Allosteric activation of the nitric oxide receptor soluble guanylate cyclase mapped by cryo-electron microscopy

Soluble guanylate cyclase (sGC) is the primary receptor for nitric oxide (NO) in mammalian nitric oxide signaling. We determined structures of full-length Manduca sexta sGC in both inactive and active states using cryo-electron microscopy. NO and the sGC-specific stimulator YC-1 induce a 71° rotation of the heme-binding β H-NOX and PAS domains. Repositioning of the β H-NOX domain leads to a straightening of the coiled-coil domains, which, in turn, use the motion to move the catalytic domains into an active conformation. YC-1 binds directly between the β H-NOX domain and the two CC domains. The structural elongation of the particle observed in cryo-EM was corroborated in solution using small angle X-ray scattering (SAXS). These structures delineate the endpoints of the allosteric transition responsible for the major cyclic GMP-dependent physiological effects of NO.

sGC is a heterodimer composed of two subunits, denoted a and b, that are each composed of four domains: an N-terminal heme nitric oxide/oxygen (H-NOX) domain, a Per/Arnt/Sim (PAS)-like domain, a coiled-coil (CC) domain, and a catalytic (CAT) domain (Derbyshire and Marletta, 2012;Montfort et al., 2017). Although each subunit contains an H-NOX domain, only the b H-NOX domain binds a heme cofactor, with direct ligation occurring through a conserved histidine residue (Zhao et al., 1998). The CC domains, together with the PAS domains, are thought to form a structured assembly upon dimerization of sGC (Campbell et al., 2014). The CAT domains form a wreath-like structure with the active site at the dimer interface (Derbyshire and Marletta, 2012;Hurley, 1998).
Biochemical aspects of sGC activation by NO have been studied in great detail. Without a ligand bound to the b H-NOX domain, sGC has a low basal activity. A stoichiometric equivalent of NO relative to the sGC heterodimer results in the cleavage of the proximal histidine-iron bond and the formation of a distal five-coordinate ferrous nitrosyl enzyme with 15% of maximal activity (Russwurm and Koesling, 2004;Cary et al., 2005;Fernhoff et al., 2009;Herzik et al., 2014). This low-activity state of sGC will be referred to here as the 1-NO state; importantly, the K D of the ferrous nitrosyl heme is 1.2 Â 10 -12 M thus in this state NO remains bound to the heme (Zhao et al., 1999). The activity of the 1-NO state can be increased to a maximally active state either by the addition of excess NO (xsNO), or by addition of small-molecule stimulators (Stasch et al., 2011). The benzylindazole compound YC-1 was the first reported small-molecule sGC stimulator (Ko et al., 1994). It was identified in a screen of compounds that inhibit platelet aggregation, one of the physiological responses of sGC activation. The molecular mechanisms by which NO and small molecule stimulator binding leads to enzyme activation remain unclear, despite the fact that the sGC-targeted drug Adempas discovered through a small molecule screen based on the YC-1 scaffold was approved by the FDA in 2013.
Visualization of the molecular steps in sGC activation has been a longstanding challenge. Crystal structures of individual truncated domains from various homologues of sGC have been reported (Pellicena et al., 2004;Purohit et al., 2013;Ma et al., 2010;Winger et al., 2008). Additionally, structures of related heterodimeric and multidomain proteins have provided insight into the higher order connectivity of sGC. The heterodimeric catalytic domain from Homo sapiens was solved in an inactive conformation (Allerston et al., 2013;Seeger et al., 2014). Structures of membrane-bound adenylate cyclases have been solved that contain catalytic domains as well as portions of or the eLife digest In humans and other animals, as the heart pumps blood around the body, the blood exerts pressure on the walls of the blood vessels, much like water flowing through a hose. Our blood pressure naturally varies over the day, generally increasing when we are active and decreasing when we rest. However, if blood pressure remains high for extended periods of time it can lead to heart attacks, strokes and other serious health conditions.
In 2013, a new drug known as Adempas was approved to treat high blood pressure in the lungs. This drug helps a signaling molecule in the body called nitric oxide to activate an enzyme that widens blood vessels and in turn lower blood pressure. Previous studies have found that the enzyme -called soluble guanylate cyclase (sGC) -contains several distinct domains and that nitric oxide binds to a domain known as b H-NOX. However, it was not clear how b H-NOX and the other three domains fit together to make the three-dimensional structure of the enzyme, or how nitric oxide and Adempas activate it.
To address this question, Horst, Yokom et al. used a technique called cryo-electron microscopy to determine the three-dimensional structures of the inactive and active forms of a soluble guanylate cyclase from a moth known as Manduca sexta. To produce the active form of the enzyme, soluble guanylate cyclase was incubated with both nitric oxide and a molecule called YC-1 that works in similar way to Adempas. The structures revealed that nitric oxide and YC-1 caused b H-NOX and another domain to rotate by 71. This in turn caused the remaining two domains -known as the coiled-coil domains -to change shape, and all of these movements together led to the activated enzyme. The structures also revealed that YC-1 bound to a site on the enzyme between b H-NOX and the coiled-coil domains.
Understanding how a drug for a particular condition works makes it much easier to develop new drugs that are more effective at treating the same condition or are tailored to treat other diseases. Therefore, these findings will allow pharmaceutical companies and other organizations to develop new drugs for high blood pressure and other cardiovascular diseases in a much more precise way. entire CC domains, helping to orient the C-terminal domains of sGC (Vercellino et al., 2017;Qi et al., 2019). However, the precise quaternary structural arrangement of domains in the N-terminal portion of sGC is not known. Consequently, the mechanism by which NO and small-molecule stimulators couple binding through the protein for activation are poorly understood.
In the absence of a high-resolution full-length structure, previous work has used alternative methods in attempts to understand interdomain interactions and how small molecules activate sGC. Crosslinking experiments and negative stain electron microscopy support the hypothesis that sGC is a flexible dumbbell-shaped particle, in which the CC domains serve to connect the H-NOX and PAS domains on one end of the dumbbell to the CAT domain on the other (Campbell et al., 2014;Fritz et al., 2013). Hydrogen deuterium exchange mass spectrometry (HDX-MS) has implicated the linker region between the PAS domains and the CC domains as critical in the activation mechanism, as both regions change in H/D exchange upon NO binding to sGC . Truncations of sGC have suggested that the b H-NOX domain directly inhibits the CAT domains . Point mutagenesis was used to identify several residues thought to transmit the ligand occupancy of the gas-binding H-NOX domain to the catalytic domains, including b D102 and b D106 (Baskaran et al., 2011a;Underbakke et al., 2013). None of these methods, however, afford sufficient resolution to discern domain organization and interdomain communication for the full-length protein.
Here, we report the first full-length structures of sGC with and without activating ligands using cryoelectron microscopy (cryo-EM) to 5.1 and 5.8 Å resolution, respectively. We find that sGC adopts two dramatically different conformations, revealing a large-scale global conformational change in response to NO and YC-1. Density consistent with YC-1 is observed in the active state map directly between the b H-NOX and both CC domains. We also show the functional relationship of NO binding to sGC on overall organization of sGC in solution by small angle X-ray scattering (SAXS). By visualizing the quaternary structural changes that activate this NO receptor, we can now answer the long-standing question as to how the ligand binding in the regulatory domains of sGC is communicated to the catalytic domains.

Results
Characterization and activity of full-length Manduca sexta sGC We selected Manduca sexta (Ms) sGC as a biochemically tractable protein which has 36% and 59% sequence identity when compared to the human homolog for the a and b monomers, respectively. A schematic of the domain organization is shown in Figure 1a. Additionally, the regulatory properties in response to NO and YC-1 mirror those of its human counterpart and YC-1 had been shown to increase the Ms sGC 1-NO state to maximal activity (Hu et al., 2008). While full-length Ms sGC has only been previously reported to be expressed in E. coli, we used a baculoviral expression protocol to produce reliable quantities of stable, full-length protein (Figure 1-figure supplements 1 and 2) (Hu et al., 2008). UV-visible absorption spectra of Ms sGC without ligands (unliganded, U), with xsNO, and with carbon monoxide (CO) show the expected ligand-dependent shifts of the Soret and Q bands, indicating the b H-NOX domain is competent to bind diatomic gases ( Figure 1b).
We confirmed that Ms sGC displays catalytic activity similar to other well-characterized mammalian homologues, including Homo sapiens sGC. Ms sGC exhibits a low, basal activity in the unliganded state (71 ± 36 nmol/min/mg) and partial activation in the 1-NO state (309 ± 77 nmol/min/ mg) (Figure 1c, Figure 1-figure supplement 3), consistent with previous reports of mammalian sGCs (Fernhoff et al., 2009). Maximal activity could be achieved by adding excess NO to the 1-NO sample (1988 ± 131 nmol/min/mg), or by the addition the sGC stimulator YC-1 to the 1-NO state (1522 ± 38 nmol/min/mg) (Figure 1c, Figure 1-figure supplement 4). Taken together, Ms sGC displays biochemical properties comparable to well-characterized mammalian sGCs.

Inactive conformation of sGC exhibits bent coiled-coils
Cryo-EM was used to solve the structure of full-length inactive Ms sGC. Two-dimensional classification after removing poor particles showed intact density with two lobes with connective density between them similar to previously reported EM envelopes obtained by negative stain (Figure 2 (c) Discontinuous cGMP activity assay for Ms sGC with various activation conditions: 1-NO, xsNO, and YC-1 ligands. Initial rates were taken from assays run at 25˚C, pH 7.5 with 2 mM Mg.GTP as the substrate (see Figure 1figure supplement 1b). cGMP formation was measured using an enzyme linked immunosorbent assay. The average initial rate is plotted, and the error bars reflect one standard deviation (n = 4). DOI: https://doi.org/10.7554/eLife.50634.003 The following source data and figure supplements are available for figure 1:   Table 1). While particles display a slight orientation preference, local resolution is uniform throughout the density map (Figure 2-figure supplements 5 and 6). Additionally, the overall map exhibits welldefined helices and continuous density. The two lobes of the structure are termed the 'regulatory' lobe and the 'catalytic' lobe, with the CC domains acting as a bridge between them ( Figure 2a). The long axis of the regulatory lobe is positioned perpendicular to the catalytic lobe and is predicted to contain the H-NOX/PAS bundle, while the CAT domains are predicted to form the catalytic lobe (Campbell et al., 2014). The CC domains are in a parallel orientation, and both domains have a clear bend, distinct from a previously determined CC X-ray structure (Ma et al., 2010). A linker extends from the C-terminus of the PAS domains and forms a rigid loop connecting the bent helix to the H-NOX/PAS domains ( Figure 2d, light and dark green). The CC density bends sharply at residues a A422 and b L343. This bent region forms a buried helix, and along with the C-terminal PAS linker, creates an interaction nexus between the H-NOX, PAS, and CC domains ( The CAT dimer displays a wreath-like fold with the monomers related by a twofold axis, typical for class III nucleotide cyclase domains (Hurley, 1998;Zhang et al., 1997). The CAT domains align well with a previous structure of an inactive guanylate cyclase (Ca RMSD of 1.3 Å to PDB ID: 4NI2), displaying an inaccessible nucleotide-binding pocket that would require a significant rearrangement for activation (Figure 2-figure supplement 9). In the absence of a substrate-bound structure of a guanylate cyclase, we compared our model to a substrate-bound adenylate cyclase. The alignment of the a chain of an active adenylate cyclase structure (PDB ID: 1CJK, gray) with the a CAT domain of our inactive Ms sGC structure reveals that the Ca of b N538 from the b CAT domain is within 2 Å of the bound nucleotide analog in the adenylate cyclase structure (Figure 2e). This state is thus in a closed conformation that is sterically incompatible with nucleotide binding, explaining the lack of guanylate cyclase activity.

Activated sGC extends the regulatory lobe from catalytic core
To elucidate conformational changes associated with sGC activation, a structure of sGC bound to NO and the small molecule stimulator YC-1 was determined. We elected to supplement the xsNO Ms sGC with YC-1 to generate the most stable activated conformation. Particles of active Ms sGC were well-dispersed in vitreous ice (Figure 3-figure supplements 1 and 2). Cleaned two-dimensional class averages exhibit a two-lobed density that is distinct from the inactive structure (  Although the density for the b H-NOX and PAS domains is similar between the inactive and active states, the overall shape of the molecule is more linear (Figure 3b). Continuous, linear helices from a L406-L457 and b A333-Y386 are seen in the active state ( Figure 3d, highlighted in purple). This conformation is more aligned with predicted CC length and is more similar to the previously published crystal structures in terms of the range of residues seen in an unbent coiled-coil (Ma et al., 2010;Qi et al., 2019).
The CAT heterodimer in the active state still forms a wreath-like geometry, but with the two monomers moved apart from one another roughly perpendicular to the CC domain axis (Figure 3e). An alignment of the a chain of an active adenylate cyclase structure (PDB ID: 1CJK, gray) with the a CAT domain of our active Ms sGC structure shows the Ca of b N538 is now greater than 4 Å from the bound nucleotide analog, providing sufficient space for the nucleotide to bind in the active site. The Ms sGC CAT domain thus adopts an open conformation with the apparent capacity to bind substrate.

Allosteric conformational rearrangement of sGC
Binding of NO to b H-NOX initiates the signal transduction event, thus interfaces were examined between the inactive and active states. A significant rearrangement of the quaternary structure of Ms sGC occurs upon activation with the regulatory domain rotating 71˚and the catalytic domain rotating 40˚ ( In the active structure, the CCs undergo a significant conformational change and rotation relative to the CAT dimer ( Figure 4d). The active state CC domains are completely extended (56 Å and 63 Å for aCC and bCC respectively in the inactive state, to 74 Å for both CC in the active state), as the bend present in the inactive state moves away from the CAT domains to be in line with the C-terminal portion of both CC domains. The CC conformational change upon activation twists the CC interface such that the b CC helix rotates relative to the a CC by 72 degrees (Figure 4d). The dramatic rearrangements of the H-NOX/PAS bundle leads to unbending of the CC domains, a change in their orientations as they project from the regulatory lobe, and finally to opening of the nucleotide-binding pocket (

SAXS shows a distribution of inactive and active states in solution
Small angle X-ray scattering (SAXS) was used to interrogate conformational changes with and without NO in solution. Experiemnts with sGC stimulators were not included as the limited solubility of YC-1 is precludes the collection of scattering data. Size exclusion chromatography (SEC)-SAXS and SEC-multi angle light scattering (MALS) chromatograms show a symmetrical peak for both samples with little variation in radius of gyration (R g ) and estimated MW across the peaks, indicating sample homogeneity (  had a slightly lower heme incorporation. The ratio for the SAXS sample with NO was lower than that expected for 1-NO, but much larger than predicted for the xsNO ratio, indicating that the SAXS samples were in the unliganded and 1-NO state during elution, respectively.  Table S2). Additionally, the Pair-distance function (P(r)), which shows a composite of the inter-atomic distances, displays a distinct shift between the unliganded and 1-NO states in the region from~70 to~100 Å . This corresponds to an extension of the regulatory lobe from catalytic lobe (Figure 5a). Shifts in the secondary peaks of the Kratky plots ( Figure 5-figure supplement 2) further suggest separation of the regulatory and catalytic lobes upon activation.
Using the inactive model obtained from cryo-EM as an initial Ca model, a rigid-body modeling pipeline was developed to systematically explore conformational space (see Online Methods) (Pelikan et al., 2009). A minimal ensemble search was performed over thousands of sGC conformations and corresponding scattering curves were calculated and compared to the experimental data. The result of this minimal ensemble search confirmed the presence of a single sGC conformation while in an inactive state (Figure 5b, Figure 5-figure supplement 3), which overlays with the inactive structure with an RMSD of 2.0 Å . The 1-NO state was best modeled as an ensemble of two states, with the majority (72%) of the sample consistent with the inactive model ( Figure 5b). However, the remainder of the sample (28%) is consistent with a more elongated conformation, one that is in between the inactive and active state obtained by cryo-EM. The Ca RMSD of the partially active SAXS model and the active state cryo-EM model is 13.5 Å . This analysis shows that sGC adopts a mixture of inactive and active conformations in the 1-NO state, suggesting that NO binding at the heme establishes an equilibrium between these two conformational extremes ( Figure 5-figure supplement 3).

Discussion
The structures of the inactive and active states of full-length sGC provide detailed insight into the molecular mechanism of activation of Ms sGC. The most unexpected feature of the inactive state of the enzyme are the bends in the CC domains. Straightening of the bent CCs is the clearest structural link between the regulatory and catalytic lobes during activation. Previous HDX-MS data showed a differential exchange pattern for the CC domains, where the N-terminal portion of the a CC and the C-terminal portion of the b CC increase in H/D exchange upon NO binding, while the C-terminal portion of the a CC and the N-terminal portion of the b CC decrease in H/D exchange (Figure 3figure supplement 7) . The changes seen in HDX in the CCs near the bend are thus consistent with the large-scale rearrangements of the CCs that we have now observed in the cryo-EM. Furthermore, the bent portion of the CC domain is highly conserved among sGC homologues, consistent with its centrality to allosteric communication.
The C-terminal portions of the CC domains undergo a 72˚twist and contain a motif known as the signaling helix (or S-helix) (Anantharaman et al., 2006). Similar to sGC, proteins with this motif contain both receptor and output domains that are connected by dimeric coiled-coils. The inactive and active sGC structures are the first full-length enzyme with a S-helix motif to be characterized. This motif aligns with C-terminal part of the CC domain, part of the 72˚twist described above (Figure 2figure supplement 10). It is possible that the twisting motion of these domains is a more general mechanism of activation for proteins with S-helices.
In transitioning between the inactive and active states, the H-NOX/PAS bundle rotates 71˚in order for b H-NOX residues 33-41 to form an interface with residues b 355-367 of the b CC domain (Figure 4c, purple). The aH helix of the b H-NOX domain contains the proximal histidine that binds to the heme cofactor (Zhao et al., 1998). Crystal structures of bacterial H-NOX domains with and without NO show a rotation in the aF helix of the b H-NOX. However, in our full-length structures, the aF of the b H-NOX retains its interface with the b PAS and the a CC domain in both states (Figure 4b and c). More strikingly, binding of YC-1 at a site bridging the b H-NOX and CC domain highlights how stabilization of this contact drives CC unbending (Figure 3c). This interface corroborates specific point mutations known to decrease activity of the full-length protein; specifically, single point mutants of b I41E in the b H-NOX domain retain basal activity but are only minimally activate with excess NO, implying that these sGC variants are catalytically competent but not as sensitive to stimulation as the wildtype enzyme (Underbakke et al., 2013;Baskaran et al., 2011b). The structural finding that residue Ile41 of the b H-NOX form contacts with the unbent b CC that are specific for the active conformation explains these phenotypes. Destabilizing these contacts by mutating Ile41 thus blocks this interface and prevents sGC from reaching the active conformation. While this paper was under review, a cryo-EM map of activated H. sapiens sGC was reported, where the activation was achieved by using xsNO (Kang et al., 2019). This xsNO map and the xsNO + YC-1 map reported here overlay well, except for the proposed density for YC-1. This lends support to our assignment of the YC-1 binding site at the b H-NOX-b CC interface. These insights suggest that the b H-NOX-b CC contact may be the most critical allosteric switch in the regulatory lobe of sGC.
While conformational changes are readily observed by comparing the two cryo-EM structures, the SAXS implies that the 1-NO state can sample similar conformations (Figure 5a). Although only a fraction of the population exhibits an extension, inspection of the Soret to protein UV-visible absorption indicates that the protein is in the 1-NO state. These data support the hypothesis that when the first NO binds to the heme of the b H-NOX, sGC adopts an equilibrium between the inactive and a partially extended conformation, with a K eq = [active]/[inactive] = [0.72]/[0.28]=0.39. This is corroborated by the activity data from the 1-NO state, which exhibits 15% of the maximal activity. The conformational heterogeneity observed in SAXS analysis could represent the physiological state of sGC at basal cellular conditions.
In the absence of increased NO concentrations, sGC stimulators can maximally activate cGMP production and are now used to treat forms of pulmonary hypertension (Koglin et al., 2002;Follmann et al., 2013). YC-1 was present in the sample used to generate the active state reconstruction of Ms sGC, and extra density is seen near the new b H-NOX: b CC interface (Figure 3c). Distinct changes in both resonance Raman and electron paramagnetic resonance spectroscopy signatures for heme ligands have been detected with both YC-1 and BAY 41-2272 supporting this proposed binding site (Derbyshire, 2008). Previously, a stimulator binding site was proposed between helices aA and aD in the b H-NOX domain based on cross linking and NMR data (Wales et al., 2018). However, there is no density for YC-1 present in the active state reconstruction between aA and aD helices. We note that our study used YC-1 compared to IWP-854, IWP-051 and BAY 41-2272 previously used, which could explain the difference in binding sites. Visualization of YC-1 bound to sGC is an important development as a proof of concept that small molecule sGC stimulators can now be characterized structurally.
Adenylyl cyclases (AC) catalyze a similar reaction mechanism as sGC. Mammalian membraneassociated ACs contain two catalytic domains which form the intrapolypeptide equivalent of a heterodimer. The catalytic domains in active mammalian AC heterodimer rotate by 7˚with respect to an inactive homodimeric counterpart bound to two molecules of forskolin (Zhang et al., 1997;  Tesmer et al., 1997;Hurley, 1999). We observe a rotation of the sGC CAT domains about the same axis, although the magnitude is larger for sGC, at 40˚. A full-length membrane bound AC (AC-9) was recently solved in the active nucleotide-bound state (Qi et al., 2019). The CC domains between the full-length active sGC and AC9 structures overlay well, consistent with a common active geometry for GCs and ACs. The AC9 has a very different angle between the CC and CAT domains. The rotation of the sGC regulatory lobe is sterically incompatible with the presence of a membrane, thus by their different natures as soluble and membrane-associated enzymes, sGC and ACs must differ in the detailed modes of allosteric communication.
To date, a detailed understanding of sGC activation has been hampered by the lack of full-length structures. The inactive and active structures in tandem with established studies lead to formation of new hypotheses for the activation of sGC. First, activity assays with sGC truncations suggested that direct interaction of the b H-NOX domain and the CAT dimer was responsible for sGC inhibition . However, the structural data shows no direct interaction between the b H-NOX and the CAT domain in either conformation. Instead, the formation of the new b H-NOX/b CC interface in the active state stimulates the active CAT conformation. Activation leads to a global conformational rearrangement of the heterodimer, elongating the structure; however, only a single equivalent of NO is required to cleave the bond between the heme Fe center and the ligating histidine residue. Maximal activation of sGC involves either the binding of a second NO molecule to a non-heme site on sGC or a small molecule stimulator stabilizing the active state (Guo et al., 2017;Horst and Marletta, 2018). Figure 6 depicts the proposed physiological activation sequence of sGC, where unliganded sGC adopts a more compact conformation with bent CC domains. Upon NO binding to the heme, an equilibrium of conformational states is established, with the partially elongated state affording about 15% of the maximal activity. Given the very tight association between NO and the heme, this represents the basal cellular activity. Finally, in the presence of either an increase in NO concentration or sGC stimulating molecules, the completely extended, fully active state is reached and sGC reaches maximal catalytic activity. Using these structures as a starting point, new avenues of exploration can now be undertaken, to elucidate the molecular mechanisms of excess NO activation. Critical residues for interdomain interactions along with the proposed stimulator binding site can be characterized in more detail. Having resolved the two structural states of an important therapeutic target, the structures of sGC will influence rational design of improved drugs for diseases associated with NO signaling impairment.

Construction of plasmids
Manduca sexta sGC a1 (Uniprot: O77105) and b1(Uniprot: O77106) genes were purchased as gBlocks from IDT with a C-terminal 6x His tag on the a1 gene with Golden Gate cloning sites. The gBlocks were subcloned into a Golden Gate entry vector (gift of the Tullman-Ercek lab, Northwestern University). PCR was used to add regions of homology to pFastBac (Bac-to-Bac Baculovirus Expresion System, Thermo Fisher Scientific), and Gibson Assembly was performed to generate pFastBac_Ms_sGC_a1_His6 and pFastBac_Ms_sGC_b1. The genes were transposed into a baculovirus bacmid using DH10Bac-GFP cells. The bacmid was isolated, validated, and then transfected into SF9 cells (Berkeley Cell Culture Facility) to generate recombinant baculovirus for both genes.  Protein expression and purification SF9 cells were maintained in monolayer and in suspension in ExCell-420 media at 27˚C (cells were shaken at 135 rpm). The recombinant baculovirus was amplified until the titer was greater than 1 Â 10 8 cfu/mL. Five liters of SF9 cells were coinfected with 50 mL of amplified virus per liter and allowed to express for 72 hr. Cells were spun down at 4300 g for 20 min, snap frozen in liquid nitrogen, and stored at -80˚C. The following steps were performed at 4˚C. Cell pellets were thawed in ice water and resuspended in lysis buffer: Buffer A (50 mM Na 2 HPO 4 , pH 8.0, 200 mM NaCl, 1 mM imidazole, 1 mM benzamidine, 5% (v/v) glycerol, 0.22 mm filtered) supplemented with 10 mM benzamidine, 1 mM AEBSF, 0.5 mg/mL bovine DNAse I, and 5 mM b-mercaptoethanol. Cells were lysed in a bead beater (BioSpec) with 0.5 mm glass beads. The resulting cell lysate was clarified by spinning at 4300 g for 5 min, followed by spinning at 158,000 g for 2 hr (Ti45 rotor, Beckman Coulter). The lysate was passed through a column containing 2 mL TALON superflow at 0.5 mL/min, and the flow-through was collected. The column was washed with 15 column volumes of Buffer A supplemented with 5 mM bmercaptoethanol at 0.5 mL/min. The protein was eluted with 10 CV of Buffer B (50 mM Na 2 HPO 4 , pH 8.0, 200 mM NaCl, 150 mM imidazole, 1 mM benzamidine, 5% (v/v) glycerol, 0.22 mm filtered) supplemented with 5 mM bmercaptoethanol. Fractions with yellow color were concentrated to <2 mL using a 30,000 molecular weight cutoff spin concentrator, supplemented with 5 mM DTT and 5 mM EDTA, and stored overnight. Next, the sample was diluted to 9 mL with Buffer C (25 mM triethanolamine, 25 mM NaCl, 5 mM DTT, 5% (v/v) glycerol, 0.22 mm filtered), and applied to a POROS HQ2 anion exchange column (Thermo Fisher Scientific). The column was washed with 3 CV of Buffer C, and then a gradient to 50% Buffer D (25 mM triethanolamine, 750 mM NaCl, 5 mM DTT, 5% (v/v) glycerol, 0.22 mm filtered) was established over 17 CV at 0.5 mL/min. Fractions with purified sGC were concentrated to 5-50 mM and stored in liquid nitrogen. A typical yield for this expression and purification procedure is 100 mg sGC per liter of insect cells.

Intact protein mass spectrometry
Purified proteins were buffer-exchanged into 25 mM HEPES, pH 7.5, 25 mM NaCl using three rounds of dilution and concentrations in a Vivaspin 500 (30,000 MWCO) spin concentrator. Samples were filtered with a 0.22 mM spin filter (Millipore). The final sample concentration was approximately 5 mM. Samples were separated with an Agilent 1200 series high-pressure liquid chromatography (HPLC) system over a ProSwift column (ThermoFisher Scientific), and subsequently analyzed by an Agilent 6224 time-of-flight (TOF) mass spectrometer with a Turbospray ion source in positive ion mode.

Absorption spectroscopy
Samples were reduced in an anaerobic glove bag (Coy) with 5 mM Na 2 S 2 O 4 for 15 min, and then desalted with a BioSpin6 column equilibrated with Buffer E (50 mM HEPES, pH 7.5, 150 mM NaCl, 5% (v/v) glycerol, 0.22 mm filtered). CO-saturated buffer (950 mM CO) was prepared by sparging 3 mL of anaerobic Buffer E for 15 min in a Reacti-Vial (Thermo Fisher Scientific). CO was added to the sample to achieve a final concentration of 425 mM. Nitric oxide was added to a concentration of 500 mM by addition of DEA NONOate, based on 1.5 moles of NO released per mole of NONOate. Protein-ligand complexes were incubated for 15 min at room temperature to establish equilibrium, and no further spectral changes were observed after this time. Samples were placed in a septum-sealed 1 cm pathlength quartz cuvette inside the glove bag, and UV-Vis spectra were recorded on a Cary 300 spectrophotometer (Agilent Technologies).

Activity assays and quantification
Steady-state kinetics for Ms sGC were measured by quantifying the amount of cGMP produced in duplicate endpoint activity assays, performed in at least biological triplicate. Samples were reduced in an anaerobic glove bag with 5 mM Na 2 S 2 O 4 for 15 min, and then desalted with a BioSpin6 column equilibrated with Buffer E. Ms sGC with 1-NO and xsNO were prepared by first adding PROLI NON-Oate to 50 mM, based on 2 moles of NO released per mole of PROLI NONOate. This sample was then buffer exchanged into Buffer E through a BioSpin6 column to generate the 1-NO state. To generate the xsNO state, PROLI NONOate was added back to a portion of the 1-NO sample and allowed to equilibrate for 5 min. YC-1 was added from a 100x stock solution in DMSO to a final concentration of 150 mM (final DMSO concentration 1%). The protein concentration was determined using the reduced heme Soret peak at 433 nm (149,000 M -1 cm -1 ) (Hu et al., 2008). Protein concentration was adjusted after desalting the excess NO by comparing the A 280 peaks to the unliganded protein. Activity assays were conducted at 25˚C and pH 7.5 in Buffer D, supplemented with 5 mM DTT and 5 mM MgCl 2 . Reactions were initiated with 2 mM GTP and timepoints from were quenched with 125 mM Zn(CH 3 CO 2 ) 2 , followed by 125 mM Na 2 (CO 3 ) to adjust the pH to 10.5. Samples were frozen at -80˚C until quantification. Quenched assays were thawed, and the zinc precipitate was spun down for 10 min at 23,150 g. The reactions were diluted by one to three orders of magnitude, and the cGMP was quantified in duplicate using an extracellular cGMP Enzyme Linked Immunosorbent Assay, following the manufacturer's instructions (Enzo Life Sciences). Concentrations of cGMP were determined from a standard curve, generated over 0.16-500 pmol/mL. Initial rates were calculated from the linear phase of the time course, where 5-10% of the GTP substrate was consumed. The experiment was repeated at least three times to ensure reproducibility.

Cryo-EM sample preparation and data collection
Samples were reduced in an anaerobic glove bag with 5 mM Na 2 S 2 O 4 for 15 min, and then desalted with a BioSpin6 column equilibrated with Buffer F (25 mM triethanolamine, pH 7.5, 25 mM NaCl, 5 mM DTT, 0.22 mM filtered). The inactive protein sample was diluted after thawing from a single frozen stock to~2-4 mM in 1-3% trehalose. The sample (3.5 mL) was applied to glow discharged UltrA-Ufoil 2/2 200 mesh gold grids (Quantifoil) and plunge-frozen using a vitrobot Mark IV (Thermo Fischer). Blotting was performed under 100% humidity with zero blot force for 3-6 s. sGC was dispersed in vitreous ice which displayed clear contrast transfer function information (Figure 2-figure supplement 1a. Active sample was prepared in a similar manner, but 500 mM NO (from DEA NON-Oate) and 150 mM YC-1 were added (Figure 3-figure supplement 1a-b).
Grids were screened and imaged on a Talos Arctica (Thermal Fischer) operated at 200 kV. Complete imaging conditions are described in Supplementary file 1. Micrographs were collected at 36,000X nominal magnification on a K3 direct electron detector (Gatan) in super-resolution counted mode at 0.5685 å /pix. Serial EM was used for automated image shift data collection of 2841 and 9330 movies for the inactive and active sample, respectively. Movies were taken in 100 ms frames, totaling an electron dose of 60 electrons per movie.

Cryo-EM data processing
Inactive sGC movies were drift-corrected, gain-corrected, and binned to 1.137 Å /pix in 7 Â 5 patches using MotionCor2 (Zheng et al., 2017). Micrographs were CTF-corrected using CTFFIND4 and single particles were manually picked for initial 2D classification within Relion 3.0 (Rohou and Grigorieff, 2015;Scheres, 2012;Nakane et al., 2018). Class averages representing the full-length complex (Figure 2-figure supplement 1c) were used for template picking with Relion autopicker, resulting in 675,956 particles. Particles were imported into Cryosparc2 and pruned using 2D classification and 3D ab initio classification (Punjani et al., 2017). Initial 2D classification yielded a large number of falsely picked background particles and particles which represented broken sGC dimers. We suspect the background particles are due to the size of the complex, while the broken particles likely stem from damage at the air-water interface (Figure 2-figure supplement 1c). These poor class averages were removed from further processing steps. In total, 59,165 final particles were refined using non-uniform refinement with default parameters (Figure 2-figure supplement 1b). Active sGC movies were binned to 2.274 Å /pix before template-free Laplacian picking for initial particle selection (Nakane et al., 2018). Micrographs were manually cleaned by visual inspection and FFT quality. Initial single particles were pruned using iterative 2D and 3D techniques (Figure 3-figure supplement 1b, c), as described above for the inactive state. Similarly, many of the initial picked particles were background or broken in the Active dataset (Figure 3-figure supplement 1c) Blurred density at low thresholds was seen in a predicted region for the alpha-HNOX but did not resolve during processing and was masked away during the final reconstruction. Of note, both Inactive and Active datasets underwent exhaustive processing schemes, including Bayesian polishing, multibody refinement and focused classification, none of which improved the resolution or showed evidence of multiple conformations in a single dataset.

Model building
Homology models for each domain were built using Phyre2 and correspond to the following PDB entries: a H-NOX (2O0C), a PAS (4GJ4), a CC (3HLS), a CAT (3UVJ), b H-NOX (2O0C), b PAS (4GJ4), b CC (3HLS), b CAT (2WZ1) (Kelley et al., 2015). Domains (with side chains removed) were rigid-body docked into the inactive reconstruction using the fit_in_map function of Chimera (Pettersen et al., 2004). Clear density for heme in the b H-NOX domain enabled clear distinction of the two H-NOX domains. Linkers between the H-NOX and PAS domains are missing in the density, likely due to flexibility, and were left out of the model (a 238-278 and b184-206). Initial placement of the CAT dimer was based on a significant extension in the sequence of the a CAT C-terminus, which is visible as unmodeled density. Continuous density from the CAT dimer enabled assignment of the CC and PAS domains. The linker region between the PAS domains and the CC were manually modeled in COOT based on secondary structure predictions from PSIPRED (Emsley et al., 2010;Vynne, 1997). Refinement using iterative rounds of phenix.real_space_refine and inspection in COOT led to the final carbon back bone trace of the inactive state. Active state modeling was based on rigid-body fitting of domains based on the inactive state. Helical density was apparent throughout the structure (with the exception of the a H-NOX, as mentioned above). Linker regions between domains were corrected with COOT and phenix.real_space_refine. Final model idealization was carried out using phenix.model_idealization and validated using Molprobity (Chen et al., 2010).

Small-angle X-ray scattering in-line with size-exclusion chromatography (SEC-SAXS)
Samples were reduced in an anaerobic glove bag with 5 mM Na 2 S 2 O 4 for 15 min, concentrated to 60 mL at 50 mM (~7.5 mg/mL), and then three rounds of dialysis were performed in Buffer G (50 mM KH 2 PO 4 , pH 7.4, 150 mM NaCl, 2% glycerol). The activated sample was prepared with 500 mM NO (from DEA NONOate). The samples were sealed and run within 3 hr of being prepared; samples did not undergo freeze-thaw after the sample was prepared (the protein was freeze-thawed once after purification for initial storage).
In situ sample purification was accomplished through SEC to isolate well-folded proteins from aggregates and other impurities immediately before exposure to synchrotron X-ray radiation using a custom designed flow cell. SEC-SAXS was collected at the SIBYLS beamline (bl12.3.1) at the Advanced Light Source at Lawrence Berkeley National Laboratory, Berkeley, California (Classen et al., 2010;Classen et al., 2013). X-ray wavelength was set at l = 1.127 Å and the sample-to-detector distance was 2105 mm, as determined by silver behenate calibration, resulting in scattering vectors, q, ranging from 0.01 Å -1 to 0.4 Å -1 . The scattering vector is defined as q = 4psinq/l, where 2q is the scattering angle. Data was collected using a Dectris PILATUS3 Â 2M detector at 20˚C and processed as previously described (Dyer et al., 2014;Hura et al., 2009). Briefly, a custom-made SAXS flow cell was directly coupled with an Agilent 1260 Infinity HPLC system using a Shodex KW-803 column. The column was equilibrated with running Buffer E with a flow rate of 0.5 mL/min for inactive sGC and 0.55 mL/min for activated sGC. To achieve activation conditions, the buffer was continuously sparged with nitrogen gas and the column was equilibrated for at least 2 hr to maintain an anaerobic environment. Several NONOates with various half-lives were added to the running buffer to achieve activation of sGC. These NONOates include DEA NONOate and DETA NONOate, which spontaneously release NO with half-lives of 16 min and 56 hr, respectively. Each sample was run through the SEC-SAXS system and 3 s X-ray exposures were collected continuously over the 30 min elution. The SAXS frames recorded prior to the protein elution peak were used to correct all other frames. The corrected frames were investigated by radius of gyration R g derived by the Guinier approximation I(q)=I(0) exp(-q 2 R g 2 /3) with the limits q*R g <1.3 ( Figure 5-figure supplement 2). The elution peak was mapped by comparing the integral of ratios to background and R g relative to the recorded frame using the program SCÅ TTER ( Figure 5-figure supplement 1). The frames in the regions of least R g variation were averaged and merged in SCÅ TTER to produce the highest signal-to-noise SAXS curves. These merged SAXS curves were used to generate the Guinier plots, volumes-of-correlation (V c ), pair distribution functions, P(r), and normalized Kratky plots. The Guinier plot indicated an aggregation-free state of the protein ( Figure 5-figure supplement 2). The P(r) function was used to determine the maximal dimension of the macromolecule (D max ) and estimate inter-domain distances ( Figure 4A) (Putnam et al., 2007). P(r) functions were normalized based on the molecular weight (MW) of the assemblies, as determined by the calculated V c .
Eluent was subsequently split (4 to 1) between the SAXS line and a multiple wavelength detector (UV-vis), set to 432 and 280 nm, multi-angle light scattering (MALS), and refractometer. The ratios of the protein (280 nm) and Soret band (432 nm) of the heme from SEC were used to evaluate the ligation state of Ms sGC upon NO binding ( Figure 5-figure supplement 1). MALS experiments were performed using an 18-angle DAWN HELEOS II light scattering detector connected in tandem to an Optilab refractive index concentration detector (Wyatt Technology). System normalization and calibration was performed with bovine serum albumin using a 60 mL sample at 10 mg/mL in the same SEC running buffer and a dn/dc value of 0.185 and 0.15 mL/g for inactive and active sGC respectively. The MALS data was used to compliment the MWs calculated by the SAXS analysis and, being furthest downstream, the MALS peaks were used to align the SAXS and UV-vis peaks along the x-axis (elution volume in mL/min) to compensate for fluctuations in timing and band broadening (Figure 5-figure supplement 1). UV-vis data was integrated using Agilent Chemstation software and baseline corrected using Origin 9.1 ( Figure 5-figure supplement 1). MALS and differential refractive index data was analyzed using Wyatt Astra seven software to monitor the homogeneity of the sample molecular weights across the elution peak, complementing the SEC-SAXS signal validation ( Figure 5-figure supplement 1).

Solution structure modeling
The cryo-EM refined crystal structure for inactive sGC was used to build an initial atomistic model; all missing loops and terminal residues were added using MODELLER (Fiser et al., 2000). A purpose-designed rigid body modeling pipeline was applied to the completed structure using BIL-BOMD to systematically explore conformational space for both the inactive and activated states of sGC (Video 4) (Pelikan et al., 2009). To obtain an inactive state model, the a H-NOX domain and both a terminal tails were moved, then all a and b domains were moved by allowing flexibility at the hinge region where the CCs meet the CAT domains, while holding the CAT domains fixed and still permitting flexibility of the terminal tails. This was then used to generate the 1-NO state model by moving the same regions as the inactive state in reverse order. Theoretical SAXS curves were produced by FOXS (Schneidman-Duhovny et al., 2010;Schneidman-Duhovny et al., 2013). Each model generated through BILBOMD as compared to the experimental SAXS profiles to assess the goodness of fit. Multistate model ensembles for activated sGC were determined using MultiFOXS (Schneidman-Duhovny et al., 2016).