Discovery of stimulator binding to a conserved pocket in the heme domain of soluble guanylyl cyclase

Soluble guanylyl cyclase (sGC) is the receptor for nitric oxide and a highly sought-after therapeutic target for the management of cardiovascular diseases. New compounds that stimulate sGC show clinical promise, but where these stimulator compounds bind and how they function remains unknown. Here, using a photolyzable diazirine derivative of a novel stimulator compound, IWP-051, and MS analysis, we localized drug binding to the β1 heme domain of sGC proteins from the hawkmoth Manduca sexta and from human. Covalent attachments to the stimulator were also identified in bacterial homologs of the sGC heme domain, referred to as H-NOX domains, including those from Nostoc sp. PCC 7120, Shewanella oneidensis, Shewanella woodyi, and Clostridium botulinum, indicating that the binding site is highly conserved. The identification of photoaffinity-labeled peptides was aided by a signature MS fragmentation pattern of general applicability for unequivocal identification of covalently attached compounds. Using NMR, we also examined stimulator binding to sGC from M. sexta and bacterial H-NOX homologs. These data indicated that stimulators bind to a conserved cleft between two subdomains in the sGC heme domain. L12W/T48W substitutions within the binding pocket resulted in a 9-fold decrease in drug response, suggesting that the bulkier tryptophan residues directly block stimulator binding. The localization of stimulator binding to the sGC heme domain reported here resolves the longstanding question of where stimulators bind and provides a path forward for drug discovery.

Nitric oxide (NO) signaling is compromised in numerous forms of vascular pathology (1, 2), and the components of the NO signaling pathways are highly sought-after therapeutic targets (3). Central to NO signaling is soluble guanylyl cyclase (sGC), 2 the NO receptor, which regulates vascular tone, platelet activation, wound healing, and other factors of importance to cardiovascular health (4,5). sGC is an ϳ150-kDa heterodimeric NO sensor composed of ␣ and ␤ subunits, with an ␣1/␤1 isoform predominating in vascular tissue (also referred to as guanylyl cyclase-1 (GC-1)), and an ␣2/␤1 isoform predominating in nerve cells (called GC-2), where it is important for memory formation. The sGC ␤1 subunit contains an N-terminal hemenitric oxide/oxygen-binding (H-NOX) domain, a Per-ARNT-Sim (PAS) domain, a coiled-coil domain, and a catalytic cyclase homology domain (Fig. 1A). The sGC ␣1 subunit was likely formed through gene duplication (5) and retains a similar domain arrangement to the ␤1 subunit except that the H-NOX domain has lost the ability to bind heme and is best referred to as a pseudo H-NOX domain.
sGC binds NO on a ferrous heme in the ␤1 H-NOX domain, leading to allosteric stimulation of cyclase activity, production of cyclic guanosine-3Ј,5Ј-monophosphate (cGMP) from guanosine-triphosphate (GTP), and a downstream signaling cascade. A single active site is formed at the interface of the ␣ and ␤ cyclase homology domains, with each subunit contributing the amino acids necessary for catalysis. sGC is therefore an obligate heterodimer.
sGC is targeted pharmaceutically to treat numerous vascular disorders, including acute coronary syndromes, congestive heart failure, and arterial hypertension, using NO donors and organic nitrates (6). Although these compounds exhibit potent vasodilatory and anti-ischemic effects, tolerance develops readily, and cellular damage by excess NO can occur (6). More recently, compounds that increase cGMP production without altering cellular levels of NO have been sought. sGC stimulators were the first compounds to overcome the limitations of NO donors and organic nitrates by enhancing cro ARTICLE cyclase activity both independently and synergistically with NO (3,(7)(8)(9)(10). Optimization of initial stimulator compounds led to the development of BAY 41-2272, which is widely used for investigating stimulator mechanism, and BAY 63-2521 (riociguat, marketed as Adempas), which is clinically approved to treat pulmonary arterial hypertension (PAH) and chronic thromboembolic pulmonary hypertension (CTEPH) (3). Additional sGC stimulators have been developed, including IWP-051, a novel compound representing a new class of stimulators with improved solubility over traditional stimulators and favorable pharmacodynamics properties (11).
Despite clinical success with compounds, where sGC stimulators bind and how they function remains unknown. Here, we have identified the binding of stimulator compounds to the sGC heme domain and bacterial H-NOX homologs using NMR approaches and a unique photoactive labeling stimulator with a signature cleavage pattern that allows unambiguous LC-MS/MS peptide assignment.

Development of a photolyzable stimulator, IWP-854
To localize stimulator binding to sGC, we synthesized a photolabile compound called IWP-854. The IWP-854 core motif is based on IWP-051, which replaces the 7-azaindazole core of BAY 41-2272 with pyrazole and isoxazole five-membered rings capable of free rotation (Fig. 1B) (11). Previous studies indicate that alteration of the benzyl ring abolishes binding, whereas modifications to the pyrimidine ring are widely tolerated (11,12). With this in mind, we modified the pyrimidine ring to have a biotin affinity tag coupled to a PEG linker and a photoactive diazirine capable of covalently attaching to all 20 amino acid side chains and the peptide backbone (13)(14)(15) (Fig. 1B). Our synthetic scheme is described in the supporting information.

IWP-854 retains stimulator activity
We examined the stimulation of recombinant human (Hs) sGC and found IWP-854 and BAY 41-2272 to stimulate to a similar extent ( Fig. 2A). Both basal and NO-stimulated activities were enhanced, as described previously for sGC stimulators (7,8,16). IWP-854 increased activity by 12-fold over basal levels in the absence of added NO and by 84-fold over basal upon the addition of NO.
Previously, we developed truncated versions of sGC from Manduca sexta for analyses of compound-enhanced CO binding (17)(18)(19)(20), which we refer to as Ms sGC-NT (Fig. 1A). Ms sGC-NT constructs are fully heme-loaded and stable in the ferrous (functional) state, as indicated by Soret band absorption (Fig. S1). One hallmark of the stimulator compounds is their ability to enhance CO and NO binding to the heme domain (reviewed in Ref. 5). CO binding to Ms sGC-NT23 in the absence of stimulator compound displays K d CO ϭ 710 nM (Fig.  2B). The addition of BAY 41-2272 enhances CO affinity 14-fold (K d CO ϭ 52 nM), whereas the addition of IWP-854 enhances CO affinity 34-fold (K d CO ϭ 21 nM). Thus, IWP-854 stimulates as well or somewhat better than the best previously described compounds.

IWP-854 and BAY 41-2272 share a common binding site
Labeling with IWP-854 was visualized by probing the biotin affinity tag through Western blot analysis (Fig. 2, C-E). A time course revealed that labeling to Ms sGC-NT23 was observed after 5 min of UV illumination and continued to increase for 15-20 min (Fig. S2). Heme was retained after 15 min of UV irradiation, as indicated by a minimal decrease in Soret band absorption and a slight shift in Soret maxima characteristic of stimulator binding (20) (Fig. S3). Longer exposures to UV light led to substantial heme loss, however, and were avoided for this reason. Using this strategy, we found that IWP-854 exclusively labels the ␤1 subunit of Ms sGC-NT23 (Fig. 2C).

Stimulator binding to the sGC ␤1 H-NOX domain
To assess compound specificity, IWP-854 labeling was monitored in the presence of increasing concentrations of BAY 41-2272 (Fig. 2C). Incubation with BAY 41-2272 attenuated IWP-854 labeling of Ms sGC-NT23 in a concentration-dependent manner, indicating that the two compounds were competing for a single binding pocket. Similar results were observed for Ms sGC-NT13 (Fig. S4A), which includes the ␣1 pseudo H-NOX domain, and full-length Hs sGC (Fig. 2D). Faint labeling was observed on the ␣1 subunit of Hs sGC; however, this labeling was not competed away with excess stimulator and is likely due to nonspecific labeling. IWP-854 also labeled the isolated ␤1 subunit of Ms sGC-NT21 (␤1 residues 1-380 (Fig.  S4B)). Curiously, BAY 41-2272 failed to diminish IWP-854 labeling of Ms sGC-NT21 ␤1. However, IWP-854 labeled the same residues in Ms sGC-NT21 ␤1 as in other Ms sGC-NT constructs (described below).
Yoo et al. (21) reported previously that BAY 41-2272 alters the release kinetics of CO not only from sGC but also from a homologous bacterial H-NOX domain of Clostridium botulinum, commonly referred to as Cb SONO (22). We therefore characterized stimulator binding to Cb SONO, as well as three previously described H-NOX proteins from Nostoc sp. PCC 7120 (Ns H-NOX), S. oneidensis (So H-NOX), and S. woodyi (Sw H-NOX) (23). IWP-854 labeled all four bacterial H-NOX proteins (Figs. 2E and S4C), suggesting that stimulator binding is conserved among ␤1 H-NOX domains. Photoaffinity labeling of bacterial H-NOX proteins required 10-fold more IWP-854 than sGC constructs, indicating a decreased affinity for com-pound binding. Labeling was reduced but not eliminated by excess BAY 41-2272, which is likely explained by the inability to reach sufficiently high BAY 41-2272 concentrations due to poor compound solubility and by the increased nonspecific labeling that occurs at higher IWP-854 concentrations. Interestingly, BAY 41-2272 (10 M) did not enhance CO binding to any of the four bacterial H-NOX proteins (Table S2). This may be due to the weaker compound binding but is also consistent with previous results with sGC, where compound enhanced CO binding to the heterodimeric protein but not to the isolated H-NOX domain (20).

Identifying labeled residues in sGC and bacterial H-NOX proteins
Residues labeled by IWP-854 were identified by liquid chromatography tandem mass spectrometry (LC-MS/MS) using an LTQ Velos Orbitrap mass spectrometer (Thermo Fisher Scientific). Initial examination of IWP-854 alone (molecular mass 1,450.743 Da) revealed a distinct and highly reproducible fragmentation pattern (Fig. S5). Key features include a singly charged peak at m/z 270.127 and a peak 1 charge state less than the precursor representing the mass of the parent ion minus 270.127 Da. MS 3 analysis of m/z 270.127 identified the fragment as the end of the biotin-containing PEG linker (C 12 H 20 O 2 N 3 S, Figs. 1B and S5). This signature cleavage pattern was observed consistently in labeled peptides (Fig. 3A), providing a robust strategy for definitively identifying peptides modified by IWP-854. A possible mechanism for the character-

Stimulator binding to the sGC ␤1 H-NOX domain
istic fragmentation pattern is for the linker amide near the cleavage site to yield a localized mobile proton that assists in the cleavage event (24), thus reproducibly generating the m/z 270.127 fragment. More broadly, judicious placement of an amide bond in a PEG linker may provide a unique cleavage pattern of general applicability.
The availability of multiple Ms sGC-NT constructs in high purity and abundance allowed for numerous experiments to be undertaken under varying conditions. Hs sGC and four bacterial H-NOX proteins were also examined under conditions similar to those developed initially with Ms sGC-NT. The results from a total of 43 experiments are reported in Tables 1 and S3. Representative sequence coverage for each protein is depicted in Fig. S6.
The identification of peptides was to high mass accuracy in all cases; however, certain peptides were detected more often than others (Table 1). Most labeled residues identified in Ms sGC-NT and Hs sGC are expected to originate from the stim-ulator-binding pocket, as evidenced by diminished labeling in the presence of excess BAY 41-2272 (Fig. 2C). In general, the diversity in labeling of sGC likely results from dynamics in both the compound and the protein. For IWP-854, the diazirine is at the fourth carbon of a 5-atom flexible linker attached to a pyrimidine ring capable of free rotation (Fig. 1B). The pyrimidine ring was shown previously to be tolerant to some modification (11), suggesting that it may exhibit greater conformational dynamics than the rest of the compound, even in the bound state.
The labeling of Ms sGC-NT23 remained the same in the presence or absence of NO or CO, consistent with a stimulatorbinding site that does not change greatly upon heme ligation. Likewise, labeling did not differ appreciably in the presence (Ms sGC-NT13) or absence (Ms sGC-NT23) of the ␣1 pseudo H-NOX domain, indicating that this domain does not harbor the stimulator-binding site. IWP-854 labeling of Ms sGC-␤1 is similar to the other Ms sGC-NT constructs, despite lacking the ␣1 chain and displaying poor competition with BAY 41-2272. (1-399)) generated through homology modeling, SAXS analysis, and chemical cross-linking (19). The subdomains of the ␤1 H-NOX domain are shown in green (N-terminal subdomain) and light blue (C-terminal subdomain). The remainder of the ␤1 subunit is depicted in tan, and the ␣1 subunit is shown in light gray. Modified residues are labeled and shown in red.

Stimulator binding to the sGC ␤1 H-NOX domain
Labeled peptides identified in the Ms sGC-NT constructs agreed well with those in full-length Hs sGC and were found nearly exclusively in the ␤1 subunit, as expected from the Western blot analyses. Likewise, many labeled peptides identified in the bacterial H-NOX proteins overlapped with those from sGC.
The labeling of residues Ms sGC-NT ␤1(195-198) was seen in 11 of 26 measurements; however, these residues lie in the linker between the H-NOX and PAS domains and in a different region of our model. This discrepancy could be due to limitations in our modeling, a slight degree of nonspecific binding, or high dynamics in this loop.
A number of additional labeled peptides, detected on a less frequent basis in these experiments, are listed in Tables 1 and   S3. Modifications to the ␤1 H-NOX/PAS linker were observed in all three Ms sGC-NT constructs but not in Hs sGC constructs. Additionally, a variety of labels were detected in individual bacterial H-NOX proteins that do not agree with the most common binding arrangement. These are likely the result of unspecific labeling introduced by increased compound concentrations and/or changes in compound affinity, as evidenced by incomplete elimination of IWP-854 labeling by competition with BAY 41-2272. For this reason, only labeled residues that were identified in multiple bacterial H-NOX proteins were considered to be part of the binding site.
Stimulators have been proposed to bind to a pseudosymmetric site in the cyclase domains similar to forskolin binding to adenylyl cyclase (25,26). Ms sGC-NT constructs retain stimulator binding and response, despite lacking both cyclase domains, suggesting that the primary stimulator-binding site resides in the N-terminal two-thirds of the protein (18 -20). We examined the possibility of a secondary stimulator-binding site in the catalytic domains using photoaffinity labeling of fulllength Hs sGC. A single label was found in the cyclase domain (Hs residue ␣1(629) ( Table S3)), which lies on the surface of the protein near where the coiled coil attaches. No labeling was found of residues in the cyclase domain active site or pseudosymmetric site, rendering the possibility of a secondary stimulator-binding site unlikely. One additional label to the human ␣1 chain was observed (Hs peptide ␣1(45-47) (Table S3)). The two ␣1 chain labels identified by mass spectrometry may represent the nonspecific ␣1 labeling observed by Western blotting. Table 1 Summary of residues multiply modified by IWP-854 Included are peptides and sequence regions modified in more than one species. Modified residues are listed where known. A range of residues is listed where the exact modified residue could not be determined due to incomplete fragmentation. All peptides were in either the ϩ3 or ϩ4 charge states and had masses between 2200 and 3900 Da. Complete mass and charge information can be found in Table S2.

Stimulator binding to the sGC ␤1 H-NOX domain Characterization of compound binding by transferred NOESY NMR
IWP-051 binding to Ms sGC-NT23, Cb SONO, and Sw H-NOX was further examined by transferred nuclear Overhauser effect spectroscopy (TrNOESY). TrNOESY provides information on the protein-bound conformation of the compound by measuring proton-proton NOE cross-relaxation enhancement. Upon binding the protein, the ligand acquires a large molecular correlation time leading to substantial enhancement of NOE cross-relaxation between protons. Thus, proteinbound compounds exhibit strong negative NOE and significantly shorter mixing times compared with small positive NOE and the longer mixing times exhibited by free ligand (27).
The NOESY spectrum of IWP-051 (Fig. 4A) in the absence of protein displayed weak positive NOE at longer mixing times (Ͼ1.2 s) (Fig. 4B, upper panel, and Table S4). Prominent NOE include those between hydrogens in the benzyl ring (H13, H14, H15, and H16) and the isoxazole ring (H9 and H10). Most NOE involving the methylene hydrogens (H11) were not observed, consistent with high rotational dynamics in the methylene bridge and benzyl ring. No NOE peaks were observed at shorter mixing times (Ͻ1.0 s).
Upon adding Ms sGC-NT23, strong negative NOE appeared at a short mixing time typical for protein molecules (400 ms). Three new NOE were observed for protein-bound IWP-051, all involving the methylene-bridging hydrogens (H2-H11, H9 -H11, and H11-H14; Fig. 4B, lower panel). One NOE peak (H10 -H20) was lost. As a control, we examined binding by compound PF-04447943, a phosphodiesterase 9A (PDE9A) inhibitor (28) that does not stimulate sGC. No NOE were observed with this compound (Fig. S7). Together, these data suggest specific binding for IWP-051 to Ms sGC-NT23. Approximate protonproton distances for IWP-051 derived from TrNOESY are reported in Table S4 with a comparison to the distances from the model structure, which are in good agreement. IWP-051 binding to Cb SONO and Sw H-NOX was also examined, with both proteins generating negative TrNOESY peaks. However, the spectra display weaker intensities than Ms sGC-NT23, likely due to the smaller sizes of the bacterial H-NOX proteins (Fig. S7).
These data reveal several key factors related to stimulator binding. First, a strong NOE peak between two protons (H2-H10) was observed upon binding (Fig. 4B), suggesting that the isoxazole and pyrazole rings are roughly planar and the protons from each ring lie near one another. Because the magnitude of NOE peaks decreases rapidly with distance and are measurable only to ϳ6 Å, the two rings must be nearly co-planar for a strong TrNOESY peak to appear in IWP-051. Second, there was a weak TrNOESY peak between an isoxazole proton and the methylene protons (H10 -H11 (Fig. 4B)), consistent with the orientation shown in this figure and the previously reported structure-activity relationship of IWP-051 (11). Third, the

Stimulator binding to the sGC ␤1 H-NOX domain
appearance of several TrNOESY peaks for the benzyl ring indicates that it occupies a single conformation in the binding pocket. Finally, and most importantly, stimulator binding was conserved from bacterial H-NOX proteins to human sGC.

Chemical shift perturbation in Sw H-NOX HSQC indicates binding pocket
Sw H-NOX displayed well-dispersed HSQC spectra, and 75% of the backbone assignments have been reported (29). We therefore completed the backbone assignment and undertook chemical shift perturbation analyses to uncover which residues would respond to binding IWP-051 (Figs. 5, S8, and S9 and Tables S5 and S6). Several residues displayed prominent concentration-dependent shifts in resonance upon titration with IWP-051 ( Fig. 5B) but not with PDE9A inhibitor. The largest shifts clustered together near the C-terminal end of helix ␣A (Sw residues 14 -18) and the N-terminal end of helix ␣D (Sw residues 61-76 (Fig. 5C)). These regions are in contact in our homology model (Fig. 5A and described below). Interestingly, they also reside near the predicted binding site of H-NOXassociated cyclic-di-GMP synthase/phosphodiesterase, the signaling partner for Sw H-NOX (29). Significant shifts were also observed for residues Glu-27 and Glu-57, near the ␣A/␣D interface. Although conserved, IWP-051 binding to Sw H-NOX is weaker than to Ms sGC-NT, with K d ϭ 1.9 mM estimated for the CO complex by chemical shift perturbation NMR titration ( Fig. 5B and Table S6) compared with K d values of 0.03-3.8 M for various stimulator compounds binding to the CO complexes of Ms sGC-NT constructs (20).

Molecular modeling of stimulator binding to sGC
Our labeling and NMR data indicate that the functional binding site for stimulator compounds resides in the ␤1 H-NOX domain. Although a high-resolution structure of an

Stimulator binding to the sGC ␤1 H-NOX domain
sGC H-NOX domain has not been reported, crystal structures for several bacterial homologs are known, including those from Caldanaerobacter subterraneus (30), also known as Thermoanaerobacter tengcongensis (22,31), Nostoc sp. PCC 7120 (32,33), and Shewanella oneidensis (34). These structures display the same overall fold and provide a solid scaffold for understanding H-NOX structure in sGC function (reviewed in Refs. 4, 5, and 23). The overall H-NOX fold is ϳ180 residues long and displays an N-terminal subdomain encompassing residues 1-60, which is dominated by a 3-helix bundle followed by a larger mixed helix/sheet subdomain that contains the heme pocket. Alignment of the larger subdomains of several H-NOX structures indicate that the smaller and larger domains can move independently of one another, altering the orientation of the two domains, proposed as the key for signal transduction by H-NOX domains and proteins (29,30,32).
Most of the residues implicated in compound binding by photoaffinity labeling and NMR lie at the interface of the H-NOX subdomains (Fig. 5A), making this interface an intriguing possibility for stimulator binding. Whereas the large domain includes the heme-binding pocket, the small domain covers the heme distal pocket and contacts the heme edge. It is easy to imagine that changes in the subdomain interface would affect heme properties, ligand affinity, and signal transduction. Thus, our working hypothesis is that stimulator compounds bind in this region, and we have modeled IWP-854 binding into a pocket at the subdomain interface (Fig. 6A).
The modeling of compound into heterodimeric sGC was more challenging because atomic-level models were unavail-

Stimulator binding to the sGC ␤1 H-NOX domain
able. For this purpose, we utilized a previously described model for Ms sGC-NT based on small-angle X-ray scattering (SAXS), chemical cross-linking, and homology modeling of the H-NOX, PAS, and coiled-coil domains (19). The labels found most frequently in the present study were to the coiled coil. Encouragingly, these residues in the coiled coil lie near the labeled residues in the ␣A, ␣C, and ␣D helixes of the ␤1 H-NOX domain (Fig. 6B).

Mutational analysis of the proposed stimulator-binding site
Mutations designed to block compound access were introduced into the ␤1 H-NOX subdomain interface. Three residues predicted to occupy different regions of the binding site, Leu-12, Thr-48, and Ile-66, were targeted for mutation and changed to tryptophan. Mutations were originally generated in Ms sGC-NT23, which lacks the ␣1 H-NOX domain (Fig. 1A); however, these constructs led to insoluble protein. We then turned to Ms sGC-NT25, which resembles Ms sGC-NT13 but is extended by 9 residues at the C-terminal end of both ␣1 and ␤1 subunits. Mutations L12W, T48W, and L66W and the double mutation L12W/T48W were engineered into Ms sGC-NT25, successfully purified, and examined for CO binding affinity in the presence or absence of BAY 41-2272 ( Table 2).
The T48W single mutation resulted in ϳ3-fold weaker binding affinity for CO in the absence of BAY 41-2272 (Table 2). No appreciable changes in CO binding affinity were observed for the L12W and I66W single mutations, whereas the L12W/ T48W double mutant bound CO ϳ2-fold weaker than the wild type. These data indicate that the mutations had a minimal effect on the heme environment.
In contrast, stimulation by BAY 41-2272 was reduced in all of the mutated proteins, quite dramatically in two cases. The addition of 10 M BAY 41-2272 to wild-type protein increased CO binding affinity by 65-fold, as expected. The single mutations T48W and I66W led to proteins with moderately less response to BAY 41-2272, whereas mutation L12W displayed only 13-fold enhancement by stimulator, and double mutant L12W/T48W lost nearly all response, displaying only 7-fold enhancement in the presence of BAY 41-2272, which was 9-fold worse than wild type.

Discussion
sGC stimulators were first discovered more than 20 years ago and have since been improved such that one compound is in clinical use and several more are likely to follow. Yet the understanding of where these compounds bind and how they function has lagged, potentially hampering the discovery of novel compounds with enhanced pharmacological properties. Here, we resolved the longstanding question of where stimulators bind, narrowing the binding site to the heme domain where NO binding stimulates catalytic activity. The stimulator-binding site is apparently evolutionarily conserved and found in the H-NOX domains, as they first appeared in bacteria.
In the present study, we used a photoactivatable stimulator compound coupled with LC-MS/MS, along with TrNOESY and chemical shift perturbation NMR, to narrow the binding site to the ␤1 H-NOX domain. No binding was observed to the cyclase domain active site or to the pseudosymmetric site, nor was there binding to the ␣1 pseudo H-NOX domain, as sug-gested by a previous labeling study (35). This discrepancy likely results from the choice of labeling reagent, with the present study utilizing a diazirine versus an aryl azide used in the former study. Diazirines improve upon aryl azides with quicker reaction times and a lower frequency of stable intermediates capable of diffusing away from the binding site (15), which in the case of aryl azides, includes ketenimine decay products that react strongly with nucleophiles such as cysteines. Because both residues previously identified were cysteines (35), the reaction may have been with the ketenimine decay product.
We modeled a possible binding complex in which stimulators bind at the interface of the two H-NOX subdomains, where most of the residues with labeling and NMR chemical shift peaks were located (Fig. 6A). This pocket was identified previously as part of a tunnel suggested to be of importance for NO, CO, and O 2 gas exchange with the heme distal pocket in bacterial H-NOX proteins (33,34). Filling this pocket with compound provides a possible mechanism for stimulation and may explain the conservation of binding in H-NOX proteins. With the most critical portions of stimulator compounds filling the gas-exchange tunnel, we modeled the IWP-854 pyrimidine ring, which contains the photoactive diazirine, to be near major H-NOX and coiled-coil labeled peptides, using our previously modeled domain arrangement (Fig. 6B) (19). Because the coiled coil was labeled in all experiments in which it was present, these results suggest a compact domain arrangement of sGC is likely to occur in high abundance.
To test our model, we introduced a series of mutations into the proposed binding pocket ( Table 2). All three mutations led to proteins with reduced response to stimulator while retaining similar CO binding affinity as wild-type protein in the absence of compound. The double mutant L12W/T48W displayed the lowest stimulator response, yielding only a 9-fold reduction in CO enhancement compared with wild type. Although it is possible that general allosteric response is altered in the mutated proteins, it is more likely that the L12W and T48W mutations physically blocked stimulator binding or distorted the pocket such that binding affinity was lost.
How do stimulator compounds stimulate? The answer to this question is still unknown, but key factors are becoming apparent (recently reviewed in Ref. 4,5). One key feature of stimulator binding is its ability to enhance affinity for NO and CO to the sGC heme, which may in turn enhance catalysis. Increased NO/CO affinity is due in part to increased geminate recombination, a process in which trapping dissociated gas molecules in the protein favors rebinding to heme over escape into bulk solution (17,36). A variety of mechanisms might contribute to increased geminate recombination upon stimulator binding, including induction of a heme conformation with enhanced ligand binding on-rates or the blocking of ligand escape paths. The sGC ␤1 H-NOX domain has inherently high affinity for NO and CO in the isolated state. This affinity is dampened as additional domains are included in the protein, with the addition of ␣1 pseudo H-NOX, ␣1 PAS, and the full-length coiled coil, each serving to lower heme affinity for gaseous ligands (19,20). The binding of stimulator partially overcomes the inhibitory effect of additional sGC domains, possibly through the release of domain contacts or through direct binding to a high-

Stimulator binding to the sGC ␤1 H-NOX domain
affinity H-NOX domain conformation, or both. Here, we show that stimulator binding is not only directly to the H-NOX domain but may also plug a proposed tunnel implicated in gas molecule release from the distal pocket. Additionally, analysis of bacterial H-NOX proteins suggests that movements in the N-and C-terminal subdomains are intimately connected to heme geometry and ligand binding affinity (23).
A second key feature for stimulator activity is its linked equilibria with NO and CO binding. The binding of NO or CO leads to an sGC conformation with higher affinity for stimulator compound just as stimulator binding leads to higher affinity for NO and CO. The simplest model for explaining such linked equilibria is through allostery: there is an H-NOX conformation that binds both gaseous ligand and stimulator with high affinity, and binding of either stimulator or NO/CO induces this conformation.
Finally, there is a linked equilibrium between the affinity of NO at the heme and GTP at the active site. The binding of NO leads to a cyclase conformation with higher GTP affinity, lowering K m for catalysis and also increasing V max . Binding of stimulator has the same effect. Binding of nucleotide alters NO affinity as one would expect from linked equilibria, although the details of this are complicated and may involve more than one nucleotide-binding site. Presumably, the affinity of stimulator for full-length sGC would also increase in the presence nucleotide, but this has not yet been examined.
In summary, we have shown that stimulators bind to the H-NOX domain of sGC as well as to bacterial homologs. Binding likely occurs at the interface of the H-NOX large and small subdomains and may act both through inducing an active conformation and through directly blocking a tunnel for gas release to bulk solvent. These data provide insight into sGC function and stimulator action as well as a roadmap for improved compounds targeting disease.

Materials
Chemicals were purchased from Sigma-Aldrich unless otherwise indicated. Uniformly labeled 15 N-ammonium chloride, 13 C-glucose, and deuterium oxide (D 2 O) were purchased from Cambridge Isotope Laboratories. 2-(N,N-Diethylamino)diazenolate-2-oxide (DEA/NO) was generously provided by Dr. Katrina Miranda (University of Arizona). Full-length human ␣1␤1 sGC expressed in Sf9 cells with C-terminal His tag was purchased from Enzo Life Sciences, Inc. (Farmingdale, NY). HEK293T cells were acquired from the American Type Culture Collection (ATCC). TurboFect was purchased from Fermentas. DMEM was purchased from Gibco/Life Technologies. Fetal bovine serum was obtained from the University of Arizona Cancer Center (Tucson). Sequencing-grade trypsin and chymotrypsin were acquired from Promega (Madison, WI), and C18 ZipTips were purchased from Pierce/Thermo Fisher Scientific.
Detailed synthetic procedures for IWP-854 are included in the supporting data.
Constructs coding for full-length human sGC were generated by amplifying cDNA coding for sequences for ␣1(1-690)-Strep-tag II and ␤1(1-619)-His 6 using primers F4, R4, F5, and R5. PCR product for Hs sGC ␣1 was cloned into plasmid pCMV_3TAG9 between the restriction sites BamHI and Hin-dIII to create plasmid pCMV_3TAG9_sGC␣1. PCR products for Hs sGC ␤1 was cloned into plasmid pCMV_3TAG3A between the restriction sites SacI and XhoI to create plasmid pCMV_3TAG3A_sGC␤1.

Expression and purification of Ms sGC-NT and bacterial H-NOX
Ms sGC-NT13, Ms sGC-NT21, Ms sGC-NT23 and Ms sGC-NT25 were expressed and purified from Escherichia coli as described previously (18 -20). Ms sGC-NT21 ␤1(1-380) was isolated from Ms sGC-NT21 by washing the column-bound sample with CO-saturated buffer, resulting in selective elution of the ␤1 subunit.
Plasmids coding for Ns H-NOX, So H-NOX, or Cb SONO were transformed into Rosetta-competent cells in Terrific Broth medium and grown at 37°C with shaking at 225 rpm.

Stimulator binding to the sGC ␤1 H-NOX domain
Once an A 600 of 0.8 -1.0 was reached, protein expression was initiated by adding 0.1 mM ␦-aminolevulinic acid and 0.5 mM IPTG. Expression continued for 18 -22 h at 16°C, and cells were harvested by centrifugation at 4500 rpm with a JLA-8.1000 rotor (Beckman Coulter) for 20 min at 4°C. Pellets were flash-frozen in liquid nitrogen and stored at Ϫ80°C.
Ns H-NOX, So H-NOX, and Cb SONO were purified by suspending the cells in buffer B (50 mM sodium phosphate, pH 7.4, 300 mM NaCl) supplemented with 0.75 mM DNase I and 1 mM PMSF. Cells were lysed by French press, and debris was removed by ultracentrifugation at 40,000 rpm in a Ti45 rotor for 35 min at 4°C. The supernatant was loaded onto a HisTrap FF nickel-nitrilotriacetic acid affinity column (GE Healthcare), and a sample was washed extensively with buffer B. Protein was eluted by supplementing buffer B with 30 mM EDTA followed by buffer exchange into 50 mM sodium phosphate, pH 7.5, 90 mM NaCl, and Ͻ0.3 mM EDTA using centrifugal filters. The His 6 affinity tag was removed by incubation overnight at 4°C with a 1:100 (protease:protein) molar ratio of TEV protease prepared as described previously (20). The sample was passed over a HisTrap FF column, and the cleaved protein was further purified using a Superdex 200 gel filtration column equilibrated with buffer D (20 mM Tris-HCl, pH 7.4, 100 mM NaCl). Protein concentrations were assessed based on Soret band absorption, and samples were frozen in liquid nitrogen for storage at Ϫ80°C.
Sw H-NOX was expressed and purified in the same manner as other H-NOX proteins with the following exceptions. For unlabeled Sw H-NOX, plasmids were transformed into E. coli Tuner (DE3) pLysS-competent cells and grown in M9 medium. Expression was induced with 0.05 mM ␦-aminolevulinic acid and 0.1 mM IPTG. Isotopically labeled Sw H-NOX was grown in M9 medium isotopically enriched for 15 N or 13 C/ 15 N, and protein expression was induced with 0.5 mM ␦-aminolevulinic acid and 0.5 mM IPTG. All buffers used to purify labeled Sw H-NOX were extensively degassed and supplemented with 0.5 mM tris(2-carboxyethyl)phosphine. Once loaded onto the HisTrap FF column, the protein was washed with buffer C (50 mM sodium phosphate, pH 7.5, 500 mM NaCl, 20 mM EDTA) and eluted by supplementing buffer C with 60 mM EDTA. A sample concentration was measured using a Pierce 660-nm protein assay (Thermo Fisher Scientific).

Determination of CO dissociation constants
CO titrations were measured using a Cary 350 UV-visible spectrophotometer (Agilent Technologies) and a cuvette with a 10-cm pathlength. K d CO measurements were determined as described previously (19,20), with the exception that samples were suspended in buffer A and BAY 41-2272 was solubilized with a final concentration of 5% EtOH.

Expression of human sGC
HEK293T cells were grown in DMEM supplemented with 10% FBS at 37°C with 5% CO 2 . A transfection mixture containing 22.5 g of pCMV_3TAG9_sGC␣1, 2.5 g of pCMV_ 3TAG3A_sGC ␤1, and 31.25 l of TurboFect (Thermo Fisher Scientific) was assembled in 2.5 ml of serum-free DMEM. The transfection mixture was incubated for 30 min at room temperature and added dropwise to a 10-cm dish containing HEK293T cells grown to 65-80% confluency. Protein expression continued for 24 h at 37°C with 5% CO 2 .

cGMP measurement
HEK293T cells transfected with human sGC were washed twice with PBS and suspended in buffer A supplemented with 8 mM MgCl 2 , 0.5 mM 3-isobutyl-1-methylxanthine (IBMX), 1 mM PMSF, and 1:100 dilution of protease inhibitor mixture (Sigma). Cells were lysed with 30 strokes of a 25-gauge needle, and debris was removed by centrifugation at 16,000 ϫ g for 20 min at 4°C. Lysate was combined with IWP-854 or BAY 41-2272 (5 M) and DEA/NO (100 M) as indicated. Samples were incubated for 10 min at room temperature prior to adding GTP (1 mM). Reactions proceeded for 5 min at 37°C and were quenched with 1.5% glacial acetic acid. The precipitated protein was removed by centrifugation at 16,000 ϫ g for 10 min at 4°C, and the supernatant was diluted at 1:100 in 50 mM sodium phosphate, pH 7.0, 0.2% BSA, and 0.2% NaN 3 . cGMP was quantified using a commercially available homogenous time-resolved fluorescence (HTRF) assay according to the manufacturer's instructions (CisBio). All samples were measured in duplicate using a Synergy H1 fluorescent plate reader (BioTek). Homogenous time-resolved fluorescence measurements were analyzed according to the manufacturer's instructions using SigmaPlot (Systat Software, Inc.).

Photoaffinity labeling sGC and bacterial H-NOX with IWP-854
The

Detecting photoaffinity labeling by Western blot analysis
Samples were suspended in 1ϫ SDS loading buffer, and 1 g of protein was run on a 15% bisacrylamide gel for 90 min at 96 V. Protein was transferred to a nitrocellulose membrane at 100 V for 1 h at 4°C. The membrane was blocked for 1 h in 5% BSA in PBS-T (0.1% Tween-20) and incubated in a 1:1000 dilution of primary antibody (Cell Signaling Technology catalog no. 5597 for biotin, Abcam ab76949 for Strep-tag II, and QED Biosciences 18814-01 for His 6 ) overnight at 4°C. The membrane was washed three times in PBS-T and incubated in a 1:10000 dilution of secondary antibody (Li-Cor, catalog no. 925-68071) for 2 h with shaking at room temperature. Membranes were washed an additional three times and imaged using the Odyssey infrared imaging system.
Western blot analyses involving Hs sGC were performed in a similar manner, with the following exceptions. A total of 200 ng of protein were run on NuPAGE 4 -12% bis-Tris gel (Invitrogen). The biotin affinity tag was detected with a 1:2000 dilution Stimulator binding to the sGC ␤1 H-NOX domain of IRdye800 streptavidin (Li-Cor, catalog no. C40403-1). A 1:1000 dilution of primary antibody (Abcam ab154841 for the Hs sGC ␤1 subunit) and a 1:15000 dilution of secondary antibody (Li-Cor 926-68071) were used as a loading control.

Preparing samples for mass spectrometry
Photoaffinity-labeled samples were buffer-exchanged into 100 mM ammonium bicarbonate, pH 8.0, using 10-kDa centrifugal filters (Amicon). Samples were reduced with 12 mM dithiothreitol for 45 min at 56°C, alkylated with 20 mM iodoacetic acid for 30 min at room temperature in the dark, and digested with a 1:30 (protease:protein) ratio of trypsin or chymotrypsin overnight at 37°C or 30°C, respectively. Samples digested with chymotrypsin contained 1 mM CaCl 2 . The digested peptides were cleaned using C18 ZipTips (Pierce/Thermo Fisher Scientific), dried by SpeedVac, and stored at Ϫ20°C.
Data-dependent scanning was performed with Xcalibur version 2.1.0 software using a survey mass scan at 60,000 resolution in the Orbitrap analyzer scanning mass/charge (m/z) range of 350 to 1600 followed by collision-induced dissociation MS/MS of the six most intense ions in the Orbitrap analyzer at 7,500 resolution. Precursor ions were selected by the monoisotopic precursor selection setting, with the instrument set to observe fragment ions once and then excluded from analysis for 45 s, allowing for interrogation of lower abundance ions. Ions were excluded with a Ϯ10 ppm window. Ionization of IWP-854 increases the charge state of the peptide, allowing for increased selection of labeled peptides by excluding precursor ions with a charge less than ϩ3.
Tandem mass spectra were searched against a protein database containing 5200 entries, including sequences for all sGC and H-NOX constructs analyzed, the proteome for E. coli BL21, and common contaminants. Searches were performed using Thermo Proteome Discoverer 1.3, version 1.3.0.339 (Thermo Fisher Scientific), considering the tryptic peptides with up to two missed cleavage sites. Iodacetamide derivatives of cysteines and oxidation of methionines were specified as variable modifications. Modification by IWP-854 (1422.717 Da) was searched against all 20 amino acid residues in an iterative fashion. The presence of IWP-854 was confirmed by a peak corresponding to the mass of the precursor ion minus m/z 270.127 Da (Fig. 3A). All labeled peptides were initially identified automatically using Discoverer software. Following the initial identification, manual assignment of peptides was performed as necessary based on precursor mass, MS/MS spectra, and column retention time.

NMR sample preparation
The diamagnetic Sw H-NOX Fe(II)-CO complex was used in backbone assignment and chemical shift perturbation experiments. To prepare the sample, the isotopically labeled Sw H-NOX was treated with 10 mM dithionite in an extensively degassed buffer containing 50 mM Tris-HCl, pH 8.0, and 50 mM NaCl and then saturated with CO in a sealed tube. Dithionite was removed by buffer exchange to an extensively degassed, then CO-saturated buffer containing 50 mM sodium/potassium phosphate, pH 7.5, 50 mM NaCl, and 10% D 2 O for NMR measurement. NMR experiments were performed in a CO-saturated and sealed 5.0-mm NMR tube. The concentration of NMR samples was ϳ0.8 mM for backbone assignment. An 15 N-HSQC spectrum was collected before and after each experiment to evaluate the stability of the protein. IWP-051 and PDE9A inhibitor were prepared by dissolving the compounds into DMSO-d 6 (99.9 atom % D, Aldrich) to a final concentration of 25 mM. 2,2-Dimethyl-2-silapentane-5-sulfonate was included as an internal reference in the TrNOESY sample and used to calibrate the concentration of IWP-051 compound.

NMR spectroscopy and data analysis
NMR experiments for protein backbone resonance assignment were carried out on Agilent 18.8 T NMR spectrometer equipped with a triple resonance cryogenic probe at the National Magnetic Resonance Facility at Madison. NMR experiments for chemical shift perturbation (CSP) and TrNOESY were carried out on Agilent 14.0 T spectrometer equipped with a triple resonance cryogenic probe at the University of Arizona. All experiments except for TrNOESY were performed at 20°C. TrNOESY experiments were performed at 25°C. NMR data were processed using NMRPipe (37) and in combination with MddNMR (38,39) or SMILE (40) for non-uniformly sampled data. NMRFAM-SPARKY (41), PINE (42), and PINE-SPARKY (43) were used for resonance assignment and spectrum analysis.

Protein backbone resonance assignment
All backbone and side-chain resonance experiments were acquired using 40-50% non-uniform sampling (38)(39)(40). Sequencespecific backbone resonance assignments were determined using the following triple resonance experiments: 3D-HNCA (44) (46) were used to verify the backbone resonance assignments. More than 90% of the residues were assigned.

Stimulator binding to the sGC ␤1 H-NOX domain
in which ⌬␦H is the change in 1 H shifts and ⌬␦N is the change in 15 N shifts in ppm. A scaling factor of 0.2 was applied to the 15 N shifts. Control experiments titrating only DMSO-d6 into the protein were used to correct the solvent perturbations from DMSO-d6. For residues with significant perturbations, their CSPs were plotted as a function of ligand concentration and fitted into the following equation (48), in which ⌬␦ is the adjusted chemical shift change, ⌬␦max is the maximum adjusted chemical shift change at saturation, [P T ] is total protein concentration, [L] is the ligand concentration, and K d is the equilibrium dissociation constant. K d was extracted by fitting the data using xcrvfit.

TrNOESY measurements
The hydrogen chemical shifts of IWP-051 were assigned as described previously (11) and confirmed by 2D-COSY, 2D-TOCSY, and 2D-NOESY experiments for samples dissolved in D 2 O. TrNOESY experiments were performed with IWP-051 or PDE9A inhibitor dissolved in 99.9% D 2 O, 50 mM sodium phosphate, pH 7.5, 100 mM NaCl, 4% DMSO-d6, 2 mM dithionite, and 100 M 2,2-dimethyl-2-silapentane-5-sulfonate. Protein/ligand molar ratios of 1/50 for Ms sGC-NT23, 1/29 for Cb SONO, and 1/19 for Sw H-NOX were used with IWP-051, and a ratio of 1/31 for Ms sGC-NT23 was used with PDE9A inhibitor. For Cb SONO, a pH of 8.0 was used to increase the stability of the protein. A conventional NOESY sequence was used with 256 ϫ 2048 data matrices and with presaturation for water suppression. TrNOESY-derived distances were calibrated using the distance between H9 and H10 from the modeled IWP-051 structure. The NOE contribution from H (11) was scaled by a factor of 2 prior to distance calculation to account for the two equivalent hydrogens on the methylene group.

Molecular modeling
A molecular model for Ms sGC-NT13 was assembled previously using SAXS, chemical cross-linking, and domain homology modeling (19). Models for compounds IWP-051 and IWP-854 were prepared by first generating a SMILES string in ChemDraw (PerkinElmer Informatics, Inc.) and submitting the string to the Grade Web Server (http://grade.globalphasing. org/cgi-bin/grade/server.cgi), 3 which generates energy-minimized coordinates and refinement parameters using known structures in the Cambridge Structural Database. Modeling of compound binding was done manually in COOT (49) followed by energy minimization in REFMAC5 (50) as encoded in CCP4i (51). Structure figures were prepared using the PyMOL Molecular Graphics System, version 1.8.6.0, Schrödinger, LLC.

Data availability
All data that support the findings of this study are in the published article (and its supporting data files) or are available from the corresponding author upon reasonable request.