Conformation Changes, N-terminal Involvement, and cGMP Signal Relay in the Phosphodiesterase-5 GAF Domain*

The activity of phosphodiesterase-5 (PDE5) is specific for cGMP and is regulated by cGMP binding to GAF-A in its regulatory domain. To better understand the regulatory mechanism, x-ray crystallographic and biochemical studies were performed on constructs of human PDE5A1 containing the N-terminal phosphorylation segment, GAF-A, and GAF-B. Superposition of this unliganded GAF-A with the previously reported NMR structure of cGMP-bound PDE5 revealed dramatic conformational differences and suggested that helix H4 and strand B3 probably serve as two lids to gate the cGMP-binding pocket in GAF-A. The structure also identified an interfacial region among GAF-A, GAF-B, and the N-terminal loop, which may serve as a relay of the cGMP signal from GAF-A to GAF-B. N-terminal loop 98–147 was physically associated with GAF-B domains of the dimer. Biochemical analyses showed an inhibitory effect of this loop on cGMP binding and its involvement in the cGMP-induced conformation changes.

(Cialis), which have been used for the treatment of male erectile dysfunction (5). Sildenafil has also been marketed as Revatio for the treatment of pulmonary hypertension (10).
Mammalian PDEs contain a conserved catalytic domain at the C terminus and a regulatory domain at the N terminus. The regulatory domains of PDEs vary considerably and contain various structural/functional motifs. These include calmodulin-binding domains in PDE1; a PAS (Per-ARNT-Sim) domain in PDE8; two upstream conserved regions in PDE4; and two GAF domains in PDE2, PDE5, PDE6, PDE10, and PDE11. The term GAF was introduced in 1997 (11) as a common structural motif found in three proteins, cGMP-binding phosphodiesterases, cyanobacterial adenylyl cyclases, and a formate-hydrogen lyase transcription activator from Escherichia coli, and now in Ͼ7000 proteins (12,13). Certain GAF motifs have been shown to function as ligand-binding domains or to provide protein-protein interactions. GAF motifs are composed of 150 -200 amino acids that are typically folded into a central six-stranded ␤-sheet flanked by two ␣-helices on both sides of the sheet. In the PDE superfamily, 5 of the 11 PDE families contain two GAF motifs that are arranged in tandem in their regulatory domains. For PDE holoenzymes, a certain type of cyclic nucleotide binds one of the two GAF motifs with high specificity: cAMP to GAF-B in PDE10; cGMP to GAF-B in PDE2; and cGMP to the GAF-A domains in PDE5, PDE6, and PDE11 (14 -18). Binding of the cyclic nucleotides to the GAF domains in these PDEs regulates the catalytic activity and also impacts the phosphorylation status in PDE5 (15, 19 -25). However, the molecular and structural features that provide for GAF functions are poorly understood.
Several x-ray crystal structures of PDE GAF domains have been determined: GAF-A ϩ GAF-B of mouse PDE2A in complex with cGMP (14), GAF-A of chicken PDE6C in complex with cGMP (26), GAF-B of human PDE10A in complex with cAMP (27), and an unliganded PDE2A fragment containing both GAF and catalytic domains (28). The NMR structure of mouse PDE5A GAF-A in complex with cGMP has also been reported (18). These structures have provided insight into the functions of the GAF domains. For example, the PDE2 structures (14,28) showed tight association of two molecules, implying that PDEs may work as a functional homodimer. However, many essential questions remain. How do the nucleotides bind to the closed GAF pockets of PDEs? How is the nucleotide signal propagated from the GAF domain to the catalytic site? Apparently, both unliganded and liganded structures of the GAF domains would be required to answer these questions. Here, we report structural and biochemical studies on human PDE5A1 fragments containing the N-terminal phosphorylation site, GAF-A, and GAF-B. These studies show a "lid" gating cGMP binding, a potential relay of the cGMP signal from GAF-A to GAF-B, and involvement of Nterminal segment 101-127 in the cGMP signaling.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The cDNA of fulllength human PDE5A1 (residues 1-875) was a gift from Drs. Jackie Corbin and Sharron Francis (Vanderbilt University). The coding regions for expression of GAF regulatory domains 29 -518, 85-518, 89 -518, 98 -518, 116 -518, 126 -518, and 142-518 were subcloned into the modified expression vector pET32a. Oligonucleotide primers were synthesized for amplification of the PDE5A fragments by PCR. The amplified PDE5A cDNAs and the expression vector pET32a were separately digested by restriction enzymes NheI and XhoI, purified on agarose gel, and ligated by T4 DNA ligase. The resultant pET-PDE5A1 plasmids were confirmed by DNA sequencing and transferred into E. coli strain BL21-Codon-Plus for expression.
The E. coli cells carrying pET-PDE5A plasmids were grown in 2ϫ yeast extract tryptone medium at 37°C to A 600 ϭ 0.7. Isopropyl ␤-D-thiogalactopyranoside (0.1 mM) was then added for further growth at 15°C for 24 h. The recombinant PDE5A proteins were passed through a nickel-nitrilotriacetic acid column (Qiagen), subjected to thrombin cleavage, and purified on Q-Sepharose and Sephacryl S-300 columns (GE Healthcare). The PDE5A fragments after thrombin cleavage contained 10 extra N-terminal residues (Gly-Ser-Ala-Met-Glu-His-Met-Ala-Ser-Met) and 2 C-terminal residues (Arg-Asn) from the expression vector. A typical purification batch yielded 20 mg of PDE5A fragments from 4 liters of cell culture. The purity of the proteins was estimated to be Ͼ95% by SDS-PAGE. To detect potential conformation or oligomerization changes, native polyacrylamide gels (10%) were prepared from ProtoGel (National Diagnostics) in 0.375 M Tris-HCl buffer and pre-run in a running buffer of 25 mM Tris and 192 mM glycine at pH 8.8 for 30 min before loading samples.
Because wild-type PDE5A fragments showed some degree of aggregation, as revealed by SDS-PAGE in the absence of reducing agent (e.g. 200 mM ␤-mercaptoethanol), each cysteine in the protein was mutated to serine to detect which cysteine plays a key role in aggregation. The C149S mutation significantly improved not only the yield of the protein purification but also crystal diffraction from 8 to 3 Å resolution. Thus, PDE5A1(C149S) (amino acids 98 -518 or 89 -518) was used in the later crystallization.
Selenomethionyl-substituted PDE5A was prepared by culturing cells in modified M9 minimal medium. Briefly, E. coli strain BL21-CodonPlus harboring plasmid pET-PDE5A was grown in 2ϫ yeast extract tryptone medium at 37°C overnight. Cells were pelleted by centrifugation and resuspended in modified M9 minimal medium that was supplemented with the following amino acids: 100 mg/ml Lys, 100 mg/ml Thr, 100 mg/ml Phe, 50 mg/ml Leu, 50 mg/ml Ile, 50 mg/ml Val, and 50 mg/ml SeMet. The cells were grown at 37°C to A 595 ϭ 0.5-0.7, and then 0.1 mM isopropyl ␤-D-thiogalactopyranoside was added for further growth at 15°C for 24 h. The protocol used for purification of selenomethionyl-substituted PDE5A from the cell paste was the same as that used for purification of wild-type PDE5A.
Crystallization and Structure Determination-Unliganded PDE5A1(C149S) (amino acids 98 -518 or 89 -518; 20 mg/ml) and selenomethionyl-substituted PDE5A1(C149S) (amino acids 98 -518) were crystallized by vapor diffusion against a well buffer of 10% PEG 3350, 5% 2-methyl-2,4-pentanediol, 0.1 M sodium citrate (pH 5.6), and 20 mM DTT. The crystals have the trigonal space group P3 1 21 with unit cell dimensions a ϭ b ϭ 144.2 and c ϭ 134.9 Å. Diffraction data were collected on beamline X29 at the Brookhaven National Laboratory (Table 1) and processed by program HKL (29). The structure of PDE5A1 was solved by program PHENIX (30). Eighteen selenium sites were located, and the refinement of these sites yielded a figure of merit of 0.465 at 3.2 Å resolution. The procedure of RESOLVE improved the figure of merit to 0.685, and the electron density map clearly revealed the most secondary structures. The atomic model was rebuilt by program O (31) against the multiwavelength anomalous diffraction maps of 2F o Ϫ F c and F o Ϫ F c , which were improved by the density modification package of CCP4. The structure was refined by CNS (32) first and then REFMAC (33).
cGMP Binding Assay-The dissociation constants of cGMP for the PDE5A1 GAF fragments were measured by its binding to Q-Sepharose Fast Flow beads (GE Healthcare). Eleven or twelve concentrations of the GAF domains were mixed with 2-20 nM (Ͻ1/10 K D ) [ 3 H]cGMP in 200 l of binding buffer (20 mM Tris-HCl (pH 7.5), 50 mM NaCl, and 1 mM dithiothreitol). After incubation at room temperature for 10 min, the mixture was added to 75 l of Q-Sepharose beads pre-equilibrated with binding buffer. Pilot experiments showed that [ 3 H]cGMP did not bind Q-Sepharose beads but that the PDE5A1 fragments did in the presence of Ͻ150 mM NaCl. The suspension was centrifuged, and the unbound [ 3 H]cGMP in the supernatant was measured by scintillation counting. The K D values were calculated using GraphPad Prism software with nonlinear regression curve fitting and a one-site saturation model. All experiments were repeated three times.

RESULTS
Architecture of the Regulatory Domain of PDE5A-Two fragments of human PDE5A1 (amino acids 89 -518 and 98 -518; abbreviated as PDE5A89 and PDE5A98, respectively), both of which contain N-terminal loop 89/98 -147, GAF-A (amino acids 148 -323 with the C149S mutation), and GAF-B (amino acids 324 -518), were crystallized in the same space group with similar unit cell parameters ( Table 1). The two structures have very similar conformations, as shown by a root mean square deviation of 0.49 Å for the comparison of all C␣ atoms in the structure.
Each of the crystallographic asymmetric units consists of a dimer of the respective fragments. Most residues in the structures have a good electron density and are traced without am-biguity, except for disordered residues 128 -145, 395-406, and 436 -445 in subunit A and residues 131-146, 209 -217, 395-408, and 435-446 in subunit B in both structures. In addition, residues 89 -100 in the PDE5A89 structure are disordered. Structural superposition between subunits A and B yielded root mean square deviations of 1.0 and 1.1 Å for the C␣ atoms of the PDE5A89 and PDE5A98 structures, respectively. Detailed examination showed no substantial differences between the conformations of subunits A and B, except for the head and tail of segments, as well as residues 117-121. These residues did not contact the main body of the structures, and their positions were poorly resolved and may thus not be biologically significant.
The monomer of the PDE5A1 fragments contains eight ␣-helices and 13 ␤-strands that are assembled into GAF-A and GAF-B domains (Fig. 1A). In GAF-A, a central ␤-sheet that comprises six antiparallel strands (B1-B6) is associated with helices H2 and H5 to form a hydrophobic core. On the other side of the sheet, the ␤-strands constitute the bottom of the cGMP-binding pocket, whereas helix H4 and strand B3 appear to serve as two lids to cover the pocket (Fig. 1A). Similarly, the six-stranded antiparallel ␤-sheet (B7-B12) of GAF-B forms a hydrophobic core with helices H6 and H8 on one side. However, on the cGMP-binding side of GAF-B, residues 439 -444 (indicated by D in Fig. 1A), which are equivalent to helix H4 of GAF-A, are disordered.
Two molecules of the PDE5A1 regulatory domain are tightly associated through hydrophobic and hydrophilic interactions with a parallel and symmetric dimer (Fig. 1A). The GAF-A dimer interface is composed of helices H1 and H5. The long helix H5 (residues 297-342) also contributes to formation of the dimer interface of GAF-B together with helix H8 but bends ϳ15°around Cys 310 (blue spheres in Fig. 1A). The tight association of both GAF-A and GAF-B domains in the crystal structure is consistent with results from biochemi-cal studies in which the isolated GAF-A or GAF-B domain of PDE5 was a dimer and had a low dissociation constant: K D Ͻ 30 nM for GAF-A and K D ϭ 1-20 pM for GAF-B (16). The smaller K D for GAF-B is supported by results of the present study demonstrating more hydrogen bonds in the GAF-B dimer interface.
The structure of the PDE5A1 GAF domain is similar to that of the dimer of the regulatory domain of unliganded PDE2A (28). However, these dimers superimposed poorly, as shown by a root mean square deviation of 4.4 Å for 459 C␣ atoms when their secondary structures were compared using the CCP4 program (Fig. 1B). In addition, GAF-A and GAF-B of PDE5A communicate with one another through interactions between Ser 194 of GAF-A and Asn 485 of GAF-B (pink spheres in Fig. 1A), but no such noncovalent contacts between GAF-A and GAF-B exist in PDE2A.
The structural differences between the regulatory domains of PDE2A and PDE5A likely reflect differences in the cGMP signaling pathways in the two PDE families. The cGMP signal in PDE5 initiates from GAF-A, passes through GAF-B, and then stimulates the enzymatic activity. However, the cGMP signal starting from GAF-B of PDE2A may not require the involvement of GAF-A because GAF-B neighbors the catalytic domain. In addition, from a comparison of the isolated regulatory domain of cGMP-bound PDE2A versus the apoform structure, cGMP binding elicits a switch of GAF-B locations (Fig. 1C). However, it is not clear if PDE5 undergoes a similar domain swap upon cGMP binding.
Conformation Changes upon cGMP Binding-Attempts were made to crystallize PDE5A fragments in complex with cGMP, but the crystals diffracted only to 5 Å resolution. Thus, to identify the cGMP-induced conformational changes, the x-ray structure of the unliganded PDE5A GAF-A domain was compared with the cGMP-bound GAF-A domain derived from NMR determination (18). This structural comparison yielded an overall root mean square deviation of 2.7 Å for 130 C␣ atoms after removal of 19 residues with differences of Ͼ6 Å. Five regions show large shifts ( Fig. 2A): 6 -8 Å for C␣ atoms of the ␤-turn between strands B1 and B2 (human PDE5A1 residues 193-197), 6 -13 Å for a loop after strand B2 (residues 207-214; indicated by L2 in Fig. 2A), 6 -10 Å for a loop after strand B6 (residues 289 -293; indicated by L6 in Fig.  2A), 4.3 Å for helix H4 (residues 254 -259), and 3.2 Å for strand B3 (residues 216 -222). Because loop L6 is the most variable loop in the NMR conformers (18), its positional difference from the x-ray structure may not be meaningful. Besides, helix H4 contacts symmetry-related helix H4 and strand B3 of another dimer in the crystal. We believe that this crystal lattice contact might not significantly change the structure as observed in many other crystal structures, but further structural study is needed for confirmation. The remaining positional differences were calculated from the well resolved NMR conformers and x-ray portions not involved in the crystal packing and may thus be statistically significant.
Functionally, loops L2 and L6 are distant from the binding pocket, and their biological influences are unclear. However, other changes appear to be biologically relevant. Helix H4 and strand B3 apparently serve as lids for the cGMP-binding pocket for opening and closing. The pocket is open in the unliganded state (Fig. 2B), as shown by ϳ10% of the solvent-accessible area of cGMP in the model of cGMP binding to unliganded PDE5A. In the cGMP-bound state, helix H4 and strand B3 moved toward one another by Ͼ3 Å relative to the unliganded state (Fig. 2C), whereas other elements of the binding pocket (helix H3 and strands B2-B6) did not significantly change their positions. These movements closed the pocket and thereby reduced the solvent-accessible area of cGMP to ϳ1.5%, as observed in the NMR structure of the cGMP-bound protein (18). Thus, helix H4 and strand B3 may play a key role in the transformation of the cGMP-binding pocket from an open to a closed state upon cGMP binding. However, a more detailed understanding of the conformational changes occurring upon cGMP binding to PDE5 GAF domains must await x-ray crystal structure determination of the cGMP-bound form.
The ␤-Turn between Strands B1 and B2 of GAF-A May Serve as a Relay of the cGMP Signal-The cGMP-induced 6 -8-Å movement of the ␤-turn between strands B1 and B2 of GAF-A also appears to be biologically relevant. This ␤-turn is the only region involved in the noncovalent interactions between GAF-A and GAF-B and may thus serve as a potential path of the cGMP signal. It makes both hydrophilic and hydrophobic interactions with the residues from N-terminal The argument that the ␤-turn serves as a relay of the cGMP signal is indirectly supported by the results of an NMR study in which the dynamic GAF-A domain was stabilized by interactions with GAF-B in the unliganded state and these interactions were disrupted when cGMP was bound to GAF-A (18). However, other pathways for the cGMP signal may also be possible. For example, the long central helix H5 that covalently link GAF-A and GAF-B may serve as a pathway for the cGMP signal, although the comparison between the x-ray and NMR structures did not show a significant positional change in the top portion of helix H5.
Association of the N-terminal Loop with GAF-B-In the PDE5A98 structure, 8 residues of the expression vector (Ala-Met-Glu-His-Met-Ala-Ser-Met) along with 3 residues of PDE5A (Thr 98 , Arg 99 , and Lys 100 ) form an amphipathic ␣-helix (H0, blue in Fig. 1A and pink in Fig. 3A). Because the corresponding residues of PDE5A1 (Asn 91 , Ser 92 , Val 93 , Pro 94 , Gly 95 , Thr 96 , and Pro 97 ) are not predicted to form a helix, the positions of the PDE5A residues preceding Thr 98 are uncer-   DECEMBER 3, 2010 • VOLUME 285 • NUMBER 49 tain. To resolve this puzzle, the structure of a longer PDE5A fragment with residues 89 -518 was determined. In this structure, residues 89 -100 are disordered, but residues 101-127 are well ordered and can be traced without ambiguity, as shown by the solid electron density (supplemental Fig. 1). The structure comparison showed that residues 101-127 in the structures of PDE5A89 and PDE5A98 have a similar conformation and interactions.

Structure of the PDE5 GAF Domain
An unusual observation is the physical association of Nterminal loop 101-127 with GAF-B. Residues 101-106 are located on the front surface of GAF-B, whereas residues 107-127 are at the back of GAF-B (Fig. 3, Left). Residues 101-114 exist in a coil conformation and interact mainly with residues in the GAF-B domain, including the hydrogen bonds between Arg 108 N1 and Thr 322 O␥1, Arg 111 nitrogen and Glu 326 O⑀2, and Arg 111 oxygen and Arg 329 N2, as well as Asp 107 O␦1 and Ser 194 nitrogen/O␥ in both subunits of the dimers (Fig. 3,  Center). In addition, Phe 124 forms a hydrogen bond with Asp 335 from another subunit and hydrophobic interactions with Phe 498 of the same subunit in the dimer (Fig. 3, Right).
However, the N-terminal loops in the dimer show some characteristics of asymmetry. Val 115 -Phe 124 form two short antiparallel ␤-strands in subunit B of the dimer, but coils in subunit A. In addition, the hydrogen-bonding partners of Ser 123 -Asp 127 are different in subunits A and B of the dimer. In subunit B, hydrogen bonds are formed between Ser 123 O␥ and Asp 335 O␦2, Phe 124 nitrogen and Asp 335 O␦1, Ser 126 nitrogen and Gln 331 O⑀1, and Asp 127 oxygen and Lys 328 N (Fig.  3, Right). In subunit A, hydrogen bonds are formed between Phe 124 oxygen and Gln 331 N⑀2, Leu 125 nitrogen and Asp 335 O␦2, Ser 126 oxygen and Lys 100 N, and Asp 127 O␦1 and Lys 100 N. These asymmetric interactions appear to result from the resolution limit of the structures because the N-terminal loops have overall similar conformations and locations.
Inhibitory Effect of the N-terminal Loop on cGMP Signaling-To further study the conformational changes induced by cGMP binding, native PAGE was performed in the presence of various concentrations of cGMP. The gels showed that 20 M cGMP converted a solid band of PDE5A89 to a smear, whereas 2 mM cAMP had no effect on the pattern (Fig.  4A). This conversion is likely caused by surface charge changes of the molecule as the consequence of the cGMP-induced conformation changes instead of the oligomerization state of the protein. FPLC showed that the PDE5A98 fragment was eluted as a sharp peak with an apparent molecular mass of 82.3 kDa in the absence of cGMP, which shifted to 125 kDa in the presence of 0.2 mM cGMP (supplemental Fig.  2). The 82.3-kDa peak basically matches the dimer molecular mass of 97.8 kDa. However, the 125-kDa peak is unlikely a trimer because the tight association of the PDE5 dimer in the crystal structure would predict a tetramer or other even oligomers. Very likely, the peak shift in the gel filtration reflects a change in the molecular shape of the PDE5 dimer.
To identify the segment of sequence that is most critical for the shifts on native PAGE, several fragments of the PDE5A GAF domain with various lengths were expressed, purified, and analyzed by native gel electrophoresis. In the presence of 2 mM cGMP, three PDE5A constructs (amino acids 116 -528, 126 -518, and 142-518) migrated slightly faster than in the absence of cGMP on the gel (Fig. 4B). This likely reflects the minor charge changes upon cGMP binding, but no major conformational changes. The high cGMP concentration used in the gel analysis might retain some cGMP in the binding pocket during electrophoresis, as reported by another group (34), thus impacting the band position. On the other hand, the bands for fragments 89 -518 and 98 -518 became smeared in the presence of 2 mM cGMP and migrated much slower, implying that N-terminal residues 98 -115 affect the conformation of the GAF domains upon cGMP binding. This phenomenon is consistent with previous reports that the N-terminal fragments or full-length holoenzyme migrated slower in the presence of cGMP on native gels (16,35).
To determine the impact of PDE5A N-terminal loop 98 -147 on cGMP binding, we measured the cGMP dissociation constants of the various fragments. These experiments showed that the dissociation constant for cGMP was dependent on the length of the fragment ( Table 2 and supplemental Fig. 3). Fragments 29 -518, 85-518, and 98 -518 had comparable dissociation constants that were at least of 17-fold lower affinity compared with those for fragments 116 -518, 126 -518, and 142-518. These constants are generally consistent with the previous report that small GAF-A fragments have high affinity with cGMP (16). However, our longer fragments showed weaker affinity compare with those reported previ- ously (16). The explanation is not clear, but it might reflect the different measurement methods. Because the N-terminal segment does not directly interact with residues around the binding pocket, its inhibitory effects must be exerted in an allosteric mode, very likely via the interactions between the N-terminal residues and the ␤-turn of strand B1/B2 of GAF-A.
Interestingly, all of the longer fragments of PDE5A1 contain Ser 102 , a unique phosphorylation site in PDE5A (16,20). Phosphorylation of Ser 102 was reported to be very low in the unliganded state of PDE5A and significantly improved by cGMP binding to the GAF-A domain (16). Controversially, activation of PDE5 was reported not to require PDE5 phosphorylation (22). In the structure, Ser 102 is located near the interface of the GAF-A and GAF-B domains and is spatially accessible (Fig. 3A). However, the proximity of the ␤-turn between strands B1 and B2 of GAF-A (Fig. 3B) may hinder the interaction of a kinase for effective phosphorylation. Thus, the phosphorylation efficacy may depend on the size of kinases, i.e. small kinases will have better efficacy than large kinases. It is also interesting to note that the N terminus of PDE11A4 and the phosphorylation of its 2 serine residues impact the activities of PDE11-cyclase hybrids (36).

DISCUSSION
The determinants for the specificity of cyclic nucleotide binding to GAF domains in PDEs continue to be puzzling. An NMR study of mouse PDE5A1 (18) identified Asp 196 as a key residue for the high specificity of this protein for cGMP. This is supported by the D196A mutation, which diminished the affinity for cGMP binding and increased the affinity for cAMP binding. This is also supported by the conservation of this aspartate in the GAF-A domain of PDE11A and the GAF-B domains of PDE2A, both of which purportedly prefer cGMP in the allosteric site. However, this argument is weakened by several observations. First, the correspondence of an asparagine to this aspartic acid in PDE6 may suggest an alternative recognition mechanism in PDE6. Second, the involvement of the backbone oxygen, instead of the side chain, of the corresponding Asp 305 in the hydrogen bond in the crystal structure of cAMP-bound PDE10A (27) implies its less critical role in nucleotide recognition. Third, the facade of the backbone atoms of Asp 206 to the binding pocket in our x-ray structure of unliganded human PDE5A1, in contrast to the side chain interaction with cGMP in the NMR structure, suggests the resolution limits of these methods, although it may also possibly imply a conformational change upon cGMP binding. Apparently, other residues (yet to be identified), as well as this aspartic acid, are important for the specificity of cyclic nucleotide binding.
Despite two GAF domains in the PDE families that have quite similar overall folding, only one binds cyclic nucleotide. The structure of unliganded PDE5A reported herein suggests a clue to this puzzle. The cGMP-binding pocket in GAF-A of PDE5A is gated by helix H4 and strand B3, but residues 438 -444, which correspond to helix H4 of GAF-A, are disordered in GAF-B. This disorder would be expected to make cGMP diffusion in/out of the pocket easier and significantly compromise the binding affinity of cGMP. Thus, our studies may imply that the three-dimensional conformation of the cGMPbinding pocket is another factor for binding of the nucleotides. However, this argument is weakened by the observation that the corresponding helix H4 in GAF-B of the unliganded PDE2A structure is disordered (28). Apparently, both the overall shape of the pocket and the chemical nature of individual residues in the pocket would determine nucleotide specificity and affinity, but further studies are required to draw a firm conclusion. In summary, our studies suggest a complicated and cooperative process of cGMP signaling, which is composed of conformation changes in the pocket upon cGMP binding, relay of the cGMP signal from GAF-A to GAF-B, and involvement of N-terminal segment 101-127 in the cGMP signaling.