Research Article

Crystal Structure of the Catalytic Domains of Adenylyl Cyclase in a Complex with G·GTPγS

See allHide authors and affiliations

Science  12 Dec 1997:
Vol. 278, Issue 5345, pp. 1907-1916
DOI: 10.1126/science.278.5345.1907


The crystal structure of a soluble, catalytically active form of adenylyl cyclase in a complex with its stimulatory heterotrimeric G protein α subunit (Gs α) and forskolin was determined to a resolution of 2.3 angstroms. When P-site inhibitors were soaked into native crystals of the complex, the active site of adenylyl cyclase was located and structural elements important for substrate recognition and catalysis were identified. On the basis of these and other structures, a molecular mechanism is proposed for the activation of adenylyl cyclase by Gs α.

Adenylyl cyclase is the effector enzyme responsible for converting adenosine triphosphate (ATP) to adenosine 3,5-monophosphate (cAMP), a ubiquitous second messenger that mediates diverse cellular responses primarily by activating cAMP-dependent protein kinases. Nine isoforms of adenylyl cyclase have been identified (ACI through ACIX). These proteins share several regulatory features: activation by the α subunit of the heterotrimeric G protein Gs, activation by the diterpene forskolin, and inhibition by a class of adenosine analogs known as P-site inhibitors. The activity of individual isoforms can also be regulated by other G protein subunits, Ca2+-calmodulin, Ca2+, or phosphorylation (1-3). This regulatory diversity allows each isoform of adenylyl cyclase to respond to intercellular and intracellular signals in a manner appropriate for its cellular context.

The first cDNA encoding a mammalian adenylyl cyclase revealed a protein of structural complexity commensurate with the difficulties encountered in its purification and expression (4). Adenylyl cyclase has a short cytosolic amino terminus, followed by two repeats of a module composed of a hydrophobic domain, modeled as six transmembrane helices, followed by a ∼40-kD cytosolic domain. Each of the two cytoplasmic domains, designated C1 and C2, respectively, contains a region of ∼200 amino acid residues (C1a, C2a) that is homologous between C1 and C2 and among all isoforms of the enzyme. Similar domains are also found in membrane-bound and soluble guanylyl cyclases (5).

The cytosolic domains of adenylyl cyclase are implicated in catalysis. The C1a and C2a domains from the same or from different isoforms of adenylyl cyclase can be expressed in bacteria as 50-kD fusion proteins that retain high enzymatic activity (6,7). The individual domains can also be expressed and purified as 25-kD proteins (8, 9). Although these individual domains form homodimers that retain little or no catalytic activity, adenylyl cyclase activity is restored when they are mixed. The cytosolic domains also retain the common regulatory properties of adenylyl cyclases (6, 8-10).

We overexpressed and purified the C1a domain from adenylyl cyclase V (VC1) and the C2 domain from adenylyl cyclase II (IIC2). The apparent affinity of VC1for IIC2 is weak (>10 μM), and the basal catalytic activity of their complex is low (∼100 nmol min-1mg-1). In the presence of forskolin or guanosine 5′-O-(3-thiotriphosphate)–activated Gs α(Gs α·GTPγS), or both, the apparent affinity of VC1 for IIC2 and their catalytic activity increase (dissociation constant K d = 150 nM, specific activity ≥150 μmol min−1mg−1) (11). Activators of adenylyl cyclase thus appear to facilitate interactions between the two cytosolic domains and stimulate catalysis. Data from sedimentation equilibrium, gel filtration, and equilibrium dialysis studies indicate that the forskolin- and Gs α-activated complex contains one molecule each of VC1, IIC2, Gs α·GTPγS, and forskolin (11,12). The activated complex is able to bind one molecule of a nonhydrolyzable ATP analog, Ap(CH2)pp, presumably at the interface between the C1 and C2 domains. Like native adenylyl cyclase, the VC1·IIC2 complex is inhibited by P-site inhibitors, a class of compounds that includes 2′-deoxyadenosine and its 3′ mono- or polyphosphates (13). These compounds are noncompetitive or uncompetitive inhibitors with respect to ATP, and they bind with highest apparent affinity to the activated enzyme and only in the presence of pyrophosphate. The VC1 and IIC2 domains are also amenable to structural analysis, as demonstrated below and by the crystal structure of the IIC2homodimer (14). Although the IIC2 homodimer has little or no enzymatic activity, its structure provides a model for the C1·C2 heterodimer and demonstrates how forskolin binds within and stabilizes the C1·C2 interface.

We now present the structure of activated Gs αalone [described in a companion paper (15)] and the structure of Gs α associated in a quaternary complex with VC1, IIC2, and forskolin. We also determined the structure of this forskolin-bound heterotrimer in a complex with the P-site inhibitor 2′-deoxy-3′-adenosine monophosphate (2′,d3′-AMP) and pyrophosphate, which provides direct insight into the mechanism of catalysis and the structural mechanism of P-site inhibition. Taken together, these structures also permit us to propose a molecular mechanism for activation of adenylyl cyclase by Gs α.

Structure determination. Bovine G· GTPγS·Mg2+ was crystallized in a ternary complex with a C1 fragment of canine ACV (residues 364 to 580) (VC1) and a C2 fragment of rat ACII (residues 874 to 1081) (IIC2) in the presence of 7-deacetyl-7(O-N-methylpiperazino)-γ-butyryl forskolin (MPFsk) (16). We used the short alternative splice form of Gsα (17), which was synthesized in Escherichia coli with a carboxyl-terminal hexahistidine tag and purified as described (11). The recombinant Gsα contains no lipid modifications. VC1, which has an NH2-terminal hexahistidine tag, and IIC2 were also produced in E. coli (8, 18). Before being crystallized, the quaternary complex of Gsα, VC1, IIC2, and forskolin was purified by gel exclusion chromatography (11).

Crystals of Gs α·VC1·IIC2 grow as 10- to 20-μm-thick plates and belong to space groupP21 2 1 2 (Table1). There is one complex per asymmetric unit of the crystal. Diffraction is anisotropic, such that data normal to the plate (the a*b* plane) extend only to 3.9 Å spacings, compared with a diffraction limit of 2.3 Å in thea* and b* directions. Therefore, reflections with |l| > 18, corresponding to spacings less than 3.9 Å in the direction along c*, were removed from the data set before scaling. To prepare complexes with P-site inhibitors, we soaked crystals in a harvesting solution containing pyrophosphate and either 2′,d3′-AMP or 2′-deoxy-3′-ATP (2′,d3′-ATP). After exposure to P-site inhibitors, crystals displayed increased mosaicity and concomitant reduction in the maximum limit of diffraction to 2.8 Å spacings in the a* and b* directions. Attempts to cocrystallize Gs α·VC1·IIC2 with either P-site inhibitors or Ap(CH2)pp have so far been unsuccessful. Complete data sets extending to 2.3 and 2.8 Å diffraction spacings were recorded at the Cornell High Energy Synchrotron Source (CHESS) for the Gs α·VC1·IIC2complexes in the absence and presence of P-site inhibitors, respectively (Table 1).

Table 1

Summary of data collection and refinement statistics. Diffraction data from crystals of the VC1·IIC2·Gs αcomplex (16) were measured on a 1k by 1k ADSC charge-coupled-device area detector with 0.908 Å radiation from the A1 beamline at CHESS. All crystals were flash frozen in nitrogencooled liquid ethane on 0.2-mm cryoloops (Hampton) and were maintained throughout data collection in a cryostream at −180°C. With a 0.2-mm collimator, the mosaicity for native crystals of the complex was 0.5°, whereas those for the P-site complexes were typically 1.0°. Diffraction amplitudes from crystals of the complex were indexed and integrated with DENZO and scaled with SCALEPACK (44). Because the diffraction was anisotropic, reflections with l indices that had an absolute value greater than 18, corresponding to a 3.9 Å resolution cutoff along thec* axis, were removed before scaling; the averageII for the omitted reflections after scaling was 1.7. The structure of the complex was initially solved by molecular replacement, using the AMORE program (45) as implemented by the CCP4 program package (46), with 3.5 Å data collected by using CuKα radiation on a MacScience 2030DIP area detector. The unmodified structure of Gi α·GTPγS (19) served as the search model for Gs α and that of the IIC2 homodimer (14) served as the search model for the VC1·IIC2 heterodimer. The initial model for the native complex was built in O (47), and all subsequent refinement steps were performed with X-PLOR (48). Anisotropic overall B-factors (Aniso.B-factor) and a bulk solvent mask were used throughout the final rounds of refinement and model building. After convergence of the native structure, a difference Fourier map was used to reveal conformational changes in the complex containing P-site inhibitors. The polypeptide backbone was altered accordingly, and after refinement, weak but interpretable difference density was observed for the P-site inhibitor, pyrophosphate, and magnesium. After addition of these molecules to the atomic model, the P-site complex was subsequently refined and rebuilt through several more cycles. A total of 91% of the amino acids in the native complex and 87% of the amino acids in the P-site inhibited complex are found in the most favored regions, and none are found in disallowed regions of the Ramachandran plot (49).

View this table:

We determined the structure of the Gs α·VC1·IIC2complex by molecular replacement using the structure of Gi α 1 (19) and the IIC2 homodimer (14) as search models. After several cycles of manual model reconstruction and refinement, the conventional R factor was 22% and the free Rfactor (20) was 28%. Difference Fourier analysis revealed substantial conformational changes upon introduction of either of the two P-site inhibitors and pyrophosphate. These changes were primarily confined to two segments of polypeptide chain in VC1 and the COOH-terminal β-ribbon of IIC2. The atomic model was revised in accordance with difference maps, and after subsequent refinement of this model, the P-site inhibitor, pyrophosphate, and Mg2+ could be discerned in difference density maps. The model of the complex containing MPFsk, 2′,d3′-AMP, and pyrophosphate has been refined to conventional and free R factors of 24 and 32%, respectively.

All three proteins in the Gs α·VC1·IIC2complex are disordered at their NH2- and COOH-termini. The refined structure of Gs α spans residues 36 to 393 (394 is the COOH-terminal residue; residue numbering is for the long splice variant of Gs α). However, no electron density is observed for five residues in the linker 1 segment between residues 66 and 86 (alternative splicing excludes 14 of the residues in this range from the long form of Gs α). The atomic model of VC1spans residues 376 to 565, and the structure of IIC2 is well ordered between residues 879 and 1077 except for residues 954 to 963 in its β3′-α3′ loop.

Structure of the VC1·IIC2complex. As expected from their amino acid sequences, the tertiary structures of IIC2 and VC1 are almost identical (Fig.1), and the overall quaternary structure of the VC1·IIC2 heterodimer is similar to that of the IIC2 homodimer (14), although the two domains are rotated by 7° with respect to each other (see below). Secondary structure elements of IIC2 are identified according to the convention of Zhang et al. (14) but are primed to distinguish them from the analogous structures in VC1. The surface of the heterodimer perpendicular to the pseudo twofold axis (Fig. 1A) is denoted the ventral surface (14). A long, shallow trough running diagonally across this face of the heterodimer contains the forskolin and substrate binding sites. In contrast, extensive contacts between IIC2 and VC1 would appear to restrict access of substrates to their binding sites from the opposite, dorsal surface. In part because the NH2-termini of VC1 and IIC2 project away from the dorsal surface of the heterodimer, we propose that this surface faces the membrane. Nevertheless, because parts of the segments that connect the cytosolic to the membrane-spanning domains are either not present in our constructs or are disordered, alternative orientations are possible.

Figure 1

Architecture of the heterodimeric complex between VC1 (mauve) and IIC2 (khaki) bound to the forskolin analog MPFsk, as observed in the complex with Gs α·GTPγS. (A) View along the pseudo-twofold axis toward the ventral surface of VC1·IIC2. The forskolin derivative is shown as a stick figure: carbon, gray; nitrogen, cyan; and oxygen, red. N and C mark the first and last ordered residues in the crystal structure of the heterodimer. (B) VC1 is depicted side-by-side with a molecule of IIC2 that has been superimposed on VC1 (rmsd of 1.3 Å for 153 Cα atom pairs). Elements of secondary structure are labeled; in all of the figures, the color of the label identifies the protein subunit to which it refers. (C) Stereo diagram of the VC1·IIC2 interface, with the Cα backbone depicted as a continuous tube. The view is the same as in (A). Ball-and-stick models of Cα−Cβ bonds are shown for residues (50) that participate in interdomain contacts (separated by less than 4 Å from an atom in the opposite domain). These residues constitute the subset of interfacial residues that are conserved in all adenylyl cyclase isoforms. Cβ atoms of residues with acidic side chains are red, basic residues are blue, and residues with polar side chains are pink. The Cβ atoms of nonpolar residues are khaki (VC1) or mauve (IIC2). Dashed gray lines show interdomain side chain–side chain or side chain–main chain hydrogen bonds or ion pairs involving polar or charged interfacial residues. Only the polar or charged residues are labeled. Figures were drawn with the program BOBSCRIPT (51) and rendered with RASTER3D (52).

VC1 and IIC2 superimpose with a root- mean-square deviation (rmsd) of 1.3 Å for 153 structurally equivalent Cα atom pairs. There are several notable structural differences between the domains. The first of these is found by comparing β1-α1-α2 with β1′-α1′-α2′ (Fig. 1B). The COOH-terminus of strand β1 is two residues longer than β1′ and is followed by a four-residue type I turn. In contrast, β1′ is followed by a three-residue 310 turn. Helix α1 is longer than α1′ by one helical turn and begins one residue earlier in the primary structure. Whereas α1 and α2 are connected by a common helical hairpin turn, the polypeptide chain succeeding α1′ forms a four-residue β-like extension, followed by a 310 and then a left-handed helical turn before entering the α2′ helix. The more complex α1′-α2′ loop is partially the consequence of a four-residue insertion in the primary structure of IIC2 relative to VC1 (Fig. 2). The structural differences between these elements reflect their distinct functional roles. We show below that the turn at the NH2-terminus of α1 forms a P-loop–like structure that binds pyrophosphate, whereas the α1′-α2′ loop forms part of the Gs αbinding site.

Figure 2

Structure-based alignment of the amino acid sequences (50) of selected adenylyl and guanylyl cyclase C1 and C2 domains. Only the sequences that correspond to the VC1 and IIC2 constructs used in this study are shown. GCα1 and GCβ1 correspond to the α and β subunits of rat soluble guanylyl cyclase (53). The remaining sequences correspond to the C1 and C2domains of various isoforms of adenylyl cyclase, as indicated (4,54, 55). The top four sequences correspond to C1domains, and the lower four correspond to C2 domains. The amino acid position of each enzyme is shown at the beginning of each line, and the residue numbers for VC1 and IIC2are indicated above and below the alignment, respectively. The secondary structure of the C1 domain is drawn in khaki above the alignment, and that of the C2 domain is drawn in mauve below the alignment. Arrows represent β strands, and coils represent α helices. Other secondary structural elements, such as random coils and turns, are represented by a solid line. Disordered regions are indicated by the lack of corresponding secondary structure. Red dots above the sequence of VC1 or below the sequence of IIC2 indicate residues from each domain that form the VC1·IIC2 interface. Red amino acid symbols correspond to residues that directly interact with Gs α. Green amino acid residues indicate the region believed to be important for interactions with the βγ subunit in IIC2 (21). Various functional residues in VC1 and IIC2 are indicated by symbols: f, forskolin binding; a, adenine binding; p, pyrophosphate binding; m, magnesium binding; and r, ribose binding.

Among the C2 domains of adenylyl cyclase isoforms, the β3′-α3′ loop is a site of sequence length variation (Fig. 2). This loop is poorly ordered in the crystal structures of both the IIC2 homodimer and the Gs α·VC1·IIC2heterotrimer. Sequences within the loop and the succeeding helix have been implicated in the interaction of adenylyl cyclase with the βγ subunits of heterotrimeric G proteins (21), although if βγ binds this peptide in the manner proposed (22), it would be upside-down with respect to the membrane if the dorsal surface of VC1·IIC2 faces the membrane as suggested above. In VC1, the β3-α3 loop is an ordered seven-residue turn that contains the only cis-proline observed in either VC1 or IIC2.

The structural similarity between VC1 and IIC2breaks down at a point that corresponds to the COOH-terminus of β7 (middle of β7′) (Fig. 2) but resumes at the COOH-terminus of α7 (NH2-terminus of β8′). In IIC2, β7′ and β8′ form a long β ribbon in which the two strands are connected by a simple reverse turn (Fig. 1B). The end of the ribbon passes over the ventral channel of the VC1·IIC2 heterodimer and forms part of the P-site inhibitor–pyrophosphate binding site (see below). This region is disordered in the crystal structure of the IIC2 homodimer (14). Strand β7 of VC1 is much shorter than its counterpart in IIC2. In place of the β ribbon is a compact Ω loop (23) that contains two short α helices. The loop does not participate in Gs α or P-site inhibitor binding. Although the primary structure is not conserved between C1 and C2 in this region, it is maintained between corresponding domains in most adenylyl and guanylyl cyclase isoforms. Therefore, this COOH-terminal subdomain may participate in regulatory interactions that are unique to C1 or C2.

The VC1·IIC2domain interface. The interface between VC1 and IIC2 buries more than 3300 Å2 of solvent-accessible surface area (24). Including the molecule of MPFsk, a total of 4200 Å2 of surface area is buried upon forming the VC1·IIC2·MPFsk complex. The pseudo-twofold symmetry of VC1·IIC2 is evident in their interface. Interfacial contacts (defined by interatomic distances of less than 4.0 Å) involve 28 residues in VC1 and 33 residues in IIC2. Only 17 of these residues from each domain are found at structurally equivalent positions. Of these residues, only 12 are conserved between C1 and C2 (Fig. 2).

There are two pseudosymmetric regions that form the VC1·IIC2 interface (Fig. 1C). In the first and much more extensive of these, closest to Gs α (upper right-hand side of Fig. 1C), β2′of IIC2 stacks on top of the β4-β5 loop of VC1. The α2′ helix contacts the NH2-terminal segment of β1 and the COOH-terminal end of β4, and the α1′-α2′ loop abuts the β5-α4 turn and the midsections of β1 and β4 (Fig. 1B). The interface is composed of both hydrophobic and polar interactions and is further strengthened by the binding of MPFsk between α1′ and β5 (see below). Two of the structural elements that form this interface, α1′ and α2′, also form part of the binding site for Gs α.

The second pseudo-equivalent domain interface, distant from the Gs α binding site, is significantly different (Fig. 1C). Strand β2 forms fewer contacts with the β4′-β5′ loop than does β2′ with β4-β5. The small number of contacts to the β2-β3 loop, which carries a catalytically essential aspartic acid, may allow this structure to be easily manipulated by regulators of adenylyl cyclase. In addition, the contact of α2 with IIC2 is less extensive, and interfacial contacts involving α1 are essentially absent. Helix α1 is therefore easily manipulated by the binding of P-site inhibitor and pyrophosphate (see below).

The C1·C2 interface and its asymmetry are crucial to the function of adenylyl cyclase. The interface is responsible for forming the forskolin and Gs αbinding sites at one end and the active site at the other. In addition, we believe that the C1 and C2 domains change their relative orientations upon binding of Gs α, representing, at least in part, a mechanism for activation of adenylyl cyclase by the G protein. Critical to this conformational change is the maintenance of the C1·C2 interfacial contacts in both the unactivated and activated states of the enzyme. As will be discussed below, the C1·C2 interface is formed by flexible elements of one domain (for example, the β4′-β5′ loop) that interact with elements from the structural core of the other (for example, α2). Mutations of interfacial residues further highlight the importance of the C1·C2 interface (Table2). For example, alanine mutants of interfacial residues equivalent to either Asp424 in VC1 or its counterpart in IIC2, Asp923, are largely inactivating.

Table 2

Structure-function correlations of various adenylyl cyclase mutants. Mutations made in different isoforms of adenylyl cyclase (first column) are mapped onto VC1 or IIC2 (second column). [Mutation name specifies the residue at the given position, mutated to the residue listed at the end (50).] Their observed phenotype is described in the third and fourth columns, and their role in the VC1·IIC2·Gs αcomplex is described in the fifth column. EC50, median effective concentration; IC50, median inhibitory concentration.

View this table:

Forskolin binding. In contrast to the pair of forskolin molecules bound to the IIC2 homodimer (14), only one molecule of MPFsk binds to the VC1·IIC2heterodimer (Figs. 1A and 3). The single forskolin binding site, flanked on one side by β5-α4 and on the other by α1′ and the β2′-β3′ loop, is structurally equivalent to and can be superimposed with both sites in the IIC2 homodimer (Fig.3). The forskolin binding site in VC1·IIC2 is adjacent to that for Gsα. Contacts between forskolin and IIC2 are generally the same as observed in the IIC2 homodimer (Lys896, Ile940, Gly941, and Ser942). Most of the contacts between forskolin and VC1 are either identical (Phe394 versus Phe889, Trp507versus Trp1020, Val511 versus Val1024) or analogous (Tyr443 versus Met945) to those between forskolin and the corresponding IIC2 subunit of the homodimer (Fig. 3). A water molecule bridges the O1 and O4 oxygens of the forskolin analog and the side chain of Ser942. The piperazino tail of the forskolin analog used in this study is poorly ordered compared with the remainder of the molecule. However, its proximity to the side chains of Cys441 and Asn515 may allow the design of similar extensions that would create an analog with higher affinity.

Figure 3

Only one molecule of forskolin binds in the ventral cleft of VC1·IIC2. MPFsk binds in the ventral cleft of adenylyl cyclase at the end closest to the Gs α binding site and is drawn in green without its methyl-piperazino group for clarity. Residues constituting the forskolin binding site of the IIC2 homodimer (14) that differ from their equivalents in the binding site of VC1·IIC2 are drawn in transparent rose (50). The side chain of Trp1020 is also shown because its side chain adopts a dramatically different conformation from that of Trp507 in the VC1·IIC2 heterodimer. To generate this figure, we superimposed one of the forskolin molecules from the IIC2 homodimer with MPFsk; this superposition does not optimally align the protein subunits of each structure.

Forskolin cannot bind to the second, pseudosymmetric site of VC1·IIC2 (12). Of the residues that would contact the putative forskolin molecule at this site, there are only two that are significantly different from the first site. The Thr512 residue is replaced by Asn1025 and, perhaps more importantly, Ser942 is replaced by Asp440. Both substitutions would result in steric conflicts with a bound molecule of forskolin. As was mentioned above, the relative orientation of the C1 and C2 domains within the Gs α·VC1·IIC2complex is significantly different from that of the C2domains in the IIC2 homodimer. One effect of this quaternary change is to alter the position of the α4′ helix in the ventral cleft such that the Asn1025 side chain overlaps the potential forskolin binding site. If the VC1 domain is superimposed on one of the IIC2 domains in the homodimer, only the overlap with Asn1025 is alleviated; that with Asp440 is not. Furthermore, the binding of a second molecule of forskolin at this site would inhibit adenylyl cyclase by interfering with substrate binding. Thus, it is very unlikely that two molecules of forskolin bind to the ventral cleft of intact adenylyl cyclase.

When bound to the domain interface, forskolin stabilizes the interaction between VC1 and IIC2(14). Accordingly, forskolin, like Gs α, increases the apparent affinity of VC1 for IIC2 from ≥10 to ∼1 μM (11). It has thus been proposed that forskolin activates adenylyl cyclase by promoting C1·C2association, resulting in formation of an active site (14). However, forskolin does not significantly affect the inhibition constant (K i) for the substrate inhibitor Ap(CH2)pp nor the Michaelis constant (K m) of ATP in either our soluble system (12) or for a IC1-IIC2 chimera (6, 7). Rather, forskolin increases catalytic activity (V max) by a factor of 60, a factor 10 times greater than its effect on domain association (11). Domain association is obviously a prerequisite for adenylyl cyclase activity. However, domain association alone is clearly not sufficient to achieve the high levels of adenylyl cyclase activity displayed in the presence of forskolin, Gs α, or both activators (11). Hence, it is likely that the binding of forskolin facilitates cAMP synthesis by directly altering the conformation of the active site. The structure suggests one way such activation could occur. The β2′-β3′ loop contacts Lys436 and Leu438 of the β2-β3 loop of VC1 (Fig. 1C), thereby linking residues that form the forskolin binding site with structural elements carrying residues important for catalysis (see below).

Interaction of G with VC1·IIC2 . The orientation of Gsα relative to VC1·IIC2 also suggests that the dorsal surface of the heterodimer is close to the plasma membrane. Thus, substrates would have direct access to the active site of adenylyl cyclase from the cytoplasm, and the palmitoylated NH2-terminus of native Gsα would be close to the membrane (Fig. 4). The orientation, as proposed here, of the α subunit with respect to the membrane is the same as that suggested for the α subunit in the heterotrimer and in models of the receptor complex (25, 26).

Figure 4

The complex between Gs α·GTPγS and VC1·IIC2 (50). The top of the figure shows two views of the complex. At top left, the complex is drawn with the wide, interdomain cleft (ventral surface) between the C1 and C2 domains facing the bottom of the page. The dorsal surface, which is presumed to face the plasma membrane, faces upward. The two adenylyl cyclase domains are colored as in Fig. 1; the ras-like domain of Gs α is colored charcoal, and the helical domain is ash gray. The switch II segment (Sw II) (which includes the α2 helix of Gs α) is red. GTPγS and MPFsk are depicted as stick figures with the atomic coloring scheme used in Fig. 1. At bottom left is a detail of the Gs α:VC1·IIC2interface in about the same orientation as that directly above. Hydrogen bonds are depicted with dashed green lines. Side chain carbon atoms are colored according to the main chain to which they are attached. Of note is the intimate cluster formed by the NH2-terminus of VC1, α2′ of IIC2, and both switch II and the α3 helix of Gs α. The GTPγS bound to Gs α is shown for reference. At top right, the view of the complex is related by a rotation of ∼70° about a horizontal axis relative to the view shown on the top left. This rotation brings the helical domain of G toward the reader and provides a view of the ventral surface of VC1·IIC2. At bottom right is a detailed view of the interface in the same orientation as that shown directly above.

The interface between Gs α and VC1·IIC2 buries 1800 Å2 of solvent-accessible surface area. Two structural elements of Gs α form the interface, primarily through contacts with IIC2 (Fig. 4). The most striking interaction is the insertion of the Gs α switch II helix (residues 225 to 240) into the groove formed by α2′ and the α3′-β4′ loop of adenylyl cyclase. Yan et al.(27) correctly identified this Gs αbinding surface of adenylyl cyclase by mutagenic analysis of the recombinant enzyme. The second contact surface is formed by the α3-β5 loop of Gs α, which interacts with both VC1 and IIC2. These two adenylyl cyclase–interacting surfaces on Gs α were identified by the scanning mutagenesis performed by Berlot and Bourne (28) and (for the α3-β5 loop) by Itoh and Gilman (29). In contrast, the α4-β6 loop of Gs α, which was also predicted to interact with adenylyl cyclase (28), is located more than 10 Å from the interface and does not interact with VC1·IIC2. Although it is difficult for us to imagine how the α4-β6 loop of Gs α could interact with adenylyl cyclase, we cannot exclude the possibility that this loop interacts with elements of adenylyl cyclase that are not present in our constructs.

All of the residues that interact with Gs αare conserved among adenylyl cyclase isoforms (Fig. 2). The Gs α·IIC2 interface is a mixture of hydrophobic and polar contacts, whereas Gs α·VC1 contacts are entirely hydrophobic (Fig. 4). Only two of the Gs αresidues involved in these interfaces with adenylyl cyclase, Gln236 and Asn239, are significantly different from analogous residues (His213 and Glu217) in the inhibitory G protein α subunit, Gi α. However, these substitutions are conservative with respect to the structure of the interface with adenylyl cyclase. Therefore, the specificity of Gs α and Gi α subunits for adenylyl cyclase appears to be dictated primarily by the backbone conformation of the residues that form the interface, not their primary structure (15).

The regions of Gi α that interact with adenylyl cyclase have recently been determined by scanning mutagenesis; they map to switch II and, less convincingly, to the α4-β6 loop, but not to the α3-β5 loop (30). Furthermore, inhibition of ACV and ACVI is not competitive with respect to Gs α (31). Therefore, Gi α appears to bind a site on adenylyl cyclase distinct from that of Gs α. Given the homology of Gs α and Gi α and their presumably similar orientation with respect to the membrane, the most attractive binding site for Gi α is between the α1-α2 loop and the α3 helix on VC1, directly opposite to the binding site of Gs α. This region contains many residues that are invariant in ACV and ACVI (for example, Glu398 and Leu472) but not in Gi α-insensitive cyclases. The binding of Gi α at this site could easily influence the structure of the active site and cause inhibition.

Many mutations inhibit activation of adenylyl cyclase by Gs α (Table 2). Several of these involve the residues that constitute the Gs α binding site; more interesting are those that do not interfere with either Gs α binding or the fold of the protein. Certain mutations, such as those that correspond to Asp923and Lys936, are involved in maintenance of the C1·C2 interface (Fig. 1C). Although Gs α still binds to these mutants, its apparent affinity is less, presumably because the α subunit is unable to effect the proper conformational change at the C1·C2 interface. Other mutants, such as Asp440 or Arg1029, are intimately involved in active site structure and catalysis and therefore have greatly inhibited activity.

Binding of P-site inhibitors. P-site analogs act as dead-end inhibitors of pyrophosphate (product) release and inhibit adenylyl cyclase only in the presence of pyrophosphate and Mg2+ or Mn2+ (32). Although these compounds are noncompetitive or uncompetitive inhibitors of cAMP synthesis, they are competitive with cAMP in the reverse reaction (32). Furthermore, kinetic studies of P-site inhibition and the structure of the complex itself both strongly suggest that the P-site and the substrate (ATP) binding site are the same. The P-site inhibitor·pyrophosphate·Mg2+ complex therefore best mimics the product complex of adenylyl cyclase (Fig. 5A).

Figure 5

Global and detailed views of the complex between the P-site inhibitor 2′,d3′-AMP, pyrophosphate, and VC1·IIC2. The view is toward the ventral face of VC1·IIC2. (A) The P-site inhibitor and pyrophosphate are bound in the ventral cleft at the end furthest from Gs α (the switch II region of Gs α is red and its surrounding structural elements are charcoal) (50). The P-site is related by a pseudo-twofold axis of symmetry to the MPFsk binding site. MPFsk is modeled as a molecule of forskolin because its methylpiperazino tail was very poorly ordered in the P-site–inhibited structure. The position of the β1-α1-α2 and α3-β4 elements of VC1 and the β7′-β8′ β ribbon of IIC2 in the absence of P-site inhibitor and pyrophosphate are depicted as transparent, rose-colored ribbons. (B) Detail of the catalytic site showing 2′,d3′-AMP, pyrophosphate, and Mg2+ (purple). The green wire cage depictsF oF c difference electron density contoured at the 2.0σ level. Crystallographic phases used in the calculation of this map were based on an atomic coordinate set obtained after modeling and refinement of the conformational changes described above, but before inclusion of coordinates for 2′,d3′-AMP, pyrophosphate, and Mg2+ (see legend to Table 1 for experimental details). Hydrogen bonds and coordination to Mg2+ are depicted with gray dashed lines.

Pyrophosphate, 2′,d3′-AMP, and Mg2+ bind to the end of the ventral cleft of VC1·IIC2 that is farthest from Gs α. The three ligands are not well ordered in the structure and exhibit high-temperature factors, suggesting incomplete occupancy or incomplete formation of the binding pocket due to constraints imposed by the crystalline lattice. However, omit maps clearly demonstrate where each molecule binds (Fig. 5B). The site is pseudosymmetric to the binding site for forskolin. Similarly, the P-site is formed by α1 and its preceding loop, the β2-β3 loop, β4, and β4′-α4′. Pyrophosphate is positioned over a P-loop at the NH2-terminus of α1 and is bound by Mg2+ and three basic residues: Arg484 (β4), Arg1029 (α4′), and Lys1065 (β7′-β8′ turn). Three periplanar ligands of the Mg2+ ion are formed by Asp396 (β1), Asp440 (β2-β3 loop), and an oxygen atom from the β phosphate of pyrophosphate. One axial ligand is donated by the carbonyl oxygen of residue 397. The unoccupied coordination sites of Mg2+ could easily be filled by water molecules, although none are apparent in our structure. In the native structure of the Gs α·VC1·IIC2complex, weak electron density, presumably that of Mg2+, also occurs between the side chains of Asp396, Asp440, and the carbonyl oxygen of residue 397.

The purine ring is stacked over the peptide plane of residues 438 and 439. Adenine nucleotides are recognized, to the exclusion of guanine, by hydrogen bonds from the purine N6 atom to Asp1018 (β5′) and the backbone carbonyl of residue 1019. A hydrogen bond is also formed between Lys938 (β2′) and the purine N1 atom. In the guanylyl cyclases, the residues corresponding to Asp1018 and Lys938 are substituted by cysteine and glutamate (Fig. 2), respectively, perhaps explaining the difference in specificity for purine nucleotides. The alanine mutant in ACI of the residue corresponding to Asp1018 in IIC2 is inactivating (Table 2).

The ribosyl moiety of 2′,d3′-AMP appears to adopt a 2′-endo conformation, with its O4′ atom packing against the side chain of Ser1028 (α4′) and its 5′-hydroxyl forming a hydrogen bond with the side chain of Thr401. The purine ring isanti to the ribose ring. The 3′-phosphate Pα-O3′ bond istrans to the ribose C3′-C4′ bond, and the 3′-phosphate binds between the Pβ of pyrophosphate and Asp440, perhaps forming hydrogen bonds to each. The complex formed with 2′,d3′-ATP is essentially identical to that formed with 2′,d3′-AMP; the two additional phosphate groups appear disordered. Potential interactions with positively charged residues surrounding the opening of the catalytic site may account for the increase by a factor of 10 in the apparent affinity for adenylyl cyclase (13) afforded by the 3′-triphosphate group.

The most potent P-site inhibitors have 3′ mono or polyphosphates, and 2′,5′-dideoxyribose moieties. In the conformation of the nucleotide bound to adenylyl cyclase, the 3′-phosphate appears to form hydrogen bonds with both Asp440 and pyrophosphate. A 2′-hydroxyl, which is absent in 2′,d3′-AMP, would collide with the 3′-phosphate. The 5′-hydroxyl of 2′,d3′-AMP is not well ordered, and its removal may permit a 5′-deoxy P-site inhibitor to bind in a slightly different orientation that has more favorable interactions with the protein.

Diffusion of P-site inhibitors into crystals of the Gs α·VC1·IIC2·MPFsk complex is accompanied by segmental conformational changes in VC1 and IIC2 (Fig. 5A). The most prominent of these is the rotation of α1 and the adjacent α3-β4 loop of VC1 toward the pyrophosphate binding site. The α1 helix forms new contacts with the α4′ helix of IIC2, most notably to Asn1025, a residue implicated in the catalytic mechanism of adenylyl cyclase (33). We suspect the α1 helix behaves like a lid for the active site that closes upon ligand binding. The β7′-β8′ β ribbon swings over the P-site, where Lys1065 forms an ion pair with pyrophosphate.

In general, mutations that affect theK i's for P-site inhibition affect theK m's for ATP to a lesser extent (Table 2). For example, the alanine mutant of the residue equivalent to Lys938 is only partially inactivating, but it increases theK i of 2′,d3′-AMP by a factor of 250 compared with a factor of 3 increase in the K m of ATP (10). These phenomena are consistent with the fact that P-site inhibitors bind most tightly to the Gs α- and forskolin-activated, pyrophosphate-bound complex of adenylyl cyclase (28). We suspect that the conformation of the ATP-bound enzyme differs from that of the 2′,d3′-AMP·pyrophosphate complex described here. Catalysis may require a conformational transition subsequent to ATP binding, resulting in a cAMP-bound conformation similar to that observed here in the presence of 2′,d3′-AMP. Such a transition would potentially involve residues such as Lys938 that form the C1·C2 interface. Thus, mutation of interfacial residues, which may hinder the conformational changes required for catalysis, necessarily reduces P-site binding to a much greater extent than ATP binding. From a kinetic rather than a structural point of view, any mutation that slows catalysis will impair accumulation of the enzyme-pyrophosphate complex, which is essential for observation of P-site inhibition (32).

Adenylyl cyclase catalytic mechanism. Analysis of the stereochemistry of the cyclase-catalyzed reaction, using diastereomers of adenosine-5′-(α-thio)triphosphate (ATPαS) or the corresponding guanine nucleotide, demonstrated that cyclization proceeds with an inversion of configuration at the α phosphate (34-37). A direct, in-line attack of the O3′ hydroxyl on the 5′-phosphate of ATP, without formation of a phospho-enzyme intermediate, most easily accounts for these results. To catalyze this reaction, the enzyme might be expected to provide a base to deprotonate the 3′-hydroxyl of ATP and a cationic residue or metal to stabilize a pentavalent phosphate intermediate.

From the above discussions, it is obvious that the binding site for ATP is formed by both domains and that the individually expressed cytosolic domains are unlikely to be catalytic, despite claims to the contrary (14, 38). A model of ATP bound to the active site can be constructed (Fig. 6A) in which the position of pyrophosphate·Mg2+ and a purine ring in the P-site complex is presumed to approximate that of the β and γ phosphates and the adenine of ATP. The conformation adopted by the α phosphate would allow in-line attack by the 3′ oxyanion with pyrophosphate as the leaving group. The Mg2+ coordination sphere would be similar to that observed in the P-site complex, with possible interaction of the α phosphate with Mg2+ in the transition state.

Figure 6

Model of ATP·Mg2+ bound to the catalytic site of VC1·IIC2 and of the mechanism for activation of the heterodimer by Gs α. (A) The catalytic site (50). The protein atomic coordinates are identical to those of the 2′,d3′-AMP·pyrophosphate·Mg2+·MPFsk complex. The coordinates of ATP were modeled into the site by superimposing its β- and γ-phosphate atoms on the phosphate atoms of pyrophosphate, and the purine ring on that of 2′,d3′-AMP. The conformations of the ribosyl C4′-C5′, C5′-O5′, and O–P bonds were adjusted to achieve a configuration in which the O3′ hydroxyl oxygen is poised for in-line attack on the 5′ α phosphate. The model of the unstimulated C1·C2 heterodimer, which was based on the structure of the IIC2 homodimer (14) [see (C) and text], is depicted as a transparent rose-colored ribbon with similarly colored transparent side chains. Thus, the transition from the transparent to solid-colored structures is indicative of two conformational changes: that from unactivated to Gs α-activated VC1·IIC2, and then from Gs α-activated to P-site–inhibited VC1·IIC2. In the catalytic site of the unstimulated heterodimer, residues involved in the binding of Mg2+ and pyrophosphate are shifted away from the adenine nucleotide. (B) Schematic of the putative transition state for the cyclization reaction, showing the predicted role of Arg1029 in stabilization of both the pentavalent phosphate intermediate and the leaving group (pyrophosphate), and the role of the two aspartate residues in coordination of Mg2+. (C) Model of the predicted conformational change that occurs upon activation of C1·C2 by Gs α. The VC1 domain in the inactive conformation is depicted with a transparent, rose-colored ribbon. The model was generated by first superimposing the IIC2 homodimer (14) on VC1·IIC2 by using only the “A” domain of the homodimer and the IIC2 domain of VC1·IIC2. The VC1 domain was then superimposed on the “B” domain of the superimposed IIC2homodimer. By binding between the α2′ and α3′ helices of C2, Gs α exerts pressure on the α1′-α2′ loop, which in turn presses against the C1domain. This results in a 7° rotation of the C1 domain with respect to the C2 domain that adjusts the positions of residues in the active site of adenylyl cyclase. ATP has been modeled in the active site as in (A).

The Arg1029 residue (α4′) is essential for catalytic activity (Table 2) and is poised to stabilize the pentacoordinate α phosphate intermediate of the reaction (Fig. 6B). This arginine may also stabilize the growing charge on the leaving β phosphate of pyrophosphate. However, mutation of Arg1029 does not perturb the K m for ATP (Table 2). Thus, Arg1029 is exactly analogous to Arg178 of Gi α 1 and Arg201 of Gs α, which stabilize the transition state of GTP hydrolysis in α subunits (19, 39), yet do not interact with GTP in the ground state (15, 19). The Arg1029 residue is, however, very important for P-site inhibition (Table 2) and directly binds pyrophosphate in the inhibited complex. The invariant Asn1025, which also appears to be essential for the catalytic mechanism (Table 2), seems important for the conformational change that occurs upon ligand binding (α1 helix contact) and may help bind ATP by coordinating a water molecule that forms a hydrogen bond with the N9 atom of ATP.

Our structures do not clearly indicate how the 3′-hydroxyl is deprotonated. The side chain of Asp440 is close to the ribose 3′-hydroxyl group and is therefore a potential catalytic base, although the basicity of this residue is presumably very low because of its participation in the ligand field of Mg2+. It is also feasible that the α phosphate serves as the general base; similar substrate-assisted mechanisms have been suggested for glutaminyl tRNA synthetase (40) and for p21ras (41). The environment in the vicinity of the ribose 3′-hydroxyl group is highly electronegative, with the preponderance of negative charges arising from Asp440, Asp396, and the α phosphate of the nucleotide. These groups would actually increase the pK a (where K a is the acid constant) of the 3′-hydroxyl, hindering deprotonation. Perhaps a second metal ion, not identified in the present study, binds to the ATP complex and polarizes the 3′-hydroxyl. The conformation of bound ATP, the active site, or both may also be significantly different than suggested by our model. This idea is supported by the fact that we failed to observe binding of Ap(CH2)pp in soaking experiments similar to those used for the P-site inhibitor.

It has been pointed out that the βαββαβ fold of adenylyl cyclase bears remarkable similarity to the catalytic scaffold of the DNA polymerase I palm domain (42). The chemical transformation catalyzed by each enzyme is also similar: intra- (cyclase) compared with inter- (polymerase) molecular attack of a 3′-hydroxyl on a 5′-phosphate. The resemblance between these two enzyme families extends to the position of two (cyclase) out of typically three (polymerase) acidic residues that are responsible for coordinating metal ions in the polymerases (43). The polymerases are proposed to bind two metal ions at the site of primer extension, whereas only one is observed in our P-site–inhibited complex. The proposed role of the first metal ion is to polarize the 3′-hydroxyl and stabilize the pentavalent intermediate. The role of the second ion is likewise to stabilize the transition state and to bind the polyphosphate tail of the nucleotide. In our structure, the observed magnesium ion is analogous to the second polymerase metal site. The role of the first metal ion could be partially assumed by the invariant Arg1029, which may stabilize the pentavalent phosphate transition state.

Our comments regarding the mechanism of the adenylyl cyclase–catalyzed reaction are speculative. The resolution to which the structure of the P-site complex is currently determined is not sufficient to define the position of potential catalytic residues with great precision. Further, we cannot rule out the possibility that additional metal ions remain undetected in the structure. Our failure, so far, to diffuse Ap(CH2)pp into crystals of Gs α·VC1·IIC2 may be indicative of potential conformational changes that accompany substrate binding. Thus, the P-site analog, presumably a good mimic of cAMP, may be a poor analog of ATP. Higher resolution studies of substrate (ATP analogs), product cAMP, and P-site inhibitors are required.

Mechanism of activation by G . There are several mechanisms by which Gsα could stimulate adenylyl cyclase. The first of these is to direct productive formation of the C1·C2 interface, a role that has been suggested for forskolin. However, as discussed with regard to forskolin above, there are a number of observations that suggest that Gsα acts primarily as an allosteric activator of adenylyl cyclase. Gsα interacts with both the C1 and C2 domains of adenylyl cyclase. Of the 1800 Å2 of surface area buried upon Gsα binding, 75% is with the C2domain and 25% is with the C1 domain. However, Gsα can bind to monomeric IIC2alone (disrupting IIC2 homodimers) (11), demonstrating that C1 is not required for binding. A C1 domain with phenylalanine at the position equivalent to 379 of VC1 is, however, important for activation of adenylyl cyclase (27). Furthermore, given the large surface area buried between C1 and C2, the reasonably large hydrophobic component of their interface, and the fact that they are covalently linked in the native protein, it seems unlikely that they are ever dissociated in vivo. The binding of Gsα to either the mixture of C1and C2 domains of adenylyl cyclase, or to fusion constructs of these, has little effect on the Km for ATP or the Ki for Ap(CH2)pp (6, 7,12). Gsα and forskolin bind synergistically (11) and thus stabilize the same form of the enzyme. Finally, together and separately, the two activators promote catalysis to a degree that exceeds their effect on domain association, and association of domains per se is an insufficient stimulus to catalysis. Therefore, Gsα, like forskolin, appears to activate adenylyl cyclase primarily by stabilizing a catalytically competent form of the enzyme.

We propose that the binding of Gs α to adenylyl cyclase induces a change in the relative orientation of the C1 and C2 domains that, in turn, primes the active site for catalysis. Because the structure of the VC1·IIC2 heterodimer in the absence of bound Gs α has not yet been determined, the conformational mechanism by which Gs αactivates adenylyl cyclase is not directly evident. However, the structure of the forskolin-bound C2 homodimer might approximate that of VC1·IIC2 in the absence of Gs α and allow us to postulate a mechanism. Several observations support this view. First, the regions of VC1 that form the interface with IIC2superimpose with their counterparts in IIC2 with a rmsd of 0.9 Å (114 Cα atom pairs). Despite the presence of residues that are not conserved between VC1 and IIC2 within the interface, superposition of VC1 on one of the subunits in the IIC2 homodimer reveals no significant steric or charge conflicts involving these residues. In addition, most of these nonconserved residues are themselves nonconserved in their respective C1 or C2 domains of other cyclase isoforms, suggesting that they play only minor roles in interface stabilization.

Comparison of the A subunit of the IIC2homodimer with that of IIC2 in the VC1·IIC2 heterodimer (difference distance plots) reveals that residues 900 to 913 (α1′-α2′ loop), 879 to 883 (NH2-terminus of β1′), and 1005 to 1020 (β4′-β5′) adopt different conformations from the remainder (the core) of the C2 domain in each structure. Omitting these elements, the two IIC2 molecules superimpose with a rmsd of 0.4 Å (127 Cα atom pairs), as opposed to 0.8 Å for the entire molecule (164 Cα atom pairs). Notably, these elements form many interfacial contacts with the analogous structural core of C1. The NH2-terminus and β4-β5 loop of C1, which form analogous contacts with the core of the C2 domain, are likewise expected to have conformational flexibility. C1and C2 thus embrace each other in flexible arms that allow the two domains to change their relative orientation, yet remain associated (Fig. 1C).

Gs α could induce domain reorientation by inserting its switch II helix into the cleft between the α1′-α2′ loop and the α3′ helix of IIC2. This intrusion widens the cleft by 3 Å at the outermost point (Fig. 6C). The α1′-α2′ loop rotates away from the core C2 domain and, through its extensive contacts with the β5-α4 region of the C1domain, forces the core of the C1 domain to rotate by 7° around an axis roughly parallel with the ventral cleft of the heterodimer. This rotation decreases the pseudo-twofold symmetry of the C1·C2 heterodimer. The NH2-terminus of VC1 is tethered between Gs α and IIC2; thus, this region may provide an anchor that allows Gs α to apply torque to the C1 core domain. In the active site of adenylyl cyclase, the result of this rotation is to move the β2- β3 loop toward and the β4 strand (carrying Arg484) away from the core of the C2 domain (Fig. 6A). Most notable in this activation model is the shift of P-loop residues at the NH2-terminus of α1 and of the β2-β3 loop relative to elements of the nucleotide binding site in β5′ and α4′. This movement would bring Asp440 closer to the 3′-hydroxyl group of the substrate. Mutation of the residue equivalent to Arg484 in VC1 eliminates activation by Gs α, highlighting the importance of this residue for the mechanism of Gs α activation (Table 2). Although this pyrophosphate-binding residue is not expected to be catalytic, its registration with other residues in the active site, in particular Arg1029 (α4′), may be essential for positioning Arg1029 to interact with the pentavalent transition state of the substrate (Fig. 6B). In these ways, Gs α can improve transition-state stabilization and potentiate a chemical step in the reaction mechanism. The structure of a C1·C2 heterodimer in the absence of Gs α, along with structures that display the position of substrate analogs, will provide much additional insight into the mechanism of activation of adenylyl cyclase by G protein α subunits.


View Abstract

Navigate This Article