G Protein Selectivity Is a Determinant of RGS2 Function*

RGS (regulator of Gprotein signaling) proteins are GTPase-activating proteins that attenuate signaling by heterotrimeric G proteins. Whether the biological functions of RGS proteins are governed by quantitative differences in GTPase-activating protein activity toward various classes of Gα subunits and how G protein selectivity is achieved by differences in RGS protein structure are largely unknown. Here we provide evidence indicating that the function of RGS2 is determined in part by differences in potency toward Gq versus Gi family members. RGS2 was 5-fold more potent than RGS4 as an inhibitor of Gq-stimulated phosphoinositide hydrolysis in vivo. In contrast, RGS4 was 8-fold more potent than RGS2 as an inhibitor of Gi-mediated signaling. RGS2 mutants were identified that display increased potency toward Gi family members without affecting potency toward Gq. These mutations and the structure of RGS4-Giα1 complexes suggest that RGS2-Giα interaction is unfavorable in part because of the geometry of the switch I binding pocket of RGS2 and a potential interaction between the α8-α9 loop of RGS2 and αA of Gi class α subunits. The results suggest that the function of RGS2 relative to other RGS family members is governed in part by quantitative differences in activity toward different classes of Gα subunits.

RGS (regulator of G protein signaling) proteins are GTPase-activating proteins that attenuate signaling by heterotrimeric G proteins. Whether the biological functions of RGS proteins are governed by quantitative differences in GTPase-activating protein activity toward various classes of G␣ subunits and how G protein selectivity is achieved by differences in RGS protein structure are largely unknown. Here we provide evidence indicating that the function of RGS2 is determined in part by differences in potency toward G q versus G i family members. RGS2 was 5-fold more potent than RGS4 as an inhibitor of G q -stimulated phosphoinositide hydrolysis in vivo. In contrast, RGS4 was 8-fold more potent than RGS2 as an inhibitor of G i -mediated signaling. RGS2 mutants were identified that display increased potency toward G i family members without affecting potency toward G q . These mutations and the structure of RGS4-G i ␣ 1 complexes suggest that RGS2-G i ␣ interaction is unfavorable in part because of the geometry of the switch I binding pocket of RGS2 and a potential interaction between the ␣8-␣9 loop of RGS2 and ␣A of G i class ␣ subunits. The results suggest that the function of RGS2 relative to other RGS family members is governed in part by quantitative differences in activity toward different classes of G␣ subunits.
Many hormones, neurotransmitters, and sensory stimuli exert their effect on target tissues by activating receptors that are coupled to heterotrimeric G proteins 1 (1,2). Receptor activation results in exchange of GTP for GDP on G␣ subunits, dissociation of GTP-bound G␣ subunits from the G␤␥ heterodimers, and activation of downstream effector pathways. Signals are terminated following G␣-catalyzed hydrolysis of GTP and reformation of G protein heterotrimers. Thus, G proteins are molecular switches that coordinate physiological responses elicited by a variety of stimuli.
RGS (regulator of G protein signaling) proteins are a family of more than 20 members that regulate G protein signaling in part by acting as GTPase-activating proteins (GAPs) for several classes of G protein ␣ subunits (3)(4)(5)(6). The GAP activity of RGS proteins decreases the lifetime of active, GTP-bound G␣ subunits, thereby attenuating responses or accelerating the kinetics of signal termination (7,8). Binding of RGS proteins to active G␣ subunits can also antagonize effector activation, thereby blocking signal production (9). These activities are mediated by the conserved RGS domain of ϳ120 amino acids that is characteristic of this protein family.
Higher eukaryotes express several types of RGS proteins, potentially to provide selective regulation of distinct types of G protein signaling pathways. Consistent with this hypothesis, RGS proteins are structurally diverse, distinguished by various domains that are likely to confer specific functions. For example, the N terminus of RGS4 confers receptor-selective regulation of G q -coupled responses (10,11), the PDZ domain of RGS12 binds peptides from the C termini of certain G protein coupled receptors (12), and the GGL domain of RGS7 selectively binds G␤5 (13,14). Differences in expression pattern (15,16), subcellular localization (17)(18)(19), and interaction with other signaling or regulatory proteins are also likely to give RGS family members distinct biological functions.
It is less clear to what extent differences in G␣ subunit selectivity govern the biological function of various RGS proteins. Whereas a few RGS proteins, such as RGSZ1 (20) and the RGS domain of p115Rho-GEF (6), are highly specific for certain G␣ subunits, many RGS proteins are promiscuous in vitro. For example, RGS1, RGS2, RGS4, and RGS-GAIP stimulate the GTPase activity of G i ␣ family members and G q ␣ in vitro. Less selective members of the RGS family do display quantitative differences in GAP activity toward various classes of G␣ subunits in vitro (21,22), but the significance of this in vivo is unknown. Transfection studies have shown that RGS proteins can be more selective for certain types of signaling pathways in vivo than they are as GAPs in vitro (23)(24)(25). Whether the regulatory specificity of RGS proteins observed in intact cells is achieved by differences in G protein selectivity or other mechanisms remains unknown. Accordingly, the mechanisms that govern the intrinsic G␣ subunit selectivity of RGS proteins remain poorly understood.
To examine the importance of G␣ selectivity as a determinant of the biological functions of RGS proteins, we have focused on RGS2 and RGS4. RGS2 or RGS4 can act in vivo and in vitro on G q or G i class ␣ subunits, although RGS2 is much less potent than RGS4 as a GAP for G i ␣ subunits in vitro. Here we have examined whether RGS2 and RGS4 differ in potency as inhibitors of G q versus G i signaling in intact cells and used mutagenesis to determine whether differences in intrinsic G protein selectivity govern the function of RGS2. The results suggest that RGS2 preferentially regulates G q , an effect that is mediated by unique structural features of its G protein-binding interface. They further implicate G protein selectivity as an important determinant of RGS2 function.

EXPERIMENTAL PROCEDURES
Materials-Cytomegalovirus promoter-driven plasmids used to express RGS4 (pCR3) and RGS2 (pEGFP-C1) were from Invitrogen (Carlsbad, CA) and CLONTECH (Palo Alto, CA), respectively. cDNAs were fused to sequences encoding three tandem copies of a Myc epitope by polymerase chain reaction as described previously (26). The pVT102U-GFP and YCp50 expression plasmids have been described (27). Bacterial expression plasmids containing RGS2 and RGS4 cDNAs were generated in previous studies (26,28). Expression constructs for RGS2/RGS4 chimeras and point mutants were made using the Quick-Change mutagenesis kit (Stratagene, La Jolla, CA) and verified by DNA sequencing of the entire protein coding region. Unless otherwise stated, all other reagents and chemicals were from Sigma.
Transfections-HEK293 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% (v/v) fetal calf serum, 2 mM glutamine, 10 g/ml streptomycin, and 100 units/ml penicillin at 37°C in a humidified atmosphere with 5% CO 2 . To generate stably transfected cell lines, HEK293 cells (7 ϫ 10 6 cells in 10-cm plates) were transfected using LipofectAMINE (Life Technologies, Inc.) and plasmids (10 g) that direct expression of RGS2 or RGS4 tagged at their C termini with three copies of the c-Myc epitope. Stable clones were selected for growth in 0.5 mg/ml geneticin. Subclones were screened for RGS expression levels by Western blotting and for clonality by immunofluorescence using the mouse 9E10 monoclonal antibody. To determine the relative expression levels of RGS proteins in stably transfected cell lines, cells from trypsinized plates were counted, pelleted, and lysed (2 ϫ 10 7 cells/ml) in Laemmli sample buffer. Following sonication and boiling, samples were resolved by SDS-polyacrylamide gel electrophoresis. Protein expression levels were determined by immunoblotting using the 9E10 antibody, ECL (Amersham Pharmacia Biotech), and densitometry.
Phosphoinositide Hydrolysis Assays-Inositol trisphosphate production was measured essentially as described by Venkatakrishnan and Exton (29). Briefly, 0.3-0.5 ϫ 10 6 cells/well in a 12-well dish were labeled in complete Dulbecco's modified Eagle's medium (without inositol) containing 4 Ci/ml [ 3 H]inositol (NEN Life Science Products) for 17-24 h. In cases where pertussis toxin (List Biologicals, Campbell, CA) was used, it was added (100 ng/ml) 2-4 h prior to completion of the labeling period. Cells were washed once with Dulbecco's modified Eagle's medium (without bicarbonate) containing 25 mM Hepes, pH 8, and incubated for 10 min in the same medium containing 5 mM LiCl. Following agonist or vehicle addition to each well, reactions were allowed to continue for 30 -60 min and then stopped with 750 l of ice-cold formic acid. Samples were incubated for 30 min at 4°C. The contents of an entire well were collected by centrifugation at 15,000 ϫ g for 15 min. The supernatant fraction (700 l) was mixed with 214 l of 0.7 M NH 4 OH, and soluble inositol-containing material was separated by ion exchange chromatography as described previously (29). Portions of the total inositol-containing material and IP 3 fraction from each sample were added to 10 ml of Ready-Safe scintillant (Beckman, Fullerton, CA) and analyzed by liquid scintillation spectrometry. IP 3 levels were expressed as the fraction of the total soluble inositol-labeled material.
Purification of Recombinant Proteins-N-terminally histidinetagged forms of RGS2, RGS4, and G o ␣ were expressed in Escherichia coli (BL21(DE3)) and purified by Ni 2ϩ -nitrilotriacetic acid chromatography as described previously (26,28). The sources of G proteins used for GAP assays have been described previously (4). Recombinant Histagged G q ␣ was prepared and purified as described (30). Baculoviruses encoding untagged G q ␣, G␤, and His-tagged G␥ subunits and methods used to express and purify untagged G q ␣ from Sf9 cells have been described previously (31). Purified phospholipase C␤1 was a generous gift of Drs. R. Ball and P. Sternweis (University of Texas Southwestern Medical Center, Dallas, TX).
GTPase Assays, G Protein Binding, and G q -mediated PLC␤1 Activation-The rate of GTP hydrolysis by G o ␣ during a single catalytic turnover was measured as described previously (4). RGS protein binding to purified G q ␣ and G o ␣ and regulation of G q ␣-mediated activation of reconstituted PLC␤1 were determined as described previously (21).
Assays of Yeast Pheromone Response-RGS2 and RGS4 cDNAs were expressed in yeast from the constitutive ADH promoter. Various portions of the coding sequences of RGS2 and RGS4 were generated by polymerase chain reaction as BamHI/XbaI fragments and inserted in frame upstream of the GFP or GFP-Ras2 (9 C-terminal amino acids of Ras2p; CAAX box) coding sequences in the high copy plasmid pVT102U (27). To generate single-copy versions of selected constructs, an SphI fragment containing the ADH promoter, RGS coding sequences, and ADH transcriptional terminator was subcloned into YCp50. The yeast strain used was BC180 (MATa leu2-3, 112 ura3-52, his3⌬1, ade2-1, sst2⌬2). Yeast cells were grown at 30°C in synthetic dextrose medium lacking uracil (SD-Ura). Pheromone responsiveness was determined by performing halo assays as described previously (27). RGS protein expression from various GFP fusion constructs was assessed by Western blotting and quantified by fluorescence activated cell sorting of yeast cells (2 ϫ 10 4 ) in a FACScan flow cytometer (Becton Dickinson).

RGS2 Is More Potent than RGS4 as an Inhibitor of G q ␣
Signaling in Vivo-To compare the inhibitory potencies of RGS2 and RGS4, we first established an assay system by identifying receptors expressed endogenously by HEK293 cells that stimulate phosphoinositide hydrolysis via G q (Fig. 1A). By stimulating cells with various agonists, we could show that HEK293 cells respond to carbachol, resulting a 4-fold increase in IP 3 production relative to untreated controls. This response was insensitive to pertussis toxin, indicating that PLC␤ activation probably occurs through G q ␣-coupled m1 or m3 muscarinic acetylcholine receptors. In contrast, lysophosphatidic acid, somatostatin (Fig. 1A), bradykinin, and angiotensin II (data not shown) did not elicit detectable stimulation of IP 3 production.
To examine the relative potencies of RGS2 and RGS4 as inhibitors of G q signaling, we appended three copies of the c-Myc epitope to the C termini of RGS2 and RGS4 and generated HEK293 cell lines that stably express either of these tagged proteins. Tagging was necessary to directly compare expression levels of RGS2 and RGS4; previous studies have shown that C-terminally tagged and untagged RGS4 have equivalent activity (26). Two stable lines were obtained that express equivalent levels of RGS2-Myc (RGS2-1 and RGS2-2), and four were obtained that express various levels of RGS4-Myc (RGS4-1 through RGS4-4). Initially we analyzed the RGS2-1 and RGS4-1 lines by comparing the levels of IP 3 produced in response to increasing concentrations of carbachol. At each concentration of carbachol, IP 3 accumulation was inhibited to a similar extent in RGS2-1 and RGS4-1 cells relative to controls (Fig. 1B). However, immunoblotting showed that RGS4-Myc expression was approximately 5-fold higher than RGS2-Myc in these cell lines (Fig. 1B, inset), indicating that RGS2 may be more potent than RGS4 as an inhibitor of G q signaling in these cells. The observed low expression levels of RGS2-Myc are likely due to inefficient translation initiation because inclusion of an optimal Kozak consensus initiation sequence (32) significantly increased RGS2-Myc expression. 2 Nevertheless, neither RGS2-Myc nor RGS4-Myc was expressed at high levels in these stably transfected lines.
As a further means of determining whether RGS2 is more potent than RGS4 as an inhibitor of G q ␣ signaling, we compared carbachol-induced IP 3 production in the RGS2-1 line with that in cell lines expressing various levels of RGS4-Myc (Fig. 1C). The results indicated that the extent of inhibition of IP 3 production correlated directly with RGS4-Myc expression levels. More importantly, IP 3 production was inhibited strongly only when RGS4-Myc expression was severalfold higher than that of RGS2-Myc (Fig. 1, B and C, compare RGS2-1 with RGS4-1 and RGS4-2). Furthermore, expression of RGS4-Myc at a level similar to that of RGS2-Myc had little inhibitory effect (Fig. 1, B and C, compare RGS2-1 with RGS4-2 and RGS4-3). The results of these experiments suggested that RGS2 is approximately 5-fold more potent than RGS4 as an inhibitor of G q ␣-mediated signaling in HEK293 cells. This difference in inhibitory potency in vivo is likely to be mediated at the G protein and/or effector level, because recombinant RGS2 is 5-10-fold more potent than RGS4 as an inhibitor of G q ␣ stimulation of purified PLC␤1 in vitro (21). Differences in potency of RGS2 and RGS4 could also be due in part to distinct patterns of subcellular localization. Accordingly, subsequent experiments were aimed at determining the role of G protein selectivity on the function of RGS2 and RGS4.
RGS4 Is More Potent than RGS2 as an Inhibitor of G i Class ␣ Subunits-Several observations prompted us to investigate whether RGS2 and RGS4 differ in their potencies as inhibitors of G i class ␣ subunits. First, although recombinant RGS2 appears to be 1000-fold less potent than RGS4 as a GAP for G i ␣ subunits (21), either RGS protein is capable of stimulating GTP hydrolysis when reconstituted with receptors and G i in lipid vesicles (5). Second, we and others have shown that RGS4 is much more potent than RGS2 as an inhibitor of the yeast mating pheromone response pathway (16), which uses a G i class ␣ subunit. However, in these studies it had not been determined whether the relative potencies of RGS2 and RGS4 are due to differences in G protein selectivity, expression level, subcellular localization, or other factors. Answering this question was the objective of the following experiments.
To explore the mechanisms that account for the different potencies of RGS2 and RGS4 as inhibitors of G i class G proteins, we tagged RGS2 and RGS4 at their C termini with GFP as a means of examining their expression level and subcellular localization when expressed in yeast. We also generated versions of the RGS-GFP fusions that contain a C-terminal prenylation sequence (CAAX) from the yeast Ras2 protein to target the chimeric proteins efficiently and specifically to the plasma membrane, as we have done previously with RGS4 (27).
By expressing each of these constructs in yeast cells lacking Sst2, the RGS homolog that normally attenuates mating pheromone signaling, we could use sensitive and quantitative assays of pheromone-induced growth arrest (halo assays) to determine the relative inhibitory potencies of RGS2 and RGS4.
Several results suggested that the relative inability of RGS2 to inhibit mating pheromone signaling is not due to a defect in plasma membrane localization (Fig. 2). As indicated by confocal fluorescence microscopy, wild type RGS2-GFP was plasma membrane-bound; nevertheless, RGS2-GFP only weakly (3fold) inhibited mating pheromone response. As expected, however, plasma membrane localization of RGS2-GFP was required for function. Removing the N-terminal non-GAP domain of RGS2 resulted in its cytoplasmic localization and loss of inhibitory activity; activity and plasma membrane localization were restored by appending a CAAX box to truncated RGS2-GFP. In contrast, RGS4-GFP-CAAX was plasma membranebound and inhibited pheromone response 200-fold, consistent with previous demonstrations that RGS4 is much more active in yeast than RGS2.
Differences in protein expression levels only partly accounted for the low activity of RGS2 relative to RGS4. The level of RGS2-GFP was severalfold lower than that of RGS4-GFP expressed from identical high copy plasmids, as indicated by immunoblotting and quantitation of GFP fluorescence. The contribution of this effect was eliminated by normalizing inhibitory activity for differences in protein levels (Table I). When corrected for differences in protein expression, RGS4 still appeared to be 8-fold more potent than RGS2 as an inhibitor of yeast mating pheromone signaling. A similar conclusion was reached by expressing RGS4-GFP from a single copy plasmid, which reduced its expression below that of RGS2-GFP expressed from a high copy plasmid (data not shown). Use of single versus high copy plasmids also indicated that the strengths of inhibitory phenotypes were proportional to RGS expression level, suggesting that differences in the effects of RGS2 versus RGS4 are unlikely to be due to misfolding of overexpressed proteins.  (left-hand panels). All images are shown at the same magnification (bar ϭ 5 m). The ability of the indicated RGS-GFP fusions to inhibit yeast mating pheromone response (right-hand panels) was measured using growth arrest (halo) assays that employed a mutant (BC180) lacking Sst2, the RGS protein that normally regulates mating pheromone response. Shown are the responses elicited by 0.15 nmol of pheromone; other doses of pheromone were also used to quantify the extent of inhibition caused by expression of the RGS-GFP fusions relative to controls expressing GFP. The data shown are representative of two transformants assayed in three independent experiments.

Features of RGS2 That Limit Its Ability to Act on G i Class G
Proteins-Based on the preceding results, it appeared that RGS2 and RGS4 differ considerably in their relative activities toward G q and G i family members. To test this hypothesis further and to investigate the mechanisms responsible for achieving different G protein selectivities, we sought to define structural features of RGS2 that may attenuate its ability to act on G i class ␣ subunits. One possibility was that differences in G␣ selectivity could be determined by the structures of the G protein-binding surfaces of RGS2 and RGS4. To test this idea, we used sequence alignments and structural data for the G protein-binding surface of RGS4 (33) to identify features of RGS2 that may account for its G protein selectivity. We assumed that the structures of the RGS domains of RGS2 and RGS4 are similar. This seemed likely because the hydrophobic residues that appear to stabilize the RGS fold are conserved even in more highly diverged members of the RGS family such as GRK3 and p115Rho-GEF.
Features of RGS2 that may govern its G protein selectivity were identified by analyzing the properties of RGS2/4 chimeras and point mutants (Fig. 3); none of the chimeras or point mutants were insoluble, indicating that they folded relatively normally. Three regions of the RGS core domain of RGS4 (overlined) contact G i ␣ 1 . With the exception of a highly conserved asparagine residue, the middle region lacked a strong consensus among the RGS proteins studied and, therefore, seemed unlikely to account for RGS2-specific G␣ selectivity. In contrast, the N-and C-terminal subregions (labeled A and B, respectively, in Fig. 3) were highly conserved among other RGS proteins but contained a significant number of RGS2-specific sequences. These regions were studied further.
To determine whether regions A and B of the RGS core domain influence G␣ selectivity, we analyzed a series of chimeric proteins in which these regions of RGS2 and RGS4 were transplanted. Region A of RGS2 (residues 99 -109) was replaced with the equivalent region of RGS4 to produce the RGS2(4A) chimera. Similarly, the RGS2(4B) chimera was made by replacing domain B (residues 181-191) of RGS2 with the corresponding region of RGS4. A third chimera, RGS2-(4AB), was generated replacing both the A and B regions of RGS2 with those of RGS4. Each protein was expressed as a His-tagged molecule in E. coli, purified, and analyzed for GAP activity using G o ␣ as a substrate (Fig. 4A). Introducing either the A or B domain of RGS4 into RGS2 did not result in detectable GAP activity. However, replacing both the A and B domains of RGS2 with those of RGS4 was sufficient to obtain GAP activity. Titration experiments indicated that the activity of RGS2(4AB) is at least 100-fold greater than wild type RGS2 and 5-10-fold less potent than wild type RGS4 (Fig. 4B).
As a means of defining the specific sequence features of regions A and B of RGS4 that confer G o ␣ GAP activity when introduced into RGS2, we constructed and analyzed a series of point mutants. Region A of RGS2 and RGS4 differ at only two sites (Phe 105 and Cys 106 of RGS2 correspond, respectively, to Tyr 84 and Ser 85 of RGS4). Accordingly, Phe 105 and Cys 106 in region A were changed singly or together to their RGS4 equivalents in the context of the RGS2 chimera containing only the B region of RGS4. This yielded RGS2(4B)F105Y, RGS2(4B)-C106S, and RGS2(4B)F105Y,C106S. A similar approach was used to identify which of three amino acid differences in region B affect the G protein selectivity of RGS2. One of these, Asn 183 , seemed less likely to be involved because changing its equivalent in RGS4 (Lys 162 ) to alanine does not affect function (26). Accordingly, we determined whether the remaining two residues (Asn 184 and Glu 191 ) in region B of RGS2 affect G protein selectivity. These residues in region B were changed singly or in combination to their RGS4 equivalents in the context of the RGS2 chimera containing the A region of RGS4. This yielded RGS2(4A)N184D, RGS2(4A)E191K, and RGS2(4A)N184D, E191K.
Results shown in Fig. 5A indicated that one residue in region A and two in region B are likely to influence the G protein selectivity of RGS2. In region A, the C106S substitution increased the G o ␣ GAP activity of the RGS2(4B) chimera,  FIG. 3. Potential G protein selectivity determinants of the RGS2 core domain. The RGS core domain sequences of RGS1, RGS2, RGS4, and RGS-GAIP were aligned by the CLUSTAL method. Amino acid residues that contribute to the G␣ interaction surface of RGS4 (33) are indicated (asterisks). Highly conserved residues in the RGS core domain are boxed. The putative G␣ interaction surface of RGS2 is indicated (overlines); amino acid residues within two of these regions (A and B) that are unique to RGS2 and proved to be functionally important are highlighted (red). whereas the F105Y substitution had little effect. In region B, the N184D or E191K substitution increased G o ␣ GAP activity of the RGS2(4A) chimera to intermediate levels, whereas the two together yielded activity similar to the RGS2(4AB) chimera.
To determine whether the three residues implicated by the preceding results are primary determinants of the G protein selectivity of RGS2, we introduced the C106S, N184D, and E191K substitutions into otherwise wild type RGS2, yielding the RGS2-triple mutant. Furthermore, we introduced the reciprocal substitutions into RGS4 (S85C,D163N,K170E ϭ RGS4-triple), reasoning that this might attenuate its GAP activity toward G o ␣. The GAP activities of recombinant the RGS2-triple and RGS4-triple mutants were compared with their wild type counterparts in titration experiments (Fig. 5B). As expected, the RGS2-triple mutant functioned as a GAP for G o ␣; like the RGS2(4AB) chimera it was 5-10-fold less potent than wild type RGS4. By contrast, even at a high concentration (1 M) the RGS4-triple mutant was ineffective as a GAP for G o ␣, further suggesting that the sequence features we have defined are involved in governing G protein selectivity.
To determine whether the RGS-triple mutants have altered potencies toward G i class G proteins in intact cells, we examined their ability to inhibit yeast mating pheromone response. This approach was chosen for several reasons. First, the sensitivity and reproducibility of pheromone response assays allows detection of quantitative differences in RGS potency. Second, inhibitory potency in yeast is determined mainly by G protein selectivity because the activities of wild type RGS proteins and their prenylated RGS core domain counterparts are similar. Third, our previous studies indicate that RGS mutants have similar activities toward yeast and mammalian G i class ␣ subunits (26).
Accordingly, we removed the N-terminal membrane targeting domains of wild type and triple mutant versions of RGS2 and RGS4 and appended GFP containing a CAAX motif to their C termini to quantify protein expression and obtain efficient plasma membrane targeting. As indicated by the results of pheromone response assays, the RGS2-triple mutant was 17-fold more active than the equivalent wild type protein, even though protein expression levels and membrane localization efficiency were similar (Table I and data not shown). Conversely, the RGS4-triple mutant was 4-fold less potent than the equivalent wild type protein, without significant differences in protein expression or membrane localization. These results are probably a more accurate indicator of the effects of these mutations on the activity of RGS2 and RGS4, because single turnover GTPase assays underestimate the activity of RGS2 toward G i class ␣ subunits. This difference between in vitro and in vivo activity toward G i ␣ family members may indicate that the membrane, G␤␥, or the receptor modulates RGS2 activity. Indeed, this is also suggested by the recent demonstration that RGS2 can mediate the receptor-stimulated steady-state GTPase activity of G i in a reconstituted system (5).
Although the preceding results suggest that G protein selectivity of RGS2 and RGS4 is affected by the point mutations, they could instead suggest that potency toward any G protein is affected. If so, the RGS2-triple mutant should be more potent than wild type RGS2 as an inhibitor of G q ␣, and the RGS4triple mutant should be less potent than wild type RGS4 as an inhibitor of G q ␣. Contrary to these expectations, titration experiments indicated that the point mutations did not dramatically affect the potencies of RGS2 or RGS4 as inhibitors of G q ␣-mediated activation of purified PLC␤1 (Fig. 6). Therefore, we suggest that these residues in RGS2 are not general inhibitors of G␣ interaction but rather act as determinants of G protein selectivity.

DISCUSSION
In this study we have provided three lines of evidence indicating that RGS2 inhibits signaling by G q in a more potent manner than it inhibits G i class G proteins. RGS2 is 5-fold more potent than RGS4 as an inhibitor of G q function in vivo and in vitro. In contrast, RGS2 is 8-fold less potent than RGS4 as an inhibitor of yeast mating pheromone signaling, which is mediated by a G i class ␣ subunit. Finally, RGS2 mutants have been identified that are more potent toward G i class ␣ subunits without affecting potency toward G q . Thus, RGS2 and RGS4 differ considerably in terms of their relative intrinsic potencies as inhibitors of G q versus G i class G proteins.
Mechanism of G Protein Discrimination by RGS2-A mech- anism whereby RGS2 discriminates between classes of G␣ subunits is suggested by our studies of RGS2 mutants and by the crystal structure of the RGS4-G i ␣ 1 transition state complex (33). This information indicates why RGS2 interacts inefficiently with G i class ␣ subunits. It also suggests why RGS2 retains affinity for G q ␣, but this aspect of the model remains to be tested experimentally. Two features of RGS2 are proposed to attenuate interaction with G i class ␣ subunits while maintaining affinity for G q ␣ (Fig. 7). First, the geometry of the pocket of RGS2 that binds a conserved threonine residue in switch I of G␣ subunits appears to decrease affinity for G i class ␣ subunits. We suggest this because increasing the potency of RGS2 toward G i class ␣ subunits requires substituting residues predicted to form part of the floor and lip of the pocket (Cys 106 and Asn 184 , respectively) with their RGS4 counterparts (Ser 85 and Asp 163 ; Fig. 7). Indeed, the pocket of RGS2 is predicted to be smaller than that of RGS4, based on modeling studies using RGS4 as a template. 2 For example, the closest approach between the side chains of Cys 106 and Asn 184 in the pocket of RGS2 is predicted to be about 1 Å shorter than that of the corresponding residues of RGS4.
We suggest two ways that the pocket of RGS2 could favor binding of G q over G i class ␣ subunits. In the first model, the observed selectivity is due somewhat to different conformations of switch I in G q and G i class ␣ subunits, as suggested by sequence differences (Lys 180 and Val 185 of G i ␣ 1 are proline and isoleucine, respectively, in G q ␣; Fig. 7). This allows the invariant threonine residue of switch I of G q ␣ to dock with the shallower or smaller pocket of RGS2 with higher affinity than the homologous threonine residue of G i ␣. RGS4 binds G q and G i class ␣ subunits because its threonine binding pocket is bigger and docks either substrate relatively well. In the second model, switch I regions of G i and G q class ␣ subunits are assumed to have approximately the same conformation. Sequence differences in and around the threonine binding pockets of RGS2 and RGS4 affect the structure of adjacent regions of their G␣ binding surfaces. These adjacent regions in turn dictate selectivity for G␣ subunits. A combination of these models could also occur.
The second structural feature of RGS2 that appears to attenuate activity toward G i class ␣ subunits is at the edge of the interaction footprint defined by the RGS4-G i ␣ 1 crystal structure. This interfering interaction is proposed to involve the putative ␣8-␣9 loop of RGS2 and ␣A of G i class ␣ subunits, because we have found that a E191K substitution in the ␣8-␣9 loop of RGS2 is required to increase activity toward G i class subunits. One interpretation is that charge repulsion occurs between Glu 191 of RGS2 and a conserved glutamic acid residue (Glu 65 ) in ␣A of G i ␣ 1 . However, the closest approach between the equivalent residue in the ␣8-␣9 loop of RGS4 (Lys 170 ) and Glu 65 of ␣A in G i ␣ 1 is 6.5 Å, a distance over which charge repulsion may be negligible. Nevertheless, this distance could be shorter earlier in the reaction mechanism, disfavoring a high affinity interaction. Alternatively, the structure of RGS2 may differ from that of RGS4, placing the ␣8-␣9 loop closer to ␣A of the G␣ subunit. G q ␣ may bind tightly to wild type RGS2 because the equivalent of E65 in ␣A (D71) is positioned farther away from Glu 191 in the ␣8-␣9 loop, either because of its shorter side chain or conformational differences of G q -relative to G i class ␣ subunits. Consistent with these hypotheses, recent studies suggest that sequences in the helical domain (which includes ␣A) of transducin ␣ subunits mediate preferential interaction with the RGS9 core domain (34). Indeed, it would be interesting to determine whether such interactions generally affect the selectivity of RGS-G␣ interaction.
Two findings suggest that the relative inability of RGS2 to interact with G i class ␣ subunits could also involve other sequence or structural features. None of the RGS2 mutants we have analyzed is as active toward yeast or mammalian G i class ␣ subunits as RGS4. Furthermore, the RGS4-triple mutant, which is identical to RGS2 at the three residues proposed to attenuate G i interaction, is 2-fold more potent than RGS2 as an inhibitor of yeast mating pheromone response. Potentially, sequences flanking the RGS domain could influence the G␣ selectivity of RGS2.
G protein selection by other RGS family members may use somewhat different mechanisms. We suggest this because the RGS domain of p115Rho-GEF, which is specific for G12/13, is highly diverged from other RGS family members (6), suggesting that its G␣ interaction surface is quite different. Furthermore, RGSZ1 (which acts on G z ␣ but not other G i family members or G q ) is identical to RGS4 (which acts on all G i family members including G z ␣) at the two of the three positions we have identified as G protein selectivity determinants in RGS2 (20,35). The selectivity of RGSZ1 may be mediated by other features of its RGS domain or by flanking sequences.
Biological Implications of the G Protein Selectivity of RGS2-G protein selectivity is likely to be one of several factors that govern the function of RGS2. G q selectivity may be important because RGS2 does not yet appear to have preference for certain types of G protein coupled receptors (10). This is in contrast to RGS1, RGS4, and RGS16, which display pronounced selectivity toward different types of G q -coupled receptors (10). Expression level is also likely to determine which types of G protein signaling pathways are regulated by RGS2. Lower level expression of RGS2 may be sufficient to attenuate G q signaling without affecting G i -mediated responses, whereas higher level expression of RGS2 could attenuate both types of signaling pathway. Thus, regulated expression of RGS2, as occurs in response to rises in cAMP or changes in neuronal activity (5,10,36), may provide a mechanism whereby various types of stimuli regulate G q and G i signaling pathways separately or in a coordinated fashion, depending on the expression level of RGS2 that is achieved. Other processes such as G␣ modification (e.g. palmitoylation or phosphorylation) or associ- FIG. 7. Proposed mechanism of G protein selection by RGS2. RGS2 is proposed to interact inefficiently with G i class ␣ subunits because of the geometry of the pocket that binds a conserved threonine residue in switch I of G␣ subunits and to charge repulsion with a glutamic acid residue in ␣A of G i ␣ subunits. The structure of RGS4 is shown (spacefill), highlighting its G␣ interaction surface (amino acids within 5.0 Å of G i ␣ 1 ; cyan) and the equivalents of three amino acids of RGS2 (yellow) that appear to prevent efficient interaction with G i class ␣ subunits. Switch I and part of ␣A of G i ␣ 1 are shown (sticks). Amino acids of RGS4 and G i ␣ 1 are designated by the prefixes r and a, respectively, followed by an arrow and the corresponding amino acid in RGS2 or G q ␣. The single-letter amino acid code is used. See text for details. ation with other signaling or regulatory molecules could further augment the ability of RGS2 to effect various regulatory outcomes.
In conclusion, it appears that quantitative differences in GAP activity toward G i and G q class ␣ subunits is one factor that governs the biological function of RGS2 and presumably other RGS isoforms as well. Expression of mutants having altered G protein selectivity should provide a means of determining to what extent intrinsic selectivity for various classes of G proteins, receptors, or effectors governs the function of RGS2 mammalian cells.