Structural Basis for Modulation of Gating Property of G Protein-gated Inwardly Rectifying Potassium Ion Channel (GIRK) by i/o-family G Protein α Subunit (Gαi/o)*

Background: Although Gβγ is known to activate GIRK, Gαi/o also modulates GIRK gating. Results: The α2/α3 helices of Gαi3 in the GTP-bound state directly bind to the αA helix of GIRK. Conclusion: The complex model explains how Gαi/o sequesters Gβγ efficiently from GIRK upon GTP hydrolysis. Significance: The structural basis for the rapid closure of GIRK by Gαi/o is provided. G protein-gated inwardly rectifying potassium channel (GIRK) plays a crucial role in regulating heart rate and neuronal excitability. The gating of GIRK is regulated by the association and dissociation of G protein βγ subunits (Gβγ), which are released from pertussis toxin-sensitive G protein α subunit (Gαi/o) upon GPCR activation in vivo. Several lines of evidence indicate that Gαi/o also interacts directly with GIRK, playing functional roles in the signaling efficiency and the modulation of the channel activity. However, the underlying mechanism for GIRK regulation by Gαi/o remains to be elucidated. Here, we performed NMR analyses of the interaction between the cytoplasmic region of GIRK1 and Gαi3 in the GTP-bound state. The NMR spectral changes of Gα upon the addition of GIRK as well as the transferred cross-saturation (TCS) results indicated their direct binding mode, where the Kd value was estimated as ∼1 mm. The TCS experiments identified the direct binding sites on Gα and GIRK as the α2/α3 helices on the GTPase domain of Gα and the αA helix of GIRK. In addition, the TCS and paramagnetic relaxation enhancement results suggested that the helical domain of Gα transiently interacts with the αA helix of GIRK. Based on these results, we built a docking model of Gα and GIRK, suggesting the molecular basis for efficient GIRK deactivation by Gαi/o.

family (G i/o ) (13). The activation and the deactivation of GIRK are also accelerated when the channels are co-expressed with G␣ i/o (14 -16).
Efficient GIRK activation seems to be attributed to the preassociation of the heterotrimeric G protein in the i/ofamily (G␣ i/o (GDP)␤␥) with GIRK, in which G␣ i/o is directly associated with both GPCR and GIRK before GPCR activation (17,18). Furthermore, the direct association of G␣ i/o (GDP)␤␥ with GIRK in the absence of GPCR-stimulation reportedly suppresses the basal activity of GIRK while maintaining the maximal evoked current by G␤␥ in the presence of GPCR stimulation (14, 15, 19 -22). This regulation mechanism is called "priming," by which G␣ i/o reportedly plays an important role by facilitating a better response to GPCR activation.
On the other hand, the deactivation is accelerated by the direct interaction of G␣ i/o (GTP) with GIRK (18). The interaction enables G␣ i/o to rapidly bind to G␤␥ upon GTP hydrolysis, which is presumably assisted by the regulator of G protein signaling (RGS) (23), resulting in the closure of the GIRK gate. Because the re-association of G␣ i/o (GDP) and G␤␥ allows G␣ i/o (GDP)␤␥ to enter the next cycle of GIRK activation, the rapid reassociation would also accelerate the activation of GIRK (15). Thus, the direct interaction between GIRK and G␣ i/o (GTP) could contribute to the efficient activation and deactivation of the GIRK channel. In a FRET study investigating the conformational changes of the GIRK1/2 channel in Xenopus oocytes, the coexpression of G␣ i/o (GTP) with G␤␥ conferred different FRET patterns, and the maximal GIRK current, as compared with those observed when G␤␥ and G␣ i/o , were separately expressed (19).
Altogether, it is now evident that G␣ i/o not only functions as the donor and acceptor of G␤␥ but also modulates the gating property of GIRK. To reveal the underlying molecular mechanism by which the G␣ i/o modulates GIRK, several studies have proved the interactions between G␣ i/o and the cytoplasmic region of GIRK. G␣ i/o binding to the N-and C-terminal fragments of GIRK was examined by pulldown assays in vitro (14,15,19,20,22,24,25), indicating that the C-terminal region of GIRK is essential for the i/o-family-specific binding to GIRK. On the other hand, little is known about the GIRK binding site on G␣ i/o . G␣ consists of a GTPase domain and a helical domain (26,27) in which GDP or GTP is sandwiched by both domains and GPCRs bind to the GTPase domain (28). An electrophysiological study using a G␣ i1 /G␣ q chimera, in which the helical domain was replaced, revealed that the helical domain of G␣ is responsible for the specific activation of the GIRK channel (25). Thus, it remains unclear how G␣ i/o modulates the GIRK gating property, and therefore, the elucidation of the molecular recognition mode of G␣ i/o and GIRK is required.
In this study we performed the NMR analyses of the interaction between the cytoplasmic region of GIRK1 and G␣ i3 (GTP), which accelerates the deactivation of GIRK. The NMR spectral changes of G␣ upon the addition of GIRK as well as the transferred cross-saturation (TCS) results indicated that G␣ and the cytoplasmic region of GIRK1 directly bind to each other, with an estimated K d value of ϳ1 mM. We identified the binding sites on G␣ and GIRK, respectively, and examined the binding mode by paramagnetic relaxation enhancement (PRE) experiments. Based on these results, we built a docking model of G␣ and the cytoplasmic region of GIRK, suggesting the molecular basis for the efficient deactivation of the channel by G␣ i/o .

EXPERIMENTAL PROCEDURES
Expression and Purification of Cytoplasmic Regions of Mouse GIRK1-The N-and C-terminal cytoplasmic regions of mouse GIRK1 (residues 41-63 and 190 -386) were fused into a single polypeptide, which is hereafter referred to as GIRK CP-L . The C-terminal region of GIRK CP-L is 15 residues longer than that of GIRK CP (residues 41-63 and 190 -371). We previously analyzed the interaction of GIRK CP with G␤␥ and confirmed its validity by the crystal structure (29). Preliminary TCS experiments using GIRK CP and G␣ i3 indicated that the C-terminal region of GIRK CP was involved in the interaction with G␣ i3 , suggesting that some interacting residues are missing in GIRK CP . Therefore, we extended the GIRK CP construct for 15 amino acids to obtain GIRK CP-L , which was used for the further analyses.
GIRK CP-L was expressed in Escherichia coli cells. The uniformly 2 H, 15 N-labeled GIRK CP-L samples for the NMR analyses were prepared by growing E. coli in M9 minimal medium containing 15 (30). GIRK CP-L mainly eluted as a tetramer from the size exclusion chromatography column. The uniformly 2 H, 15 N-labeled and uniformly 2 H-labeled GIRK CP-L samples for the TCS experiments were prepared without the denaturing and refolding procedure to preserve the amide hydrogen atoms as 2 H in the core of the protein. The GIRK CP-L samples were incubated at 303 K for Ͼ48 h before the TCS experiments to prevent 2 H/ 1 H exchange during the TCS measurements.
Assignments of NMR Signals of GIRK CP-L -The 1 H, 15 N transverse relaxation-optimized spectroscopy (TROSY) signals of GIRK CP-L were well dispersed and mostly overlapped with those of GIRK CP . The backbone assignments of residues 41-63 and 190 -366 of GIRK CP-L were transferred from those of GIRK CP (BioMagResBank accession number 11067) (30). The assignments of residues 367-377 were established by triple resonance experiments (HNCACB and HNCA) (31, 32) using uniformly 2 H, 13 C, 15 N-labeled GIRK CP-L as shown in supplemental Table 1. The backbone amide resonances for Asn-378 -Val-386 could not be assigned because their signals were very weak or not observed, probably due to either the fast exchange of their amide hydrogen atoms with those of water molecules or the line-broadening due to conformational exchange.
Expression and Purification of G␣ i3 -The G␣ i3 protein, with an N-terminal decahistidine tag followed by an HRV3C protease cleavage site, was expressed in E. coli cells. The uniformly 2 H, 15  G␣ i3 was purified as described (12,33). Briefly, G␣ i3 was purified to homogeneity by chromatography on a HIS-Select Nickel Affinity Gel (Sigma) column, His-tag cleavage by Pre-Scission TM Protease (GE Healthcare), and removal of the cleaved His tags and the PreScission TM Protease (GE Healthcare) by HIS-Select Nickel Affinity Gel (Sigma).
Spin Labeling of G␣ i3 -For PRE experiments, we first prepared the G␣ i3 mutant referred to as Hexa III, in which all six exposed Cys residues were substituted (G␣ i3 (C3S-C66A-C214S-C305S-C325A-C351I)), according to the previous report (34). Using this construct as a template, cysteine substitutions were separately introduced to Ile-82 and Ser-153 by the QuikChange system (Stratagene). All mutations were confirmed by DNA sequencing.
Spin-labeling was performed in a buffer containing 10 mM Hepes-NaOH (pH 7.0), 50 mM KCl, 10 mM MgCl 2 , and 0.60 mM GTP␥S. The G␣ i3 mutants were incubated with S-(1-oxy-2,2,5,5-tetramethylpyrroline-3-methyl)-methanethiosulfonate (MTSL) at a molar ratio of 1:3 for each protein:MTSL at room temperature for 4 h. Under these conditions we confirmed that only the most reactive cysteine residue was modified, whereas the remaining buried native cysteine residues were not modified. The excess MTSL was removed by extensive washes with the buffer by ultrafiltration using an Amicon Ultra filter unit (Millipore). For the diamagnetic state experiments, ascorbate was added to the MTSL-labeled G␣ i3 mutants at a molar ratio of 1:3 protein:ascorbate, and the solution was incubated at 4°C for 12 h. The ascorbate was then removed by ultrafiltration. Thus, we prepared the MTSL-labeled Hexa III in the diamagnetic state.
NMR Analyses-All NMR experiments were performed on a Bruker Avance 600 spectrometer equipped with a cryogenic probe. The 1 H, 15 N TROSY spectra of G␣ i3 in the presence of various amounts of GIRK CP-L were observed at 308K. TCS and PRE experiments were performed at 303 K. All spectra were processed by the Bruker TopSpin 2.1 software, and the data were analyzed by Sparky (T. D. Goddard and D. G. Kneller, Sparky 3, University of California, San Francisco, CA). The error bars are based on the signal-to-noise ratio calculated by the Sparky software. The backbone NMR signal assignments of G␣ i3 were reported previously (35).
The TCS experiments were performed as described with minor modifications (12,36). In the TCS experiments observing G␣ i3 , a solution containing uniformly 2 H, 15 N-labeled G␣ i3 (0.3 mM) and unlabeled GIRK CP-L (0.25 mM as a tetramer) was prepared in buffer (10 mM HEPES-NaOH (pH 6.5), 50 mM KCl, 5 mM DTT, 1 mM DSS, 0.8 mM GTP␥S, 20% 1 H 2 O, 80% 2 H 2 O). The saturation frequency was set at 0.83 ppm, and the maximum radiofrequency amplitude was 0.17 kHz for WURST-2 (adiabatic factor Q 0 ϭ 1). The saturation duration and the relaxation delay were set at 1.5 and 2.5 s, respectively. To evaluate the effect of the residual aliphatic protons within G␣ i3 , TCS experiments were also performed under the same conditions as those mentioned above, with the sample containing uniformly 2 H, 15 N-labeled G␣ i3 (0.3 mM) and uniformly 2 H-labeled GIRK CP-L (0.25 mM as a tetramer). It should be noted that the residues with signal overlapping and/or signal-to-noise ratios less than 10 were excluded from the analyses.
In the TCS experiments observing GIRK CP-L, a solution containing uniformly 2 H, 15 N-labeled GIRK CP-L (0.25 mM as a tetramer) and unlabeled G␣ i3 (0.4 mM) was prepared in buffer (10 mM HEPES-NaOH (pH 6.5), 50 mM KCl, 5 mM DTT, 1 mM DSS, 0.8 mM GTP␥S, 20% 1 H 2 O, 80% 2 H 2 O). The saturation frequency was set at 0.83 ppm, and the maximum radiofrequency amplitude was 0.17 kHz for WURST-2 (adiabatic factor Q 0 ϭ 1). The saturation duration and the relaxation delay were set at 3.0 and 2.0 s, respectively. To evaluate the effect of the residual aliphatic protons within G␣ i3 , TCS experiments were also performed under the same conditions as those mentioned above, with the sample containing uniformly 2 H, 15 N-labeled GIRK CP-L (0.25 mM as a tetramer) and uniformly 2 H-labeled G␣ i3 (0.4 mM). It should be noted that the residues with signal overlapping and/or signal-to-noise ratios less than 10 were excluded from the analyses.
In the PRE experiments of the paramagnetic state, samples containing uniformly 2 H, 15 N-labeled GIRK CP-L (0.075 mM as a tetramer) mixed with 0.2 mM oxidized Hexa III-Cys-MTSL(ox) were prepared. In the experiments of diamagnetic state, samples containing uniformly 2 H, 15 N-labeled GIRK CP-L (0.075 mM as a tetramer) mixed with reduced Hexa III-Cys-MTSL(red) in buffer (10 mM HEPES-NaOH (pH 7.0), 50 mM KCl, 1 mM DSS, 10% 2 H 2 O, 90% 1 H 2 O) were prepared. The 1 H, 15 N TROSY spectrum of each sample was recorded. The residues with overlapping resonances were omitted from the analyses. PRE was calculated as paramagnetic to diamagnetic signal intensity ratios (I para /I dia ) (37).
Construction of Complex Models-The complex models of G␣-GIRK and G␣-GIRK-G␤␥ were obtained with the HADDOCK software (38).
First, we built a homology model of G␣ i3 (GTP) by the MOD-ELLER software (39) using the crystal structure of G␣ i1 (GTP␥S) (PDB code 1GIA) as a template, whose amino acid sequence is 94% identical to that of G␣ i3 . The crystal structure of GIRK CP (PDB code 1N9P) and the structure of G␤␥ in the crystal structure of G␣ i1 (GDP)␤ 1 ␥ 2 (PDB code 1GP2) were used, respectively, for dockings. The active residues used in the definition of the ambiguous interaction restraints for docking are listed in supplemental Table 2.
The G␣-GIRK-G␤␥ ternary complex model was built as follows. First, we built a docking model of the G␤␥-GIRK complex with parameters listed in supplemental Table 2. Then, the G␣-GIRK and G␤␥-GIRK complex models were superimposed by the GIRK structure.

RESULTS
NMR Spectral Changes of G␣ i3 upon Binding to GIRK CP-L -To investigate the direct binding between G␣ i3 and GIRK CP-L , we observed a series of 1 H, 15 N TROSY spectra of 0.28 mM uni-formly 2 H, 15 N-labeled G␣ i3 in the absence or presence of 0.25, 0.50, 0.75, and 1.0 mM GIRK CP-L as a tetramer. As the concentration of GIRK CP-L increased, most signals exhibited decreased intensity due to line-broadening without changing their chemical shifts, whereas a number of signals exhibited further intensity reductions and eventually disappeared in the presence of 0.75 mM GIRK CP-L . In addition, several signals exhibited small but significant chemical shift changes (Fig. 1, A  and B). Although the overall intensity reductions are caused by the slowing of the overall tumbling motion upon binding to GIRK CP-L , the further intensity reductions and apparent chemical shift changes reflect the direct binding of GIRK CP-L to G␣ i3 . The weighted averages of the chemical shift differences (⌬␦) were calculated using the equation (⌬␦ ϭ [(⌬␦ HN ) 2 ϩ (⌬␦ N / 6.5) 2 ] 1/2 ). The titrations curves of ⌬␦ were fit to the following theoretical formula to obtain the value of the dissociation con- where ⌬␦ sat is the ⌬␦ value when a saturating amount of GIRK CP-L is added (ppm), and [G␣ i3 ] tot ϭ 0.28 mM. The fitting of the titration curves of the chemical shift changes for the signals from Lys-209 and Trp-258 resulted in dissociation constant (K d ) values of 0.6 and 1.1 mM, respectively (Fig. 1C).
We evaluated the apparent chemical shift differences (⌬␦) of G␣ i3 in the absence of GIRK CP-L and the presence of 0.75 mM GIRK CP-L , in which 37-50% of G␣ i3 was in the GIRK CP-Lbound state, as estimated by the K d value of 0.6 -1.1 mM (Fig.  1D). The residues with significant chemical shift changes are Glu-207, Lys-209, Trp-211, His-213, Phe-215, Glu-216, Ser-246, Trp-258, Arg-312, and Thr-316, for which the minimum values of ⌬␦ within the error ranges (⌬␦ min ϭ ⌬␦ Ϫ error) are larger than 0.008 ppm. It should be noted that the threshold, 0.008 ppm, corresponds to 0.016 -0.022 ppm between the free and GIRK CP-L -bound states.
We also evaluated the signal intensity reductions, as the accelerated intensity reductions are caused by the differential line broadening, i.e. the chemical shift changes in the slow to intermediate exchange regime, upon binding to GIRK CP-L ( Fig.  1B) (40). Fig. 1E shows the intensity ratios (R) in the presence and absence of GIRK CP-L , which were corrected by multiplying by the scaling factor of 1.78, for the increase in the molecular weight upon binding (see supplemental Methods for details). The scaling factor of 1.78 is obviously too large for the N-and C-terminal residues (residues 1-31 and 349 -354), because these residues tumble faster than those in the other regions of G␣ i3 , as evidenced by the small line widths and the small values of the chemical shift indices (35,41), resulting in the R values larger than 1.0. Except for these signals, most signals showed the R values ranging from 0.40 to 0.80, suggesting that the tumbling motion is slowed in the presence of GIRK CP-L , presumably due to the increase in sample viscosity. The residues with significantly larger signal intensity reductions, which exhibited the maximum R value within the error range (R max ϭ R ϩ error) lower than 0.42, are Ala-41, Ser-44, Gly-45, Lys-46, Ser-47, Thr-177, Gly-203, and Ile-253. Fig. 1F shows the mapping of these affected residues on the crystal structure of G␣ (PDB code 1GIA), where the residues with significantly large ⌬␦ min and small R max are colored red and blue, respectively. Most of the residues with chemical shift changes are located in two regions; that is, the region from the ␣2 helix and the following loop (Glu-207, Lys-209, Trp-211, His-213, Phe-215, and Glu-216) and the region from the ␣3 helix and the following loop (Ser-246 and Trp-258). The other residues, Arg-312 and Thr-316, are located on the loop between the ␣4 helix and the ␤6 strand. On the other hand, the residues with significant intensity reductions are located around the GTP binding site, in which Ser-44, Gly-45, Lys-46, Ser-47, and Gly-203 directly interact with the phosphate group of GTP␥S.
As shown in Fig. 1F, the affected residues exhibited significant chemical shift changes, indicating that they are involved in the direct GIRK CP-L binding site and/or in the site(s) exhibiting a conformational change upon binding.
GIRK CP-L Binding Site on G␣ i3 Revealed by Transferred Crosssaturation Experiments-To identify the GIRK CP-L binding site on G␣ i3 , TCS experiments were performed. The saturation of the GIRK CP-L resonances by the irradiation with radio frequency pulses caused the signal intensity reductions of the 1 H, 15 N TROSY signals of the G␣ i3 residues, which should be located in the GIRK CP-L binding site ( Fig. 2A) (42,43).
In the case of a larger protein system, such as G␣ i3 and GIRK CP-L , with molecular masses of 41 and 103 kDa, respectively, the enhanced 1 H homonuclear dipolar-dipolar interactions might cause intramolecular saturation transfer from the residual protons in G␣ i3 (the exchangeable hydrogen atoms in the NH, OH, and SH groups and/or the hydrogen atoms, due to the incomplete 2 H-labeling of 2 H, 15 N-labeled G␣ i3 ). To exclude these effects, we also performed a control experiment by using 2 H-labeled GIRK CP-L instead of the unlabeled (i.e. 1 H-labeled) protein to reflect only the artificial effects described above (Fig. 2B, supplemental Fig. 1, gray) and subtracted the intensity reduction ratios of this control experiment from those obtained by using unlabeled GIRK CP-L (supplemental Fig. 1, orange). The differences in the reduction ratios, ⌬RR, are shown in Fig. 2C, with the error bars calculated based on the signal-to-noise ratios. The minimum values of ⌬RR within the error ranges (⌬RR min ϭ ⌬RR Ϫ error) were utilized for the evaluation.
The residues with large intensity reductions (⌬RR min Ͼ 0.08, Fig. 2C) are located on the helical domain (Gly-89 in the ␣A helix; Gly-112 and Ala-114 in the ␣B helix) and the GTPase domain (Arg-208, Lys-209, Trp-211, Ile-212, His-213, Glu-216, and Gly-217 in the ␣2 helix and the following loop; Met-240, Lys-248, Leu-249, Ile-253, Asn-256, and Trp-258 in the ␣3 helix and the flanking loops; Glu-186 in the ␤2 strand). The mapping of these residues on the structure of G␣ revealed that the residues identified on the GTPase domain of G␣ i3 are clustered, indicating that this site (hereafter, referred to as the "␣2/␣3 site") mainly contributes to the GIRK CP-L binding (Fig. 2D). This is also supported by the two alanine mutants of the identified residues (I212A and W258A) that exhibited impaired binding affinity for GIRK CP-L (supplemental Fig. 2). It should be noted that Glu-186 seems separated from the cluster by the intervening residue, Phe-199 (supplemental Fig. 3). The amide group of Phe-199 is buried in the protein and is more than 6 Å away from the contiguous surface for GIRK CP-L binding, resulting in the lack of cross-saturation for Phe-199. Therefore, we conclude that the ␣2/␣3 site is the major GIRK CP-L binding site.
The residues identified on the helical domain (Gly-89, Gly-112, and Ala-114), which are distant from the ␣2/␣3 site, do not form a contiguous surface but are located on the same side of the protein surface as the ␣2/␣3 site, suggesting that these residues transiently contact GIRK CP-L . We further investigated the interactions of the helical domain with GIRK CP-L (see below).
G␣ i3 Binding Site on GIRK CP-L Revealed by TCS Experiments-Conversely, to identify the G␣ i3 binding site on GIRK CP-L , TCS experiments observing GIRK CP-L were performed. In the same manner as the TCS experiments to identify the binding site on G␣ i3, we evaluated the cross-saturation from unlabeled G␣ i3 (supplemental Fig. 4, orange) by subtracting the intensity reduction ratios of the control experiment in the presence of 2 H-labeled G␣ i3 (supplemental Fig. 4, gray). The differences in the reduction ratios, ⌬RR, are shown in Fig. 3A, with the error bars calculated based on the signal-to-noise ratios. The minimum values of ⌬RR within the error ranges (⌬RR min ) were utilized for the evaluation.
The residues with large intensity reductions (⌬RR min Ͼ 0.02) are Glu-242, Val-358, Leu-365, Leu-366, Met-367, Ser-368, Ser-369, Leu-371, Ile-372, and Ala-373 as well as one of the C-terminal unassigned residues (hereafter, referred to as CT1). These residues were mapped on a single subunit (Fig. 3B) and two adjacent subunits (Fig. 3C) of GIRK CP (PDB code 1N9P). All of these residues, except for Glu-242 are located on the eq of GIRK CP-L (0.75 mM as a tetramer). The signals with chemical shift differences larger than 0.008 ppm are labeled. The signal from Gly-45, which is one of the residues showing significant intensity reduction without a significant chemical shift change, is also labeled in parentheses. B, overlays of the spectra in the presence of 0 -2.7 eq of GIRK CP-L (as a tetramer) are displayed for Gly-45 (left) and Trp-258 (right) as typical examples of the signals with significant intensity reductions and chemical shift changes, respectively. C, titration curves of the chemical shift differences for Lys-209 (left) and Trp-258 (right) are shown. D, shown are plots of the chemical shift differences of G␣ i3 in the absence and presence of 2.7 eq of GIRK CP-L (as a tetramer). The error bars were calculated based on the digital resolutions of the spectra. The minimum values of ⌬␦ within the error ranges (⌬␦ min ) were utilized for the evaluation. Bars corresponding to the residues with significant chemical shift differences, which have the minimum values of ⌬␦ within the error ranges (⌬␦ min ) larger than 0.008 ppm, are colored red. Asterisks in D and E indicate the residues with no data due to signal overlapping, lack of assignments, or insufficient signalto-noise ratio. The secondary structure elements of G␣ i3 are depicted in gray below the sequence, and ␣2, ␣3, ␣4, and ␤6 are colored green. E, shown are plots of the normalized intensity ratios (R) of uniformly 2 H, 15 N-labeled G␣ i3 upon the addition of 1.8 eq of the GIRK CP-L tetramer to G␣ i3 . The intensities of the free G␣ i3 were divided by a scaling factor of 1.78 (see supplemental Methods) and then were used for the calculation of the intensity ratios. The error bars were calculated based on the signal-to-noise ratios. Bars corresponding to the signals with the maximum value of R, within the error ranges (R max ) lower than 0.42 are colored cyan. F, mapping of the affected residues on the G␣ i1 structure in which 333 of 354 residues (94%) are identical to those of G␣ i3 (PDB code 1GIA) is shown. The amide nitrogen atoms of the residues with apparent chemical shift differences and intensity reductions are shown as balls colored red and cyan, respectively. Proline residues and residues with no data are colored black. The ␣2, ␣3, ␣4, and ␤6 are colored green, and GTP␥S is depicted by teal sticks.
C-terminal region around the ␣A helix of GIRK CP-L (Fig. 3). Therefore, we concluded that these residues are the G␣ i3 binding residues. Glu-242 would be involved in the direct binding to the G␣ i3 , as explained under "Discussion." It should be noted that the first two residues, Ile-372 and Ala-373, and CT1 were detected among the extended C-terminal 15 residues from GIRK CP. The other C-terminal residues showed no significant intensity reductions. Thus, we concluded that the extension for 15 residues is sufficient for the interaction with G␣ i3 .

Contribution of Helical Domain of G␣ i3 to Interaction with GIRK CP-L , as Investigated by PRE Experiments-
The interaction of the helical domain of G␣ i3 with GIRK CP-L was investigated by PRE experiments. PRE arises from magnetic dipolar interactions between the nuclear spin and the unpaired electron spin of the paramagnetic center, which enhances the relaxation of the nuclear spins, leading to the line-broadening and thus the intensity reduction of the NMR signals of the residues within about 20 Å of the spin label (37,44). In addition, PRE can detect a transient interaction, because the PRE effect occurs within 250 -500 s (45), which is a much shorter duration than that of the cross-saturation (the effective saturation time of 200 -300 ms for the TCS experiments in this study) (43).
The spin-labeling reagent, MTSL, which can be chemically introduced to cysteine side chains, was used to label G␣ i3 . First, we prepared a mutant of G␣ i3 (C3S-C66A-C214S-C305S-C325A-C351I) in which the six natively existing cysteine residues that are exposed and reactive are substituted with other residues (hereafter, referred to as Hexa III). It should be noted that the corresponding mutant of G␣ i1 (C3S-C66A-C214S-C305S-C325A-C351I), named Hexa I, was folded properly and functional (34). We also confirmed the proper folding of Hexa III by the 1 H, 15 N TROSY spectrum of 2 H, 15 N-labeled Hexa III (data not shown). Then, either Ile-82 or Ser-153 was substituted by Cys and chemically modified by MTSL (Hexa III-I82C-MTSL and Hexa III-S153C-MTSL). The Ile-82 residue is located at the center of the three residues identified by TCS (Gly-89, Gly-112, and Ala-114), and Ser-153 exists on the opposite side from I82. Because the paramagnetism of MTSL disappears upon reduction by the addition of ascorbate, the PRE effects were evaluated by the signal intensity ratios of the 1 H, 15 N TROSY spectra of 2 H, 15 N-labeled GIRK CP-L between the oxidized and reduced states in the presence of Hexa III-I82C-MTSL or Hexa III-S153C-MTSL. Fig. 4A shows that the PRE effects from Hexa III-I82C-MTSL were observed for 13 signals of GIRK CP-L : Thr-353, Tyr-356, Ser-357, Leu-365, Leu-366, Met-367, Ser-368, Ser-369, Leu-371, Ile-372, Ala-373, A375, and CT1, which reside in the C-terminal region of GIRK CP-L . It should be noted that most of the other residues of the C-terminal region could not be analyzed due to signal overlapping, and thus these residues might also be affected by MTSL. On the other hand, only Cys-53 was detected for Hexa III-S153C-MTSL (Fig. 4B), which might be caused by the partial MTSL modification of Cys-53. Because the spin label was introduced at I82C, which is at the center of the three residues identified by TCS, we conclude that the Ile-82 side of the helical domain can approach within 20 Å of the C-terminal region of GIRK CP-L .

DISCUSSION
Direct Interaction between G␣ i3 and Cytoplasmic Region of GIRK-NMR analyses were performed to probe the interaction between GIRK CP-L and G␣ i3 . Because cross-saturation is a phenomenon depending on the intermolecular 1 H-1 H distances, the TCS results from GIRK CP-L to G␣ i3 (Fig. 2) and vice versa (Fig. 3) indicated that the ␣2/␣3 site of G␣ i3 and the ␣A helix of GIRK CP-L directly interact with each other. Our relaxation matrix calculations (43,46) in which the on and off rates of the GIRK CP-L -G␣ i3 interaction are also considered, suggested that under the current experimental conditions the cross-saturation effect should be observed for the amide hydrogen atoms of GIRK CP-L within 6 Å from G␣ i3 and for those of G␣ i3 within 5 Å from GIRK CP-L . Furthermore, the direct interaction was verified by the mutations of the interacting residues on G␣ i3 , which impaired the affinity for GIRK CP-L (I212A and W258A, supplemental Fig. 2).
Most of the residues with apparent chemical shift changes are located in the GIRK CP-L binding site consisting of the ␣2 and ␣3 helices of G␣ i3 , whereas the other residues, Arg-312 and Thr-316, exist on the ␣4/␤6-loop that is adjacent to the ␣3 helix, reflecting the conformational changes of these residues upon GIRK CP-L binding (Fig. 1). The fitting of the titration curves of the chemical shift changes of Lys-209 and Trp-258, the GIRK CP-L binding residues, resulted in the K d values of 0.6 and 1.1 mM, respectively, which are within the range of the fitting error (Fig. 1C). Thus, the K d value for the binding of G␣ i3 and GIRK CP-L was estimated as 1 mM. Although this K d value is quite large as a value for a proteinprotein interaction, it would fall in the nanomolar to micromolar range on the cell membrane, considering the reduced dimensionality effects (12,47,48). The K d value on the order of 1 mM is 4 times of the K d value of 0.25 mM that we previously reported for the GIRK CP -G␤␥ interaction (12), which is consistent with the report that the affinity of the cytoplasmic region of GIRK1 for G␤␥ was 4 -5-fold stronger than that for G␣ i3 (15).
As shown in Fig. 1, E and F, accelerated signal intensity reductions were observed for the G␣ i3 residues at the GTP binding region, which reflect the larger chemical shift changes for the residues in the intermediate exchange regime. These residues are distant from the ␣2/␣3 site, which is the direct GIRK CP-L binding interface, suggesting that the conformational change around the GTP binding site is induced by GIRK CP-L binding to the ␣2/␣3 site. In particular, Ser-44, Gly-  25 mM as a tetramer), which was recorded without (left) and with (right) radio frequency irradiation. Cross-sections are also shown for the signals from Ala-99 and Trp-258. B, procedures were the same as A, except that 2 H-labeled GIRK CP-L was used instead of unlabeled ( 1 H-labeled) GIRK CP-L , as a negative control. C, shown is a plot of the difference in the reduction ratios (⌬RR) originating from the backbone amide groups with and without irradiation in the presence of unlabeled GIRK CP-L and 2 H-labeled GIRK CP-L (see also supplemental Fig. 1). The residues indicated by asterisks are those with no data mostly due to overlapping of the resonances or insufficient signal-to-noise ratio. The error bars were calculated based on the signal-to-noise ratios. Bars corresponding to the residues with significant intensity reductions (minimum values of ⌬RR (⌬RR min ) Ͼ 0.08) are colored red. The secondary structure elements of G␣ i3 are depicted in gray below the sequence, and ␣A, ␣B, ␣2, and ␣3 are colored green. D, mapping of the affected residues in the TCS experiment on the G␣ structure (PDB code 1GIA) is shown. The backbone nitrogen atoms of the affected residues are shown as red balls in the ribbon diagram of the G␣ structure (left). The affected residues are colored red on the surface representations of the G␣ structure, whereas the residues with no data, including proline residues, are colored black (center and right). The ␣A, ␣B, ␣2, and ␣3 are colored green. 45, Lys-46, Ser-47, and Gly-203 are the residues directly interacting with the ␤or ␥-phosphate groups of GTP. This might be related to the report that the GTPase activity of G␣ s is enhanced upon binding to its effector, adenylate cyclase (49), which also binds to the ␣2/␣3 site of G␣ s (50). However, the significance of the conformational change at the GTP binding FIGURE 3. TCS from G␣ i3 to GIRK CP-L . A, shown is a plot of the difference in the reduction ratios (⌬RR) originating from the backbone amide groups with and without radio frequency irradiation in the presence of unlabeled G␣ i3 and 2 H-labeled G␣ i3 (see also supplemental Fig. 3). The residues indicated by asterisks are those with no data, mostly due to overlapping resonances or insufficient signal-to-noise ratio. The error bars were calculated based on the signal-to-noise ratios. Bars corresponding to the residues with significant intensity reductions, which have the minimum values of ⌬RR within the error ranges (⌬RR min ) larger than 0.02, are colored red. The primary sequence of GIRK CP-L (residues 41-63 and 190 -386) is displayed in the single-letter amino acid code followed by the residue number. The secondary structure elements of GIRK CP-L are depicted in gray below the sequence, based on the crystal structure (PDB code 1N9P), and the ␣A helix is colored green. B, mapping of the affected residues in the TCS experiment on a single subunit of the GIRK CP tetramer (PDB code 1N9P) is shown. Residues 371-386 are depicted by gray tubes. Side views of the GIRK CP-L structures parallel to the membrane plane, where the membrane side is above the molecules, are depicted by ribbon diagrams with balls for the amide nitrogen atoms of the affected residues (left). The affected residues are colored red on the surface representations of the GIRK CP-L structure, whereas the residues with no data, including proline residues, are colored black (center and right). The ␣A helix is colored green in the ribbon diagrams. Views from the outside (left and center) and the inside of the tetramer (right) are shown. The four subunits of the GIRK CP tetramer are shown in a surface representation for reference. C, mapping of the affected residues (colored red) in the TCS experiment on both of the two adjacent subunits of a GIRK CP tetramer (white and gray), viewed from the outside (left) and the inside (right) of the tetramer. Schematic drawings of the GIRK CP tetramer (subunits A, B, C, and D), viewed from the membrane side, are included to indicate the view of the surface models. The dashed blue line indicates the boundary between the two neighboring subunits. site in terms of GIRK regulation remains unclear and is beyond the scope of this paper.
Binding Mode of G␣ and Cytoplasmic Region of GIRK-Although the TCS results revealed that the ␣A helix of GIRK CP-L directly binds to the ␣2/␣3 site of G␣ i3 , the PRE results indicated the ␣A helix of GIRK CP-L is within 20 Å of Ile-82 of G␣ i3 . However, the ␣A helix of GIRK CP-L , which binds to the ␣2/␣3 site, should be more than 20 Å away from the spin-labeled site, suggesting that another ␣A helix in a neighboring subunit of the GIRK CP-L tetramer must come close to the spin-labeled site (Fig. 5).
To build a model of the G␣ i3 -GIRK CP-L complex satisfying the NMR-derived structural information, rigid body docking was performed by using the HADDOCK program (38). Because the structure of G␣ i3 (GTP) is not available, we built a homology model by the MODELLER software of G␣ i3 (GTP) from the crystal structure of G␣ i1 (GTP␥S) (PDB code 1GIA) in which 94% of the 354 residues are identical to G␣ i3 (39) and used it for the construction of the complex model. The residues in the ␣2/␣3 site of G␣ and in the ␣A helix of GIRK CP-L , identified by the TCS experiments, were specified as the "active residues" in the program so that they formed the interface of the complex. Although the information about the residues on the helical domain derived from the TCS and PRE experiments was not used for the calculation, the helical domain of G␣ was calculated to be proximal to the ␣A helix in a neighboring subunit of GIRK CP-L , probably due to the restraint of Glu-242 on the subunit interface of GIRK CP-L , as determined from the TCS experiments (Fig. 5). In this binding mode, Glu-242 of GIRK CP-L at a distance from the ␣A helix, approaches the side chain of Glu-186 of G␣ i3 , which accounts for its intensity reduction in the TCS experiment.
The three residues that TCS identified on the helical domain of G␣ i3 (Gly-89, Gly-112, and Ala-114) showed relatively weak cross-saturation, and they do not form a continuous binding surface, presumably because the ␣A helix of GIRK CP-L transiently accesses the G␣ i3 helical domain. This can be accounted for by the conformational flexibility of the ␣A helix. In the crystal structure, the ␣A helix of GIRK CP is stabilized by the crystal contacts with the ␣A helix of another tetramer. In addition, the C-terminal part of the helix (residues 367-368) exhibited very weak or no NOEs that are typical for an ␣ helix (data not shown), and the chemical shift index calculated by the 13 C chemical shifts indicated that the C-terminal residues after Met-367 are unstructured. Altogether, we concluded that the ␣2/␣3 site of G␣ i3 mainly recognizes the ␣A helix of GIRK CP-L in one subunit of the tetramer, whereas the helical domain of G␣ i3 transiently approaches the ␣A helix in another neighboring subunit.
The N terminus of G␣ is modified by a lipid moiety in vivo and anchored to the cell membrane (51), which can be accounted for by this binding mode. Although the most N-ter-  A and B, left, shown is a plot of the signal intensity RRs of paramagnetic to diamagnetic for Hexa III-I82C-MTSL (A) and Hexa III-S153C-MTSL (B). The residues indicated by asterisks are those with no data due to overlapping resonances or insufficient signal-to-noise ratio. The error bars were calculated based on the signal-to-noise ratios. Bars corresponding to the residues with significant intensity reductions, which have the minimum values of RR within the error ranges (RR min ) larger than 0.20, are colored red. The primary sequence of GIRK CP-L (residues 41-63 and 190 -386) is displayed in the single-letter amino acid code followed by the residue number. Center, mapping of the affected residues in the PRE experiment on a single subunit of the GIRK CP structure (PDB code 1N9P) is shown. Residues 371-386 are depicted by gray tubes. Side views of the GIRK CP structures parallel to the membrane plane are depicted by surface representations viewed from the outside of the tetramer, where the membrane side is above the molecules. The affected residues are colored red. The proline residues and the residues with no data are colored black. Right, shown is mapping of the spin-labeled site of G␣ i3 on the structure of G␣ (PDB code 1GIA).
minal G␣ residue in the crystal structure, Val-34, resides at about 35 Å away from the possible intracellular cell membrane in the current binding mode (supplemental Fig. 5), the N-terminal residues 1-33 of G␣, which would form a 40 Å-long ␣ helix in the complex with G␤␥, are presumably unstructured and adopt a longer length than 40 Å in the GTP-bound state.
Molecular Recognition Mode between G␣ i3 and GIRK CP-L - Fig. 6 shows the surface properties of the ␣2/␣3 site of G␣ i3 and the ␣A helix of GIRK CP-L . The ␣2/␣3 site possesses a hydrophobic cleft at its center, which is surrounded by polar residues, whereas the ␣A helix also contains a cluster of hydrophobic residues that are exposed to the outside. This suggests that the complementarity of the shape at the direct binding sites, formed by these hydrophobic residues, is important for the binding. Indeed, the mutations of the hydrophobic residues of G␣ i3 , I212A and W258A, resulted in the impaired affinity for GIRK CP-L (supplemental Fig. 2).
The ␣2/␣3 site on G␣, which is highly conserved among the G␣ family members, is known to be the effector binding surface of G␣ in the GTP-bound form, as revealed by the crystal structures of G␣ in complex with its effectors such as G␣ s -adenylate cyclase (50) and G␣ q -G protein receptor kinase (52), and with peptides (53). In these complexes, the hydrophobic and acidic side chains from the effectors and peptides are inserted into the hydrophobic cleft of the ␣2/␣3 site on G␣. GIRK is also recognized in a similar manner to those of the effectors and the peptides. It should be noted that the conformational flexibility of the ␣A helix might play a role in its interaction with the hydrophobic cleft of ␣2/␣3 on G␣.
The helical domain of G␣ is reportedly important for the i/o-family-specific activation of the GIRK channel, as determined with a G␣ i1 /G␣ q chimera (25). However, the main GIRK binding site for the G␣ i3 (GTP) revealed in this study was the ␣2/␣3 site on the GTPase domain, not on the helical domain. The activation specificity should be accomplished by the bind-ing of G␣ i/o (GDP)␤␥ to GIRK, which seems to occur at a different site from that for G␣ i/o (GTP).
Previously, the G␣ i/o binding regions in GIRK1 were investigated by pulldown assays using the cytoplasmic fragments of GIRK (14,15,20,22,24). The cytoplasmic C-terminal residues 320 -369 of GIRK1 were indicated to be important for strong binding to G␣ i3 (14), whereas the N-terminal fragment of GIRK1 (residues 1-84) also bound to G␣ i/o (GTP) (15,20). The G␣ i3 binding residues of GIRK1 revealed in this study are involved in the previously defined C-terminal region, and we identified three additional residues, Leu-371, Ile-372, and Ala-373. These residues vary among the GIRK subtypes, and GIRK1 possesses more hydrophobic and fewer charged residues than the other subtypes (supplemental Fig. 6), suggesting that the G␣ i/o binding affinity might differ among the subtypes. Unfortunately, the N-terminal binding region was not identified here, as the construct used in our study lacks residues 1-40 and 64 -84.
Modulation of GIRK by G i/o -GIRK is activated by the G␤␥ binding, and G␣ i/o modulates the gating property of GIRK. G␣ i/o (GDP)␤␥ is assumed to be precoupled with GPCRs and GIRK, which facilitates the efficient gate opening of GIRK, with the high specificity of GPCR signaling to GIRK (17,18). G␣ i/o (GDP)␤␥ binding to GIRK suppresses the basal K ϩ current of GIRK while maintaining the maximal current evoked by the GPCR stimulation, which is referred to as priming (14, 15, 19 -22). On the other hand, G␣ i/o (GTP) binding to GIRK accelerates its deactivation (15).
Recently, a two-site model was proposed for the G protein binding sites on GIRK ("anchoring site" and "activation site") (19,22). The anchoring site is relevant to precoupling and priming for GIRK when the site accommodates G␣ i/o (GDP)␤␥ and to the binding of G␣ i/o (GTP) when G␤␥ shifts to the activation site, which was revealed by our NMR analyses as the border of the two neighboring subunits of the GIRK tetramer (12). In this  (39) is shown in a surface representation. Residues 371-386 of GIRK1 are depicted schematically. The nitrogen atoms of the GIRK CP-L residues identified in the TCS experiments and the residues with PRE effects from MTSL, modified to I82C on the helical domain of G␣ i3 (green), are shown as red and magenta balls, respectively. The ␣2/␣3 residues of G␣ i3 identified in the TCS experiments are colored blue. Center, shown is the binding model of G␣ i3 in the GTP-bound state and the cytoplasmic region of GIRK1 (PDB code 1N9P), obtained with HADDOCK software (38). The residues on GIRK CP-L and the ␣2/␣3 residues on G␣ i3 identified by TCS were defined as the active residues in the program so as to form the interface of the complex. The residues 358 -370 on GIRK were defined as "semi-flexible segments" to allow them to move during the simulated annealing. Only the two adjacent subunits of the GIRK tetramer are shown from the inside of the tetramer. Right, the model displayed at the center is rotated by 180 degrees, and all of the subunits of the GIRK tetramer are shown. study we have identified the anchoring site for G␣ i/o (GTP) on GIRK (Figs. 5 and 7). Notably, it is unknown whether the site is identical to the one for G␣ i/o (GDP)␤␥. Upon ligand stimulation of the precoupled GPCR, the GDP-GTP exchange on G␣ i/o in the complex with G␤␥ at the anchoring site for G␣ i/o (GDP)␤␥ decreases the affinity of G␣ i/o for G␤␥, allowing G␤␥ to activate GIRK by binding to the activation site. When the GPCR stimulation ends, GTP on G␣ i/o at the anchoring site for G␣ i/o (GTP) is rapidly hydrolyzed to GDP with assistance from RGS (23). Thus, G␣ i/o , at the anchoring site for G␣ i/o (GTP) immediately re-associates with G␤␥, leading to the closure of the GIRK gate.
In this study we revealed that the cytoplasmic region of GIRK1 and the GTP-bound G␣ i3 directly bind to each other, and the ␣A helix of GIRK1 corresponds to the anchoring site for G␣ i/o (GTP). The mapping of the G␤␥ binding residues (12) on the structure of the G␣-GIRK complex indicated that the binding sites of G␣ i3 (GTP) and G␤␥ on GIRK do not overlap with each other, and therefore, G␣ i3 (GTP) and G␤␥ can simultaneously bind to GIRK (Fig. 7). Furthermore, the mapping of the RGS binding site on the structure of G␣ suggested that RGS can bind to G␣ i3 in the complex with GIRK (supplemental Fig. 7). Therefore, the G␣ i/o -GIRK-G␤␥ ternary complex model obtained here provides the structural basis for the rapid closure of the GIRK channel upon signal termination through the efficient removal of the proximal G␤␥ from GIRK. The residues on GIRK CP-L and the ␣2/␣3 residues on G␣ i3 identified by TCS (labeled by bold letters) and their surrounding residues with C␣ atoms within 6.0 Å, are colored according to their side-chain properties: acidic, basic, hydrophobic, and hydrophilic residues are colored red, blue, green, and yellow, respectively. FIGURE 7. A model of the G␣ i/o -GIRK-G␤␥ ternary complex. Left, the nitrogen atoms of the G␤␥ binding residues on GIRK1 (12) are mapped as magenta balls on the complex model of G␣ and the GIRK tetramer (Fig. 5). Right, shown is a ternary complex model of G␣-GIRK-G␤␥. First, the binding model of G␤␥-GIRK was built by using HADDOCK with parameters listed in supplemental Table 2. The G␤␥ binding residues on GIRK CP-L identified in our previous study (12) and the residues on G␤␥ reported to be important for GIRK binding (54,55) were defined as the active residues in the software. Then the G␣-GIRK complex model shown in Fig. 5 and the G␤␥-GIRK complex model were superimposed by the GIRK structure, rendering the G␣-GIRK-G␤␥ ternary complex model. G␤ and G␥ are colored magenta and violet, respectively. The residues involved in the intermolecular interactions between G␣ and G␤␥, within a distance of 4 Å in the G␣ i1 (GDP)␤ 1 ␥ 2 (PDB code 1GP2) structure, are colored orange.
It should be noted that the ␣2 helix of G␣ i3 in the GIRK binding site (the ␣2/␣3 site) is located in the switch II region, which is known to alter its conformation upon GDP-GTP exchange and is included in the G␤␥ binding site. Thus, the anchoring site of G␣ i/o (GDP)␤␥ is different from that of G␣ i/o (GTP) revealed here. Structural analyses of the interaction between G␣ i/o (GDP)␤␥ and GIRK will provide a complete understanding of the modulation mechanism of the GIRK-gating by G i/o proteins.