Three-dimensional Structure of Guanylyl Cyclase Activating Protein-2, a Calcium-sensitive Modulator of Photoreceptor Guanylyl Cyclases*

Guanylyl cyclase activating protein-2 (GCAP-2) is a Ca2+-sensitive regulator of phototransduction in retinal photoreceptor cells. GCAP-2 activates retinal guanylyl cyclases at low Ca2+ concentration (<100 nm) and inhibits them at high Ca2+ (>500 nm). The light-induced lowering of the Ca2+ level from ∼500 nm in the dark to ∼50 nm following illumination is known to play a key role in visual recovery and adaptation. We report here the three-dimensional structure of unmyristoylated GCAP-2 with three bound Ca2+ ions as determined by nuclear magnetic resonance spectroscopy of recombinant, isotopically labeled protein. GCAP-2 contains four EF-hand motifs arranged in a compact tandem array like that seen previously in recoverin. The root mean square deviation of the main chain atoms in the EF-hand regions is 2.2 Å in comparing the Ca2+-bound structures of GCAP-2 and recoverin. EF-1, as in recoverin, does not bind calcium because it contains a disabling Cys-Pro sequence. GCAP-2 differs from recoverin in that the calcium ion binds to EF-4 in addition to EF-2 and EF-3. A prominent exposed patch of hydrophobic residues formed by EF-1 and EF-2 (Leu24, Trp27, Phe31, Phe45, Phe48, Phe49, Tyr81, Val82, Leu85, and Leu89) may serve as a target-binding site for the transmission of calcium signals to guanylyl cyclase.

The calcium ion (Ca 2ϩ ) in retinal rod cells plays a critical role in regulating the recovery phase of visual excitation and adaptation to background light (1)(2)(3)(4). Ca 2ϩ enters rod outer segments through cGMP-gated cation-specific channels in the plasma membrane. These channels are kept open in the dark by the binding of cGMP. Light triggers the hydrolysis of cGMP, leading to channel closure. The cytosolic Ca 2ϩ level decreases following illumination from ϳ500 to ϳ50 nM (5-7), because channel closure blocks the entry of Ca 2ϩ , whereas its extrusion by a light-independent Na ϩ /K ϩ , Ca 2ϩ exchanger continues (8). The light-induced lowering of the Ca 2ϩ level promotes restoration of the dark state by stimulating the synthesis of cGMP (9). cGMP is synthesized in retinal photoreceptor cells by two membrane guanylyl cyclases, RetGC-1 and RetGC-2 (10 -12).
Photoreceptor guanylyl cyclases are regulated by homologous Ca 2ϩ -sensing proteins, guanylyl cyclase activating protein-1, -2, and -3 (GCAP-1, GCAP-2, and GCAP-3) 1 (13)(14)(15). Mammalian GCAP-1, GCAP-2, and GCAP-3 activate guanylyl cyclase at low Ca 2ϩ (Ͻ100 nM). GCAP-2 in addition inhibits cyclase at high Ca 2ϩ (16). In frogs, a GCAP homolog called guanylyl cyclase inhibitory protein (GCIP) inhibits cyclase at high Ca 2ϩ (17). The amino acid sequences of GCAP-1, GCAP-2, GCAP-3, and GCIP ( Fig. 1) showed that they are members of the EF-hand superfamily of Ca 2ϩ -binding proteins (18). They are similar in sequence to recoverin (19), a retinal rod outer segment protein that inhibits rhodopsin kinase at high Ca 2ϩ (20,21). The recoverin branch of the EF-hand superfamily includes neuronal Ca 2ϩ sensors such as neurocalcin, frequenin, visinin, and hippocalcin (reviewed in Ref. 22). Indeed, there is a homolog in yeast, 2 indicating that these calcium sensors arose early in the evolution of eukaryotes. The members of this family have a myristoylated amino terminus and four EFhands. They all contain a Cys-Pro sequence in EF-1 that prevents Ca 2ϩ binding by this EF-hand. The three-dimensional structures of the myristoylated and unmyristoylated forms of recoverin in the Ca 2ϩ -free and Ca 2ϩ -bound states have been determined by x-ray crystallography (24) and nuclear magnetic resonance (NMR) spectroscopy (25,26). A striking feature of these structures is the large Ca 2ϩ -induced conformational change. The binding of Ca 2ϩ to recoverin leads to the extrusion of its myristoyl group, which is highly sequestered in the Ca 2ϩfree state, and to a large rotation of the two domains of the protein. The Ca 2ϩ -induced exposure of the myristoyl group, termed the calcium-myristoyl switch, enables recoverin to bind to membranes at high Ca 2ϩ (27,28 We report here NMR spectroscopic studies of the three-dimensional structure of the Ca 2ϩ -bound form of GCAP-2 as a step toward understanding the molecular mechanism of regulation of photoreceptor guanylyl cyclases. Ideally, one would like to solve the structures of the Ca 2ϩ -free and Ca 2ϩ -bound form of myristoylated GCAP-2, the physiologic species, but this is not yet feasible because of the low solubility of the myristoylated protein. We chose instead to solve the structure of Ca 2ϩbound unmyristoylated GCAP-2, which is soluble and gives clearly resolved NMR spectra. Moreover, the structure of the unmyristoylated form of GCAP-2 is likely to be biologically pertinent. Unmyristoylated GCAP-2 is nearly as effective as myristoylated GCAP-2 in activating guanylyl cyclase at low Ca 2ϩ and inhibiting it at high Ca 2ϩ (29). Hence, structural studies of unmyristoylated GCAP-2 should reveal the Ca 2ϩinduced conformational changes underlying its regulation of cyclase.

EXPERIMENTAL PROCEDURES
Sample Preparation-Unmyristoylated recombinant GCAP-2 protein uniformly labeled with nitrogen-15 and carbon-13 was expressed in Escherichia coli strain BL21(DE3)pLysS using pET11d vector (Novagen) grown in M9 minimal medium (containing 15 N-labeled NH 4 Cl and [ 13 C 6 ]glucose) according to previously published procedures (29). Specific labeling of GCAP-2 with valine or leucine, whose methyl carbons were stereospecifically labeled with 13 C, was prepared as described previously (30). Recombinant GCAP-2 protein expressed in E. coli forms insoluble inclusion bodies that were conveniently isolated and solubilized using 8 M urea (29). The urea-solubilized protein was then dialyzed extensively to remove urea. More than 80% of the refolded GCAP-2 (after dialysis of urea) remained soluble. The soluble GCAP-2 was then further purified using gel filtration chromatography described previously (29). In addition, anion-exchange chromatography (DEAE-Sepharose, Amersham Pharmacia Biotech) was performed at pH 6.1 and at room temperature. GCAP-2 eluted from the DEAE-Sepharose column (50-ml bed volume) using a salt gradient (0 -0.5 M KCl over 60 min at flow rate, 2 ml min Ϫ1 ).
Samples for NMR experiments were prepared by dissolving 15 Nlabeled or 13  NMR Spectroscopy-All NMR experiments were performed at 45°C on a Varian UNITY-plus 500 or UNITY-600 spectrometer equipped with a four channel interface and a triple resonance probe with an actively shielded z gradient together with a pulse field gradient accessory.
The exchange rates of amide protons were measured as described previously (31) by recording a series of 15   All triple resonance experiments were performed as described previously (31) on the uniformly 13  ) 512, 64 ms). The triple resonance spectra were analyzed as described previously (31) and provided a nearly complete sequencespecific assignment of the backbone resonances.
Ca 2ϩ -binding Measurements-Tryptophan fluorescence titrations ( Fig. 3) were performed with 1 M GCAP-2 in 2 ml of 0.1 M KCl, 50 mM HEPES (pH 7.5), 1 mM dithiothreitol at 25°C. The free calcium concentration (30 nM to 2 M) was set using an EGTA buffer system. The protein samples initially contained an equal molar ratio of total Ca 2ϩ and EGTA (2 mM); the free Ca 2ϩ concentration was adjusted by adding aliquots of 0.1 M EGTA. The free Ca 2ϩ concentration was calculated based on the total amount of Ca 2ϩ and EGTA present using the computer algorithm by Brooks and Storey (37). The calculated free Ca 2ϩ concentrations agreed closely with measured Ca 2ϩ concentrations using fluorescent indicator dyes fluo-3 and rhod-2 (Molecular Probes, Eugene, OR) with K d of 0.4 and 1.0 M, respectively (38).
Ca 2ϩ -binding curves (Fig. 3) were obtained by the equilibrium dialysis method using a Dispo-equilibrium Biodialyzer (Sialomed, Columbia, MD). The apparatus consisted of two fluid-containing chambers (protein and buffer chambers) separated by a thin dialysis membrane (molecular mass cutoff, 10 kDa). The protein chamber contained 100 l of 50 M GCAP-2 in the same buffer used in the fluorescence titration above plus the addition of 1 M 45 Ca 2ϩ (total radioactivity, 1.4 Ci). The buffer chamber contained 100 l of buffer (excluding any GCAP-2) plus the addition of a known amount of cold Ca 2ϩ . The fluid in the two chambers was allowed to come to equilibrium after 12 h at 25°C. Fifteen different dialysis experiments were performed at various cold Ca 2ϩ concentrations (0, 1, 2, 10, 20, 25, 35, 45, 65, 100, 125, 140, 150, 160, 170 M). At equilibrium, the free Ca 2ϩ concentration is defined by where Ca tot 2ϩ is the total Ca 2ϩ concentration in the system, r b is the radioactivity (counts/min) of 45 Ca 2ϩ measured from an aliquot of the buffer chamber, and r p is the radioactivity measured from an equal aliquot of the protein chamber. The concentration of Ca 2ϩ bound to protein is as follows.
The fractional saturation is then defined as where P tot is the total protein concentration in the system, P bound is the concentration of protein species bound by Ca 2ϩ , ␣ is the Hill coefficient, and K d is the apparent dissociation constant.

RESULTS
The structure of recombinant GCAP-2 uniformly labeled with carbon-13 and nitrogen-15 was studied by heteronuclear NMR spectroscopy. Two-dimensional heteronuclear single quantum coherence ( 15 N-1 H HSQC) NMR spectra, which serve as fingerprints of the conformation of main chain and side chain amide groups, were obtained. The HSQC spectra of unmyristoylated GCAP-2 are presented in Fig. 2. The Ca 2ϩ -bound unmyristoylated protein exhibits many sharp and well resolved peaks. In contrast, the Ca 2ϩ -free form exhibits broad and poorly resolved peaks, suggesting that Ca 2ϩ -free GCAP-2 may represent an unfolded, aggregated protein. However, circular dichroism studies (data not shown) indicate that Ca 2ϩ -free GCAP-2 is well folded with greater than 60% helical content. In addition, the Ca 2ϩ -free GCAP-2 sample used in the NMR study is biologically active and was shown to activate photoreceptor guanylate cyclase. Hence, Ca 2ϩ -free GCAP-2 in our study represents a well defined and folded protein. The observed broadening of the NMR peaks suggests that Ca 2ϩ -free GCAP-2 most likely forms a dimer or other multimeric species under the conditions of the NMR experiment. The HSQC spectrum of Ca 2ϩ -free myristoylated protein (data not shown) is similar to that of Ca 2ϩ -free unmyristoylated protein. The low solubility of the Ca 2ϩ -bound myristoylated protein prevented us from obtaining its HSQC spectrum.
The striking Ca 2ϩ -induced spectral differences point to a large Ca 2ϩ -induced structural change in the unmyristoylated protein. The characteristic NMR peaks of the Ca 2ϩ -bound form saturate on addition of three molar equivalents of Ca 2ϩ to the protein. Ca 2ϩ -binding measurements using equilibrium dialysis and tryptophan fluorescence titrations also showed that three Ca 2ϩ bind to unmyristoylated GCAP-2 (Fig. 3). The apparent affinity is 300 Ϯ 40 nM, and the Hill coefficient is 2.1 Ϯ 0.2. A stoichiometry of three Ca 2ϩ bound to GCAP-2 is also supported by site-directed mutagenesis studies of the EF-hand motifs (16). Substituting glutamine for glutamate at position 12 of the EF-hand loops (EF-2, EF-3, EF-4) prevents the binding of Ca 2ϩ and produces a constitutively active form of GCAP-2.
The strong and well resolved peaks observed in the HSQC spectrum of Ca 2ϩ -bound, unmyristoylated GCAP-2 (Fig. 2B) suggested that it would be feasible to determine its threedimensional structure. To elucidate the structure, resonances in the NMR spectrum were assigned to specific amino acid residues. Triple resonance experiments correlating 15 N, 13 C, and 1 H were performed to facilitate making assignments. Over 95% of the backbone resonances were assigned as indicated in Fig. 2B. These backbone assignments served as the basis for assigning about 80% of the side chain resonances. Nuclear Overhauser effect spectroscopy experiments were analyzed to establish nearly 2000 proton-proton distance relationships (ϳ11 nuclear Overhauser effects/residue) throughout the protein. In addition, 216 dihedral angle restraints ( and ) were deduced from J-coupling and chemical shift data. Finally, the three-dimensional structure was calculated by distance geometry and restrained molecular dynamics.
A superposition of 22 structures of Ca 2ϩ -bound unmyristoylated GCAP-2 consistent with the NMR data is shown in Fig. 4, and their average is depicted as a ribbon diagram (Fig. 5A) and a space-filling model (Fig. 5B). The entire polypeptide chain has been traced except for the disordered region at the carboxyl terminus (residues 191-204). The structure near the amino terminus (residues 2-18) and the region between EF-3 and EF-4 (residues 132-144) are rather poorly defined (the RMS deviation of the main chain atoms is greater than 2 Å) because of a relatively small number of nuclear Overhauser effect contacts observed in these regions. Also, chemical shift data indicate a structurally disordered, random coil secondary structure in most of these regions.
Three Ca 2ϩ are bound to GCAP-2, as anticipated on the basis of its amino acid sequence and site-directed mutagenesis. The structure of EF-3 is strikingly similar to that of EF-3 in Ca 2ϩbound recoverin and calmodulin. The RMS deviations of the 116 main chain atoms of EF-3 are 0.66 Å in comparing GCAP-2 with recoverin and 0.80 Å in comparing GCAP-2 with calmodulin. Likewise, the coordination of Ca 2ϩ is virtually identical in all three. The interhelical angle or helix packing angle of EF-3 is 94°(GCAP-2), 95°(recoverin), and 96°(calmodulin).
The structures of the 12-residue Ca 2ϩ -binding loop of the EF-hands are depicted in Fig. 7. The loop of EF-1 is quite similar to that of recoverin and again shows why this motif does not bind Ca 2ϩ . EF-1 is distorted from a favorable Ca 2ϩbinding geometry by Pro 36 at the fourth position of the 12residue loop. Also, the third residue in the loop (Cys 35 ) is not suitable for ligating Ca 2ϩ . The bulky sulfhydryl group sterically blocks the entry of Ca 2ϩ . The EF-2 loop adopts a favorable structure for binding Ca 2ϩ , despite the tight turn centered at Asn 74 (position 6 of the loop). Normally, a glycine residue is conserved at position 6 in most other EF-hands (Fig. 1). The loop of EF-3 is very typical of Ca 2ϩ -occupied EF-hands and closely resembles the EF-3 loop of recoverin and calmodulin. The EF-4 loop of GCAP-2 is quite different from that of recoverin. In recoverin, the second residue in the loop (Lys 161 ) forms a salt bridge with residue 12 (Glu 171 ) that disables Ca 2ϩ binding. In GCAP-2, the second residue of the EF-4 loop (Glu 159 ) is negatively charged and cannot form a salt bridge that would impede Ca 2ϩ binding. Furthermore, residues 1 and 3 of the EF-4 loop (Asp 158 and Asn 160 ) contain oxygen atoms in their side chains that can ligate Ca 2ϩ , in contrast with the corresponding residues of recoverin (Gly 160 and Lys 162 ). Thus, Ca 2ϩ binds to EF-4 similarly to EF-2 and EF-3.
GCAP-2 has a solvent-exposed, hydrophobic surface formed by residues from EF-1 and EF-2 (Fig. 8A). The exposed patch of hydrophobic residues is formed by the clustering of several aromatic side chains (Trp 27 , Phe 31 , Phe 45 , Phe 48 , Phe 49 , and Tyr 81 ) and several aliphatic residues (Leu 24 , Leu 40 , Ile 76 , Val 82 , Leu 85 , and Leu 89 ) (Fig. 8B). These exposed hydrophobic residues are highly conserved in members of the family (22) and form a similar nonpolar patch in Ca 2ϩ -bound recoverin (24,25). In Ca 2ϩ -free recoverin, these residues make close contacts with the highly sequestered myristoyl group (26). Ca 2ϩ -induced extrusion of the myristoyl group causes these residues to become solvent exposed, suggesting that they may serve as a targetbinding site.

DISCUSSION
In this study we present the three-dimensional structure of unmyristoylated GCAP-2 with three Ca 2ϩ bound. This structure is an important step toward 1) understanding the regulatory mechanism of photoreceptor guanylyl cyclases and 2) elucidating the novel membrane-targeting mechanism of GCAPs. Although the precise structure of the amino-terminal myristoyl group of GCAP-2 could not be studied, our structure shows the amino-terminal region (residues 2-18) to be solvent exposed, suggesting that the covalently attached myristoyl group may be extruded as in Ca 2ϩ -bound recoverin (25). Recent NMR studies on the myristoyl group of GCAP-2 also suggest that the myristoyl group may be solvent exposed (41). An extruded myristoyl group of Ca 2ϩ -bound GCAP-2 may not necessarily interact with bilayer membranes (as demonstrated for recoverin), because the Ca 2ϩ -bound, myristoylated GCAP-2 appears to be cytosolic at low ionic strength (29). Instead, the myristoyl group of GCAP-2 might interact with the cyclase or perhaps with itself to form a soluble dimer. Structural studies of the myristoylated GCAPs are needed to more rigorously determine the structural role of the myristoyl group and to test whether the myristoyl group can be sequestered in Ca 2ϩ -free GCAP-2 as was seen for Ca 2ϩ -free recoverin (26).
The exposed hydrophobic patch of GCAP-2 ( Fig. 8) may serve a role in regulating guanylyl cyclase. Recent site-directed mutagenesis studies reveal that many of these exposed residues are important in the cyclase interaction (42). In particular, replacement of residues 78 -110 (that includes the exiting helix of EF-2) with corresponding residues of neurocalcin results in a chimeric protein that fails to inhibit guanylyl cyclase at low Ca 2ϩ levels but activates it at high Ca 2ϩ . Also, the replacement of residues in EF-1 (residues 24 -49) with the corresponding residues of neurocalcin renders the chimera completely inactive. It will be interesting to make point mutations of individual residues in the exposed patch to more precisely map their effect on the cyclase interaction.
The hydrophobic patch of GCAP-2 may also serve as a possible dimerization site. The crystal structures of Ca 2ϩ -bound unmyristoylated recoverin (24) and neurocalcin (36) both show the presence of a stable dimer in the asymmetric unit. Dimerization of GCAP-2 might enable a Ca 2ϩ -bound monomer to tie up a Ca 2ϩ -free monomer to prevent activation of the cyclase. Alternatively, a dimer of Ca 2ϩ -bound GCAP-2 might bind directly to the cyclase and inhibit it. However, GCAP-2 does not appear to dimerize in our NMR experiments perhaps because detergent (20 mM octyl glucoside) was present in our samples to dramatically sharpen the peaks in the NMR spectrum. This detergent does not appear to denature or inactivate GCAP-2 as was demonstrated in the original purification of GCAP-2 from the retina. 3 Additional studies are needed to test whether GCAP-2 forms a functional dimer under physiological conditions.
The structure of GCAP-2 near the amino terminus (residues 2-18) appears different from that of recoverin. There is virtually no sequence similarity between recoverin and the GCAPs in this region. Recoverin contains a long, amphipathic helix (residues 4 -18) that packs against the sequestered myristoyl group (26). This amino-terminal helix is considerably shorter in bovine GCAP-2 (residues 7-13) because four residues have been deleted in this region (Fig. 1). The orientation of the 3 A. M. Dizhoor, personal communication. amino-terminal helix is different in recoverin and GCAP-2. This helix in recoverin extends close to the interdomain linker, whereas it interacts primarily with the entering helix of EF-1 in GCAP-2. The helix orientation in GCAP-2 is also characterized by contacts between Ser 6 and Leu 79 . We note, however, that these apparent structural differences in the amino-termi-nal region between recoverin and GCAP-2 may result from the very low precision of our structure in this region (RMS deviation, 4 Å) because of dynamical disordering. Substitution of this amino-terminal region with the corresponding residues of neurocalcin has little effect on the function of GCAP-2 (42), consistent with our finding that this region is structurally FIG. 8. Space-filling representation (A) and ball-and-stick model (B) of side chain atoms of the exposed hydrophobic patch of GCAP-2. Hydrophobic, negatively charged, and positively charged residues are highlighted in yellow, red, and blue, respectively. Solventexposed hydrophobic residues from EF-1 and EF-2 are indicated.

disordered.
The carboxyl-terminal helix (residues 180 -186, highlighted in white in Fig. 5) interacts with the helices of EF-3 and EF-4, similar to that seen for recoverin (43). The association of the COOH-terminal helix with these EF-hands resembles the interaction of calmodulin with its helical target peptides (44). The carboxyl-terminal helix may enhance the specificity of GCAP-2 and recoverin by blocking their adventitious binding to targets of calmodulin.
The GCAP-2 structure is likely to be similar to that of GCAP-1 (40% sequence identity), GCAP-3 (35% identity), and GCIP (37% identity), because the overall main chain structure appears so similar to recoverin (RMS deviation, 2.2 Å; identity, 30%) and to neurocalcin (RMS deviation, 2.0 Å; identity, 40%). Most of the hydrophobic residues in the hydrophobic core and in the exposed patch (Fig. 8) are highly conserved. Also conserved are the residues that ligate Ca 2ϩ in the EF-hand loops (Fig. 7). Interestingly, important residues in the entering helix of EF-2 at the domain interface (Ala 57 , Ala 63 , and Ala 67 ) are not conserved. Other structurally important and nonconserved residues include Asn 74 , Leu 79 , Thr 93 , His 95 , and Thr 100 . Considerable differences are also found in the amino-terminal (residues 2-18) and carboxyl-terminal (residues 191-204) regions. These differences suggest that the interaction and/or orientation between the NH 2 -terminal and COOH-terminal domains might be different in GCAP-1, GCAP-3, and GCIP. Indeed, a point mutation at the domain interface causes very different phenotypes in GCAP-1 and GCAP-2. The mutation (Y99C) causes GCAP-1 to be constitutively active (45,46), resulting in autosomal dominant cone dystrophy in humans (47). In contrast, the corresponding mutation in GCAP-2 (Y104C) does not alter its Ca 2ϩ sensitivity and partially inactivates GCAP-2 (45).
In summary, we have determined the structure of unmyristoylated GCAP-2 with three bound Ca 2ϩ by NMR spectroscopy. The overall main chain structure of GCAP-2 is similar to that of Ca 2ϩ -bound recoverin except for structural differences near the amino terminus (residues 2-18) and the binding of Ca 2ϩ to EF-4. We see an exposed hydrophobic patch of residues belonging to EF-1 and EF-2 that may play a role in regulating guanylyl cyclase. Our next goal is to solve the structure of Ca 2ϩfree GCAP-2, a formidable challenge because of its lower stability and solubility, to fully elucidate the Ca 2ϩ -induced structural changes that enable GCAP-2 to activate guanylyl cyclases in the absence of Ca 2ϩ .