Inhibition of adenylyl and guanylyl cyclase isoforms by the antiviral drug foscarnet.

The pyrophosphate (PP(i)) analog foscarnet inhibits viral DNA-polymerases and is used to treat cytomegalovirus and human immunodeficiency vius infections. Nucleotide cyclases and DNA-polymerases catalyze analogous reactions, i.e. a phosphodiester bond formation, and have similar topologies in their active sites. Inhibition by foscarnet of adenylyl cyclase isoforms was therefore tested with (i) purified catalytic domains C1 and C2 of types I and VII (IC1 and VIIC1) and of type II (IIC2) and (ii) membrane-bound holoenzymes (from mammalian tissues and types I, II, and V heterologously expressed in Sf9 cell membranes). Foscarnet was more potent than PP(i) in suppressing forskolin-stimulated catalysis by both, IC1/IIC2 and VIIC1/IIC2. Stimulation of VIIC1/IIC2 by Galpha(s) relieved the inhibition by foscarnet but not that by PP(i). The IC(50) of foscarnet on membrane-bound adenylyl cyclases also depended on their mode of regulation. These findings predict that receptor-dependent cAMP formation is sensitive to inhibition by foscarnet in some, but not all, cells. This was verified with two cell lines; foscarnet blocked cAMP accumulation after A(2A)-adenosine receptor stimulation in PC12 but not in HEK-A(2A) cells. Foscarnet also inhibited soluble and, to a lesser extent, particulate guanylyl cylase. Thus, foscarnet interferes with the generation of cyclic nucleotides, an effect which may give rise to clinical side effects. The extent of inhibition varies with the enzyme isoform and with the regulatory input.

The second messenger cAMP controls an array of cellular responses ranging from lipid and glucose metabolism, motility and contraction, proliferation and differentiation, to synaptic transmission and memory formation. The formation of cAMP is catalyzed by the enzyme adenylyl cyclase. In mammals, there are at least 9 membrane-bound isoforms; these differ in their susceptibility to regulation by G proteins (stimulatory ϭ G␣ s ; inhibitory ϭ G␣ i ; ␤␥-dimers ϭ dual action), Ca 2ϩ , Ca 2ϩ -liganded calmodulin, protein kinases, and the plant diterpene forskolin (1). However, all isoforms share a similar channel-or transporter-like topology: two hydrophobic domains (of ϳ20 kDa each) contain 6 putative transmembrane spanning ␣-helices. These are linked by a cytosolic portion (of ϳ40 kDa) which contains the first catalytic domain (referred to as C1, of ϳ30 kDa). The carboxyl terminus (also of ϳ40 kDa) comprises the second catalytic domain (referred to as C2, of ϳ30 kDa) which is internally homologous to C1. Each domain is per se enzymatically inactive, but catalysis is restored, if the two domains are combined. In addition, there is a soluble isoform, the expression of which is restricted to sperms; this enzyme is not regulated by G proteins and is more closely related to the bacterial isoforms (2).
The substrate for the reaction catalyzed by adenylyl cyclase is Mg⅐ATP, the reaction product is cAMP and PP i (pyrophosphate). The formation of the intramolecular phosphodiester bond (5Ј-PO 4 linked to the ribose 3Ј-OH) is analogous to the reaction catalyzed by DNA-polymerases (phosphodiester bond between incoming nucleotide and DNA strand). Thus, although adenylyl cyclases and DNA-polymerases have little, if any, sequence homology, the overall topology of the active site is similar in the two classes of enzymes (3) and catalysis is thought to involve two metal ions (Mg 2ϩ as the physiological ligand which can be substituted for by Mn 2ϩ ) that are bound in the active site of both, adenylyl cyclases and DNA-polymerases (4).
The PP i analog foscarnet was originally discovered as an inhibitor of herpesvirus DNA-polymerase (5) but later also found to inhibit the reverse transcriptase of HIV, the human immunodeficiency virus (for review, see Ref. 6). Viral DNApolymerases are more sensitive to foscarnet than their mammalian counterparts; depending on the virus strain, IC 50 values in the range of ϳ10 to ϳ250 M have been observed. Foscarnet is currently used to treat infections with cytomegalovirus, other herpesviruses, and human immunodeficiency virus. The most common side effect of foscarnet is reversible nephrotoxicity. The phosphaturia associated with foscarnet treatment has been linked to a direct inhibition of Na ϩ /P i symport in the renal brush border (7). The mechanism underlying the many additional side effects (in particular the neurological abnormalities) is unknown. Previous experiments rule out an effect of foscarnet on cAMP accumulation induced by parathyroid hormone in proximal renal tubules (7). In contrast, the mechanism by which foscarnet blocks the action of antidiuretic hormone is consistent with inhibition of cAMP formation (8). Because adenylyl (and guanylyl) cyclases and DNA-polymerases catalyze related reactions, it is reasonable to assume that foscarnet can, in principle, inhibit adenylyl and guanylyl cyclases. In the present work, we show that this is the case; the inhibitory potency of foscarnet depends on the nature of the isoform and varies with the state of enzyme activation.
Protein Purification and Membrane Preparations-The catalytic domains of adenylyl cyclase were expressed in Escherichia coli BL21 and purified from bacterial lysates using metal affinity chromatography, anion exchange chromatography on MonoQ, and gel filtration on Superose HR12 (9). Similarly, recombinant G␣ s (12) and myristoylated rG␣ i-1 (13) were purified from bacterial lysates, G protein ␤␥-dimers from porcine brain membranes (14). Membranes were prepared from the following sources: guinea pig cerebral cortex (15), calf myocardium (16), and human platelets (17). Sf9 cells were infected with recombinant baculoviruses encoding adenylyl and guanylyl cyclases, lysed, and fractionated into membranes and cytosol as described previously (11).
Cell Culture and cAMP Formation-CHO-K1 cells were grown in Ham's F-12 nutrient mixture supplemented with 10% fetal calf serum, 2 mM L-glutamine, 100 units/ml penicillin G, and 100 g/ml streptomycin; cells were transiently transfected by electroporation with the plasmid pSVL encoding the particulate guanylyl cyclase-A/atrial natriuretic factor receptor (18) and harvested 55 h later. The generation and propagation of stably transfected HEK293 cells that express the A 2Aadenosine receptor (HEK-A 2A ) has been described (10). PC12 cells were propagated in Opti-MEM medium containing 10% horse serum, 5% fetal calf serum, L-glutamine, penicillin G, and streptomycin. The adenine nucleotide pool was metabolically labeled by incubating confluent monolayers for 16 h with [ 3 H]adenine (1 Ci/well); if this preincubation period was varied between 12 and 24 h, there was no appreciable difference in the amount of [ 3 H]cAMP formed in response to the receptor agonist or forskolin; hence, the adenine pool was assumed to be labeled to equilibrium. After the preincubation, fresh medium was added that contained adenosine deaminase (1 unit/ml), 100 M RO201724, and the indicated foscarnet concentrations; after 30 min, cAMP formation was stimulated by the A 2A -selective agonist CGS21680 or 25 M forskolin for 15 min. Assays were performed in triplicate. The formation of [ 3 H]cAMP was determined according to Salomon (19).
Enyzme Assays-For measuring the generation of cAMP by the purified catalytic domains, IC1 or VIIC1 (each at 20 -60 ng/assay) were combined with IIC2 (0.02-2 g/assay) and incubated for 5 min at 20°C in a final volume of 50 l containing 50 mM Hepes/NaOH (pH 7.5), [␣-32 P]ATP (specific activity 10 -100 cpm/pmol), 2.5 mM MgCl 2 , 1 mM MgSO 4 (carry over from the preactivation of rG␣ s or of rG␣ i-1 ), 0.01% Lubrol; unless indicated otherwise in the figure legends, the concentrations of [␣-32 P]ATP-Mg and forskolin were 0.5 and 0.1 mM, respectively. The concentration of foscarnet and PP i was varied between 0.01 and 10 mM; prior to dilution, equimolar MgCl 2 was added to the stock solution of foscarnet and PP i to keep the free Mg 2ϩ concentration constant. Where applicable, rG␣ s and rG␣ i-1 (each at 20 M) were preactivated at 30°C for 30 and 120 min, respectively, in buffer containing 50 mM Hepes/NaOH (pH 7.5), 1 mM EDTA, 10 mM MgSO 4 , 100 M GTP␥S, and 0.01% Lubrol; free GTP␥S and GDP (released from the proteins) was removed by gel filtration over Sephadex G-50 pre-equlibrated in 50 mM Hepes/NaOH (pH 7.5), 1 mM MgSO 4 , and 0.01% Lubrol (20). Both, the catalytic domains of adenylyl cyclase and the G protein ␣-subunits, are soluble in the absence of detergent. However, Lubrol prevents adsorptive losses that occur at low concentrations of G␣ subunits (20). The presence of Lubrol had no effect on the activity of C1/C2 heterodimers. The activity of membrane-bound adenylyl cyclase was assayed in a similar manner with the following modifications: the incubation time was 20 min, the assay mixture contained 10 mM MgCl 2 , 0.1 mM RO201724, 10 mM creatine phosphate, 1 mg/ml creatine kinase. The formation of [ 32 P]cAMP was quantified according to Johnson and Salomon (21).
If not otherwise indicated, experiments were done at least three times. Data were subjected to nonlinear least-squares curve-fitting using the appropriate equations (rectangular hyperbola, Hill equation) to obtain parameter estimates.

Comparison of IC1/IIC2 and VIIC1/IIC2 Heterodimers-
The stimulatory effect of forskolin and G␣ s differs in individual adenylyl cyclase isoforms (1,24). To rule out that these differences may distort the results obtained in subsequent experiments, we have defined the conditions under which IC1 and VIIC1 were saturated with IIC2 and activators ( Fig. 1). We observed a (slightly) higher potency of forskolin in stimulating catalysis by the dimer IC1/IIC2 than that by VIIC1/IIC2 ( Fig.  1C; EC 50 ϭ 8.0 Ϯ 0.9 and 25.8 Ϯ 6.0 M, respectively; n ϭ 3). Conversely, the apparent affinity of preactivated G␣ s was modestly higher for VIIC1/IIC2 than for IC1/IIC2 ( Fig. 1E; EC 50 ϭ 72 Ϯ 28 and 245 Ϯ 87 nM, respectively; n ϭ 3). Accordingly, the apparent affinity of each C1 domain for IIC2 varied with the activating agent employed: for IC1 (Fig. 1A), the EC 50 of IIC2 was 227 Ϯ 109, 276 Ϯ 24, and 153 Ϯ 37 nM in the presence of forskolin, G␣ s , and the combination thereof, respectively. For VIIC1 (Fig. 1B) the EC 50 of IIC2 was 510 Ϯ 149, 39 Ϯ 2, and 17 Ϯ 7 nM in the presence of forskolin, G␣ s and the combination thereof, respectively. The combination of forskolin and G␣ s resulted in overadditive stimulation regardless of the C1-subtype ( Fig. 1, D and F). In summary, these experiments indicated (modest) differences in the affinity of IC1 and VIIC1 for IIC2 which depended on the activator; thus, enzymatic activity was subsequently assessed under conditions where the C1 domain was limiting (ϳ30 ng/assay), while IIC2 (ϳ2 g/assay), G␣ s (2 M), and forskolin (100 M) were saturating.
Foscarnet Inhibits cAMP Formation Catalyzed by IC1/IIC2 and VIIC1/IIC2 via Interaction with the PP i -binding Site-Adenylyl cyclases are subject to product inhibition by both, cAMP and PP i . Product release is random and, at least in part, rate-limiting (25). In the forward reaction, i.e. the formation of cAMP ϩ PP i from ATP, the product PP i is not a competitive inhibitor of adenylyl cyclase with respect to the substrate ATP. Depending on the assay conditions, the inhibition is noncompetitive or mixed competitive (25). If foscarnet acted as a PP i analog, the two compounds should inhibit the reaction in a similar way. This was the case. For both, IC1/IIC2 ( Fig. 2A) and VIIC1/IIC2 (Fig. 2B), the V max of the forskolin stimulated activity was suppressed by raising the concentration of foscarnet and PP i . Similarly, the K m for ATP increased with higher concentrations of foscarnet and PP i ; this is more readily seen from the replots shown in Fig. 2, C and D.
The product inhibition imposed by (nonphysiological concentrations of) cAMP is mimicked by adenosine and some analogs (3Ј-phosphorylated and deoxyadenosine analogs, see Refs. 25 and 26), which are referred to as P-site inhibitors. When combined, PP i and P-site ligands inhibit adenylyl cyclase synergistically and this is thought to reflect the formation of a dead-end complex (25,27). If foscarnet inhibited adenylyl cyclase via interaction with the binding site for PP i , the combination of foscarnet and a P-site ligand also should act in a synergistic manner; in contrast, the combination of PP i and foscarnet is expected to result in mutual antagonism. Dixon plots (where the reciprocal of enzymatic velocity is plotted as a function of one inhibitor at several fixed concentrations of the second inhibitor) allow to test if two inhibitors can occupy an enzyme simultaneously or whether their binding is mutually exclusive (28). If foscarnet was combined with fixed concentrations of 2Ј,3Ј-dideoxyadenosine, the slope of the individual regression lines depended on the concentration of this second inhibitor for both, IC1/IIC2 (Fig. 3A) and VIIC1/IIC2 (Fig. 3B); this observation shows that the two compounds can be bound simultaneously (28). As expected, this was also seen for the combination of 2Ј,3Ј-dideoxyadenosine and PP i (Fig. 3, C and D). In all cases, the lines intersected above the x axis indicating that 2Ј,3Ј-dideoxyadenosine facilitated inhibition by foscarnet (or by PP i ). A similar synergism was observed with adenosine and 2Ј,5Ј-dideoxyadenosine (data not shown). In contrast with both, IC1/IIC2 (Fig. 4A) and VIIC1/IIC2 (Fig. 4B), Dixon plots for the combination of PP i and foscarnet yielded a family of parallel regression lines. This is the diagnostic feature indicative of mutually exclusive binding (28). Thus, the presence of a fixed concentration of foscarnet impeded the inhibitory action of PP i . Taken together, the data are consistent with the interpretation that foscarnet and PP i occupy the same site in the catalytic core of adenylyl cyclases.
Inhibition of cGMP Formation by Foscarnet and PP i -The other cyclic nucleotide second messenger in cells, cGMP, is generated by the isoforms of guanylyl cyclase. Although not investigated in detail, the catalytic mechanism of guanylyl cyclase is generally thought to resemble that of adenylyl cyclase. It is evident from Fig. 5A that both, PP i and foscarnet, inhibit basal soluble guanylyl cyclase; the difference in potency between PP i and foscarnet was very modest (IC 50 ϭ 0.35 Ϯ 0.06 and 0.52 Ϯ 0.09 mM for PP i and foscarnet, respectively). Addition of the NO-donor diethylamine/nitric oxide (0.1 mM) stimulated catalysis 8.2 Ϯ 1.8-fold; the IC 50 of PP i (0.40 Ϯ 0.06 mM) was not affected, but the potency of foscarnet was somewhat lower (IC 50 ϭ 0.90 Ϯ 0.17 mM) in the presence of the NO-donor (not shown). In contrast, cGMP formation catalyzed by atrial natriuretic peptide-stimulated particulate guanylyl cyclase was resistant to inhibition by PP i ; the enzyme was only inhibited by foscarnet (Fig. 5B), albeit with lower potency (IC 50 ϭ 3.5 Ϯ 0.3 mM) than the soluble isoform.
G␣ s Affects the Potency of Foscarnet-Foscarnet was ϳ5-6fold more potent than PP i in inhibiting the basal as well as the forskolin-stimulated activity of IC1/IIC2 and VIIC1/IIC2 (Table   FIG. 1. Complementation of adenylyl cyclase activity of IC1 and VIIC1 by IIC2 in the presence of the activators forskolin, rG␣ s , and the combination thereof. A and B, increasing amounts of IIC2 were added to IC1 (A) or VIIC1 (B) (each at 30 ng/assay); catalysis was activated by 100 M forskolin (q), 2 M GTP␥S-liganded rG␣ s (E), or a combination of forskolin and rG␣ s (). Panels C-F, the amount of IIC2 (2 g/assay) and IC1 (q; 20 -30 ng/assay) or VIIC1 (E; 30 ng/assay) was kept constant and the concentration of forskolin (C and D) and preactivated rG␣ s (E and F) was varied. In D and F, preactivated rG␣ s and forskolin were held constant at 0.3 and 30 M, respectively. Data are means of duplicate determinations; each comparison of IC1 and VIIC1 was done in parallel. The experiment is representative for two additional experiments. I). However, the physiological activator of adenylyl cyclase is G␣ s ; we have therefore also determined the inhibitory potency of foscarnet and PP i on the activity stimulated by GTP␥Sliganded G␣ s , forskolin, and their combination. Preactivated G␣ s was employed at a saturating concentration to eliminate the difference in affinity of IC1/IIC2 and VIIC1/IIC2 (see Fig.  1). The inhibition by foscarnet was blunted in the presence of rG␣ s while the IC 50 of PP i decreased. This effect was more pronounced for the heterodimer VIIC1/IIC2 than for IC1/IIC2 (Table I). We rule out that the difference between the IC 50 can simply be accounted for by the higher activity that is achieved through stimulation by G␣ s . If catalysis was further activated by the combination of forskolin and G␣ s , foscarnet suppressed cAMP formation with an intermediate potency (Table I); the IC 50 values were lower than those observed in the presence of G␣ s but higher than those seen with forskolin.
Inhibition of Membrane-bound Adenylyl Cyclase Isoforms-Because the C1/C2-dimers employed are artificial, we have also analyzed the effect of foscarnet on adenylyl cyclase in membranes prepared from guinea pig cerebral cortex, calf ventricular myocardium, and human platelets. These preparations are likely to contain a mixture of several isoforms due to the cellular heterogeneity and due to the fact that many cells express more than one isoform. However, brain membranes are enriched in the type I isoform (1). Platelets contain an isoform that is activated by ␤␥-dimers in the presence of activated G␣ s (14). This is a characteristic feature of the type II and type IV isoforms (1,29). The myocardium expresses predominantly the type V (and to lesser extent the type VI) isoform (1,30). Adenylyl cyclase in these membrane preparations was assayed under four different conditions. (i) The enzyme was directly stimulated by forskolin. Since the stimulation is greatly augmented by the presence of G␣ s (1), in membranes forskolinstimulated catalysis reflects the sum of the direct action of the compound and the basal level of G␣ s activation. (ii) GTP␥S was added to the reaction; this results in activation of both G i (and G o in brain membranes) and G s ; thus the activity reflects the combined effect of inhibition and stimulation. (iii) Purified G␣ s was preactivated by incubation with GTP␥S and MgSO 4 , unbound GTP␥S was removed by gel filtration and GTP␥S-liganded G␣ s was added to the membranes. (iv) Basal activity was assessed without any exogenous activator. It has to be pointed out, though, that this basal activity does not only reflect the intrinsic rate of catalysis of the enzyme, but also the sum of stimulatory and inhibitory input occurring at low level. Trace amounts of GDP are present in the membrane (e.g. bound to monomeric and heterotrimeric G proteins); commercial ATP preparations are contaminated by low levels of GTP (15). In the assay, GTP is also formed by transphosphorylation of GDP (31).
As can be seen from Table II, the apparent potency of foscarnet depends on the nature of activating ligand. In general, activation of GTP␥S-liganded G␣ s reduced the sensitivity of the enzyme to foscarnet. This effect was most pronounced in cardiac membranes. In addition, the IC 50 values varied which presumably reflected the expression of different adenylyl cyclase isoforms. Basal activity, for instance, was most readily inhibited in brain membranes. In cardiac membranes, the enzyme was most susceptible to foscarnet in the presence of GTP␥S and there was a striking left-shift of the curve when compared with the activity in the presence of GTP␥S-liganded G␣ s . This was also seen, albeit to a lesser extent, in cerebrocortical membranes. As mentioned above, addition of GTP␥S activates both, G i and G s endogenous to the membranes. Hence, it appears likely that the presence of inhibitory subunits (G␣ i and/or G␤␥-dimers) renders the enzyme more sensitive to foscarnet. Regardless of the underlying mechanism, it is safe to conclude that the potency of foscarnet is affected by the activation state of the individual isoform.

FIG. 3. Inhibition of forskolin-activated adenylyl cyclase activity by the combination of foscarnet (A and B) or PP I (C and D) with 2,3-dideoxyadenosine (ddA). IC1 (A and
A comparison of Tables I and II shows that foscarnet inhibited the purified catalytic domains more potently than the membrane-bound holoenzymes. This discrepancy may result from the fact that the heterodimers represent nonphysiological forms of the enzyme. Alternatively, domains of the holoenzyme that are not part of the catalytic core may modify access of foscarnet to the PP i -binding sites. To differentiate between these two possibilities, we have used the adenylyl cyclase isoforms type I and type II. These were expressed in Sf9 cells either as intact holoenzymes or as half-molecules (12). In the latter case, the site of truncation is within the region preceding the second hydrophobic domain (positions 571 and 556 in type I and type II, respectively). Thus, IM1C1 comprises the entire amino-terminal half of type I (including the first transmembrane domain M1) and IIM2C2 the entire COOH-terminal half of type II (including the second transmembrane domain M2). Sf9 cells were simultaneously infected with baculoviruses encoding IM1C1 and IIM2C2. The control consisted of cells infected with viruses encoding either type I or type II holoenzyme. In addition, the type V isoform was also investigated because it is the predominant enzyme in the myocardium (1,30). Membranes prepared from these cells were used to evalu-ate the effects of foscarnet and PP i on forskolin-stimulated catalysis (Fig. 6). Foscarnet inhibited holoenzymes type I and type II with comparable potency (Fig. 6A). Similarly, the variation in the IC 50 values of PP i was trivial (Fig. 6B). Importantly, the combination of IM1C1 and IIM2C2 yielded an enzymatic activity that was inhibited over the same concentration range as the holoenzymes. Finally, the potency of foscarnet (IC 50 ϭ 0.4 Ϯ 0.1 mM) and PP i (IC 50 ϭ 1.3 Ϯ 0.2 mM) was lower than on the catalytic core IC1/IIC2 (see Table I).
In both, myocardial and brain membranes, cAMP formation that was stimulated by the addition of GTP␥S was more susceptible to inhibition by foscarnet than catalysis activated by GTP␥S-liganded G␣ s (Table II). This discrepancy can be rationalized if the GTP␥S-induced increase in inhibitory subunits (GTP␥S-liganded G␣ i and G␣ o and ␤␥-dimers) is assumed to sensitize the catalyst to foscarnet. To test this conjecture, we have mimicked this situation by adding GTP␥S-liganded G␣ i-1 and free ␤␥-dimers to Sf9 membranes expressing the type I and type V isoforms in the presence of preactivated G␣ s (Fig. 7). Sole addition of G␣ i-1 caused a modest inhibition of the G␣ sactivated type I enzyme which was substantially augmented by the presence of free ␤␥-dimers (Fig. 7A). Importantly, addition of G␤␥ shifted the concentration-response curve of foscarnet to the left; this is most readily seen from the inset in Fig. 7A, where the data were normalized (IC 50 ϭ 0.9 Ϯ 0.1, 1.0 Ϯ 0.1, and 0.3 Ϯ 0.1 mM in the presence of rG␣ s , rG␣ s ϩ rG␣ i-1 and rG␣ s ϩ rG␣ i-1 ϩ ␤␥, respectively). In contrast, adenylyl cyclase type V was only inhibited by G␣ i-1 ; G␤␥ had no additional effect (Fig. 7B). Furthermore, the presence of G␣ i-1 and G␤␥ did not affect the potency of foscarnet; the inset of Fig. 7B shows that the three concentration-response curves were superimposable (IC 50 ϭ 1.2 Ϯ 0.1, 1.3 Ϯ 0.1, and 1.3 Ϯ 0.2 mM in the presence of rG␣ s , rG␣ s ϩ rG␣ i-1 and rG␣ s ϩ rG␣ i-1 ϩ ␤␥, respectively). Hence, the type V isoform did not adequately reproduce the properties of the catalyst(s) present in cardiac membranes. However, the results with the type I isoform clearly showed that an inhibitor was capable of rendering the enzyme more susceptible to foscarnet.
Effect of Foscarnet in Intact Cells-Taken together, the ob-TABLE I Inhibition by foscarnet and by pyrophosphate (PP i ) of cAMP formation catalyzed by IC1/IIC2-and VIIC1/IIC2-heterodimers IC1 or VIIC1 (each at 20 -60 ng/assay) were combined with IIC2 (2 g/assay); activity was determined in the absence (basal) and presence of the indicated concentrations of activators under assay conditions described in the legend to Fig. 1. Enzymatic activity is expressed as cAMP formation/mg of IC1 or of VIIC1; 6 logarithmically spaced concentrations of foscarnet (10 M to 3 mM) or PP i (30 M to 10 mM) were used. Data are mean Ϯ S.D. from three experiments.

TABLE II
Inhibition by foscarnet of adenylyl cyclase activity in membranes prepared from heart, brain cortex and platelets Membranes (10 -20 g) were incubated in the absence of any exogenous activator (basal) and in the presence of the indicated activators; the carry-over of detergent and MgSO 4 was corrected by adding appropriate amounts to the other reactions; 6 logarithmically spaced concentrations of foscarnet (10 M to 3 mM) were used. The data represent mean Ϯ S.D. of three separate experiments that were done in duplicate and conducted in parallel.  8A); the similar potency of CGS21680 (EC 50 ϭ 23 Ϯ 9 and 24 Ϯ 6 nM in HEK-A 2A and PC12 cells, respectively) indicates that the A 2A -receptor is efficiently coupled to G␣ s in both PC12 and HEK-A 2A . However, the action of foscarnet was clearly distinct. In PC12 cells, forskolin-and A 2A -agonist-dependent cAMP formation were suppressed by foscarnet over a similar concentration range (Fig. 8B). In contrast, in HEK-A 2A cells, the A 2A -agonist-dependent stimulation was resistant to foscarnet, while the response to forskolin was blunted (Fig. 8C). Thus, after receptor-dependent activation of G␣ s , adenylyl cyclase in HEK-A 2A cells was no longer susceptible to inhibition by foscarnet.

DISCUSSION
The observations demonstrate that foscarnet directly inhibits adenylyl cyclase isoforms; several lines of evidence indicate that foscarnet binds to the PP i -binding site. (i) Similar to PP i , foscarnet caused mixed-competitive inhibition of the forward reaction; (ii) foscarnet substituted for PP i in synergizing with P-site ligands; (iii) adenylyl cyclase could only be inhibited by either foscarnet or PP i indicating that they bound to the catalytic core in a mutually exclusive manner. The PP i -binding site is formed by a loop in the C1 domain (3,4). We have observed that foscarnet (and PP i ) prevented binding of the fluorescent ATP analog TNP-ATP (9) to VIIC1 but not to IIC2. 2 Taken together, our findings are consistent with the notion that foscarnet and PP i bind to the same site. In addition, the data suggest that amino acid residues, which are not part of the catalytic core, impinge on the PP i -binding site to regulate catalysis. We observed that the soluble heterodimers formed by the C1 and C2 domains were more susceptible to inhibition by foscarnet and by PP i than were the holoenzymes (i.e. the mol- ecules comprising the catalytic core, the additional cytoplasmic stretches and the transmembrane domain); this difference was seen regardless of whether intact holoenzymes or the artificial holoenzyme IM1C1/IIM2C2 were employed. Several lines of arguments suggest that the stretch that links the first catalytic C1 domain to the second transmembrane portion participates in the regulation of catalysis (32). This region, for instance, is required for activation of the type I isoform by calmodulin (33,34) and is thought to contain the inhibitory Ca 2ϩ site of the type V (and VI) isoforms (35). Foscarnet also inhibited soluble guanylyl cyclase and, to a lesser extent, particulate guanylyl cyclase-A. This is to be anticipated. Adenylyl cyclases and guanylyl cyclases catalyze the same reaction; accordingly, the substrate specificity can be switched by exchanging appropriate residues between adenylyl cyclase and soluble (36) or membrane-bound guanylyl cyclases (37).
Previous studies drew opposite conclusions, namely that foscarnet did (9) or did not (8) inhibit receptor-dependent cAMP production; our experiments resolve this controversy. We observed that G␣ s relieved the inhibition of adenylyl cyclase by foscarnet. The effect of G␣ s on the inhibitory potency of foscarnet, however, varied with individual isoforms; it was, for instance, more pronounced with the VIIC1/IIC2 than with the IC1/IIC2 heterodimer. In PC12 and HEK-A 2A cells, the coupling efficiency of the signaling cascade A 2A -adenosine receptor/G s /adenylyl cyclase was similar as reflected by the virtually identical EC 50 for the agonist. Nevertheless, foscarnet discriminated between the receptor (and hence G␣ s -)-dependent cAMP formation in PC12 and HEK-A 2A cells. We thus conclude that the potency of foscarnet in intact cells depends on the cellular complement of adenylyl cyclase isoforms. In addition, an inhibitory input renders some isoforms more susceptible to the action of foscarnet. G␤␥-Dimers (but not by Gi ␣-1 ) enhanced the potency of foscarnet on adenylyl cyclase type I. In contrast, the experiments with the type V isoform failed to reproduce the sensitization that occurred in cardiac membranes upon activation of endogenous G proteins by GTP␥S. The reason for this discrepancy is not clear. Suppression of cAMP accumulation was observed, if cells were co-transfected with plasmids encoding adenylyl cyclase type V and G␤␥ (38). However, previous reconstitution experiments also did not detect an inhibitory action of ␤␥-dimers on adenylyl cyclase type V (and VI) (29). The reconstitution assay may fail to restore the correct interaction between ␤␥ and the type V enzyme and this may account for our inability to recapitulate the sensitization to foscarnet that was seen in cardiac membranes. The alternative explanation is the cellular heterogeneity of the myocardium. A large proportion of cardiac membranes is actually derived from the endothelium (39); hence, the presence of ubiquitously expressed isoforms such as type VII (30) may give rise to the distinct findings when cardiac membranes are compared with the type V isoform.
It has recently been appreciated that inhibition of adenylyl cyclase may account for side effects of antiviral and cytostatic adenosine analogs, which are converted to acyclic adenine nucleoside phosphonates (40). Contrary to these experimental drugs, foscarnet is widely used in man. Foscarnet permeates into cells, the volume of distribution is in the range of 0.5 liters/kg (7) which is indicative of distribution in the total body water. Hence, plasma concentrations (in the range of 0.25 to 0.5 mM) presumably reflect the intracellular levels. The experiments that were carried out in intact cells show that adenylyl cyclase inhibition can occur within the therapeutic concentration range. Thus, our findings indicate that some clinical side effects of foscarnet may be linked to inhibition of cAMP and/or cGMP accumulation. At the cellular level, the actual adenylyl cyclase activity reflects the integrated response to stimulatory and inhibitory input and is presumably subject to wide interindividual variation. The same consideration holds true for guanylyl cyclase isoforms. It is attractive to speculate that our observations can, in principle, explain the variable extent to which foscarnet elicits untoward reactions in individual patients. This is, in particular, relevant to the neurological manifestations of foscarnet toxicity. Since cAMP and cGMP levels in neurons are subject to diverse regulatory influences, the sensitivity to foscarnet may vary depending on the individual level of catalytic activity.