Conformational Stability Analyses of Alpha Subunit I Domain of LFA-1 and Mac-1

β2 integrin of lymphocyte function-associated antigen-1 (LFA-1) or macrophage-1 antigen (Mac-1) binds to their common ligand of intercellular adhesion molecule-1 (ICAM-1) and mediates leukocyte-endothelial cell (EC) adhesions in inflammation cascade. Although the two integrins are known to have distinct functions, the corresponding micro-structural bases remain unclear. Here (steered-)molecular dynamics simulations were employed to elucidate the conformational stability of α subunit I domains of LFA-1 and Mac-1 in different affinity states and relevant I domain-ICAM-1 interaction features. Compared with low affinity (LA) Mac-1, the LA LFA-1 I domain was unstable in the presence or absence of ICAM-1 ligand, stemming from diverse orientations of its α7-helix with different motifs of zipper-like hydrophobic junction between α1- and α7-helices. Meanwhile, spontaneous transition of LFA-1 I domain from LA state to intermediate affinity (IA) state was first visualized. All the LA, IA, and high affinity (HA) states of LFA-1 I domain and HA Mac-1 I domain were able to bind to ICAM-1 ligand effectively, while LA Mac-1 I domain was unfavorable for binding ligand presumably due to the specific orientation of S144 side-chain that capped the MIDAS ion. These results furthered our understanding in correlating the structural bases with their functions of LFA-1 and Mac-1 integrins from the viewpoint of I domain conformational stability and of the characteristics of I domain-ICAM-1 interactions.

Introduction b 2 integrins, as important cellular adhesion molecules, play critical roles in a variety of pathophysiological processes, such as inflammation and tumor metastasis [1,2]. Binding of the two members of b 2 integrin, lymphocyte function-associated antigen-1 (LFA-1 or a L b 2 ) and macrophage-1 antigen (Mac-1 or a M b 2 ), to their common ligand intercellular adhesion molecule-1 (ICAM-1) mediates leukocyte-endothelial cell (EC) adhesion in inflammation cascade [3,4,5]. How do the two b 2 integrins work together during this process? While a prevailing view had emphasized for years that LFA-1 and Mac-1 cooperatively promoted leukocyte firm adhesion, their distinct roles recently attracted much attention especially when an indispensable step so-called intraluminal crawling was added to the paradigm [6]. For neutrophils, LFA-1 mainly mediates the slow rolling and firm adhesion and Mac-1 governs the following intraluminal crawling [6,7,8]. For monocytes, LFA-1 supports the firm adhesion and crawling in unstimulated condition while Mac-1 is responsible for TNF-astimulated crawling [4,9]. For lymphocytes, multiple functional states of LFA-1 with different affinities are required to facilitate the cell rolling, adhesion, and crawling since no Mac-1 molecules are presented onto the cell surface [10,11,12]. It is also noticed that LFA-1 dominates the long-distance, meandering crawling for resident monocytes patrolling but Mac-1 mediates the shortdistance, straight crawling for activated monocytes and neutrophils in the effective (trans-)migration [4,6,9]. Moreover, LFA-1 is able to interact with ICAM-1 in three known affinity states [13] but Mac-1 does not support neutrophil attachment to ECs without fMLP activation [14]. Distinct dynamics of b 2 integrin activation and functioning are observed, e.g., LFA-1 responses within first 300 s after IL-8 stimulation whereas Mac-1 begins to be engaged in from 350 s after activation [15]. In short, LFA-1 is versatile in inflammation cascade whereas Mac-1 is a specific crawling mediator, and the two molecules differ in their activation dependence and dynamics.
Even with the distinct functionality of the two b 2 integrins, they possess the high similarity in structure [2,16,17]. On one hand, the two molecules have the identical b 2 subunit that is critical in transmitting the allostery bi-directionally [2]. This subunit involves I-like domain pocket capturing the bottom conservative glutamate of a 7 -helix of a subunit I domain [18,19,20], downward displacement of a 7 -helix of I-like domain, swinging out of hybrid domain [21,22], extending on the knee to translocate laterally b subunit lower leg, and exerting forces on cytoskeleton by b subunit cytoplasmic domain binding to talin or kindlins [23,24]. On the other hand, the two molecules have highly similar a L and a M subunits sharing 34% identical amino acids (calculated using Basic Local Alignment Search Tool (BLAST)). An I (inserted) domain of about 180-190 residues is located on the top of each a subunit. Both the a subunit I domains contain a metal ion-dependent adhesion site (MIDAS) for ligand binding, adopt Rossmann fold with 7 a-helices surrounding 6 central b-strands [16,25], and share similar allosteric pathways by relating MIDAS re-orientation to a 7 -helix downward movement, which facilitates the capture by Ilike domain pocket and transfers the conformational change to b subunit [25,26]. Currently it is hard to correlate their distinct functionality to their highly-similar structure and allosteric patterns. One possible way to bridge the gap is to elucidate the structural bases at microscopic scale. Since the allosteric pathway of the two molecules starts from or ends at their a subunit I domain or the ligand binding domain, it is speculated that the conformational differences between their I domains, especially movement of the a 7 -helix should be, at least partially, responsible for the differences in their biological functions.
In the current work, we performed molecular dynamic (MD) simulations to test the structural evolution of LFA-1 and Mac-1 I domains upon existing crystal structures. Conformational stability was compared between two molecules as well as among three affinity states. Impact of ICAM-1 ligation on conformational stability and I domain-ICAM-1 interactions were also analyzed. Mechanical features of the complexes were tested using steered MD (SMD) simulations. Our results provided a new insight in the structure-function relationship for both LFA-1 and Mac-1.

Methods
Two sets of molecular systems were employed in the MD simulations. The first set consists five I domain alone systems including high affinity (HA)/low affinity (LA) Mac-1 I domain (PDB codes of 1IDO [16] and 1JLM [26]) and HA/intermediate affinity (IA)/LA LFA-1 I domain (PDB codes of 1T0P [27], 1MJN/ 1MQ8 [25], and 1LFA [17]), which is used to evaluate the conformational stability. The second set is composed of five I domains from the first set ligated respectively with ICAM-1, which is used for elucidating the impact of ligand presence on I domain conformational stability as well as the binding strength of integrinligand complex. Since Mn 2+ ion does not exist in physiological condition and is always referred as an activator [13,28,29], we replaced it by Ca 2+ ion in the MIDAS of LA Mac-1/LFA-1 to avoid ionic activation possibility. IA LFA-1-ICAM-1 complex was adopted from the crystal structure with D1 domain of residue ID 1 to 85, and HA/LA LFA-1-D1 complexes were obtained by aligning HA/LA LFA-1 I domain to IA LFA-1 I domain. Noting the sequence and structure similarity between LFA-1 and Mac-1 I domains and between ICAM-1 D1 and D3 (residue ID 186 to 284) domains [30], HA/LA Mac-1-D3 complexes were constructed by aligning HA/LA Mac-1 I domain and D3 domain (PDB code of 2OZ4 [31]) onto IA LFA-1 I domain and D1 domain, respectively, using VMD [32] Plugin MultiSeq [33] software.
Each simulation system was built up by solvating the target molecule(s) into a rectangular water box and neutralized with ,100 mM Na + and Cl 2 ions. NAMD program [34] with CH ARMM22 all-atom force field [35] was used for the simulations. An integration time step of 1 femtosecond (fs) and the periodic boundary conditions were applied in the simulations. A smooth (10-12 Å ) cutoff and the Particle Mesh Ewald (PME) method were employed to calculate van der Waals forces and full electrostatics, respectively. 300 K heat bath was manipulated under Langevin thermostat, and 1 atm pressure was controlled by Nosé-Hoover Langevin piston method. Prior to the equilibration process, energy minimization with 10000 steps of fixed backbone atoms followed by additional 10000 steps with all atoms free, and system heating from 0 to 300 K at 30 K increment every 5 picosecond (ps) were performed. 10 nanosecond (ns) or longer equilibration was performed for each system. SMD simulations were also conducted to unbind the LFA-1/ Mac-1-ICAM-1 complex [36]. Complex conformations resulted from the above equilibrations served as the initial conformation for SMD simulations. Here N-, C-, or N-/C-terminal C a atoms of I domain were fixed independently and constant force of 200, 800, 1200 or 1600 piconewton (pN) was applied on C-terminal C a atom of ICAM-1 D1 or D3 along the vector from the fixed atom or geometry center to the pulled end. All the simulation set-up was summarized in Table 1.
Geometry parameters were defined to represent the characteristics of conformational change. Root mean squared deviation of entire I domain (global RMSD) was used to quantify I domain stability, and average RMSD of each residue over time was employed to identify the unstable amino acids within I domain.
Aligning the stable core residues of trajectory conformations to that of crystal structure of HA, LA (for both LFA-1 and Mac-1), or IA (only for LFA-1) I domain respectively, the RMSD of a 7 -helix was used to probe if the trajectory tends to switch among different affinity states. MIDAS residues [2] were regarded within the primary coordination sphere if the side-chain oxygen atom was close to the metal ion by 3.5 Å and within secondary sphere if not. Complex lifetime was defined from SMD simulations as the time interval required to separate the MIDAS ion away from side-chain oxygen of D229 of D3 or E34 of D1 around 10 Å . VMD program was used for data analysis and conformation presentation [32].
Simulations were conducted on the DeepComp 7000 supercomputer at the Computer Network Information Center, Chinese Academy of Science. Altogether, 690-ns MD and 783-ns SMD simulations were performed in this work, and spent 5,795 h on 64 compute nodes with two 3.0 GHz quad-core Intel Xeon processors.

Conformational stability of LFA-1/Mac-1 I domains
The crystal structures of HA/LA Mac-1 I domains and HA/IA/ LA LFA-1 I domains were superposed in Figs. 1A and 1B, respectively. Two states of HA and LA Mac-1 and one state of LA LFA-1 structures are defined for wild-type molecules while two states of HA and IA LFA-1 structures stem from the mutated molecules which are stabilized by a disulfide in b 6 -a 7 loop and a disulfide locking C-terminals of a 1 and a 7 -helix, respectively. It is noted that both the disulfide bonds distort the a 7 -helix conformation. Two common features were observed for the structure of IA and HA states by comparing with LA state: MIDAS re-orientation with a 1 -helix inward movement and the distinct a 7helix axially downward shift [2,26] (Figs. 1A-B). Thus, we first conducted the equilibration simulations to test their conformational stabilities. Stable conformations were visualized for the four structures except of LA LFA-1 I domain. To confirm this interesting observation, the simulation was repeated ten times for LA LFA-1 and three times for each of the other four. Global RMSDs relative to their respective crystal structures were calculated and averaged for each system. RMSD of LA Mac-1 I domain yielded as small value as ,1 Å (purple) and that of HA Mac-1 and HA/IA LFA-1 I domains retained ,2 Å (black, blue and green), while it fluctuated from 2 to 4 Å for LA LFA-1 (red) (Fig. 1C).
To further understand the intramolecular bases of structural instability of LA LFA-1 I domain, global RMSD of each trajectory and average RMSD of each residue were compared between LA LFA-1 and LA Mac-1. It was found that LA Mac-1 state remained stable with a small RMSD of 0.9-1.5 Å in three repeats ( Fig. 2A) while RMSD for LA LFA-1 state was much higher (1.2-5.5 Å ) and dramatically diverse from one to another simulation (Fig. 2B). Average RMSD of each residue exhibited similar trends as global one did. In fact, all the residues for LA Mac-1 state yielded low RMSD values (Figs. 2C and 2E) except of those moderate fluctuations around a 1 -b 2 loop (L164-S167), b 4 -a 5 loop (D242-G251), and b 5 -a 6 loop (V271-S277) (grey stripes in Fig. 2C). By contrast, all the residues of LA LFA-1, except of those around central b-strands and a 4 -helix, seemed to be unstable (Figs. 2D and 2F), especially for MIDAS residues and a 7 -helix (E293-K305; grey stripe in Fig. 2D). RMSDs of a 7 -helix residues were significantly high (1.3-14.1 Å ) and remarkably fluctuating in repeated simulations, implying that there exists the diversified conformations. Together, the conformational stability is distinct between two b 2 integrin I domains and from one state to another, of which LA LFA-1 state is most likely unstable.
Conformational diversity of LA LFA-1 I domain Next we analyzed those diversified conformations observed in LA LFA-1 I domain equilibrations. Here the core residues of equilibrated snapshots were aligned to that of crystal structure of HA/IA/LA LFA-1 I domain, respectively, and RMSD of a 7 -helix was calculated with time which is used to define the movement of a 7 -helix. Ten repeated simulations were classified into four categories: 1) a 7 -helix stabilized in LA state (as observed from two of ten trajectories). Here RMSD was low when aligning equilibrated snapshot to LA LFA-1 crystal structure (dark red and dark blue lines), high when aligning to HA LFA-1 structure (red and blue lines) (Fig. 3A), and intermediate when aligning to IA LFA-1 structure (4-6 Å below that of HA LFA-1; lines not shown for clarity), which implied that the equilibrated a 7 -helix (red newcartoon in opaque) retained similar conformation as that of LA crystal structure (silver newcartoon in opaque) (Fig. 3A9). 2) a 7 -helix swung in and moved downward slightly (Fig. 3B9) (from three of ten trajectories). RMSDs relative to LA (dark-colored lines) or to HA (colored lines) reference exhibited a transition phase with time to approach each other followed by a stable phase at .6 ns (Fig. 3B). These conformations were different from the established HA/IA/LA crystal structures, and more and longer simulations are needed to confirm whether they are stable ones. 3) a 7 -helix moved downward significantly (from two of ten trajectories). This movement resembles much to the state transition allostery of I domain with a descending phase of RMSDs to HA reference (colored lines) and an ascending phase to LA reference (dark-colored lines) (Fig. 3C). One simulation run was extended to 30 ns, and the resulted steady-going phase in the last 20 ns indicated that the allosteric conformation was stable (Fig. S1). Unlike the forced or ligand-induced a 7 -helix downward-movement previously described [20,22], the equilibrations reported here first visualized the spontaneous conformational change of a 7 -helix with a remarkable displacement of 8-11 Å for each residue (Fig. 3C9). 4) a 7 -helix swung outwards (from three of ten trajectories). Here RMSDs were significantly high and fluctuated from one to another run of simulation (Fig. 3D). Defining the angle of a 7 -helix swing-out by variation of the vector along C a atom of K305 to C a atom of E293, 30u to 45u rotations were observed for the equilibrated a 7 -helix ( Fig. 3D9), which are much smaller than that found in a crystal structure (106u) [37].
While a 7 -helix presented multiple patterns of conformational change, corresponding MIDAS conformation re-orientated similarly. It was found from the ten simulations that the ion in MIDAS (vdW, silver and red for crystal and equilibrated structures, respectively) moved inwards about 2.1 to 2.8 Å (Figs. 3A9-D9). The coordination between the ion and MIDAS residues altered in the same way (Figs. 3A0-D0). In crystal structures, S139, S141, and D239 coordinate the ion in the primary coordination sphere and D137 and T206 in the secondary sphere. During the equilibrations, S141 moved backward to secondary sphere, and D137 and T206 moved forward to coordinate the ion tightly (residues shown in licorice).
Allostery from LA to IA LFA-1 I domain It was noticed that four categories of conformational states derived from LA LFA-1 I domain. The first reserved its original LA state (Fig. 3A), the forth might be artificial due to the spatial repulsion of neighboring domains (Fig. 3D), and the second was not proved by any crystal structures (Fig. 3B). The most striking is the third one that underwent a large downward movement of a 7 -helix (Fig. 3C). The resulted low a 7 -helix RMSD value to IA or HA LFA-1 crystal structure implied the transition to IA or HA state (Fig. S1). To further understand the structural features of the third category of equilibrated LA LFA-1 I domain conformations, the conjunction between a 1 -and a 7 -helices was tested and a novel zipper-like hydrophobic junction was identified. Here, the sidechains of F292 and F299 (blue in licorice) in a 7 -helix orientated downwards and inserted between I150 and F153 and between V157 and L161 (yellow in licorice) of a 1 -helix, respectively (Figs. 4A and 4C). Fig. 4A shows the conformation in crystal structure of LA LFA-1 and Fig. 4C shows what it looks like after 30-ns equilibration. C b -atom distances from F153 to F292, D F153Cb-F292Cb , and from L161 to F299, D L161Cb-F299Cb , defined to quantify the extent of the junction formation, decreased within first 10-ns duration followed by a steady phase in another 20-ns equilibration (Fig. 4B), which is consistent with the evolution of the a 7 -helix RMSDs (Fig. S1).
Conformations of those hydrophobic residues were compared among different Mac-1 or LFA-1 I domain states. Three types of zipper-like hydrophobic junction were found in the five crystal structures: the first was defined as an inadequately-zippered one where the a 7 -helix F292/F302 and F299/Q309 residues located apart in which F292/F302 wandered over the zipper head I150/ M153 residues and F299/Q309 over the zipper stalk V157/V160 residues respectively in both LA LFA-1 and Mac-1 (Figs. 4A and 4D); the second was denoted as a over-zippered one where a 7 -helix underwent a dramatic downward movement and translocated F292/F302 residues around respective F153/F156, and F299/ Q309 residues to the zipper tail L161/L164 residues in both HA LFA-1 and Mac-1 (Figs. 4F and 4E); the third was illustrated as a properly-zippered one where the large downward movement of a 7 -helix enforced F292 and C299 (mutated from F299) interacting strongly with their counterpart residues of a 1 -helix and closed the hydrophobic core perfectly (Fig. 4G). It was evident that same hydrophobic upper junction formed in equilibrated LA LFA-1 I domain as mutated IA LFA-1 I domain crystal structure while the difference was found in the lower junctions which were locked by hydrophobic interaction for the former and by a disulfide bond for the latter (Figs. 4C and 4G). In fact, the introduction of a disulfide bond resulted in a distinct upper junction in HA LFA-1 as compared with that in HA Mac-1 (Figs. 4E and 4F). A supplemental simulation by breaking up the disulfide bond in IA LFA-1 via resetting the mutated residues back to the original ones, using SWISS-MODEL workspace [38], also indicated that the large-sized side-chain of F299 pushed the lower half a 7 -helix away to rejoin with the upper half and form the properly-zippered junction (Fig. S2), as the one shown in Fig. 4C.
Time courses of the hydrophobic junction were quantitatively presented by the distance evolving between typical residues of HA/ LA Mac-1 and HA/IA/LA LFA-1 I domains during at least 10-ns simulations (Fig. S3). The results indicated that three states have distinct hydrophobic junctions with different typical residue distances: the HA (A, C) and LA (B, E) states yielded large distances  (Fig. S3). Together, these analyses support that two distinct features of hydrophobic residues exhibited: different states in same b 2 integrin member share different hydrophobic junction conformations and the same state yields the similar hydrophobic junction in different members. And these results also indicate that the third category of equilibrated conformations of LA LFA-1 I domain acts as IA LFA-1 I domain and the allostery is able to take place spontaneously.

Impact of ICAM-1 ligation on conformational stability
The above simulations of LFA-1/Mac-1 I domain alone indicated that LA LFA-1 presented the conformational instability and the spontaneous allostery of state transition but LA Mac-1 retained its original state stable. Meanwhile it has also been reported that the ligand binding could induce the conformational change of a v b 3 headpiece [22,39]. To test the impact of ICAM-1 ligation on conformational stability of LFA-1 and Mac-1 I domain, we further conducted the equilibration simulations for five I domain-ICAM-1 complexes and, specifically, the repeated runs for LA LFA-1-ICAM-1 D1 and LA Mac-1-ICAM-1 D3 complex for comparison. Again, LA Mac-1 I domain remained low global RMSDs to its crystal reference in four simulations whereas ten runs of LA LFA-1 exhibited the unstable features with diverse RMSDs (Figs. 5A-B), confirming that both isolated and ligated LA LFA-1 states were unstable. Similar four patterns of I domain conformation, as those found for I domain alone (cf. Fig. 3), were also observed with different proportions in ten simulations of LA LFA-1-D1 complex, of which four simulations denoted to LA state, three to slight swing-in and downward movement of a 7 -helix, two to large a 7 -helix downward movement, and one to 60u swing-out movement of a 7 -helix (Fig. S4). Moreover, the resulted conformation with large a 7 -helix movement (cf. Figs. 3C and S2C) was stable with steady a 7 -helix RMSDs relative to HA/LA LFA-1 references found in an extended 30-ns equilibration (Fig. 5C). Calculated C b -atom distance of D F153Cb-F292Cb , (black arrow) and D L161Cb-F299Cb (red arrow) also supported the same hydrophobic junction as that of IA LFA-1 (Figs. 5D-E). Thus, these results indicated that ICAM-1 ligation played a limited role in conformational stability and allosteric features of LFA-1 and Mac-1.

The difference between LFA-1 and Mac-1 I domain-ICAM-1 interactions
Finally we compared the difference between LFA-1 and Mac-1 I domain-ICAM-1 interactions. Detailed conformational comparison illustrated that the main difference between HA and LA Mac-1 I domain-ICAM-1 D3 domain interactions attributed to the orientation of I domain S144 side-chain. While there exists free space enabling D229 residue of D3 domain to interact readily with MIDAS ion in HA Mac-1 state (Fig. 6A), the tight interaction of S144 side-chain with MIDAS ion most likely reduced the accessibility of D229 residue (Fig. 6B), even though the ion located ,2 Å more closer to D229 residue for LA Mac-1 than that for HA Mac-1 for the systems built. The ligated HA LFA-1 I domain yielded a distinct MIDAS conformation where b 1 -a 1 loop (blue in newcartoon) pointed inwards to enable S141 residue coordinating the ion in the primary coordination sphere and the re-orientation of b 4 -a 5 loop (purple in newcartoon) induced E241 substitution for D239 residue in the primary sphere (Fig. 6C). Interestingly, MIDAS conformations were similar for ligated IA (Fig. 6D) and LA LFA-1 (four cases in Figs. 6E-H) where D137 and T206 directed forwards to the primary coordination sphere and S141 pointed backwards to the secondary sphere (except for the ligated swung-out one, of which both S139 and S141 pointed backwards). The repeated simulations presented the similar MIDAS conformations and ion-D229/E34 interactions, especially for the wide type systems (Fig. S5), which confirmed, to some extent, the strategy of complex construction in the current work. Remarkably, the simulation was repeated for LA Mac-1-ICAM-1 four times from 10 to 30-ns, in which not even once D229 was able to break in and catch the ion, most likely due to the shielding of S144 (Fig. S5B).
We further tested the mechanical strength of the complex to illustrate if and how the conformational change of I domain would affect the binding function of b 2 integrin to ICAM-1 ligand. Upon the end-point equilibrated structures of those five complexes, SMD simulations of 800 pN constant force were performed by pulling C-terminal C a atom of ICAM-1 D1 or D3 domain along the vector from fixed atom of N-terminal C a atom of LFA-1 or Mac-1 I domain to the pulled end. Since the interaction of MIDAS ion and D229 or E34 residue of D3 or D1 domain provides key resistance to external force, the complex lifetime, defined as the duration to separate MIDAS ion beyond 10 Å from D229 or E34 side-chain oxygen atom, was used to quantify the binding strength. It was found that the lifetime significantly reduced from HA (6.1561.53 ns) to LA Mac-1 I domain (0.1160.04 ns) (1 st and 2 nd bars), which is consistent to that for the shielding of S144 to abrogate ion-D229 binding (Fig. 6I). And comparable lifetime was found for HA (36.53611.16 ns) or IA (35.89614.70 ns) LFA-1 I domain to D1 domain, which was dramatically higher than those for Mac-1 I domain (3 rd and 4 th bars), presumably due to the introduced disulfide bonds. Three cases of LA LFA-1 I domain, corresponding to the first three categories shown in Fig. S4, yielded the intermediate lifetime values of 12.7468.40 (A), 18.3367.01 (B), and 11.5962.69 ns (C), respectively (5 th -7 th bars) (Fig. 6I). The lifetime for another case of LA LFA-1 I domain (D), the fourth category presented in Fig. S4, was not determined, mainly due to that the swing-out conformation of a 7 -helix may not be practical. These results indicated that all states of LFA-1 I domain were able to bind to D1 domain effectively but LA Mac-1 I domain was unfavorable in binding to D3 domain with poor mechanical resistance. Such the nature was further confirmed by varying the fixed end from N-terminal to C-terminal and to jointed N-and Cterminals and altering the external forces from 200 to 1600 pN. In fact, even though the favorable unfolding of a 7 -helix was visualized when only C-terminal C a atoms was fixed, both the fixation of C-terminal and of jointed N-and C-terminals yielded similar feature with that from N-terminal fixation (data not shown), implying the fixation strategy used here is appropriate. Since applying the force of 200 pN is unable to dissociate HA/LA LFA-1-ICAM-1 and HA Mac-1-ICAM-1 complexes in 36-ns simulation but applying a force of 1200 or 1600 pN dissociates the complex too fast, considering the balance between simulation time and conformational details, most of SMD simulations employed a typical force of 800 pN. Total fifty-one SMD tests were done in different cases, as summarized in Table 1

Discussion
This work aimed at elucidating the structural bases for distinct functionality of LFA-1 and Mac-1 by analyzing the conformational stability of their a subunit I domains of different affinity states in the absence or presence of ICAM-1 ligand. The novelty of our work lies in the following aspects: First, LA LFA-1 I domain is found to be more flexible than that of LA Mac-1 (Figs. 1C and 2, 3, 5) and thus, it readily possesses conformational change to implement the versatile function of LFA-1 in inflammation cascade. Second, LA LFA-1 I domain is able to interact stably with ICAM-1 D1 domain and resist applied force effectively (Figs. 6, S5). Combined with the conformational instability of I domain, one possible mechanism is proposed that LA LFA-1 binds ICAM-1 first and then its flexible a 7 -helix of I domain moves downward to connect with b subunit for further conformational modulation, which is crucial to cellular mechanotransduction and biological functions such as slow rolling [7,8]. Recently published MD simulations showed that the interaction between bent a v b 3 integrin and its RGD ligand could bear the pulling and then induce a v b 3 to extend [40]. Third, the stubborn LA Mac-1 I domain is unable to open its pocket properly for ICAM-1 binding with the unfavorable orientation of S144 (Fig. 6, S5), supporting that Mac-1 is insufficiently functional without activation. This may also provide a hint to interpret why Mac-1 is unable to react as fast as LFA-1 does when the inflammation occurs. Together, these simulations offer important structural clues for understanding the functional differences between LFA-1 and Mac-1.
Spontaneous transition from LA state to IA state of wild type LFA-1 I domain is crucial to the conformational variations as well as biological functions of b 2 integrin. On one hand, this is the first time to visualize the spontaneous allostery of LFA-1 I domain in MD simulations (Figs. 3, 4, 5), although it has been reported that applied forces drives the transition of LA to IA and then to HA of I domain [20] and that ligand binding induces the downward movement of a 7 -helix of b 3 I-like domain [22,39]. The minor restriction of a 7 -helix exhibited in newly-crystallized a X b 2 structures indicates that I domain a 7 -helix has enough space for spontaneous allostery within whole molecule [18], so the allostery from LA LFA-1 to IA LFA-1 presented in this study is structurally feasible. Combined with the indication that resting b 3 integrin could spontaneously extend when being solvated in hydrodynamic octylglucoside [41], this new finding provides a clue that IA LFA-1 with both extended legs and IA I domain may exist on resting leukocytes and cooperate with LA LFA-1 in cell adhesions as a result of the balance of multiple conformations. On the other hand, it is further observed from zipper-like hydrophobic junction of IA LFA-1 that the lockage of a 1 and a 7 -helice together is a general mechanism to retain certain conformational state for integrin [20,39]. Upon the similarity of hydrophobic junction conformations and related residues between LA/HA Mac-1 I domains and those of LA/HA LFA-1 (Figs. 4A, 4D, 4E, and 4F), it seems reasonable to predict the existence of stable IA state of Mac-1 I domain with a zipper-like junction similar to that for IA LFA-1 (Figs. 4C, 4G, and S2C). Thus, the finding of the novel IA state of LFA-1 I domain, which spontaneously evolved from LA state, furthers the understandings in structural variations.
It has long been recognized that the conformational change in different affinity states of integrin a subunit I domain lies in two major regions, one is for MIDAS and the other is for a 7 -helix, where the MIDAS conformation usually re-orientates distinctly corresponding to multiple conformational patterns of a 7 -helix. In contrast, our simulations illustrate no significant difference in MIDAS conformations for the residues reserved in LA state and those turned to IA or other states in the absence (Fig. 3) or presence (Fig. S4) of ICAM-1 ligation. Even though further investigations are required to elucidate the subtle difference, this speculation of conformational irrelevance between MIDAS and a 7 -helix regions seems to be reasonable, since these two regions are so flexible in LA LFA-1 that they may work independently in the absence of supporting domains. And the results may also attribute to the limitation of insufficiently long equilibrations and to the uncertainty of conformational transition dynamics.
In the current work, five complex systems were constructed upon high similarity in structure by aligning the target molecules to those crystallized complex available in PDB database, which worked quite well for both LFA-1 and Mac-1. All states of LFA-1 I domain binds to ICAM-1 with the vital coordination of MIDAS ion to E34 of D1 finely formed (Figs. 6C-H and S5C-G). Our SMD simulations of the resulted complexes indicate that the LA states of LFA-1 I domain with distinct conformations are able to bind to its ligand (Fig. 6I), which is consistent with the observation that ICAM-1 bound to LFA-1 at different conformations, including the bent one with the lowest affinity [13]. For Mac-1, HA state I domain is able to bind to ICAM-1 ligand effectively with the key coordination of MIDAS ion to D229 of D3 stably formed and the moderate mechanical resistance (Figs. 6A, 6I, and S5A). LA state I domain consistently prevents its MIDAS ion from binding to D229 of D3 domain in four 10 to 30-ns simulations (Figs. 6B, 6I and S5B), which resembles to the mutation of D229 that completely abrogated the binding of Mac-1 to ICAM-1 [42].
It should also be pointed out that the identity of divalent cation has insignificant impacts on I domain stability and I domain-ICAM-1 interactions for both LFA-1 and Mac-1, since the replacement of Ca 2+ by Mg 2+ retained the similar global RMSDs of I domain and similar I domain-ICAM-1 interaction patterns (Fig. S6). Such the observations are consistent with those crystal studies, in which the conformation of I domain is not altered in the presence or absence of MIDAS ion or by the replacement of MIDAS ion [43,44], as well as with those measurements that isolated I domain is not activated by Mg 2+ or Mn 2+ [28,29]. The structural bases for the effect of Mg 2+ or Mn 2+ activation on wild type LFA-1 but not isolated I domain require further investigations [13,29].
In summary, the flexibility of LFA-1 I domain and the rigidity of Mac-1 I domain are basic features for the two b 2 integrins, which are highly correlated to their allosteric pathways and biological functions. Flexible LFA-1 is usually ready for ligand binding and easy to perform state transition, while the functionality of stubborn Mac-1 requires additional cytokine signaling [14,45]. Thus, these structural analyses suggest that LFA-1 is able to interact quickly in multiple biological functions such as cell rolling, firm adhesion, transient migration while Mac-1 seems to mediate the slow interactions mainly on crawling in inflammation cascade.

Supporting Information
Figure S1 a 7 -helix RMSD of a 30-ns equilibration of LA LFA-1 I domain alone when aligned the core residues to those of HA/IA/LA LFA-1 crystal structures. Noting that the green line is 4 to 6 Å underneath the blue line during first 10 ns, as mentioned in the text, but merges together at the last 18 ns.