Crystal Structure of the GAF-B Domain from Human Phosphodiesterase 10A Complexed with Its Ligand, cAMP*

Cyclic nucleotide phosphodiesterases (PDEs) catalyze the degradation of the cyclic nucleotides cAMP and cGMP, which are important second messengers. Five of the 11 mammalian PDE families have tandem GAF domains at their N termini. PDE10A may be the only mammalian PDE for which cAMP is the GAF domain ligand, and it may be allosterically stimulated by cAMP. PDE10A is highly expressed in striatal medium spiny neurons. Here we report the crystal structure of the C-terminal GAF domain (GAF-B) of human PDE10A complexed with cAMP at 2.1-Å resolution. The conformation of the PDE10A GAF-B domain monomer closely resembles those of the GAF domains of PDE2A and the cyanobacterium Anabaena cyaB2 adenylyl cyclase, except for the helical bundle consisting of α1, α2, and α5. The PDE10A GAF-B domain forms a dimer in the crystal and in solution. The dimerization is mainly mediated by hydrophobic interactions between the helical bundles in a parallel arrangement, with a large buried surface area. In the PDE10A GAF-B domain, cAMP tightly binds to a cNMP-binding pocket. The residues in the α3 and α4 helices, the β6 strand, the loop between 310 and α4, and the loop between α4 and β5 are involved in the recognition of the phosphate and ribose moieties. This recognition mode is similar to those of the GAF domains of PDE2A and cyaB2. In contrast, the adenine base is specifically recognized by the PDE10A GAF-B domain in a unique manner, through residues in the β1 and β2 strands.

atory activity and delayed acquisition of conditioned avoidance behavior, suggesting that PDE10A is involved in regulating striatal output, possibly by reducing the sensitivity of medium spiny neurons to glutamatergic excitation (20). Papaverine, which inhibits PDE10A, improves the behavior of animal models of psychosis, and therefore the inhibition of PDE10A may have the potential to treat psychosis (21,22).
In this study, we determined the crystal structure of the human PDE10A GAF-B domain complexed with cAMP. The PDE10A GAF-B domain forms a parallel dimer mediated by hydrophobic interactions between helical bundles. The dimer interface is completely different from those of the GAF domains in PDE2A and cyaB2. Interestingly, the cAMP molecule tightly binds to a cNMP-binding pocket. The residues in the ␤1 and ␤2 strands specifically recognize the adenine base in a unique manner.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The GAF-B domain (residues 246 -427) of human PDE10A1 was produced as a 212-amino acid protein with an N-terminal histidine affinity tag and a tobacco etch virus protease cleavage site. The selenomethionine-substituted protein was synthesized by the Escherichia coli cell-free system, using the dialysis method (23)(24)(25)(26). The reaction solution was centrifuged at 16,000 ϫ g at 4°C for 20 min. The supernatant was loaded on a HiTrap Chelating (GE Healthcare) column (5 ml), previously equilibrated with 20 mM Tris-HCl buffer (pH 8.0) containing 1 M NaCl and 15 mM imidazole, and was eluted with 20 mM Tris-HCl buffer (pH 8.0) containing 500 mM NaCl and 500 mM imidazole. The sample buffer was exchanged to 20 mM Tris-HCl buffer (pH 8.0) containing 1 M NaCl and 15 mM imidazole, with a HiPrep 26/10 desalting column. The histidine affinity tag was cleaved by 250 l of tobacco etch virus protease (4 mg/ml) at 25°C for 3 h and was removed by a second passage through the His-Trap column. The protein sample was desalted on a HiPrep 26/10 desalting column, and was eluted with 20 mM Tris-HCl buffer (pH 8.5) containing 10 mM NaCl and 5 mM ␤-mercaptoethanol. Next, the protein sample was loaded on a HiTrap Q (GE Healthcare) column (5 ml), previously equilibrated with 20 mM Tris-HCl buffer (pH 8.5) containing 10 mM NaCl and 5 mM ␤-mercaptoethanol, and was eluted with a linear gradient of 10 mM to 1.0 M NaCl in 20 mM Tris-HCl buffer (pH 8.5) with 5 mM ␤-mercaptoethanol. Finally, the protein sample was loaded on a HiLoad 16/60 Superdex 75 (GE Healthcare) column, previously equilibrated with 20 mM Tris-HCl buffer (pH 8.0) containing 300 mM NaCl and 2 mM dithiothreitol, and was eluted with this buffer. The native protein prepared for the analytical ultracentrifugation experiments was synthesized and purified in the same manner as the selenomethionine-substituted protein.
Crystallization and Data Collection-The best crystals of the selenomethionine-substituted protein were grown at 20°C by the sitting-drop vapor-diffusion method by mixing equal volumes of the protein (2.8 mg/ml in 20 mM Tris-HCl buffer, pH 8.0, containing 300 mM NaCl, 2 mM dithiothreitol, and 4 mM sodium 3Ј,5Ј-cAMP), and reservoir solutions (0.1 M BisTris buffer, pH 5.5, containing 0.2 M ammonium sulfate and 25% PEG 3350). The crystals belong to the space group P3 1 21, with unit cell constants of a ϭ b ϭ 74.58 Å, c ϭ 146.68 Å. There is one dimer in the asymmetric unit. X-ray diffraction data for the MAD method were collected at three different wavelengths at BL26B2 of SPring-8 (Harima, Japan). All data were processed using the HKL2000 and SCALEPACK programs (27). The redundancy-independent merging R factor (R r.i.m. ) and the precision-indicating merging R factor (R p.i.m. ) were calculated using the program RMERGE (28,29). The data processing statistics are summarized in Table 1.
Structure Determination and Refinement-The positions of the selenium atoms and the initial MAD phases were determined using the program SOLVE (30), and the MAD phases were improved with RESOLVE (31). The resulting electron density map was clear. Two cAMP molecules in the asymmetric unit were traced unambiguously. The model was built with the program TURBO-FRODO, and multiple cycles of model building and refinement were performed. The model was refined using CNS 1.1 (32). TLS refinement with 8 groups, defined by the TLSMD server, was used in Refmac5 for the final refinement stage (33)(34)(35). In the electron density map, the N-terminal artificial linkers and residues 423-427 in molecule A, and residues 246 -247 and 420 -427 in molecule B were disordered. The final model has good geometry, as examined by MolProbity (36): 97.4% of the residues have / angles in the "favored region" of the Ramachandran plot, and 99.8% are in the "allowed regions." The refinement statistics e Figure of merit after SOLVE phasing. f R work ϭ ⌺͉F obs Ϫ F calc ͉/⌺F obs for all reflections and R free was calculated using randomly selected reflections (5%).
are summarized in Table 1. The ribbon, tube, ball-and-stick, and solvent-accessible surface models in the figures were drawn using PyMOL (DeLano Scientific, Palo Alto, CA). Analytical Ultracentrifugation-All analytical ultracentrifugation experiments were carried out with a Beckman Optima XL-I analytical ultracentrifuge. The sample buffer was 20 mM Tris-HCl (pH 8.0), 300 mM NaCl, and 5 mM ␤-mercaptoethanol, and all experiments were performed at 4°C. The solvent density ( ϭ 1.0129 g/ml) and the protein partial specific volume (v ϭ 0.728 ml/g) were estimated with SEDNTERP (37). Sedimentation velocity data were obtained at 40,000 rpm using an Epon double-sector centerpiece, with loading concentrations of 1.0 mg/ml of native protein. The data were collected every 5 min for 500 min. The data were analyzed with the program SEDFIT (38).
The GAF-B domain of human PDE10A forms a dimer in the asymmetric unit of the crystal (Fig. 1, C and D). Each GAF-B domain has a bound cAMP. The sedimentation velocity data from the analytical ultracentrifugation experiments at 4°C showed that the PDE10A GAF-B domain sedimented as a single major peak, with an estimated s-value of 2.2013 S (Fig. 1B). The corrected s-value under water conditions at 20°C is 3.3957 S. The molecular weight was estimated to be 41,144, which is in agreement with the formation of a dimer, with a calculated molecular weight of 42,430.
The two molecules (A and B) superimpose each other well in the region of the six-stranded ␤-sheet and the three short helices (Fig. 1A), with a root mean square deviation of 0.74 Å for the 104 C␣ atoms of residues 284 -387. In contrast, the structure of the helical bundle is different: the ␣1 helix of molecule B is much shorter than that of molecule A (16 and 9 residues in molecule A and B, respectively); and the ␣5 helix has a kink at Phe 404 only in molecule B (Fig. 1, A and E).
Dimer Interface-Most of the dimer interface involves the three-helix bundle (Figs. 1, C and D, and 2). The interaction between the molecules buries a total surface area of 3,061 Å 2 , calculated with the AREAIMOL program in the CCP4 suite (41). The homodimer interaction is asymmetric. The ␣1 and ␣2 helices of molecule B interact with the ␣5 helix of molecule A,    but only the ␣1 helix of molecule A interacts with the ␣5 helix of molecule B (Fig. 1, C and D). The ␣1 helix of molecule B also interacts with the ␤5-␤6 loop, the ␤5 strand, and the ␣3-␤4 loop of molecule A, but the ␣1 helix of molecule A interacts only with the ␤5-␤6 loop and the ␤5 strand, and not with the ␣3-␤4 loop of molecule B (Fig. 1C). There is an intensive interaction between the ␣5 helices (Fig. 1E). The ␣5-␣5 interaction surface is also asymmetric, with a slight translation along the long axis of the helices, and the conformations of the side chains of the corresponding amino acids are different between the two molecules (Fig. 1E). The homodimer interface is stabilized mainly by hydrophobic interactions (Fig. 1E). cAMP-binding Pocket-The bound cAMP molecules were unambiguously identified in the electron density map (Fig. 3A). cAMP is located between the ␤-sheet and the ␣4 helix of each GAF-B domain molecule (Figs. 1A and 3B). Because the structures of the cAMP-binding sites and cAMP recognition of the two molecules in the asymmetric unit are essentially identical, we will describe only molecule A. The cAMP molecule is buried in the binding pocket, and only the C2 carbon of cAMP is visible through an opening on the protein surface (Fig. 3A). Furthermore, the B factors for the cAMP atoms (21.1-24.6 Å 2 ) are similar to the average B factor for the protein molecule (21.4 Å 2 ). Thus, the interaction between the protein and cAMP is tight. This may explain the fact that even if cAMP was never added during either the purification or crystallization, cAMP complex crystals of PDE10A GAF-B, which diffract to 3.0-Å resolution, were obtained. The bound cAMP might have originated from the E. coli extract and remained bound tightly to the protein. A crystal diffracting to 2.1-Å resolution was obtained when cAMP was added in the crystallization drop.

D V S K D K R F P W T T E N T G N V N Q Q C I R S L L C D G R E I N F Y K V I D Y I L H G K E D I K V I P N P P P D H W A L V S G L P A Y V A Q N G L I C N I M . N A P A E D F F A F Q K E P L D E . S G W M I K N V L S D G R E I V F Y K V I D Y I L H G K E E I K V I P T P S A D H W A L A S G L P S Y V A E S G F I C N I M . N A S A D E M F K F Q E G A L D D . S G W L I K N V L S D G R E V N F Y K I I D Y I L H G K E E I K V I P T P P A D H W T L I S G L P T Y V A E N G F I C N M M . N A P A D E Y F T F P K G P V D E . T G W V I K N V L S
The adenine base of cAMP is sandwiched between the aromatic ring of Phe 304 and the aliphatic side chain of Val 385 (Fig.  3B). The N1 forms a hydrogen bond to the side chain of Arg 286 , and the side chain of Arg 286 is sandwiched between the aromatic ring of Tyr 362 and the main chain of Ile 306 (Fig. 3B). The N3 forms water-mediated hydrogen bonds to the side chains of Asp 357 and Thr 360 . The N6 forms hydrogen bonds to the main chains of Cys 287 and Asp 305 . The O2Ј forms a hydrogen bond  A) and cyan (molecule B) asterisks. The highly conserved NNKFDE motif is shown below the alignment. The alignment was produced by Clustal_X (47) and was manually modified. The figure was generated using ESPript (48  The bound cAMP (yellow) and the interacting residues (white) are shown by ball-and-stick models, with oxygen, nitrogen, phosphorus, and sulfur atoms shown in red, blue, purple, and orange, respectively. Two water molecules are shown as red spheres. In the ribbon model, the ␤ strands are cyan, the ␣ helices are salmon, the 3 10 helix is green, and the random coils are gray. Hydrogen bonds between cAMP and the protein are indicated by broken red lines.
to the side chain of Thr 364 , and water-mediated hydrogen bonds to the main chains of Asn 353 and Asp 357 . The oxygen atoms of the phosphate group form hydrogen bonds to the main chains of Ile 330 , Ala 331 , and Asn 353 , and to the side chain of Gln 383 (Fig. 3B).

DISCUSSION
Comparison with Structures of Other cNMP-binding GAF Domains-Two crystal structures of GAF domains bound to a cNMP have been reported previously: the tandem GAF domains of PDE2A complexed with cGMP (7), and the tandem GAF domains of cyaB2 cyclase complexed with cAMP (12). In PDE2A, the GAF-A domain is involved in dimerization, and the GAF-B domain binds cGMP (7). In cyaB2, both the GAF-A and GAF-B domains are involved in dimerization and cAMP binding (12). The superposition of the PDE10A GAF-B domain over the GAF-A and GAF-B domains of PDE2A yielded root mean square deviation values of 2.9 and 4.4 Å for 149 and 156 C␣ atoms, respectively. The superposition over the GAF-A and the GAF-B domains of cyaB2 yielded root mean square deviation values of 3.5 and 3.1 Å for 154 and 157 C␣ atoms, respectively. The main difference in the conformation exists in the helical bundle, whereas the conformations of the ␤ sheet and the three short helices are similar. The PDE10A GAF-B domain also contains the conserved NNKFDE motif, which is reportedly essential for nucleotide binding (11,12). The conformation and the mutual interaction network of the motif are similar to those in PDE2A and cyaB2 (data not shown).
The PDE10A GAF-B domain dimer and the PDE2A GAF-A domain dimer are parallel, whereas the cyaB2 dimer is antiparallel. Although the homodimer interaction mainly occurs at the helical bundle in both the PDE10A GAF-B and PDE2A GAF-A domains (7), the dimer interaction mode is completely different. The PDE10A GAF-B domain homodimer interaction is asymmetric, whereas the PDE2A GAF-A domain homodimer interaction is symmetric. The buried area of the PDE10A GAF-B domain is much broader than that of the GAF-A domain of PDE2A (3,061 and 1,338 Å 2 , respectively). In addition, the connecting helix between the GAF-A and GAF-B domains of PDE2A is also critically involved in the homodimer interaction, which buries 1,476 Å 2 , and the PDE2A GAF-B domain is not involved in dimerization (7). The interaction of the cyaB2 antiparallel dimer mainly occurs through the connecting helices, the two N-terminal helices of GAF-A, and the one C-terminal helix of GAF-B (12). In PDE5, homodimerization occurs between the GAF-A domains and the GAF-B domains, and the connecting region between the GAF domains also contributes to the stability (42,43). In retinal rod PDE6, the catalytic core is a heterodimer formed by the ␣ and ␤ subunits (43), and the GAF-A domains of PDE6␣ and PDE6␤ determine the specificity of dimerization (44). The amino acid residues corresponding to the helical bundle in the GAF domains of the PDE family proteins are weakly conserved (Fig. 2). It seems that the dimer interaction mode differs among the PDE GAF domains, although the helical bundle is the main interacting portion in all of the PDE GAF domains studied thus far.
cNMP Recognition by GAF Domains-All of the bound cNMPs in the GAF domains of PDE10A, PDE2A, and cyaB2 have an anti conformation and a C3Ј-endo ribose. The recognition manner of the sugar and phosphate group of cNMP is similar among the three crystal structures (Fig. 4A). The O2Ј forms a hydrogen bond to the strictly conserved Thr residue (Fig. 4A, indicated in sky blue). In contrast, the corresponding residues of the non-cNMP-binding GAF domains of the PDE families are all aliphatic residues (Fig. 2). Therefore, this Thr residue may be one of the major determinants of the cNMP binding of the GAF domains (12). In the cases of PDE10A and cyaB2 cyclase, the phosphodiester ring is in the chair conformation, and the phosphate group forms two hydrogen bonds to the backbone amides of the Ile/Phe-Ala sequence in the ␣3 helix, and a hydrogen bond to the side chain of Gln (Fig. 4A, indicated in green). In the PDE2A GAF-B domain, although Ile 458 , Ala 459 , and Glu 512 , corresponding to Ile 330 , Ala 331 , and Gln 383 in PDE10A, respectively, are at similar positions to those of PDE10A, the phosphodiester ring of the bound cGMP is in the boat conformation (7).
The cNMP is sandwiched between the side chains of Val/Leu and Phe/Ile on one side and the side chain of Val/Ile on the other (Fig. 4A, indicated in orange), except for the GAF-A domain of cyaB2. These residues are highly conserved among the GAF domains of the PDE families (Fig. 2). The N3 of cAMP or cGMP is recognized through water-mediated hydrogen bonds by the side chains of the conserved Thr/Asn and Asp residues in the ␣4 helix (Fig. 4A, indicated in pink).
The GAF domains specifically recognize the 1 and 6 key positions of the purine ring in different manners. In PDE10A, the side chain of Arg 286 forms a hydrogen bond to the N1 of cAMP (Figs. 3B and 4B). These interactions are similar to those of the cyaB2 complex, where the side chain of Arg 103 /Arg 291 , equivalent to Arg 286 in PDE10A, forms a hydrogen bond to the N1. In contrast, the side chain of Asp 439 forms a hydrogen bond to the N1 in PDE2A (Fig. 4B).
The remarkable difference in cAMP recognition is the N6 recognition. The main chain carbonyl of Cys 287 forms a hydrogen bond to the N6 in PDE10A (Fig. 4B). In cyaB2, the side chain of Thr 105 /Thr 293 , just after the corresponding residues of Cys 287 in PDE10A, hydrogen bonds with the N6 and N7 of cAMP (Fig. 4B). In PDE2A, the side chain of the corresponding Ser 424 , at the same position as Thr 105 /Thr 293 in cyaB2, also hydrogen bonds with the N7 of cGMP (Fig. 4B). The residue equivalent to these residues in the PDE10A GAF-B domain is Ala 288 . The N6 atom is also recognized by the PDE10A GAF-B domain and the cyaB2 GAF-A domain through hydrogen bonds to the carbonyl oxygens of Asp 305 and Ala 122 , respectively (Fig. 4B). The corresponding residue in PDE2A is Asp 439 , which forms hydrogen bonds to the O6 and the N1 through the main chain amide and the side chain, respectively, and contributes to the cGMP-specific recognition (Fig. 4B). Whereas Asp 439 in PDE2A is at the ␤2-␤3 turn, Asp 305 in PDE10A is within the ␤2 strand (Fig. 2). The backbone directions of these Asp residues are completely different between the two structures (Fig. 4B). The side chain of Asp 305 in PDE10A hydrogen bonds with the main chain amide of Gly 307 (Fig. 3B). This interaction fixes the conformation of Asp 305 , and may contribute to the base recognition.
Tight and Specific Interaction with cAMP-The expression of PDE10A is restricted to the brain and testis, and high levels of expression are only found in the GABAergic medium spiny neurons in the striatum (13,(17)(18)(19). In Huntington disease transgenic mice, PDE10A mRNA and protein levels decrease in the striatum prior to motor symptom development (45). Experiments using a chimeric construct of the PDE10A GAF domain and the cyanobacterial cyaB1 adenylyl cyclase suggested that the PDE10A GAF-B domain weakly binds cAMP (11). To the contrary, we found that cAMP is bound tightly and specifically in the binding pocket of the PDE10A GAF-B domain in the crystal. The weak binding estimated by the experiments might be due to the prebound cAMP, which is bound during protein preparation. It has also been suggested by the experiments using a chimeric construct that cAMP binding stimulates the catalytic activity of PDE10A, like PDE2A and PDE5 (11). The best crystals were obtained only when cAMP was added in the protein solution, suggesting that the cAMPbinding pocket of the purified PDE10A GAF-B domain sample was not fully occupied at low cAMP concentrations. In addition, Na ϩ inhibits the activity of the cyaB1 and cyaB2 adenylyl cyclases by binding to GAF domains at less than 1 M cAMP, and this regulation is conserved in mammalian GAF domains specific to cGMP (46). Allosteric regulation of PDE10A may occur at relatively low cAMP concentrations. The tight and specific binding of cAMP to the PDE10A GAF domain may be important to critically control the basal ganglia circuit of the mammalian brain.
PDE10A inhibitors may have the potential to treat schizophrenia and Parkinson disease (21,22). PDE10A is the only mammalian PDE family protein in which the GAF domain ligand is cAMP. Although the targets of structure-based drug design are usually the catalytic sites, the regulatory GAF-B domain of PDE10A is also an attractive target. The crystal structure presented here can be used for the development of drugs treating such neuropsychiatric disorders.