Binding of cGMP to GAF Domains in Amphibian Rod Photoreceptor cGMP Phosphodiesterase (PDE)

Retinal cGMP phosphodiesterase (PDE6) is a key enzyme in vertebrate phototransduction. Rod PDE contains two homologous catalytic subunits (Pαβ) and two identical regulatory subunits (Pγ). Biochemical studies have shown that amphibian Pαβ has high affinity, cGMP-specific, non-catalytic binding sites and that Pγ stimulates cGMP binding to these sites. Here we show by molecular cloning that each catalytic subunit in amphibian PDE, as in its mammalian counterpart, contains two homologous tandem GAF domains in its N-terminal region. In Pγ-depleted membrane-bound PDE (20–40% Pγ still present), a single type of cGMP-binding site with a relatively low affinity (K d ∼ 100 nm) was observed, and addition of Pγ increased both the affinity for cGMP and the level of cGMP binding. We also show that mutations of amino acid residues in four different sites in Pγ reduced its ability to stimulate cGMP binding. Among these, the site involved in Pγ phosphorylation by Cdk5 (positions 20–23) had the largest effect on cGMP binding. However, except for the C terminus, these sites were not involved in Pγ inhibition of the cGMP hydrolytic activity of Pαβ. In addition, the Pγ concentration required for 50% stimulation of cGMP binding was much greater than that required for 50% inhibition of cGMP hydrolysis. These results suggest that the Pαβ heterodimer contains two spatially and functionally distinct types of Pγ-binding sites: one for inhibition of cGMP hydrolytic activity and the second for activation of cGMP binding to GAF domains. We propose a model for the Pαβ-Pγ interaction in which Pγ, by binding to one of the two sites in Pαβ, may preferentially act either as an inhibitor of catalytic activity or as an activator of cGMP binding to GAF domains in frog PDE.

phosphorylations are presumed to occur in inactive PDE; however, under dark conditions, P␥ phosphorylation has not been observed in vivo (27). On the other hand, P␥ phosphorylation by Cdk5 appears to be important for the turnoff of GTP-T␣-activated PDE. In this mechanism (26 -29), after P␥ is complexed with GTP-T␣, Cdk5 phosphorylates Thr 22 in P␥, and the phosphorylated P␥ is released from GTP-T␣. Pro 20 , Pro 23 , and Arg 24 are also crucial for P␥ phosphorylation. Phosphorylated P␥, which has higher affinity for P␣␤ compared with non-phosphorylated P␥, returns to P␣␤ and inhibits the cGMP hydrolytic activity of P␣␤. This P␥ phosphorylation is also observed in vivo (27). The in vivo phosphorylation has two phases. In the rapid phase, P␥ phosphorylation was detected when ϳ0.03% of the rhodopsin was bleached in 1 s, and ϳ4% of total P␥ was estimated to be phosphorylated after Ͻ0.3% of the rhodopsin was bleached in Ͻ10 s. In the slow phase, ϳ10% of P␥ was phosphorylated after bleaching of ϳ10% rhodopsin, and ϳ16% of P␥ was phosphorylated after bleaching of 20% rhodopsin. The phosphorylated P␥ was rapidly dephosphorylated in Ͻ1 s when the light was turned off. These observations suggest that GTP-T␣-activated PDE can be deactivated without GTP hydrolysis and that the lifetime of GTP-T␣-activated PDE can be regulated by P␥ phosphorylation. Thr 22 in P␥ was also reported to be phosphorylated by MAPK (30). Under the published conditions (100 M ATP and 5 h of incubation), ϳ12% of P␥ was phosphorylated in situ. However, as discussed by Paglia et al. (30), the significance of this observation is unclear because the reported kinetics of phosphorylation are not consistent with the rate of regulation of visual signal transduction. ADP-ribosylation of P␥ has also been suggested as a potential mechanism for regulation of the cGMP hydrolytic activity of P␣␤ in frog ROS membranes (31,32). In this P␥ modification, one of two arginine residues (Arg 33 and Arg 36 ) is ADP-ribosylated by two endogenous ADP-ribosyltransferases. Interestingly, P␥ complexed with GTP-T␣ is not a substrate for these enzymes, and ADP-ribosylated P␥ lacks the ability to interact with GTP-T␣. Together, these various P␥ modifications suggest that P␥ has numerous functional amino acid residues in the N-terminal half and that these residues may be linked to various functional domains.
In addition to a catalytic site(s), P␣␤ has high affinity, cGMP-specific, non-catalytic binding sites (33)(34)(35)(36)(37). Recently, these non-catalytic sites were termed GAF domains, a ubiquitous motif for cGMP binding, because of their presence in cGMP-regulated cyclic nucleotide PDEs, certain adenylyl cyclases, and the bacterial transcription factor FhlA (38,39). In frog ROS membranes, it has been shown that the binding of cGMP to non-catalytic sites is stimulated by P␥ (40 -44). It has been reported that P␥ is also involved in regulation of cGMP binding to GAF domains in mammalian P␣␤ (45)(46)(47). Thus, it is reasonable to conclude that P␥ affects P␣␤ in two different ways: inhibition of the cGMP hydrolytic activity of P␣␤ and regulation of cGMP binding to GAF domains in P␣␤. However, whether a single P␥ can affect both functions simultaneously is controversial.
Although cGMP binding to non-catalytic sites in P␣␤ was initially found in frog PDE (33), and regulation of cGMP binding by P␥ has been mainly studied using frog PDE systems, GAF domains in frog P␣␤ have not been identified. In addition, the amino acid sequence of frog P␥ has not been determined. Thus, bovine P␥ and its mutants have been used to investigate P␥ regulation of cGMP binding to frog P␣␤. In this study, using molecular cloning of frog PDE subunits, we have identified GAF domains in frog P␣␤ and found that frog P␥ is almost identical to bovine P␥. Then, using various P␥ mutants, we have also identified P␥ amino acid residues important for stim-ulation of cGMP binding to GAF domains in frog P␣␤. We focused mainly on amino acid residues involved in various post-translational modifications of P␥; however, we also investigated amino acid residues in other regions to make a comprehensive list of P␥ sites involved in stimulation of cGMP binding. The effects of these P␥ mutants on the cGMP hydrolytic activity of P␣␤ were also determined. Our results strongly suggest that the P␣␤-P␥ interaction required for stimulation of cGMP binding to GAF domains in P␣␤ is different from that required for inhibition of cGMP hydrolysis by P␣␤ in frog P␣␤␥␥. The suggestion is consistent with previous results (43,47,48). Based on this study, we propose a new model for the P␣␤-P␥ interaction in frog P␣␤␥␥.

EXPERIMENTAL PROCEDURES
Materials-Chemical reagents were purchased from the following sources: [ 3 H]cGMP from PerkinElmer Life Sciences; GTP and cGMP from Roche Molecular Biochemicals; leupeptin, pepstatin A, phenylmethylsulfonyl fluoride, benzamidine, and snake venom (Ophiophagus hannah) from Sigma; and AG 1-X2 (100 -200 mesh) from Bio-Rad. Native frog rod P␥ was isolated as described previously (18). Recombinant bovine rod P␥ was expressed in Escherichia coli and purified as described previously (31). Frog GDP-T␣ and the transducin ␤␥ subunits were isolated as described previously (49).
Cloning of Frog Rod P␣␤ and P␥-All DNA manipulations were carried out using standard procedures. Frog rod P␣ cDNAs were isolated by screening a frog (Rana pipiens) retinal cDNA library with a 1.1-kb fragment encoding mouse P␤ cDNA under reduced stringency (50). Plaques (1 ϫ 10 5 ) were screened with fragments of mouse P␤ cDNAs under reduced stringency (25% formamide and 6ϫ SSC (0.15 M NaCl and 0.015 M trisodium citrate (pH 7.0)) at 35°C for 12 h). Filters were washed four times with 1ϫ SSC and 0.1% SDS at 35°C and exposed to Kodak XAR5 films. Positive clones were purified, and the inserts were incorporated into pBluescript SK(Ϫ) (Stratagene) according to the manufacturer's protocol. Frog P␤ cDNAs were isolated by PCR and library screening. Briefly, a cDNA fragment (370 bp) was amplified from the same frog library using degenerate primers FPB9 (5Ј-TGYCAYGAYATYGAYCAYMGNGG) and FPB4Y (5Ј-CATGCNGT-CATCATCATNGCC). The fragment was cloned into pCR2.1 vectors (Invitrogen), sequenced, and identified as a partial clone encoding P␤. Additional frog P␤ clones were identified from the library using the fragment as a probe. The remaining sequence was amplified from the library by nested PCR using P␤-specific primer FPBB1 (5Ј-CAAAT-GGGTTGATTACTTATCG) and universal primer M13R and then nested primers FPBB7 (5Ј-CCTCTTCTTCTTCTTCCTTC) and T7. Frog rod P␥ cDNAs were isolated by screening the same frog retinal cDNA library with 300-bp mouse rod P␥ cDNA as described above. The fragments were cloned into the pCR2.1 vector for sequencing.
Preparation of P␥-depleted Membrane-bound PDE-Under dim red light, ROS were prepared from dark-adapted bullfrogs (Rana catesbeiana or Rana grylio) as described previously (18). Bleached ROS membranes from 40 frogs were suspended in 6 -8 ml of Buffer A (10 mM Tris-HCl (pH 7.5), 5 mM dithiothreitol, 5 mM MgCl 2 , 10 M CaCl 2 , 5 M leupeptin, 5 M pepstatin A, 100 M phenylmethylsulfonyl fluoride, 1 mM benzamidine, and 200 mM NaCl) and passed seven times through a 21-gauge needle. Membranes were then isolated by centrifugation at 200,000 ϫ g for 15 min at 4°C. ROS membranes were washed seven times in this manner to minimize the contamination of soluble proteins in membranes. PDE and transducin were not extracted under these conditions. P␥ was then extracted with GTP-T␣ (18) by washing the membranes seven times with 6 ml of Buffer A containing 400 M GTP and 1 mM EGTA. Preliminary experiments indicated that these extensive washings were required for the maximum extraction of P␥, although most of the P␥ release was observed in the first and second washes. Hydrolysis-resistant GTP analogs were not used to avoid the possible effect of residual T␣ complexed with the GTP analog on PDE activity in membranes. These membranes were further washed two times with 6 ml of Buffer A to remove traces of GTP, suspended in the same buffer (5.6 mg of protein/ml), and stored at Ϫ70°C. We assume that residual GTP, if present, was hydrolyzed to GDP by residual T␣ and that the high PDE activity measured was due totally to the release of P␥ rather than activation by residual GTP-T␣. PDE activity in membranes stored at Ϫ70°C was stable for at least 6 months under these conditions. T␣ (Ͼ90%) and P␥ (ϳ60 -80%) were released from membranes during the washing; however, only negligible amounts of P␣␤ were released (18). Thus, these membranes primarily contained P␥-less PDE (P␣␤␥ and/or P␣␤) and were termed P␥-depleted membrane-bound PDE. EGTA (1 mM) was added to Buffer A during preparation of P␥-depleted membrane-bound PDE because our preliminary study indicated that, in the absence of Ca 2ϩ , more P␥ was released then in the presence of CaCl 2 . 3 Buffer A also contained 200 mM NaCl because 200 mM salt increases the release of P␥ from P␣␤ by GTP-T␣ (42). Control membranes were prepared with Buffer A containing 1 mM cGMP, which prevents P␥ release (51). Membranes were further washed twice with Buffer A to eliminate residual cGMP.
Equilibrium Binding of cGMP to P␥-depleted Membrane-bound PDE-Frog P␥-depleted membrane-bound PDE (ϳ15 g of protein/100 l) was incubated with [ 3 H]cGMP (1 mCi/ml) for 30 min on ice, and [ 3 H]cGMP bound to P␥-depleted PDE was assayed using a filter method (33,40,42,43). To ensure cGMP binding to high affinity, cGMP-specific, non-catalytic sites in P␣␤, 0.5 M [ 3 H]cGMP was used in the presence of 1 mM 1-methyl-3-isobutylxanthine and 2 mM EDTA. Under these conditions, ϳ5% (at most) of the added cGMP was hydrolyzed, and cGMP binding reached the maximum level when ϳ1 M P␥ (recombinant and purified) was added. Under these conditions, the linearity of cGMP binding was established against amounts of P␥-depleted membranebound PDE (5-30 g of protein/100 l), and the P␥ concentration required for the maximum level of cGMP stimulation did not vary over the range of membrane-bound PDE added. When excess amounts of P␥-depleted membrane-bound PDE were added to the reaction mixture, washing of membrane filters might not be sufficient in the measurement of cGMP binding. However, under these conditions, the level of [ 3 H]cGMP binding will be unusually high and irregular, and these kinds of failures can be identified instantly. cGMP-specific binding was also confirmed by competitive binding of [ 3 H]cGMP in the presence of 100 M cAMP as described previously (33). However, the effect of cAMP is not shown in this report because it has been shown that, under our conditions, only cGMP binds to non-catalytic sites in P␣␤ in frog ROS membranes (33,40). Under our conditions, the increase by added P␥ in cGMP binding to these non-catalytic sites is not due to P␥ inhibition of the cGMP hydrolytic activity of P␥-depleted PDE (43). In addition, we acknowledge the possibility that the P␣␤ conformation may be changed by the binding of P␥ to P␣␤ (52), and the altered conformation might alter P␣␤ binding to a membrane filter during assay of cGMP binding. However, P␥-depleted PDE appeared to firmly associate with membranes, and addition of P␥ did not weaken the association. 4 Thus, it is expected that P␥-depleted membrane-bound PDE, with or without P␥, is similarly trapped by filters and that the level of [ 3 H]cGMP binding is not changed during washing filters. For Scatchard plot analysis, binding of cGMP to P␥-depleted membrane-bound PDE was carried out in the presence of 0.01, 0.025, 0.5, 0.075, 0.1, 0.2, 0.5, and 1.0 M [ 3 H]cGMP with or without 1 M P␥ or its mutants. Data were analyzed graphically to obtain K d values, but similar results were obtained using the GraphPAD computer program Prism. Siliconized tubes and pipette tips were used in all experiments with P␥ and its mutants.
Measurements of P␥ Inhibitory Activity-PDE activity was measured by incubating 1 mM cGMP (with ϳ0.9 Ci of [ 3 H]cGMP) and P␥-depleted membrane-bound PDE (50 -100 ng of protein) in 100 l of buffer containing 20 mM Tris-HCl (pH 7.5), 5 mM MgCl 2 , and 1 mM EGTA for 10 min (at 33°C). Although the content of P␥-depleted membranebound PDE in the reaction mixture was small, we did not add any protein to stabilize the membrane-bound PDE because P␥-depleted PDE appears to be stable during incubation due to tight binding to membranes. 4 The enzyme reaction was initiated by addition of cGMP. The PDE inhibitory activities of P␥ and its mutants were measured under the conditions that the linearity of PDE activity was established against amounts of P␥-depleted membrane-bound PDE, and ϳ30% (the maximum) of the added cGMP was hydrolyzed during incubation. The concentrations of P␥ and its mutants used were 1 ϫ 10 Ϫ10 to 1 ϫ 10 Ϫ6 M. Preliminary studies indicated that these concentrations were enough to detect inhibition of cGMP hydrolytic activity by wild-type P␥ in P␥-depleted membrane-bound PDE. The reaction was terminated by incubation (80°C) for 2.5 min. Twenty-five l of a snake venom solution (1 mg/ml of 10 mM Tris-HCl (pH 7.5) containing 1 mM dithiothreitol and 5 mM MgCl 2 ) was added to the reaction mixture, and the mixture was incubated for 15 min (at 33°C). The resulting [ 3 H]guanosine was isolated using AG 1-X2 resin (53), and its radioactivity was quantified. Each reaction was carried out in duplicate. For certain P␥ mutants, the concentrations required for 50% inhibition of the cGMP hydrolytic activity of P␣␤ were found to be similar to that for wild-type P␥ using P␥-depleted membrane-bound PDE prepared similarly (26,31,43). These P␥ mutants include P␥(N16Sub) (43), P␥(P20G) (26), P␥(P23G) (26), P␥(P20G,P23G) (26), P␥(R24K) (31), P␥(R24E) (31), P␥(R33K) (31), and P␥(R36K) (31). Because we have previously reported the characterization of these P␥ mutants and have replicated the previous findings as part of this study, we have not shown all of the individual IC 50 values of each P␥ mutant in this study, although such characterization is crucial and has been done.
Other Analytical Methods-Protein concentrations were assayed with bovine serum albumin as a standard (54). The amounts of P␥ were assayed as described previously (18). In this method, the amounts of P␥ (frog or bovine) were on the linear range up to at least 1.25 g. SDS-PAGE was performed as described previously (48). It should be emphasized that all experiments were carried out more than two times, and the results were similar. Moreover, individual points obtained in all experiments represent the average values of duplicate assays. Data shown are representative of these experiments.

Cloning of Frog Rod PDE Subunits
Controversies have arisen regarding possible differences between mammalian and amphibian PDE systems in regulation of cGMP binding to non-catalytic sites in P␣␤ (35,40,41). Differences in sequence and functional domains between mammalian and amphibian PDE subunits might be responsible for the discrepancy. However, recent studies of cGMP binding and 4 V. A. Bondarenko and A. Yamazaki, unpublished data.
a Underlined letters indicate mutation sites.
its regulation (41)(42)(43)(44)55) have been based on the assumption that frog catalytic subunits have non-catalytic sites similar to those found in bovine catalytic subunits (36,37), and that frog rod P␥ has functional domains similar to those in bovine rod P␥ (9 -11). Thus, it was important to identify both these noncatalytic sites in frog P␣␤ and the functional domains of frog rod P␥.
To characterize non-catalytic sites in frog P␣␤, we cloned cDNAs encoding these catalytic subunits from a frog (R. pipiens) retinal library using a combination of direct screening with mammalian PDE cDNA probes, direct PCR from the library, and reverse transcription-PCR. The open reading frames of these cDNAs predict that unprocessed frog P␣ and P␤ are polypeptides consisting of 866 and 857 amino acids, respectively, and that, as expected, these subunits are homologous (overall 72-76%) to their bovine counterparts (36,37). Both catalytic subunits contain two tandem repeats of the GAF domains (ϳ90 amino acids) in their N-terminal regions, and the consensus sequence for cGMP binding (N(K/R) X n GX n FX 3 DE) is located in these repeats (Fig. 1A). These repeats are highly homologous to their bovine counterparts (78% for the first repeat and 94% for the second repeat in P␣ and 84% for the first repeat and 94% for the second repeat in P␤). We also found that both frog P␣ and P␤ contain a CAAX box at the C terminus (CSIL in P␣ and CRIL in P␤), signaling for isoprenylation (Fig. 1B). These CAAX box sequences predict that frog P␣ and P␤ are geranylgeranylated and farnesylated, respectively, opposite the patterns observed in mammalian PDE (56,57). Consequently, frog P␣ is predicted to be more tightly associated with ROS membranes than frog P␤. The biological significance of the difference in prenylation patterns between mammalian and amphibian is unclear.
Frog rod P␥ cDNA predicts a polypeptide of 87 amino acid residues (Fig. 1C). Only four amino acid residues (Ala 6 , Pro 8 , Val 17 , and Ala 21 ) are considered to be non-conservative substitutions compared with bovine rod P␥ (58) and other mammalian rod P␥ subunits (59,60). Thus, we conclude that all known rod P␥ subunits are highly homologous. This conclusion is also supported by biochemical data showing that P␥ subunits from frog (R. catesbeiana) and bovine ROS have very similar abilities to inhibit the PDE activity of P␣␤ and to stimulate cGMP binding to P␣␤ in frog P␥-depleted membrane-bound PDE (42). In this study, we further confirmed that P␥ subunits from different frog species (R. catesbeiana and R. pipiens) similarly inhibited PDE activity and stimulated cGMP binding in P␥depleted membrane-bound PDE (data not shown). Thus, bovine FIG. 1. Alignment of predicted amino acid sequences of frog (R. pipiens) and bovine rod PDE subunits. The deduced amino acid sequences are shown in single-letter code. The following residues are considered identical or conservative substitutions: Arg ϭ Lys; Tyr ϭ Trp ϭ Phe; Ala ϭ Gly; Ile ϭ Leu ϭ Val; Glu ϭ Asp; Gln ϭ Asn; and Ser ϭ Asn. Alignment was performed with ClustalX (82) in a multiple alignment mode using the default parameters. The figure was drawn using ALSCRIPT (83). A, alignment of tandem repeats (GAF domains) of frog (f) P␣ (GAF1, residues 133-223; and GAF2, 339 -432), frog P␤ (GAF1, residues 132-222; and GAF2, 336 -431), bovine (b) P␣ (GAF1, residues 133-223; and GAF2, 339 -432), and bovine P␤ (GAF1, residues 131-221; and GAF2, 337-430). Residues that are identical in all repeats are shown on a black background, and conservative substitutions are on gray background. The consensus sequence for cGMP binding, N(K/R)X n GX n FX 3 DE, is boxed. B, alignment of the C-terminal regions of frog and bovine P␣ and P␤. The C-terminal CAAX box is shown, and the predicted post-translational modifications are indicated as gg (geranylgeranylation) and fa (farnesylation). The numbers represent the lengths of the unprocessed predicted amino acid sequences of individual subunits. C, predicted amino acid sequence of frog rod P␥ aligned with the bovine rod P␥ sequence (58). The symbols indicate the amino acid residues investigated in this study: q, amino acid residues involved in stimulation of cGMP binding; E, amino acid residues not involved directly in the stimulation. Sites 1-4 indicate the domains involved in stimulation of cGMP binding to P␣␤ in frog P␥-depleted membrane-bound PDE. All amino acid residues between positions 11 and 18 and positions 78 and 87 were not necessarily involved in stimulation of cGMP binding because these residues were identified by deletion. We also noted that the amino acid residues 1-10 were identified by a deletion mutant, P␥(N16Del), and that Lys 39 , Lys 41 , Lys 44 , and Lys 45 were identified using the quadruple mutant P␥(K39L,K41L,K44L,K45L). Thus, we cannot rule out the possibility that the P␥ ability to stimulate cGMP binding was lost by combination of these amino acid residues. and frog P␥ subunits have identical biological properties, and the different amino acid residues in these P␥ subunits are not crucial for the biological properties of P␥. Together, these observations indicate the validity of the use of bovine rod P␥ and its mutants in the frog rod PDE system.

P␥-depleted Membrane-bound PDE Used to Measure P␥ Stimulation of cGMP Binding to Non-catalytic Sites in P␣␤
The P␥-depleted membrane-bound PDE used in this study contained 20 -40% of the P␥ present in control membranes, and its cGMP hydrolytic activity was ϳ10 times that in control membranes (data not shown). This PDE activity was sensitive to bovine and frog P␥ subunits (purified or E. coli cellexpressed), and 50% inhibition was achieved by ϳ1.0 nM bovine P␥, as shown below (see Fig. 6A). These observations are consistent with previous data (18,42) that were obtained using P␥-depleted membrane-bound PDE prepared similarly. Thus, we conclude that the P␥ interaction required for inhibition of the PDE activity of frog P␣␤ is significantly weaker than the P␥ interaction with bovine P␣␤ (IC 50 ϭ 10 pM (at most)) (47,61). The reason for the differences observed between the bovine and frog systems may lie in the use of soluble versus membranebound PDEs. Typically, the bovine P␣␤-P␥ interaction for inhibition of PDE activity has been measured using trypsinactivated PDE, which lacks the ability to interact with membranes due to loss of the isoprenylated C terminus (61). We have used frog P␥-depleted PDE, which is activated by GTP-T␣ (without the use of trypsin) and binds tightly to membranes. 4 It seems possible that the kinetic properties of soluble PDE may be different from those of PDE bound tightly to membranes.
Interpretation of the data obtained using trypsin-activated, soluble bovine PDE (47) has given rise to the concept that the two P␥ subunits that bind to P␣␤ are both involved in inhibition of PDE catalytic activity. In such a model, P␥ depletion by treatment of membranes with GTP␥S-bound T␣ (GTP␥S-T␣), as used in our study, would remove one P␥ from the lower affinity site, whereas the second higher affinity site would still be occupied by one P␥. Thus, by extrapolation from the soluble bovine system, high concentrations of P␥ would be expected to be necessary for inhibition of PDE activity in frog P␥-depleted membrane-bound PDE. However, Melia et al. (62) have indicated the importance of membrane binding of the phototransduction proteins in their interactions and have shown that, on positively charged membranes, GTP-T␣ activates PDE maximally when GTP␥S-T␣ binds to PDE at a 1:1 molar ratio. These authors concluded that, under their conditions, the one P␥binding site for GTP-T␣ on PDE that stimulates catalysis must have a higher affinity for GTP-T␣ than another sites that are silent with respect to activation of PDE. If these concepts can be applied to frog P␥-depleted membrane-bound PDE, the explanation using two P␥ subunits for inhibition of PDE catalytic activity appears to be unlikely in membrane-bound systems.
The level of cGMP binding to non-catalytic sites in P␣␤ in P␥-depleted membrane-bound PDE was low, but consistently detected (Figs. [2][3][4][5]. Scatchard plots indicate that P␥-depleted membrane-bound PDE contains one kind of cGMP-binding site and that its affinity for cGMP is relatively low (K d ϭ 102 nM) (the average K d of two K d values shown in Fig. 2, B and D). It is unclear whether cGMP binding observed is due to stimulation of cGMP binding by residual P␥ or represents the basal binding activity of P␣␤. Additional P␥ stimulated the level of cGMP binding. P␥ (1 M) stimulated cGMP binding ϳ5-fold (average of the data shown in Figs. 2 (A and C), 3 (A and C), 4A, and 5A). When cGMP binding was maximally stimulated by P␥, ϳ2 mol of [ 3 H]cGMP were estimated to be bound to 1 mol of P␣␤. This stoichiometry is based on the assumption that, after washing with GTP, Ͼ95% of the protein in frog membranes is rhodopsin and that the molecular ratio between rhodopsin and PDE in frog ROS membranes is ϳ300:1 (34). Analysis of cGMP binding by Scatchard plots indicates that, in the presence of 1 M P␥, the average K d value for cGMP is 10 nM (Figs. 2 (B and D), 3, (B and D), 4B, and 5B). We also found that 50% stimulation was accomplished by ϳ160 nM P␥ (the average P␥ concentration of the data shown in Figs. 2 (A and C), 3 (A and C), 4A, and 5A). These observations indicate that, in frog membrane-bound PDE, the P␣␤-P␥ interaction for stimulation of cGMP binding is weak, consistent with previous studies (40 -44). As described above, the P␣␤-P␥ interaction for inhibition of the cGMP hydrolytic activity of frog PDE appears to be also much weaker than that of bovine PDE. These weak interactions may be one of the reasons for the easy extraction of P␥ by GTP-T␣ from frog membrane-bound PDE (18).
It should be emphasized that the P␥ concentration required for 50% inhibition of cGMP hydrolysis (ϳ1 nM) and the P␥ concentration required for 50% stimulation of cGMP binding to P␣␤ (ϳ160 nM) differ by 2 orders of magnitude. We believe that this difference is significant and reliable based on the following reasons. (a) These EC 50 values (concentration yielding 50% stimulation of binding or inhibition of activity) were obtained using the same preparations of P␥ and P␥-depleted membranebound PDE, indicating that this difference is not due to the different sample preparations. (b) The EC 50 value for P␥ inhibition was obtained in the linear range of PDE activities against P␣␤ concentrations. When P␥ inhibition was measured outside of the linear range of PDE activity, higher concentrations of P␥ were required to detect its inhibition of PDE activity (data not shown). However, values obtained under those conditions are not related to the EC 50 value for P␥ inhibition. It is true that the amounts of P␥-depleted membrane-bound PDE used for the measurement of PDE activity and cGMP binding to GAF domains are different. Thus, the concentration difference of an unknown factor in membranes might contribute the difference in these EC 50 values. However, when the P␥ effect on cGMP binding was measured using a 1:10 dilution of P␥-depleted membrane-bound PDE in a 1000-l reaction mixture, both the level of cGMP binding and the P␥ concentration required for 50% stimulation of cGMP binding were essentially the same as those without dilution (data not shown). This suggests that, under our conditions, the EC 50 value for P␥ stimulation is independent of the concentration of P␥-depleted membrane-bound PDE. (c) Various amounts of P␥-depleted membrane-bound PDE used for the measurement of cGMP binding stimulation by P␥ yielded similar EC 50 values for P␥ (data not shown). This indicates that nonspecific binding of P␥ to ROS membranes is negligible and that the high EC 50 value is not due to nonspecific binding of P␥ to membranes. This result is consistent with a previous study showing that P␥ binds very weakly to ROS membranes stripped of PDE and other peripheral proteins (19). (d) These EC 50 values were independent of other assay conditions. For example, the reaction mixture used for the measurement of cGMP hydrolysis contained Mg 2ϩ , but Mg 2ϩ was not present in the mixture used for the assay of cGMP binding. However, the EC 50 values for P␥ stimulation of cGMP binding were similar in the presence and absence of Mg 2ϩ (data not shown). Moreover, the reaction mixtures used to measure these EC 50 values contained different concentrations of Tris-HCl and EGTA (for PDE activity) or EDTA (for cGMP binding). However, neither buffer concentrations nor presence of EGTA or EDTA in the reaction mixtures was critical (data not shown). Because the EC 50 values for P␥ inhibition of cGMP hydrolytic activity and stimulation of cGMP binding are significantly different, we conclude that the P␣␤-P␥ interaction required to stimulate cGMP binding to P␣␤ is different from that required to inhibit the cGMP hydrolytic activity of P␣␤.

Involvement of a P␥ Site Containing Amino Acids between Positions 11 and 18 in Stimulation of cGMP Binding to Non-catalytic Sites in P␣␤
We showed previously that when 16 amino acids in the N terminus of P␥ were replaced by 16 unrelated amino acids (from a different reading frame of the cDNA), the resulting P␥ mutant inhibited the PDE activity of P␥-depleted PDE exactly as did wild-type P␥, but its ability to stimulate cGMP binding to non-catalytic sites in P␣␤ was weakened (43). This suggests that these 16 amino acid residues (or a domain contained within) are involved in stimulation of cGMP binding. Using mutant forms of bovine P␥ in which amino acid residues in its N terminus were deleted stepwise, we investigated which amino acid residues in the N terminus are involved in P␥ stimulation of cGMP binding to non-catalytic sites in P␣␤ (Fig.  2, A and B). We found that the ability of P␥(N10Del) (a P␥ mutant in which 10 amino acid residues were deleted from its N terminus) to stimulate cGMP binding was similar to that of wild-type P␥ and that the K d value for cGMP obtained in the presence of P␥(N10Del) was similar to that obtained in the presence of wild-type P␥ (data not shown). However, P␥(N16Del) and P␥(N18Del) (P␥ mutants in which 16 and 18 amino acid residues were deleted from the N terminus, respectively) exhibited diminished ability to stimulate cGMP binding compared with wild-type P␥ ( Fig. 2A). Moreover, when 1 M P␥(N16Del) was added, two kinds of cGMP-binding sites with affinities of K d ϭ 18 and 62 nM were observed in P␥-depleted membrane-bound PDE (Fig. 2B). In the case of P␥(N18Del), two kinds of cGMP-binding sites were observed, and their affinities for cGMP were lower than those of P␥(N16Del) (K d ϭ 31 and 100 nM) (Fig. 2B).
These observations suggest that amino acid residues between Lys 11 and Gly 18 in P␥, but not the first 10 amino acid residues in the N terminus, are involved in P␥ stimulation of cGMP binding. It is also possible that a certain length of amino acid sequence could be required for P␥ to express the ability to stimulate cGMP binding. However, when 16 amino acids in the P␥ N terminus were randomly replaced by other amino acids, the ability of P␥ to stimulate cGMP binding to non-catalytic sites in P␣␤ was also weakened (43). Thus, this possibility seems unlikely. We observed two kinds of cGMP-binding sites with P␥ mutants, indicating that P␣␤ has two kinds of cGMPbinding sites, although the cGMP affinity of these sites seems to be identical in the presence of saturating amounts of wildtype P␥. We also noted that these P␥ mutants had similar EC 50 values for inhibition of cGMP hydrolysis by P␥-depleted membrane-bound PDE compared with wild-type P␥ (data not shown).

Involvement of a P␥ Site Containing Prolines 20 and 23 in Stimulation of cGMP Binding to Non-catalytic Sites in P␣␤
The involvement of the N-terminal amino acid residues in P␥ stimulation of cGMP binding to P␣␤ was further investigated using P␥(N22Del) (a P␥ mutant in which 22 amino acid residues were deleted from its N terminus) ( Fig. 2A). We found that the mutation greatly abolished the ability of P␥ to stimulate the level of cGMP binding. This suggests that an amino acid(s) between Gly 19 and Thr 22 in bovine P␥ (Fig. 1C) is deeply involved in P␥ stimulation of cGMP binding. On the other hand, we have shown that Pro 20 and Pro 23 in P␥ are required for phosphorylation of Thr 22 by Cdk5 (26) and that the phosphorylation increases the P␥ interaction with P␣␤, but weakens the P␥ interaction with GTP-T␣ (28,29). Thus, we speculated that Pro 20 might be crucial for maintaining a conformation of P␥ that is able to stimulate cGMP binding. We found that the P␥ mutant P␥(P20G) stimulated the level of cGMP binding much less compared with wild-type P␥ (Fig. 2C). In addition, the ability of another proline mutant, P␥(P23G), to stimulate cGMP binding was also much less than that of wildtype P␥. Moreover, the double mutant P␥(P20G,P23G) greatly decreased the ability of P␥ to stimulate cGMP binding. Scatchard plots indicate two kinds of cGMP-binding sites (K d ϭ 20 and 67 nM) in the presence of 1 M P␥(P20G) (Fig. 2D). These affinities were clearly lower than that obtained in the presence of wild-type P␥ (11 nM). Similar effects were detected in the presence of P␥(P23G) (K d ϭ 20 and 73 nM). In the case of the P␥(P20G,P23G) double mutant, the mutation also greatly decreased the cGMP affinity of these sites (K d ϭ 19 and 110 nM).
These observations clearly indicate that replacement of Pro 20 and Pro 23 with glycine greatly reduces the ability of P␥ to stimulate cGMP binding to non-catalytic sites in P␣␤. In other words, the P␥ site containing these two proline residues appears to be deeply involved in P␥ stimulation of cGMP binding to non-catalytic sites in P␣␤.
We also investigated the effect of Thr 22 substitution on the ability of P␥ to stimulate cGMP binding to non-catalytic sites in P␣␤. First, we checked the effect of substitutions with amino acids with a negative charge because Thr 22 is phosphorylated by Cdk5, as described above, and phosphorylation imparts a negative charge to the target amino acid residue. We found that substitution of Thr 22 with aspartic acid or glutamic acid, P␥(T22D) or P␥(T22E), respectively, reduced the P␥ ability to stimulate cGMP binding (Fig. 3A). Scatchard plots indicate two kinds of cGMP-binding sites with different affinities (K d ϭ 15 and 86 nM) in the presence of 1 M P␥(T22D) (Fig. 3B). In the case of P␥(T22E), two kinds of binding sites also were observed (K d ϭ 21 and 45 nM) (Fig. 3B). These affinities are clearly lower that that obtained in the presence of wild-type P␥. These observations suggest the interesting possibility that P␥ phosphorylation by Cdk5 also functions to regulate cGMP binding to non-catalytic sites in P␣␤.
Substitution of Thr 22 with valine also reduced the P␥ ability to enhance the level of cGMP binding to non-catalytic sites in P␣␤ in P␥-depleted membrane-bound PDE (Fig. 3C). Scatchard plots indicate that P␥(T22V) had less ability to enhance the affinity of non-catalytic sites for cGMP compared with wildtype P␥ (Fig. 3D). For example, in the presence of 1 M P␥(T22V), two kinds of cGMP-binding sites with lower affinities (K d ϭ 13 and 110 nM) were detected (Fig. 3D).
In bovine P␥, the sequence PVTP (residues 20 -23) represents the consensus motif for phosphorylation by Cdk5 (26). As described above, substitutions of both Pro 20 and Pro 23 with glycine and of Thr 22 with valine reduces the ability of P␥ to stimulate cGMP binding to non-catalytic sites in P␣␤. Thus, we also explored the effect of triple mutation of these prolines and threonine. We expected that the triple mutation would profoundly reduce the P␥ ability to stimulate cGMP binding to P␣␤. However, the triple mutation P␥(P20G,T22V,P23G) completely reversed the deleterious effects caused by the single and double mutations on the level of cGMP binding (Fig. 3C). Scatchard plots also indicate that the triple mutation reversed the effect of each mutation on the ability of P␥ to increase the cGMP affinity of these sites (Fig. 3D). These observations were unexpected.
We speculate that this result may emphasize the importance of the rigidity of the peptide bonds in this region as follows. Studies of model substrate conformations have suggested that the sequence PX(T/S)P is crucial for protein phosphorylation by Cdk5 (63) and that a turn or loop in the peptide backbone induced by the proline residue following the Thr/Ser phosphorylation site is an especially important determinant for Cdk5 phosphotransferase activity (64,65). A previous study of P␥ phosphorylation by Cdk5 (26) has shown that the single mutants P␥(P20G) and P␥(P23G), especially P␥(P23G), as well as the double mutant P␥(P20G,P23G) are ineffective as substrates for Cdk5. These same mutants exhibit reduced the P␥ ability to stimulate cGMP binding. Both effects may be reflections of the loss of critical secondary structure in this region of P␥ because the introduction of glycine residues in the peptide backbone would be expected to allow relative conformational freedom around the peptide bond with the adjacent amino acid, in contrast to the situation with a proline residue, whose pyrrolidine ring allows only two conformations about the peptide bond. The replacement of Thr 22 with valine in the P20G,P23G background places two valine residues adjacent to each other (GVVG, residues 20 -23). The staggered placement of the valine side chains necessitated by the vicinal arrangement may induce a relative conformational rigidity resembling that of wild-type P␥, thus restoring the ability to stimulate cGMP binding. Taken together, these observations strengthen our conclusion that the P␥ site involved in phosphorylation by Cdk5 is critically involved in stimulation of cGMP binding to non-catalytic sites in P␣␤. Whether the T22V mutation restores the conformation needed for Cdk5 substrate activity cannot be determined because the essential Thr 22 has been mutated to valine. Arg 24 is also crucial for P␥ phosphorylation by Cdk5 (26). However, the ability of Arg 24 mutants P␥(R24E), P␥(R24H), and P␥(R24K) to stimulate the level of cGMP binding was similar to that of wild-type P␥ (data not shown), suggesting that Arg 24 is not involved in stimulation of cGMP binding. Amino acid 21 in P␥ is not conserved in all known species (28, 58 -60). This amino acid appears not to be crucial for P␥ regulation of cGMP binding because bovine and frog P␥ subunits stimulate cGMP binding in a similar way (42), although they have different amino acid residues at position 21 (Fig. 1C). It is possible that the site containing amino acids between positions 11 and 18 and the site containing amino acid residues required for P␥ phosphorylation by Cdk5 would form one functional domain required for stimulation of cGMP binding to non-catalytic sites in P␣␤. However, amino acid residues between positions 11 and 18 are not directly involved in P␥ phosphorylation by Cdk5 (26). Thus, we favor two distinct sites. In addition, it should be emphasized that these proline and threonine mutants inhibited the PDE activity of P␥-depleted PDE in a manner similar to wild-type P␥ (data not shown).

Involvement of a P␥ Site Containing Arginines 33 and 36 in Stimulation of cGMP Binding to Non-catalytic Sites in P␣␤
Arg 33 and Arg 36 in P␥ are similarly ADP-ribosylated by endogenous ADP-ribosyltransferases when P␥ is complexed with P␣␤, but not with T␣ (31). We found that substitution of Arg 33 in P␥ with lysine or glycine reduced its ability to stimulate the level of cGMP binding to non-catalytic sites in P␣␤ (Fig. 4A). In the presence of 1 M P␥(R33K), two kinds of cGMP-binding sites (K d ϭ 14 and 80 nM) were detected (Fig.  4B). In the case of P␥(R33G), two kinds of cGMP-binding sites (K d ϭ 18 and 74 nM) were observed. Substitution of Arg 36 with lysine also reduced the P␥ ability (Fig. 4A). Scatchard plots indicate two kinds of cGMP-binding sites (K d ϭ 23 and 84 nM) in the presence of 1 M P␥(R36K) (Fig. 4B). The effect of this mutation was much less than that of P␥(R33G); however, these results were consistently observed. Thus, we conclude that the P␥ site containing Arg 33 and Arg 36 , especially Arg 33 , is involved in P␥ stimulation of cGMP binding to these sites in P␣␤. Thr 35 in P␥ is located in the same domain containing Arg 33 and Arg 36 . However, substitution of this threonine with valine did not change the ability of P␥ to increase the level of cGMP binding to non-catalytic sites in P␣␤ (data not shown). Importantly, the arginine and threonine mutants described inhibited the PDE activity of P␥-depleted PDE in a manner similar to wild-type P␥ (data not shown).

Involvement of the C-terminal Region of P␥ in Stimulation of cGMP Binding to Non-catalytic Sites in P␣␤
We investigated whether P␥(C10Del), a P␥ mutant in which 10 amino acid residues in the C terminus were deleted, has the ability to stimulate cGMP binding to non-catalytic sites in P␣␤. We found that P␥(C10Del) had less ability to stimulate the level of cGMP binding than wild-type P␥ (Fig. 5A). Scatchard plots indicate two kinds of cGMP-binding sites (K d ϭ 48 and 92 nM) in the presence of 1 M P␥(C10Del) (Fig. 5B). These obser-vations suggest that the C-terminal region of P␥ is also involved in its stimulation of cGMP binding to non-catalytic sites in P␣␤. The effect of P␥(C10Del) on the cGMP hydrolytic activity of P␣␤ is described below (Fig. 6).
We also tested other P␥ mutants for their abilities to stimulate cGMP binding to non-catalytic sites in P␣␤ to compile a more comprehensive list of P␥ sites involved. These P␥ mutants were the quadruple mutant P␥(K39L,K41L,K44L,K45L) and the single mutants P␥(M57L), P␥(C68S), and P␥(H75K). We found that these P␥ mutants stimulated the level of cGMP binding in a manner similar to wild-type P␥ (data not shown). Moreover, these P␥ mutants inhibited the PDE activity of P␥depleted PDE in a fashion similar to wild-type P␥ (data not shown). Taken together, these results strongly suggest that up to four distinct P␥ sites (site 1, containing amino acid residues between Lys 11 and Gly 18 ; site 2, containing residues 20 -23, involved in phosphorylation by Cdk5; site 3, containing residues 33-36, involved in ADP-ribosylation; and site 4, containing the C terminus) may be required for maximum stimulation of cGMP binding to non-catalytic sites in P␣␤ (indicated in Fig.  1C). Among them, the site involved in P␥ phosphorylation by Cdk5 is most profoundly involved. During preparation of this manuscript, several groups reported that some of these sites are involved in regulation of cGMP binding to non-catalytic sites in P␣␤. Using peptides, Mou and Cote (47) reported that multiple sites in the P␥ N terminus, especially sites in the region of amino acid residues 18 -41, are involved in regulation of cGMP binding to GAF domains in bovine PDE. It was also reported that phosphorylation of Thr 22 in P␥ reduces the P␥ ability to regulate cGMP binding to GAF domains in bovine PDE, implying that a region containing Thr 22 may be involved in regulation of cGMP binding to GAF domains (30). Moreover, direct interaction of a P␥ peptide (amino acids 21-45) with GAFa of bovine P␣ has been reported (48).
It is possible that the diminished cGMP-binding activity of the P␥ mutants used in this study was due to altered P␥ structure and/or misfolding of the mutant peptides. However, we showed that the ability of these P␥ mutants to inhibit PDE activity was similar to that of wild-type P␥, implying that the P␥ structure required for inhibition of PDE activity was not adversely affected. In addition, the difficulty of P␥ crystallization 3 suggests that P␥ may not have a rigid, defined structure. If so, mutations in these P␥ mutants are expected to change only those functions of the target amino acid residues. Thus, we infer the presence of these four interaction sites in P␥. We also noted that, from our limited experiments, this study cannot rule out the possibility that an unidentified P␥ site (or unidentified amino acid residue(s)) is also involved in stimulation of cGMP binding to non-catalytic sites in P␣␤.

P␥ Sites Required for Inhibition of cGMP Hydrolysis by P␣␤
According to studies from several groups, the C-terminal region of P␥ is involved in its inhibition of the cGMP hydrolytic activity of P␣␤ (9 -12, 43). In this study, we confirmed this conclusion using P␥(C10Del) and P␥(C18Sub). These mutant forms of P␥ did not inhibit the cGMP hydrolytic activity in P␥-depleted membrane-bound PDE even when 10 M concentrations of these mutants was added (Fig. 6A). We also investigated the possibility that a site other than the C terminus is involved in P␥ inhibition of cGMP hydrolysis. We found that, in the presence of 3 nM wild-type P␥ (which inhibited ϳ90% of the cGMP hydrolytic activity in P␥-depleted membrane-bound PDE), addition of 300 nM P␥(C18Sub) or ϳ5 M P␥(C10Del) restored 40% or more of the PDE activity measured without wild-type P␥ (Fig. 6B). The simplest explanation is that P␥(C10Del) and P␥(C18Sub) competitively bound to P␥-de-pleted P␣␤ and suppressed the inhibitory effect of wild-type P␥. We also observed that a high concentration of P␥(C10Del) and P␥(C18Sub) slightly, but consistently stimulated PDE activity in P␥-depleted membrane-bound PDE (Fig. 6A). This stimulation may be due to release of the inhibitory effect of residual P␥ by competitive binding of these P␥ mutants to P␣␤. (As described above, P␥-depleted membrane-bound PDE contains 20 -40% of the P␥ present in control membranes.) These observations imply that, in addition to the C-terminal region, P␥ may have another site(s) for interaction with P␣␤ to express its ability to inhibit the cGMP hydrolytic activity of P␣␤. The important point is that the unknown site is different from P␥ sites involved in stimulation of cGMP binding to non-catalytic sites in P␣␤ because various mutations in these P␥ sites, except the C-terminal region, did not affect the ability of P␥ to inhibit cGMP hydrolysis by P␣␤ in P␥-depleted membrane-bound PDE.
The P␥-depleted membrane-bound PDE used in this study contained one kind of cGMP-binding site with a relatively low affinity (K d ϭ ϳ100 nM). Both the level of cGMP binding and cGMP affinity were increased ϳ5and ϳ10-fold, respectively, if saturating amounts of wild-type P␥ were added to P␥-depleted membrane-bound PDE. However, if P␥ mutants in which amino acid residues involved in stimulation of cGMP binding were substituted with other amino acid residues were used to stimulate cGMP binding, two kinds of cGMP-binding sites with medium affinities for cGMP were detected. The simplest explanation of these results is that P␣␤ in P␥-depleted membranebound PDE contains two kinds of high affinity, cGMP-specific, non-catalytic binding sites; however, the difference in these two distinct binding sites is not detected during stimulation of cGMP binding to these sites by wild-type P␥. Is there any step in phototransduction in which the activities of these two distinct sites may be separately expressed? At present, we do not know. However, it is possible that these two distinct sites may be expressed when P␥ is modified by Cdk5 or ADP-ribosyltransferase because these two distinct sites were observed when these modification sites in P␥ were mutated.
In this study, we have suggested that the P␣␤-P␥ interaction involved in stimulation of cGMP binding to GAF domains in P␣␤ is different from that involved in inhibition of cGMP hydrolysis by P␣␤ in frog P␣␤␥␥. It should be emphasized that several studies, including ours, have already suggested or implied that both P␥ subunits in P␣␤␥␥ do not contribute equally to inhibition of the cGMP hydrolytic activity of P␣␤. We earlier reported that the first bound P␥ might be sufficient to inhibit much of the activity of frog P␥-free PDE (18). We then suggested that, under several conditions, removal of P␥ by GTP-T␣ does not necessarily result in a corresponding stimulation of the cGMP hydrolytic activity of P␣␤ in frog P␣␤␥␥ (42,43), implying that P␥ does not function solely as an inhibitor of cGMP hydrolysis by P␣␤. Berger et al. (66) monitored the competition between wild-type P␥ and a P␥ mutant that binds to P␣␤ with increased affinity, but that has a decreased ability to inhibit PDE. They found that wild-type P␥ binds to two sites in bovine P␣␤, whereas the P␥ mutant binds to and competes with wild-type P␥ at only one of the two sites and that wild-type P␥ is able to fully inhibit cGMP hydrolysis even when the P␥ mutant occupies one site in P␣␤. This study provides strong direct support for the notion of two functionally distinct P␥binding sites in P␣␤. Finally, as described above, Melia et al. (62) reported that, on positively charged membranes, GTP␥S-T␣ activates PDE maximally when GTP␥S-T␣ binds to PDE at a 1:1 molar ratio, indicating that one site is silent with respect to activation of PDE. Although they discussed the PDE activation by GTP␥S-T␣, this also implies that, under their conditions, only one P␥ is involved in PDE inhibition.

Hypothetical Models of P␣␤-P␥ Interactions
The P␣␤-P␥ interaction in P␣␤␥␥ depicted in Fig. 7A, has been used in one form or another to explain P␥ inhibition of the cGMP hydrolytic activity of P␣␤. The interaction shown is based on the following assumptions. (a) Each catalytic subunit (P␣ and P␤) has the ability to hydrolyze cGMP. It has been shown that one catalytic pocket of PDE4 has a volume large enough to accommodate a cyclic nucleotide (67) and that a monomeric form of PDE5 is catalytically active (68). Because each catalytic subunit of PDE6 has a potential catalytic domain homologous to other PDEs and PDE5 is the closest relative to PDE6 in the PDE family, these studies have been interpreted as indicating that each catalytic subunit in PDE6 harbors one active site. However, when catalytic subunits of bovine rod PDE were expressed in vitro, individually and simultaneously, no convincing evidence was presented to show that individual catalytic subunits were biologically active (69). Furthermore, we have suggested that both catalytic subunits are required to express the cGMP hydrolytic activity, whereas individual subunits are completely inactive (70). Moreover, a study of Irish Setters affected by rod/cone dysplasia indicates that P␣ expressed alone, due to null mutation of P␤, is completely inactive in vivo (71). This suggests that neither a monomeric form of the catalytic subunit nor homodimeric forms of the catalytic subunit (such as P␣-P␣) have PDE activity, although homodimers of P␣ and P␤ have been reported to exist in vitro on the basis of cross-linking studies (72). (b) Each catalytic subunit binds one P␥. Cross-linking studies have been interpreted as indicating that each catalytic subunit in P␣␤␥␥ binds one P␥ (7,73). The P␥ cross-linking to P␣␤ (7) appears to show the same amount of P␥ binding to each catalytic subunit, but the data are not necessarily quantitative because different antibodies against P␣ and P␤ have to be used to detect the complexes. A previous study (73) only speculated on the interaction of P␥ with each catalytic subunit because cross-linked products of T␣ (but not P␥) and each catalytic subunit were detected. Previous P␥ titration studies reported that two P␥ subunits might be FIG. 7. Model of P␣␤-P␥ interactions in frog P␣␤␥␥. The peripherally membrane-associated PDE is shown as a heterotetramer consisting of P␣␤ and two P␥ subunits. Membrane association is mediated by two prenyl groups attached to the C-terminal Cys (Cys 15 or farnesyl for P␤ and Cys 20 or geranylgeranyl for P␣). In each catalytic subunit (P␣ and P␤), there are two GAF domains (GAF1 and GAF2) and one catalytic domain (CD). The numbers of functional catalytic sites in P␣␤ are shown under each model. A, P␣␤-P␥ interactions as commonly portrayed; B, P␣␤-P␥ interactions with one active catalytic site; C, P␣␤-P␥ interactions with two active catalytic sites. In C, the two catalytic sites function alternately. One P␥ binds to either P␣ or P␤ and causes total inhibition. See "Results and Discussion" for details. required to completely inhibit active PDE (6,7,47,74), implying that each catalytic subunit has the ability to hydrolyze cGMP and that each P␥ binds to each catalytic subunit. These studies used soluble P␥-less or P␥-free P␣␤ preparations as active PDE, isolated (6) or trypsin-treated (7,47,74). However, active PDE appears to bind tightly and specifically to membranes in bovine (75) and frog 4 ROS. It is possible that the kinetic properties of soluble PDE may not be same as those of PDE bound tightly to membranes as discussed above. (c) The P␥-binding site is the same or similar in both catalytic subunits. A study using cross-linking with mass spectrometry (76) showed a P␥ interaction with P␣ near the putative catalytic site, but a corresponding interaction with P␤ was not found. (d) Each P␥ functions to inhibit the cGMP hydrolytic activity of P␣␤ and to stimulate cGMP binding to non-catalytic sites in P␣␤ in frog P␣␤␥␥. Although P␥ titration studies have been interpreted as implying a functional equivalence of the two P␥ subunits bound (47), another group reported non-equivalence of bound P␥ using mutant forms of P␥ as indicated above (66). Considering all of these observations, several aspects of the P␣␤-P␥ interaction shown in Fig. 7A appear to be supported, but not conclusively in our opinion.
The unique heterodimeric nature of rod PDE (P␣␤) as contrasted with the homodimeric nature of the other PDE family members implies that PDE6 membranes may have unique characteristics that are more complex than those of other PDEs. Based on this P␣-P␤ heterodimerization, how can we explain the following two phenomena observed in this study? (a) The P␥ responsible for inhibition of cGMP hydrolysis can be distinguished from the P␥ responsible for stimulation of cGMP binding; and (b) because the same wild-type P␥ shows these two effects on P␣␤, these two P␥ effects are expressed by binding of P␥ to different sites in P␣␤. We suggest two possible mechanisms as follows.
P␣␤ Containing a Single Functional Catalytic Site (Fig.  7B)-In this model, the two potential catalytic sequences in P␣ and P␤ are combined in P␣␤ to form one highly active functional catalytic site, and one P␥ inhibits the catalytic site. A similar kind of mechanism has been observed in mammalian adenylyl cyclase (77,78). Mammalian adenylyl cyclases contain two catalytic cores, C 1a and C 2a . Expression of each catalytic core alone does not show any detectable enzymatic activity. However, coexpression of these catalytic cores produces very high activity in the presence of an activator. The important point is that, although the interface of the C 1a and C 2a domains can accommodate two potential ATP-binding sites, only one of these sites binds substrate, indicating that the enzyme has only one functional catalytic site. Thus, if the P␣-P␤ heterodimerization provides a similar mechanism for PDE activity, P␣ and P␤ may form a single functional catalytic site. Under these conditions, one P␥ may block the active site and inhibit the entry of cGMP into that site.
P␣␤ Containing Two Functional Catalytic Sites (Fig. 7C)-In this model, P␣␤ has two functional catalytic sites, and one P␥ inhibits total PDE activity. As an example, in the alternating two-catalytic-sites mechanism for succinyl-CoA synthetase (79), the transfer of the phosphoryl group from the enzyme to succinate at one active site is not favored until the neighboring active site is phosphorylated by ATP, implying alternating reciprocal changes in the conformations of the two catalytic sites of the enzyme. Similar interactions have been reported in other enzyme systems with two active sites (80,81). In such alternating-sites mechanisms, if the function of one site is inhibited, the entire activity of the enzyme is suppressed (79 -81). If the two potential catalytic sites in P␣␤ express PDE activity in a similar fashion, binding of one P␥ would be suffi-cient to inhibit the entire PDE activity of P␣␤ even though the P␥ blocks only one of the two active sites.
Both models shown in Fig. 7 (B and C) also imply that GAF domains may function in dimeric units and that just one cGMP binds to a dimeric form of two GAF domains in P␣␤, i.e. two cGMPs bind to non-catalytic sites in P␣␤␥␥. This implication may be supported by the observations that the maximum number of cGMPs binding to P␣␤ detected so far is only two (but not four) cGMPs per one P␣␤ (35,44). This study has also suggested that P␣␤ has two types of non-catalytic cGMP-binding sites and that only two cGMPs bind to P␣␤ under conditions of maximum cGMP binding. However, it should be emphasized that these arguments are totally hypothetical. To define the P␣␤-P␥ interaction in P␣␤␥␥ at a molecular level, more information is required. The information may be obtained by analysis of structures of P␣␤␥␥, P␣␤␥, and P␣␤ using electron microscopy and x-ray crystallography with biochemical characterization of these PDE species.