An N-terminally truncated form of cyclic GMP–dependent protein kinase Iα (PKG Iα) is monomeric and autoinhibited and provides a model for activation

The type I cGMP-dependent protein kinases (PKG I) serve essential physiological functions, including smooth muscle relaxation, cardiac remodeling, and platelet aggregation. These enzymes form homodimers through their N-terminal dimerization domains, a feature implicated in regulating their cooperative activation. Previous investigations into the activation mechanisms of PKG I isoforms have been largely influenced by structures of the cAMP-dependent protein kinase (PKA). Here, we examined PKG Iα activation by cGMP and cAMP by engineering a monomeric form that lacks N-terminal residues 1–53 (Δ53). We found that the construct exists as a monomer as assessed by whole-protein MS, size-exclusion chromatography, and small-angle X-ray scattering (SAXS). Reconstruction of the SAXS 3D envelope indicates that Δ53 has a similar shape to the heterodimeric RIα–C complex of PKA. Moreover, we found that the Δ53 construct is autoinhibited in its cGMP-free state and can bind to and be activated by cGMP in a manner similar to full-length PKG Iα as assessed by surface plasmon resonance (SPR) spectroscopy. However, we found that the Δ53 variant does not exhibit cooperative activation, and its cyclic nucleotide selectivity is diminished. These findings support a model in which, despite structural similarities, PKG Iα activation is distinct from that of PKA, and its cooperativity is driven by in trans interactions between protomers.

The type I cGMP-dependent protein kinases (PKG I) serve essential physiological functions, including smooth muscle relaxation, cardiac remodeling, and platelet aggregation. These enzymes form homodimers through their N-terminal dimerization domains, a feature implicated in regulating their cooperative activation. Previous investigations into the activation mechanisms of PKG I isoforms have been largely influenced by structures of the cAMP-dependent protein kinase (PKA). Here, we examined PKG I␣ activation by cGMP and cAMP by engineering a monomeric form that lacks N-terminal residues 1-53 (⌬53). We found that the construct exists as a monomer as assessed by whole-protein MS, size-exclusion chromatography, and small-angle X-ray scattering (SAXS). Reconstruction of the SAXS 3D envelope indicates that ⌬53 has a similar shape to the heterodimeric RI␣-C complex of PKA. Moreover, we found that the ⌬53 construct is autoinhibited in its cGMP-free state and can bind to and be activated by cGMP in a manner similar to full-length PKG I␣ as assessed by surface plasmon resonance (SPR) spectroscopy. However, we found that the ⌬53 variant does not exhibit cooperative activation, and its cyclic nucleotide selectivity is diminished. These findings support a model in which, despite structural similarities, PKG I␣ activation is distinct from that of PKA, and its cooperativity is driven by in trans interactions between protomers.
PKG I is expressed as two splice variants that form homodimers (␣/␤) (2,12,13). In both isoforms, each protomer is composed of a dimerization domain formed by a leucine zipper (LZ) followed by an autoinhibitory domain segment, a regulatory domain containing two cGMP-binding sites (A and B), and a catalytic domain (see Fig. 1A) (14). PKG I␣ and I␤ have low sequence conservation in their dimerization and autoinhibitory domains, and this difference is thought to both control the activity and response of the kinase to cGMP and mediate its targeting to specific substrates (6,(15)(16)(17)(18)(19)(20). How the regulatory and catalytic domains interact in the inactive state and communicate to control cooperative activation have remained unsettled questions for the field (21,22). In the inactive state, it has been hypothesized that the autoinhibitory domain of PKG I occupies the catalytic cleft formed between the N-and C-terminal lobes and acts as a pseudosubstrate in a manner similar to the cAMP-dependent protein kinase (PKA) (23).
PKG isoforms share sequence homology with PKA (28 and 41% identity in its regulatory domain and catalytic domain, respectively). Consequently, current models of the relationship between PKG I structure and function have been overwhelmingly influenced by studies of PKA. However, PKA is expressed as separate catalytic and regulatory domains and PKG exists as a single polypeptide chain (24 -26). Structural and biochemical studies of PKA have demonstrated an in cis mechanism for PKA activation (27)(28)(29). The sequential binding of cAMP to two coupled binding sites (A and B) within the regulatory subunit constitutes the physical basis for cooperativity in both cAMP binding and kinase activation. Although it is known that fulllength PKG I dimers also exhibit cooperative activation, it is unknown whether cooperativity is driven by the regulatory domain (in a fashion similar to PKA) or whether the dimeric state influences this effect (15,25,30).
It has been suggested that the N-terminal domains in the hinge region between the leucine zipper and autoinhibitory domains in PKG I are highly flexible, and structural models of PKG I have depicted the N terminus protruding from the holoenzyme (regulatory and catalytic domain complex) (14,23,31). Although it has been widely suggested that dimerization is essential for catalytic function, it has never been experimentally tested in the I␣ variant. Informed by previous models, we hypothesized that PKG I␣ could be truncated to generate a functional monomer by removing the flexible N terminus (zipper and hinge). Moreover, we surmised that this construct would retain cGMP-dependent activation due to the inclusion of the autoinhibitory domain. In this study, we provide the biochemical and biophysical characterization of PKG I␣ ⌬53 (subsequently referred to as ⌬53) by demonstrating its monomeric architecture, phosphorylation state, cGMP binding properties, and the kinetic characteristics of its cGMP-dependent activation. Using the ⌬53 construct, we address the following: 1) whether PKG I␣ is inhibited in cis or in trans, 2) how cyclic nucleotide binding and selectivity of the A-site is linked to activation, and 3) a putative mechanism by which cGMP-mediated cooperativity is derived in PKG. Our data indicate that PKG I␣ can form an in cis autoinhibited complex. Furthermore, cooperative activation of PKG I␣ by cGMP, in contrast to PKA, relies on interactions between the homodimeric regulatory domains. We conclude that cooperativity is driven in part by the N-terminal dimerization domain by localizing PKG monomers within close proximity during cGMP binding to form in trans interprotomer interactions.

Design, expression, and purification of ⌬53
The design of a monomeric form of PKG I␣ was accomplished using a bioinformatics approach by comparing with a previously solved structure of the homolog, PKA, in a heterodimeric complex with its regulatory domain (Fig. 1, A and B). The complex RI␣-C (Protein Data Bank code 2QCS) was chosen because of the high sequence conservation of its autoinhibitory segment with PKG I␣ and the availability of its three-dimensional coordinates (Fig. 1C). A multiple sequence alignment of  (25); N/C-lobe, catalytic domain. B, a multiple sequence alignment shaded by BLOSUM62 score is shown for PKG I␣ relative to PKA regulatory domain isoforms. The autoinhibitory sequences of PKG I (red) and PKA-RI␣ (cyan) are outlined. C, the crystal structure of the RI␣-C heterodimer (Protein Data Bank code 2QCS) detailing the contacts between the autoinhibitory domain segment beginning at Lys 92 (cyan) and the catalytic domain (gray).

⌬53 PKG I␣
the PKG I␣ autoinhibitory fragment with that of known isoforms of PKA was performed. The distance from the first residue (Lys 92 ) to the start of the autoinhibitory sequence in the RI␣-C structure was used to determine that the residue equidistant from the autoinhibitory sequence in PKG I␣ was Pro 56 (Fig. 1B). We hypothesized that, in a similar fashion to RI␣-C, the autoinhibitory segment preceding the canonical (A/G)ISAE pseudosubstrate segment would confine the P ϩ 1 loop of the catalytic domain and make further contacts with the C-terminal tail, glycine-rich loop, and the F-helix from the large lobe (32,33). As a consequence and so as not to exclude residues that may be important for autoinhibition, ⌬53 was engineered to begin at Ile 54 .
Following expression and purification of ⌬53 ( Fig. 2A), we examined the kinase by size-exclusion chromatography (Fig.  2B). Both the apo and cGMP-bound forms exhibited reproducible elution profiles centered at 14.8 and 14.7 ml, respectively, indicative of a 70-kDa species according to molecular mass standards (Fig. S1). The addition of cGMP appeared to have a negligible effect on the elution volume of the protein, suggesting that dimer formation was not present in the apo or cGMPsaturated samples. ⌬53 was further examined by time-of-flight (TOF) MS to determine its mass and phosphorylation state (Fig. 2C). A mass of 70,511 Da, corresponding to a monophosphorylated, monomeric species, was observed. We also discovered the presence of a doubly phosphorylated species compris-ing 20% of the total ion content (70,591 Da). Tandem MS analyses of trypsin-digested samples were used to confirm the primary phosphorylation site at Thr 517 (full-length enzyme numbering scheme) located in the activation loop of the catalytic domain (Table 1). This phosphorylation site has been previously identified in full-length PKG I␣ and is required for catalytic activity (34). To confirm this, we also examined preparations of full-length PKG I␣ and found a single phosphorylation site corresponding to Thr 517 (Table S1). Based on previous studies, the minor phosphorylation site observed for ⌬53 was suspected to reside in the autoinhibitory domain (35,36). However, the peaks corresponding to the peptide containing the autophosphorylated pseudosubstrate sequence, R2AQGISA-EPQTYR2S (residues 61-72), were of high intensity and suggested no phosphorylations were present.

Small angle X-ray scattering (SAXS) analysis
The results presented thus far indicate that ⌬53 exists as a monomer in solution. To characterize the low-resolution solution structure of the autoinhibited complex, we utilized sizeexclusion chromatography (SEC) coupled to SAXS wherein during analysis a single species associated with ⌬53 was observed ( Table 2). Concurrent Guinier calculation during data collection indicated a constant R g across the sample peak (Fig.  3A). The averaged scattering intensity curve (I q versus q) for ⌬53, which was obtained by averaging frames 410 -459 of the SEC-SAXS peak, exhibited a linear Guinier plot (Fig. 3, B and inset, and Table 2). The Guinier plot indicated the presence of a monodisperse system with no evidence of aggregation. A Kratky plot of the data suggested the presence of a well-folded species (Fig. 3C). The P(r) curve calculated from the scattering intensity contained a single peak at ϳ30 Å that smoothly decayed to a maximum linear dimension (D max ) of 97 Å (Fig.  S2). When compared with the P(r) curve of the homologous PKA heterodimer upon which we based our construct, RI␣-C (green), both showed a single peak at 34 Å, but the RI␣-C heterodimer had an extended D max of ϳ130 Å. Comparison of the ⌬53 P(r) with that observed for a different PKA heterodimer, RII␤-C (red), demonstrated that they have nearly identical dimensions (Table S2) (37).
Predicted X-ray scattering from the crystal structures of the PKA RI␣-C, PKA RII␤-C, and PfPKG from Plasmodium falciparum (a monomeric species containing four cGMP-binding sites and a catalytic domain) was conducted using both CRYSOL and FoXS (Table 2f) (38,39). The best fit of the experimental SAXS data to the predicted scattering was obtained from RI␣-C using CRYSOL with constant subtraction ( 2 ϭ 0.82; Fig. 3D and Table 2). Using the RII␤-C heterodimer in the same analysis resulted in a poorer fit to the experimental data ( 2 ϭ 1.25). The PfPKG fits were poor using CRYSOL ( 2 ϭ 9.21) but acceptable with the FoXS server ( 2 ϭ 1.29; Table 2f and Fig. S3). We also found that the R g values calculated using the predicted scattering curves for both the PKA RI␣-C and RII␤-C heterodimer crystal structures were in closer agreement with the measured and calculated values for ⌬53 (Table  2f). Next, DAMMIF and DAMMIN modeling was implemented. Using SUPCOMB, the resulting averaged and filtered DAMMIF/DAMMIN ab initio three-dimensional envelope A, Coomassie-stained 12% SDS-PAGE of PKG I␣ full length and ⌬53 under denaturing and reducing conditions. B, size-exclusion chromatography traces of ⌬53 in its apo (black) and cGMP-saturated (red) forms. C, a raw m/z trace (inset) and the deconvoluted TOF-MS displaying the two predominant masses observed for ⌬53 corresponding to singly and doubly phosphorylated species. The mass correlated to the location of an unphosphorylated monomer is also denoted. Abs, absorbance; mAU, milli-absorbance units.

⌬53 PKG I␣
accommodated both crystal structure models of the RI␣-C and RII␤-C heterodimers with little protrusion outside of the envelope (Table 2e and Fig. 3, E and F). Among the 3D models examined, these data cumulatively suggest that ⌬53 adopts a shape most similar to PKA RI␣-C.

Binding and activation of full-length PKG I␣ and ⌬53 with cNMP
Finally, we studied the activation kinetics of ⌬53. No significant differences in basal activity were observed between PKG I␣ full length and ⌬53, indicating that the monomeric construct was autoinhibited at levels comparable with previous reports (15,30). Moreover, the cGMP-dependent activation of ⌬53 and full-length PKG I␣ were analogous with respect to their maximum velocities of 4.0 and 3.9 mol/min ϫ mg, respectively (Fig. 4B, panel a). These results indicate that the cGMP-dependent -fold stimulation of ⌬53 is indistinguishable from full length. However, differences between the two constructs were observed with respect to their activation constants and degree of cooperativity. We observed K a values and Hill coefficients corresponding to 182 nM (n H ϭ 1.6) for PKG I␣ full length and 250 nM (n H ϭ 1.0) for ⌬53 (p Ͻ 0.0001 for both measurements). When we tested the activation profiles of full-length and ⌬53 PKG I␣ with cAMP, we observed that the full-length enzyme exhibited cooperative activation, whereas again ⌬53 was noncooperative. Comparisons of the -fold difference in activation by cGMP and cAMP for both enzymes demonstrated that the removal of the N terminus of PKG I␣ reduced the selectivity for cyclic nucleotide from 58-(full length) to 3.5-fold (⌬53), representing an overall 16-fold decrease in selectivity for cGMP over cAMP.
To probe how binding of cyclic nucleotides correlates with the activities of the enzymes, we measured cGMP and cAMP binding to PKG I␣ constructs by surface plasmon resonance (SPR) spectroscopy. Binding of cyclic nucleotides was fit using a two-site binding model, because the regulatory domain of PKG I␣ contains two cGMP sites per monomer. To determine cooperativity of the binding, we also fit the data using a one-site model with Hill coefficient. PKG I␣ full length exhibited an almost 3-fold weaker affinity for cGMP than ⌬53 (K D(FL) ϭ 7.9 M and K D(⌬53) ϭ 2.9 M) when fit with a one-site model (Table  3). However, when fit with a two-site binding model, the ratio dropped to 1.6-fold. Moreover, both constructs displayed negative cooperativity in binding cGMP (n H(FL) ϭ 0.60 and n H(⌬53) ϭ 0.74). When cAMP binding was measured for the two constructs, the K D values for their high-affinity sites increased by over 30-fold. However, cAMP binding remained negatively cooperative in both enzymes. In binding to either cyclic nucleotide, we found that the low-affinity site demonstrated much weaker binding than the high-affinity site.
To distinguish the cyclic nucleotide sites responsible for activation in the full-length PKG I␣ kinase, we mutated the glutamic acid residues in the Phe-Gly-Glu (FGE) motif of the phosphate-binding cassettes within the A-site (E168G) and the B-site (E292A). Because the glutamate residue is involved in binding to the 2Ј-hydroxyl from the ribose, mutation of this site in PKA has been shown to disrupt cyclic nucleotide binding (28, 40 -43). Mutation of the A-site (PKG I␣(E168G)) completely abolished cGMP-dependent activation of the full-length kinase (Fig. S4). These results demonstrate that a functional A-site is necessary for kinase activity. Moreover, when we examined the cGMP-dependent activation of PKG I␣(E292A), we observed cooperative (n H ϭ 1.45) stimulation of the kinase with a K a of 95 nM.
In an effort to further examine the contribution of the N terminus and the catalytic domain on cyclic nucleotide binding selectivity, we purified two constructs of the regulatory domain of PKG I␣. The first construct, PKG I␣(1-326), contains the N terminus and cGMP-binding regulatory domain, whereas the second construct, PKG I(78 -326), contains only the regulatory domain. When we examined binding of cGMP to PKG I␣(1-326) and PKG I(78 -326), we found that the one-site model K D shifted significantly compared with values observed for fulllength PKG I␣ and ⌬53 (Fig. S5). For PKG I␣(1-326), the twosite binding model showed that the K D value for cGMP binding to the A-site increased by 8-fold compared with the full-length enzyme (Table 3). Furthermore, binding of cAMP showed a drastic decrease in selectivity for cGMP over cAMP. The B-site showed a clear selectivity for cGMP over cAMP (4-fold). However, when the N terminus was excluded in PKG I(78 -326), we observed no difference in the affinities for cGMP and cAMP in the high-affinity A-site. The B-site retained a clear selectivity for cGMP over cAMP (10-fold).

Discussion
PKG activation has been extensively studied by traditional biochemical methods and, more recently, using modern biophysical approaches (18,24,25,44,45). These investigations sought to determine the architecture of the homodimeric kinase in its basal and activated states, the order and selectivity of cGMP binding, and, by extension, the origin of its cooperativity. The generation of a functional, monomeric form of PKG I␣ is a useful tool for investigating the molecular basis of these long-observed biochemical phenotypes.
PKG I␣ ⌬53 purified from Sf9 cell extracts exists as a mixture of mono-and diphosphorylated forms; no unphosphorylated form was detected. The primary phosphorylation site, Thr 517 , was found in high abundance by TOF-MS ( Fig. 2C and Table 1). Phosphorylation of Thr 517 , which is located in the activation loop of the catalytic domain, is essential for catalytic activity   ⌬53 PKG I␣ (34). In the PKA catalytic domain, the analogous phosphorylation at Thr 197 forms contacts with other activation loop residues, the catalytic loop, and the ␣ C-helix to integrate the active site components (46,47). It has been hypothesized that phosphorylated Thr 517 in PKG serves a similar function because the lack of phosphorylation or mutation of this residue renders PKG inactive (34). The presence of this phosphorylation site in ⌬53 corroborates the cGMP-dependent activation observed by our phosphotransferase assays. In regard to the second, less abundant phosphorylation site, we had hypothesized that this residue would be located in the autoinhibitory domain based on previous published data (35,36,48). Analysis of autoinhibitory domain residues suggested that Ser 64 and Thr 70 were not phosphorylated (parent peptide Ala 61 -Arg 72 ; Table 1). Another potential site, Thr 58 , has previously been identified as the major site that is autophosphorylated most rapidly in the presence of cGMP in vitro or cAMP in native PKG I preparations isolated from bovine lung (35,48). Because Sf9 cells endogenously express adenylyl cyclases, we hypothesized that Thr 58 could be the minor phosphorylation site (49,50). However, our LC-MS/MS data could not confirm the precise location. These results are in agreement with previously published results that observed the phosphorylation at Thr 517 , but were also unsuccessful in identifying the second site (23). Furthermore, these results suggest that neither its dimerization through the N-terminal dimerization domain nor exposure to cGMP are required for activation loop phosphorylation.
Despite the presence of these two phosphorylation states, we consistently observed a compact, monodisperse, monomeric species in solution by SEC (Fig. 2B). In addition, the elution profile of ⌬53 did not change with the addition of cGMP, suggesting that it maintains its monomeric state even during longlived exposure. To reconstruct the three-dimensional shape of inactive ⌬53 and confirm its monomeric character, we collected data by SEC-SAXS and compared the reconstructed 3D envelope with the crystal structures and SAXS data from two PKA heterodimers (Protein Data Bank codes 2QCS and 4WBB) and PfPKG (Protein Data Bank code 5DYK). We found that ⌬53 adopts a similar shape to that of the PKA RI␣-C heterodimer (Protein Data Bank code 2QCS) based on the best fit of the atomistic modeling in CRYSOL to the scattering data ( 2 0.82) ( Fig. 3D and Table 2e). In addition, the crystal structure is readily accommodated by the 3D envelope calculated from the ⌬53 SAXS data (NSD ϭ 1.06; Fig. 4F). These data concurred with previously published results for the apo state of monomeric PKG I␤ ⌬1-52 (Table S2) (45). PKG I␤ ⌬1-52 contains 18 additional residues upstream of its autoinhibitory domain compared with PKG I␣ ⌬53. Considering the high overlap between these two truncated constructs in the SAXS analyses, this suggests that the linker in I␤ ⌬1-52 adopts a compact conformation relative to the regulatory-catalytic domain complex. Based upon these data, we propose a model wherein each monomer within the dimeric PKG I␣ complex can be autoinhibited by its own regulatory domain and autoinhibitory sequence (Fig.  5, top).
Finally, we report that ⌬53 is activated by cGMP at concentrations similar to the full-length kinase (Fig. 4B, panel a). A previous study by Richie-Jannetta et al. (51) and a more recent  These results support the longstanding conclusion that PKG I␣ and -␤ are enzymatically distinct due to differences in their N termini (1,15,20,30,50). Interestingly, ⌬53 does not exhibit cooperative cGMP-dependent activation (Fig. 4B, panel a). The loss of cooperativity has been previously reported for PKG I␤ wherein full-length PKG I␤ (n H ϭ 2.1) becomes noncooperative when the N terminus is removed (⌬1-52 PKG I␤) (30,52). In addition, we observed that cAMP also cooperatively activates the full-length kinase

⌬53 PKG I␣
(n H ϭ 1.75) but not ⌬53 (n H ϭ 0.81). These results suggest that the cooperative activation is mediated by the N terminus through the facilitation of homodimer formation. The N-terminal dimerization domains of PKG I␣ and I␤ are left-handed, coiled-coil, LZ motifs. The LZ is characterized by a heptad repeat of amino acids (abcdefg) where residues a and d are typically hydrophobic and residues e and g are typically hydrophilic. The leucine zipper of PKG I␣ spans residues 1-47. In addition to its leucine and isoleucine residues, there are also four non-leucine/isoleucine residues at the d positions that provide additional nonhydrophobic interhelical contacts (Phe 7 , Lys 14 , Lys 28 , and Cys 42 ) (53). One of these residues, Cys 42 , is unique to the I␣ isoform and forms an interprotomer disulfide bond in the presence of oxidizing agents such as hydrogen peroxide (54,55). The leucine zipper of PKG I␤ spans residues 4 -53 and also contains four non-leucine/isoleucine residues at the d positions (Lys 13 , Arg 20 , Lys 34 , and Tyr 48 ) (18). Although the majority of the noncanonical residues are basic, their differences in position within the helices and the overall discrepancy in length of the helices are reasons why PKG I protomers have not been shown to form heterodimers either in vitro or in vivo.

⌬53 PKG I␣
It is well-characterized that the cGMP A-site in the regulatory domain is the high-affinity binding site and thus is the first to bind cyclic nucleotide (24,26,44). Our data of PKG I␣ full length and ⌬53 agree with these previously published results and indicate a clear selectivity for cGMP over cAMP in the A-site (Table 3). To further test the contributions of the A-and B-sites in PKG I␣ activity, we measured the activities of PKG I␣(E167G) and PKG I␣(E292A). We found that disruption of cyclic nucleotide binding to the B-site did not have an appreciable effect on activity (Fig. S4). In contrast, disruption of cyclic nucleotide binding to the A-site completely abolished activity. These data collectively support the importance of the A-site in regulating activation of PKG I␣. However, previous studies using isolated regulatory domain constructs have suggested that the A-site is nonselective for cyclic nucleotide (44,25). These results suggest that selectivity for cGMP binding to the A-site must be controlled through regions outside of the regulatory domain.
To investigate the influence of regions outside of the regulatory domain on the cyclic nucleotide-dependent characteristics of PKG I␣, we expressed and purified two additional constructs, PKG I␣(1-326) and PKG I␣(78 -326). PKG I␣ without its catalytic domain (PKG I␣(1-326)) showed a significant decrease in binding affinity for cGMP and loss of selectivity for cGMP over cAMP in the A-site compared with the full-length construct (Table 3). Selectivity was only slightly reduced in the B-site, and the values agree with those measured previously for the isolated B-domain (44). These results differ from the binding and nucleotide selectivity results observed for PKG I␣ full length and ⌬53. Thus, these characteristics seem to be integrated into the architecture of the inactive holoenzyme.
The initial comparison of the activation curves for full-length and ⌬53 with the one-site binding model indicates a large disparity between the concentration of cGMP required for full saturation of the A-and B-sites and the minimum concentration required for full activation. In early measurements of cyclic nucleotide binding with [ 3 H]cGMP, a significant difference in the K D values for cGMP binding at 4 and 30°C was observed (56). Our measurements for cGMP are consistent with these early experiments. When a two-site model is applied to the binding data, we can directly correlate the K D for the A-site with the V max for activation. This model is reasonable because there are two cGMP-binding sites with distinct binding affinities. Furthermore, this effect is observed for both cGMP and cAMP equally (Fig. 4). This implies that full-length PKG I␣ only requires that half of the A-sites are occupied to stimulate full activation. These data, which also suggest that the B-site is nonessential for kinase activity, are further supported by the fact that the K D for the B-site (ϳ90 M) is higher than estimated intracellular concentrations of cGMP (ϳ10 nM-10 M) (57,58,59). Moreover, we observed that mutation of the nucleotidebinding cassette in the B-site does not appreciably affect activation of the full-length kinase (Fig. S4). Thus, we can also conclude that binding studies that are either directed toward the isolated PKG regulatory domains or use equilibrium exchange of radiolabeled cyclic nucleotides are limited, and future studies should examine these effects in the context of the intact holoenzymes.
These SPR measurements also indicate that binding of cGMP to both full length and ⌬53 is negatively cooperative  The model depicted hypothesizes that one cyclic nucleotide bound to the high-affinity A-site is sufficient to activate both the full-length and ⌬53 PKG I␣ constructs. Cooperativity arises from the in trans interaction of protomers at the knobnest site and not from cooperative cyclic nucleotide binding. In the ⌬53 construct, cooperativity is not observed and cannot be facilitated due to the lack of dimerization. The organization of the full-length enzyme must also lend itself to generate a cGMP selectivity filter for the A-site based upon our fulllength and truncated PKG I␣ data. CAT, catalytic domain.

⌬53 PKG I␣
(n H Ͻ 1), which indicates that binding of cGMP to subsequent sites becomes more difficult with increasing concentrations. These results suggest a lack of avidity between cyclic nucleotide-binding domains A and B, and this has been corroborated by recent structural evidence (52). These observations stand in contrast to PKA where the origins of cooperativity have been directly linked to the regulatory domain wherein binding of cAMP into the B-site enhances binding for the A-site through intrasubunit contacts (27,28,32). It has been suggested that the cooperativity observed for PKG activation arises from a similar mechanism (23,24,44). However, the loss of cooperative activation in PKG I␣ through disruption of the dimer via two mechanisms, 1) by removing residues that allow for N-terminal dimerization (this study) and 2) via mutation of residues forming the knob-nest interface, suggests that this is not the case (25). Finally, we propose a model based on our findings wherein autoinhibition of PKG I␣ is driven in cis. The high-affinity A-site provides cyclic nucleotide selectivity through contacts outside of the regulatory domain, and cooperative activation of PKG I␣ is driven in trans, facilitated by the N terminus, which ensures monomers are close enough to form necessary interprotomer contacts (Fig. 5). These data reinforce our earlier hypothesis that cooperative activation of the kinase is not driven through cooperative binding of cyclic nucleotides to the regulatory domain within the same protomer. Rather, it is driven through an inducible dimeric interface mediated by the switch helix (30,60). Therefore, we conclude that the mechanism driving cooperativity in PKG I␣ lies in the interface between protomers and not between cGMP-binding sites within the same protomer.

Sequence alignments
Sequences for PKG I␣ and PKA regulatory domain isoforms were downloaded from the NCBI and UniProt repositories and

Maintenance of Sf9 cells
Sf9 cells were purchased from Life Technologies. Frozen cell stocks were prepared according to the Growth and Maintenance of Insect Cell Lines Manual (Life Technologies). Typically, one vial containing Sf9 cells (1 ϫ 10 6 cells/vial in freezing medium (80% supplemented Grace's insect medium (Gibco), 10% fetal bovine serum (FBS; Sigma), and 10% DMSO) was suspended in 15 ml of Grace's medium (Life Technologies) supplemented with 5% fetal bovine serum (Sigma) and 10 g/ml gentamicin (Sigma). Cells were grown in a 75-cm 2 flask until cells reached 80% confluence. Cells were detached from the flask and suspended in Sf900 III medium (Life Technologies) supplemented with 1ϫ lipids (Sigma), 1% poloxamer 81 (Sigma), and 10 g/ml gentamicin (Sigma). Sf9 cells were maintained in suspension at 27°C at 80 rpm and passaged every 3.5 days to 1.2 ϫ 10 6 cells/ml.

Expression of PKG I regulatory domain constructs in E. coli
The PKG I regulatory domain constructs (PKG I(78 -326) and PKG I␣(1-326)) were amplified by PCR from the fulllength PKG I␣ gene using forward and reverse primers containing BamHI-and EcoRI-cut sites, respectively: PKG I-78-f, 5Ј-CGG GAT CCA TGC AGG CAT TCC GGA AGT TC-3Ј (sense); PKG I-1-f, 5Ј-TGG GGA TCC AGC GAG CTG GAG-3Ј (sense); PKG I-326-r, 5Ј-TCG AAT TCC ATG CTA TTA TAA TCC TCC AAT CAA ATG-3Ј (antisense). The respective fragments were ligated into BamHI/EcoRI-digested pRSET-A using T4 ligase (New England Biolabs). Both constructs were expressed in Rosetta2 E. coli (Novagen). Overnight cultures were used to inoculate 1 liter of terrific broth supplemented with 50 g/ml ampicillin and 25 g/ml chloramphenicol and subsequently grown at 37°C at 220 rpm to an A 600 of 0.8. Expression was induced by the addition of isopropyl 1-thio-␤-D-galactopyranoside (1 mM final), and the induced cultures ⌬53 PKG I␣ were incubated overnight at 25°C at 220 rpm. E. coli were harvested by centrifugation at 1500 ϫ g for 30 min, and the resulting bacterial pellets were flash frozen in liquid nitrogen and stored at Ϫ80°C.

Purification of PKG I␣ constructs
Protease inhibitors (Roche Applied Science) were dissolved in 50 ml of lysis buffer and added to 1 liter of pelleted Sf9 cells containing either PKG I␣ full length, E167G, E292A, or ⌬53 (B. taurus). Sf9 cell pellets were thawed on ice for 40 min followed by gentle mixing. Resuspended cell pellets were lysed using a French pressure cell (two passes at Ͼ1,500 bars; SLM-Aminco) or by passing cells through a 22.5-gauge needle (three passes). Bacterial pellets containing PKG I(78 -326) were resuspended by vortexing in 50 ml of lysis buffer supplemented with protease inhibitors (Roche Applied Science) and lysed using a French pressure cell (five passes at Ͼ1,500 bars; SLM-Aminco).
All cell lysates were clarified by centrifugation at 30,000 ϫ g for 30 min at 4°C and passed through a 0.22-m polyethersulfone syringe filter (Millipore). The resulting clarified, filtered lysates were loaded onto a 5-ml prepacked nickel immobilized metal affinity chromatography column (Bio-Rad) using a P-1 peristaltic pump (GE Healthcare) and washed with 5 column volumes of lysis buffer and 6 column volumes of mid-wash buffer (lysis buffer supplemented with 30 mM imidazole). Proteins were eluted from the column with elution buffer (lysis buffer supplemented with 250 mM imidazole) using a Profinia FPLC (Bio-Rad). Peak fractions containing PKG I␣ were pooled and analyzed by SDS-PAGE. PKG I␣ was dialyzed against 4 liters of 50 mM MES, 300 mM NaCl, 1 mM TCEP, pH 6.9, using 12-14-kDa molecular mass cutoff dialysis tubing and against 1 liter of 50 mM MES, 300 mM NaCl, 1 mM TCEP, 10% glycerol, pH 6.9. Aliquots of PKG I␣ were flash frozen in liquid nitrogen and stored at Ϫ80°C.

Analytical SEC
Samples containing 0.5 mg of PKG I␣ ⌬53 (apo or preincubated with 5 M cGMP) were loaded sequentially onto a Superdex 200 10/300 column (GE Healthcare) connected to an Ä kta PURE FPLC (GE Healthcare) and eluted isocratically in 50 mM MES, 300 mM NaCl, 1 mM TCEP, pH 6.9 (apo), or buffer supplemented with 5 M cGMP (bound). Peak fractions were examined for the presence of PKG I␣ ⌬53 by SDS-PAGE. Peaks were analyzed with Unicorn 6.4 (GE Healthcare) and plotted using DataGraph (Visual Data Tools).

Phosphotransferase assays
Activation of PKG I␣ constructs by cGMP (BioLog) was assessed by measuring phosphorylation of a synthetic peptide substrate (W15; TQAKRKKSLAMA) with [␥-32 P]ATP similar to the method described previously (63). Reactions were quenched after 1.5 or 3 min by spotting onto P81 phosphocellulose (Whatman and Reaction Biology Corp.) and washing with 0.8% phosphoric acid. Determination of 32 P incorporation into W15 substrate was measured by liquid scintillation. Data were analyzed using Excel (Microsoft) and Prism v7 (GraphPad Software) and then plotted using DataGraph.

Measurement of cGMP binding by SPR spectroscopy
SPR measurements were conducted on an SR7500 dualchannel surface plasmon resonance spectrometer connected to an SR7100 autosampler utilizing a standard, dual-channel flow cell (Reichert Technologies). All steps were performed at 25°C. Gold sensorchips coated with a 1500-nm linear polycarboxylate hydrogel (HC1500M, Xantec) were installed on the instrument and pre-equilibrated with SPR buffer (50 mM MES, 150 mM NaCl, 1 mM EDTA, 0.05% Tween 20, 1 mM TCEP, pH 6.9) at 20 l/min. Sensorchips were washed sequentially with a solution of 2 M NaCl and 10 mM NaOH for 5 min each followed by 2 min with SPR buffer. The sensorchip was activated for 5 min with a solution of freshly prepared, degassed 100 M N-hydroxysuccinimide (NHS) and 200 M 1-ethyl-3-(-3-dimethylaminopropyl)carbodiimide hydrochloride dissolved in 1 ml of 500 mM MES, pH 5.5. PKG constructs were diluted to 1 M in 20 mM NaOAc, pH 5.0 (500 l), and injected into the left channel for 10 min until 10,000 -15,000 micro-refractive index units was observed in the difference channel. Excess protein was washed from the surface for 5 min. Any remaining reactive groups on the sensorchips were deactivated by injecting 1 M ethanolamine, pH 8.5, into both channels for 5 min and then washing with SPR buffer for 1 min. Solutions of cyclic nucleotides were prepared using the same SPR buffer used during the PKG I coupling steps. During cNMP binding experiments, solutions were injected at 50 l/min for 1 min (association) and then washed with SPR buffer for 3 min (dissociation). All injected solutions were prepared in clear or amber borosilicate screwtop glass vials with robotic screw-top vial closures (Fisher Scientific). Data were reduced using Scrubber (BioLogic). Data were fit to one-site (with Hill coefficient) and two-site (specific binding) models using GraphPad Prism and plotted using DataGraph.
LC-MS/MS was performed to determine specific phosphorylation sites as follows. Two 150-pmol aliquots of dried PKG ⌬53 PKG I␣ ⌬53 or PKG I␣ full-length enzyme were resuspended in 25 l of 50 mM ammonium bicarbonate buffer, pH 8.5, and 30 l of 0.02% ProteaseMAX (Promega). A 5-l aliquot of sequencing grade trypsin (Promega) was added at a 1:20 ratio (enzyme to substrate). Samples were incubated overnight at 37°C and acidified with formic acid to obtain a final concentration of 0.5%. The samples were evaporated and resuspended in 10 l of 98% water, 2% acetonitrile, 0.1% formic acid prior to LC-MS analyses. UPLC separation of the tryptic peptides was performed on an Acquity high-strength silica T3 1.8-m, 1.0 ϫ 150-mm column (Waters Corp.). UPLC conditions were as follows: LC flow, 50 l/min; column temperature, 45°C; A, water and 0.1% formic acid; B, acetonitrile and 0.1% formic acid; linear gradient from 98% A and 2% B to 65% A and 35% B over 50 min. Approximately 30 pmol of digested PKG ⌬53 was injected. All samples were analyzed by data-independent acquisition between 100 and 2000 m/z using the MS E mode with alternating low-(4-V) and high-energy (15-40-V) acquisitions. The MSE data were processed using Protein-Lynx Global Server v3.0.1 (Waters Corp.) that searched against a consolidated database that included human cGMPdependent protein kinase 1 (Q13976) and possible modifications, including phosphorylations.

SEC-SAXS
SAXS data pertaining to ⌬53 was collected at the Stanford Synchrotron Radiation Lightsource BL4-2 using an MX225-HE detector (Rayonix). A 35-l sample of 6 mg/ml PKG I␣ ⌬53 was injected onto a Superdex 200 3.2/30 column (GE Healthcare) and eluted isocratically using an Ettan FPLC (GE Healthcare) at 0.05 ml/min in 50 mM MES, 300 mM NaCl, 1 mM TCEP, pH 6.9, supplemented with 5 mM DTT as a radical scavenger. The eluant stream was connected in line to a 1.5-mm quartz capillary positioned 1.7 m from the detector. Data were collected using a wavelength of 1.127 Å with a 1 s/frame exposure rate at 22°C. A total of 600 frames were collected with a measurement range of 0.0087-0.5126 Å Ϫ1 . Frames 410 -459 containing the peak corresponding to ⌬53 PKG were averaged and subtracted from the background (45 averaged frames of buffer alone). The data were truncated to exclude the range of 0.0087-0.016 Å Ϫ1 due to radiation-induced aggregation and above 0.19 Å Ϫ1 for all subsequent analyses. The scattering curve, Guinier and P(r) functions, and Porod volume were calculated using the PRIMUS program suite (64). DAMMIF/DAMMIN/DAMMAVER/ DAMFILT was used to generate a total of 20 models of which 19 were used to calculate the final 3D envelope. Models were excluded based upon their NSD to the mean using DAMSEL. The excluded model was determined to have an NSD that was 2 times the standard deviation from the mean of all NSD scores using a cross-correlation matrix. SUPCOMB was further used to fit the averaged 3D envelope to existing heterodimeric structures of PKA RI␣-C (Protein Data Bank code 2QCS), RII␤-C (Protein Data Bank code 4WBB), and PfPKG (Protein Data Bank code 5DYK) (65)(66)(67). Radii of gyration were calculated from the predicted scattering curves of the RI␣-C and RII␤-C heterodimers (using Protein Data Bank codes 2QCS and 4WBB, respectively) using the FoXS and ATSAS servers (https://modbase.compbio.ucsf.edu/foxs/ and https://www. embl-hamburg.de/biosaxs/atsas-online) 6 (38,39). Data were deposited in the Small Angle Scattering Biological Data Bank (SASBDB) using the identifier SASDDS4 (68).