Mechanism of cAMP Partial Agonism in Protein Kinase G (PKG)

Background: Protein kinase G (PKG) is selectively activated by cGMP, but the mechanism of cGMP- versus -cAMP selectivity is not fully understood. Results: cAMP-bound PKG exists in a dynamic exchange between three states, inactive, active, and a partially autoinhibited state, which results in partial agonism. Conclusion: Partial agonism contributes to cGMP- versus -cAMP selectivity. Significance: The cGMP- versus -cAMP selectivity controls the cross-talk between cGMP- and cAMP-dependent signaling pathways. is not controlled uniquely through affinities. Here, we show that cAMP is a partial agonist for PKG, and we elucidate the mechanism for cAMP partial agonism through the comparative NMR analysis of the apo, cGMP-, and cAMP-bound forms of the PKG cyclic nucleotide-binding domain B. We show that although cGMP activation is adequately explained by a two-state conformational selection model, the partial agonism of cAMP arises from the sampling of a third, partially autoinhibited state.

Protein kinase G (PKG) is a major receptor of the cGMP second messenger. By binding to PKG, cGMP regulates intracellular signaling pathways that control a wide range of intracellular processes, such as cell differentiation, platelet activation, memory formation, and vasodilation (1)(2)(3)(4)(5). The PKG signaling pathways are often distinct from those regulated by cAMP-dependent proteins such as protein kinase A (PKA) (6 -12). Thus, a key element of the cross-talk between PKG and PKA signaling pathways is the selective activation of PKG by cGMP rather than by cAMP. However, the mechanism of cGMP-versus-cAMP selectivity in PKG is currently only limitedly understood.
One possible determinant of cyclic nucleotide selectivity is the lower binding affinity of PKG for cAMP than for cGMP (13)(14)(15)(16). Indeed, the C-terminal cyclic nucleotide-binding domain (CNB-B) of PKG I␤ was previously found to exhibit greater binding affinity for cGMP, which has been attributed to specific interactions between the base-binding region (BBR) 5 of CNB-B and the cGMP base moiety, as observed in previously solved crystallographic x-ray structures ( Fig. 1e) (17,18). However, the intracellular concentration of cAMP is typically significantly greater than that of cGMP (19 -23), thus potentially allowing cAMP to bind to PKG despite its lower binding affinity and suggesting that other means of cyclic nucleotide selectivity must be present in PKG.
Another key difference in the response of PKG I␤ to cAMP versus cGMP is that in contrast to cGMP, cAMP is only a partial agonist of PKG I␤, as shown by Cook photometric kinase assays of PKG I␤ activity (Fig. 1f). Such partial agonism of cAMP is not fully explained by previously solved structures of PKG I␤, as they reveal very similar structural rearrangements of both CNB domains in response to either cAMP or cGMP binding ( Fig. 1, b-d) (2,17,18) that are analogous to structural rearrangements observed for other CNB domains (8, 12, 24 -29). Hence, we hypothesize that the partial agonist response of PKG I␤ to cAMP may reflect differences in the dynamics of cAMP versus cGMP-bound PKG I␤. For example, in the context of the two-state conformational selection mechanism, the partial agonism of cAMP could simply arise from a reduced inactive-to-active shift of the dynamic conformational equilibrium underlying PKG activation (30 -33). Indeed, such conformational dynamics have been shown to play a role in the varying responses of other CNB domains to cyclic nucleotides (13,34). Alternatively, in the presence of cAMP, the regulatory region of PKG may dynamically sample a third conformation, which is distinct from both the apo-inactive and cGMP-bound active structures (30) and is at least partially auto-inhibitory of the kinase function of PKG.
As a first step toward testing our hypotheses and differentiating between the two-state versus three-state hypothetical models for the partial agonism of cAMP, here we report the comparative NMR analyses of the apo, cAMP-, and cGMPbound forms of the PKG I␤ CNB-B domain (i.e. "PKG I␤(219 -369)"; see Fig. 1a). This domain is adjacent to the catalytic domain (Fig. 1a), and it has been previously identified as a critical control unit for auto-inhibition and cGMP selectivity in PKG I␤ (17,18). In addition, we examined the CNB-B domain because although the CNB-A domain contributes to the cyclic nucleotide modulation of PKG I function (17,(35)(36)(37), the CNB-A domain undergoes only minor structural changes upon cyclic nucleotide binding (2,17). We found that the partial agonism of cAMP cannot be explained by a simple reversal of the two-state inactive/active conformational equilibrium that rationalizes cGMP activation (2,17). Rather, our data show that FIGURE 1. Introduction to PKG I␤ domain and structural architecture, and to cGMP versus cAMP binding and activation. a, schematic overview of the domain organization of PKG I␤. The regulatory and catalytic regions of PKG are indicated, as are the major underlying structural domains, as follows: the N-terminal dimerization domain (D/D); the autoinhibitory linker region (AI); the cyclic nucleotide-binding domains (CNB-A and CNB-B); the switch helix region (SW); and the two lobes of the catalytic region (N-lobe and C-lobe). Residue numbers for the regulatory region domain boundaries are indicated, and the monomer fragment examined in the current study (referred to herein as PKG I␤(219 -369)) is highlighted in gray. b-d, ribbon structure overlays of the previously solved apo (red ribbon), cAMP-bound (blue ribbon), and cGMP-bound (green ribbon) structures of PKG I␤(219 -369) (17,18). Bound cAMP and cGMP are shown as blue and green sticks, respectively, and the following key structural elements are indicated: the N-terminal ␣-helix bundle (N3A), the ␤-barrel (␤-core); the BBR element involved in cGMP binding; and the C-terminal ␣B-helix (␣B) and switch helix region. To clearly demonstrate the differences in structure, the structures are overlaid at their ␤-barrels. e, zoomed-in view of the cGMP-binding site of the cGMP-bound structure. Key cGMP-interacting residues (17) are indicated as sticks, hydrogen bonds as black dotted lines, and cGMP functional groups that are different/absent in cAMP are highlighted with black circles. f, activation of PKG I␤ by cGMP and cAMP, highlighting the difference in maximal activity levels in the presence of saturating amounts of cAMP versus cGMP. The phosphotransferase activity of PKG I␤ was monitored at various concentrations of cAMP (blue plot) and cGMP (red plot) by a spectrophotometric kinase assay. Here, 1 unit is equal to a substrate conversion rate of 1 mol per min. Each data point was measured at least in duplicate, and the data sets were fitted using a sigmoidal dose-response algorithm. the partial agonism exhibited by cAMP is the result of cAMPbound CNB-B sampling a distinct third conformational state. In this third state, the lid region that caps the cyclic nucleotide base in the fully active state is disengaged, and kinase inhibition is only partially effective. By populating this third conformation, and to a lesser degree the inactive conformation, cAMP reduces the overall level of PKG activity compared to cGMP, thereby providing an explanation for the observed partial agonism of cAMP.

Experimental Procedures
Preparation of the PKG I␤(219 -369) Construct-The PKG I␤(219 -369) construct was expressed with an N-terminal poly-L-histidine tag in Escherichia coli strain BL-21(DE3). The E. coli cells were grown at 37°C in isotopically enriched minimal media supplemented with trace metals, D-biotin, and thiamine-HCl. Expression was induced with 0.4 mM isopropyl ␤-D-1thiogalactopyranoside at an optical density of 1.0 (at ϭ 600 nm), and the cells were further incubated for 16 h at 20°C before being harvested by centrifugation.
The harvested cells were lysed using a cell disruptor, and cell debris was subsequently removed by centrifugation (20,000 ϫ g for 1 h). For this process, the cells were resuspended in 50 mM Tris buffer (pH 7.6) with 150 mM NaCl, 1 mM ␤-mercaptoethanol, and 0.2 mM 4-(2-aminoethyl)benzenesulfonyl fluoride. An initial purification of the PKG construct was then performed using a Ni 2ϩ -Sepharose resin in a 5-ml gravity column. After passing the cell lysate through the column, the column was rinsed with more cell resuspension buffer, followed by 50 mM Tris buffer (pH 7.6) with 500 mM NaCl, 20 mM imidazole, and 1 mM ␤-mercaptoethanol. The protein was then eluted from the column using 50 mM Tris buffer (pH 7.6) with 50 mM NaCl, 300 mM imidazole, and 1 mM ␤-mercaptoethanol.
The collected gravity column eluant was dialyzed in 50 mM Tris buffer (pH 7.6) with 100 mM NaCl and 1 mM ␤-mercaptoethanol, and cleaved with tobacco etch virus protease for 48 h to remove the N-terminal poly-L-histidine tag from the PKG I␤(219 -369) construct. The cleaved poly-L-histidine tag and tobacco etch virus protease were removed from the dialyzed eluant by passing the eluant through a Ni 2ϩ -Sepharose gravity column, followed by a rinse with dialysis buffer. All column flow-through and rinse were collected, and the PKG I␤(219 -369) construct was further purified by size-exclusion FPLC (HiLoad Superdex 120-ml column). The FPLC was performed using a 50 mM Tris running buffer (pH 7.0) with 100 mM NaCl, 1 mM DTT, and 0.02% (w/v) NaN 3 , which also served as the NMR sample buffer for most of the NMR experiments (except for H/D hydrogen exchange, which was performed in a pH 7.0 2 H 2 O-based phosphate buffer).
Spectrophotometric Kinase Activity Assay-Kinase activity was measured using the coupled spectrophotometric assay as described by Cook et al. (38). The phosphotransferase activity of full-length PKG I␤ purified from Sf9 cells was determined in 100-l reaction volumes containing 35 nM PKG I␤, 1 mM Kemptide (LRRASLG), varying concentrations of cGMP or cAMP, respectively, and the assay mix. The depletion of NADH directly correlating with the conversion of ATP was monitored using a spectrophotometer (SPECORD 205; Analytik Jena).
The buffer used for the spectrophotometric assay consisted of 100 mM MOPS (pH 7), 10 mM MgCl 2 , 1 mM phosphoenolpyruvate, 1 mM ATP, 150 units of lactate dehydrogenase, 84 units of pyruvate kinase, 220 M NADH, and 5 mM ␤-mercaptoethanol. Phosphotransferase activity was plotted against the logarithmic cyclic nucleotide concentration, and data points were fitted using a sigmoidal dose-response fit in GraphPad Prism 6.01.
NMR Data Acquisition and Analysis-All NMR spectra for the PKG I␤(219 -369) construct were acquired at 306 K with a Bruker Avance 700-MHz NMR spectrometer equipped with a 5-mm TCI cryoprobe. For the chemical shift-based analyses of PKG I␤(219 -369), two-dimensional { 1 H, 15 N} HSQC NMR spectra were acquired for apo, cAMP-, and cGMP-bound samples of PKG I␤(219 -369). A PKG I␤(219 -369) concentration of 20 M was used for all HSQC samples, and 80 M 15 N-labeled N-acetylglycine was added to the samples as an internal reference for subsequent alignment of the HSQC spectra with one another. The experiments for cAMP-and cGMP-bound PKG I␤(219 -369) were performed with a cyclic nucleotide concentration of 2 mM to achieve saturated ligand binding, thus minimizing the influence of differing ligand binding affinities on the results of the comparative analyses. The spectra were processed with NMRPipe (39) and analyzed using Sparky (40). Peak assignments for the HSQC spectra were obtained from standard three-dimensional triple-resonance NMR spectra (i.e. HNCACB, CBCA(CO)NH, HNCA, and HN(CO)CA), using the automated PINE-NMR server (41) to facilitate the initial assignment process. The assigned HSQC spectra were aligned using Sparky, and the resulting ( 1 H, 15 N) chemical shift data for apo, cAMP-, and cGMP-bound PKG I␤(219 -369) were utilized to compute ( 1 H, 15 N) combined chemical shifts, perform chemical shift projection analyses (CHESPA), and build chemical shift correlation plots, as described previously (13,(42)(43)(44)(45). The CHESPA "outlier" residues are defined as those deviating significantly from the two-state equilibrium between conformations resembling the apo and cGMP-bound PKG I␤(219 -369) structures (i.e. residues with projection angle cosine values less than 0.8). The chemical shift correlation plots were confined to residues with CHESPA fractional activation values less than 1.2 and projection angle cosine values greater than 0.95.
To further examine PKG I␤(219 -369) dynamics, backbone 15 N relaxation (R 1 , R 2 , HN-NOE) and hydrogen exchange (H/D and H/H) NMR data for the apo, cAMP-, and cGMP-bound samples of PKG I␤(219 -369) were acquired and analyzed using a protocol similar to that described previously (13,46,47), with the T 1 and T 2 data sets acquired as pseudo-three-dimensional matrices (48). Reduced spectral densities (RSDs) were then computed from the 15 N relaxation rates as described previously (13,46,47). Considering the rigidity typically exhibited by the inner ␤-strands (i.e. ␤3, ␤4, ␤7, and ␤8) of the ␤-barrel of CNB domains (27,46), R 1 R 2 and/or J(0) values significantly greater than the inner-␤-strand average were considered indicative of the presence of millisecond-microsecond backbone dynamics, whereas R 1 R 2 , J(0), and/or HN-NOE values significantly less than the inner-␤-strand average, and/or J( H ϩ N ) values significantly greater than the inner-␤-strand average, are indicative of the presence of picosecond-nanosecond backbone dynamics. Differences in the R 1 R 2 and J(0) values between dif-ferent samples of PKG I␤(219 -369) were deemed significant if the error bars of the compared data points did not overlap one another. Meanwhile, because of the larger errors affecting the HN-NOE and J( H ϩ N ) values, differences in these values were deemed significant if the error bar of one of the compared data points did not overlap the other data point, and vice versa. Finally, significant differences in hydrogen exchange rates were indicated by non-overlapping protection factor (PF) value error bars or transitions between different time scales of hydrogen exchange. The latter are defined, from fast to slow exchange, as follows: H/H-time-scale exchange, H/D-dead-time exchange, quantifiable exchange (i.e. measurable PF), and finally, exchange that is too slow for reliable PF quantification.
Determination of State Populations in the Three-state Model-The state populations were estimated based on the fractional activations measured through CHESPA. Specifically, the state populations were determined based on the average fractional activations observed for the "pre-lid" region, which is reflective of the population of the state in which this region adopts an active conformation (i.e. the active state), and for the CNB-B domain region preceding residue 340, which is reflective of the total population of states in which this region is active-like (i.e. the intermediate and active states). Thus, the population of the intermediate state is reflected by the difference between the two average fractional activations, whereas the population of the inactive state is the percentage not accounted for by the other two states.

Initial Assessment of cAMP-Versus-cGMP Differences in PKG I␤(219 -369) through Compounded Chemical Shift Changes-
As an initial assessment of the differences between cAMP-and cGMP-bound PKG I␤(219 -369), we examined the pairwise compounded chemical shift changes between apo, cAMP-, and cGMP-bound PKG I␤(219 -369) (Fig. 2c). Chemical shifts from the apo state revealed that binding of both cAMP and cGMP results in major widespread structural changes within the CNB-B domain, including part of the switch helix region (Fig.  2c, black and green points), in agreement with the available x-ray structures (17,18). In addition, as expected, the cAMPversus-cGMP chemical shifts detected differences for residues at/near the base-binding pocket, including residues in the ␤5-strand of the BBR, and the cGMP-capping residue from the switch helix region (i.e. residue Tyr-351; Fig. 2c, red points). However, Fig. 2c also shows that the cAMP-and cGMP-bound states exhibit differences that are not apparent from previous structural comparisons. For example, unanticipated cAMPversus-cGMP chemical shift differences were observed for residues outside the binding site, including the residues spanning the 340 -350 region, and the N-terminal N3A motif spanning residues 220 -250 (Fig. 2c, red points marked with dotted boxes). To better understand the origin of these long range differences, we analyzed the cAMP-bound chemical shifts using CHESPA (Fig. 2, b, d, and e).
CHESPA of cAMP-bound PKG I␤(219 -369) Reveals Three Major Types of Long Range Perturbations Caused by the Replacement of cAMP with cGMP-The CHESPA analysis was performed for cAMP-bound PKG I␤(219 -369), utilizing apo and cGMP-bound PKG I␤(219 -369) as reference states (Fig.  2b). The resulting computed fractional activations (Fig. 2d) and projection angle cosine values (Fig. 2e) were complemented by a cAMP-apo versus cGMP-apo chemical shift correlation plot analysis, whose slope provides an alternative estimation of the relative cAMP-versus-cGMP fractional activation (Fig. 2f). Examination of the CHESPA (Fig. 2, d and e) and chemical shift correlation results (Fig. 2f) revealed three groups of non-binding site residues demonstrating notable cAMP-versus-cGMP differences. The first group includes residues 340 -350 (referred to herein as the "pre-lid" region), which link the ␣B-helix to the C-terminal capping residue (i.e. Tyr-351; a residue referred to herein as the "lid"). The residues in this first group exhibited an average fractional activation of only ϳ40% (Fig. 2d), as was also evident from partial HSQC peak shifts for these residues (Fig. 2h), suggesting that in the presence of cAMP the pre-lid region is subject to a partial disengagement from its active-state structural arrangement. The second group of residues is composed of the residues N-terminal to residue 340, which exhibited an average fractional activation of ϳ87% in the presence of cAMP (Fig. 2, d, f, and g), reflecting a markedly different degree of activation compared with the pre-lid region. The third notable group spans multiple "outlier" residues, which deviate from a two-state conformational equilibrium in the presence of cAMP, as seen through low projection angle cosine values (Fig. 2e, purple highlights) and non-linear arrangements of apo, cAMP-, and cGMP-bound state HSQC cross-peaks (Fig. 2, i and j). The identified outlier residues include cGMP-binding residue Arg-297 (from the ␤5-strand), as well as multiple non-binding-site residues in the pre-lid region N terminus and the N3A motif (Fig. 2e, purple highlights). When mapped on the structure of cGMP-bound PKG I␤(219 -369), the non-binding-site outlier residues form a cluster spanning the interface between the pre-lid region and the N3A motif (Fig. 3a, gray surface), indicating that the partial disengagement of the pre-lid region caused by the replacement of cGMP with cAMP may in turn influence the N3A motif as well. In addition, the low degree of fractional activation observed for the pre-lid region (Fig. 2d) points to a loss of structure in this segment, suggesting that dynamics are involved in the observed perturbations caused by cAMP. To test this hypothesis, we examined PKG I␤(219 -369) dynamics more closely through 15 N relaxation NMR measurements.
Assessment of Apo-Versus-cGMP Differential Dynamics through Backbone 15 N Relaxation Rates and Reduced Spectral Densities-As a first assessment of PKG I␤(219 -369) dynamics, we measured backbone 15 N relaxation rates and reduced spectral densities (RSDs) for the apo, cAMP-, and cGMPbound states (Figs. 4 and 5). The apo versus cGMP-bound state comparison revealed several unanticipated changes in dynamics. First, the ␤4-␤5 loop in the base binding region (BBR) exhibited consistent picosecond-nanosecond dynamics in both the apo and cGMP-bound states (Figs. 4, c and d, and 5a and c, gray highlights). In addition, the phosphate-binding cassette (PBC) exhibited a shift from millisecond-microsecond to picosecond-nanosecond dynamics upon cGMP binding (Figs. 4, c  and d, and 5a and c, gray highlights), suggesting a retention of residual PBC dynamics. This is in marked contrast to the more quantitative quenching of PBC dynamics observed previously for the bound states of other CNB domains, which unlike PKG include a proline residue in the PBC (8, 12, 27, 46, 49, 50). In addition, our data indicated that the region C-terminal to the ␣B-helix (i.e. C-terminal to residue 340) is subject to picosecond-nanosecond dynamics in the apo state (Figs. 4, c and d, and The secondary structural elements from the apo-state x-ray structure are indicated across the top of each graph as follows: black bars ϭ ␣-helices; brown bars ϭ ␤-strands. c, compounded chemical shifts of cAMP-bound (black points) and cGMP-bound (green points) PKG I␤(219 -369) from apo PKG I␤(219 -369), and of cAMP-bound from cGMP-bound PKG I␤(219 -369) (red points). Black dotted boxes indicate notable cAMP-versus-cGMP chemical shift differences observed for residues outside the cGMP-binding site. d and e, fractional shifts toward activation ("X"; d) and projection angle cosines ("cos"; e) achieved by cAMP binding, relative to apo and cGMP-bound PKG I␤(219 -369). The average fractional shift toward activation computed for the residue 340 -350 region (average of X values from d) and the overall fractional shift toward activation computed for residues N-terminal to residue 340 (computed slope in f) are marked by arrows and black dotted lines. The residue 340 -350 region is indicated in all plots by gray highlighting, and notable low-cos-value residues (outliers, as defined under "Experimental Procedures") are indicated by purple highlights. f, chemical shift correlation plot of 1 H (black points) and scaled 15 N (brown points) chemical shifts of cAMP-bound versus cGMP-bound PKG I␤(219 -369) (both relative to the apo state). The black line delineates the line of best fit as determined from linear regression, and the corresponding slope and correlation coefficient are indicated. g-j, expansions of the overlaid { 1 H, 15 N} HSQC spectra of apo (red contours), cAMP-bound (blue contours), and cGMP-bound (green contours) PKG I␤(219 -369), illustrating the cAMP-associated perturbations of representative high-cos-value residues from the residue 340 -350 region (h) and N-terminal to residue 340 (g), as well as representative low-cos-value (outlier) residues from the N3A (i) and residue 340 -350 region (j).   Backbone 15 N relaxation rates for apo (red points), cAMP-bound (blue points), and cGMP-bound (green points) PKG I␤(219 -369). a, R 1 relaxation rates; b, R 2 relaxation rates; c, product of R 1 and R 2 ; d, ( 1 H, 15 N) NOE values, computed as I sat /I nonsat . Black horizontal lines denote the average values computed for the inner ␤-strands of the ␤-barrel, which, due to the rigidity of this region of the protein, are assumed to represent the overall tumbling motion of the protein in solution. Residues exhibiting notable enhancements of millisecond-microsecond/picosecond-nanosecond dynamics in cAMP versus cGMP-bound PKG I␤(219 -369) are indicated by red/orange highlights, respectively, while other regions exhibiting notable dynamics are indicated by gray highlights, and an apo-state residue for which the R 2 rate could not be properly quantified (due to a poor signal-to-noise ratio) is marked with an asterisk. The secondary structure elements from the apo-state x-ray structure are indicated across the top of each graph (black bars ϭ ␣-helices, brown bars ϭ ␤-strands), and residues for which no data are shown are prolines or were not successfully assigned in the relaxation NMR spectra for one or more states. (17). The binding of cGMP produces a dramatic quench of picosecond-nanosecond dynamics in the pre-lid region residues (i.e. residues 340 -350) that bridge the ␣B-helix to the lid residue Tyr-351 (Figs. 4, c and d, and 5, a and c). However, despite this dramatic quenching of dynamics, Tyr-351 retains partial picosecond-nanosecond flexibility (Figs. 4, c  and d, and 5, a and c).

Assessment of cAMP-Versus-cGMP Differential Dynamics through Backbone 15 N Relaxation Rates and Reduced Spectral
Densities-The region C-terminal to the ␣B-helix is also a site of previously unanticipated cAMP-versus-cGMP differences in dynamics. In particular, partially enhanced picosecond-nanosecond dynamics were observed in the cAMP-bound state throughout the region C-terminal to residue 340, including the lid residue Tyr-351 (Figs. 4, c and d, and 5, a and c, orange  highlights). Furthermore, several other residues exhibited enhanced millisecond-microsecond dynamics in the cAMPbound state (Figs. 4c and 5a, red highlights). The latter enhance-ments include binding site residues from the ␤5-strand and the PBC N terminus, as well as non-binding-site residues such as those in the N3A motif and the pre-lid region N terminus (Figs.  4c and 5a, red highlights). These cAMP-versus-cGMP differences cannot be explained by variations in diffusional anisotropy, given the similarity of the cAMP-and cGMP-bound structures of the CNB-B domain (Fig. 1, b-d). Therefore, the pre-lid region emerges as a site affected by cAMP-versus-cGMP enhancements in both picosecond-nanosecond and millisecond-microsecond dynamics. This conclusion is further supported by the correlated changes in high-and low-frequency spectral densities for this region (i.e. J( H ϩ N ) versus J(0); Fig. 5d). Fig. 5d shows that the cAMP-versus-cGMP changes deviate from the simple rigid rotor model, suggesting the simultaneous presence of both picosecond-nanosecond and millisecond-microsecond flexibility in the pre-lid region of the cAMP-bound state. These observations corroborate the partial disengage-  N ) values. The RSDs were computed as described previously (13,46,47) based on the relaxation rates reported in Fig. 4. Black horizontal lines denote the average values computed for the inner ␤-strands of the ␤-barrel, which, due to the rigidity of this region of the protein, are assumed to represent the overall tumbling motion of the protein in solution. All color codes illustrated are as in Fig. 4. d, two-dimensional J( H ϩ N ) versus J(0) plots for residues 341-350. The values expected for a rigid rotor in the absence of millisecond-microsecond dynamics are indicated by the black plot, and the cAMP-bound state data are indicated by a blue dashed oval. ment of the lid, as suggested by the CHESPA analysis (Fig. 2d). Furthermore, the map of non-binding site loci subject to enhanced millisecond-microsecond dynamics in the presence of cAMP coincides quite well with that of the outlier residues identified from CHESPA (Fig. 3b, red surfaces versus Fig. 3a, gray surfaces). Overall, the data of Figs. 4 and 5 confirm that cGMP replacement with cAMP leads to increased structural dynamics not only at binding site residues but also at selected allosteric sites of PKG I␤(219 -369) (e.g. the pre-lid and N3A loci; Fig. 3b). Unlike the cAMP-versus-cGMP changes, the apoversus-cGMP variations in the correlated J( H ϩ N ) versus J(0) spectral densities follow, within error, the values expected for a rigid rotor in the absence of millisecond-microsecond dynamics (Fig. 5d, black line), further confirming that cGMP binding quenches picosecond-nanosecond dynamics in the pre-lid region.
Further Assessment of Dynamics through Hydrogen Exchange Rates-Next, we measured the backbone hydrogen exchange rates for apo, cAMP-, and cGMP-bound PKG I␤(219 -369) (Fig. 6). The comparison of the apo and cGMP-bound states revealed unanticipated changes in dynamics. For example, multiple residues of both the ␣B-helix and the N3A motif showed significantly reduced hydrogen exchange rates, pointing to a reduced access to partially unfolded and solvent-exposed excited states upon cGMP binding (Fig. 6). However, when cGMP is replaced by cAMP, a partial recovery of solvent exposure is observed, as several residues exhibited greater hydrogen exchange rates in the cAMP-bound state than in the cGMPbound state (Fig. 6). The affected areas include residues in the vicinity of the binding site (e.g. the ␤5and ␤6-strands, the ␤2-3 loop adjacent to the PBC, and the PBC N and C termini), as well as non-binding-site residues, such as the N terminus of the pre-lid region and the N3A motif (Fig. 6, red highlights). Overall, the identified non-binding site residues affected by cAMP overlap quite well with the outlier residues identified from CHESPA (Fig. 3c, red surfaces with dotted outline versus Fig. 3a, gray surfaces with dotted outline), suggesting that in agreement with the 15 N relaxation and RSD results, the cAMP-versus-cGMP perturbations observed for PKG I␤(219 -369) involve an increase in dynamics at the interface between the ␣-helical regions flanking the ␤-barrel (Fig. 3). Furthermore, the pre-lid and lid regions exhibited high solvent exposure in all three states, as suggested by fast hydrogen exchange (Fig. 6), thus confirming the finding from 15 N relaxation and RSDs that the lid retains significant dynamics even in the cGMP-bound state.

Discussion
The comparative NMR analysis of the CNB-B construct (i.e. PKG I␤(219 -369)) in its apo, cGMP-, and cAMP-bound states revealed that the cAMP-versus-cGMP differences are more extensive than previously anticipated, reaching well beyond the cGMP contact sites to crucial allosteric structural elements. In particular, the chemical shift, 15 N relaxation, and hydrogenexchange data (Figs. 2 and 4 -6) consistently show that cAMP binding results in only a partial recruitment of the pre-lid region to its active-state conformation, which in turn leads to perturbations of the adjacent N3A region (Fig. 3). Furthermore, when bound to the PKG I␤ CNB-B domain, cAMP exhibits a ; residues with hydrogen exchange rates quantifiable by H/D NMR are plotted as the logarithms of PFs computed from the H/D NMR data, whereby higher values indicate slower hydrogen exchange; and residues whose hydrogen exchange rates were too slow to be properly quantified are denoted by S.E. and triangles. Residues exhibiting notable enhancements of hydrogen exchange rates in cAMP versus cGMPbound PKG I␤(219 -369) are indicated by red highlights, and the secondary structure elements from the apo-state x-ray structure are indicated across the top of each graph: black bars ϭ ␣-helices; brown bars ϭ ␤-strands. Residues for which no data are shown are prolines or were not successfully assigned in the NMR spectra.
higher propensity for an anti base orientation compared to cGMP, which mainly adopts a syn conformation (17,18). A possible explanation for the differing syn/anti conformational propensities of bound cAMP versus cGMP is the intrinsic preference of free cAMP for the anti orientation together with a lack of stable interactions between the 6-NH 2 moiety of cAMP, which is replaced by a less bulky hydrogen-bond-accepting 6-oxo moiety in cGMP, and the residues of the cyclic nucleotide-binding site responsible for cGMP selectivity, including residue Arg-297 (in ␤5), which interacts with the 6-oxo moiety of cGMP (Fig. 1e). The anti versus syn difference may reduce the efficacy with which the lid residue (i.e. Tyr-351) is recruited to its active-state capping interaction with the bound cyclic nucleotide, thus providing an explanation for the partial pre-lid region disengagement observed in the cAMP-bound state. Indeed, the analogous capping residue of the cAMP-bound PKA CNB-B domain (i.e. Tyr-371) adopts a structural arrangement similar to that of Tyr-351 in the cGMP-bound PKG CNB-B domain, and the bound cyclic nucleotides in both of these domains adopt a syn conformation (17), suggesting that replacement of cGMP by cAMP without a syn-to-anti transition is not sufficient to achieve the observed lid disengagement.
The inability of cAMP to recruit the pre-lid region into the active state to the same extent as observed for the majority of  Fig. 2d), and the overall fractional shift toward activation computed for residues N-terminal to residue 340 (computed slope in Fig. 2f).
the CNB-B domain residues preceding residue 340 (Fig. 2d) suggests that the partial agonism of cAMP cannot be explained by a simple two-state active-inactive model of conformational selection. A purely two-state model dictates that all of the CNB-B domain residues should exhibit a similar fractional activation, which is not the case here (Fig. 2d). Hence, the dual pattern of fractional activations observed in Fig. 2d (horizontal  arrows) implies the presence of a third state in which kinase inhibition is only partially effective. The deviation from a twostate model is also independently confirmed by the observation for multiple non-binding-site residues of clear deviations from a linear apo-cAMP-cGMP CHESPA pattern (Fig. 2, e, i, and j). These results consistently suggest that the partial agonist response of PKG I␤ to cAMP (Fig. 1f) cannot be explained by a simple reversal of the two-state inactive/active conformational equilibrium (Fig. 7, red and green structures). Instead, our results point to a dynamic exchange of the cAMP-bound CNB-B domain with a third, "intermediate," conformational state (Fig. 7, orange structure), leading to a three-state model (Fig. 7).
In the proposed three-state model (Fig. 7), the population of the intermediate state is ϳ47%, whereas that of the apo-like inactive state is ϳ13%, as assessed based on the fractional activations measured through CHESPA (Fig. 2d). The intermediate state sampled in the presence of cAMP maintains an active-like conformation with the exception of the pre-lid region (i.e. residues 340 -350). Upon cAMP binding, the pre-lid region becomes disengaged and more dynamic in the picosecondnanosecond and millisecond-microsecond time scales (Fig. 7, orange structure). Notably, the disengagement of the pre-lid region would alter the relative positions of the PKG regulatory and catalytic regions, which are connected to one another by the CNB-B domain/switch helix region (Fig. 1a) (2,4,17,18). Such repositioning is likely to affect the access of substrates to the PKG catalytic site (6), explaining why the intermediate state retains at least partial kinase inhibition. Hence, by promoting population of the intermediate conformation, and to a lesser degree the inactive conformation (Fig. 7), cAMP reduces the overall level of PKG activity compared to cGMP, explaining the partial agonism observed for cAMP (Fig. 1f). Such partial agonism represents a key determinant of the cGMP-versus-cAMP selectivity of PKG, and together with differing binding affinities for cAMP versus cGMP (17,18), it contributes to the minimization of cross-talk between cGMP-and cAMP-controlled signaling pathways.
Finally, it is worth noting that while the partial agonism of cAMP contributes to cyclic nucleotide selectivity for cGMPcontrolled signaling pathways, previous studies have suggested that selected signaling responses are influenced by cAMP via modulation of PKG activity, rather than cAMP-dependent proteins such as PKA (5,51). A possible explanation for this phenomenon is that as the intracellular cAMP concentration rises, the minimal cGMP concentration necessary for full PKG activation increases, and the shape of the cGMP-dependent activation profile of PKG may also change. Such modulation of PKG activity by cAMP represents a notable contribution to the control of PKG, and may thus have relevant physiological implications for cGMP-dependent intracellular signaling.
Author Contributions-B. V. S., R. S., and G. M. conceived and designed the study. B. V. S., R. S., and R. G. performed and analyzed the NMR experiments. R. L. and F. W. H. performed and analyzed the kinase assay. C. K. designed and constructed vectors for expression. B. V. S. and G. M. wrote the paper. All authors reviewed the results and the manuscript.