Recent advances in the AC (adenylate cyclase)/cAMP field reveal overarching roles for the ACs. Whereas few processes are unaffected by cAMP in eukaryotes, ranging from the rapid modulation of ion channel kinetics to the slowest developmental effects, the large number of cellular processes modulated by only three intermediaries, i.e. PKA (protein kinase A), Epacs (exchange proteins directly activated by cAMP) and CNG (cyclic nucleotide-gated) channels, poses the question of how selectivity and fine control is achieved by cAMP. One answer rests on the number of differently regulated and distinctly expressed AC species. Specific ACs are implicated in processes such as insulin secretion, immunological responses, sino-atrial node pulsatility and memory formation, and specific ACs are linked with particular diseased conditions or predispositions, such as cystic fibrosis, Type 2 diabetes and dysrhythmias. However, much of the selectivity and control exerted by cAMP lies in the sophisticated properties of individual ACs, in terms of their coincident responsiveness, dynamic protein scaffolding and organization of cellular microassemblies. The ACs appear to be the centre of highly organized microdomains, where both cAMP and Ca2+, the other major influence on ACs, change in patterns quite discrete from the broad cellular milieu. How these microdomains are organized is beginning to become clear, so that ACs may now be viewed as fundamental signalling centres, whose properties exceed their production of cAMP. In the present review, we summarize how ACs are multiply regulated and the steps that are put in place to ensure discrimination in their signalling. This includes scaffolding of targets and modulators by the ACs and assembling of signalling nexuses in discrete cellular domains. We also stress how these assemblies are cell-specific, context-specific and dynamic, and may be best addressed by targeted biosensors. These perspectives on the organization of ACs uncover new strategies for intervention in systems mediated by cAMP, which promise far more informed specificity than traditional approaches.
- adenylate cyclase
- A-kinase-anchoring protein (AKAP)
- focal adhesion
- protein kinase
ACs CAN BE SUBJECT TO MULTIPLE REGULATORY INFLUENCES
The nine mammalian TmACs [transmembrane ACs (adenylate cyclases] are highly conserved in general structural terms, yet they vary considerably outside their catalytic and transmembrane domains (Figure 1), which allows them to display individual properties [1–4]. The complexity of their transmembrane architecture prompts comparisons with ion channels and membrane transporters , and, as with ion channels, this may also provide domains for interacting with other membrane proteins.
Early studies of overexpressed ACs, coupled with inferences based on structural homologies, led to a consensus view on specific AC regulation as presented in Table 1. Evidence accumulated over 20 years has indicated that that ACs are multiply responsive to, e.g., G-protein subunits (αs, αi and βγ), PKC (protein kinase C), PKA (protein kinase A) and tyrosine kinases [1,4]. However, the evidence as applied to individual ACs is often less than systematically established . A major goal of the present review is to make the case for determining how the specific cellular context and the cadre of the required regulatory elements will determine how specific ACs will be regulated. Assessing the ACs may be quite challenging in the native contexts in which they occur, as we develop below. By being multiply responsive to, e.g., βγ subunits of G-proteins, PKC and Ca2+, the ACs can interdigitate the activities of many cellular signalling pathways, such as those governed by ERK (extracellular-signal-regulated kinase), MAPK (mitogen-activated protein kinase) and β-arrestins [6–9]. Thus there is a clear need to organize these elements precisely in cells.
The best understood of these regulatory events, because of its dependence on in vivo responses, and the best characterized at the molecular, cellular and physiological levels, is the interaction between Ca2+ and ACs . This illustrates the organizational abilities of ACs.
REGULATION OF ACs BY CA2+ IS ELABORATE AND CRITICAL
The organizing powers of ACs are exemplified in their regulation by Ca2+, in which the concept of AC-recruitment of effectors clearly emerges along with the notion of AC-centred microdomains of cAMP and Ca2+. Exploration of the regulation of susceptible ACs by Ca2+ in vitro is relatively straightforward, although it does reveal features that are exploited in vivo.
Inhibition in vitro
AC5 and AC6 are inhibited directly by submicromolar concentrations of Ca2+ in a manner that exploits the octa-chelating properties of Ca2+ compared with the hexa-chelating powers of the far more abundant cellular cation Mg2+ . Mg2+ cannot simultaneously chelate both cAMP and pyrophosphate (PPi) (the inhibitory by-product of cAMP formation), so that PPi is released. However, in the presence of Ca2+, PPi can remain at the AC catalytic site and inhibit activity. The crystal structure reveals the precise interactions .
Stimulation in vitro
AC1 and AC8 are activated by Ca2+ acting via CaM (calmodulin). In vitro studies showed differences in the regulation by Ca2+/CaM of AC1 and AC8, which suggested distinct applications of these properties in intact cells. AC1 appears to be regulated by Ca2+/CaM binding directly to the C1b domain in a direct activation mechanism. The critical residues that bind CaM near the catalytic domain have been identified [15,16].
A more complex mechanism is used by AC8, which is reminiscent of the use of CaM by L-type Ca2+ channels [17,18]. CaM is bound tightly by a classic 1-5-8-14 loop at the N-terminus of AC8. Nevertheless, this is not where CaM acts to stimulate activity; this site appears to garner limited CaM from its other competitors in the cell . In fact, the N-terminal binding site can be deleted or mutated and the enzyme remains fully stimulable in vitro. The vital role of the N-terminal binding site becomes apparent in vivo (see below). Upon binding Ca2+ by the C-terminal EF-hand pair on CaM, a conformational state is induced that allows the transition of the CaM to an IQ-like domain at the C-terminus of AC8 and a disinhibitory activation ensues [16,20–24]. The use of shuttling between a classic 1-5-8-14 motif and a ‘Ca2+-independent’ IQ motif is mirrored in L-type Ca2+ channels .
Regulation in vivo
The ACs that are regulated by Ca2+ in vitro (AC1, AC5, AC6 and AC8) are regulated by physiological transitions in Ca2+. In non-excitable cells, this occurs as a consequence of SOCE (store-operated Ca2+ entry)  either triggered by physiological means, such as agonist-mediated depletion of inositol 1,4,5-trisphosphate-sensitive stores, pharmacologically, as a result of passive store depletion by inhibition of Ca2+-reuptake (e.g. by thapsigargin), or indirectly, as a result of the depletion of stores by ionophore-driven emptying . Remarkably, neither release of Ca2+ from stores nor the very large accumulation of Ca2+ entry triggered by high concentration of ionophore can regulate these enzymes. Thus the efficacy of Ca2+ is not due to the global cytosolic concentration that is achieved, but to the mechanism by which Ca2+ is elevated. The underlying mechanism in the case of AC8 relies on a direct interaction between the AC and the channel element of SOCE channels, Orai1 (see below) (Figure 2). Strong circumstantial evidence suggests that the same dependence applies to AC1 in blowfly, pancreatic MIN6 cells  and the sino-atrial node [28,29]. AC5 and AC6 which show the same requirement for Ca2+ entry over other modes of Ca2+ elevation might be expected to show similar interactions. Another channel, the cGMP-stimulated wild-type rat olfactory CNG (cyclic nucleotide-gated) channel, when expressed by adenovirus, can mediate inhibition of the endogenous AC6 of C6-2B cells .
ACs are regulated by L-type Ca2+ channels in excitable cells, both when exogenously and when endogenously expressed. For instance, AC8 when expressed in GH3 and pancreatic MIN6 cells is stimulated [29,31], the endogenous AC5/AC6 in cardiomyocytes is inhibited, and the endogenous AC1/AC8 in hippocampal slices is stimulated [32,33]. The issue of the selectivity of ACs in excitable cells for VGCCs (voltage-gated Ca2+ channels) over other forms of Ca2+ entry has not received the same exploration as has SOCE in non-excitable cells. Nevertheless, in MIN6 cells, Kitaguchi et al.  showed a regulation of the endogenous AC1 by L-type Ca2+ channels, but not by SOCE or receptor-induced release from stores. It has been suggested that the ACs may form a complex with L-type channel elements and, although the evidence is not as direct as in the case of the AC8–Orai1 complex, ACs and channels do bind shared elements, e.g. AKAPs (A-kinase-anchoring proteins) (see below), so this is quite a viable possibility .
Regulation by the emptying of stores
AC activity has been proposed to be stimulated by depletion of Ca2+ stores in a manner that is independent of cytosolic Ca2+ levels, but responsive to the state of filling of the stores . The AC species involved has been proposed to be AC3 . There is no suggestion that the effect involves any regulation by direct effects of Ca2+; indeed, in these experiments, Ca2+ entry appears to inhibit the AC3, but the effect is some indirect consequence of store depletion. Given the major cellular consequences that occur upon Ca2+ store depletion, including cytoskeletal movements and ER (endoplasmic reticulum) reorganization , it may not be too surprising that the trafficking of membrane-directed proteins (such as AC) might be altered, so that new activity becomes measurable.
Insensitivity to ionophore
The independence of Ca2+-sensitive ACs from ‘trivial’ forms of Ca2+ rise is a continuing puzzle. Certainly, a major factor must be the extreme closeness of the ACs to Ca2+ channels. But the ineffectiveness of IM (ionomycin)-mediated Ca2+ elevation (when IM-triggered store-depletion is taken into account) is an abiding conundrum. One possibility is that the elevation in cytosolic Ca2+ achieved by IM does not occur in the domain of the ACs. For instance, IM is reported not to dissolve well in lipid membranes that contain cholesterol , so that the presence of the Ca2+-sensitive ACs in cholesterol-rich rafts may distance them from IM-induced transitions in Ca2+. Another possibility is that the IM-induced Ca2+ rise is precluded from reaching the AC by the presence of Ca2+ buffers or Ca2+ pumps in the vicinity of the AC, which mutes the Ca2+ rise. When Ca2+ was measured in the vicinity of an AC using an attached fluorescent protein sensor for Ca2+, GCaMP [39,40], a very small IM component, as well as a SOCE component, was detectable in response to IM [40,41], although this did not compare with the Ca2+ rise reported by a cytosolic sensor . So it may well be that insufficient Ca2+reaches the vicinity of the AC, and the impression conveyed by a cytosolic sensor exaggerates the Ca2+ in the AC domain. Another consideration is that the GCaMP only measures Ca2+; this leaves open the possibility that the AC8–GCaMP sensor ‘sees’ the Ca2+, but the AC does not respond. This option raises the possibility that the nature of the Ca2+ rise achieved by ionophore is different from that achieved by a channel. Clearly ionophores facilitate the transport of ions across membranes in a continuous, concentration-driven, manner, whereas channels open and close and yield rapid rises and falls in very high (up to 100 μM) ion concentration in the immediate vicinity of the channel mouths . Thus it is conceivable that the ACs respond to these increments rather than to the absolute amounts of Ca2+ achieved. An analogous situation has been nicely dissected in the case of the control of sperm turning by Ca2+entry via CNG channels. It turns out that, by modelling the rate of change, not the absolute amount of Ca2+, it was possible to relate how changes in Ca2+ concentration controlled turning. Alvarez et al.  showed that the sperm responded to a time derivative of [Ca2+]i which they proposed could be based, for instance, on the different time constants and co-operative Ca2+ binding to the EF-hands of CaM. This situation could readily apply to the Ca2+/CaM-regulation of AC1 and AC8. Detailed analysis of the effects of the relevant CaM-binding Nt (N-terminal), Ct (C-terminal) or C1b peptides from AC8 and AC1 respectively show quite different effects on Ca2+-binding kinetics by the N- and C-lobes of CaM , which could readily introduce a ‘chemical differentiation’ step  that could allow ACs to respond to rapid Ca2+ gradients. The N-lobe of the CaM has extremely high affinity for Ca2+ when bound to the N-terminus of AC8, but this is reduced by two orders of magnitude upon its transition to the C-terminus . Thus both a highly discerning affinity for very high Ca2+ concentrations is built in, as are found near the mouths of ion channels, but not at ionophores, as well as a temporal differentiation step due to the transition of the CaM to the C-terminus, resulting in a reduced affinity for Ca2+, which delays the process and allows a temporal differentiation step to be present (see also ).
ACs ARE SCAFFOLDS
ACs are much more than producers of cAMP. Table 2 lists proteins that have been demonstrated to interact with ACs; the evidence varies from extremely robust to suggestive, but it is clear that a wide variety of proteins can be involved. The best-represented group are the AKAPs, which are bound by every AC species . The discovery of AKAPs as devices that minimally recruit PKA, a major target for cAMP, along with, often, an effector of cAMP, as well as a cellular targeting domain was a major conceptual advance in imparting great efficiency to cAMP signalling by minimizing diffusion distances for interacting components [45,46]. The fact that all of the ACs bind some selection of AKAPs allows even more immediacy in co-localizing the source and the target of cAMP. Well-established examples of AC–AKAP complexes include AC5/AC6–AKAP79–L-type channel–PDE4D3 (PDE is phosphodiesterase) , AC8–AKAP79–PDE4D3  and AC5–mAKAP (muscle-specific AKAP) in cardiomyocytes . However AC–AKAP complexes are more than just co-localized interacting assemblies of AC–PKA and PKA targets; they are also devices for integrating with other signalling systems. For instance, AKAP79 binds PKC as well as PKA, and this allows AC2 to be regulated by the PKC pathway, as stimulated by muscarinic cholinergic receptors . AC9 through binding the AKAP Yotiao permits the regulation of the IKs (slow outward K+ current) by both calcineurin and PKA in cardiac myocytes . In another example, the relaxin RXFP1 receptor recruits an AC2–AKAP79–PDE4D3 complex which associates with β-arrestin signalling, allowing a novel pathway to be activated . AC–AKAP complexes can also act as bridges to effectively permit the AC recruitment of other targets of PKA such as L-type channels  and K+ channels  to yield other versions of AC-centred complexes.
The interaction between the ACs and the AKAPs can be expected to be dynamic and regulated. Indeed, the PKA bound to the AKAP79–AC8 complex phosphorylates the N-terminus of AC8 and reduces its responsiveness to Ca2+ rises, which may provide a device for a feedback inhibition or the generation of cAMP oscillations, depending on the context .
ACs bind scaffolds other than AKAPs, such as PP2A (protein phosphatase 2A), a complex multiply interacting scaffold, which organizes numerous pathways [71,72]. In any proteomic analysis or yeast two-hybrid analysis of putative AC-interacting proteins, cytoskeletal elements are generally encountered ([71,72], and K.L. Everett, A. Schotten and D.M.F. Cooper, unpublished work). Such interactions would be expected from either stable interactions or trafficking involving the cytoskeleton (see below).
The ACs are encountered in regulatory complexes that include not only the downstream targets of cAMP, such as voltage-gated Ca2+ channels and K+ channels, but also regulators of AC such as Orai1 (the likely basis for the dependence of AC8 on SOCE) and L-type channels which can feedback inhibit (or stimulate) the ACs . AC8 binds Orai1 directly via their respective N-termini, as established by peptide array and FRET analysis of truncated and mutated forms, co-immunoprecipitations and functional assessments of Ca2+ and cAMP in the domain of the complex  (Figure 2). AC5/AC6 may bind L-type channels indirectly by sharing AKAP79 .
Two extensive reviews [3,75] summarized the consequences of AC-knockout and -overexpression studies which implicate ACs in processes as diverse as learning and memory (AC1/AC8), olfaction, nociception (AC3), control of movement (AC5) and cardiac output (AC5/AC6). However, within the context of the present review, one might expect difficulties to arise in distinguishing the importance of the loss of the ability to produce cAMP from the loss of AC scaffolds. The separable roles of AC compared with cAMP in cardiac physiology in terms of its effects on phospholamban phosphorylation, sarcoplasmic reticulum Ca2+ pump activity and L-type Ca2+ channel activity is illuminated by a series of reports on the effect of the knockout of AC6 and the knockin of a catalytically dead AC6 [76,77]. In AC6 knockouts, Ca2+ handling is enhanced in the left ventricle despite the absence of the cAMP that is normally considered to regulate the relevant proteins. However, cardiac function is preserved in knockin AC-dead mice regardless of the fact that the production of cAMP is reduced. The complete deletion of AC6 severely impairs cardiac function . This draws attention to the likely role of the AC as a scaffold. The proteins involved and the mechanisms are still not understood.
The latter example highlights the inevitable consequences of knocking out a scaffold rather than a simple source of cAMP. Although the first knockouts of AC1 and AC8 were enormously insightful in establishing the roles of these ACs in learning and memory , some questions might now be posed on whether there had been any consequences of the mislocation of the proteins scaffolded by the ACs. Is it possible that the scaffolded proteins had complicating or compensatory actions in the development of those mice, or recruited/scaffolded proteins might be protected from degradation so that the absence of the scaffold would shorten their half-life, dampening any consequences of mislocation. The ideal approach seems to be to knock back in ACs with targeted deletions or mutations, notwithstanding that overexpression studies of dead enzymes may also have unforeseen consequences. For instance, in the study cited above, the mutant AC6 is apparently expressed at the nuclear membrane (by immunohistochemistry ) and it is possible that other ACs can be up-regulated so that the cAMP levels are unimpaired. Nevertheless, the role of ACs as scaffolds is an unavoidable experimental consideration.
Another significant issue, which is apparent from Table 2, is that the ACs and their binding partners have binding sites for multiple partners, but presumably not at the same time, given that some of the binding regions overlap. Consequently, dynamic binding and release events should be expected. For instance, in AC8, the Nt-binding domain for CaM is the same domain as for AKAP79 and Orai1, as established by extensive mutagenesis, peptide array, functional and FRET studies [22,41,48]. The Orai1-binding domain for AC8 can also bind phosphatidylinositol 4,5-bisphosphate and be phosphorylated by PKC , so it can be anticipated that access to the various targets is regulated. Thus exchange of partners in different settings should be anticipated. In addition, there is a short form of Orai1  (Orai1b) with an alternative start site which lacks the first 60 residues of Orai1a, rendering it incapable of interacting with, or regulating, AC8, so that the relative abundance of these forms will determine which is in complex with AC8. These complex issues underline the challenges and opportunities now presented by ACs.
DYNAMIC CELLULAR ORGANIZATION OF ACs
The foregoing discussion has made the case for considering the ACs as scaffolds, which group responsive proteins in compact arrays. The AC scaffolds are also organized in a broader cellular context, so that discrete lipid and cytoskeletal interactions must be considered. Lipid rafts are seen as cellular devices that can concentrate and promote interactions between signalling targets and effectors . These rafts are domains of the plasma membrane that are rich in cholesterol, sphingolipids and unsaturated side-chain phospholipids, a subset of which are caveolae, which are identifiable structures particularly prominent in endothelial cells . The concept of lipid rafts has evolved significantly with the development of super-resolution microscopic and other live-cell methods and a growing awareness of the heterogeneity and dynamism of lipid elements within the plasma membrane . Consequently, whereas numerous proteins have been found in biochemically prepared rafts by proteomic analyses, individual rafts are seen as often transient ‘nanodomains’ with sizes estimated between 5 and 300 nm, which can accommodate 10–30 protein molecules [84–87]. Some ACs (AC1/AC3/AC5/AC6/AC8) are consistently associated with such rafts and others (AC2/AC4/AC7/AC9) are not [88–90]. The evidence gathered to support this selection ranges from crude biochemical cellular fractionation to more elaborate live-cell methods [69,90,91]. The cytosolic regions appear to be responsible for the residence in rafts of AC5 and AC6 [92,93] which suggests that protein–protein interactions might underlie these associations. It is also interesting that AKAP79 can be palmitoylated and as a consequence occur in rafts along with AC8 [94,95]. Orai1 and TRPC1 (transient receptor potential canonical 1), channel components of SOCE, are also raft-resident proteins [96,97], as are CNG channels  and L-type channels . Identification of the precise targeting mechanisms of proteins to rafts is rendered difficult by the fact that the forces involved in sustaining raft residence must be a summation of low-affinity interactions with cholesterol and other cholesterol-binding or palmitoylated protein partners.
Rafts are more dynamic than simple platforms for signalling molecules. Besides diffusing laterally in the bilayer, rafts also undergo intracellular assembly and trafficking to and from the plasma membrane as well as endocytosis and inward trafficking to an array of subcellular compartments [100–102]. Movement of these nanoplatforms is dictated in part by their sorting signals and maintained through the integrity of various cytoskeletal structures [103–106]. The involvement of cytoskeletal elements in trafficking of rafts to the plasma membrane  make it unsurprising that proteins such as actin, actinin, filamin, paxilin, vinculin, myosin and raft markers (e.g. caveolin) are not uncommonly associated with ACs by a range of direct co-immunoprecipitation, proteomic or live-cell approaches [65,69,90,108,109]. The associations of ACs with these cellular features can cloud interpretations of specific regulatory interactions from more generic trafficking interactions. Doubtless trafficking is specific and likely to be individual for every protein; nevertheless the overall trafficking process, as might be perturbed by, e.g., brefeldin, is not centred on specific proteins, such as the ACs. Yet specific interactions do seem to occur between actin and ACs and it can be difficult to separate a vital role in permitting the precise cellular placement of ACs from a general role in trafficking and recycling ACs between various destinations [69,90].
Ric8A, which was originally proposed as a guanine-nucleotide-exchange factor, bound to the N-terminus of AC5 and inhibited its activity . This may now be a good example of a trafficking rather than a regulatory cofactor. Ric8a has been mooted as a molecular chaperone for G-proteins  and so may simply interfere with the PM (plasma membrane) delivery of AC5 or its components. Another protein, snapin, was one of the first proteins shown to interact with AC6  and was originally posited as an element of the SNARE (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor) complex. This protein is now viewed to have a more widespread distribution and to be more widely involved in vesicular trafficking .
A related problem arises with caveolin. Caveolin has been considered as a specific regulator of AC5  and, indeed, a peptide derived from the putative AC5-binding sequence of caveolin potently inhibited AC5 activity, which led to the proposal that caveolin exerted a specific inhibitory regulatory influence on AC5 . Developments in eNOS (endothelial nitric oxide synthase) research have cast a different light on this issue. eNOS has been shown to bind caveolin, to be inhibited by caveolin, and this inhibition is relieved and antagonized by Ca2+/CaM binding to a domain on eNOS. Consequently, a rather elegant mechanism was suggested involving activation/inactivation cycles of caveolin binding/unbinding that were controlled by transitions in Ca2+ . Recently, it has been demonstrated that in the mature eNOS structure, the caveolin-binding domain of eNOS is quite inaccessible to caveolin because of the three-dimensional conformation adopted by the active form of eNOS [100,113]. It turns out that many caveolin-binding domains are similarly inaccessible in mature proteins . Consequently, the previous mechanism for eNOS is suggested to be inadequate. A compromise might be that where, as part of the undoubted trafficking and selective residence of eNOS to caveolae, eNOS, although initially unfolded, does form a tight association with caveolin through this binding site; this binding to caveolin ensures its trafficking to caveolae and, upon elevation of Ca2+ levels, caveolin is released and eNOS adopts its active conformation permanently in a manner no longer subject to caveolin. A similar situation may apply to ACs where trafficking to caveolae in endothelial cells or cardiomyocytes requires an explicit (and inactivating) association with caveolin which is relieved upon the delivery of the ACs to its appropriate environment [88,115].
Biogenesis, delivery and endocytosis of ACs is a topic that has not been explicitly explored to date. We should expect that the targeting of Ca2+-regulated ACs to cholesterol-rich domains will rely on specific sorting signals, interactions with cholesterol, cytoskeletal elements, other raft-targeted proteins and possibly association with chaperones (including caveolin [69,91]). A summary of these interactions is presented in Figure 3, from which it can be observed that disruption of these essential processes at any step can ultimately have consequences for overall AC localization, association with lipid rafts and AC activity.
The foregoing discussion proposed that AC scaffolds are organized in their cellular destination as part of a regulated, but continuous, process. A more dynamic situation may occur at FAs (focal adhesions), where a scaffolding AC could play a key regulatory and dynamic structural role. FAs are large macromolecular assemblies through which both mechanical force and regulatory signals are transmitted . They link the cytoskeleton to the ECM (extracellular matrix) and are considered as subcellular macromolecular signalling hubs that mediate the regulatory effects of ECM adhesion on cell movement. As well as ERK signalling processes, both Ca2+ and cAMP play major roles in FA dynamics [119,120]. Transiently forming rafts are also part of the FA cycle of release and attachment to the cellular substrata . The Ca2+-inhibitable AC6 could be poised to play an important role in such situations given that high cAMP promotes actin depolymerization and low cAMP allows actin to stay polymerized and maintain contact with integrins at the FA. Thus AC6, which binds actin at FAs , is susceptible to the influx of Ca2+ to keep cAMP levels low. AKAP220 has already been reported at FAs , as have the elements of SOCE  in addition to their previously known raft associations [93,124]. Thus it seems a small step to propose that a raft-localized Ca2+-inhibitable AC scaffolding a palmitoylated AKAP and subject to inhibition by a raft-localized SOCE complex could play a key role in maintaining FAs. Removal of the Ca2+-elevating stimulus would allow cAMP to rise, the actin to depolymerize, the FA to collapse and cell migration to resume (Figure 4).
The foregoing discussion aimed to convey that the interactions of ACs will be determined by context, that the organization of ACs in the cell will be dynamic and can be expected to change throughout the behavioural activities of the cell . The ACs may act as foci for other regulatory molecules, which have their own coterie of associated proteins, and the ACs may interact with the cyto-architectural elements, so that they can be viewed as sculptors of their own domains both by recruiting proteins and by producing cAMP, which affects actin polymerization.
ACs MAY BE INTERNALIZED AND SIGNAL INSIDE CELLS
The organization of ACs at the PM, coupled with a proposed firewall of cytosolic PDEs has sometimes prompted speculation as to how cAMP can reach different internal areas of the cell. It has been clearly argued  that the placement of PDEs will have a major influence on where cAMP may, or may not, reach. Tethered internal sensors for cAMP do rapidly report cAMP in response to external stimuli so that, apart from simple diffusion, the formal possibility can be considered that there may be internal sources of cAMP [apart from sAC (soluble AC), which is mentioned below]. Against this background, a body of intriguing and provocative papers have suggested that ACs are internalized and can be a source of internal cAMP. The typical experiment traces fluorescently tagged receptors, G-proteins and perhaps BODIPY–forskolin binding (as a surrogate for the catalytic unit of ACs). For instance, Calebiro et al.  presented an elegant study showing that, whereas short stimulation by TSH (thyroid-stimulating hormone) of thyroid follicles showed transient increments in cAMP, a longer treatment, e.g. 30 min, resulted in a sustained elevation in cAMP. Associated with this sustained elevation was a sustained internalization of the TSH receptor, Gsα and even AC3, as well as BODIPY–forskolin. Although it is advisable to be mindful of the likely receptor and Gs internalization that always accompanies prolonged receptor stimulation, as well the functionality of the tools, such as the ability of a putative anti-AC3 antibody to report the likely endogenous levels of AC3 in cells, nevertheless the most attractive interpretation of such findings amounts to a major additional signalling option for ACs. A different interpretation of these results comes from Neumann et al.  who suggested that inhibition of TSH receptor internalization by dynamin inhibition did not affect persistent cAMP accumulation, and concluded that TSH remaining at the cell surface was responsible even for the prolonged effects. A possible explanation builds on the longstanding observation that extended receptor activation is accompanied by coupling of β-arrestins to receptors, which leads to a switch in coupling from Gs to Gi, with consequent activation of the ERK pathway ; the internalized β-arrestin activates ERK which in turn can lead to phosphorylation and inhibition of the activity of long PDE isoforms . Thus ERK activation would result in increased cAMP levels independently of acute AC stimulation. Such is the interpretation of the effects of low- compared with high-dose isoprenaline (isoproterenol) on acute or sustained cAMP accumulation in cardiac myocytes . Thus, although the possibility that ACs might continue to signal even when (and if) internalized remains an attractive hypothesis deserving continuing reassessment, alternative interpretations of the data such as normal recycling of ACs to and from the PM (Figure 3) or β-arrestin effects on PDEs, along with some reservations on some of the reagents, dilutes enthusiasm for embracing internal ACs as a significant cAMP signalling option at present.
An alternative source of internal cAMP can be the sAC which possesses no transmembrane segments. First discovered in high concentrations in sea urchin sperm and responsive to pH , it was later found to be widespread in mammalian sperm, where, as part of the acrosomal reaction, it reacted with bicarbonate . It is not stimulated by forskolin or G-proteins, and, for activity to be measured in vitro, high concentrations (2–5 mM) of either Mn2+ or Mg2+ and high millimolar concentrations of ATP are required (Km ~2 mM compared with 0.2 mM for TmACs). This requirement for millimolar concentrations of ATP has been taken to suggest that sAC may act as a sensor of ATP . For instance, when cells overexpressing sAC are poisoned with mitochondrial toxins such as azide, the cAMP levels fall along with the cellular ATP. Even with mitochondrial poisons, cellular ATP levels would not approach limiting concentrations for the TmACs, but, for sAC, because of its high Km when poisoned, cAMP levels due to overexpressed sAC fall, whereas those stimulated by forskolin (i.e. TmACs) are unaffected.
From targeted disruption experiments, there seems little doubt that sAC plays a vital role in sperm motility , but the role of sAC in tissues where it is expressed at low levels is a difficult experimental challenge. An observation that can be made is that the classic inhibitor of TmACs, 2′,5′-dideoxyadenosine, can only ever inhibit cAMP accumulation by ~95%. (In the context of the present review, it is not surprising that compartmentalized ACs may not be fully accessible to inhibitors, and inhibitors can rarely be expected to reach intracellular concentrations where they can achieve 100% inhibition.) Under such circumstances, the putative sAC inhibitor KH-7 can be shown to inhibit the residual cAMP accumulation and some downstream events, thus the presence and involvement of sAC can be proposed. A potential problem in such experiments is that KH-7 also inhibits glucose oxidation and so leads to decreased ATP levels , so it is difficult to separate the role for a sAC that depends on very high ATP levels from a process that depends on high ATP levels.
AC-CENTRED MICRODOMAINS AND KINETICS OF cAMP AND CA2+ ARE REVEALED WITH TARGETED SENSORS
The foregoing discussion of AC-based signalling hubs implies that cAMP dynamics might be expected to vary in different AC microdomains. Support for such concepts required the development of live-cell sensors for cAMP. Traditional methods that measured the cAMP produced by a multitude of cells at a single time point constrained thoughts on this messenger to the conceptual doldrums of no spatial and limited temporal resolution. Following the clever, but technically demanding, efforts of patch-cramming of CNG channels  and the microinjection of chemically fluorescently labelled PKA subunits , the leap to genetically encoded sensors brought temporal and spatial awareness of intracellular cAMP as a signal to the same level of sophisticated insight and framing of hypotheses as had been the case for Ca2+. The development of FRET-based cAMP sensors built around Epacs (exchange proteins directly activated by cAMP) allowed the single polypeptide to be targeted to discrete domains of the cell by the use of PM-targeting sequences derived from, e.g., Lyn kinase, or directly attached to ACs, to sample cAMP in the domain from which it emanated or to which it might diffuse from other sources [140–143]. (The placing of a substantial Epac tag on either the N- or C-terminus of AC8 has no deleterious effect on AC responsiveness either in vitro or in vivo.) Although compartmentalization had been proposed as the only likely explanation for phenomena such as the fact that two agents which caused equivalent elevations in cellular cAMP did not cause equivalent physiological outcomes , the concept was not universally embraced, given the prevailing ignorance at the time of how cells, and ACs, were organized. Twenty years ago, the apparent differences in diffusion between Ca2+and cAMP  were invoked to explain presumed targeted signalling by Ca2+, but not by cAMP. With the growing realization of mechanisms such as PDEs to limit cAMP diffusion , the organization by scaffolding of cAMP targets, combined with single-cell imaging of dynamic cAMP kinetics, these inevitably naïve discussions have become moot. At the same time, it remains critical to acknowledge that any measurements (and concepts derived therefrom) do reflect the measuring methodology rather than the absolute entity being measured. In the case of Ca2+, the ready acknowledgement was instructive that, since fluorescent derivatives of EGTA were the basis for quin-2, fura-2 and indo-1 measurements, the apparent cellular dynamics of Ca2+ must reflect the presence of the fluorescent Ca2+ reporter, its concentration and its ability to bind and release Ca2+ . Probe concentration and kinetics have profound implications for the manifestation of Ca2+ blips, puffs and sparks. Analogous reservations should attend the measurement of cAMP.
Microdomains of cAMP have now been demonstrated using targeted sensors such as PKA, Epac and CNG channels [142,147,148]. Even more striking are the microdomains of both cAMP and Ca2+ that are manifest when sensors are targeted directly to the ACs, which reveal unique signals surrounding differently regulated ACs [41,143] (Figure 5). Targeted sensors are obviously the new key to exploring these domains. More than one AC is expressed in cells, with different responsive properties; hence microdomains can be expected to be the key level of organization for ACs, since their targets may also be extremely local. The likely local or targeted signalling by cAMP implies that a global elevation of cAMP or a generic inhibition of AC may give quite unintended and uninsightful consequences. Coincident with this reservation is the confusion attending elevation of cAMP by forskolin (which can activate every AC in a cell) the use of cAMP analogues with unknown permeation properties or PDE inhibitors that globally elevate cAMP to mimic the effects of hormones.
Targeted cAMP and Ca2+ sensors revealed that, in fact, the ACs detect selective changes not only in cAMP in their domains in response to different signals, but also in Ca2+, that are quite distinct from the dynamics in the global cytosol [40,73] (Figure 6). A particular power of AC-tagged sensors could be to probe the composition of AC microdomains, or the proteins which affect cAMP or Ca2+ in those domains. For instance, peptide or small-molecule disruptors of scaffolding interactions or AC channel interactions, or siRNAs against components, could all be usefully assessed using targeted compared with global sensors of the second messengers.
AC-CENTRED MICRODOMAINS ALLOW cAMP LEVELS TO OSCILLATE AND THE OSCILLATIONS CAN BE DECODED BY PKA
The information content provided to Ca2+ signalling by the manifestation and development of Ca2+ oscillations is widely acknowledged, particularly in the context of CaMKIII (Ca2+/CaM kinase III) [149–151] and in the different transcriptional regulation afforded by distinct temporal patterns of Ca2+ rises . However, whether the opportunities offered by digital frequency-encoded signalling could apply to cAMP has, until recently, been more theoretical. It had been predicted that cAMP levels could oscillate as a function of feedback loops between ACs and Ca2+ channels under their control . Indeed, over 40 years ago, Brooker  had shown that cAMP levels oscillated in the beating frog heart. That system had the advantage of a synchronized organ and the ability to freeze and extract cAMP at different parts of the contractile cycle. Clearly, in independent cells, the diffusion of cAMP away from its site of synthesis along with the influences that give rise to oscillations are unlikely to sustain the fidelity of cAMP oscillations distant from the enzyme source, a consideration which draws attention to the potential added value of localized cAMP microdomains. Using live-cell sensors, cAMP oscillations have been shown to be inducible artificially, or to occur spontaneously. [154–157]. In addition, the absolute need for these oscillations as distinct from a simple rise in cAMP has been demonstrated in the case of neuronal path-finding. Gaspar and colleagues’ exuberant axon pathfinding paradigm is an explicit example of the temporal dynamics of localized cAMP signalling . In that system, pulses of cAMP determined by the entry of Ca2+ via L-type channels or released by pulsatile uncaging exerted regulatory effects, but global elevation of cAMP or inhibition of PDE induced spatially uninformative cAMP rises of no regulatory consequence . Those effects were apparently mediated by AC1 and were not demonstrable in AC1-knockout mice. cAMP in those domains might be expected to oscillate if measurements could be made using a cAMP-sensor tagged AC1. Curiously, when AC1 and AC8 were expressed in the non-excitable HEK (human embryonic kidney)-293 cell line, the Ca2+ oscillations induced by muscarinic cholinergic receptor stimulation of SOCE could induce measurable cAMP oscillations using a global cAMP sensor in cells expressing AC8, but not AC1 . These data were interpreted to reflect the different mechanism of AC activation by Ca2+/CaM mentioned above. However, it is not inconceivable that AC1 could generate cAMP oscillations in the AC1 microdomain in response to the Ca2+ transients induced in exuberant neurons by L-type channels.
Even though the fact that cAMP oscillations occur and that they can have explicit consequences has been shown, until very recently how these oscillations could be read by PKA had not been understood. Traditional, textbook views of the regulation of PKA had represented the dissociation of the PKA tetrameric holoenzyme in response to high concentrations of cAMP, such as those that are achieved by forskolin, the global activator of ACs and IBMX (3-isobutyl 1-methylxanthine) a global inhibitor of (most) PDEs. Obviously (as we have noted above), such experiments cannot address potential spatiotemporal properties of cAMP. But Taylor et al.  have now shown how low concentrations of cAMP can lead to activation of PKA without full dissociation of the subunits. By such means, they propose that PKA can decode temporally complex patterns in cAMP .
AC INHIBITORS IN THE LIGHT OF MORE THAN ONE AC PER CELL TYPE
Specific inhibitors of ACs still hold promise as therapeutic agents or as tools for dissecting the contribution of particular ACs to defined cellular events . The problem with inhibitors, apart from their expected ‘off-target’ effects, when they are nucleotide or sugar derivatives, is magnified by the fact that no cell expresses only one AC, which exacerbates the problem of specificity. A recent screen of supposedly selective AC inhibitors against a full panel of ACs revealed little specificity among sugar-modified purine derivatives. A structure-based virtual screen identified a potential non-adenine-binding pocket in the catalytic site that showed promise for selective compounds . The issue of access to the cells, and to the compartment where the ACs reside, magnifies this challenge enormously. In vitro comparisons just cannot anticipate in vivo results and the apparent efficacy of some of these inhibitors in intact cells may not even have to do with their action on ACs. In one example, the purportedly AC5-selective compound vidarabine was fully effective against residual AC activity in AC5-knockout mice . The continued efficacy of vidarabine in AC5-knockout mice may reflect off-target (non-AC) effects as much as poor selectivity for AC subtypes. Although investigators acknowledge the possibility of off-target effects, little effort is made to exclude such effects at early stages of inhibitor development. Another insightful study of the widely used AC inhibitor SQ 22,536 shows that not only is it a potent inhibitor of AC activity, but it also inhibits the downstream effects of cAMP (or dibutyryl-cAMP) on PKA-mediated CREB (cAMP-response-element-binding protein) phosphorylation .
A recent study used an intact cell screening method for inhibitors against PMA-stimulated AC2 activity and later validated the efficacy of the optimal compounds against AC2 activity in in vitro assays . Other approaches that target isoform-specific properties of the ACs such as the binding sites for CaM might also be of interest, though obviously ridden with an equal array of possibly off-target effects on other CaM-binding proteins.
Given the problems associated with AC inhibitors, within the concept of the present review scaffold disruptors seem particularly worth exploring. Clearly, the involvement of AKAP complexes in numerous disease predispositions makes them viable and precise therapeutic targets as discussed recently . Indeed, the first successful attempts at a small-molecule inhibitor of cAMP-directed PKA/AKAP disruptors in cardiac myocytes have been reported [166,167]. Drugs which aim to disrupt the AC–AKAP interaction may offer even greater precision.
FUTURE DIRECTIONS AND APPROACHES
The central role played by cAMP in physiological homoeostasis via its downstream targets, PKA, CNG channels and Epac has been appreciated for many years, nevertheless the sources of the cAMP, the ACs, offer even more insights into the elaborate devices used in these pathways. Specific ACs are now identified with distinct physiological processes and disease states, and unravelling their important contributions poses sizeable challenges, not least because most, if not all, cells express more than one AC, and the ACs can act as scaffolds. How or whether individual ACs with demonstrated susceptibility to indirect regulation, such as PKC, βγ subunits of G-proteins and tyrosine kinases, are regulated in cells which possess more than one AC within restricted cellular domains is a particular challenge. This issue may be most easily be pursued with targeted sensors and, of course, selective inhibitors.
The mechanism of stimulation of AC8 by Ca2+ entry in non-excitable cells seems to be solved by the AC8–Orai1 complex. However, whether the other similarly discriminating Ca2+-sensitive ACs (AC1 and AC5/AC6) form similar complexes remains to be determined. In excitable cells, the mode of interaction between all of the Ca2+-sensitive ACs with L-type Ca2+ channels remains to be clarified.
The full range of AC partners scaffolded by ACs, particularly in explicit physiological contexts, is unknown. Although a daunting challenge, this is the level of complexity at which the ACs operate. Inferences based on simple AC knockdowns are ambiguous; far more insight is likely from knockbacks with defined mutations in regions scaffolding particular partners. Measurements of the consequences for cAMP dynamics in microdomains upon the knockout of putative partners will also be insightful.
The dynamic behaviour of ACs is another demanding question, calling for elaborate methodologies. Single-molecule analyses of ACs labelled with photoactivatable probes along with putative partners  could allow kinetic assessments of encounters, combined with high-resolution microscopic techniques [169,170].
Whether ACs signal internally still seems be an open question. The notion that internalized β-arrestin via ERK inactivates PDEs and thereby allows a sustained accumulation of cAMP seems the more likely interpretation of the experiments described to date. However, whether acceleration of AC internalization provides a means of allowing cAMP to originate from internal zones remains worth considering.
ACs tagged with cAMP or Ca2+ sensors are a major opportunity for studying the enzymes in their own microenvironment and to address the intimate signalling dynamics of specific ACs among a medley of different ACs in a cell. Such experiments are challenging and, whereas some ACs tolerate N-terminal modifications, not all may; significant optimization, calibration and extensive controls will be needed. Nevertheless, if AC-centred microdomains are the base cAMP-signalling unit, the challenge may need to be embraced or alternative avenues into these domains devised [171,172]. The fact that mechanisms have been proposed that allow PKA to respond to cAMP oscillations should stimulate a major exploration of the possible role of cAMP oscillations. Techniques such as generating pulses of cAMP by flash photolysis of caged cAMP will be valuable. Even less disruptive would be to impose cAMP oscillations by controlled membrane depolarization in excitable cells to yield predictable Ca2+ oscillations linked to cAMP via Ca2+-dependent ACs. Such experiments will be highly insightful in systems such as cardiomyocytes, islet β-cells and neurons.
The development of live-cell sensors for cAMP (and Ca2+) has allowed significant maturing of the cAMP field. The theoretical postulate of cAMP compartments has materialized and the dynamic behaviour of cAMP in cellular spaces and over time has been demonstrated. The underpinnings of this behaviour in the shape of AC-centred and sculpted microdomains is beginning to emerge, and this new awareness promises new routes to precise tools to manipulate this centrally important physiological regulator.
We thank the Newton Trust and the Gates Cambridge Trust for support.
Space constraints dictated that this overview could not fully represent the work of the many investigators in the relevant areas.
Abbreviations: AC, adenylate cyclase; AKAP, A-kinase-anchoring protein; CaM, calmodulin; CNG, cyclic nucleotide-gated; ECM, extracellular matrix; eNOS, endothelial nitric oxide synthase; Epac, exchange protein directly activated by cAMP; ER, endoplasmic reticulum; ERK, extracellular-signal-regulated kinase; FA, focal adhesion; IM, ionomycin; Nt, N-terminal; PDE, phosphodiesterase; PKA, protein kinase A; PKC, protein kinase C; PM, plasma membrane; sAC, soluble AC; SOCE, store-operated Ca2+ entry; TmAC, transmembrane AC; TSH, thyroid-stimulating hormone
- © The Authors Journal compilation © 2014 Biochemical Society