Structure, Mechanism, and Regulation of Mammalian Adenylyl Cyclase*

The discovery of 39,59-cyclic adenosine monophosphate (cAMP) in the late 1950s by Sutherland and co-workers was the pivotal event that led to our current paradigm of hormone signaling through second messengers. Despite the subsequent discovery of many other second messengers, cAMP has never left center stage. The adenylyl cyclases are the family of enzymes that synthesize cAMP (1–5). Breakthroughs in determining the first structures of the mammalian adenylyl cyclase catalytic core (6, 7) provide a new context for understanding the action of many regulators, both physiological and pharmacological: free metal ions, P-site inhibitors, forskolin, G-proteins, Ca/calmodulin, and protein phosphorylation. Understanding the catalytic mechanism of an enzyme is a prerequisite to understanding its regulation. Here I will describe the essentials of catalysis and then consider how these elements are controlled by each of the major regulators.

The discovery of 3Ј,5Ј-cyclic adenosine monophosphate (cAMP) in the late 1950s by Sutherland and co-workers was the pivotal event that led to our current paradigm of hormone signaling through second messengers. Despite the subsequent discovery of many other second messengers, cAMP has never left center stage. The adenylyl cyclases are the family of enzymes that synthesize cAMP (1)(2)(3)(4)(5).
Breakthroughs in determining the first structures of the mammalian adenylyl cyclase catalytic core (6, 7) provide a new context for understanding the action of many regulators, both physiological and pharmacological: free metal ions, P-site inhibitors, forskolin, G-proteins, Ca 2ϩ /calmodulin, and protein phosphorylation. Understanding the catalytic mechanism of an enzyme is a prerequisite to understanding its regulation. Here I will describe the essentials of catalysis and then consider how these elements are controlled by each of the major regulators.

Structure of Adenylyl Cyclase
The nine cloned isoforms of mammalian adenylyl cyclase share a primary structure consisting of two transmembrane regions, M 1 and M 2 , and two cytoplasmic regions, C 1 and C 2 (8) (Fig. 1). The transmembrane regions each contain six predicted membrane-spanning helices. The function of M 1 and M 2 , aside from membrane localization, is unknown despite their topological analogy to transporters. The C 1 and C 2 regions are subdivided into C 1a and C 1b ; and C 2a and C 2b . The C 1a and C 2a are well conserved, homologous to each other, and contain all of the catalytic apparatus (9). C 1a and C 2a domains heterodimerize with each other in solution (10,11). These domains can also form homodimers. Domains derived from different isoforms can form chimeric heterodimers. The C 1b region is large (ϳ15 kDa), variable, and contains several regulatory sites. The C 2b is vanishingly short in some isoforms and lacks identified functions; hence C 2 and C 2a are sometimes referred to interchangeably.
The structure of the type II adenylyl cyclase C 2 region revealed a homodimer with two C 2 monomers in a wreath-like arrangement (6). A deep ventral groove runs between the two in the center of the wreath. Two forskolin molecules bind to this groove in the homodimer. The monomer is built around a large ␤ sheet that folds back onto itself on the "inside" facing the dimer interface. The "outside" is ␣-helical. A ϳ80-amino acid substructure within the monomer is similar to the palm domains of the DNA polymerase I and reverse transcriptase families (12,13).
The type V C 1a region and type II C 2 region arrange themselves in a heterodimeric wreath that is nearly identical in overall structure to the C 2 homodimer, with some critical differences in detail (7). The active site is at one end of the ventral groove. The single forskolin binding site, as anticipated by equilibrium binding (14), is at the other. The active site is formed at the interface by residues contributed by both C 1a and C 2 . Because the active site is shared between the two domains, association of two catalytic domains in the proper orientation is an absolute prerequisite of catalytic activity. The activity of mammalian adenylyl cyclases depends on the heterologous association of C 1a and C 2 . This is not the case for many other related cyclases (15,16). Mammalian membrane guanylyl cyclases and many microbial homologues of mammalian adenylyl cyclases are active as homodimers. The mammalian C 2 homodimer has measurable activity (9,17,18), although reduced by many orders of magnitude because of the loss of two catalytic Asp residues relative to the heterodimer.

Nucleotide Binding Site and Specificity
The ATP binding site has been revealed by the structure of the P-site inhibitor complex (7), molecular modeling (15), and mutagenic analysis (15, 19 -22). Lys-923 1 and Asp-1000 from C 2 interact directly with the N-1 and N-6 of the adenine ring. Gln-417 of C 1 plays a supporting role by orienting the Lys. Mutation of these three residues destroys the ATP versus GTP nucleotide specificity of adenylyl cyclase, although it does not convert it into a guanylyl cyclase (22). This is because of a main chain carbonyl that hydrogen bonds to the adenine N-6 and disfavors guanine. Guanylyl cyclases can be converted completely into adenylyl cyclases by mutating their guanine binding Glu and Cys to their adenylyl cyclase counterparts, Lys and Asp (21,22).
The ATP binding site is rounded out by hydrophobic residues that pack against the purine ring and by charged interactions with phosphate groups. Hydrophobic contacts are contributed mainly by C 2 . Charged interactions are formed by Arg (C 1 ) and Lys (C 2 ). The Lys is part of a flexible lid over the active site that is capable of undergoing an order-disorder transition (6,7).

Mg 2؉ Binding Site
The Mg 2ϩ binding site consists of two mutationally sensitive Asp residues (15,19). The P-site inhibitor complex shows a single Mg 2ϩ ion interacting with both phosphate moieties of pyrophosphate (7). There is abundant kinetic evidence for a two-ion mechanism (23,24). One ion acts kinetically as free Mg 2ϩ whereas the other binds as a complex with ATP. An analogous situation holds for the DNA polymerase family. The polymerases carry out the intermolecular attack of a primer 3Ј-hydroxyl on the ␣-phosphate of a deoxynucleotide rather than the intramolecular attack of a nucleotide 3Ј-hydroxyl on its own ␣-phosphate.
Two Mg 2ϩ ions bind to DNA polymerase-primer-templatenucleotide complexes (25)(26)(27)(28), providing a model for the coordination of two ions in the adenylyl cyclase active site. The polymerase "B" metal ion corresponds to the observed ion in the adenylyl cyclase complex. It binds all three nucleotide phosphates in a tridentate arrangement (25,27). These tight interactions leave little doubt that this is the ion that acts kinetically as an ATP complex. The polymerase "A" metal ion is less tightly bound as judged by higher temperature factors and fewer interactions with protein and nucleotide (25). It interacts with the 3Ј-hydroxyl of the primer and the nucleotide ␣-phosphate. Both metal ions are coordinated by both Asps. This coordination geometry fits the adenylyl cyclase structure consistent with known stereochemistry (29).

Mechanism of Cyclic AMP Formation
Cyclic AMP formation requires the deprotonation and activation of the ATP 3Ј-hydroxyl for nucleophilic attack; stabilization of the transition state at the ␣-phosphate; and stabilization of increased negative charge on the leaving group, pyrophosphate. Metal ion A activates the 3Ј-hydroxyl, and both metal ions share in transition state stabilization. Asn-1007, Arg-1011, and Lys-1047 approach the phosphate moieties (Fig.  2). Their modeled interactions with the non-bridging ␣-phosphate oxygens have poor geometry. It seems likely that at least one of these residues stabilizes the leaving group. Their precise positions in the ATP complex and, therefore, their precise roles in catalysis have yet to be determined. The fate of the proton on the 3Ј-hydroxyl is unknown. It has been suggested that Asp-354 in adenylyl cyclase or its counterpart in DNA polymerase could act as a general base in these reactions (15,26). Substrate-assisted catalysis is the other leading possibility suggested for base catalysis (7,20).

A Conformational Change That Controls Domain Orientation and Active Site Structure
The determinants of both nucleotide binding and catalysis are shared between C 1 and C 2 . This insight is central to understanding regulation of adenylyl cyclase catalytic activity. It means that any factor that alters the relative orientation of the C 1 and C 2 domains can alter the structure of the active site and thereby alter substrate affinity, catalytic velocity, or both.
The structure of the catalytic core is known in two conformations: that of the forskolin-bound homodimer (6) and that of the forskolin and G s ␣ 2 -bound heterodimer (7). The two structures differ by a 7 o rotation of the heterodimer C 1 domain relative to the correspondent C 2 in the homodimer. The domain rotation brings key catalytic elements from the two domains about 2 Å closer to each other. The structural differences between the two might be caused by differences in interface residues in the two different dimers, by occupancy of one versus two forskolin molecules, or, most probably, by the binding of G s ␣. Despite some uncertainties about which of these factors are driving the observed structural change, there is no doubt that the two boughs of the adenylyl cyclase wreath are capable of moving into more or less active conformations.

Regulation by Free Metal Ions
Mammalian adenylyl cyclases are strongly activated by Mn 2ϩ and inhibited by millimolar concentrations of free Ca 2ϩ . These effects are unlikely to have any physiological meaning or to reflect distinct binding sites for these ions. Many otherwise unrelated Mg 2ϩ -dependent enzymes can be activated by replacing Mg 2ϩ with Mn 2ϩ , probably because the latter is nearly the same size but is a stronger Lewis acid. At high concentrations free Ca 2ϩ binds competitively to Mg 2ϩ sites on enzymes but fails to replace it catalytically. A high affinity and possibly physiological inhibition of type V and VI adenylyl cyclase by free Ca 2ϩ has been reported (30), and the possibility of a high affinity binding site in the C 1b domain of these isoforms has been raised (31). Numbering is for type I. The crystal structure is shown for the type V C 1 (green)/type II C 2 (red) heterodimer bound to forskolin, adenosine 2Ј-deoxy-3Ј-monophosphate, pyrophosphate, and one Mg 2ϩ (G s ␣ is omitted). Membrane (cream) and C 1b (magenta) regions are drawn. Binding sites for G s ␣, G i ␣ (types V and VI), and G␤␥ (type II) are black, blue, and yellow, respectively. The Ca 2ϩ /calmodulin binding helix shown is for type I, and the protein kinase C phosphorylation site shown is for type II. Atom colors are: carbon, light gray; nitrogen, blue; oxygen, red; phosphorus, green; and magnesium, purple. The orientation of the catalytic core relative to the membrane is arbitrary in this view.

FIG. 2.
Model for the mechanism of adenylyl cyclase. Metal coordination by phosphates and Asp is derived from the T7 DNA polymerase structure (25), and the mechanism is adapted from that of Steitz (28). C 1 residues are green, and C 2 residues are red. Numbering is for type I. Interactions regulated by conformational changes are marked. Asp-310 moves about 1 Å between the two conformations, and Asp-354 and Gln-417 move about 2 Å. A hypothetical hydrogen bond between the putative base Asp-354 and the ATP 3Ј-hydroxyl is shown. Arg-1011 probably interacts with the ␣-␤ bridging oxygen of ATP, but its other interactions are less certain.

Regulation by P-site Inhibitors
P-site inhibitors are a class of nucleoside inhibitors of adenylyl cyclase so-called because they all contain a purine ring (32). The most potent lack a 2Ј-hydroxyl and are polyphosphorylated at the 3Ј-position (32)(33)(34). P-site inhibition is potentiated when adenylyl cyclase is activated. P-site inhibition is hypersensitive to certain mutations that slightly reduce enzyme activity. Psite inhibitors are non-or uncompetitive with respect to the forward reaction but compete with the product cyclic AMP in the reverse reaction (35). P-site inhibitors are effective against engineered ATP-specific guanylyl cyclases that are in all other ways regulated by different mechanisms than the adenylyl cyclases (22). P-site inhibitors bind to the active site primarily through conserved residues. Different adenylyl cyclases do show some differences in P-site inhibition, and a physiological role for this type of regulation has been postulated (36).

Regulation by Forskolin
Forskolin is a hydrophobic activator of all the mammalian adenylyl cyclases except type IX. It is an extremely powerful activator of some of the synthetic soluble adenylyl cyclase systems, increasing activity by up to 10 3 , although other forms of soluble adenylyl cyclase barely respond. Forskolin binds to the catalytic core at the opposite end of the same ventral cleft that contains the active site (7,15). It activates the enzyme by gluing together the two domains in the core using a combination of hydrophobic and hydrogen bonding interactions that are distributed equally between the two domains (6). Type IX adenylyl cyclase is non-responsive to forskolin because of a Ser 3 Ala and a Leu 3 Tyr change in the binding pocket. When these changes are reversed by site-directed mutagenesis, the resulting type IX mutant can be activated by forskolin as well as other adenylyl cyclases (37).
The forskolin binding pocket is a narrow hydrophobic crevice that almost completely buries the forskolin molecule once bound. The pocket residues are absolutely conserved in types I-VIII and differ only subtly in type IX. The presence of a hydrophobic crevice in a protein is highly destabilizing in the absence of bound ligand. It seems improbable that such a destabilizing feature would be so highly conserved if it had no function. This paradox led us to revive the idea that there exists an endogenous forskolin-like small molecule activator of adenylyl cyclase.

Regulation by G-protein Subunits
All mammalian adenylyl cyclases are potently and physiologically activated by the GTP-bound G-protein ␣-subunit G s ␣. This activation is synergistic, not competitive, with respect to forskolin. GTP-G s ␣ binds to a crevice on the outside of the wreath formed by ␣2Ј and ␣3Ј of C 2 and by the N-terminal portion of C 1 (7,38,39). GTP-G s ␣ is capable of gluing together C 1 and C 2 as does forskolin, but mutational analysis suggests this cannot be its only function. If the C 1 contact is abolished, activation can be partially rescued when forskolin is used to dimerize C 1 and C 2 . Therefore there must be a non-glue role for GTP-G s ␣ (38). This role is probably to induce a conformational change that allosterically stimulates catalysis. The 7 o rotation of C 1 , which moves the catalytic residues into their proper positions, is probably the result of a torque applied by G s ␣ as it "pushes" the C 1 away from its binding site (7). G i ␣ selectively inhibits adenylyl cyclase types V and VI. Symmetry and sequence homology arguments led to the suggestion that G i ␣ binds to the adenylyl cyclase catalytic core on a groove pseudosymmetrically related to the G s ␣ binding groove (7,38). Mutational analysis confirmed that the groove formed by ␣2 and ␣3 of C 1 is the primary site for binding of G i ␣ to type V (40). The inhibitory mechanism postulates a rotation of the C 1 in the opposite sense as that induced by G s ␣.
G␤␥ subunits conditionally regulate several adenylyl cyclases. Type II adenylyl cyclase is activated by G␤␥ when G s ␣ is bound. At least part of the G␤␥ binding site of type II has been located using peptide competition studies (41). The site spans a flexible loop between ␤3Ј and ␣3Ј and the first two-thirds of ␣3Ј (6). The G␤␥ site is adjacent to, but does not overlap, the G s ␣ site, consistent with conditional activation.
G-protein interactions with non-catalytic regions of adenylyl cyclase seem likely. The ␣4 -␤6 region of G s ␣ was predicted to interact with adenylyl cyclase based on mutagenic analysis (42,43), but no such contact with the catalytic domain was seen in the crystal structure. G␤␥ regulation of the soluble adenylyl cyclase model has not been established, even though the known binding site is located within the type II C 2 domain. C 1b (44), M 1 , or M 2 might be involved in either of these processes.

Regulation by Ca 2؉ /Calmodulin
Ca 2ϩ /calmodulin activates type I adenylyl cyclase by binding to a putative helical region on the C 1b (45,46). The precise activation mechanism is unknown. If other Ca 2ϩ /calmodulinactivated enzymes are a precedent, it is likely that Ca 2ϩ /calmodulin binding will disrupt an autoinhibitory interaction between the C 1a /C 2 catalytic core and sequences within the C 1b .

Regulation by Protein Phosphorylation
Protein kinase C activates type II adenylyl cyclase by phosphorylating it on Thr-1057 (47). This site is within a region known to be required for protein kinase C activation (48). This Thr is at the edge of the "lid," a flexible region that is disordered in the homodimer structure but folds over the top of the active site in the heterodimer-P-site complex. Phosphorylation might enhance the ability of the lid to adopt the correct conformation. CaM kinase II inhibits type III adenylyl cyclase by phosphorylating it at Ser-1076 (49). This Ser is at the outer lip of the active site, hence its phosphorylation could directly interfere with catalysis. Protein kinase A phosphorylates Ser-674 in the C 1b of type VI (50) and appears to regulate a low affinity secondary binding site for G s ␣. CaM kinase IV phosphorylates type I adenylyl cyclase in its C 1b domain and disables Ca 2ϩ / calmodulin activation by interfering with the calmodulin binding site (51).

Conclusions and Perspectives
A great deal of regulatory complexity has been layered onto the rather simple core structure of adenylyl cyclase. The core consists of two parts. The all important two metal ions bind to one part, C 1 . The nucleotide binding pocket and other catalytic residues are contributed primarily by the other part, C 2 . Both parts need to be aligned to carry out catalysis. The most potent activators of the broad range of mammalian adenylyl cyclases, G s ␣ and forskolin, bind to the domain interface and thereby control domain orientation in a powerful and direct manner. The small molecule forskolin binds on the inside, whereas the large protein activator binds to the outside of the wreath. More specialized regulatory sites have been added to the surface of the catalytic domain (G␤␥, protein kinase C) or appended to it (Ca 2ϩ /calmodulin, protein kinase A). The structural work reinforces the concept of mammalian adenylyl cyclases as sophisticated coincidence detectors and provides a new framework for a precise understanding of regulation.
Progress in understanding the structure and function of the C 1a and C 2 regions has not been matched by information on the rest of adenylyl cyclase. Proposed roles for the transmembrane segments M 1 and M 2 as a transporter or ion channel (8), membrane potential sensor (52, 53), or Ca 2ϩ channel interaction domain (54) have yet to be proved or conclusively disproved.
Resolution of this mystery would be a major advance. There are no structural data for C 1b , M 1 , or M 2 . For now we must live with a Who Framed Roger Rabbit picture in which we see the catalytic core in vivid three-dimensional "live action" whereas the remainder is just a cartoon.
What has the recent burst of progress in the structure and biochemistry of adenylyl cyclase contributed to the physiology of cAMP signaling? The possibility has been raised that there are endogenous forskolin-like or P-site inhibitor small molecule regulators. The locations of most key regulatory sites are now known. In many cases we can selectively alter the specificity of these sites, eliminate them, or even create them at will by site-directed mutagenesis. Transfection experiments with engineered cyclase isoforms should provide powerful new tools to determine in vivo how a particular regulator controls a particular adenylyl cyclase isoform in the context of many simultaneously active pathways.