pN).
We found that force induced the ␣1/␣7 T-junction to form within the 5-ns time frame in seven (of nine) simulations. These include two in which the hinge was free to open under force and five in which the closed hinge was constrained. Even in the latter case, force varied the constrained hinge angle by as much as 15°. For representative depictions, see Fig. 6 and supplemental Videos 3 and 4.
We found that force could induce the T-junction to form prior to inducing a significant change in the hinge angle. This is illustrated in Fig.  6 by the decrease in the Leu 333 -Leu 134 C ␤ -atom distance (black trace) that occurs prior to the corresponding hinge increase (orange trace). When the ␣1/␣7 T-junction did not form under force (two of nine times) the split between the ␤1-␣1 loop and the ␣1 helix displayed greater prevalence and the ␤1-␣1 loop and ␤6-␣7 loop remained in close proximity (supplemental Video 5, in contrast to supplemental Videos 3 and 4).
Once the ␣1/␣7 T-junction was formed in a mechanically strained integrin complex, it remained stable (seven simulations). In three additional simulations, forces of 500, 800, and 900 pN were applied for several nanoseconds after the junction was formed. Even at these high pulling forces, the junction was not disrupted.
Next, we asked if force could induce formation of the ␣1/␣7 T-junction in Ca 2ϩ -occupied complexes, because we did not observe this event under equilibrium conditions (three of three Ca 2ϩ -occupied equilibrations). When the hinge of Ca 2ϩ -occupied complexes was opened under forces of 100, 150, FIGURE 3. The ␣1 helix split. The helical conformation of the middle of the ␣1 helix (residues 132 to 135) was assessed by residues over time using the "ssrecalc" command in VMD. A, when the hinge opened spontaneously, all of these residues were maintained in the ␣ helix conformation. B, in the unliganded structure, Asn 133 fluctuated in and out of the ␣ helix conformation, resulting in a split helix. C, when Ca 2ϩ replaced Mg 2ϩ at the LIMBS and ADMIDAS, Gln 132 and Asn 133 fluctuated in and out of the ␣ helix conformation, again resulting in a split helix. In all of our simulations, we found the prevalence of the split ␣1 helix to be a characteristic feature of complexes in which the hinge did not open on the nanosecond time scale (e.g. unliganded or Ca 2ϩ -occupied complexes).

FIGURE 4. How Ca 2؉ in place of Mg 2؉ at the ADMIDAS inhibits formation of the ␣1/␣7 helix T-junction.
A, top, the coordination shell of the Ca 2ϩ -occupied ADMIDAS. Water molecules are shown in cyan licorice representation. When Ca 2ϩ replaces Mg 2ϩ , the second carboxyl oxygen from Asp 126 joins the coordination sphere and a split becomes more prevalent between the ␤1-␣1 loop and ␣1 helix (see Fig. 3C). Under equilibrium conditions, the hinge does not open on the nanosecond time scale and the ␣1/␣7 T-junction does not form in Ca 2ϩ -occupied complexes (compare with Fig. 2A). B, when the charge on the second Asp 126 oxygen was turned off, in Ca 2ϩ -occupied complexes, creating a Ca 2ϩ -specific Asp 126 -charge off complex, the most prevalent conformation of the ␤1-␣1 loop and ␣1 helix switched to an uninterrupted helical structure. Also, the ␣1/␣7 T-junction formed (bottom, black trace) and a greater increase is evident in the hinge (middle, orange trace). When force was applied to this Ca 2ϩ -specific Asp 126 -charge off complex, hinge opening was accelerated relative to Ca 2ϩ -occupied complexes (supplemental Fig. S3). For further depiction of how the bidentate ADMI-DAS coordination governs the ␤1-␣1 loop conformation, see supplemental Fig. S2. DECEMBER 25, 2009 • VOLUME 284 • NUMBER 52 and 250 pN, according to the same protocol described above for Mg 2ϩ -occupied complexes, formation of the ␣1/␣7 T-junction accompanied hinge opening two of three times (at 100 and 250 pN).

Integrin Activation Dynamics
Hinge opening in Ca 2ϩ -occupied complexes occurred with a delay relative to Mg 2ϩ -occupied complexes. Thus, we asked if the same structural feature regulates hinge opening and ␣1/␣7 T-junction formation under force that we found to regulate this pathway under equilibrium conditions: namely, the ␤1-␣1 loop/ADMIDAS coordination. We found that when the same constant forces were applied to the Ca 2ϩ -specific Asp 126charge off complex, force-induced hinge opening was accelerated toward the time scale of Mg 2ϩ -occupied complexes (supplemental Fig. S3). In other words, like ligand-induced hinge opening, force-accelerated hinge opening was found to be regulated by the influence of the bidentate ␤1-␣1 loop ADMIDAS coordination.
Alterations of the ␤1-␣1 Loop Shield the Force-bearing RGD-Integrin Bond from Water Access-Having established a link between ␤1-␣1 loop contacts and the ␣1/␣7 T-junction, we next asked if the ␤1-␣1 loop could regulate the lifetime of the principal RGD-␣ v ␤ 3 integrin bond. Previous SMD studies have shown that force-bearing bonds break when free water molecules compete with the bonding partners to form hydrogen bonds (21,28). This indicates that the lifetime of a force-bearing bond can be increased by small structural perturbations that lessen access of free water to the bond. In the ␣ v ␤ 3 integrin complex, the principal force-bearing bond occurs between a carboxyl oxygen on Asp RGD and the MIDAS ion (21). Thus, we investigated whether the ␤1-␣1 loop can regulate access of free water molecules to this bond.
The top of the ␤1-␣1 loop packs flush against the force-bearing Asp RGD -MIDAS ion bond via direct contacts with both the MIDAS ion and Asp RGD (Fig. 7A). The contact with Asp RGD involves the "second" carboxyl oxygen (i.e. the Asp RGD side chain oxygen that is not bound to the MIDAS ion) and Tyr 122 . In our simulations, we found that this contact was sometimes lost in Mg 2ϩoccupied complexes when the second Asp RGD side chain oxygen joined the first in coordinating the MIDAS ion (Fig. 7D). Although formation of the bidentate MIDAS ion coordination by Asp RGD may not be physiologically relevant, as it is not currently found in crystal structures, we used this structural event to compare the influence of the Asp RGD -Tyr 122 bond on the access of water molecules to the forcebearing Asp RGD -MIDAS ion bond.
Analysis of the number of collision events between free water molecules and the MIDAS ioncoordinating Asp RGD oxygen (O␦-2) reveals a considerable shielding when the Asp RGD -Tyr 122 bond is in place (Fig. 7, C versus E). During SMD simulations begun with the bidentate Asp RGD -MIDAS ion coordination in place, it either remained formed (six of nine) or switched to monodentate Asp RGD -MIDAS ion coordination together with Asp RGD -Tyr 122 bond formation (three of nine). In the latter cases, we found that the frequency of water attacks on the ion-coordinating Asp RGD oxygen decreased when the Asp RGD -Tyr 122 bond was formed, as illustrated in Fig. 7B. These findings suggest a critical role for the Asp RGD -Tyr 122 bond, and thus of the ␤1-␣1 loop, in regulating the lifetime of the principal RGD-␣ v ␤ 3 integrin bond.

DISCUSSION
Although experiments have linked the switch to high affinity binding in the integrin headpiece with the opening of the ␤A/hybrid domain hinge (2)(3)(4), the structural basis for this link has remained unclear. On the one hand, the first crystal structure of the ligand-bound integrin headpiece (13) was formed under conformational constraints (17) and displays a closed ␤A/hybrid domain hinge. On the other hand, this crystal structure displays a conformation of the ligand-bound binding pocket that is nearly identical to that of the open-hinge ␤ 3 integrin headpiece structure, which was solved subsequently by cocrystallization with ligand (12). Thus, these crystallographic snapshots beg the question of how the conformation of the ␤A/hybrid domain hinge governs structural events in the vicinity of the ␤A domain binding pocket.
The simulations presented here were based on a semi-equilibrated crystal structure, in which activating structural changes local to the binding pocket had already occurred as a result of ligand binding but long-range propagation of the activation signal had been arrested by domain-domain contacts in the preexisting crystal lattice. Previously, we showed with molecular dynamics simulations based on this structure that a small inwards (versus outwards) movement of the ␣1 helix promotes (versus impedes) opening of the ␤A/hybrid domain hinge (17). Here, we have shown how a contact of the ␣1 helix that occurs approximately halfway between the RGD-binding pocket and the ␤A/hybrid domain hinge links the signal propagation between these two sites. This contact is an ␣1/␣7 T-junction between hydrophobic amino acid side chains in the middle of the ␣1-helix, the top of the ␤6-strand, and the top of the C-terminal ␣7-helix. We found that formation of this junction results from RGD occupation of the binding pocket and, as expected for allosterically regulated proteins, can also be induced by ␤A/hybrid domain hinge opening. A characteristic of this ␣1/␣7 T-junction is the fusion of the ␣1-helix and ␤1-␣1 loop, resulting in one continuous helix structure flanking the side of the ␤A domain (Figs. 1, H and J, and  3A). A change in ADMIDAS ion coordination, resulting from Ca 2ϩ in place of Mg 2ϩ , inhibits formation of the ␣1/␣7 T-junction by stabilizing a break between the ␤1-␣1 loop and the ␣1-helix (Figs. 3C and 4A). Mechanical force was found to accelerate formation of the ␣1/␣7 T-junction (Fig. 6). Taken in consideration with mutational, monoclonal antibody, electron micrographic and tomographic experiments that link the ␤A/hybrid domain hinge opening with integrin activation (2)(3)(4), together with recent evidence of force-accelerated integrin activation (9, 10), formation of the ␣1/␣7 T-junction appears to be a key event along the integrin activation pathway.
Recently, five activating point mutations have been identified in the ␤-subunit headpiece domain (29). A preliminary analysis of the solvent accessible surface areas of these residues in our simulations indicates changes before and after ␣1/␣7 T-junction formation that are consistent with the model proposed here. These include decreased fluctuations in residues located close to junction formation on the ␣1 and ␣2 helices and increased fluctuations in residues located at the ␤A/hybrid domain interface. 4 Residues of the ␣1/␣7 Helix Tjunction Are Highly Conserved-The hydrophobic ␣1/␣7 T-junction residues Leu 138 , Leu 333 , Leu 343 , Val 247 , Ile 307 , and Ala 309 are completely conserved across the 8 integrin ␤-subunits, whereas Leu 134 and Ile 344 are conserved across 6 ␤-subunits. These residues are shown in green and red surface representations in Figs. 1 and 6. In the integrin ␤ 8 A domain, both Leu 134 and Ile 344 are replaced by the similarly 4 E. Puklin-Faucher and V. Vogel, unpublished data. FIGURE 6. Force-induced ␣1/␣7 helix junction formation. Traces from three separate SMD simulations are shown, with corresponding structural snapshots beneath. In each case, force was applied to FnIII 10 after 1 ns of equilibration and was found to induce ␣1/␣7 T-junction formation. Force was applied to the C terminus of the FnIII 10 module with the C terminus of the ␤ hybrid domain fixed. When force was applied at 500 pN (E and F), opening of the hinge angle was constrained by also fixing the C terminus of the ␣ subunit ␤ propeller domain. All snapshots are from the 3rd ns under force. In the bottom panels, each snapshot is overlaid with the starting crystal structure (shown in gray).
hydrophobic Val, whereas in the ␤ 4 A domain, they are replaced by hydrophobic residues Met and Leu, respectively.
In accord with the premise of the model described here for the ␣ v ␤ 3 integrin, experiments have shown that mutations of hydrophobic residues on the ␣1 and ␣7 helices influence activation of the ␣ 5 ␤ 1 integrin as well (30). Although the key residues identified experimentally differ from the ␣1/␣7 T-junction residues identified here, there is a large degree of hydrophobic redundancy across highly conserved residues of the ␣1and ␣7-helices (supplemental Fig. S4). For example in a preliminary study, we found that when Leu 131 , located above Leu 134 in the ␣1 helix, is computationally "mutated" to Ala that a different kind of ␣1/␣7 T-junction formed. In this case, a hydrophobic residue from the top of the ␣7 helix, Ile 344 , became inserted into a hydrophobic pocket in the ␣1 helix between Leu 138 and Met 142 and the headpiece hinge angle displayed an increase of ϳ20°after 5 ns of equilibration (data not shown). Experimentally mutating the equivalent Leu residue in the ␣ 5 ␤ 1 integrin was shown to promote activation (30). Thus, we suggest that a variety of putative hydrophobic interactions between the ␣1 and ␣7 helices could influence the dynamics of the ␤1-␣1 loop and thereby influence binding affinity of the RGDbinding pocket.
A Unique Role for the ADMIDAS in ␤A Domain Allostery-Although all 24 known integrin heterodimers include the three headpiece domains investigated here, 9 types of integrins additionally contain the ␣A domain (also called the I domain) inserted into the extracellular end of the ␣-subunit. In those integrins that contain it, the ␣A domain mediates ligand binding directly. Although the ␣A and ␤A domains share considerable functional and structural homology, our simulations indicate that their mechanisms of activation differ. Although ␣A domain activation has been shown to proceed via the ratchetlike movement of the ␤6-␣7 loop through an underlying pocket of the hydrophobic protein core (31), ␤A domain activation involves a distinctive, key role for the ␣1 helix and the ADMI-DAS, a metal ion binding site that is not present in the ␣A domain (17,26). Our simulations indicate that the change in bidentate ADMIDAS coordination that accompanies the change in occupation from Mg 2ϩ or Mn 2ϩ to Ca 2ϩ regulates ␤A domain activation via an influence on ␣1/␣7 T-junction formation.
Recent mutational studies of the ␤1-␣1 loop ADMIDAS coordinating residues have shown disparate effects on integrin binding affinity. In one case, mutations equivalent to Asp 126 induced firm adhesion in flow chamber experiments of ␣ 4 ␤ 7 integrins (32). In the other case, the corresponding mutation induced weakened adhesion in solid phase binding assays of FIGURE 7. The ␤1-␣1 loop regulates the stability of the force-bearing Asp RGD -MIDAS bond. A, the force-bearing bond between O␦-2 of Asp RGD (black) and the MIDAS ion (red) occurs at the top of the ␤1-␣1 loop, shown in green, schematic, and transparent surface representations. When the Asp RGD and Tyr 122 hydrogen bond is formed (dashed line), the ␤1-␣1 loop packs in tightly with the Asp RGD side chain. Water molecules within 5 Å of the O␦-2 atom are shown in blue van der Waals representation. B, when the bond between Asp RGD -Tyr 122 was formed under force, as evident from the trace of the heavy atom distance (right side, traced in black), shielding of force-bearing O␦-2 atom from attack by free water molecules was increased (left side, traced in blue). Water contact events to the O␦-2 atom were counted for each time step (picosecond intervals) by identifying the number of water molecules within 3 Å of the O␦-2 atom. C, when the Asp RGD -Tyr 122 bond was already formed under force, this bond remained formed (right side, black trace) and the O␦-2 atom remained shielded from attack by water molecules (left side, blue trace). When instead both of the O␦-1 and O␦-2 atoms coordinated the MIDAS ion (D) and the Asp RGD -Tyr 122 bond did not form (E, right side), the Asp RGD atom that forms the force-bearing bond with the MIDAS ion, O␦-2, was subject to a relatively higher level of water molecule attacks. ␣ 5 ␤ 1 (33) and ␣ 2 ␤ 1 integrins (27). In this case, mutationally induced activation inhibition was partially rescued by monoclonal antibodies 12G10 or TS2/16, which bind at or near the ␣1 helix (2,26). In all cases, ADMIDAS was found to play a critical role in relaying allostery from the ligand binding pocket to the rest of the integrin molecule. Notably, cell rolling assays probe integrin binding under tensile force conditions imposed by fluid flow, whereas solid phase binding assays probe integrin binding interactions in the absence of tensile forces. Thus, these seemingly disparate findings are also consistent with the model described here, whereby we predict that force applied to Ca 2ϩoccupied complexes could override the ␣1 helix split that otherwise occurs, allowing the ␣1/␣7 T-junction to form. In other words, this is how force could reverse the inhibitory influence of the Ca 2ϩ -occupied ADMIDAS. It should also be pointed out that ADMIDAS residues are completely conserved across all integrin ␤ subunits except the ␤ 8 subunit, where the two ␤1-␣1 loop/␣1 helix aspartates are replaced by similarly acidic asparagines.
Force-accelerated Allostery and the Integrin Catch Bond-Several receptor-ligand complexes have been shown to increase their binding affinity under force, the hallmark of a catch bond. These include the bacterial adhesin FimH with mannose (34), the myosin-actin motor protein complex (35), the cellular adhesins P-and L-selectin with their respective ligands (36,37) and, most recently, the ␣ 5 ␤ 1 integrin with FnIII 7-10 (10). In each case, a structural change has been implicated that is first induced or accelerated by force and then extends the lifetime of the receptor-ligand bond (for reviews, see Refs. 38 -40). Like integrins, the FimH ligand binding site is allosterically coupled to a distal hinge region (40). Also like integrin catch bond complexes, activation of the FimH and selectin catch bond complexes can be stabilized by point mutations or monoclonal antibodies that promote the opening of the headpiece hinge, even if their epitopes are located distal to the ligand-binding site. Yet, for each of the receptor-ligand catch bond complexes currently known, the structural changes in the binding pocket that enable the extension of the bond lifetime remain experimentally undetermined (36,41).
Here, we have shown that in the case of the RGD-␣ v ␤ 3 integrin complex, formation of a bond between the ␤1-␣1 loop (Tyr 122 ) and Asp RGD will extend the lifetime of the principal RGD-integrin bond by shielding it from frequent attacks by free water molecules (Fig. 6). This finding is consistent with available pharmacophoric refinement data, which identified the same bond as critical for RGD-integrin binding stability (42). It is also consistent with a general mechanism for allosteric proteins, whereby the dual character of regions of low and high structural stability in the binding site leads to the propagation of structural change via the region of low structural stability (43). In accord with this general mechanism, the ␤1-␣1 loop both connects directly to the ␣1/␣7 T-junction and, as shown in comparison across crystal structures of various presumed affinities (Fig. 1, D-H), is the only loop in the immediate vicinity of the Asp RGD -MIDAS ion bond that displays significant structural change. This model is also consistent with the recent finding that monoclonal antibodies that activate integrins by their influence on the ␣1 helix also shift the integrin catch bond to a lower force regime (10).
As evidenced in crystallographic data (12,13), formation of the ␣1/␣7 T-junction occurs only once the ␤A domain is free to re-equilibrate after a binding event. Equilibration of the ␤A domain might be hindered by domain-domain contacts such as those formed in the bent-knee conformation of inactivated integrins. Although the two available liganded crystal structures display near identical configurations of the liganded ␤ 3 integrin binding pocket, differing only in backbone hydrogen bonding of the ␤1-␣1 loop, the ␣1/␣7 T-junction is absent where the ligand was soaked into preformed integrin crystals (13) that where constrained in the closed-hinge conformation by the pre-existing crystal lattice (17). In contrast, the junction is formed where the ligand was co-crystallized with the integrin headpiece in the absence of the legs domains (12). Thus, we propose that the physiological role of the bent-knee conformation is to tune, by domain-domain contact, the height of the energy barrier necessary to transition the headpiece to the high affinity state via hinge opening and ␣1/␣7 T-junction formation. This may be the mechanism by which force accelerates the integrin activation process.