Negative Feedback Exerted by cAMP-dependent Protein Kinase and cAMP Phosphodiesterase on Subsarcolemmal cAMP Signals in Intact Cardiac Myocytes AN IN VIVO STUDY USING ADENOVIRUS-MEDIATED EXPRESSION OF CNG CHANNELS*

Intracardiac cAMP levels are modulated by hormones and neuromediators with specific effects on contractility and metabolism. To understand how the same second messenger conveys different information, mutants of the rat olfactory cyclic nucleotide-gated (CNG) channel (cid:1) -subunit CNGA2, encoded into adenoviruses, were used to monitor cAMP in adult rat ventricular myocytes. CNGA2 was not found in native myocytes but was strongly expressed in infected cells. In whole cell patch-clamp experiments, the forskolin analogue L-858051 (L-85) elicited a non-selective, Mg 2 (cid:2) -sensitive current observed only in infected cells, which was thus identified as the CNG

gets are essential for the speed and specificity of signal transduction events (1,2,3). However, how such modules maintain specificity when small diffusible molecules are generated during the signaling cascade is difficult to investigate. This question is particularly relevant for cAMP in the heart, where this cyclic nucleotide second messenger exerts diverse effects in response to a number of different neuromediators and hormones. For instance, the ␤-adrenergic agonist isoprenaline (ISO), 1 prostaglandin E 1 (PGE 1 ), and glucagon-like peptide 1 (GLP-1) elevate intracardiac cAMP levels with different effects on contractility; ISO augments the force of contraction, PGE 1 does not, and GLP-1 exerts a negative inotropic effect (4,5). In order to explain these results, subcellular compartmentation of cAMP was proposed more than 20 years ago (6).
Localized cAMP signals may be generated by the interplay between discrete production sites and restricted diffusion within the cytoplasm. In addition to specialized membrane structures that may circumvent cAMP spreading (6,7), degradation of cAMP into 5Ј-AMP by cyclic nucleotide phosphodiesterases (PDEs) appears critical for the formation of dynamic microdomains (8 -14). Cardiac PDEs fall into four families: PDE1, which is activated by Ca 2ϩ /calmodulin; PDE2, which is stimulated by cGMP; PDE3, which is inhibited by cGMP; and PDE4. Whereas PDE1 and PDE2 can hydrolyze both cAMP and cGMP, PDE3 preferentially hydrolyzes cAMP, and PDE4 is specific for cAMP. There is abundant biochemical evidence that PDE3 and PDE4 are activated by cAMP-dependent protein kinase (PKA) phosphorylation in several tissues, providing a putative negative feedback mechanism by which cAMP may regulate its own levels (15,16,17).
A deeper understanding of the mechanisms involved in cAMP homeostasis requires appropriate methods for the direct and continuous measurement of the second messenger in intact cells. The approaches developed so far in cardiac myocytes are based on the use of fluorescent PKA as a biosensor of cAMP (18,14). However, careful evaluation of PKA-based indicators re-* This work was supported in part by the Fondation de France. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  veals a number of drawbacks, which complicate the interpretation of the results. Among these, the kinase activity inherent to the probe and the poor dynamic range may be regarded as the most limiting (18,19). An alternative approach, initially developed by Rich et al. (7), uses genetically modified ␣-subunits of rat olfactory cyclic nucleotide-gated channel (CNG) as cAMP sensors. The wild-type ␣-subunit (CNGA2), on its own, forms a cationic channel directly opened by cyclic nucleotides with fast kinetics, wide dynamic range, and lack of desensitization. Mutants of CNGA2 with increased cAMP sensitivity and selectivity have proven to be valuable tools for monitoring subsarcolemmal cAMP in model cells (19).
In this article, we report the first real-time measurement of cAMP with CNG channels in cardiac myocytes. This method was used to directly examine some of the regulatory mechanisms involved in shaping physiological subsarcolemmal cAMP signals. EXPERIMENTAL April 12, 1991). Male Wistar rats (160 -180 g) were subjected to anesthesia by intraperitoneal injection of pentothal (0.1 mg/g), and hearts were excised rapidly. Individual ventricular myocytes were obtained by retrograde perfusion of the heart as previously described (20). Freshly isolated cells were suspended in minimal essential medium (MEM: M 4780; Sigma) containing 1.2 mM Ca 2ϩ , 2.5% fetal bovine serum (FBS, Invitrogen, Cergy-Pontoise, France), 1% penicillin-streptomycin and 2% HEPES (pH 7.6) and plated on laminin-coated culture dishes (10 g/ml laminin, 2 h) at a density of 10 4 cells per dish. The cells were left to adhere for 1 h in a 95% O 2 , 5% CO 2 incubator at 37°C, before the medium was replaced by 200 l of FBS-free MEM containing the E583M CNGA2-encoding adenovirus (Ad-CNG), the E583M C460W CNGA2-encoding adenovirus or the green fluorescent protein-encoding adenovirus (Ad-GFP). Ad-GFP was used at a multiplicity of infection (MOI) of 100 plaque-forming units (pfu) per cell, whereas both CNG channel-encoding adenoviruses were generally used at 3000 pfu/cell (see results). In co-infection experiments of Ad-CNG with an adenovirus encoding for rabbit muscle cAMP-dependent protein kinase inhibitor (Ad-PKI) (21), each virus was used at MOI 600. After 2 h, the same volume of FBS-free medium without adenovirus was added, and the cells were placed overnight in an incubator. The medium was changed the next morning for adenovirus-and FBS-free MEM. Patch-clamp experiments were performed the same day.
Electrophysiological Experiments-The whole cell configuration of the patch-clamp technique was used to record the L-type Ca 2ϩ current (I Ca,L ) and the CNG current (I CNG ). For I Ca,L measurement, the cells were depolarized every 8 s from Ϫ50 to 0 mV during 400 ms. The use of Ϫ50 mV as holding potential allowed the inactivation of voltage-dependent sodium currents. Potassium currents were blocked by replacing all K ϩ ions with external and internal Cs ϩ . For I CNG measurement, the cells were maintained at 0 mV holding potential and routinely hyperpolarized every 8 s to Ϫ50 mV test potential during 200 ms. Current-voltage relationships were obtained by varying the test potential amplitude to values ranging from Ϫ50 to ϩ50 mV. The 0 mV holding potential was chosen because it corresponds to the reversal potential of I CNG under our experimental conditions (Fig. 2C). Indeed, I CNG was recorded in the absence of divalent cations in the extracellular solution (see below) allowing monovalent cations to flow through the channels in a nonspecific manner. All the experiments were done at room temperature (21-27°C), and the temperature did not vary by more than 1°C in a given experiment.
Solutions and Drugs for Patch-Clamp Recording-Control zero Ca 2ϩ / Mg 2ϩ extracellular Cs ϩ Ringer solution contained (in mM): NaCl 107.1, CsCl 20, NaHCO 3 4, NaH 2 PO 4 0.8, D-glucose 5, sodium pyruvate 5, HEPES 10, adjusted to pH 7.4. For I CNG recording, this solution was supplemented with nifedipine (1 M) to block nonspecific cation current through L-type Ca 2ϩ channels. For I Ca,L recording, nifedipine was omitted and 1.8 mM CaCl 2 and 1.8 mM MgCl 2 were added to the external solution. Control and drug-containing solutions were applied to the exterior of the cell by placing the cell at the opening of a 250-m inner diameter capillary tubing. Patch electrodes (0.8 -1.2 M⍀) were filled with control internal solution containing (in mM): CsCl 118, EGTA 5, MgCl 2 4, sodium phosphocreatine 5, Na 2 ATP 3.1, Na 2 GTP 0.42, CaCl 2 0.062 (pCa 8.5), HEPES 10, adjusted to pH 7.3. ISO and IBMX were purchased from Sigma. L-858051 (L-85, a hydrosoluble analogue of forskolin), H89 and cilostamide were purchased from France Biochem (Meudon, France), and RO 20-1724 was gently provided by Hoffman-La-Roche (Basel, Switzerland).
Data Analysis-The maximal amplitude of I Ca,L was measured as the difference between the peak inward current and the current at the end of the 400-ms duration pulse (22). I CNG amplitude is time-independent and was measured at the end of the 200-ms pulse. Currents were not compensated for capacitance and leak currents. In a total of 154 rat ventricular myocytes, mean capacitance was 157.3 Ϯ 5.1 pF. I CNG density (dI CNG ) was calculated for each experiment as the ratio of current amplitude to cell capacitance. All the data are expressed as mean Ϯ S.E. When appropriate, the Student's t test was used for statistical evaluation. The concentration-response curves (CRC) for the effects of ISO and L-85 on I CNG were fitted to the Hill equation: where EC 50 is the drug concentration ([drug]) required to produce half-maximal stimulation, E max is the maximal effect, and n the Hill coefficient. For each agonist, the CRC obtained with the two CNGA2 channels were compared by a Fisher test. For both Student and Fisher tests, a p value of Ͻ0.05 was considered statistically significant.
Immunocytochemistry-Cells attached onto coverslips were rinsed once in phosphate-buffered saline solution (PBS) for 5 min, fixed in paraformaldehyde 4% (5 min) and washed in PBS (3 ϫ 5 min). The cells were then permeabilized in Triton X-100 0.5% (15 min), washed in PBS (2 ϫ 5 min) and once in PBS-BSA 1% (5 min). Next, they were incubated with a mouse monoclonal antibody against CNGA2 (dilution 1:800, 1 h, 37°C) This antibody was a generous gift from Drs. F. Mueller and B. Kaupp (Juelich, Germany). After three washes in PBS and one in PBS-BSA 1%, the cells were revealed with the Alexa fluor 488 goat anti-mouse antibody (30 min, 37°C). After three additional washes in PBS, the coverslips were mounted in Mowiol antifadent mounting medium (France Biochem) and examined under a Carl Zeiss (Oberkochen, Germany) LSM 510 confocal scanning laser microscope. Optical sections series were obtained with a Plan Apochromat ϫ63 objective (NA 1.4, oil immersion). The fluorescence was observed with a LP 505-nm emission filter under 488-nm laser illumination.
PDE Assays-For each assay, 5 ϫ 10 5 freshly isolated rat ventricular myocytes were plated on 60-mm dishes and infected with Ad-CNG as described above. After 24 h of culture, the cells were preincubated. in control external Ringer (similar to that used in patch-clamp experiments, see composition above) supplemented or not with H89 (1 M). After 30 min, cells were stimulated or not with 100 nM ISO or 10 M L-85 during 5 min. Cells were homogenized at 4°C in extraction buffer (20 mM Tris-HCl, pH 7.5, 5 mM EGTA, 150 mM NaCl, 20 mM glycerophosphate, 1 M H89, 10 mM NaF, 1 mM NaVO3, 1% Triton X-100, 160 M of the serine protease inhibitor 4-(2-aminoethyl)benzenesulfonyl fluoride (AEBSF), 130 nM aprotinin, 8.3 M bestatin, 2.5 M E64, 3.3 M leupeptin, and 1.6 M pepstatin A) and centrifuged 15 min at 15,000 rpm. The supernatant was kept at Ϫ80°C until use. PDE activities were measured by radioenzymatic assay as previously described (23) at a substrate concentration of 1 M cAMP in the presence of 15,000 cpm [ 3 H]cAMP as a tracer. The buffer solution was of the following composition: 50 mM Tris-HCl, pH 7.5, 2 mM magnesium acetate, and 1 mM EGTA. H89 (1 M) was added to this buffer to prevent phosphorylation of PDEs by exogenous cAMP. Assays of total cAMP hydrolytic activity and isoform-specific PDE activity were run in the same batch of experiments. Samples were diluted in order to have around 15% of hydrolysis in absence of specific inhibitors. PDE isoform-specific activities were determined as the difference between PDE activity in the absence of inhibitor and the residual hydrolytic activity observed in the presence of the selective inhibitor (24). The results were expressed as pmol min Ϫ1 mg prot Ϫ1 . Proteins were determined according to Lowry et al. (25) using BSA as standard.

Functional Expression of Cyclic Nucleotide-gated Channels in Cardiac
Myocytes-The E583M mutant of the ␣-subunit of rat olfactory CNG channels encoded into an adenoviral vector (Ad-CNG) was used as an indicator of cAMP in rat ventricular myocytes. This mutation increases the sensitivity of the chan-cAMP Dynamics in Cardiomyocytes nel to cAMP and decreases the sensitivity to cGMP (12). Preliminary patch-clamp experiments were designed to test whether functional CNG channels could be detected in native and Ad-CNG-infected rat ventricular myocytes. Freshly isolated cells were infected or not with Ad-CNG at different MOI (see below), and the whole cell current at Ϫ50 mV was recorded 24 h after in Ca 2ϩ -and Mg 2ϩ -free external solution (7). The forskolin analogue L-85 was used to elevate intracellular cAMP. As shown in the individual current traces of Fig. 1A, application of L-85 at 100 M had no effect on the current at Ϫ50 mV in non-infected cardiomyocytes (NI, n ϭ 6) but activated a robust, time-independent inward current at this voltage in Ad-CNG-infected cells (n ϭ 8). This current displayed other characteristic features of CNG currents, such as Mg 2ϩ block (Fig. 1B) and a linear current-voltage relationship crossing at 0 mV (Fig. 1C). Such current was not detected in rat ventricular myocytes infected with a GFP-encoding adenovirus (n ϭ 4, data not shown) and was thus attributed to the functional expression of E583M CNGA2.
To determine the optimal virus to cell ratio to use in subsequent experiments, the effect of L-85 (100 M) on the CNG current at Ϫ50 mV (I CNG ) was compared in ventricular myocytes infected at three different MOI: 100, 600, and 3000 pfu/ cell. As shown in Fig. 1D, the mean I CNG density (which represents the ratio of I CNG to cell membrane capacitance) upon application of 100 M L-85 was 17.3 Ϯ 7.2 pA/pF (MOI 100), 36.1 Ϯ 9.0 pA/pF (MOI 600), and 50.1 Ϯ 4.5 pA/pF (MOI 3000, n ϭ 8 for each). Thus, the higher dose of MOI 3000 provided the wider dynamic range and increased the liability of the measure. The toxicity of Ad-CNG infection was evaluated by calculating the percentage of round myocytes at all three MOI and in non-infected cells. In 24 fields from 4 different dishes, totaling about 400 cells in each condition, this parameter was 14.8 Ϯ 2.2% (non-infected), 10.2 Ϯ 2.7% (MOI 100), 9.4 Ϯ 1.3% (MOI 600) and 10.2 Ϯ 0.3% (MOI 3000). Thus, Ad-CNG infection did not increase mortality at 24 h, even at a MOI of 3000. This virus to cell ratio was used in all subsequent experiments except for co-infection protocols where each adenovirus was used at MOI 600.
Subcellular Localization of Recombinant CNG Channels-In a second series of experiments, CNGA2 expression was investigated by immunofluorescence in non-infected and Ad-CNGinfected myocytes after 24 h of culture. Fig. 2 shows confocal images of representative cells in both groups labeled with the monoclonal anti-CNGA2 or the secondary antibody alone. In non-infected cells, a weak longitudinal staining was observed in both conditions, and was thus attributed to nonspecific binding of the secondary antibody. In contrast, in rat ventricular myocytes infected with Ad-CNG, a strong fluorescent signal was observed throughout the cell, with accentuation at the extremities, around the nucleus, and in transversal stripes with a periodicity of Ϸ1.9 m. Such pattern was not observed cAMP Dynamics in Cardiomyocytes when the primary antibody against CNGA2 was omitted. These results suggest that the recombinant channel distributes ubiquitously at the membrane, with preferential localization in particular structures.
Comparative Effects of ISO and L-85 on I CNG and I Ca,L -We next tested whether CNG channels were able to detect cAMP increases elicited by a hormonal-like stimulation. Fig. 3A presents a typical experiment in which the effect of the sympathomimetic amine ISO was tested and compared with the effect of L-85 on I CNG at Ϫ50 mV. ISO at 100 nM induced a clear increase of I CNG , although Ϸ5 times smaller than that elicited by 100 M L-85. In 7 similar experiments the mean current density was augmented from 4.3 Ϯ 1.3 pA/pF in control Ringer to 10.9 Ϯ 2.1 pA/pF with 100 nM ISO and to 51.4 Ϯ 7.8 pA/pF with 100 M L-85 (Fig. 3B). Comparable differences between ISO and L-85 were obtained at MOIs of 100 and 600 (data not shown). In order to link these results to a physiological process governed by ␤-adrenergic receptors in the heart, the regulation of the L-type Ca 2ϩ current (I Ca,L ) was examined with 1.8 mM [Ca 2ϩ ] in the bath. At variance with what was observed on I CNG , ISO (100 nM), and L-85 (100 M) exerted similar, maximal stimulatory effects on I Ca,L , both in Ad-CNG-infected myocytes (MOI 3000) and in non-infected cells (Fig. 3C). This indicates: (i) that the ␤-adrenergic regulation of I Ca,L was apparently not affected by CNG channel overexpression, and (ii) that limited cAMP signals elicited by a ␤-adrenergic stimulation are sufficient to fully activate L-type Ca 2ϩ channels.
Comparative Effects of Isoprenaline and L-858051 on I CNG Elicited by Two CNGA2 Mutants with Different Sensitivity to cAMP-The smaller current density obtained with ISO compared with L-85 may result from a weaker activation of the channels, or by the fact that only a given subset of channels is activated by the ␤-adrenergic agonist. To address this point, we compared the effects of ISO and L-85 on I CNG elicited by the E583M CNGA2 channel with those obtained with another CNGA2 variant, C460W/E583M, with a higher sensitivity to cAMP (12). As expected, the CRC for the effect of L-85 on I CNG in cells expressing C460W/E583M was leftward-shifted compared with E583M (Fig. 4A). The forskolin analog started to activate I CNG significantly at 10 M with C460W/E583M and at 30 M with E583M, with a maximal effect obtained at 100 M for both channels. The Hill fit of the data yielded an apparent half-maximal activation (EC 50 ) and Hill coefficient (n) of 12.7 M and 1.78 for C460W/E583M, and 36.7 M and 1.53 for E583M, respectively, while the apparent maximal effects (E max ) were similar with both channels (59.9 pA/pF versus 61.4 pA/pF, respectively). The two curves were statistically different as indicated by Fisher test (p Ͻ 0.001). The E max values reflected a saturation of the channels by cAMP since addition of the non-selective PDE inhibitor IBMX (100 M) to 300 M L-85 failed to further increase I CNG through E583M CNGA2 (data not shown). As illustrated in Fig. 4B, the results obtained with ISO differed substantially from those obtained with L-85. With both CNGA2 mutants, ISO started to increase I CNG density significantly at 10 nM, and the effect was maximal at 100 nM since a 10-times higher concentration produced no additional effect. The apparent EC 50 and n values for ISO were 12.8 nM and 1.20 for C460W/E583M, and 22 nM and 1.45 for E583M, respectively, and apparent E max was 20.5 pA/pF with C460W/ E583M and 13.6 pA/pF with E583M. Comparison of the two CRC using the Fisher test indicated that the two curves were significantly different (p Ͻ 0.05).
If one assumes that cAMP is uniformly distributed under both L-85 and ISO stimulations, then the lower saturating effect of ISO should be ascribed to its smaller capacity to raise the cAMP level. As indicated in Fig. 4C by the intersection between ISO CRC and L-85 CRC obtained with E583M, this level corresponds to that induced by Ϸ15 M L-85. However, To further confirm the implication of PKA in the control of ␤-adrenergic cAMP signals, the effect of ISO and L-85 were assessed in myocytes co-infected with Ad-CNG and Ad-PKI, an adenovirus encoding the substrate inhibitor PKI (21), each at a MOI of 600. In these cells, the basal current density and the effect of L-85 (100 M) were not significantly different (6.5 Ϯ 1.7 pA/pF and 58.7.Ϯ9.9 pA/pF, respectively, n ϭ 10) from those obtained in myocytes infected with Ad-CNG alone at MOI 600. However, similar to what was obtained with H89, the effect of ISO on I CNG density was Ϸ3 times higher in PKI-overexpressing cells (26.4 Ϯ 2.9 pA/pF, n ϭ 10) than in myocytes infected with Ad-CNG alone at MOI 600 (8.4 Ϯ 2.8 pA/pF, n ϭ 8, p Ͻ 0.005 versus control). Altogether, these results indicate that PKA acts as a negative regulator of cAMP accumulation in cardiac cells.
PDE Subtypes That Regulate cAMP Signals-Among the various targets of PKA that could control cAMP levels, PDEs appear as obvious candidates (8,12,13,14). Thus, in a first series of experiments, the effect of the non-selective PDE inhibitor IBMX on cAMP homeostasis reported by I CNG was investigated. As shown in Fig. 6A, IBMX (100 M) alone barely affected I CNG . However, it dramatically potentiated the response of I CNG to 100 nM ISO, which reached values similar to that obtained with L-85. Indeed, in 8 similar experiments, I CNG density was 7.9 Ϯ 2.3 pA/pF with 100 nM ISO alone and 53.0 Ϯ 9.0 pA/pF when the ␤-adrenergic agonist was applied in the presence of 100 M IBMX (Fig. 6B). In 6 of these myocytes, the effect of 100 M L-85 was tested, and the stimulation of I CNG was similar (47.0 Ϯ 10.8 pA/pF). Parallel experiments were conducted to investigate the consequences of PDE inhibition on L-85 activation of cAMP production (Fig. 6, C and D). cAMP Dynamics in Cardiomyocytes cardiac myocytes (20,28) to the attenuation of the ␤-adrenergic response. In the experiment shown in Fig. 7A, selective inhibition of PDE3 with cilostamide further increased I CNG prestimulated with ISO, although it did not reach the level obtained with L-85 alone. Fig. 7B shows that on average PDE3 inhibition with 1 M cilostamide exerted no significant effect on basal I CNG density, but augmented the response to 100 nM ISO about 3-fold (7.8 Ϯ 2.0 pA/pF versus 24.2 Ϯ 6.3 pA/pF, n ϭ 6). This latter value represented about 40% of the current density in the presence of 100 M L-85 (59.3 Ϯ 3.8 pA/pF, n ϭ 6). Similarly, selective PDE4 inhibition with 10 M RO 20 -1724 had no effect on basal I CNG density (Fig. 7, C and D). However, application of 10 M RO 20-1724 in the presence of ISO (100 nM) induced an increase of I CNG that was comparable to the stimulation obtained with 100 M L-85 (65.1 Ϯ 11.0 pA/pF and 65.5 Ϯ 9.8 pA/pF, respectively, n ϭ 6). These results emphasize the role of PDE3, and even more of PDE4, in regulating cAMP levels in cardiac myocytes.
Regulation of cAMP-hydrolyzing PDEs-Considering the above results, one may wonder whether PKA and PDE represent independent or linked regulators of cAMP dynamics. To answer this question, we examined whether PDE3 and PDE4 were regulated by ISO or L-85 in a PKA-dependent manner in rat ventricular myocytes. As shown in Fig. 8 (panels A and B), in 5 independent experiments ISO (100 nM) stimulated total PDE3 activity by Ϸ70% and total PDE4 activity by Ϸ75%. However, these effects were not blocked by H89 (1 M). Upon application of L-85 at 10 M (Fig. 8, panels C and D), total PDE3 activity was increased by Ϸ65%, while total PDE4 activity was augmented by Ϸ60%. As with ISO, this latter effect was insensitive to H89.
PDE Inhibition Prevents PKA Negative Feedback-The results above suggest that the potentiating effect of PKA inhibi-tion on ISO-and L-85-stimulated I CNG (Fig. 5) might not involve PDE inhibition. To check this, we tested the ability of H89 to potentiate the ␤-adrenergic stimulation of I CNG when PDEs are inhibited by IBMX (Fig. 9). In the absence of ␤-adrenergic stimulation, H89 and IBMX, used alone or in combination, had virtually no effect on I CNG (Fig. 9B). Low concentrations of the ␤-adrenergic agonist (0.3 nM) did not produce a detectable cAMP increase either alone or when PKA was blocked by H89, but cAMP accumulation became apparent in the presence of IBMX (4.1 Ϯ 0.2 pA/pF, n ϭ 5, versus 1.5 Ϯ 0.2 pA/pF for ISO 0.3 nM alone, n ϭ 3). Addition of H89 did not further increase I CNG augmented by IBMX ϩ ISO (Fig. 9B). These results suggest that PKA inhibition has no consequence on cAMP accumulation when PDEs are blocked. However, the failure of H89 to potentiate the effect of ISO alone may indicate that cAMP synthesis was too small for a retrocontrol inhibition to occur or to be detected. Thus, the same protocol was repeated using a higher concentration of ISO (30 nM). In the representative experiment of Fig. 9A, the activation of the CNG current by 30 nM ISO was enhanced by H89, as well as by IBMX. However, in the presence of the PDE inhibitor, H89 was ineffective in potentiating the ␤-adrenergic stimulation. On average, ISO (30 nM) increased I CNG density to 8.1 Ϯ 1.2 pA/pF alone and to 18.2 Ϯ 3.9 pA/pF in the presence H89 (1 M, n ϭ 9). In the presence of IBMX (100 M) the effect of ISO reached 47.6 Ϯ 4.9 pA/pF (n ϭ 9) and could not be further augmented by simultaneous H89 application (45.4 Ϯ 5.6 pA/pF, n ϭ 6). This was not caused by saturation of the CNG channels with cAMP because a subsequent application of L-85 (100 M) was still able to further increase the current amplitude (62.6 Ϯ 4.8 pA/pF, n ϭ 5). Thus, PDE inhibition with IBMX eliminated the effect of H89 on ISO-stimulated I CNG .

DISCUSSION
The use of recombinant CNG channels as cAMP biosensors was developed in a series of elegant studies in model cell lines (7,12,13,29). Here, we have applied this methodology to differentiated adult cardiomyocytes in primary culture. The results obtained confirm and extend the suggestions made in a previous study of ours (8). We show that whole cell patch-clamp recording of I CNG provides a reliable readout of subsarcolemmal cAMP fluctuations in a single cardiac myocyte. The use of two CNGA2 variants with different affinities for cAMP indicates that the ␤-adrenergic cAMP signal is compartmentalized in these cells. PKA activation of plasma membrane PDE negatively regulate cAMP increases triggered by ␤-AR or direct AC activation. This negative feedback controls global cAMP homeostasis beneath the membrane and contributes to the maintenance of restricted hormonal cAMP signals.
Although molecular evidence exists for the presence of CNG channels in the heart (30 -33) our immunofluorescence experiments did not reveal a significant expression of CNGA2 in native rat ventricular myocytes. Moreover, in patch-clamp experiments, a cAMP-activated current was never observed, either in native cells or in Ad-GFP-infected myocytes. These results exclude that the signal measured in Ad-CNG-infected myocytes was contaminated by an endogenous CNG-like current. In these cells, immunolocalization attested expression of CNGA2. The observed perinuclear localization could reflect an ongoing maturation process in the Golgi apparatus, and the staining at the ends of the cells, an accumulation near intercalated disks. The striated pattern may represent a facilitated access to less constrained regions of the sarcomere, such as the I-bands. However, the important point is that although not totally homogeneous, CNGA2 distributed throughout the plasma membrane and was thus expected to detect cAMP variations occurring anywhere below.
Because CNG channels are blocked by external and internal Ca 2ϩ , the results presented in this study were obtained at nominal external [Ca 2ϩ ] and internal pCa 8.5. Since Ca 2ϩ inhibits cAMP synthesis by cardiac AC (34), the cAMP signals reported herein are certainly more robust than at physiological [Ca 2ϩ ]. Nevertheless, a larger cAMP accumulation with forskolin than with ISO was also shown to occur in intact cardiac myocytes in the presence of external Ca 2ϩ (35,36). Similarly, it seems very unlikely that the regulatory mechanisms described in this study are caused by the low Ca 2ϩ context. Indeed, overexpression of PKI was shown to potentiate the effect of ISO on total cAMP in the presence of external Ca 2ϩ (35). Our initial observation that L-85 had a much stronger effect than ISO on I CNG , while both agents activated I Ca,L maximally, can intuitively be explained in two ways. The first is that ␤-adrenergic stimulation produces a modest [cAMP] elevation that activates CNG channels only partially while maximally activating Ca 2ϩ channels. This could result from the lower affinity of E583M CNGA2 for cAMP (Ϸ10 M) compared with the affinity of PKA (Ϸ100 nM) (7). The other possibility is that ␤-adrenergic stimulation activates only a subset of CNG channels located next to the ␤-adrenoreceptor-coupled adenylyl cyclases and Ca 2ϩ channels, whereas L-85 activates all cyclases and thus a larger population of CNG channels.
Comparison of the CRC obtained for each agonist using two CNGA2 mutants whose cAMP sensitivities differ by a factor of Ϸ10 (12) suggests that both possibilities might occur (Fig. 4). Indeed, the stronger activation of C460W/E583M than E583M by each concentration of ISO (Fig. 4B) means that the [cAMP] produced by the ␤-adrenergic stimulation does not fully activate the channels. However, this difference is too small to involve a large number of channels. This is attested by the larger variation observed between both CNGA2 variants with L-85 (Fig. 4C). The data are consistent with ISO generating discrete microdomains of high [cAMP] where a limited number of channels are saturated. In this scheme, the dose dependence of I CNG to the ␤-adrenergic agonist may be viewed as an increased cAMP microdomain number with increasing concentrations of the agonist. Because maximal concentrations of forskolin and G s similarly activate AC (37,38), the ratio of the current densities of the E583M CNGA2 channel at maximal ISO and L-85 concentrations provides an estimate of the fraction of CNG channels, and thus the fraction of membrane area that is activated by the ␤-adrenergic agonist. By doing so, we can estimate that the maximal ␤-adrenergic response spreads over Ϸ25% of the total sarcolemmal area. This experimental value is in agreement with the prediction of a recently developed model for ␤-adrenergic control of cardiac myocyte contractility (39).
Our results also underline the importance of regulatory mechanisms that prevent the spreading of small and diffusible cAMP molecules upon ␤-adrenergic receptor or direct AC stimulation. In this study, we focused on postreceptor mechanisms, which could presumably apply to ISO and L-85. This of course does not exclude that ␤-adrenergic receptor desensitization, in particular mediated by ␤-adrenergic receptor kinase, participates in the differential effect of ISO and L-85 on I CNG .
First, we show that PKA exerts a tonic inhibition of cAMP accumulation upon ISO or submaximal concentration of L-85. A similar result was reported by Cui and Green (35) in avian embryonic ventricular myocytes, whereas such regulation was not observed in frog cardiomyocytes, suggesting species differences (40). Second, in agreement with a number of previous studies, we identified PDE3 and PDE4 as critical regulators of cAMP signals in cardiac myocytes. This finding is also consistent with the predominant impact of PDE3 and PDE4 activities for the ␤-adrenergic regulation of I Ca,L in rat ventricular myocytes (20) and the positive inotropic effect of PDE3 and PDE4 inhibitors in rats (41,42).
These regulations can be overcome by maximal stimulation of all AC with 100 M L-85 but not by saturation of the ␤-adrenergic receptor with ISO (Fig. 4). However, when disrupted by H89, IBMX, or Ro 201724, ISO and L-85 stimulations cannot be distinguished. This strongly suggests that these regulations ensure global homeostasis of cAMP but are also fundamental to confine receptor-triggered cAMP signals.
These findings raise the question of whether PKA acts through cAMP-PDE or through other targets. PKA activation of PDE3 (43)(44)(45)(46) and PDE4 (17, 46 -48) has been well characterized in vitro, although the evidence for this in the heart remains scarce (49,50). Here, while ISO and L-85 clearly enhanced the total PDE3 and PDE4 activities, these effects were not blocked by H89 (Fig. 8). This result is at variance with studies by Oki et al. (51) and MacKenzie et al. (52) showing that H89 prevents the stimulatory effect of cAMP-elevating agents on PDE4. However, these were performed in cell lines and, in the case of MacKenzie et al. (52), after overexpression of long PDE4 isoforms. PKA regulation of endogenous cAMP-PDE in adult cardiomyocytes is likely to be more difficult to prove since only certain PDE isoforms are phosphorylated by PKA (17,53,54), and PDEs can be regulated by other mechanisms, for instance involving other kinases (17,55,56).
Thus, while our electrophysiological data suggest that PDE activities controlling cAMP beneath the membrane are regulated by PKA, this was not apparent when total cellular activities of PDE3 and PDE4 were measured. These results may be reconciled by two kinds of explanations. Membrane-associated cAMP PDEs represent Ϸ20 -30% of the total cAMP hydrolyzing activity in guinea pig heart (57) and rat ventricular myocytes (preliminary results from this study, data not shown), an estimation that includes PDEs associated to intracellular organelles such as SR (58). Thus, one could imagine that a PKA-independent activation of "deep" pools of PDE might mask a PKA-dependent activation of the minor PDE fraction associated to the plasma membrane. On the other hand, one should not really expect to be able to mimic in an in vitro assay a regulation that presumably requires much cellular integrity. Indeed, it is possible that recruitment of the phosphorylated PDEs by a scaffold protein is part of the activation process (see below). Upon breaking the cells, such a complex would fall apart and the overall stimulation would be underestimated and might thus not be readily inhibitable by H89.
Alternatively, another mechanism but PDE activation could be involved in PKA-mediated negative feedback of cAMP signals revealed by CNG channels. In particular, adenylyl cyclases type V and VI, the major isoforms present in the adult heart (59), can be phosphorylated and inhibited by PKA (46,60). In order to check this hypothesis, the effect of H89 on ISO-stimulated I CNG was assessed in the presence of IBMX. Since IBMX completely prevented the effect of H89 (Fig. 9), AC inhibition by PKA was unlikely to play a role. This result demonstrates that cAMP-PDEs are the relevant PKA targets for the control of subsarcolemmal cAMP gradients in cardiac myocytes.
This study documents a typical functional consequence expected from multimolecular signaling complexes of PKA and PDE such as shown recently for the isoform PDE4D3 (50,61). We provide a direct validation of the negative feedback model suggested by such a molecular organization and prove that it applies to subsarcolemmal cAMP dynamics. Dodge et al. (50) only found a modest PDE activity associated with AKAP15/18, the AKAP thought to associate with L-type Ca 2ϩ channels at the membrane (62,63). This fits well with the fact that limited cAMP increases induced by ISO fully activate I Ca,L , and suggests that the feedback loop identified here is essentially useful to prevent broadcast of cAMP outside of microdomains where it is needed (8,14). This suggests that other proteins are responsible for targeting PDE3 and PDE4 around these signaling microdomains. Other AKAPs expressed in heart are known to associate to PDE4, such as AKAP450 (61,64) or myomegalin (65). Alternatively, ␤-arrestin could be responsible for PDE4 translocation to the ␤-adrenergic receptors (66 -68), but this deserves further investigation in adult cardiac cells.
In conclusion, we show that PKA activation of cAMP-PDE limits subsarcolemmal cAMP increases in intact cardiac cells. This mechanism can be triggered by ␤-adrenergic receptor stimulation or nonspecific activation of cellular AC and thus appears to control global cAMP homeostasis and to be required for compartmentation of ␤-adrenergic receptor cAMP signals. Further experiments are required to identify the PDE subtype(s) involved and to examine whether cAMP signals elicited by other G s -coupled receptors are similarly regulated. The methodology presented here should contribute to a better understanding of cyclic nucleotides pathways in healthy and diseased heart cells.