We have determined the structure of a C. reinhardtii soluble guanylate cyclase catalytic domain dimer by molecular replacement, using the structure of the mammalian adenylate cyclase heterodimer 35 as a search model. The structure contains one catalytic domain dimer per asymmetric unit. During refinement, we observed unexplained peaks in electron density maps around five of the fourteen cysteine residues in the dimer (Cys 499 and Cys 592 of both monomers, and Cys 621 of monomer B). High concentrations of reductant [5–10 mM dithiothreitol and 10 mM tris(2-carboxyethyl)phosphine] were present during crystallization, making it unlikely that cysteine oxidation or disulfide bond formation are responsible for these features. We wondered whether the apparent modification of the cysteine sidechains might be due to the addition of dimethylarsenic groups via reaction of the cysteine thiol group with the sodium cacodylate [sodium dimethylarsenate, (CH₃)₂AsO₂Na] buffer and dithiothreitol reductant present during crystallization 36. This chemical modification has been observed previously for several proteins crystallized from solutions containing this buffer 3738; it has even been used to obtain experimental phases for protein structure solution 39. Accordingly, we confirmed the presence and location of the dimethylarsenic-modified cysteines by taking advantage of the arsenic anomalous signal to calculate an anomalous difference map, which showed unambiguous peaks of electron density around the dimethylarsenic-modified cysteines (see Additional File 1: Dimethylarsenic cysteine modifications).

The three C-terminal residues of each monomer are disordered, as are residues 564–566 in monomer A and residues 562–565 in monomer B. The final model of the guanylate cyclase dimer includes residues 467–563 and 567–653 for monomer A and residues 467–561 and 566–653 for monomer B. The model also includes 8 phosphate ions and 99 solvent molecules, and was refined to 2.55 Å resolution. After refinement, the final model had working and free R-values 40 of 17.2% and 21.5%, respectively (Table 1).

Each guanylate cyclase domain contains a central seven-stranded β sheet surrounded by several α helices. Secondary structure elements are named according to the convention for adenylate cyclases 3541 and are indicated in Figure 2a. The first four β strands are part of a βαββαβ arrangement of secondary structure elements, a hallmark of the class III nucleotide cyclase fold, several classes of polymerase, and other nucleotidyltransferases 424344. Indeed, a search of the structure database with the program DALI 45 identifies the mammalian adenylate cyclase catalytic domains as the closest structural match, followed by several bacterial adenylate cyclases and polymerases. A smaller 3-stranded β sheet, formed by strand β5 and extensions of β1 and β4 (β1a and β4b), extends from the core of the domain, and interacts with strands β2 and β3 of the opposite monomer to form part of the dimer interface. The two monomers in the wreath-like dimer 41 are related by a twofold axis, and a central cleft, formed by the dimer interface, contains the two symmetric active sites (Figure 2b). The two monomers superimpose on each other with a r.m.s. deviation of 0.3 Å for 160 structurally equivalent Cα atom pairs. The primary structural differences between the monomers are found in their C-terminal subdomains, particularly in the α6–β7 and β7–β8 loops, and are due to differences in crystal packing interactions.

The structure of the guanylate cyclase catalytic domain differs from that of the adenylate cyclases primarily in the elements that connect strands and helices, and in the less-conserved C-terminal subdomains (Figure 1). Biochemical and structural studies have shown that these regions in the adenylate cyclases couple to regulatory proteins such as the heterotrimeric G-protein subunits G_sα, G_iα, and protein kinase C (reviewed in 46). Differences in sequence and structure between individual catalytic subunits in heterodimeric adenylate and guanylate cyclases also localize to the same regions (Figure 1).

The catalytic residues necessary for synthesizing cyclic nucleotides, conserved across all adenylate cyclases and guanylate cyclases (Figure 1), are contributed by structural elements of both monomers at each active site. We will focus on one active site, and refer to each domain as monomer A or B. In the C. reinhardtii guanylate cyclase described here, the catalytic residues are the metal-binding residues Asp 482(A) and Asp 527(A) located in the β2–β3 loop and in strand β1b, respectively, the ribose-orienting residue Asn 599(B) and the transition-state-stabilizing residue Arg 603(B), both located in helixα4, and the triphosphate-positioning residues Arg 571(A) in strandβ4a and Lys 640(B) in the β7–β8 loop (Figure 3a). With the exception of the critical metal-binding residue Asp 527(A) (see below), all of these active site residues are located at positions analogous to their location in the adenylate cyclase active site (Figure 3a); minor conformational differences likely reflect the absence of metals and nucleotides, which would assist in organizing the active site into a catalytically competent conformation.

Modeling based on the adenylate cyclase structures has indicated the mechanism of nucleotide base discrimination 4748. A glutamic acid and a cysteine, conserved in guanylate cyclases, have been proposed to mediate recognition of the exocyclic amine and carbonyl group of the guanine base, respectively. The specificity-determining residues in mammalian adenylate cyclase, a lysine and an aspartic acid, are located at the same relative positions in the adenylate cyclase protein sequence. In fact, swapping those residues into a guanylate cyclase catalytic domain results in conversion into an adenylate cyclase 4748, underscoring the equivalence of the catalytic machinery between guanylate and adenylate cyclases. In our structure, the corresponding residues are Glu 523(B) and Cys 592(B), which are situated close to the locations of their adenylate cyclase counterparts (Figure 3b). Local distortions caused by the dimethylarsenic modifications in each monomer appear to prevent optimal sidechain orientation for base recognition. For Cys 592(B), the dimethylarsenic modification of the thiol side chain prevents potential hydrogen bonding interaction with the exocyclic carbonyl group of a substrate GTP molecule (Figure 3b). The modification also results in distortion of the β2–β3 loop that contains the metal-binding residue Asp 527(A) and the base-recognition residue Glu 523(A), causing it to adopt a conformation incompatible with binding a metal-nucleotide complex (Figure 3a). Together, these local distortions would likely preclude any nucleotide binding in the active site, providing a rationale for our failure to visualize nucleotides and metals that have been soaked into the crystals.

Activation of the mammalian adenylate cyclase has been proposed to proceed via two steps, based on structures of active and inactive forms of the protein 354149. In the first step, binding of G_sα between the α1–α2 and α3–β4 loops of the C2 subunit causes a 7° rotation of the core of the C1 subunit around an axis that runs parallel to the central cleft, priming the active site for catalysis by bringing the catalytic residues from one subunit ~2 Å closer to the catalytic residues of the other 35. The second step involves the closure of the active site around the bound nucleotide. This closure brings structural elements that bind the metal cofactors and the nucleotide triphosphate moiety, and residues in the opposite subunit that orient the ribose ring and stabilize the transition state, into optimal alignment for catalysis 49 (see Additional file 2: Activation mechanism of mammalian adenylate cyclase). In particular, helix α1 moves towards helix α4 of the opposite subunit, such that a triphosphate interaction site is formed from the helix α1 dipole and a P-loop-like structure between strand β1b and helix α1. The N-terminal end of strand β4, the C-terminal end of strand β1, and the N-termini of helices α2 and α3 also shift towards the active site, properly orienting metal-binding and triphosphate-binding residues for catalysis.

Comparison of our guanylate cyclase catalytic domain structure to structures of the mammalian adenylate cyclase suggests that the guanylate cyclase catalytic domain is in an inactive conformation. The dimethylarsenic modifications described above clearly distort several active site residues. But, in addition to these localized changes, the overall orientation of one subunit with respect to the other corresponds to an open state that must close considerably for catalysis to occur, as we discuss further below.

The signature structural change upon activation for all adenylate cyclases is the movement of helix α1 towards the active site and helix α4 in the opposite subunit, regardless of other changes in domain orientations. In our structure, a phosphate ion is bound to the N-terminal end of helix α1 and to the P-loop-like site, suggesting the presence of a likely triphosphate-coordination site. However, when monomer A of the guanylate cyclase structure is superimposed on the active adenylate cyclase C1 domain, helix α1 has not moved towards the active site, which must occur to properly align all the catalytic residues, bind nucleotide, and achieve an active conformation (Figure 4). Instead, the conformation of helix α1 in the guanylate cyclase structure is much closer to that of helix α1 in the inactive adenylate cyclase structure (compare Figure 4 and Additional file 2: Activation mechanism of mammalian adenylate cyclase). Activation of the guanylate cyclase domains in the structure reported here would require the N-terminus of helix α1 of each monomer to move ~3 Å towards helix α4 of the other monomer, resulting in the concomitant shifting of the ends of strands β1b and β4a inwards towards the opposite monomer, and leading to an active site configuration similar to that observed for the mammalian adenylate cyclases (Figure 4).

The mammalian adenylate cyclase catalytic domain heterodimer contains one active site and one catalytically non-functional site – each monomer is missing residues required for catalysis, which are provided by the other monomer. The homodimeric nature of the guanylate cyclase catalytic domain described here suggests that it contains two active sites. In fact, the existence of two active sites has been postulated for mammalian membrane guanylate cyclases, all of which function as homodimers 50. We sought to confirm this possibility by looking for evidence of cooperativity in activity assays. Activity assays were carried out in the presence of Mn²⁺, because the activity in the presence of Mg²⁺ was less than 1% of that in the presence of Mn²⁺ (data not shown), as seen for mammalian soluble guanylate cyclase catalytic domains 21. We found that the enzyme exhibits positive cooperativity, with a Hill coefficient of 1.5 (Figure 5), indicating the presence of more than one active site and providing evidence that the active sites interact with each other. Cooperativity has also been observed for other homodimeric cyclases, such as the mammalian membrane-bound guanylate cyclases 515253, as well as some bacterial adenylate cyclases 5455. A possible mechanism for communication between the mammalian adenylate cyclase active site and the pseudosymmetric site, where the activator forskolin binds, has been proposed. In the mammalian adenylate cyclase catalytic domain heterodimer, the β2–β3 loop of the C1 monomer, which contains a catalytically essential aspartic acid, is connected via a hydrogen bond and a hydrophobic interaction to the β2–β3 loop of the C2 monomer, which forms part of the forskolin binding site. These interactions between the forskolin binding site and the active site provide a possible mechanism by which the binding of forskolin is communicated to the catalytic residues in the active site, resulting in activation 35.

In the absence of nucleotide, it is not clear in our structure how occupation of one active site by nucleotide is detected by the other. The analysis is also complicated by the local distortions near the active site caused by the dimethylarsenic-modified cysteines. However, inspection of our structure provides two possible mechanisms for communication between the active sites. The first mechanism involves a direct connection between residues in the two active sites. The β2–β3 loop of each monomer carries both the invariant catalytic residue Asp 527 for one active site and the conserved guanine-binding residue Glu 523 for the opposite active site. We propose that the interaction of either Asp 527 or Glu 523 with nucleotide in one active site could alter the conformation of the loop in which they both reside, resulting in a change in nucleotide affinity or enhanced catalysis in the other active site. The second mechanism involves propagation of local changes in one active site, through changes in elements of secondary structure, to the other active site. As noted above, upon substrate binding, helix α1 and strands β1 and β4 in monomer A move inward towards crucial residues in helix α4 of monomer B, bringing residues from both monomers into correct alignment for catalysis in one active site (Figure 4). This movement also brings helix α4 of monomer A, which lies directly above, and packs against, strands β1 and β4 of monomer A, towards monomer B. As helix α4 of monomer A carries residues necessary for catalysis at the other active site, this movement might allow nucleotide binding at one active site to effectively begin pre-organizing the other active site for nucleotide binding.

The activating conformational transition of helix α1 and attendant shift in adjacent β strands might be facilitated by a domain rotation like that observed for the mammalian adenylate cyclase C1 monomer upon G_sα binding 35. In the adenylate cyclases, the α1–α2 and α3–β4 loops of the C2 domain form a groove into which the switch II helix of G_sα is docked. The docking of G_sα brings about the rotation of the C1 domain. A similar groove is found in the analogous location on the guanylate cyclase dimer structure (Figure 6). It is tempting to speculate that some regulatory element, such as a soluble guanylate cyclase H-NOX sensor domain, might interact with this region in an analogous fashion, altering the balance of conformations in the catalytic domain. It is also possible that interaction of regulators with entirely different structural elements, such as the C-terminal subdomains, may be required to activate the guanylate cyclase catalytic domain. Answers to such questions await the structure of an active guanylate cyclase domain in the presence of regulatory elements.

PCR was used to amplify the gene encoding a 991-residue soluble guanylate cyclase homolog CYG12 (GenBank: XM_001700795) from a C. reinhardtii cDNA library obtained from the Chlamydomonas Center 56. Forward and reverse PCR primers were 5'-ATGCTGGGCTGGTATGACCGT-3' and 5'-TTACTCCAAACACGGGTTGTCA-3', respectively. PCR products were phosphorylated, blunt-cloned into the vector pGEM, and verified by sequencing (UC Berkeley DNA Sequencing Facility). The guanylate cyclase catalytic domain was expressed and purified using a SUMO-based system (LifeSensors) as follows: residues 468–655, which comprise the catalytic domain of CYG12, were subcloned into a vector containing the yeast SUMO homolog SMT3 with an N-terminal His-tag, and the fusion protein was expressed in Escherichia coli Tuner(DE3) (Invitrogen) for 18 h at 20°C. The fusion protein was purified from supernatant by passage over a HisTrap Ni Sepharose affinity column (GE Healthcare). A His-tagged version of the SMT3-specific protease was used to cleave the N-terminal SMT3 fusion partner from the guanylate cyclase domain, which was separated from the protease and SMT3 by a second Ni affinity step. The guanylate cyclase domain was further purified by Q Sepharose anion-exchange chromatography, followed by gel filtration into a final buffer of 25 mM triethanolamine, pH 7.5, 25 mM NaCl, and 10 mM dithiothreitol. Purified protein was concentrated and stored at -20°C until use. Protein concentrations were determined by absorbance using the calculated extinction coefficient ε₂₈₀ = 7680 M^-1 cm^-1.

Crystals were grown using the sitting-drop vapor diffusion method. Equal volumes (200 nl) of protein [40–60 mg/ml in 25 mM triethanolamine, pH 7.5, 25 mM NaCl, 5–10 mM dithiothreitol, 10 mM tris(2-carboxyethyl)phosphine] were mixed with crystallization solution [0.1 M sodium cacodylate, pH 5.0–6.4, 42–62% saturated (NH₄)₂HPO₄] and then equilibrated with a 100-μl reservoir of the same crystallization buffer at 20°C. Crystals grew in the trigonal space group and appeared within 1–2 days. Crystals were transferred to a solution of mother liquor containing 28% glycerol as a cryoprotectant, and cryo-cooled and stored in liquid nitrogen. Diffraction data were collected at 100 K using synchrotron radiation at beam line 8.2.2 at the Advanced Light Source, Lawrence Berkeley National Laboratory. Reflections were integrated and scaled with the programs MOSFLM 57 and SCALA 58. The structure was solved by molecular replacement with the program PHASER 59 using the mammalian adenylate cyclase catalytic domain (PDB code 1AZS) 35 as the search model, and the space group was identified as P3₂21. Map improvement was carried out using ARP/wARP 60 and RESOLVE 61. The model was built using COOT 62 and refined using PHENIX 63. Six TLS domains were used during refinement: residues 467–475 and 578–595, monomer A/B; residues 476–560, 569–577, and 596–608, monomer A/B; and residues 613–651, monomer A/B. Analysis of model quality was carried out using MOLPROBITY 64. Figures were prepared using PYMOL 65. The atomic coordinates and structure factors have been deposited in the Protein Data Bank (3ET6).

Guanylate cyclase assays were performed in duplicate at 24°C as described previously 66. Assays contained 5 μg of guanylate cyclase in 25 mM triethanolamine, pH 7.5, 25 mM NaCl, 4 mM MnCl₂ or MgCl₂ and 5 mM dithiothreitol. Assays were initiated by addition of indicated amounts of GTP, and were quenched after 2 minutes by addition of 400 μl of 125 mM Zn(CH₃CO₂)₂ and 500 μl of 125 mM Na₂CO₃. cGMP was quantified using a cGMP enzyme immunoassay kit, format B (Biomol), per the manufacturer's instructions. Experiments were repeated three times to ensure reproducibility.