Regulators of G-protein Signaling Form a Quaternary Complex with the Agonist, Receptor, and G-protein

Regulators of G-protein signaling (RGS) proteins modulate signaling through heterotrimeric G-proteins. They act to enhance the intrinsic GTPase activity of the Gα subunit but paradoxically have also been shown to enhance receptor-stimulated activation. To study this paradox, we used a G-protein gated K+ channel to report the dynamics of the G-protein cycle and fluorescence resonance energy transfer techniques with cyan and yellow fluorescent protein-tagged proteins to report physical interaction. Our data show that the acceleration of the activation kinetics is dissociated from deactivation kinetics and dependent on receptor and RGS type, G-protein isoform, and RGS expression levels. By using fluorescently tagged proteins, fluorescence resonance energy transfer microscopy showed a stable physical interaction between the G-protein α subunit and RGS (RGS8 and RGS7) that is independent of the functional state of the G-protein. RGS8 does not directly interact with G-protein-coupled receptors. Our data show participation of the RGS in the ternary complex between agonist-receptor and G-protein to form a “quaternary complex.” Thus we propose a novel model for the action of RGS proteins in the G-protein cycle in which the RGS protein appears to enhance the “kinetic efficacy” of the ternary complex, by direct association with the G-protein α subunit.

studies have identified a large RGS gene family, each endowed with a conserved RGS domain of 120 -130 amino acids that is flanked by N and C termini of varying lengths (3)(4)(5)(6)(7). By itself, the RGS domain is capable of interacting with G-protein ␣ subunits to accelerate the GTP hydrolysis rate of the G␣ subunit, thereby promoting termination of the G-protein signal (3)(4)(5)(6)(7). Based on primary sequence similarities, mammalian RGS proteins have been grouped into five subfamilies (7). In this study we focus largely on RGS8, belonging to the R4 subfamily of RGS proteins that are generally considered prototypical in that they appear to have little function other than to act as GTPase-activating proteins (GAPs) on G i/o and G q/11 G-protein ␣ subunits. However, we also examine RGS7, which belongs to the R7 family, and GAIP, which belongs to the RZ family. RGS7 is particularly interesting because it contains a number of protein-protein interaction domains in the N terminus, but it is of particular relevance for our study because it has a substantially attenuated GAP activity compared with other RGSs (8).
The discovery of the RGS protein family had important ramifications for the study of the G-protein gated K ϩ channel (GIRK) that was first identified in atrial myocytes and activated by acetylcholine acting at muscarinic M2 receptors (9). Activation of GIRK is responsible, in part, for slowing of the heart rate in response to vagal stimulation (10,11). Analogous GIRK currents are present in neurons and neuroendocrine cells (12). The native channel is a heterotetrameric complex composed of the inwardly rectifying K ϩ (Kir) channel subunits, Kir3.1-3.4 (12). Activation of native and cloned G-protein gated K ϩ channels has been shown to involve a direct, membranedelimited interaction with the G␤␥ subunit (13,14). A number of studies have sought to explain why these channels, when expressed in Xenopus laevis oocytes, deactivate more slowly than the native atrial current upon termination of the receptor stimulus. Identification of the RGS family has resolved this discrepancy. For example, expression of RGS4 or RGS8 in X. laevis oocytes accelerates GIRK deactivation kinetics such that the measured time constants are more consistent with those occurring after stimulation of native channels in atrial cells (15,16).
A paradox arose, however, because RGS protein expression not only enhanced channel deactivation but also accelerated GIRK activation (15,16). Furthermore, RGS expression did not attenuate the overall signal, an effect that would be predicted if the only function of the RGS were to act as a GAP (16,17). This interesting phenomenon has resulted in a number of new proposals for the mechanism of action of RGS proteins and has important consequences for our understanding of the G-protein cycle. RGS proteins may function as scaffolds to help preassemble signaling complexes of receptor with G-protein, for example (17)(18)(19). Alternatively, RGS proteins may promote entry into the G-protein cycle by physically aiding the dissociation of G␣-GTP and G␤␥ subunits or by encouraging the formation of fast cycling signaling complexes, acting according to a "kinetic scaffolding" mechanism (7,20,21). Previously, we have presented data showing the importance of the ternary complex in determining the selectivity and kinetics of GIRK activation (22)(23)(24). Briefly stated, this hypothesis ascribes a unique conformation for the complex formed between various combinations of agonist, GPCR, and G-protein, which dictates the speed and efficiency of G-protein activation. This hypothesis is consistent with many of the recent models used to explain GPCR signaling (25,26). In this paper we present data to show that the paradoxical action of RGS proteins on GIRK activation kinetics can be resolved if the RGS enters into the ternary complex to form a "quaternary complex." Furthermore, based on our FRET studies, we propose a stable physical interaction between the G-protein ␣ subunit and the RGS protein that is independent on the functional state of the G-protein.

MATERIALS AND METHODS
Molecular Biology, Cell Culture, and Transfection-The human clones of RGS8, RGS7, and GAIP were obtained from the Guthrie cDNA Resource Centre. A PCR-based approach was employed to clone RGS8, RGS7, and GAIP in-frame into pEYFP-N1 (Clontech) by using HindIII and BamHI (RGS8 and GAIP) and XhoI and BamHI (RGS7) as cloning sites. Fluorescently tagged receptors (D2-CFP, A1-CFP, ␣2A-CFP, and M4-YFP) were constructed by using a similar general approach in the pECFP-N1 and pEYFP-N1 vectors using KpnI and HindIII as the cloning sites. Inducible expression of RGS8-YFP was achieved using the TRex system (Invitrogen). RGS8-YFP was excised from pEYFP-N1 and ligated into pcDNA5/TO with a HindIII/NotI restriction digest. The methods for cell culture and transient transfection and the techniques for establishing stable cell lines with HEK293 cells have been described previously (22,27). "Quadruple" inducible stable cell lines were established after the transfection of RGS8-YFP in pcDNA5/TO and pcDNA6/TR (both Invitrogen) into the HKIR3.1/3.2/M4 cell line and subsequent selection with 727 g/ml G418, 364 g/ml Zeocin, 400 g/ml hygromycin, and 5 g/ml blasticidin. In this stable cell line, designated as HKIR3.1/3.2/M4/R8YFP-T, gene expression was conditional upon the addition of the antibiotic tetracycline (Sigma). Treatment of cells with tetracycline (0.01-10 g/ml) for 24-36 h resulted in graded expression of the fluorescently tagged protein. G o ␣-YFP and G s ␣-CFP were constructed in an analogous fashion to G o ␣ A -CFP (28). G␥2-CFP was the kind gift from Dr. S. Ikeda. All other plasmids were used as described previously. A CFP-YFP dimer was constructed by cloning an in-frame PCR fragment of enhanced YFP (Clontech) into enhanced CFP-N1 in the XhoI-BamHI sites of the polylinker.
Electrophysiology-Whole-cell membrane currents were recorded with an Axopatch 200B amplifier, and data were acquired with a Digidata 1200B interface (both Axon Instruments) and analyzed with pClamp software (version 6.0; Axon Instruments). Cell capacitance was ϳ15 picofarads, and series resistance (Ͻ10 megohms) was at least 75% compensated using the amplifier circuitry. After an equilibration period of ϳ5 min, cells were voltage-clamped at Ϫ60 mV; records were digitized at 100 Hz, and agonist-induced currents were measured at this potential. Rapid drug application and removal were achieved as described previously (24,29,30). Agonist was applied for 20 s. Upon agonist application, the current was activated with an initial lag and then a subsequent rise to peak current amplitude (time-to-peak). Activation kinetics are described as lag ϩ time-to-peak. On removal of the agonist, the current deactivated back to base-line levels. Deactivation was generally well described by a single exponential function, I(t) ϭ A⅐exp(Ϫt/) ϩ C, where A is the current amplitude at the start of the fit; t is time; is the deactivation time constant, and C is the steady state asymptote. Curve fitting was performed by using Clampfit software (pClamp version 6.0). After removal of the agonist dopamine from HKIR3.1/3.2/D2 cells, the deactivation phase was found to be better described by the "sum of two exponentials" equation: I(t) ϭ A1⅐exp(Ϫt/ 1) ϩ A2⅐exp(Ϫt/2) ϩ C. After removal of the agonist adenosine from HKIR3.1/3.2/A1 cells, the deactivation phase was not adequately fit by a single exponential function. Therefore, the measure "time to halfcurrent decay" was used to describe the rate of deactivation. This is defined as the time taken from removal of agonist for the agonistinduced current to deactivate to half its peak value.
Fluorescence Microscopy-Cells for imaging were subcultured onto 25-mm glass coverslips or 35-mm culture dishes with integral number 0 glass coverslip bottoms (Mattek) and, if transiently transfected, were allowed 24 -48 h to express the protein(s) of interest.
Confocal Microscopy-Prior to imaging, a coverslip was placed into a watertight cell imaging chamber at room temperature, and cells were overlaid with HEPES-buffered Opti-MEM without phenol red (Invitrogen). HEK293 cells were imaged by using a Bio-Rad Radiance 2100 confocal microscope using a 60ϫ Nikon Plan Apo oil objective (1.40 NA). CFP was excited with a 457-nm laser line, and images were obtained using a 470 -500-nm bandpass filter. YFP was excited with a 514-nm laser line, and emission was measured between 530 and 570 nm. The FRET imaging conditions were obtained with excitation by using a 457-nm laser line, and emission was measured between 530 and 570 nm. Multiple images were acquired sequentially. Intensities in the CFP, YFP, and FRET set of imaging conditions were determined from membrane-delimited regions of interest drawn by hand at high magnification using the LaserPix software. The background-subtracted intensities were analyzed to determine FRET ratios using three-cube protocols (see below). 16-Bit images were obtained with identical laser powers, photomultiplier gain, and pinhole size and were optimized to examine cells with moderate expression of both constructs. Care was taken to avoid saturating images.
CCD-based Fluorescence Microscopy-We also used a standard fluorescent microscope (Nikon TE200 60ϫ Plan Apo oil objective 1.40 NA), equipped with a back-illuminated digital CCD camera (Roper Scientific MicroMax 1024 EB) and high speed CCD detector control, to measure FRET ratios and changes in these ratios with perfusion. Samples were excited by using a mercury lamp with an excitation filter wheel, and emission filters were selected through an emission filter wheel (Sutter Instruments, Lambda-10/2). The following filter sets were used (excitation and emission): YFP (500 Ϯ 10 nm; 535 Ϯ 15 nm), CFP (430 Ϯ 12.5 nm; 465 Ϯ 20 nm), and FRET (430 Ϯ 12.5 nm; 535 Ϯ 15 nm). 16-Bit images were acquired and analyzed with the Universal Imaging Corp. metamorph software (version 6.1). Background values employed were regions containing no cells in the viewing field. 3-Cube parameters and FRET ratios were calculated as above.
It worth noting that on both microscopy systems the S.E. of the FRET ratio was less than 5% of the mean, and this compares very favorably with other methods (31). Given the different imaging conditions in the two setups, the relationship between FRET efficiency and FRET ratio will be different as it depends on the ratio of the CFP and YFP molar extinction coefficients at the respective excitation wavelength (32). Practically, exciting CFP at the suboptimal 457 nm on the laser-scanning confocal microscope leads to a smaller FRET ratio given a particular FRET efficiency.
FRET Ratio and Other Calculations-Colocalization ratios were determined using the LaserPix software. The FRET ratio (FR) measures the fractional increase in YFP emission because of FRET and was calculated according to Equation 1 (32), ). If the FRET ratio ϭ 1 (i.e. no FRET) then the above equation will equal 0. Any excess signal (i.e. FRET ratio Ͼ1) will manifest as increased intensity above background. It is possible to manipulate images on a pixel-by-pixel basis and thus obtain spatial information on the FRET signal. Images were pseudocolored, filtered, and converted to 24-bit RGB files for display (TIFF or JPEG).
Data Analysis-Membrane currents were measured at Ϫ60 mV, and all data are presented as mean Ϯ S.E., where n indicates the number of cells recorded. Data were analyzed for statistical significance using either Student's t test or one-way analysis of variance with Dunnett's post-test where appropriate; * indicates p Յ 0.05; ** indicates p Յ 0.01, and *** indicates p Յ 0.001. Time measurements were reciprocated prior to statistical analysis.

RESULTS
As in our previous studies (22,23), we used a HEK293 stable cell line that expresses the GIRK channel subunits Kir3.1 and Kir3.2A (HKIR3.1/3.2) and dual channel plus GPCR stable lines (adenosine A1, HKIR3.1/3.2/A1; adrenergic ␣2A, HKIR3.1/3.2/␣2; dopamine D2S, HKIR3.1/3.2/D2; muscarinic M4, HKIR3.1/3.2/M4; and the heterodimeric GABA-B1b ϩ GABA-B2, HKIR3.1/3.2/GGB). Cells were studied with the whole-cell configuration of the patch clamp technique, and drugs were applied using a rapid agonist application system. Initial transfection studies were carried out to confirm that RGS8-YFP and RGS7-YFP cDNAs could express functional proteins that mediated effects similar to the untagged protein. Fig. 1A illustrates the effects of expression of RGS8, RGS8-YFP, and RGS7-YFP in a cell line expressing the heterodimeric GABA-B1b/2 receptor. Application of a saturating dose of the appropriate agonist, in this case 100 M baclofen, rapidly activated an inward K ϩ current. Activation was characterized by an initial delay (lag) and then a subsequent rise to a peak value (time-to-peak, "ttp") (the activation kinetics were quantified by lag ϩ ttp). After expression of either RGS8 or RGS8-YFP, there was enhancement of both GIRK activation ( Fig. 1B (i)) and deactivation kinetics ( Fig. 1B (ii)) by both constructs. Thus RGS8-YFP is functional with effects equivalent to that of the untagged protein: there was no significant difference between RGS8 and RGS8-YFP in the acceleration of activation and deactivation. Saitoh et al. (57) have reported that expression of RGS7 profoundly accelerated activation kinetics but had a much weaker effect on deactivation kinetics. We were able to reproduce this observation, and we also observed that RGS7-YFP had a similar effect (Fig. 1B). Thus, RGS7-YFP is functional with effects equivalent to that of the untagged protein; there were no significant differences between RGS7 and RGS7-YFP in the acceleration of activation and in the failure to accelerate deactivation kinetics (Fig. 1B, (i) and (ii)). We did not find it necessary to coexpress G␤5 with RGS7 to observe pronounced functional effects and presumably RGS7 complexes with endogenously expressed G␤5 (33,34). It is also known that G␤5 containing G␤␥ complexes can inhibit GIRK channels (35), and overexpression of this protein would potentially complicate our electrophysiological analysis.
We have found previously that with certain agonist/GPCR combinations, the deactivation rate is not determined by the rate of GTP hydrolysis by the G-protein but by processes occurring at the level of the receptor, particularly the agonistunbinding rate (30). For example, for the synthetic agonists 5Ј-N-ethylcarboxamidoadenosine at the A1 receptor or quinpirole at the D2S receptor, the deactivation rate was slow and was not altered by overexpression of RGS8 (30). However, with the agonists adenosine and dopamine at their respective GPCRs, the rate of current deactivation was faster and strongly enhanced by RGS8-YFP expression (Fig. 2). We then questioned whether the kinetic effects of RGS8 on activation might be selective for different GPCRs. We examined the effects of RGS8-YFP expression on GIRK channel kinetics in cells stably expressing different GPCRs, in response to fast application and removal of agonists. We have previously used radioligand binding to show equivalent levels of receptor expression in the HKIR3.1/3.2/A1, HKIR3.1/3.2/␣2A, and HKIR3.1/3.2/D2 cell lines (24).
When RGS8-YFP was expressed in HKIR3.1/3.2/M4, HKIR3.1/3.2/D2, and HKIR3.1/3.2/GGB cells, both current activation and deactivation were significantly accelerated upon FIG. 1. Effect of RGS on GIRK channel kinetics. Transient transfections were performed in the HKIR3.1/3.2/GGB cell line with 1 g of the relevant cDNA (50 ng of enhanced green fluorescent protein was cotransfected with the untagged RGS). Fluorescent cells were recorded 24 -72 h post-transfection. A, representative whole-cell current traces recorded at a holding potential of Ϫ60 mV were made from cells transfected as indicated. The bar above, positioned at the zero current level, indicates a 20-s agonist application (100 M baclofen). The sum of the lag and time-to-peak was used to characterize activation kinetics ("lag ϩ ttp"). Deactivation was characterized using the time constant (deactivation tau, OFF ) from single exponential curve fits to the declining current phase. B, graphs summarize these kinetic parameters. Activation (i) and deactivation (ii) are significantly accelerated by RGS8 and RGS8-YFP. In contrast, RGS7 and RGS7-YFP accelerate only the activation kinetics but not the deactivation kinetics. the application and removal of an appropriate agonist (Figs. 1A and 2). In contrast, in the HKIR3.1/3.2/A1 and HKIR3.1/3.2/␣2 cell lines, only deactivation kinetics were accelerated by RGS8-YFP (Fig. 2). In the HKIR3.1/3.2/A1 cell line, basal currents were significantly reduced after RGS8-YFP expression (control, Ϫ125 Ϯ 26 pA/pF; ϩRGS8-YFP, Ϫ69 Ϯ 6 pA/pF; p Ͻ 0.05, t test), which is consistent with a GAP function. Thus it appears that the effects of RGS8-YFP on the kinetics of current activation are dependent on the GPCR studied. For those GPCRs that exhibit relatively slow GIRK activation, RGS8-YFP can accelerate the kinetics, whereas for those that are intrinsically fast (A1 and ␣2A), the expression of RGS8-YFP has no effect. Finally, to ensure these observations were not because of YFP tagging the C terminus of RGS8, untagged RGS8 (1 g) was expressed in the HKIR3.1/3.2/␣2A cell line, and the kinetics were analyzed. Activation was similarly unchanged (␣2A control, lag ϩ ttp (s), 1.13 Ϯ 0.05; ϩRGS8, 1.42 Ϯ 0.25, n ϭ 12; not significant), and deactivation was significantly enhanced (␣2A control, OFF (s), 5.05 Ϯ 1.00; ϩRGS8, 2.39 Ϯ 0.60s, n ϭ 12, p Ͻ 0.01). Therefore, activation kinetics appeared to approach a limiting value.
There are some additional points to note. First, with dopamine as an agonist, deactivation was significantly accelerated by RGS8-YFP overexpression, but on closer kinetic analysis, two distinct deactivation time constants were revealed, a fast and slow component ( Fig. 2A). Second, it was necessary to use the measurement time to half-current decay (see "Materials and Methods") in the A1 recordings due to difficulty in fitting single exponential curves to the current deactivation phase as noted previously (30). Finally, we have shown previously that untagged RGS8 overexpression did not significantly accelerate activation kinetics in the M4 line (30); however, this was based on fewer observations, and there is more intrinsic biological variability in this line as evidenced by the error bars (Fig. 2B).
It is known that different G i/o -coupled GPCRs interact preferentially with certain G i/o ␣ isoforms and that RGS proteins may show selectivity for different G-proteins. Therefore. we addressed the question of whether RGS8 was acting selectively on certain G i/o ␣ isoforms. G-proteins of the G i/o class are characteristically inhibited by pertussis toxin (PTx), which acts by ADP-ribosylating a cysteine four amino acids from the end of the C terminus. Mutation of this residue to either glycine or isoleucine renders the G-protein ␣ subunit resistant to the action of the toxin but still functional (36 -38). Thus by PTxtreating cells to inactivate endogenously expressed G i/o ␣ subunits, it is possible to selectively study the interactions of a transfected mutant G-protein. We have shown that these mutant G-proteins can substitute for native G-proteins, and we have used this approach to dissect out the interaction of different GPCRs with distinct G i/o isoforms (23,24). When initially characterized, RGS8 was found to coimmunoprecipitate with G o ␣ A and G i ␣ 3 subunits (16). Therefore, we asked if RGS8 showed substrate specificity for specific G-protein ␣ isoforms. In Fig. 3, PTx-resistant forms of either G i ␣ 2 or G o ␣ A were expressed in the HKIR3.1/3.2/GGB cell line. The activation and deactivation kinetics of currents mediated by either G i ␣ 2 or G o ␣ A were significantly enhanced by coexpression of RGS8-YFP (Fig. 3B). As noted previously (24), the GABA-B1b/2 receptor activates currents significantly faster via the G o ␣ A subunit when compared with G i ␣ 2 . However, RGS8-YFP had a significant effect on current activation rates mediated by both G-protein isoforms. Next, the A1 receptor was constrained to signal via the G o ␣ A subunit with or without RGS8-YFP. The results (Fig. 3, C and D) demonstrate that activation of the A1 receptor-mediated currents is not accelerated by RGS8-YFP expression when currents are activated exclusively via G o ␣ A . Deactivation was significantly enhanced. Clearly the differential effects of RGS8 on channel activation do not occur solely at the level of the G-protein, but it is the combination of receptor with G-protein that matters.
We next investigated how varying the concentration of RGS8-YFP affected the kinetics of current responses. We used a "Tet-On" system and developed a monoclonal cell line on the background of the HKIR3.1/3.2/M4 cell line, additionally expressing RGS8-YFP in an inducible vector. In the absence of tetracycline (Tet), there was little basal RGS8-YFP expression. Concentrations of Tet were selected that yielded intermediate (0.3 g/ml) and saturating (10 g/ml) levels of RGS8-YFP expression (Fig. 4A). We observed a graded acceleration of deactivation kinetics with increasing Tet concentrations (Fig. 4B). In contrast, the change in activation kinetics was smaller and did not reach statistical significance. Thus it seems that at lower expression levels there is a graded acceleration of deactivation kinetics before activation kinetics are enhanced.
We have shown previously that HEK293 cells possess endogenous RGS proteins, and we found that endogenous RGS proteins solely modulate deactivation kinetics via the GABA-B receptor, without affecting activation kinetics (30). We further investigated the role of endogenous RGS proteins in the HKIR3.1/3. Activation through the A1 receptor, constrained to signal through G o ␣ A (C-G) subunits, is not enhanced by RGS8-YFP expression, whereas deactivation, measured using the time to half-current-decay parameter, is significantly accelerated by RGS8-YFP. channel activation. The effect on activation (less than a 3-fold change) was not as substantial as that on deactivation (greater than 10-fold change).
To investigate the trafficking and the interaction of components in this signaling cascade, we used laser-scanning confocal microscopy to examine the subcellular distribution of RGS8-YFP. Another RGS protein from a separate family, namely GAIP (RGS19), was also YFP-tagged and imaged to test the generality of our observations. We expressed RGS8-YFP and GAIP-YFP in D2S receptor-expressing cells (HKIR3.1/3.2/D2) with and without cotransfection of the G o ␣ A subunit. Alone, RGS8-YFP was located in the cytosol and nucleus and GAIP-YFP in the cytosol (Fig. 6A). Cotransfection of G o ␣ A resulted in a dramatic redistribution of RGS8-YFP to the membrane (Fig.  6A). This effect was quantified by colocalization studies with CFP-tagged G-protein ␣ subunits. We have shown previously (28) such constructs to be functional and able to participate in receptor-mediated coupling. However, more subtle differences between untagged and tagged G-protein cannot be excluded. In control experiments, RGS8-YFP was coexpressed with mem-CFP (CFP targeted to the membrane by a dual palmitoylation sequence). Our findings were that coexpression of G o ␣ A -CFP trafficked RGS8-YFP to the plasma membrane (Fig. 6B, upper  two panels), whereas G s ␣-CFP (Fig. 6B, lower panel) and mem-CFP did not (colocalization data are given in Fig. 6D). Coexpression of GAIP-YFP with G o ␣ A -CFP or G i ␣ 3 -CFP resulted in a similar translocation to the plasma membrane; again this did not occur with mem-CFP (Fig. 6, C and D). RGS7-YFP also showed similar membrane translocation with G o ␣ A -CFP (see below). Finally, the potential for RGS8-YFP to translocate to the plasma membrane after activation of a G i/o -coupled receptor was examined. RGS8-YFP was expressed in HKIR3.1/ 3.2/D2 cells, and no translocation was observed on application of dopamine (n ϭ 7, Fig. 6E).
To investigate the potential for protein-protein interaction between RGS8 and G o ␣ A in living cells, we performed FRET measurements between RGS8-YFP and G o ␣ A -CFP. By using laser-scanning confocal microscopy, we examined whether there was basal FRET between these two proteins transiently expressed in HEK293 cells. We collected images corresponding to CFP, FRET, and YFP wavelengths in cells expressing RGS8-YFP and G o ␣ A -CFP alone (Fig. 7A, top two panels) or the combination of the two proteins (Fig. 7A, bottom panel). It can be seen from the images (Fig. 7A, top two panels) that there is only modest bleed through of signal into the FRET wavelength. Parameters obtained from such images were used to correct for bleed through using the three-cube method. Fig. 7B (upper panels) shows application of the three-cube method as follows: the "net FRET" image clearly demonstrates that cotransfection of G o ␣ A -CFP and RGS8-YFP led to a significant net FRET signal delimited to the plasma membrane. A further confirmation of FRET between G o ␣ A and RGS8 comes from examining acceptor photobleaching on cells cotransfected with G o ␣ A -CFP and RGS8-YFP (Fig. 7B, lower panels). Under conditions where FRET occurs, photobleaching the acceptor leads to an increase in donor fluorescence intensity. Our experiments show that intensity of G o ␣ A -CFP fluorescence increases by 11 Ϯ 2% (n ϭ 11) if RGS8-YFP is bleached (RGS8-YFP intensity decreases by 75 Ϯ 4%, n ϭ 11). In addition, under these microscopy conditions, the CFP signal bleaches by ϳ10% after obtaining each image, and thus the true increase is probably larger. We preformed a similar analysis on RGS7-YFP. Coexpression of G o ␣ A -CFP with RGS7-YFP led to a significant net FRET signal delimited to the membrane (Fig. 7C).
We also performed experiments with a mutant RGS8-YFP, RGS8(N122H)-YFP. When introduced into RGS4, this mutation had no effect on deactivation kinetics but preserved acceleration of GIRK activation (40) thereby converting RGS4 to act in a kinetic manner equivalent to RGS7. In the HKIR3.1/3.2/ GGB cell line, we found that expression of RGS8(N122H)-YFP did not significantly accelerate deactivation, whereas activation was enhanced (deactivated is 1.8 Ϯ 0.26 s, not significant; lag ϩ ttp ϭ 1.07 Ϯ 0.11 s (n ϭ 13) p Ͻ 0.05). The RGS8(N122H)-YFP protein was more diffusely distributed in the cytoplasm on coexpression with G o ␣ A -CFP, and there was still significant but reduced FRET (FRET ratio ϭ 1.31 Ϯ 0.10, n ϭ 15, p Ͻ 0.01).
It is possible to quantify the FRET signal by calculating three-cube FRET ratios from regions of interest defined on the relevant cellular compartment. We first performed control experiments to verify that three-cube methods provide sensitive and non-artifactual detection of FRET. Cells coexpressing CFP and YFP showed no FRET (FRET ratio ϳ1; Fig. 9). Coexpression of two membrane-localized GPCRs D2-CFP and M4-YFP, which are not thought to form heterodimers (41), also did not show FRET (FRET ratio ϳ1; top panel of Fig. 8B and Fig. 9). Cells expressing the CFP-YFP tandem dimer showed a FRET ratio of ϳ2.5, a positive control that calibrated the dynamic range of the system. Coexpression of RGS8-YFP and G o ␣ A -CFP with and without overexpression of G␤1 and G␥2 and RGS7- YFP and G o ␣ A -CFP gave a FRET ratio of ϳ1.7 to 1.8. Measurement of the FRET ratio after cotransfection of the CFPtagged GPCRs with RGS8-YFP revealed the value was not different from 1 (lower panels of Figs. 8B and 9).
To test further the validity of these FRET measurements, it was necessary to exclude the possibility that FRET had arisen from random collision of fluorescent species in the membrane (42)(43)(44). This phenomenon is known to happen in regions of high fluorophore intensity. Therefore, we plotted the FRET ratio against donor and acceptor intensity per pixel, which is equivalent to the concentration (Fig. 10, A and B). Three-cube measurements show a tendency for increased FRET ratios at higher donor concentration probably because of random collision (44). This trend was present in our data, although at similar intensity levels, the FRET ratio was significantly higher for RGS8-YFP and G o ␣ A -CFP compared with the two non-interacting membrane proteins, D2-CFP and M4-YFP, expressed at comparable levels but below that for the tandem CFP-YFP dimer (Fig. 10A). The dependence of FRET between RGS8-YFP and G o ␣ A -CFP on the donor concentration is also likely to reflect complex formation occurring at higher levels of expression. Edidin and co-workers (42,43) have shown that FRET occurring from stable interaction should result in a FRET ratio that is independent of the acceptor intensity at a fixed donor:acceptor ratio. Indeed there was only a weak correlation between FRET ratio and acceptor intensity for the RGS8-YFP and G o ␣ A -CFP interaction (Fig. 10B).
The G o ␣ A -CFP construct contains a mutation to render it resistant to the action of PTx (28). We examined whether RGS8-YFP was able to functionally interact with G o ␣ A -CFP; overexpression of RGS8-YFP in PTx-treated cells coexpressing G o ␣ A -CFP led to an acceleration of deactivation and also under these conditions activation kinetics via the ␣2A receptor (Fig.  11A). Therefore, there is functional coupling between the two fluorescently tagged proteins. To examine whether the FRET signal is dependent on the functional state of the G-protein, we Net FRET signal is delimited to the plasma membrane. In the lower panels acceptor (i.e. RGS8-YFP) photobleaching was carried out using the 514-nm laser line. There is a subtle but significant increase in donor fluorescence. C, representative images of HEK293 cells expressing G o ␣ A -CFP and RGS7-YFP. used a CCD-based microscopy system by which we could more easily perfuse the cell under investigation. As conditions for the excitation of CFP are better optimized on the CCD-based system (see "Materials and Methods"), the dynamic range of the FRET ratio for a given FRET efficiency is greater. For example, the CFP-YFP dimer yielded a FRET ratio of 5.25 Ϯ 0.2 (n ϭ 12), whereas coexpressed CFP/YFP gave a FRET ratio close to 1, and RGS8-YFP/G o ␣ A -CFP had an FR significantly greater than 1 (Fig. 11B). To encourage the formation of the heterotrimer, we first coexpressed RGS8-YFP with G␤1 and G␥2. The FRET signal between RGS8-YFP/G o ␣ A -CFP was marginally reduced but was still substantial (Fig. 11B). Equivalent experiments on the confocal microscope yielded a similar result (Fig. 9). Second, we examined for FRET between RGS8-YFP and G␥2-CFP (cotransfected with G o ␣ A and G␤1 for efficient membrane localization). Once again we observed a significant FRET ratio suggesting close proximity of RGS8-YFP to the assembled heterotrimer (Fig. 11B). The next step was to look at the possible modulation of the FRET signal between G o ␣ A -CFP and RGS8-YFP by agonist activation of the GPCR. In the D2S and ␣2A cell lines, FRET ratios were unchanged by activation of the GPCR with the relevant agonist (Fig. 11C).

DISCUSSION
There are two major hypotheses to account for enhancement of GIRK channel activation by RGS proteins: "physical scaffolding" of the GPCR, G-protein, and possibly channel by the RGS protein, or a kinetic mechanism. The physical scaffolding hypothesis has the potential to account for receptor-selective effects and kinetic effects on GIRK channel activation by bringing certain components of this signaling system into closer proximity. In pancreatic acinar cells, RGS4 was found to preferentially inhibit Ca 2ϩ signaling initiated at the M3 receptor compared with the cholecystokinin receptor (45). A scaffolding function for RGS4 has also been proposed in the preassembly of a signaling complex of the M2 receptor, heterotrimeric G-protein and GIRK channel, and the ␣2A adrenergic receptor, RGS8 and GIRK channel (15,18,19). More recently, it has been proposed that these effects can be explained solely by the GAP activity of RGS proteins: a process termed kinetic scaffolding (7,20,21). In this formulation, in an area of high receptor density with a saturating dose of agonist, receptor to G-protein signaling could potentially be attenuated by the rapid depletion of G-protein heterotrimers. By increasing G-protein turnover, the GAP activity of RGS proteins should prevent a rate-limiting depletion of heterotrimeric G-protein. Currently, the only evidence in support of this hypothesis is biochemical: Zhong et al. (21) demonstrated increased binding of GTP␥S to membranes in the presence of RGS4, the ␣2A receptor, and agonist.
In our studies there is dissociation between the effects of RGS proteins on activation and deactivation kinetics. The data obtained with the GAP-defective RGS7 is particularly telling; RGS7 is ineffective at enhancing GIRK deactivation, which is consistent with the biochemical data (8), but we found RGS7 to be as effective as RGS8 on activation. Another example is that low concentrations of RGS8-YFP enhance deactivation without having an effect on activation kinetics. Therefore, the effects of RGSs on GIRK activation cannot be explained solely by a kinetic scaffolding mechanism as one would expect a parallel increase in activation and deactivation kinetics. However, we cannot completely exclude a kinetic component from the mechanism of action of RGS proteins. For example, an N-terminally deleted RGS8 construct was found to exert equivalent effects to the full-length protein. 2 Our data support a novel model that contains many elements of the physical scaffolding hypothesis. We propose that RGS8 and RGS7 can form a stable complex with the G-protein ␣ subunit at the plasma membrane that is independent of the state of the G-protein. To support this claim, we have performed a series of experiments using FRET microscopy between fluorescently labeled signaling components. Our data indicate that the RGS binds the heterotrimeric G-protein in addition to G␣-GTP and transition complex. Furthermore, data obtained in living cells show an interaction that is not competed away by G␤␥ overexpression and occurs in the presence and absence of receptor activation. In addition, FRET can occur between the RGS and G␥. Given our current methodology, it would be difficult to detect a subtle change in FRET ratio indicative of conformational rearrangement, and this cannot be excluded. In agreement with the findings of other investigators (46), the G-protein ␣ subunit appears to recruit the RGS (RGS8, RGS7, and GAIP) to the plasma membrane. Recent studies (47) indicate that some activity in the G-protein cycle may be necessary to promote membrane attachment through RGS7 palmitoylation. It is worth emphasizing that the fluorescent G␣ subunits used here tend to preferentially adopt an inactive conformation (28), and no further increase in translocation or in the FRET signal was observed upon G-protein activation. FRET is purely a distance-dependent phenomenon, and it is possible that intermediary proteins may be involved in supporting the interaction between the RGS and G␣ subunit.
Our theory goes against the grain of conventional biochemical thinking that RGS proteins only associate with the active G␣ subunit, with the highest affinity for the transition state of the G-protein and little or no affinity for G␣-GDP (6). In addition, from the crystal structure it is clear that the binding sites for RGS and G␤␥ overlap on G␣ in its hydrolytic transition state (48). However, there are some recent indirect indications that other RGS proteins may be able to interact more weakly with other states of the G-protein (49,50). Therefore, it is plausible that the interaction we observe between RGS8 and 2 A. Benians and A. Tinker, unpublished observations. the heterotrimeric G-protein may be weaker in affinity and may occur over different domains of both proteins. Other investigators (50) have detected an ability of the M2 receptor to traffic RGS2 and RGS4 to the plasma membrane. We were unable to observe a direct interaction between three different GPCRs (␣2A-CFP, D2-CFP, and A1-CFP) and RGS8-YFP. Thus, at least for RGS8, scaffolding with the receptor appears to only occur by virtue of interaction with the G-protein.
The second important point to our model is the role that the RGS may have in shaping the behavior of the ternary complex. Although acceleration of deactivation kinetics (using an appropriate agonist) was an invariable feature of RGS8 overexpression, enhancement of activation kinetics only occurred with a subset of receptors. We have shown previously that the particular combination of agonist/receptor/G-protein can have a profound influence on the channel activation kinetics (24). Activation kinetics with a saturating dose of agonist at the D2S, GABA-B1b/2, and M4 receptors was accelerated by RGS8 whereas that via A1 and ␣2A receptors was not. Closer inspection of the data reveals that activation via A1 and ␣2A is intrinsically fast. Our interpretation is that activation kinetics essentially reach a limiting value that may or may not require RGS, depending on the receptor. Hence, we propose that the intrinsically rapid kinetics seen with the A1 and ␣2A receptors reflect an efficient ternary complex of agonist, receptor, and G-protein. Essentially the kinetic efficacy cannot be improved upon. In the case of the D2, M4, and GABA-B receptors, in which the ternary complex displays less kinetic efficacy, RGS8-YFP strengthens it. Our data show that with all five receptors the RGS participates by direct protein-protein interaction in the ternary complex with the G-protein to form a quaternary complex; it is just the functional consequences that differ. We have shown previously that endogenous RGSs may significantly influence receptor-mediated GIRK channel activation and deactivation kinetics (30), and this unavoidable complication should be borne in mind in the interpretation of receptorselective effects. Such factors may explain inconsistencies between different expression systems.
This complex formation, accounting for the acceleration of activation kinetics, may only occur at higher RGS8 and Gprotein expression levels as both the functional effects (Fig. 4) and FRET ratio (Fig. 10A) are dependent on concentration.
That the level of expression should influence signaling is important. There is evidence that RGS concentrations may be dynamically regulated by intracellular signaling pathways (51), in a use-dependent manner by persistent activation of a FIG. 11. Interaction between G␣ and RGS8 is independent of the functional state of the G-protein. A, RGS8-YFP accelerates both activation (i) and deactivation (ii) kinetics of GIRK currents activated at the ␣2A receptor (HKIR3.1/3.2/␣2A line) when constrained to signal via G o ␣ A -CFP in PTx-treated cells. NE, norepinephrine. B, FRET ratios calculated using three-cube methods from images obtained with a backilluminated CCD camera (see "Materials and Methods"). HKIR3.1/3.2/A1 cells were transfected with cDNA constructs as indicated, and n values indicate number of cells imaged. C shows images obtained with the back-illuminated CCD microscope. Top panels show images taken at the CFP, FRET, and YFP wavelengths. Below are images showing the three-cube net FRET signal obtained with the three-cube FRET module (Metamorph, Universal Imaging). Perfusion of an appropriate agonist does not change the membrane-localized FRET signal in the HKIR3.1/3.2/D2 and HKIR3.1/3.2/␣2A lines. Bar graphs summarize the FRET ratios measured under the conditions indicated. GPCR or in disease states, for example (52). How this relates to signaling in neurons and cardiac cells, in which these signaling components may be clustered in specialized microdomains at high concentration, is intriguing. Furthermore, whether all RGSs behave in such a fashion is an important topic for future investigation.
In our view, it is the ternary complex that matters and not just the receptor. In particular, we have examined the role of the G-protein ␣ subunit. At one level, RGS8 receptor selectivity does not reflect G-protein coupling profiles (23) as the pattern of receptor selectivity is unaffected when signaling is constrained through a single G-protein ␣ isoform (as shown in Fig.  3). However, the data are more subtle than this. In Fig. 11A, we make the observation that when the ␣2A receptor is constrained to signal through G o ␣ A -CFP, the activation kinetics are slower and now become modulated by RGS-YFP. In addition, the GABA1b/2 receptor activates current more slowly via G i ␣ 2 than G o ␣ A , and RGS8-YFP accelerates this, but the absolute rates of activation differ. These observations all support the idea that it is the conformation of the signaling complex that dictates the rate of activation, and this is uniquely dependent on the nature of all of the agonist, receptor, G-protein, and associated RGS. These ideas are coherent with much of the recent signaling literature (25,26). It is also interesting that for the ␣2A and A1 receptors, with which we see intrinsically fast activation, are those for which receptor-G-protein "precoupling" has been described (53)(54)(55)(56). The kinetic consequences of this should be to intrinsically accelerate the activation process. A further feature is that it opens the possibility that different RGSs may have differential effects on activation and deactivation kinetics. For example, it has been shown here (Fig. 1) and previously (57) that RGS7 is particularly effective at accelerating activation kinetics.
In summary, our data reveal that the RGS can enter into the ternary complex to form a quaternary complex, thus accounting for the behavior of accelerated activation kinetics. Our data contribute to an emerging picture in which channel, G-protein, GPCR, and now RGS may all be scaffolded into a macromolecular signaling complex (22-24, 58 -61). Furthermore, they illustrate the central role of the G␣ subunit in organizing the specificity and dynamics of signaling to the G-protein-gated inwardly rectifying K ϩ channel.