Molecular switches of the κ opioid receptor triggered by 6′-GNTI and 5′-GNTI

The κ opioid receptor (κOR) is a member of G-protein-coupled receptors, and is considered as a promising drug target for treating neurological diseases. κOR selective 6′-GNTI was proved to be a G-protein biased agonist, whereas 5′-GNTI acts as an antagonist. To investigate the molecular mechanism of how these two ligands induce different behaviors of the receptor, we built two systems containing the 5′-GNTI-κOR complex and the 6′-GNTI-κOR complex, respectively, and performed molecular dynamics simulations of the two systems. We observe that transmembrane (TM) helix 6 of the κOR rotates about 4.6o on average in the κOR-6′-GNTI complex. Detailed analyses of the simulation results indicate that E2976.58 and I2946.55 play crucial roles in the rotation of TM6. In the simulation of the κOR-5′-GNTI system, it is revealed that 5′-GNTI can stabilize TM6 in the inactive state form. In addition, the kink of TM7 is stabilized by a hydrogen bond between S3247.47 and the residue V691.42 on TM1.

Scientific RepoRts | 6:18913 | DOI: 10.1038/srep18913 provide active and inactive models of the GPCR based on the action of the so-called "molecular switches" [32][33] buried in the receptor. TM6 of the κOR was observed to rotate about 4.6 o on average in the κOR-6′ -GNTI complex. Helices TM7 and TM2 highly responded to the activation signal in 6′ -GNTI systems too. In the simulation of the κOR-5′ -GNTI system, 5′ -GNTI could stabilize TM6 in the inactive state form.

Results
Initial interaction modes obtained from molecular docking. The docking poses of 6′ -GNTI and 5′ -GNTI are almost identical with each other as shown in Fig. 2. Strong electrostatic interaction is formed between the protonated nitrogen on the heterocyclic ring and D138 3.32 in each complex. The phenolic hydroxyl group of both ligands interacts with H291 6.52 bridged by two water molecules and the cyclopropyl group interacts with W287 6.48 . Meanwhile, Y139 3.33 and Y320 7.43 also contribute to the stabilization of the two ligands at the orthosteric site. Because of the difference in the substitution site of the guanidinium group in 6′ -GNTI and 5′ -GNTI, different interaction patens were observed when the two ligands bind to the κOR. The guanidinium group of 5′ -GNTI is located in the cavity formed by the residues on TM6 and TM7 and forms three hydrogen bonds with E297 6.58 , whereas the guanidinium group of 6′ -GNTI is positioned in a cavity formed by the residues on TM5 and TM6 and forms only one hydrogen bond with the residue E297 6.58 . In addition, the docking poses of the two GNTIs are almost identical with the conformation of NTI adopted in the crystal structure of the NTI-δOR complex. With the structures of 5′ -GNTI or 6′ -GNTI in the orthosteric pocket of the κOR and the κOR in the apo state, we built three systems, 5′ -GNTI-κOR, 6′ -GNTI-κOR, and the κOR in the apo state, to be used for the following simulations.
Overview of the MD simulation results. In our simulations, the systems reached the equilibrium at about 100 ns as indicated from the root mean square deviation (RMSD) values in the simulations (Fig. 3A). The RMSD values of the κOR with respect to its crystal structure range from 2.0 to 3.6 Å (Fig. 3A). The ligands undergo comparably small conformational changes as indicated by their RMSD values, which range from 0.2 to 1.2 Å (Fig. S1). In addition, we also calculated the root mean square fluctuations (RMSFs) of the systems to analyze the fluctuations of the receptor. The loop section of the receptor exhibits a much larger fluctuation as compared to the conserved TM section (Fig. 3B). IL (intracellular loop) 2, EL (extracellular loop) 2 and IL3, show significant fluctuations, which are related to the fact that there are many residues on these three loops. The large fluctuation of the loop section reflects the significant conformational changes of the residues adjacent to these loops.
Ligand cluster analysis was performed on the two 600 ns GNTIs-κOR systems, and two sub-state conformations were obtained under a 0.2 Å RMSD cut-off (Table 1). The main diversity between the two clusters was the tilt direction of indole ring of GNTIs (Fig. S2). One direction was outward if viewed from the extracellular side, whereas the other one was inward. Interestingly, the tilt direction of indole ring was inward among Cluster 1 of 6′ -GNTI-κOR system, while that was outward in 600 ns 5′ -GNTI-κOR system. Meanwhile, the population of Cluster 1 ran up to a half in 6′ -GNTI-κOR system, whereas that was in the majority, amount to 81.4% in 600 ns 5′ -GNTI-κOR complexes. Given the corresponding changes of guanidinium group (Fig. S2) and their important roles in GNTIs binding, we think the above-mentioned differences were mainly due to the diverse performances of guanidinium group between the two GNTIs-κOR systems.

Movements of the transmembrane helices.
Comparisons of GPCR crystal structures in the inactive and active states have revealed several conserved switches to explain the inhibition or activation mechanisms of GPCRs 34 . One of the conserved switches is the large-scale rearrangements of the TM helices, including the rotation    of TM6, the movement of TM5, the slight rotation and upward movement of TM3, and the inward movements of TM7 and TM1 35 . TM6 and TM3 are at the heart of any common activation pathways, because they are coupled with critical conserved residues and are in direct contact with other helices except for TM1 36 . We thus analyzed the conformational changes of TM6 and TM3. A sharp kink of TM6 has been observed in the crystal structure of the κOR as well as in other family A GPCR structures with P 6.50 as the pivot point [36][37] . Such a kink is preserved in our MD simulations (Fig. 4B). However, the kink angle of TM6 with P289 6.50 as the pivot point varies with the simulation time. It changes from 160 o to 140 o at the initial stage of the simulation and then increases to about 159 o again ( Fig. 4B and Table S1). Besides the kink of TM6, TM6 in the 6′ -GNTI-κOR complex counterclockwise rotates about 4.6 o on average, even 11.4 o at 600 ns conformation viewed from the extracellular side of the receptor (Fig. 4A). This observation is consistent with the published results 35 . In contrast, the TM6 kink in the 5′ -GNTI-κOR complex or the apo κOR form was found to be stabilized at about 160 o and the counterclockwise rotation of TM6 was not observed from our simulations. We thus believe that the variation of the kink angle and the counterclockwise rotation of TM6 can be used to explain the different behaviors of the κOR triggered by 5′ -GNTI and 6′ -GNTI.
Normal mode analysis (NMA) 38 is an efficient method for predicting inherent flexibilities in biological macromolecules. We performed NMA on the typical structures of the initial crystal structure, principal component and final 600 ns state in Fig. 4A, to detect the intrinsic motions of κOR. The low-frequency modes of κOR produced by the NMA reflected the global motions of the receptor and were often related to biological functions 39 . The first two lowest-frequency motion modes (modes 1 and 2) on the final 600 ns state and principal component were relevant to the transition from active-like to inactive (Fig. 5). Interestingly, NMA of the initial crystal structure revealed that the inactive state had an intrinsic potential to change back to the active-like conformation. However, the modes relevant to this transition were only among the fifth and third lowest-frequency motion modes. Therefore, unless an external force or ligand was present, κOR would favour the inactive state within the circuit. The consistency between the results of MD and NMA supports the efficiency of both methods in studying the large-scale motions of TM6 domains.
The conformational changes of TM3 and TM7 in the simulations can also be used to disclose the role of 5′ -GNTI and 6′ -GNTI in the activation of the κOR. The binding of GNTIs causes large conformational change of TM3 in the GNTIs− κOR complexes as shown in our simulations. In the simulation of the 5′ -GNTI-κOR complex, we observed a clockwise rotation of TM7 if viewed from the extracellular side (Fig. S4A). We attribute the clockwise rotation of TM7 to the fluctuation of ECL3 (Video S1A). ECL3 is mainly composed of hydrophilic residues, such as Ser or His. These residues form strong interactions with the guanidino group of 5′ -GNTI, which is responsible for the clockwise rotation of TM7.

Comparison of the binding interactions of GNTIs with the κOR.
Hydrogen bonds between the residues D138 3.32 , Y139 3.33 , H291 6.52 , and E297 6.58 on the receptor and the ligands are formed for most of the time in the simulations of the 5′ -GNTI-κOR and 6′ -GNTI-κOR complexes ( Table 2 and Fig. 2). E297 6.58 forms two HBs  with the ligands and contributes significantly to the stabilization of the ligands at the orthosteric site. In addition, the protonated nitrogen in either 6′ -GNTI or 5′ -GNTI is not able to form any HBs with D138 3.32 (Table 2). This can be considered as a common binding mechanism of opioid towards ORs 40 .
Contribution of I294 6.55 to the conformational change of TM6. Detailed analysis of the trajectory from the simulation of the 5′ -GNTI-κOR system showed that the minimum distance between C α of I294 6.55 and 5′ -GNTI is mainly stabilized at about 5.5 Å ( Fig. 6 and Table S1). The corresponding distance in the 6′ -GNTI-κOR complex fluctuates between 4 and 6 Å, with the average at about 5.2 Å. The average minimum distance between C α of I294 6.55 and 6′ -GNTI is about 0.3 Å shorter than that between C α of I294 6.55 and 5′ -GNTI ( Fig. 6A and Video S1B). The shorter distance between C α of I294 6.55 and 6′ -GNTI can raise the steric clash between 6′ -GNTI and I294 6.55 . The existence of the steric clash between I294 6.55 and 6′ -GNTI can be further indicated from the temporal evolutions of an angle between residue I294 6.55 and TM6 (Fig. 6B). Because of the steric clash, this angle undergoes a significant fluctuation between 40 o and 120 o , with an average value at about 63 o . On the contrary, the corresponding angle is stabilized at about 81 o in the 5′ -GNTI-κOR complex. The RMSF values also reflect that the fluctuation of the angle in the 6′ -GNTI-κOR complex is larger than in the 5′ -GNTI-κOR complex (Fig. 3B). In the simulation of the κOR in the apo state, this angle fluctuated between 40 o and 100 o freely (Fig. S5A).
The strong steric clash between I294 6.55 and 6′ -GNTI has a significant influence on the conformational states of E297 6.58 (Fig. 6C). The angles between residue E297 6.58 and TM6 is mainly stabilized at about 77 o in the 6′ -GNTI-κOR complex, whereas in the 5′ -GNTI-κOR complex it is at about 86 o on average. The large influence could also be reflected from the diverse performances of ligand cluster among 6′ -GNTI-κOR interaction (Table 1 and Fig. S2).
Conformational change of TM7. A sharp kink on TM7 was observed in the crystal structure of the κOR 26 with P327 7.50 as the pivot point 36,37 . Although this kink was preserved in our simulations, the kink angles behave differently in the three systems. Temporal evolutions of the kink angle of TM7 indicate that the kink in the 5′ -GNTI-κOR system is more stable than that in the other two systems. The kink angle is stabilized at about 150 o in the 5′ -GNTI-κOR complex ( Fig. 7 and Table S1). However, the kink angle in the 6′ -GNTI-κOR complex or in the apo form shows an obvious deviation. Specifically, this kink angle in the 6′ -GNTI complex changed from about 150 o to 130 o at the initial stage of the simulation. The temporal evolution of the kink angles could be used to explain the responses of TM7 caused by GNTIs.
A detailed analysis of the simulation results reveals that the kink is stabilized by a hydrogen bond formed between S324 7.47 and V69 1.42 ( Table 2). The occupancy rate of this hydrogen bond is 24% from the simulation of the 5′ -GNTI-κOR system, which is much higher than that in the 6′ -GNTI-κOR system (2.7%) and in the apo κOR (6.1%). A recent study on CRF 1 R also proved the significance of the kink in explaining the antagonistic mechanism 41 . Allosteric action mediated by the sodium ion. The role of the sodium ion in the function of some GPCRs has become an attractive topic recently. The crystallographic structure of the adenosine A2A receptor (A2AR) at an ultra-high resolution provides us with the first evidence of the binding of the sodium ion to the GPCR, followed by the crystallographic structures of the β1-adrenergic receptor and δOR 28,42,43 . In the δOR, the sodium ion is positioned around the negatively changed D 2.50 , and coordinated by two polar residues (N131 3.35 and S135 3.39 ) and two water molecules 28 . Many studies have been carried out to investigate the allosteric activation of GPCRs mediated by the sodium ion 44 . A recent molecular dynamics simulation of the μOR suggested that a sodium ion penetrates into the allosteric pocket from the extracellular side of the receptor to perform its allosteric effects on the receptor 40 . The dynamic properties of the sodium ion in the apo-κOR have also been described in the latest work of Filizola et al. 45 .
In the crystal structure of the κOR, the sodium ion is not co-crystallized with the κOR. We thus placed a sodium ion in the allosteric pocket of the κOR by referring to the position of the co-crystallized sodium ion in the crystal structure of the NTI-δOR complex 28 . In our simulations, the sodium ion is stablized in the allosteric pocket, reflecting that the position of the sodium ion is reasonable (Fig. 8A).
In our simulations, although the sodium ion is stablized in the allosteric pocket, the residues coordinating the sodium ion undergo conformational changes and result in the fluctuation of the distances between the sodium ion and the residues (Fig. 8A). For example, the distance between the sodium ion and D105 2.50 increased significantly during 180-300 ns in the simulation of the 6′ -GNTI-κOR system due to the conformational change of D105 2.50 (Fig. 8B); D105 2.50 deviated from the allosteric pocket and N322 7.45 was involved in the direct interaction with the sodium ion during 180-280 ns.

HB NO.
5′ -GNTI system 6′ -GNTI  Table 2. Occupancies of the hydrogen bonds in the two 600 ns GNTIs-κOR systems and the apo-κOR system. a the symbol "-" represents that there is no hydrogen bond interaction.
Scientific RepoRts | 6:18913 | DOI: 10.1038/srep18913 The 3-7 lock. Formation of a hydrogen bond between D138 3.32 and Y320 7.43 , which is named as the 3-7 lock in the following text, is suggested to play a key role in the activation of the κOR. To investigate the 3-7 lock in detail, we monitored the temporal evolution of the distance between D138 3.32 and Y320 7.43 (Fig. 8C). In the 5′ -GNTI-κOR system, a hydrogen bond between D138 3.32 and Y320 7.43 formed at about 80 ns and broke at about 500 ns of the simulation. The occupation rate of the hydrogen bond was 69.6%. In the simulation of the 6′ -GNTI-κOR system, a hydrogen bond between D138 3.32 and Y320 7.43 formed at about 340 ns and was preserved until the end of the simulation, resulting in an occupation rate of 44.4%. On the contrary, the 3-7 lock was not found in the whole simulation of the κOR in the apo form. We thus contribute the formation of the hydrogen bond between D138 3.32 and Y320 7.43 to the binding of 5′ -GNTI or 6′ -GNTI to the orthosteric pocket of the κOR.
The antagonistic effects of 5′-GNTI. To investigate whether 5′ -GNTI can trigger a change of the conformation of the κOR induced by 6′ -GNTI to the inactive conformation, we docked 5′ -GNTI to the last snapshot the κOR generated in the simulation of the 6′ -GNTI-κOR system (Fig. S6), and performed another MD simulation with a time scale of 100 ns. Interestingly, we found that TM6 gradually returned to the antagonistic state under the action of 5′ -GNTI (Fig. 9A) as indicated by the kink angle of TM6. This angle changed from about 160 o to 145 o at the beginning of the simulation, and then increased to 157 ± 4 o ( Fig. 9B and Table S1). Structural alignments of TM6 also indicated that 5′ -GNTI triggered the conformational change of the κOR induced by 6′ -GNTI to the inactive state gradually. In addition, the tilt direction of the indole ring was outward in Cluster 1 of the additional 5′ -GNTI, which was identical with the 600 ns 5′ -GNTI-κOR system (Table 1 and Fig. S2).

Discussion
In our simulations, we found that E297 6.58 in the κOR is directly interacting with 6′ -GNTI. This interaction, together with the steric effect from I294 6.55 , contributes to the rotation of TM6 as well as the movements of other TMs. In contrast, this ligand shows no agonistic effect towards the μOR and δOR 23 . From sequence alignments of the κOR, μOR and δOR, we can easily find that E 6.58 in the κOR corresponds to K 6.58 in the μOR and to W 6.58 in the δOR 46 . On the other hand, it has been suggested that both K 6.58 in the μOR and W 6.58 in the δOR are not directly interacting with the guanidinium group of 6′ -GNTI, which leads to the absence of the agonistic effect of this ligand towards the μOR and δOR 23 . All these observations demonstrate that the direct interaction between the residue E297 6.58 and 6′ -GNTI plays an indispensible role in the activation of the κOR and the residues E297 6.58 and I294 6.55 function as a critical pair in the activation of the κOR.  Besides the nontrivial function of E297 6.58 and I294 6.55 in the activation of the κOR, large conformation changes of TM7 and TM2 were also identified in our simulations as indicated by the kink angles of TM7 with P327 7.50 as the pivot point (Fig. 7) and TM2 with D105 2.50 as the pivot point (Fig. 8B). Such conformational changes can further help us understand the mechanism of κOR activation. In addition, D105 2.50 is the key residue in positioning the sodium ion in the allosteric pocket. Therefore, we believe that the sodium ion also contributes to the activation of the κOR. S324 7.47 is located adjacent to the pivot point P327 7.50 on TM7 and forms a hydrogen bond with a residue on TM1 in the simulation of the 5′ -GNTI-κOR system ( Fig. 7 and Table 2). Because of the formation of the hydrogen bond, the conformational change of TM7 is hampered. As a result, TM7 is stabilized in its initial conformation and leads to that the κOR adopts the inactive conformation. We thus suggest S324 7.47 plays an important role in stabilizing the κOR at the inactive state.
The 3-7 lock is found to be formed in our simulations of the 5′ -GNTI-κOR and 6′ -GNTI-κOR systems. In contrast, such a lock is breaking-down in the crystal structure of the JDTic-κOR complex. Both 5′ -GNTI and 6′ -GNTI are derivatives of morphinan, while the ligand JDTic 47 is a pure antagonist with (3R,4R)-3,4-dimethyl-4-(3-hydroxyphenyl)piperidine as the scaffold. We can thus use the ligand in the orthosteric pocket to explain the formation and breaking-down of the 3-7 lock in the κOR -the derivatives of morphinan can induce the formation of the 3-7 lock, while a pure antagonist results in the breaking-down of such a lock in the κOR.
In addition, we observed a large rotation of TM6 around the pivot point P289 6.50 in the simulation of the 6′ -GNTI-κOR system, which is an indispensible step for the activation of the κOR. In our opinion, the receptor needs to visit a set of intermediate states for its activation from the inactive state to the fully active state 48 . The rotation of TM6 allows the receptor to visit one of the key intermediate states for the activation of the κOR. However, we did not observe full activation of the receptor probably due to the limitation of the available computational resources.

Conclusions
In this work, we studied the molecular switches of the κOR triggered by 5′ -GNTI and 6′ -GNTI using molecular dynamics simulations. We observed about 4.6 o rotation of TM6 on average in the κOR-6′ -GNTI complex. Detailed analysis of the simulation results revealed that E297 6.58 and I294 6.55 play a crucial role for the rotation of TM6. On the other hand, we found that the hydrogen bond between S324 7.47 and residue V69 1.42 on TM1 contributes to the stabilization of the κOR in the inactive state as revealed from the simulation of the κOR-5′ -GNTI complex.

Materials and Methods
Protein preparations. The published crystal structure of human inactive JDTic-bound κOR (PDB code: 4DJH), was obtained from an engineered κOR mutant protein where part of the intracellular loop between transmembrane helices TM5 and TM6, was replaced by T4 lysozyme (T4L) 26 . In order to do molecular dynamics simulations on the wild type receptor based on the crystal structure of the T4L mutant κOR dimer, we removed the T4L from the mutant receptor structure, the monomer B and other unnecessary parts. Then we reconstructed the loop of the remaining chain A by adding the missing residues S262 and T302-S303-H304-S305-T306 using the loop refinement protocol in Discovery Studio 3.5; 10 loops were generated and the most reasonable one was chosen for receptor construction. According to the latest crystal structure of inactive NTI-complexed δ OR, ion Na+ was exactly located at the allosteric pocket position 28 .
The above three-dimensional integrated κOR structure was then imported into the Schrödinger software package. The protein structure was prepared with Protein Preparation Tool (ProPrep) in the Schrodinger 2012 suit software. Asn, Gln, and His residues checked for protonated states automatically in ProPrep. Hydrogen atoms were added into the κOR crystal structure at the physiology pH environment by the PROPKA tool in Maestro with optimized hydrogen-bond network. No non-standard protonation state of the amino acids was found. The pocket Grid file of the κOR was generated by 20 Å around residue D138 3.32 with the Receptor Grid Generation module, and docked with Glide Docking module (Glide 5.8) 49,50 . The Van der Waals (vdW) scaling was set to 0.8 for nonpolar atoms of receptor and ligand. During docking, the number of docking output was set as 10000 poses per docking run at most. The most reasonable conformation was picked out for molecular dynamics simulations.
MD preparations. Two 5′ -GNTI-κOR complexes, one 6′ -GNTI-κOR complex and one apo-κOR system, were built for the simulations. A POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) bilayer with the surface area of 75 × 75 Å 2 on the X-Y plane was generated under Charmm36 force field by VMD program (Version 1.9.1). For each system, the receptor was first embedded into the POPC bilayer using our in-house program pre-aligned in the OPM (Orientations of Proteins in Membranes) database [51][52][53][54] . Thereafter, Pre-equilibrated 103 POPC lipids coupled with 11067 TIP3P water molecules in a box ~75 × 75 × 100 Å 3 were used solvate the protein. Lipid molecules within 0.85 Å of the heavy atoms on the protein structure and water molecules in the bilayer were removed. 51 Na+ and 59 Cl-ions were used to produce neutral systems with 0.15 M NaCl in the water phase of GNTI systems, whereas the number of Cl-ions was 57 in apo-κOR system. We described the protein using CHARMM36 force field with cmap correction.
Ligand force field was generated by Paramchem webserver, a program coupled with CHARMM Force field [55][56][57] . Small molecules with a correct configuration were imported into Paramchem webserver and force field parameters of these compounds were then obtained through a script CHARMM General Force Field (CGenFF).

Molecular dynamics simulations.
All simulations were performed using Gromacs V.4.6.5 58 and the CHARMM36 parameters for all compositions. In the first simulation step, the system was subjected to a 10000-step energy minimization with 1000.0 kJ/mol/nm as the force threshold. Then, the systems were gradually heated from 0 K to 310 K followed by 50 ps initial equilibration at constant volume and temperature at 310 K (NVT). An additional 1 ns equilibration was performed at constant pressure and temperature (NPT ensemble; 310 K, 1 bar) with two thermostats (stabilizing temperature independently for protein-ligand system, and the lipids-water-ions system) at 0032 ps time steps. vdW and short-range electrostatic interactions were cut off at 12 Å. Long-range electrostatic interactions were computed by the Particle Mesh Ewald (PME) summation scheme. The MD simulations of the additional 5′ -GNTI system was performed for 100 ns, while all the other systems were performed for about 600 ns under NPT conditions with Parrinello-Rahman pressure coupler methods and Nose-Hoover thermostat for temperature coupling. The time step for MD simulation was 2 fs and the integrator leap-frog algorithm was employed. While the MD simulations of the additional 5′ -GNTI-κOR system was performed for 100 ns, all the other systems were performed for about 600 ns using the NPT ensemble with the Parrinello-Rahman pressure coupling and the Nose-Hoover temperature coupling methods.
Analysis of simulations. RMSD and RMSF calculations, ligand cluster statistics, hydrogen bond analysis, angle and distance evolutions were produced by the program Gromacs. The interval time of trajectory calculations, including RMSD, RMSF, angle, distance and principal component analysis, was 500 ps in 600 ns systems, whereas that in 100 ns 5′ -GNTI systems was 100 ps. RMSD values were calculated through comparing to the initial simulation conformations. Principal components analysis was carried on through the g_covar tool. All the smooth curves in Figs 4, 7 and Fig. S3 were fitting groups, which were identical with corresponding Angle/RMSD calculations in color.
Scientific RepoRts | 6:18913 | DOI: 10.1038/srep18913 To determine the sub-state conformations that close ligands bound to κOR, structural clustering was performed using the g_cluster tool in Gromacs with the linkage algorithm based on the RMSD of each ligand molecule after alignment of the Cα atoms in the transmembrane helices in each frame to the starting structure 59 . A 0.2 Å RMSD cut-off was chosen because it best captured spatially distinct clusters and allowed the top clusters to be representative of the predominant binding sites explored. The clustering was performed on the simulation frames with 100 ps interval time. The populations of each cluster are given in Table 1. We examined the two most populated clusters and calculated the RMSDs for each ligand found in these clusters (Table 1).
Angle analysis was carried on by g_angle tool in Gromacs. The specific input parameters were as follows. The kink angle of TM6 was measured with atom C α of V285 6.46 , P289 6.50 and F293 6.54 ; The angle between I294 6.55 and TM6 was measured with atom C δ and C α of I294 6.55 and atom C α of W287 6.48 , while that angle between E297 6.58 and TM6 was computed with atom C δ and C α of E297 6.58 and atom C α of F293 6.54 ; The kink angle of TM7 was measured with atom C α of Y320 7.43 , P327 7.50 and A331 7.54 ; The angle between D105 2.50 and TM2 was measured with atom C γ and C α of D105 2.50 and atom C α of I98 2. 43 .