Prediction of GABARAP interaction with the GABA type A receptor

Abstract We have performed docking simulations on GABARAP interacting with the GABA type A receptor using SwarmDock. We have also used a novel method to study hydration sites on the surface of these two proteins; this method identifies regions around proteins where desolvation is relatively easy, and these are possible locations where proteins can bind each other. There is a high degree of consistency between the predictions of these two methods. Moreover, we have also identified binding sites on GABARAP for other proteins, and listed possible binding sites for as yet unknown proteins on both GABARAP and the GABA type A receptor intracellular domain.


| INTRODUCTION
The GABA A -receptor associated protein, GABARAP, was first described by Wang et al. 1 It is a protein of 117 amino acids and has a relative molecular mass of 13 900. These authors also determined that it interacted with amino acids 394-411 of the intracellular domain of the γ2-subunit of the GABA A receptor. If this sequence was shortened to 399-411 or 389-402, then the interaction was no longer observed.
These authors also reported that GABARAP 36-117 and GABARAP 1-68 both interacted with the γ2-subunit in the GST pull-down assay, indicating that the interaction domain spanned GABARAP amino acids 36-68. In a subsequent paper, Nymann-Andersen et al. 2 concluded that the octadecapeptide RTGAWRHGRIHIRIAKMD from the GABA A receptor γ2-subunit was necessary and sufficient for interacting with the GABARAP, but the interaction, as determined by the glutathione-Stransferase pull-down assay, was not as high as that given by the tricosapeptide CFEDCRTGAWRHGRIHIRIAKMD. This molecule gave the highest level of activity in the assay.
Knight et al. 3 examined the NMR shift of the GABARAP crosspeaks when the octadecapeptide RTGAWRHGRIHIRIAKMD was present. They noticed that the NMR signals from GABARAP amino acids Coyle et al. 4 measured intrinsic tryptophan fluorescence to study the binding between GABARAP and the γ2-subunit of the GABA A receptor. They used native GABARAP, GABARAP with the first 10 amino acids truncated (ΔN10) and GABARAP with the first 27 amino acids truncated (ΔN27). They found that the dissociation constant between the octadecapeptide RTGAWRHGRIHIRIAKMD and native GABARAP was 1.29 AE 0.09 μM, between the octadecapeptide and ΔN10 was 1.17 AE 0.06 μM, and between the octadecapeptide and ΔN27 was 6.10 AE 0.29 μM. The dissociation constant between native GABARAP and the tridecapeptide RTGAWRHGRIHIR was 3.33 AE 0.34 μM, and between native GABARAP and the undecapeptide GAWRHGRIHIR was 5.52 AE 0.52 μM. These dissocation constants are much smaller than that determined from NMR by Knight et al., 3 and it is still unclear where the source of the large discrepancy lies. 5 The function of GABARAP is most probably 2-fold: anchoring the GABA A receptor to the cytoskeleton, and modulating the function of the receptor. Amino acids near the N-terminal of GABARAP could bind to tubulin, 4 while the amino acids nearer the C-terminal bind to the GABA A receptor. 2 Chen et al. 6 showed that GABARAP caused GABA A receptor clustering, and clustered receptors exhibited lower affinity for GABA (EC 50 increased from 5.74 AE 1.4 μM to 20.27 AE 3.8 μM), and they desensitized less quickly (the desensitization time constant τ increased from 1 second to 2 seconds). Everitt et al. 7 performed electrophysiology experiments and showed that GABARAP promotes the clustering of GABA A receptors, and increases the conductance of the GABA A receptor from below 40 pS to above 50 pS.
Despite all these advances on the interaction between the GABA A receptor and GABARAP, we still do not know the structural details of this interaction. Weiergräber et al. 5 cocrystallized GABARAP with the K1-peptide (sequence DATYTWEHLAWP) and determine the structure to 1.3-Å resolution. They used these data and previous published data to infer the interaction between GABARAP and the GABA A receptor.
In this work, we used experimental structures of the GABARAP and a modeled structure of the intracellular domain of the GABA A receptor, and performed docking simulations. We also carried out simulations of the docked structures. Independently, we also used inhomogeneous fluid solvation theory (IFST) 8,9 to calculate the free energy of displacing all reasonable clusters of water containing 7-18 molecules from the surface of the intra-cellular domain of the GABA A receptor, and from the surface of experimental structures of GABARAP. This information was applied to validate the docking interaction between the GABA A receptor and GABARAP, in the context of surface hydration following the methods of Vukovi c et al. 10 2 | METHODS

| Molecular coordinates
In this research, we used the coordinates of a GABA A receptor model from the work of Mokrab et al. 11 This model used as template the nicotinic acetylcholine receptor (nAChR) structure from the work of Unwin,12 where five intracellular helices were resolved (Protein Data Bank code: 2BG9). Thus, this is the only model of the GABA A receptor that includes part of the intracellular domain. The subunit composition of this receptor is (α1) 2 (β2) 2 γ2.
There exist five stand-alone structures of GABARAP, and their Protein Data Bank codes are, respectively, 1GNU, 1KJT, 1KOT, 1KLV, and 1KM7. 1GNU and 1KJT come from X-ray crystallography experiments, and we chose 1GNU because of its higher resolution of 1.75 Å. 1KOT, 1KLV, and 1KM7 all come from NMR experiments; 1KM7 contains only one conformer, while residues 1-17 in 1KLV could not be located and so we chose 1KOT with 15 conformers. We thus used two structures of GABARAP. One is an NMR solution structure, PDB code 1KOT, 13 and the other is an X-ray crystallography structure, PDB code 1GNU. 3  showed that this tricosapeptide gave full binding to GABARAP. We had tried docking GABARAP to the complete GABA A receptor, but this was rejected by SwarmDock as the GABA A receptor contained too many atoms (14 900 nonhydrogen atoms). Therefore we used only part of the γ2-subunit in the docking. In this work, we did not specify the interface amino acids and only used 'blind' docking. A maximum of five normal modes were allowed for each molecule.

| Docking
SwarmDock produced 468 docks for each GABARAP conformation. The output consisted of 10 764 coordinates of different conformations of GABARAP and the tricosapeptide from the GABA A receptor. The coordinates of the latter were slightly different from the original tricosapeptide coordinates, as SwarmDock flexible docking has changed the structure of both the receptor and the ligand. We used a least-squares fit to superimpose the SwarmDock structure of the receptor onto the original tricosapeptide coordinates; the translation vector and rotation matrix used were noted. The same vector and matrix were subsequently used to move GABARAP to a model of the complete GABA A receptor whose γ2 tricosapeptide position were coincident with that of the tricosapeptide used in the docking. We then tested for steric clashes between GABARAP and the GABA A receptor. If two atoms, one from each protein, were found to be within 1 Å of each other, that dock was rejected.
The results filtered for steric clashes were then selected using the  3 In this paper, Ile 64 was also identified as an important interface amino acid, but its position means that we were unable to obtain any docking poses with Ile 64 at the interface. Criteria 2 and 3 were applied to extract docks consistent with the yeast two-hybrid assay. 1 161 docks were selected after these procedures.
We undertook further filters to select the optimal docks from these 161 docks: we examined the distribution of these 161 docks according to the following seven criteria: 4. The SwarmDock energy score should be in the more favorable half of the energy score distribution. 10. The number of atomic contacts from the receptor to any ligand amino acids should be in the higher half of the corresponding distribution.
In the above criteria, a contact was defined as an atom which was less than the sum of the van der Waals radii of the two atoms +20%. 14,15 A dock was selected from these 161 configurations if all of these additional seven criteria were met.
These seven additional criteria were chosen to enforce that the best ligand structure should have a competitive energy score such that the structure is stable (criterion 4), maintain an overall high contact to the receptor (criterion 5) to multiple sites which are distributed between the upper (criterion 6) and lower (criterion 7) portions of the receptor sequence. The best structures must also reciprocate contact across many sites on the ligand (criterion 8) and the strength of all contacts should be a close and strong as possible on the receptor (criterion 9) and ligand (criterion 10).

| Simulation of GABARAP and intracellular helices
We took two representative docked structures of GABARAP and three intracellular helices of the GABA A receptor, and performed simulations on these complexes. Figure 1 shows the docking of GABARAP to the GABA A receptor.
The two docked structures chosen were 1KOT model 15 dock 54a and 1GNU bbb-conformer dock 41d. Each structure consisted of GABARAP in the docking position beside the intracellular helix of the γ2-subunit of the GABA A receptor, from Asp 413 to Asp 442, together with the intracellular helices of the two adjacent subunits. They were included to provide a more realistic environment for GABARAP. These two helices comprised the α1-subunit of the GABA A receptor from Lys 391 to Ser 417, and the β2-subunit from His 421 to Thr 444. The GABARAP/trihelix complex is shown in Figure 2.
The GABARAP/trihelix complexes were placed in a periodic box with at least 10 Å between the protein and its image.  The time-step was lengthened to 2 fs over 30 000 time-steps, while all main-chain nitrogen atoms of the three helices were tethered with a force constant of 2 kJ/mol/Å 2 . A 40-ns equilibration was carried out on the initialised system, followed by a data collection period of 100 ns. Configurations were output every 2 ps.
We calculated the r.m.s. deviation of the simulated structures from the original starting structure. We also evaluated the distance between the γ2-subunit helix Asp 423 and GABARAP Lys 46, and γ2-subunit Ile 438 and GABARAP Gln 59. As can be seen in Figure 2, Asp 423 is at the membrane end of the intracellular helix, and Ile 438 is at the cytoplasmic end of the helix. We monitor these distances to see if GABARAP stays bound to the GABA A receptor throughout the simulations.

| Free energy change calculations
The molecules were prepared using the CHARMM-GUI freely available on the web (as of the time of writing this article, the address of the CHARMM-GUI is http://charmm-gui.org). 17 The molecular dynamics package NAMD 2 18 was used in this work.
In the simulation of the intracellular helices of the GABA A receptor we first selected atoms from the following amino acids: α1-subunit Lys 391-Leu 422, β2-subunit His 421-Ile 449 and γ2-subunit Asp 413-Ser 443. The helices were placed in a periodic box with at least 10 Å between the protein and its image. The system consisted of 19 708 water molecules, 56 K + ions and 73 Cl − ions to achieve a [KCl] of 0.15 mM. The system comprised a total of 61 857 atoms.
The system was minimized for 10 000 steps with all the protein atoms frozen. Molecular dynamics was initialised for 10 000 timesteps of 0.1 fs each, with all main-chain nitrogen atoms frozen. Langevin dynamics was applied; the thermostat was set with a time constant of 1 ps -1 , and the barostat set with a piston decay time of 10 ps and a piston period of 20 ps. The van der Waals cut-off was 12 Å, and Ewald summation was used for long-range electrostatics. The time-step was lengthened to 2 fs over 30 000 time-steps, while all main-chain nitrogen atoms were frozen. A 2-ns equilibration was carried out on the initialised system. A data collection simulation was then carried out for 5 ns, again with all main-chain nitrogen atoms fixed. Configurations were output every 0.5 ps. We obtained a total of 10 000 configurations of the intracellular helices of the GABA A receptor.
For the simulation of GABARAP, we chose model 3 of 1KOT and the 1GNU structure (AAA) as the starting structures. The 1KOT structure of 117 amino acids was placed in a periodic box with at least 10 Å between the protein and its image; 9161 water molecules, 24 K + ions and 26 Cl − ions were placed in this box. The system consisted of a total of 29 508 atoms. The 1GNU structure of 117 amino acids was placed in a periodic box with at least 10 Å between the protein and its image; 9115 water molecules, 25 K + ions and 27 Cl − ions were placed in this box. The system consisted of a total of 29 372 atoms.
These systems were minimized for 10 000 steps with all mainchain nitrogen atoms frozen. Langevin dynamics was applied; the thermostat was set with a time constant of 1 ps -1 , and the barostat set with a piston decay time of 1 ps and a piston period of 2 ps. The van der Waals cut-off was 12 Å, and Ewald summation was used for longrange electrostatics. The time-step was lengthened to 2 fs over 40 000 time-steps. The system was then equilibrated for 2 ns. Data collection was carried out for 5 ns, again with all main-chain nitrogen atoms frozen, with configurations output every 0.5 ps. We obtained a total of 10 000 configurations for each model of the hydrated GABARAP.
The MD trajectory for the GABA A receptor was processed as described by Vukovi c et al. 10 First, hydration sites as defined by Haider and Huggins 19 were created on all surface regions of the GABA A receptor. The hydration sites were time averaged water molecules assigned positions, densities and occupancies. 20,21 Hydration sites with a radius of 1.2 Å were picked starting from the densest patch of water in order of decreasing density and no sites were picked within 2.4 Å of an already existing site. Next, an IFST calculation for the free energy was carried out for each of the hydration sites according to IFST described in Vukovi c et al. 10 IFST had previously been used on water molecules around proteins where the proteins are involved in binding small ligands [22][23][24] and in protein-protein interactions. 25 All 10 000 snapshots of the protein sampled at 0.5 ps intervals were used to calculate the free energy difference associated with hydrating each site with a single water molecule. These free energy differences were mostly negative because solvation was favorable.
At this stage some hydration sites were removed to improve the efficiency of the combinations algorithm. Hydration sites inside the ion channel of the GABA A receptor were removed; the ion channel was aligned to the z-axis, the positions of all protein atoms were converted to cylindrical coordinates with a height z, and a radius and angle in the xy-plane. The cylindrical mid-plane of the protein atoms as a function of height and averaged over angle was found by fitting a quadratic polynomial to the protein atom data. Hydration sites on the inside of this mid-plane were removed. Hydration sites with coordinate z > − 48 Å were also removed as this region was close to the lipid bilayer in the full GABA A receptor model. Then a combinatoric search scheme was employed to search for up to the best 1000 clusters containing from 7 to 18 hydration sites within 12.5 kJ/mol of the best cluster. The search was run three times with these parameters, the first time searching for "near" clusters with hydration sites at most 3.1 Å away from nonhydrogen atoms and 3.6 Å away from hydrophobic nonhydrogen atoms, the second time searching for "regular" clusters with hydration sites at most 3.6 Å away from nonhydrogen atoms and 4.1 Å away from hydrophobic nonhydrogen atoms, as originally performed by Vukovi c et al. 10 The third search was for "far" clusters with hydration sites at most 4.1 Å away for nonhydrogen atoms and 4.5 Å away for hydrophobic nonhydrogen atoms. These three ranges were selected to observe how the hydration patches changed on variation of the hydration site cutoff distance from the protein that is, the degree to which bulk-like distal waters are included in hydration patches.
The method used by Vukovi c et al. 10 predicts ligandability of drug molecules to a protein, and advances in combinatoric search allow clusters of this size to be found. These authors conclude that, for a small peptide, clusters of 30 hydration sites may need to be considered. Finding clusters with volumes commensurate with the ligand in this case is computationally infeasible, especially as GABARAP is much larger than a small peptide. As the free energy change of displacing hydration sites relative to bulk water atoms tends to zero at distances as small as 7 Å-8 Å from the surface, 10 one could instead search for a clustering of clusters with the most favorable displacement free energy scores to estimate candidate regions for larger objects to bind, namely proteins. This method was employed for the GABA A receptor. The set of hydration sites within the best 1000 clusters for each size of 7 to 18 hydration sites were filtered, and turned into hydration patch data for all three classes of clusters, "near," "regular" and "far." For GABARAP, multiple "regular" passes were made of the hydration patch combinatoric search, and after each iteration, the hydration sites associated with patches identified previously were removed. There were 5 passes for the 1KOT file and 4 passes on the 1GNU file, after which no more sites could be found.
The first-pass sites take the least energy to displace and hence are the most displaceable and the fifth-pass ones are the least displaceable.

| Docking
SwarmDock produced 10 764 docks, and 161 docks were selected according to the first three criteria described in the previous section.
Using seven additional criteria, we identified 11 docks, two of them coming from 1GNU and nine from 1KOT. The configuration of these docks are shown in Figures 1 and 3, and the coordinates are deposited  In Table 2

| Simulation of GABARAP and intracellular helices
The r.m.s. deviation of the simulated structures is shown in Figure 6.
The 1GNU structure shows a slightly higher r.m.s. deviation than the 1KOT structure, but the deviations remain stable throughout the simulation. The distances between the key amino acids are shown in, respectively, Figures 7 and 8. In both the 1KOT and 1GNU simulations, the distance between Lys 46 and Asp 423 is shorter than that between Gln 59 and Ile 438, and the latter also shows less variation than the former. We can rationalize this observation by noting that Lys 46 and Asp 423 are both charged, whereas Gln 59 is a polar amino acid and Ile 438 is a nonpolar one. In the 1GNU simulation, the distance between Lys 46 and Asp 423 atoms were generally below 5 Å, but sometimes it increased to above 10 Å. Visual inspection of the structures show that, in the case of the larger distances, the main chain of GABARAP has moved further away from the γ2-subunit intracellular helix and there is a dihedral angle change in the side chain of Lys 46. All this can cause the Nζ-atom of Lys 46 to move by as much as 5 Å.
It can be seen that GABARAP interacts in a stable manner with the GABA A receptor intracellular helices.

| Hydration of the GABA A receptor intracellular domain
The top panel of Figure 9 shows the most displaceable "close" hydration sites near the intracellular domain of the γ2-subunit of the      Diagram showing a model of the intracellular helices of the GABA A receptor; the γ2-subunit is shown in cyan. In the top panel, the hydration sites from the best "close" clusters of sizes 7-18 as red, orange and yellow spheres. In the middle panel, the "regular" clusters are shown, while in the bottom panel, the "far" clusters are shown. The hydration sites are shown in color as described in Table 3 [Color figure can be viewed at wileyonlinelibrary.com] The three classes of hydration site clustering, "close," "regular," and "far" all show a set of most displaceble clusters: those primarily situated on the γ2-subunit (red), those between the γ2 and β2-subunits (orange) and those on the lower, cytoplasmic portion of the β2-subunit (yellow). In addition to this a patch was found on the β2-subunit (green) in the "regular" and "far" classes. As can be seen in Table 3, the red patch on the γ2-subunit is the easiest to displace on average across all classes.
The amino acids within 5 Å of the red patch, in order of highest contact to lowest contact (name followed by frequency), are listed in  Table 5 and Figure 11 show the location of the main hydration patches on the surface of GABARAP. It is useful to divide these patches up into two: those with known binding proteins and those without. We define two kinds of hydration sites, "overlapping" sites where the hydration patch is directly over the binding face of the protein, and "surrounding" sites where the hydration patch is near the binding face of the protein. Note that these GABARAP hydration sites FIGURE 10 Diagrams comparing the overlaid main chains of predicted docking positions of GABARAP (multiple colors), and all the hydration sites (red, orange, yellow, green) identified in this work from "close," "regular," and "far" searches. The γ2-subunit is shown in cyan.

| GABARAP hydration
The hydration sites are shown in color as described in Table 3 [Color figure can be viewed at wileyonlinelibrary.com] are different from the GABA A receptor hydration sites but some of them share the same color codes. There are a large number of hydration sites not involved in the binding of these three proteins. However, when we examine the crystallographic datasets, we find that these sites are involved in dimerisation or trimerisation. It is still unknown how GABARAP dimerises in the cell, so it is uncertain if these crystallographic oligomers represent the natural state of oligomerisation. Table 6  form of GABARAP is in a dimer form, and also simultaneously binds tubulin and the GABA A receptor. We have access only to structural data of the "closed" form of GABARAP so the overlapping site identity is less certain than other sites.

| Summary
Using SwarmDock and subsequent filtering based on available experimental evidence, we have identified 11 docked poses of GABARAP.
These docked positions are all very similar, and they are all in contact with highly displaceable GABA A receptor hydration sites. We note that the GABA A receptor amino acids in Table 4 match those in Table 2 very well. Hydration analysis of water molecules around GABARAP has identified a large number of possible binding sites, and some of them are found to match the binding face for the GABA A receptor γ2-unit intracellular domain (see Figure 12). Figure 13 shows a global comparison of the results from docking and from hydration patch analysis.
However, in both cases, we have discovered hydration patches that might suggest a binding site, but we could not find any known binding molecule. In the case of the GABA A receptor intracellular domain, there are hydration patches next to the β2-subunit (green and yellow patches in Figure 9) which are distant from the GABARAP-binding site, and do not seem to bind any known protein.  receptor binding site and the site for other proteins. It is interesting to note that the first-pass and third-pass sites are often involved in binding autophagy-related proteins, but the second-pass sites are used for dimerisation and trimerisation under crystallography conditions. Figure 11 also shows the hydration patches classified around GABARAP. The hydration patches from the 1GNU structure do not exactly match those from the 1KOT structure; the patches are defined by the 1KOT structure. Nevertheless, Table 5 shows that the first-pass and second-pass sites around 1KOT and 1GNU are very similar. Moreover, all the possible locations for hydration are identified in both cases, though they appear at different passes.

| DISCUSSION
Cys-loop ligand-gated ion channels often interact with cytoplasmic proteins, and this interaction serves many purposes, amongst them the clustering of ion channels and the modulation of channel function.
One of the best studied examples is the interaction between the nAChR and the cytoplasmic protein rapsyn. Rapsyn has a molecular weight of about 43 000, 29 and it interacts with the intracellular domain of the nAChR. 30 Electron microscopy showed that the nAChR are interconnected by rapsyn dimers. Up to three rapsyn dimers can contact each nAChR in specific regions in the nAChR intracellular domain. This tight network probably underlies the low mobility of nAChR in the plane of the cell membrane, and also allows nAChR to be concentrated at the neuromuscular junction motor end-plate. 30 FIGURE 11 GABARAP with the hydration sites listed in Table 5. The CPK-colored atoms are from residues Lys 48, Val 51, Phe 60, and Ile 64. The angles of view of these three panels are approximately the same as those for the three panels in Figure 3 [Color figure can be viewed at wileyonlinelibrary.com] FIGURE 12 Diagram showing the GABA A receptor red and orange hydration sites on the surface of its γ2-subunit. In this diagram, the GABA A receptor "close," "regular" and "far" red hydration sites, as described in Table 3 are combined to give the red sites, and the "close," "regular," and "far" orange sites are combined to give the orange sites. The GABARAP residues are colored to correspond to their nearest sites according to the convention in Table 5: GABARAP  sites 11, 32 The interaction between the glycine receptor and gephyrin has been studied experimentally. Gephyrin was first identified as a protein which bridged the glycine receptor and tubulin. 31 Sola et al. 32 cocrystallized a segment of the glycine receptor β-subunit and a partial dimer of the cytoplasmic protein gephyrin (Protein Data Bank code: 1T3E).
They were able to resolve the structure of a pentapeptide portion of the glycine receptor β-subunit and the gephyrin domain E dimer. They proposed a network of gephyrin molecules linking the glycine receptors. Unfortunately, only the structure of five amino acids of the receptor was resolved, so it is difficult to draw any conclusion from this dataset.
Gephyrin also interacts with the GABA A receptor through its α2-subunit 33 and α3-subunit. 34 It is unclear if gephyrin binds the α1-subunit of the GABA A receptor; some experiments failed to show any interaction, 35 but others showed a weak interaction. 36 40 It thus appears that gephyrin has more general actions on both the GABA A receptor and the glycine receptor, and that the action of gephyrin and collybistin appear to be confined to receptor clustering. The action of GABARAP is more specific to the GABA A receptor, and, in addition to receptor positioning, it also modulates the electrophysiology of this ion channel.
In this work, we have used a flexible protein-protein docking programme to identify the interaction between the GABA A receptor and GABARAP. We have also used a novel method to predict hydration sites on the two proteins, and suggest docking poses. We have identified possible binding faces on the GABA A receptor and on GABARAP.
To confirm our theoretical predictions would require a high-resolution structure of the GABA A receptor with an intact intracellular domain. Some of the GABARAP binding faces we have identified are at the GABARAP/GABA A receptor interface, but others are involved in binding other proteins. In addition, we have also identified possible faces not known to bind any protein. It is interesting to note that, in the case of GABARAP, hydration patches appear on five out of six faces of this protein. As so many interfaces are involved in different types of interaction, it is possible that the last face is not active to remove the burden of constraints on protein architecture.
Currently, this method only examines the hydration details around proteins. We could envisage including details such as shape and electrostatic properties, and develop a molecular docking method based on this hydration site survey.
The GABA A receptor in neurons have different ion channel properties from recombinant receptors. 41 Luu et al. 40 and Everitt et al. 7 show that GABA A receptor conductances in neurons is similar to that obtained from recombinant receptors associated with GABARAP.
GABARAP is thus of importance in physiological functioning of the GABA A receptor in the central nervous system, and this underlies the importance of understanding the physiological role of the intracellular domain of this receptor. It would be interesting to investigate the interaction between GABARAP and the GABA A receptor further, to understand how GABARAP changes the ion channel functioning of the receptor. This would require a high-resolution structure of the GABA A receptor with an intact intracellular domain. FIGURE 13 GABARAP with the hydration sites listed in Table 5. The CPK-colored atoms are from residues Lys 48, Val 51, Phe 60, and Ile 64. The three panels on the left show the docking results with the intracellular MA helix of the γ2-subunit of the GABA A receptor present (transparent blue), and the three panels on the right show the results from hydration patch analysis. The angles of view on each row for the two structures are identical