Interaction sites of the G protein beta subunit with brain G protein-coupled inward rectifier K+ channel.

G protein-coupled inward rectifier K(+) channels (GIRK channels) are activated directly by the G protein betagamma subunit. The crystal structure of the G protein betagamma subunits reveals that the beta subunit consists of an N-terminal alpha helix followed by a symmetrical seven-bladed propeller structure. Each blade is made up of four antiparallel beta strands. The top surface of the propeller structure interacts with the Galpha subunit. The outer surface of the betagamma torus is largely made from outer beta strands of the propeller. We analyzed the interaction between the beta subunit and brain GIRK channels by mutating the outer surface of the betagamma torus. Mutants of the outer surface of the beta(1) subunit were generated by replacing the sequences at the outer beta strands of each blade with corresponding sequences of the yeast beta subunit, STE4. The mutant beta(1)gamma(2) subunits were expressed in and purified from Sf9 cells. They were applied to inside-out patches of cultured locus coeruleus neurons. The wild type beta(1)gamma(2) induced robust GIRK channel activity with an EC(50) of about 4 nm. Among the eight outer surface mutants tested, blade 1 and blade 2 mutants (D1 and CD2) were far less active than the wild type in stimulating GIRK channels. However, the ability of D1 and CD2 to regulate type I and type II adenylyl cyclases was not very different from that of the wild type beta(1)gamma(2). As to the activities to stimulate phospholipase Cbeta(2), D1 was more potent and CD2 was less potent than the wild type beta(1)gamma(2). Additionally we tested four beta(1) mutants in which mutated residues are located in the top Galpha/beta interacting surface. Among them, mutant W332A showed far less ability than the wild type to activate GIRK channels. These results suggest that the outer surface of blade 1 and blade 2 of the beta subunit might specifically interact with GIRK and that the beta subunit interacts with GIRK both over the outer surface and over the top Galpha interacting surface.

Heterotrimeric G proteins transduce a variety of regulatory signals from a large number of heptahelical receptors to effec-tors such as adenylyl cyclases, phosphodiesterases, phospholipases, and ion channels (1). Each G protein oligomer contains a guanine nucleotide binding ␣ subunit and a high affinity dimer of ␤ and ␥ subunits. The agonist-bound receptor activates the G proteins and generates GTP-bound ␣ subunits and free ␤␥ subunits. Both GTP ␣ and ␤␥ can regulate downstream effectors. The hydrolysis of GTP to GDP on ␣ subunits leads to the reassociation of ␣ and ␤␥ subunits to form inactive heterotrimers.
The crystal structure of the ␤␥ subunit reveals that the ␤ subunit consists of an N-terminal ␣ helix followed by a symmetrical seven-bladed propeller structure based on WD repeat sequences, repeating motif of about 40 amino acids (2,3). Each blade consists of four antiparallel ␤ sheets. The top surface of the propeller structure interacts with the ␣ subunit. The bottom surface of the torus is the major site for interaction with the ␥ subunit. The outer surface of the torus is largely made up of the outer ␤ strands of seven blades. ␤␥ subunits directly regulate various effectors, including phospholipase C␤s (PLC␤s), 1 adenylyl cyclases, and ion channels. Experiments using mutated ␤ subunits have demonstrated that some of the amino acid residues, which are located at the interaction sites between G␣ and -␤ subunits, are crucial for the interaction between ␤␥ and the effectors, and each effector demonstrates its specific domain of interaction on the ␤ subunit (4,5). Further mutational analysis showed that the activation of PLC␤ 2 also involves residues in the outer strands of blades 2, 6, and 7. However, mutations of ␤ subunits that affected PLC␤ 2 activity did not influence the interaction between ␤␥ and type I or type II adenylyl cyclase (6). These results suggest that ␤␥ interacts with effectors through both the G␣ binding surface and the outer surface of the propeller structure and that each effector is interacting with ␤␥ using different regions of its outer surface.
The present study was undertaken to investigate the interaction sites of ␤ 1 with brain GIRK channels. We focused our investigation on the outer surface of the ␤ 1 subunit. We made sets of ␤ 1 mutants in which the amino acid residues located on or near the outermost ␤ strand of each blade were replaced with the corresponding residues of the yeast ␤ subunit (STE4), a most distantly related member of ␤ from mammalian ␤ subunits. In this study we show that mutating residues located on or near the outer strands of blade 1 and blade 2 disrupt GIRK activation by ␤ 1 ␥ 2 . In contrast, the same mutations did not substantially affect the ability of ␤ 1 ␥ 2 to regulate adenylyl cyclases, indicating that no global disruption of ␤ 1 occurred because of the mutations. It is possible that the sites of the ␤ 1 subunit in blade 1 and blade 2 could specifically be involved in brain GIRK activation. Sf9 cells were cultured in suspension in IPL-41 medium containing 1%  Pluronic F68 and 10% heat-inactivated fetal bovine serum at 27°C with  constant shaking (150 rpm). The site-directed mutagenesis of ␤ 1 cDNA was performed using the Muta-Gene in vitro mutagenesis kit (Bio-Rad). The amino acid sequence of each mutation is shown in Fig. 3. The mutations were confirmed by sequencing the mutating region. Mutated ␤ 1 cDNAs were subcloned into the pVL1392 transfer vector, and the resulting plasmids were cotransfected into Sf9 cells with BacPac6 viral DNA linearized with Bsu36I (CLONTECH) by LipofectAMINE (Life Technologies, Inc.). Recombinant baculoviruses were plaque-purified and amplified as described (7). Baculoviruses encoding L117A, D224S, D228R, and W332A were generated and generously provided by the late Dr. Eva J. Neer (Harvard University) (4). Recombinant baculoviruses encoding wild type ␤ 1 , ␥ 2 , and His 6 -G␣ i1 have been described previously (8).
Cultures of Locus Coeruleus Neurons-Cultured neurons from the locus coeruleus were made from 2-4-day-old postnatal Long-Evans rats (Charles River Breeding Laboratories). The culture methods were described previously (9,10). Rats were anesthetized with ether, their brainstems were removed, and the rats were killed by decapitation. Brain slices were made from isolated brainstems using a Vibratome (Lancer 1000). The locus coeruleus was visually identified under a dissecting microscope and excised out. The excised pieces were incubated in a papain solution, dissociated by trituration, and cultured. The culture medium contained a minimum essential medium with Earle's salts (Life Technologies, Inc., catalog no. 11430 -030), modified by adding D-glucose (5 mg/ml), NaHCO 3 (3.7 mg/ml), and L-glutamine (0.292 mg/ml). The medium was supplemented by heat-inactivated rat serum (2 or 5%, prepared in our laboratory), L-ascorbic acid (10 g/ml), penicillin (50 units/ml), and streptomycin (50 g/ml). Dissociated neurons were cultured on a squared plastic piece placed in a small well made at the center of a 35-mm culture dish. Before plating the neurons, the well, containing the plastic piece, was coated with rat collagen and a feeder layer of glia cells obtained from rat brains. The cultures were incubated at 37°C in 10% CO 2 and 90% air with saturated humidity for 13.4 Ϯ 2.5 days (mean Ϯ S. D.). Experiments were performed on large neurons (soma diameter, 24.8 Ϯ 2.9 m; mean Ϯ S.D.).
The procedures of data analysis were similar to those previously described (12). The data were analyzed with pCLAMP programs (version 6) (Axon Instruments). The overall frequency response was set at 1 kHz (Ϫ3 db by an 8-pole Bessel filter) and digitized at 10 kHz. Transitions between the closed and the open states were registered if a level crossed a threshold and lasted for more than 100 s. The threshold was set at 2.0 pA from the baseline when the membrane was held at Ϫ101 mV (1.6 pA when the membrane was held at Ϫ80.4 mV); this procedure excluded most of the small background channels (12). We expressed the channel activity by a variable NP o (i.e. the open probability of an elementary channel multiplied by the number of channels in the patch).
The experimental protocol was as follows. After an inside-out patch was formed, the basal channel activity (in the GDP-containing bathing solution) was recorded for a few minutes. The solution in the bath (ϳ0.1-0.17 ml) was then exchanged with 0.5 ml of various kinds of ␤␥ subunits dissolved in the same GDP-containing bathing solution. The exchange was done manually by using a pipetter wrapped up with a grounded aluminum foil. The degree of solution exchange was tested by using osmolarity changes as an indicator; the test showed that this manual method resulted in the replacement of about 86% of the original solution (using the bath volume of 0.15 ml).
When the inside-out patch configuration was established, usually the patch produced a very infrequent activity of GIRK channels (basal GIRK activity) and an activity of channels of small amplitude (of unknown origin) in the GDP-containing bathing solution. In some patches, the occurrence of large flickering channels (about 100 picosiemens with [K ϩ ] o of 156 mM), whose activity was not dependent on GTP, was observed (12). Frequent occurrence of these large background channels hindered the analysis of GIRK channel, and thus such patches were not included in our sample. Occasionally, instead of an inside-out patch, vesicle formation occurred. (The vesicle formation was inferred by the appearance of a current drooping because of the existence of membrane resistance and capacity on the opposite side of the patch.) (11). These patches were not included. Sometimes we observed rather vigorous basal activity of GIRK-like channels in the GDP-containing bathing solution before the application of ␤␥. This basal activity might have originated from the local presence of overexpressed ␤␥s or from the basal activity of the receptors. Because the present objective was to analyze the GIRK channel activation induced by the application of exogenous ␤␥ proteins, we did not pursue the investigation of the patches with a high frequency basal activity (more than about once per second).
Stock solutions of ␤␥ proteins were diluted with the GDP-containing bathing solution to a final concentration of 1-100 nM. At 10 nM G␤␥, the solution contained the following buffer/detergent: 0.0033% octyl glucoside, 0.066 mM HEPES, 0.0033 mM EDTA, and 0.24 -2.1 M dithiothreitol. Experiments were done with a bath temperature of ϳ21°C.
Statistical Treatment of Electrophysiological Data-The distribution pattern of NP o was almost always non-Gaussian; this can be inferred from substantial discrepancies between the mean and the median values (see Figs. 4A and 6). This non-Gaussian distribution suggests the possible existence of more than one type of channel (e.g. solitary channels and aggregates of channels). Because the mean value in our samples is influenced greatly by a small number of patches with a large NP o , the median was a more appropriate parameter to represent the channel activity of a group. Also, statistical comparisons of NP o were done, unless otherwise noted, by using the nonparametric statistics (the Kruskal-Wallis ANOVA and a posttest using the Mann-Whitney with the Bonferroni adjustment).
We used two different types of wild type ␤ 1 ␥ 2 : with and without hexahistidine tagging at the N terminus of ␤ 1 . The median value of NP o for the channel activated by the hexahistidine-tagged wild type ␤ 1 ␥ 2 was 0.109 (n ϭ 47), and the median value of NP o by nontagged ␤ 1 ␥ 2 was 0.099 (n ϭ 35) (difference not significant; p Ͼ 0.5). We will refer to these two types simply as ␤ 1 ␥ 2 .
To measure adenylyl cyclase activity, purified ␤␥ subunits were reconstituted with 10 g of membranes from Sf9 cells expressing type I or type II adenylyl cyclase for 3 min at 30°C in a final volume of 20 l. Assays were then performed for 7 min at 30°C in a total volume of 50 l containing 4 mM MgCl 2 and 0.2% octyl glucoside as described (4).

Activation of GIRK Channels by ␤ 1 ␥ 2 in Locus Coeruleus
Neurons-When a gigaseal was formed in the on-cell mode in locus coeruleus neurons, some channel activity was usually observed. Upon making an inside-out patch, this activity started to subside, reaching a low level within a minute (presumably because the intracellular GTP was washed away in exchange of the GDP-containing solution). We then applied wild type ␤ 1 ␥ 2 (10 nM), which induced, after a latency, vigorous channel activity (Fig. 1A) (13, 14). The channels showed typical GIRK-like characteristics (12,14) with a chord conductance of ϳ30 -35 picosiemens, exhibiting a mixture of short openings and long openings, the latter sometimes showing bursts (Fig.  1A, record b). The current-voltage relationship of the channels showed an inward rectification ( Fig. 2A). Thus, these channels activated by ␤ 1 ␥ 2 in locus coeruleus neurons probably belong to the GIRK (Kir3) subfamily (15,16).
10 nM Is an Almost Saturated Concentration for ␤ 1 ␥ 2 - Fig.  2B shows a concentration-response relation in wild type ␤ 1 ␥ 2 . Values of NP o over 5 to 9 min after the introduction of ␤ 1 ␥ 2 were plotted. The wild type ␤ 1 ␥ 2 activated the brain GIRK with an EC 50 of about 4 nM, and the activation almost saturated with 10 nM. These results are in approximate agreement with previous studies on cardiac and cloned GIRK channels (17)(18)(19)(20).
␤ 1 Mutations at Outer Blades: K ϩ Channel-Within the ␤ subunit family, the STE4 gene product of Saccharomyces cerevisiae is most distantly related to mammalian ␤ subunits. So far, no evidence has been presented to suggest the regulation of adenylyl cyclase, PLC, or K ϩ channel activity by yeast ␤␥ subunits. Therefore, we suspected that exchanges of the effector-interacting domains of the ␤ subunit with the correspond-ing sequences from STE4 would result in mutant proteins, which are incapable of interacting with mammalian effectors. Indeed, Peng et al. (21) have recently reported that STE4 does not activate GIRK channels.
To test the hypothesis that regions on the side surface of the ␤␥ torus are important for the regulation of effectors, we mutated residues of an outer strand of each blade into the corresponding sequence of STE4 (Fig. 3). Each ␤ mutant was coexpressed with His 6 -G␣ i1 and ␥ 2 in Sf9 cells, and mutant ␤␥ subunits were purified as described under "Experimental Procedures." First, we surveyed all these outer blade mutants (eight mutants altogether) for their ability to activate GIRK channels.  2. A, current-voltage relation of GIRK channels from locus coeruleus neurons; single channel recordings using the inside-out mode. The channel activity was induced by 10 nM wild type ␤ 1 ␥ 2 . The solid line is the fit of the data to a second order polynomial. Five patches (represented by different symbols) were used. The inset records show single channel currents at different membrane potentials. The experiment was done in the 156 mM potassium gluconate pipette solution. B, doseresponse relationship of wild type ␤ 1 ␥ 2 . For each patch, two or three different concentrations (including the standard concentration, 10 nM) of wild type ␤ 1 ␥ 2 were applied in ascending order. The channel activity was determined by averaging NP o during the 5-9 min of introducing a new concentration of ␤ 1 ␥ 2 (except that values at zero concentration were determined by NP o during 1 min before introducing ␤ 1 ␥ 2 ). For each patch, the activity was normalized to the NP o value at the standard concentration (10 nM We used ␤ 1 ␥ 2 mutants at a concentration of 10 nM, which is, for the wild type ␤ 1 ␥ 2 , an almost saturating dose for activating the K ϩ channel (Fig. 2B). The variability of NP o among different patches was quite large (see "Experimental Procedures") ( Fig.  4A). Nevertheless, the mutant D1 clearly showed far less ability to activate the GIRK channel than did the wild type (p Ͻ 0.001) (Figs. 1B and 4A). The mutant CD2 was also significantly less effective in activating the channels compared with the wild type (p Ͻ 0.05). Mainly because of the large variation of NP o , we could not obtain significant differences in NP o between the wild type and each of the other mutants (D2F, D2R, D3, D4, D6, and D7) (Fig. 4A). We, therefore, focused on D1 and CD2 mutants and tested a higher concentration for their ability to activate GIRK channels. Even at 100 nM, both D1 and CD2 produced only a small amount of channel activity. Comparison of the channel responses by D1 and CD2 with the wild type dose-response curve clearly indicates the impaired ability of these two mutants (CD1 and CD2) to activate the K ϩ channel (Fig. 4B).
␤ 1 Mutations at Outer Blades: Adenylyl Cyclases and PLC␤ 2 -Despite the impaired ability of D1 and CD2 to activate GIRK channels, these two mutants were as active as the wild type to stimulate type II adenylyl cyclase (Fig. 5A). They could also inhibit type I adenylyl cyclase, similarly to wild type ␤ 1 ␥ 2 (data not shown). Both D1 and CD2 were capable of stimulating PLC␤ 2 . The mutant D1 was, however, more active and the mutant CD2 was less active than the wild type within the concentration range of the assays (Fig. 5B).
Effects of Detergent-We also tested whether channel activities were affected by the detergent in which the protein was dissolved. We tested the buffer with 0.033% octyl glucoside, 0.66 mM HEPES, 0.033 mM EDTA, and 6 -17 M dithiothreitol, corresponding to those used for 100 nM ␤ 1 ␥ 2 . This buffer alone did not induce channel activity during ϳ10 min of application (mean NP o : 0.000189 before application; 0.000232 after application; n ϭ 5; p Ͼ 0.7; paired t test). We also compared the FIG. 3. Mutation of outer blades of ␤ 1 subunit. The amino acid sequence of bovine ␤ 1 (␤ 1 ) is aligned with the corresponding STE4. For each type of mutation (D1 through D7, except D2R), the amino acid sequence of the outer blade of the bovine ␤ 1 subunit (underlined) was replaced with the corresponding sequence of STE4 (aligned underneath, underlined). WD represents WD repeat motif. Because blade 5 (WD 6) is interacting extensively with the ␥ 2 subunit, it was excluded from the mutational analysis. In mutant D2R, valine was changed into glycine; this mutation has been shown to inhibit the ability of STE4 to transmit mating signaling (25). The names of the mutants are indicated in bold letters above the sequence (such as D1 and CD2). The empty parentheses in WD6 of STE4 represents the following sequence: (LFRGYEERTPT-PTYMAANMEYNTAQSPQTLKSTSSSYLDNQ).
FIG . 4. A, effect of replacing residues on the outermost strands of ␤ 1 with the corresponding sequences of yeast STE4 on GIRK activity in locus coeruleus neurons. The thick horizontal lines represent median values, and the heights of the columns represent mean values of NP o. The vertical lines represent S.E. The number in parentheses indicates the number of patches; for each patch, the channel activity was determined by averaging NP o during the 5-9 min after introducing various types of ␤ 1 ␥ 2 . The mean basal activity (before applying various types of ␤ 1 ␥ 2 ) was 0.0051 in NP o (n ϭ 74) and mainly originated from the small background channel activity (12). The 156 mM KCl pipette solution was used. Comparison with the wild type: *, p Ͻ 0.05 (p ϭ 0.035); ***, p Ͻ 0.001 (nonparametric ANOVA and the posttest). B, comparison between wild type dose-response relation with the responses to D1 and CD2 mutants at 10 and 100 nM. Square symbols represent the doseresponse relationship of the wild type. For each patch, the data were normalized to the response at the standard ␤ 1 ␥ 2 concentration (10 nM). This figure was derived from the same experiment as in Fig. 2B; here, NP o was averaged during the 5-7 min (instead of 5-9 min) of the solution exchange (see legend for Fig. 2B for details). Circles and triangles, respectively, represent data for D1 and CD2 mutants (at 10 and . The difference between 10 nM wild type and 10 nM D1 was significant (p Ͻ 0.01; nonparametric ANOVA and posttests). The difference between 10 nM wild type and 10 nM CD2 was marginally significant (p ϭ 0.13). Because we performed two independent tests on the differences between wild type and CD2 (10 nM) (A and B), we calculated the overall significance level by which wild type and CD2 (10 nM) are different; it gives p ϭ 0.03. In contrast, no significant differences were obtained (ANOVA, posttest) between 10 nM D1 and 100 nM D1 or between 10 nM CD2 and 100 nM CD2. effect of 10 nM wild type ␤ 1 ␥ 2 in the standard buffer with that in the high detergent buffer. Again, no marked difference was observed. The NP o during the 4 -6 min after the start of application of the wild type ␤ 1 ␥ 2 in the standard buffer was 0.233 (mean, n ϭ 7), and that in the high buffer/detergent was 0.14 (n ϭ 5). (The medians were 0.065 and 0.17, respectively; p Ͼ 0.6.) ␤ 1 Mutations at G␣/␤ Interaction Sites-␤ 1 mutants (L117A, D228R, D246S, and W332A), which are mutated at a residue of the interface between ␤␥ and ␣ subunits, were previously tested for their ability to regulate PLC␤s and adenylyl cyclases (4). We also tested these four mutants at 10 nM for their ability to activate the GIRK channels. Fig. 6 summarizes the values of NP o during the 5-9 min after the start of ␤ 1 ␥ 2 application. Two of these mutants (D246S and W332A) produced channel activity significantly lower than the wild type (p Ͻ 0.01 and p Ͻ 0.001, respectively). DISCUSSION GIRK channels are activated directly by the G protein ␤␥ subunit. The interaction sites of ␤␥ with GIRK have been investigated on the G␣ interacting surface of the ␤␥ subunit (5). In the present study, we have demonstrated that regions outside of the G␣/␤ interaction surface of the ␤ subunit also participate in the interaction with GIRK channels. We characterized the interaction of ␤ with brain GIRK by using ␤ 1 mutants on the outer strands of the seven-bladed ␤-propeller structure. Mutations of certain residues on the outer strands of blade 1 (D1) and blade 2 (CD2) resulted in the severe disruption of their ability to activate GIRK channels. However, these mutants could regulate adenylyl cyclases similarly to the wild type, suggesting that the mutations of the D1 and CD2 areas did not produce a global disruption of ␤ 1 structure. Thus, the results suggest that the mutated residues on the side of blade 1 and blade 2 of the ␤␥ torus might be specifically involved in the regulation of GIRK channels.
It should be noted that in this study, we have concentrated on analyzing mutations that produced a large functional deterioration in the K ϩ channel activation. This study does not exclude the possibility that other mutants may have a moderate defect in their ability to interact with K ϩ channels.
It was previously shown that the activation of PLC␤ 2 involves the outer strands of blades 2, 6, and 7 (6). Interestingly, CD2 also showed a defect in its ability to activate PLC␤ 2 . The results further support the importance of blade 2 for the interaction with PLC␤ 2 . As shown in Fig. 5, D1 was 2-3-fold more potent than the wild type in its ability to stimulate PLC␤ 2 . The exact reason for this difference is currently unclear.
Our results on ␤ mutations over the G␣/␤ interaction surface indicate that W332A showed far less ability to activate GIRK channels. Although only one concentration (10 nM) of ␤ 1 ␥ 2 was tested, we observed a very large difference in their activity between W332A and the wild type. Because 10 nM is an almost saturating dose for the wild type, this result suggests that W332A is much less active than the wild type in its ability to stimulate GIRK channels. Thus, ␤/GIRK interaction sites partially overlap with the G␣/␤ interaction sites, but they are not identical. This is similar to the case of ␤ subunit interaction with PLC␤ 2 or adenylyl cyclase (4,5).
It is known that ␤␥ can interact with effectors only if G␣ is dissociated from ␤␥. X-ray crystallographic studies have shown that the ␤ subunit does not undergo conformational changes when it is dissociated from G␣ (3). It has been demonstrated that GDP-bound G␣ could sever the ␤/GIRK association quickly (22,23). Because the interaction sites of GIRK overlap with the G␣ interaction surface on G␤␥, this effect of G␣ could be explained by a simple spatial (three-dimensional) competition on the ␤ subunit between G␣ and GIRK (4). It is also possible that the association of GIRK to ␤␥ induces a conformational change of the ␤ subunit, and this change would favor the binding of ␤␥ and GIRK. Conformational changes of the ␤␥ subunit are demonstrated in the complex of ␤␥ with phosducin (24). The bind- ing of phosducin to the ␤␥ subunit produces a distinct conformational change in blade 6 and blade 7 of the ␤ subunit. The phosphorylation of phosducin on Ser-73 reduces its affinity for ␤␥ and the released ␤␥ subunit and then switches back to the conformation of free ␤␥ or that of the heterotrimer (24). In the case of ␤␥ complexed with GIRK, when G␣ i interacts with a certain region of G␣/␤ interaction sites, the conformation of ␤␥ could return to the resting state (before GIRK was attached), and this could decrease the affinity of GIRK with ␤␥. The determination of the structure of ␤␥ complexed with GIRK will be necessary to answer this question.