Mechanism of Gαi-mediated Inhibition of Type V Adenylyl Cyclase*

The topology of mammalian adenylyl cyclase reveals an integral membrane protein composed of an alternating series of membrane and cytoplasmic domains (C1 and C2). The stimulatory G protein, Gαs, binds within a cleft in the C2 domain of adenylyl cyclase while Gαi binds within the opposite cleft in the C1domain. The mechanism of these two regulators also appears to be in opposition. Activation of adenylyl cyclase by Gαs or forskolin results in a 100-fold increase in the apparent affinity of the two domains for one another. We show herein that Gαireduces C1/C2 domain interaction and thus formation of the adenylyl cyclase catalytic site. Mutants that increase the affinity of C1 for C2 decrease the ability of Gαi to inhibit the enzyme. In addition, Gαi can influence binding of molecules to the catalytic site, which resides at the C1/C2 interface. Adenylyl cyclase can bind substrate analogs in the presence of Gαi but cannot simultaneously bind Gαi and transition state analogs such as 2′d3′-AMP. Gαi also cannot inhibit the membrane-bound enzyme in the presence of manganese, which increases the affinity of adenylyl cyclase for ATP and substrate analogs. Thus homologous G protein α-subunits promote bidirectional regulation at the domain interface of the pseudosymmetrical adenylyl cyclase enzyme.

The classic mammalian adenylyl cyclases (AC) 1 consist of two repeats of a unit that includes six transmembrane spans and a cytoplasmic domain. Nine isoforms of adenylyl cyclase have been cloned that share this common topology. However, each of these enzymes display distinct patterns of regulation (reviewed in Refs. 1 and 2). For example, types V and VI adenylyl cyclase can be activated by GTP␥S-G␣ s (3), forskolin (4), and protein kinase C (␣ and , Ref. 5) and inhibited by G␣ i (6,7), calcium (8,9), cAMP-dependent protein kinase (10,11), and protein kinase C ␦ (type VI only, Ref. 12). Regulation by heterotrimeric G proteins has been the hallmark of adenylyl cyclase activity, but the details of this regulation have remained unclear until recently. In vitro, G␣ i can inhibit both G␣ s and forskolin-stimulated adenylyl cyclase activities in a non-competitive manner (7,13). But much of our recent knowledge is based upon experiments utilizing the cytoplasmic (or soluble) domains of adenylyl cyclase.
The two cytoplasmic domains of adenylyl cyclase (C 1 and C 2 ) create a pseudosymmetrical heterodimer that forms the catalytic moiety of the enzyme and is the target for most known intracellular regulators (reviewed in Ref. 14). The cytoplasmic domains contain a 200 -250 amino acid region that is ϳ50% similar to each other and 50 -90% similar to corresponding regions of other adenylyl cyclase isoforms. The C 1 and C 2 domains can be independently expressed as soluble proteins in Escherichia coli and mixed to reconstitute full adenylyl cyclase activity, including activation by G␣ s and forskolin and inhibition by P-site inhibitors and G␣ i (15)(16)(17)(18)(19). These soluble adenylyl cyclase fragments have proven invaluable in determining the stoichiometry of G␣ s , forskolin, and ATP binding. They have also proven invaluable in localizing binding sites for G␣ s and G␣ i and in discerning the catalytic mechanism of P-site inhibition of adenylyl cyclase. Much of this early work utilized C 1 and C 2 proteins from dissimilar isoforms of adenylyl cyclase (C 1 from type I and the C 2 domain from type II adenylyl cyclase), although more recent systems contain both domains from a single adenylyl cyclase isoform (19 -22).
Crystal structures of complexes containing the C 1 and C 2 domains, G␣ s , and forskolin reveal that forskolin and ATP analogs bind at the interface of the C 1 and C 2 domains (23-25), while the major binding site for G␣ s is located on the C 2 domain in the cleft formed by the ␣2Ј and ␣3Ј helices (17,23,26). A similar groove is formed by the corresponding structural elements of C 1 (23). Mutagenesis and binding studies indicate that G␣ i binds to this corresponding site in the C 1 domain (19); however, the mechanism for G␣ i -mediated inhibition is still unclear. Activation of adenylyl cyclase by G␣ s or forskolin results in a 100-fold increase in the apparent affinity of the two domains for one another (15,16). We show herein that G␣ i works in opposition to G␣ s to reduce domain interaction and thus the formation of the adenylyl cyclase catalytic site.
G Protein Subunits-All G protein ␣-subunits were synthesized in E. coli as described by Lee et al. (27). G␣ i was co-expressed with yeast protein N-myristoyltransferase (27,28) for synthesis of myristoylated protein. Purified ␣-subunits were activated by incubation with 50 mM NaHEPES (pH 8.0), 5 mM MgSO 4 , 1 mM EDTA, 2 mM dithiothreitol, and 400 M [ 35 S]GTP␥S at 30°C for 30 min for G␣ s or 2 h for G␣ i . Free GTP␥S was removed by gel filtration. All G proteins were activated with GTP␥S unless stated otherwise.
Expression and Purification of Adenylyl Cyclase in E. coli-The C 2 domains of type II and type V adenylyl cyclase (IIC 2 , VC 2 ) and the C 1 and C 1a domains of type V adenylyl cyclase were expressed in E. coli and purified by metal affinity chromatography followed by ion exchange as described previously (15,17,19).
Adenylyl Cyclase Assays-Adenylyl cyclase activity was measured as described (29). All assays were performed for 7-10 min at 30°C in a final volume of 50 -100 l with 5 mM MgCl 2 and 100 M ATP for the membrane-bound enzyme. In assays containing C 1 and C 2 domain proteins (reconstitution assays), limiting concentrations of the C 1 domain protein were first incubated with G␣ i for 15 min on ice followed by addition of G␣ s and VC 2 prior to initiation of the assay. The final concentration of C 2 was at least 0.5 M to promote interaction between the C 1 and C 2 proteins except where indicated otherwise. The final concentration of ATP was 1 mM for reconstitution assays, unless stated otherwise. All determinations were performed in duplicate and are representative of at least two experiments. Filter membranes were dried in a vacuum desiccator, and the bound ligand was quantified by liquid scintillation counting.

RESULTS
We have previously shown that G␣ i binds within the cleft formed by the ␣2 and ␣3 helices of the C 1a domain of type V adenylyl cyclase. Using [ 35 S]GTP␥S-G␣ i a clear shift in molecular weight of G␣ i is observed upon addition of the C 1a domain as determined by gel filtration. However, the tight binding of the C 1a domain and G␣ i is partially disrupted by formation of a C 1 -C 2 complex in the presence of forskolin (Fig. 1A). Further- more, the complex of C 1a and G␣ i is completely abolished in the presence of both G␣ s and forskolin. G␣ s does not compete for binding of G␣ i to the C 1 domain; however, full activation of adenylyl cyclase by G␣ s and forskolin does limit the extent of inhibition by G␣ i (7,19). Therefore we examined the formation of a potential G␣ i -C 1 -C 2 -G␣ s complex in the absence of forskolin ( Fig. 1B) using the full-length C 1 domain, which has a 10-fold higher affinity for G␣ i (19). Even in the presence of high concentrations of C 1 , C 2 , G␣ s , and G␣ i , complexes between C 1 -G␣ i and C 2 -G␣ s are formed, but no heterotetramer is ever observed. Note that myristoylated G␣ i and complexes containing the myristoylated protein tend to be slightly retarded on gel filtration columns most likely because of the hydrophobic nature of the myristate group.
Both forskolin and G␣ s have been shown to increase the affinity of C 1 and C 2 by 100-fold more than basal, and the combined action of the activators increases the apparent affinity of C 1 and C 2 by more than 1000-fold. Our gel filtration results suggest that G␣ i may work in opposition to these regulators to decrease the affinity between C 1 and C 2 . We therefore examined the ability of G␣ i to inhibit adenylyl cyclase with increasing concentrations of C 2 protein to drive the interaction between the two domains ( Fig. 2A). At low concentrations of C 2 , G␣ i greatly inhibits adenylyl cyclase activity. However, as the concentration of C 2 increases, the interaction between C 1 and C 2 increases, reducing the ability of G␣ i to inhibit the enzyme. At maximal concentrations of C 2 protein, no inhibition by G␣ i is observed. Therefore, the interaction between the C 1 and C 2 domains not only decreases the binding of G␣ i to C 1 but also decreases the ability of G␣ i to inhibit the enzyme. A similar phenomenon is observed with a mutant of C 1 (E418A), which has a 6-fold higher affinity for G␣ i (19). Once again, maximal inhibition occurs at low C 2 concentrations, and the effect of G␣ i decreases with increasing C 2 (Fig. 2B). In this case, however, the effect of G␣ i is not eliminated at maximal concentrations of C 2 . This mutation is within the G␣ i binding cleft and has no effect on the apparent affinity of the C 1 and C 2 domains (Fig.  2C) and does not increase dimerization of C 1 as measured by gel filtration (data not shown). The inability to eliminate the effect of G␣ i at high C 2 concentrations is most likely caused by the increased efficacy of G␣ i -mediated inhibition of both the membrane-bound and cytoplasmic domains of adenylyl cyclase when this mutation is present (19). These data suggest that the mutation of Glu-418 to alanine induces a conformational change that facilitates G␣ i inhibition independent of changes at the C 1 -C 2 interface.
We have also examined the interaction between the C 1 and C 2 domains utilizing a mutant within the C 2 domain (K1014N in type II AC) with an increased affinity between the C 1a and C 2 domains in the presence and absence of G␣ s (30). When paired with the full-length C 1 domain from type V, the C 2 -K1014N mutant shows a similar 10-fold shift in affinity for C 1 (Fig. 3A). This point mutant is located 16 -28 Å from G␣ iinteracting residues and on the opposite domain from which G␣ i binds. Thus it is not predicted to directly interfere with G␣ i binding. However, the inhibition of C 1 and the mutant C 2 domain by G␣ i is remarkably diminished as compared with wild type (Fig. 3B). Therefore, once again as the affinity of C 1 and C 2 is increased, the ability of G␣ i to inhibit adenylyl cyclase is decreased.
The G␣ i binding cleft is located in close proximity to the catalytic site, particularly to residues contacting the magnesium ions at the active site and the phosphate moieties of substrate or P-site molecules. Therefore, G␣ i may affect binding of molecules to the active site. To address this possibility, kinetic analysis of inhibition by G␣ i and P-site inhibitors or substrate analogs was performed. P-site inhibitors bind at the active site and are postulated to mimic a product-like transition state. Classic P-site inhibitors (such as 2Јd3Ј-AMP) require the product pyrophosphate for binding to the enzyme and show uncompetitive inhibition with respect to MgATP upon activation with G␣ s (18). The P-site inhibitor 2Јd3Ј-AMP and G␣ i behave as mutually exclusive inhibitors of adenylyl cyclase giving rise to a family of parallel lines on a Dixon plot obtained at different G␣ i concentrations (Fig. 4A). This kinetic pattern is true for both the recombinant membrane-bound enzyme and the cytoplasmic domains of type V adenylyl cyclase (Fig. 4B). Therefore any complex of adenylyl cyclase containing G␣ i has significantly reduced ability to obtain a conformation capable of binding this type of transition state analog. This same pattern of inhibition is also observed for the membrane-bound adenylyl cyclase with the more potent P-site inhibitor, 2Ј5Јdd3Ј-ATP (data not shown) (31). This is an important feature of catalysis because product release is one of the rate-limiting steps along the reaction coordinate for adenylyl cyclase (18).
Ap(CH 2 )pp is a non-hydrolyzable analog of ATP that competitively competes for ATP binding (18). As shown in the Dixon plot for the membrane-bound adenylyl cyclase (Fig. 5A), Ap(CH 2 )pp also shows a nearly parallel family of curves obtained at different G␣ i concentrations with respect to Ap(CH 2 )pp, indicating that each of these two inhibitors binds with greatly reduced affinity in the presence of the other. The intersection of these lines well below the x axis provides a reduction in the K i for Ap(CH 2 )pp of at least 6-fold in the presence of G␣ i . However, the cytoplasmic domains of type V show an intersection of lines on the x axis, suggesting no significant reduction in the affinity of the enzyme for Ap(CH 2 )pp in the presence of G␣ i (Fig. 5B). This non-competitive interaction is consistent with the formation of a G␣ i -C 1 -C 2 -ATP complex.
To further understand the binding of Ap(CH 2 )pp to the cy- toplasmic domains of adenylyl cyclase, we measured the direct binding of Ap(CH 2 )pp to C 1 , C 2 , and G␣ s in the presence or absence of G␣ i . These measurements were made using Mn 2ϩ , which greatly increases the affinity of adenylyl cyclase for Ap(CH 2 )pp as compared to Mg 2ϩ (18). An identical pattern of non-competitive inhibition was obtained for the soluble type V domains with Ap(CH 2 )pp and G␣ i in the presence of Mn 2ϩ as previously shown for Mg 2ϩ (Fig. 6). G␣ s is required to form a complex at reasonable concentrations of C 1 and C 2 , but G␣ s will reduce the effectiveness of G␣ i in this assay by reducing the binding of G␣ i to the C 1 domain. Therefore a complete loss of binding is not anticipated, but clearly a 50% reduction in Ap(CH 2 )pp is observed by both filter binding assays (Fig. 6B) and equilibrium dialysis (data not shown). Scatchard plot analysis of the filter binding data would suggest non-competitive binding of Ap(CH 2 )pp, although we were unable to fully saturate binding.
One unusual aspect of G␣ i -mediated inhibition of adenylyl cyclase is the inability of G␣ i to inhibit the membrane-bound adenylyl cyclase in the presence of manganese. This has been reported previously by Hildebrandt and Birnbaumer (32), but it was unclear at that time if this was caused by an effect of Mn 2ϩ on the G protein or catalytic subunit. We show that G␣ i is unable to inhibit type V adenylyl cyclase in the presence of Mn 2ϩ alone, Mn 2ϩ and G␣ s , or Mn 2ϩ and forskolin (Fig. 7A). In fact, G␣ i actually increases forskolin-stimulated activity in the presence of Mn 2ϩ . This may be due to the weak ability of G␣ i to compete at the G␣ s binding site (7). The inability of G␣ i to inhibit adenylyl cyclase in the presence of Mn 2ϩ is not caused by an inactivation of G␣ i , since G␣ i can inhibit the Mn 2ϩ -G␣ sstimulated cytoplasmic domains of adenylyl cyclase activity under identical conditions (Fig. 6A). In addition to 3 mM Mn 2ϩ , these assays all contain 0.5 mM Mg 2ϩ , which should be sufficient to maintain activation of the G proteins. The inclusion of an additional 1 mM Mg 2ϩ has no effect on the membrane-bound enzyme in the presence of Mn 2ϩ (Fig. 7B). The mechanism for this effect of metals on G␣ i -mediated inhibition and the difference between the soluble and membrane-bound adenylyl cyclase is discussed below.

DISCUSSION
Preliminary kinetic analysis suggests that G␣ s and G␣ i bind simultaneously to adenylyl cyclase; however, a complex of C 1 -C 2 -G␣ s -G␣ i has not been observed (7,19). This suggests a model in which G␣ i inhibits the enzyme by decreasing the affinity of one domain for another. This is in stark contrast to G␣ s and forskolin stimulation where the affinity between C 1 and C 2 increases with increased activation. We show that binding of G␣ i to the C 1 domain is weakened or lost in the presence of a C 1 -C 2 -forskolin or a C 1 -C 2 -G␣ s -forskolin complex (Fig. 1).
However, G␣ i inhibits the fully stimulated G␣ s -forskolin-activated enzyme very poorly, and we would not expect to observe a complex with both G␣ s and G␣ i under these conditions. Therefore, we also examined whether a heterotetrameric complex could be formed with G␣ i , G␣ s , and adenylyl cyclase in the absence of forskolin. Even under conditions of high protein concentrations, a complex of C 1 -C 2 -G␣ s -G␣ i is never observed.
A direct test of our hypothesis is the measurement of C 1 /C 2 affinity in the presence of G␣ i . It is clear that increased C 1 -C 2 complex formation, driven by increasing C 2 concentrations, decreases the ability of G␣ i to inhibit adenylyl cyclase activity (Fig. 2). This is consistent with the limited ability of G␣ i to inhibit the most stimulated forms of adenylyl cyclase that display the highest C 1 /C 2 affinity (7,19). In fact, a mutant C 2 protein (K1014N, Ref. 30) with increased affinity for C 1 displays dramatically decreased inhibition by G␣ i (Fig. 3). Although the membrane spans of adenylyl cyclase physically link C 1 and C 2 in close proximity, the structural changes that we measure through a change in affinity are still occurring in the native enzyme. Supporting a structural change at the interface of C 1 /C 2 is the fact that residues on one face of helix ␣2 interact with G␣ i , whereas residues on the opposite face of ␣2 interact with the C 2 domain (19,23).
A conformational change at the C 1 /C 2 interface and the close proximity of the G␣ i binding site to the catalytic site might also suggest that G␣ i influences the binding of molecules to the active site of adenylyl cyclase. Kinetic analysis of inhibition by P-site inhibitors and G␣ i reveals that the actions of these inhibitors are mutually exclusive. Therefore, binding of G␣ i prevents formation of specific conformations at the active site, particularly those capable of binding these transition state analogs. This is true for both the soluble and membrane-bound enzymes, and this pattern of inhibition is observed with both uncompetitive (2Јd3Ј-AMP) and non-competitive (2Ј5Јdd3Ј-ATP) types of P-site inhibitors.
It is clear that the key to regulation of adenylyl cyclase is the conformational state of the C 1 /C 2 domain interface. The change in affinity is a measurement of the changes in conformation at the active site located at the domain interface. This type of movement is mimicked somewhat with P-site inhibitors. Even with bound G␣ s and forskolin, the unliganded active site maintains an open conformation. But upon addition of P-site inhibitors that mimic a product-like transition state, the active site clamps down on the inhibitor as one might observe in an induced fit model of regulation (33). In the absence of activator, it is speculated that an even more open active site would be observed, similar to the structure obtained for the C 2 homodimer (34). This open to closed transition involves the inward collapse of structural elements around the active site. The G␣ i binding site is directly adjacent to these regions and may affect binding of molecules to the catalytic site. In fact, the most flexible regions of the C 1 domain are located on the back side of the G␣ i binding site. We put forth evidence that G␣ i mediates its effects by reducing the ability of the enzyme to obtain a closed conformation. This is observed as a reduced affinity between the C 1 and C 2 domain and a reduced ability to form a transition state conformation necessary for binding P-site inhibitors relative to the ground state.
The kinetic analysis of inhibition by the substrate analog Ap(CH 2 )pp reveals a different pattern. Kinetic and binding data of the soluble domains and the membrane-bound enzyme show non-competitive aspects between G␣ i and the substrate analog Ap(CH 2 )pp, as displayed by intersecting lines on a Dixon plot. However, the soluble domains have no reduction in the K i for Ap(CH 2 )pp in the presence of G␣ i , while the membrane-bound enzyme displays a reduced affinity for Ap(CH 2 )pp and G␣ i in the presence of each other. Despite this difference, clearly both G␣ i and a substrate analog can bind simultaneously to adenylyl cyclase. However, it is not clear whether a G␣ i -ATP-bound enzyme can lead to cAMP production. G␣ i binding may allow limited binding of substrate to an inactive enzyme. Alternatively, G␣ i binding may produce a catalytically competent enzyme but with greatly reduced activity. The latter rationale might explain why G␣ i inhibition of either the soluble domains or membrane-bound type V AC never leads to zero or basal activity levels. Generally, the maximal inhibition by G␣ i is 50 -70% when stimulated with modest levels of G␣ s (7), suggesting that an AC-G␣ i complex might retain low activity and hence bind substrate. Additional experiments are required to differentiate between these two possibilities.
The effects observed with manganese may also point to an inability of G␣ i to interact with a closed conformation. Residues from both C 1 and C 2 domains are required for binding ATP and ATP analogs (23,35,36). In fact, P-site analogs bound at the active site can stabilize complex formation between C 1 and C 2 (30). Manganese acts to increase the affinity of adenylyl cyclase for ATP analogs. This is observed as a 22-fold decrease in the K d for Ap(CH 2 )pp and a 2.5-fold decrease in the K m for ATP with the G␣ s -stimulated cytoplasmic domains of type V adenylyl cyclase in the presence of Mn 2ϩ versus Mg 2ϩ (36). The difference between Mn 2ϩ and Mg 2ϩ is even more dramatic for the basal or forskolin-stimulated soluble enzyme (36). This phenomenon is also present in the membrane-bound enzyme yielding a 4 -7-fold reduction in the K m for ATP in the presence of Mn 2ϩ (37) and a 5-fold reduction in the K i for competitive ATP analogs (38). We measure a 30-fold reduction in the IC 50 for Ap(CH 2 )pp in the presence of 100 M ATP and manganese versus magnesium (data not shown). A similar reduction in K i is observed in the presence of manganese for all P-site analogs; however, the mechanism of this reduction may be due to the increase in activity of the enzyme rather than the increase in P-site affinity (18,39). By increasing the affinity of metal-ATP for the enzyme, manganese may be serving to increase the interaction between C 1 and C 2 , driving the enzyme to a more closed conformation.
The question is why is this observed only for the membranebound enzyme. The soluble enzyme may be more susceptible to G␣ i inhibition because of its very nature of two separate proteins. The membrane-bound enzyme holds the C 1 /C 2 domains in the optimal orientation for catalysis. Although V max is comparable to the soluble domains, the K m for metal-ATP and the affinity for ATP analogs are considerably lower for the membrane-bound enzyme (ϳ13-fold difference in the K m for Mg-ATP, data not shown). This may be part of the reason that G␣ i reduces the affinity of Ap(CH 2 )pp for the membrane-bound enzyme and not for the soluble domain. The increased affinity for metal-ATP, particularly Mn-ATP, may reflect an inability of G␣ i to inhibit this enzyme as compared with the C 1 /C 2 domains that have substantially weaker affinity for substrate. For the C 1 /C 2 domains an increase in affinity for Mn-ATP is not sufficient to significantly reduce the inhibition by G␣ i . As with all model systems, the cytoplasmic domains have been an invaluable tool but may have their limits in faithfully reproducing the kinetic features of adenylyl cyclase.
In summary, these data suggest a model where G␣ i decreases the interaction of C 1 and C 2 for each other and also decreases catalytic activity by decreasing the formation of the active site. This is directly opposite to the actions of G␣ s and highlight the pseudosymmetry of the adenylyl cyclase structure that is poised for bidirectional regulation.