Cyclic AMP Analog Blocks Kinase Activation by Stabilizing Inactive Conformation: Conformational Selection Highlights a New Concept in Allosteric Inhibitor Design*

The regulatory (R) subunit of protein kinase A serves to modulate the activity of protein kinase A in a cAMP-dependent manner and exists in two distinct and structurally dissimilar, end point cAMP-bound “B” and C-subunit-bound “H”-conformations. Here we report mechanistic details of cAMP action as yet unknown through a unique approach combining x-ray crystallography with structural proteomics approaches, amide hydrogen/deuterium exchange and ion mobility mass spectrometry, applied to the study of a stereospecific cAMP phosphorothioate analog and antagonist((Rp)-cAMPS). X-ray crystallography shows cAMP-bound R-subunit in the B form but surprisingly the antagonist Rp-cAMPS-bound R-subunit crystallized in the H conformation, which was previously assumed to be induced only by C-subunit-binding. Apo R-subunit crystallized in the B form as well but amide exchange mass spectrometry showed large differences between apo, agonist and antagonist-bound states of the R-subunit. Further ion mobility reveals the apo R-subunit as an ensemble of multiple conformations with collisional cross-sectional areas spanning both the agonist and antagonist-bound states. Thus contrary to earlier studies that explained the basis for cAMP action through “induced fit” alone, we report evidence for conformational selection, where the ligand-free apo form of the R-subunit exists as an ensemble of both B and H conformations. Although cAMP preferentially binds the B conformation, Rp-cAMPS interestingly binds the H conformation. This reveals the unique importance of the equatorial oxygen of the cyclic phosphate in mediating conformational transitions from H to B forms highlighting a novel approach for rational structure-based drug design. Ideal inhibitors such as Rp-cAMPS are those that preferentially “select” inactive conformations of target proteins by satisfying all “binding” constraints alone without inducing conformational changes necessary for activation.

as Rp-cAMPS are those that preferentially "select" inactive conformations of target proteins by satisfying all "binding" constraints alone without inducing conformational changes necessary for activation. cAMP is an important second messenger that allows both prokaryotes and eukaryotes to sense cues from the environment and then translate extracellular signals into a biological response (1). The cyclic nucleotide binding (CNB) 1 domain is an ancient motif that appears to have co-evolved as the primary receptor for cAMP (2). CNBs function as regulatory modules in different classes of proteins, such as catabolite activator protein, cyclic nucleotide gated channels, Epac proteins, and guanine-nucleotide-exchange factors, where effector protein and domain activity is controlled by cAMP binding to CNB (3). In eukaryotes, one of the principal targets of cAMP is the cAMP-dependent protein kinase, protein kinase A (PKA), whose regulatory subunit (R-subunit) contains two tandem CNBs. In the absence of cAMP, PKA exists as an inactive tetrameric holoenzyme with the homodimeric R-subunits bound to two catalytic (C) subunits. cAMP binding to the CNBs of the R-subunits induces conformational changes, leading to dissociation of the C-subunit and activation of PKA (4). PKA is the only known example where cAMP-dependent conformational changes lead to dissociation of the effector protein (C-subunit), whereas in all other CNB-containing proteins, the consequences of cAMP removal and binding result in induced conformational changes without dissociation of the target domains and proteins (3). Of the many isoforms of the C-and R-subunits, C␣ and RI␣ are most widely distributed across all mammalian tissue and are the best studied isoforms of PKA (5). Of the two CNBs (CNB-A and CNB-B) in RI␣, the N-terminal CNB-A, together with a pseudo substrate region (IS) (6) and corresponding to the deletion fragment, RI␣(91-244), is sufficient for mediating high-affinity interactions with both cAMP and the C-subunit (7) (Fig. 1A). The most critical interaction sites for the C-subunit are localized entirely within this fragment (8). Although regions N-terminal to the pseudosubstrate site have been predicted to associate weakly with the C-subunit on the basis of limited proteolysis footprinting experiments (9), there were no significant differences in amide hydrogen/deuterium (H/D) exchange that could be attributable to regions N-terminal to the pseudosubstrate site between the full-length apo RI␣ and RI␣:C (10) complexes. Furthermore comparisons of the amide H/D exchange between apo and cAMP-bound RI␣ showed no differences in the N-terminal region of the protein (10). RI␣(91-244) thus serves as a generic prototype for the CNB domain as well as a minimal model for mapping interactions with both cAMP as well as the C-subunit, and has been analyzed comprehensively in solution using amide H/D exchange mass spectrometry (11) and nuclear magnetic resonance spectroscopy (NMR) (12).
The structural biology of CNBs from each of the above classes of molecules reveals a highly conserved architecture with two subdomains: a ␤-subdomain with an eight-stranded ␤-sheet, containing a solvent-shielded pocket for binding cAMP and a noncontiguous ␣-subdomain. The structure of the PKA holoenzyme complex (8,13) revealed a dramatic conformational change in the R-subunit compared with the cAMP-bound form and was termed the "H-conformation" (Fig.  1B). Structures of the CNB domain have been solved with the ligand cAMP (14,15) and this cAMP-bound conformation has been denoted the "B-conformation" (Fig. 1C). A comparison An N-terminal dimerization/docking domain (D/D (gray)) is followed by an inhibitory region/pseudo substrate region (IS, (red)) and two tandem cAMP binding domains, CNB-A and CNB-B at the C-terminal end. A deletion mutant, RI␣ (74 -379) (R AB ) with IS, CNB-A and CNB-B. A smaller deletion mutant containing IS and CNB-A alone, RI␣ (91-244) (R A ), is sufficient for high affinity binding to the C-subunit (7). B, Structure of the R-subunit in the C-subunit-bound conformation (H-conformation) (from the R A : C complex structure, PDB: 3FHI). The ␣-subdomain is in pale green and the ␤-subdomain is in tan. C, Structure of the R-subunit (bound to cAMP, PDB: 1RGS) in the B-conformation. D, Apo R-subunit toggles between cAMP-bound and C-subunit-bound states. E, The width of the Phosphate Binding Cassette (PBC) pocket in the H form is 10.1Å (between the C␣ atoms of Leu 201 and Pro 208). F, The corresponding width of the PBC pocket in B form is 8.7 Å. cAMP binds with high affinity to the pocket with two critical and conserved amino acids, Glu 200 and Arg 209, which coordinate binding to the ribose 2Ј-OH and the equatorial oxygen of cAMP, respectively. Asp 170, present in the ␤1-␤2 loop, is seen forming a salt bridge with Arg 209, which is critical for allosteric communication between the PBC and the C-subunit-interface. All the above figures were generated with the program pyMOL. of the structures of the B and H forms demonstrated for the first time that the CNB domain has a remarkable dynamic range in terms of the conformations it can assume. This established the R-subunit as a conformational toggle switch between B (14,15) and H (8) conformational states (Fig. 1D). The phosphate binding cassette (PBC) within the ␤ subdomain reveals a conserved hydrogen bonding network, connecting residues that are critical for specific binding to cAMP (3). This presumably provides a solvent-shielded and phosphodiesterase-resistant environment for cAMP. One of these residues is Arg 209 (in PKA, CNB-A), which is positioned with precise geometry to anchor the equatorial oxygen of the critical phosphodiester bond of cAMP (Figs. 1E, 1F). A second residue is Glu 200, which is within H-bonding distance of the 2ЈOH of the ribose moiety. Substitutions at both these residues abolish cAMP binding to the PKA R-subunits (16,17).
One of the major puzzles in the field of cAMP signaling is that out of numerous cAMP analogs screened, only the Renantiomers of thio-substituted cAMP analogs or phosphorothioates ((RpcAMPS and abbreviated to Rp), where the equatorial oxygen of cAMP is substituted with sulfur) have been observed to be antagonists of PKA (17,18). Recent solution studies of Rp bound to PKA indicate that it functions more as an inverse agonist to increase intersubunit interactions via increased dynamics of each of the subunits (19). In contrast, substitution of the oxygen at the axial position with sulfur yields the cAMP mimetic and agonist SpcAMPS (abbreviated to Sp). In this study, we have used a combination of x-ray crystallography and solution mass spectrometry (solution phase amide H/D exchange and ion mobility mass spectrometry) to describe the structure and conformational dynamics of the R-subunit by comparing the apo, cAMP-bound, the agonist Sp and inverse agonist Rp-bound states. In addition to establishing the molecular basis for antagonism of Rp, our results address the basis for the 1000-fold difference in affinity between the cAMP-bound and apo R-subunit for the Csubunit (20).
Because RI␣(91-244) containing CNB-A is a versatile minimal model for examining cAMP as well as C-subunit interactions and conformational changes, we have used this double deletion construct in this study, henceforth referred to as R A . Our results reveal that the apo and cAMP-bound R-subunit crystallized in the B-conformation and the two structures are nearly superimposable. Surprisingly, the Rp-bound R A crystallized in the H-form, in the same conformation as the inactive R A :C holoenzyme complex (8). Our results show that whereas x-ray crystallography reveals a snapshot of the apo R A -subunit in the B conformation, an analysis of temperature factors, amide deuterium exchange measurements and collisional cross sectional areas of molecules from ion mobility mass spectrometry indicate that the R-subunit is highly dynamic and exists as an ensemble of at least two end-point B (15) and H (8) conformations. It is clearly evident that cAMP and a cAMP-agonist (Sp) preferentially recognize and bind the B conformation, whereas the C-subunit and the cAMP-antagonist (Rp) preferentially recognize and bind the H conformation. We propose this represents "conformational selection" in apo RI␣. In contrast, interconversion of cAMP-bound B-form and C-subunit-bound H-form likely occurs via "induced fit." Our study provides valuable insights into the allosteric regulation of PKA by cAMP and has important implications for designing small molecule disruptors of protein-protein interactions.
Expression and purification of PKA C-subunit-The PKA C-subunit with an N-terminal hexahistidine tag was expressed in E. coli [BL21 (DE3)] and purified using the Talon resin. Large scale expression was achieved by culturing bacteria at 37°C until mid exponential phase, followed by induction with 500 M IPTG overnight at 20°C. Cells were harvested at 6000 ϫ g (Beckman Coulter JA-10 rotor) for 20 min and the cell pellet was resuspended in lysis buffer [50 mM potassium monobasic phosphate, 20 mM Tris-HCl (pH 8.0), 100 mM NaCl, 5 mM ␤-mercaptoethanol, 5 mM imidazole]. Cells were lysed by a sonicator (Misonix) and centrifuged at 17,000 ϫ g (Sigma, Sartorius, 19776-H rotor,) at 4°C for 40 min and the supernatant was incubated with talon resin at 4°C for 1 h. The resin was then transferred into columns (Bio Rad). Washes were performed with both lysis buffer and wash buffer (Lysis buffer, pH 7.5) followed by elution buffer containing lysis buffer with 200 mM imidazole, pH 7.0. Further purification was achieved by size-exclusion chromatography [S200 column, AKTA system (GE Healthcare)].
Expression and Purification of PKA R A -PKA R A was expressed in E. coli [BL21 (DE3)] and purified using cAMP Sepharose affinity chromatography as described earlier (14). Cells, grown upto midexponential phase, were induced with 500 M IPTG overnight at 20°C. Cells were harvested at 6000 ϫ g (Beckman Coulter JA-10 rotor) for 20 min and the cell pellet was resuspended in lysis buffer (20 mM 2-(N-morpholino)ethanesulfonic acid pH 6.5, 100 mM NaCl, and 2 mM EDTA) and lysed by sonication. Centrifugation of cell lysates was carried out at 17,000 ϫ g for 40 min and the supernatant was precipitated with 45% ammonium sulfate. The ammonium sulfate precipitate was separated from supernatant by centrifugation at 6500 ϫ g for 15 min and resuspended in lysis buffer followed by incubation with cAMP Sepharose resin overnight at 4°C. The resin was then transferred into columns and purified R A was eluted with 40 mM cGMP (50 mM 2-(N-morpholino)ethanesulfonic acid pH 5.8, 200 mM NaCl, 2 mM EDTA, 40 mM cGMP). The protein was further purified by size-exclusion chromatography [S75 column, AKTA (GE Healthcare)].
Purification of PKA Holoenzyme-R A and C-subunit in a 3:1 molar ratio were dialyzed for 16 h, against buffer containing 10 mM Mops cAMP Antagonist Promotes Selection and Preferentially Binds Inactive of Protein Kinase A (pH 7.0), 100 mM NaCl, 1 mM EGTA, 2 mM MgCl 2 , 1 mM dithiotreitol and 10% glycerol using Spectra/pore 3.5 kDa molecular weight cut off membrane. The holoenzyme was further purified by size-exclusion chromatography (S75 column, AKTA FPLC system).
Crystallization, Data Collection, Structure Solution, and Refinement, of apo R A and cAMP-Bound R A -PKA R A was set up for crystallization at 25°C in hanging drops using the vapor diffusion method in 0.1 M sodium cacodylate trihydrate pH 6.5, and 30% w/v polyethylene glycol 8000.
The crystals were transferred to a cryoprotectant solution (mother liquor containing 20% glycerol) and flash-frozen in liquid nitrogen. X-ray diffraction data were collected at the Beamline 9.1 (The Stanford Synchrotron Radiation Lightsource, CA). Diffraction data were processed and scaled using HKL2000. The final data were integrated and scaled in the space group P6 5 22 (a ϭ b ϭ 56.4, c ϭ 168 Å) with satisfactory statistics shown in Table I. Initial phases of apo R A were generated by molecular replacement using the A-domain (residues 113-244) (PDB code 1RGS) (15) as a search model using AMoRe of the CCP4 program suite (21). There was one molecule in the asymmetric unit, corresponding to a solvent content of 45% (V M ϭ 2.2 Å 3 /Dalton). The phases were improved by solvent flattening using Density Modification. The resulting F o map calculated from the improved phases showed a well-defined electron density for R A . The model was rebuilt manually using Turbo-Frodo and this was followed by iterative cycles of structure refinement using CNS 1.2 (22). The final refinement for each chain converged to R and R free values of 23.5% and 28.5%, respectively with good geometry ( Table I). The final model contained residues 108 to 244 and evaluated using PROCHECK (Table I) (23). Data collection and refinement for cAMP-bound R A were carried out as for apo R A and details are shown in Table I.
Crystallization, Data Collection, Structure Solution, and Refinement of Rp-Bound PKA R A -The purified R A :C holoenzyme was concentrated to 9 mg/ml using Vivaspin concentrators (Sartorius Stedim Biotech GmbH, Goettingen, Germany). A 2 mM aliquot of AMP-PNP was added and incubated for 3 days to allow for intrinsic ATP hydrolysis (24) and complete displacement of hydrolyzed ADP, to ensure saturation with the nonhydrolysable ATP analog (AMP-PNP). This was followed by overnight incubation with 1 mM Rp. Integrity of the holoenzyme was confirmed by analyzing the sample by size-exclusion chromatography (data not shown).
Crystal screenings were performed at room temperature with Hampton and JB screens by the hanging drop vapor diffusion method. Reproducible crystals were obtained after 3 days in the JB 6 (D6) condition (100 mM 2-(N-morpholino)ethanesulfonic acid (pH 6.5), 2 M ammonium sulfate). Crystallization of PKA R A :C holoenzyme in the presence of 0.2 mM AMP-PNP and Rp generated crystals of R Abound to Rp alone. In both cases there was no trace of the C-subunit in the crystal. We again suspect the high salt in the crystallization conditions resulted in dissociation of the C-subunit. Prior to data collection, crystals were briefly soaked in a cryoprotectant consisting of paratone and mineral oil (1:1 ratio) and flash frozen in liquid nitrogen. A complete dataset was collected on a Bruker Platinum 135 CCD detector mounted on a Bruker MicrostarH x-ray generator. The dataset was processed and scaled using the HKL2000 program (25) Matthew's coefficient confirmed that the crystal was that of Rpbound R A and not that of Rp-bound to the R A :C holoenzyme. The presence of Rp (and not cAMP or cGMP) was unequivocally confirmed, based on bromine density at the C8 position of Rp (Fig. 3). The structure was solved by molecular replacement, using the Molrep program (21) and the R A coordinates of the holoenzyme complex from the PDB file 3FHI (8) as the search model. Interestingly, the coordinates of the holoenzyme complex gave better solutions than when the coordinates of cAMP-bound R AB (PDB: 1RGS) (15) were used. Automatic model building was achieved with Arp/wARP (26) and manual adjustment was made with Tomfrodo (27). Alternating cycles of model building and refinement, using the CNS program (22), were performed until the final R-factor and R-free dropped to 25% and 32% respectively. A PROCHECK analysis did not show any residue in the disallowed regions of the Ramachandran plot. Separately, we also set up crystallization of Rp-bound R A as follows. R A was purified and concentrated to 6 mg/ml. This protein was incubated with 1 mM Rp and screened for crystals. JB 5 condition (20% polyethylene glycol 8K, 100 mM 2-(cyclohexylamino)ethanesulfonic acid, pH 9.5) yielded crystals that we used for data collection, structure determination and refinement as described above.    (32,33) and searched against sequence of R A with no enzyme specified and no modifications of amino acids. Identifications were only considered if they appeared at least twice out of three replicate runs. The precursor ion mass tolerance was set at Ͻ10 ppm and fragment ion tolerance was set at Ͻ20 ppm. Only those peptides that satisfied the above criteria through Database search pass 1 were selected and are listed in Table III (32). The default criteria for false positive identification (Value ϭ 4) was applied. A decoy database containing the reverse sequence of R A was created. No peptide identifications were found to be assigned to the decoy sequence and that of the C-subunit of PKA when the above default parameter for false positive identifications was applied. Increasing this parameter increased the number of identifications seen for the decoy sequence underscoring the stringency of sequence identifications with the default parameter setting. It should be noted that MS E does not produce quadrupole isolated tandem MS (MS/MS) spectra, and hence they are not optimal for submission to traditional search engines. Therefore to further confirm the identity of the pepsin fragments, we also carried out independent data dependent analysis of an undeuterated sample of R A . The MS/MS files generated were converted to .pkl files and searched against the R A sequence along with the decoy sequence used above with the Mascot search program (v2.1.0). A total of 14 peptide identifications corresponding to MOWSE scores Ն20 are listed in Supplemental Table I. This threshold MOWSE score was chosen because ions with scores of 12 and below also appeared in the decoy sequence. These results showed that MS E data searched with PLGS 2.4 maximized identification of peptides and were used for deuterium exchange analysis. These identifications were mapped to subsequent deuteration experiments using prototype custom software (HDX browser, Waters, Milford). Data on each individual peptide at all time points were extracted using this software, and exported to HX-Express (34) for analysis. A total number of 37 peptide fragments yielded primary sequence coverage of 93%.

Amide Hydrogen/Deuterium (H/D) Exchange Mass Spectrometry-
Continuous instrument calibration was carried out with Glu-fibrinogen peptide at 100 fmol/l. We also visually analyzed the data to ensure only well-resolved peptide isotopic envelopes were subjected to quantitative analysis. The lowest signal to noise ratio among all analyzed peptides was four. A control experiment was carried out to calculate the deuterium back exchange loss during the experiment by incubating ligand-free R A with deuterated buffer A for 24 h at room temperature. All reported deuterium exchange values were corrected

FIG. 2. Structures of apo and cAMPbound R A in the B-conformation and Rp-bound R A in the H-conformation.
A, Structure of apo R A in the B-conformation showing the helical ␣-subdomain in brown and ␤-subdomain in tan. B, Structure of the cAMP-bound R A also in the B conformation is completely superimposable with apo R A except for cAMP bound to the PBC pocket. C, Structure of Rp-bound R A in the H-conformation showing the helical ␣-subdomain in smudge green and ␤-subdomain in tan. Structures determined independently from crystals of Rp bound to R A :C holoenzyme or to apo R A were identical and in the H-conformation.  (19).
Gas Phase Protein Structure Measurement by Ion Mobility Mass Spectrometry-Gas phase measurements of collisional cross-sectional area (CCS) were conducted on a SYNAPT HDMS Mass Spectrometer (Waters, Manchester, UK). Ion mobility measurements were carried out on native conformations using a traveling wave ion mobility method as described (35). Briefly, samples were infused from a solution (0.3 mg/ml in 200 mM ammonium acetate, pH 7.2) of apo, cAMP and Rp-bound RI␣(91-379)(henceforth referred to as R AB )-(cAMP, Rp at 1 mM final concentrations) into the source region using borosilicate glass nano-electrospray tips (Waters, Manchester, UK). Operating and separation voltages were optimized for the experiment, ensuring that no measureable loss of conformation was produced from the instrument (data not shown). CCS measurements and calibration were carried out using Driftscope 2.1 (Waters, Manchester, UK) working with a modified version of an existing protocol (36) by infusing 0.2 mg/ml myoglobin in a 50% methanol solution containing 0.1% formic acid, in reference to previously published CCS values (http://www.indiana.edu/ϳclemmer/Research/research.htm).
For comparison, coordinates from Protein Data Bank (PDB) files for apo and protein conformations, based on x-ray crystallography data were used to generate theoretical CCS measurements using the open source code MOBCAL (37), (http://www.indiana.edu/ϳnano/Software.html).

RESULTS
So far, structures of several end-point conformations of RI␣ have been solved by x-ray crystallography. These include structures of cAMP-bound R AB (15), C-subunit-bound R A (8), and C-subunit-bound R AB (13). These structures have provided molecular insights into the cAMP binding pocket and R-C interface, as well as provided details of the distinct B and H-conformational states. However a structural basis for Rp functioning as an inverse agonist and mechanistic details of cAMP-dependent activation of PKA, specifically by mediating conversion of the holoenzyme (H-form) to the cAMP-bound (B-form) conformation are still unclear. Also unclear is how apo RI␣ binds both cAMP as well as the C-subunit with equally high affinity. We believe that the apo as well as Rpbound states represent critical intermediates in the interconversion of B and H-forms and mapping the structure and dynamics of these two states is critical to understanding the overall mechanism of cAMP action. Therefore we set out to solve the structures of the apo and cAMP-bound R-subunit fragment R A as well as Rp-bound R A :C.
Structures of apo and cAMP-Bound R A -In our 2.7 Å structure, a single apo R A molecule occupies the asymmetric unit and only residues 108 to 243 could be traced, indicating high disorder in the N-terminal 17 residues spanning the pseudosubstrate site, as observed in the x-ray structure of cAMPbound R AB (15) (PDB:1RGS). Furthermore, the overall topology of the molecule is identical to the cAMP-bound conformation (B-form), indicating that the apo form ( Fig. 2A) is completely superimposable with the cAMP-bound form. However the structure of apo R A is highly dynamic as it is characterized by high B-factor values for the main chain atoms throughout the molecule (Supplemental Fig. 2A).
The structure of cAMP-bound R A was solved using molecular replacement with cAMP-bound R AB (PDB: 1RGS) as the input model and this structure is highly similar to cAMPbound R AB (15) (Fig. 2B). The asymmetric unit of this 1.5 Å resolution structure contains two molecules. Interestingly, absence of the aromatic capping residue Trp 260, which is missing in this deletion mutant, is compensated by stacking interactions mediated by Tyr 188 from an adjacent molecule in the asymmetric unit.
Structure of R A Bound to Rp-Although the two superimposable structures of apo and cAMP-bound R A show differences in temperature factors (Supplemental Figs. 2A and 2B), there is no clear explanation to address the 1000-fold difference in affinity between the cAMP-bound and apo R-subunit for the C-subunit (20). Based on our recent report that Rp binding to R A leads to a higher affinity for the C-subunit (19) wherein it functions more as an inverse agonist, we next set out to solve the structure of the PKA holoenzyme bound to Rp, which would represent an intermediate in cAMPmediated activation. Despite using a nonhydrolyzable ATP analog (8), the obtained crystals were those of Rp-bound to R A alone. Dissociation of the C-subunit presumably could have resulted from the high ionic strength in the crystallization buffer (38). Rp-bound R A crystallized in the trigonal space group P3 1 21. The structure was solved by molecular replacement using the conformation of the R A in the holoenzyme complex (8) (PDB: 3FHI). The identity of the Rp ligand was confirmed by electron density for bromine at the C8 position. Rp binds in the syn conformation (Fig. 3) and the structure of Rp-bound R A is identical to the C-subunit bound conformation (Hform) but not to the B-form (Supplemental Figs. 1 and 2). This was entirely unexpected and indicated that the basis for Rp functioning as an inverse agonist for PKA was because of its inability to induce a conformational change from the H to the B form (Fig. 2C). To test whether Rp functions as an inverse agonist only upon binding to the holoenzyme (preformed Hconformation), we crystallized apo R A bound to Rp in the absence of C-subunit. The structure of Rp-bound R A solved using these crystals was identical to that obtained with crystals of Rp-bound R A generated from cocrystallization of R A :C and Rp. These revealed that Rp-bound R A crystallized in the H-conformation (with Rp in the binding pocket) regardless of whether the C-subunit was absent (apo form) or present (holoenzyme; H-form).
Structural Differences Between Rp-Bound R A and apo, cAMP-and C-Subunit-Bound States-Superposition of the structures of Rp-bound R A with that of R A :C (8) shows a lower RMSD (0.910 Å) (Supplemental Fig. 1A) compared with the superposition with cAMP-bound R A (2.858 Å) (Supplemental Fig. 1B). This reflects that the Rp-bound R A assumes the H-form. More details are summarized below.
␤-Subdomain-Rp is anchored within the PBC at the center of the ␤-barrel, by a similar network of contacts found in cAMP-bound R A (Table II, Fig. 3). Distances between the sulfur of Rp and critical residues (Ala 202 and Arg 209) of the PBC pocket are different from those seen for the equatorial oxygen of cAMP (Fig. 4). A particularly large increase in distance (ϳ 0.7 Å) is seen between the sulfur of Rp and the backbone amide nitrogen of Ala 202, a critical residue along with Arg 209 responsible for anchoring the cyclic phosphate of cAMP (15,39). In contrast, the distances between the guanidinium moiety of Arg 209 and backbone carbonyl oxygen of Gly 199 are shorter, reflecting a stronger bond among these atoms in the Rp-bound structure.
Although, overall topologies of the PBC in the Rp-free and bound states are the same, a comparison of the communication networks in these structures reveal important differences. These are primarily seen in the PBC and ␤-strands 2 and 3.
Residues 161-163, which are part of ␤7 in the structure of the R A :C holoenzyme, are disordered in the structure of Rpbound R A . This reflects C-subunit-induced conformational changes in the R-subunit. In addition to the above residues, the carbonyl group of the highly conserved Gly 169, which is critical for the cAMP mediated activation of PKA (40), makes a H-bond with Arg 226 in both the R A : Rp and R A :C structures. ␣-Subdomain-The conformationally mobile region of the ␣ subdomain, the ␣:B/C helix region that forms part of the C-subunit-binding surface, adopts the H-conformation in R A :Rp structure.
Amide Hydrogen/Deuterium (H/D) Exchange Mass Spectrometry Analysis-To further probe conformational dynamics of the R-subunit, we carried out amide H/D exchange mass spectrometry experiments of apo, cAMP-, Rp -, Sp-and C-subunit bound R A . Earlier studies on the H/D exchange of the apo, cAMP-bound and holoenzyme states had shown clear evidence of allostery between the two subdomains of the protein (11).We were interested in extending this analysis to cover the two thio-substituted analogs. We set out to use electrospray ionization-quadrupole time-of-flight MS to measure the amide exchange, which yielded primary sequence coverage of ϳ93%.
Our amide H/D exchange results indicate that apo R A is a highly dynamic molecule that shows high rates of amide deuterium exchange across the entire molecule, which directly correlate to the B-factor values of the backbone atoms (Fig. 5, Supplemental Fig. 2). cAMP binding causes a large decrease in amide exchange, suggesting higher ordering of several residues both within the PBC as well as in peripheral regions of the molecule. A nearly identical effect was also observed for Sp-bound R A (Fig. 5, Table III). Although Rp binding did cause greater ordering of the molecule relative to the apo protein, there were significant differences in exchange at multiple regions, compared with the cAMP/Sp-bound proteins (Fig. 5). Most importantly, at the PBC, decreased exchange in the cAMP and Sp-bound states and intermediate exchange between the apo and cAMP-bound states was observed upon Rp binding (Fig. 5F). Within the PBC, the apo R A and R A :C showed highest amide deuterium exchange, consistent with the absence of ligand whereas cAMP and Sp-bound proteins showed lowest exchange. Interestingly, peptides within this region in Rp-bound R A showed increased exchange, compared with other ligand bound proteins (Fig.  5F), reflecting potentially weaker interactions of Rp with Arg 209 and Ala 202. Subtractive analysis of the overlapping peptides showed that there were no significant differences at residues 203-206 as well as 213-217 between the Rp-bound and other ligand-bound states indicating that the most significant differences could be localized to peptides spanning residues Glu 200 and Ala 202 on one hand and Arg 209 and Ala 210 on the other. Our results also show that amide deuterium exchange at the N-terminal pseudosubstrate region (92-102) of R A :C is lower than in the ligand-bound states (Fig.  5C), consistent with this region binding to the substrate binding cleft of the C-subunit. It is interesting to note that the highly disordered conformation of the pseudosubstrate region and N terminus appear to be independent of the B or H-conformational states of the cAMP-binding domain.
Crucial differences between the Rp-bound and cAMP/Spbound states are evident in the PBC. This is completely consistent with the differences in H-bonding distances seen in these residues from the crystal structures described in Table II pared with Sp/cAMP-bound R A ) and a second region from 207-212 which includes important residues such as Arg 209 and Ala 210 that shows increased exchange of ϳ1.6 deuterons in Rp:R A compared with the other ligand-bound states. Overall, Rp-bound R A shows significantly increased exchange across the entire PBC pocket reflecting a more dynamic occupancy of the PBC by Rp compared with Sp or cAMP.
Ion Mobility Mass Spectrometry-Although apo R A crystallized in the B-conformation, it still did not provide mechanistic insight into the basis for the 1000-fold difference in binding affinity for the C-subunit seen between the cAMP-bound and apo states of the protein. Amide H/D exchange did however reveal that the apo state was highly dynamic and distinct from the ligand-bound states. To further highlight

FIG. 5. Amide H/D exchange mass spectrometry shows that apo R A is highly dynamic and highlights clear differences in deuterium exchange in R A between Rp and other ligand-bound states.
A, ESI-Q-TOF mass spectra for a peptide spanning residues R A (202-212) (m/z ϭ 567.32(2)) comparing amide exchange in the apo, ligand-bound and C-subunit-bound states. (i) Undeuterated apo R A . (ii) The isotopic envelope for the same peptide from Sp-bound R A following 10 min deuteration (iii) The isotopic envelope for the same peptide from cAMP-bound R A following 10 min deuteration; (iv) The isotopic envelope for the same peptide from Rp-bound R A following 10 min deuteration; (v) The isotopic envelope for the same peptide from R A :C following 10 min deuteration; (vi) The isotopic envelope for the same peptide from apo R A following 10 min deuteration. B, Time course of deuterium exchange at residues (202-212). Open circle (E), apo R A ; Closed circle (•), cAMP-bound R A ; Open diamond (छ), R A C complex; closed triangle (OE), Sp-bound R A ; Closed Square (f), Rp-bound R A . C, Time course of deuterium exchange at residues 92-102. Symbols are as in Fig. 5B. D, Structure of apo R A highlighting regions showing enhanced deuterium exchange in the absence of ligands relative to cAMP-bound R A , in red. E, Structure of apo R A highlighting regions showing enhanced deuterium exchange in the Rp-bound state relative to Sp/cAMP-bound R A . F, Summary of deuterium exchange data for peptides spanning PBC in ligand-free and ligand-bound states. Overlapping nested peptides obtained greatly improve data resolution for the PBC region. The PBC region in R A : Rp is more solvent accessible when compared with all other ligand bound states of the protein.

cAMP Antagonist Promotes Selection and Preferentially Binds Inactive of Protein Kinase A
the differences in conformation between apo and ligandbound states, we carried out ion mobility mass spectrometry studies to specifically probe differences in molecular shape between all the above states. Theoretical collisional cross-sectional area (CCS) measurements of the cAMPbound (B-conformation) and C-subunit-bound R A (H-conformation) from the structural coordinates of these two conformations predicted only small (Ͻ5% changes in CCS) changes that were below the resolution of our Synapt HDMS mass spectrometer. However, the changes predicted for the larger deletion construct encompassing both CNB-A and CNB-B, R AB , between the two conformations were much greater (ϳ18%) and hence, we set out to determine experimental values for CCS for apo and ligand-bound states of R AB .
Three to four charge states (z ϭ ϩ9, ϩ10, ϩ11, and ϩ12) were observed (Fig. 6A). A comparison of the drift times observed showed distinct differences between the Rp and Sp-bound states with the Sp-bound state showing a shorter drift time indicative of a more compact molecule compared with the Rp-bound state. This is entirely consistent with theoretical CCS measurements of the B and H-conformations with the H-conformation being larger (Fig. 6B). Interestingly, apo R AB shows a drift time distribution that spans both Sp and Rp-bound states indicating that this represents an ensemble of multiple conformations encompassing both the B and Hconformational states (Fig. 6C).

Conformational Selection in the R-Subunit: Rp Stabilizes
Inactive H-Conformation-Conformational dynamics forms the molecular basis for the functioning of all signaling proteins. The ability of regulatory proteins to toggle between multiple conformations in response to ligand binding, other proteins or post-translational modifications forms the basis for signal transduction. Recent experimental evidence has allowed postulation of a "conformational selection" hypothesis, which suggests that the ability of proteins to exist in ensembles of multiple conformational states is an inherent property of the protein (41). This introduces an additional dimension to earlier models that suggested binding events with extraneous proteins or ligands or post-translational modifications might be solely responsible for inducing conformational changes in proteins.
From our solution studies we demonstrate that the ability of the apo form of the R-subunit to mediate mutually exclusive high affinity interactions with cAMP or C-subunit depends on its ability to exist in an ensemble of multiple conformations. This has been demonstrated clearly in the CNB from Epac by elegant NMR experiments where it has been estimated that the protein existed in an 80:20 distribution of B and H-forms (42). Although hints of such an ensemble were obtained from our crystallographic studies, amide exchange results show that the apo protein displayed the greatest amide deuterium exchange among all the states of the R-subunit tested. This suggested that the apo form of the R-subunit was conformationally highly dynamic and metastable. Our ion mobility mass spectrometry results show the apo R-subunit with a distinct profile for the collisional cross-sectional area that straddles the profiles of the cAMP-bound and Rp-bound states (Fig.  6C). This provides clear evidence that the apo R-subunit is indeed an ensemble of multiple conformational states and encompasses the two end-point B and H-conformations and is completely consistent with elegant NMR experiments comparing dynamics of Rp and Sp-bound R A (43). Binding of cAMP or C-subunit would then occur through 'conformational selection' where cAMP, which is a natural ligand, specifically binds to and selects the B-conformation whereas the C-subunit binds preferentially to the H-conformation. This clearly explains the 1000-fold difference in affinity between apo and cAMP-bound R-subunit for the C-subunit. Thus the higher affinity of apo R-subunit for the C-subunit is entirely because of "conformational selection" whereas C-subunit binding to cAMP-bound R-subunit (B-conformation) requires bindinginduced dissociation of cAMP for stable holoenzyme complex formation; "induced fit." This is supported by the observation that Rp functions more as an inverse agonist rather than a cAMP antagonist in the activation of PKA (19). It is interesting that a single thio substitution in Rp enables it to bind selectively to the H-form (Fig. 2C) in contrast to the cAMP-favored B-form (Fig. 2B). This indicates that the sulfur substitution at the equatorial oxygen is uniquely important for determining the preferred conformation that a ligand can bind. In summary, apo R-subunit is uniquely poised to interact with all ligands and the C-subunit via conformational selection whereas interconversion of stable end-point conformations as seen in the cAMP-dependent activation of PKA holoenzyme is through induced fit (41).

Mechanism of cAMP Action and Basis for Antagonism of
Rp-Although conformational selection describes the ability of Rp and cAMP to preferentially select the H-inactive and B-active conformations respectively, comparison of structures of the two states highlights the unique binding interactions of Rp relative to cAMP. Our structure of cAMP-bound R A clearly shows the important contacts between the specific moieties of cAMP and residues within the PBC. These provide high specificity for cAMP binding to the conserved PBC pocket (2,15). The important contacts include the anchoring of the 2ЈOH by both the side chain of Glu 200 and the backbone nitrogen of Gly 199. The axial oxygen of cAMP is anchored by a single contact with the backbone amide nitrogen of Ala 210. The adenine ring mediates stacking interactions with Trp 188 from a neighboring molecule in the asymmetric unit while in solution, stacking interactions have been proven to be mediated with Tyr 244 (44). All the above interatomic distances are nearly identical between structures of cAMP-bound and Rp-bound R A ( Fig. 1 and Fig. 4). The importance of all these contacts for binding interactions have been confirmed by mutagenesis (7) and by the observation that 2Јdeoxy-cAMP is incapable of binding PKA R-subunit (45). These are listed in Table II and shown as black dashed lines in Fig. 4 and represent contacts important for binding and occupancy of PBC by cAMP and Rp. However, the distance between the equatorial sulfur and residues Ala 202 and Arg 209 are significantly different in Rp and is summarized in Table II and the H-bonds are represented by red dashed lines in Fig. 4. These confirm the importance of the equatorial oxygen not only for anchoring of cAMP to the pocket through critical interactions with the backbone nitrogen of Ala 202 and the guanidinium side chain of Arg 209, but also for allosteric communication through the above residues to activate PKA.
It is clear that the mechanism of cAMP action is through induced fit and involves binding to PBC followed by allosteric conformational changes. A combination of our x-ray crystallography and solution mass spectrometry results clearly indicates that whereas all moieties including the equatorial oxygen are important for binding, the equatorial oxygen is uniquely important for mediating allosteric relays necessary for propagation of the effects of cAMP binding. This is entirely consistent with this being the critical atom for determining whether preferred binding exists for the B or H-conformations. Details of the allosteric networks of residues leading from the equatorial oxygen are described below.
Comparison of our cAMP-bound and Rp-bound structures of R A confirm the role of specific allosteric communication relays, proposed to propagate the effects of cAMP binding throughout the molecule to facilitate dissociation of the Csubunit. Amide exchange results further show very clearly that substitution of the equatorial oxygen by sulfur in Rp increases deuterium exchange within two specific regions of the PBC. These are the regions spanning Glu 200, Ala 202, and 207-212, spanning Arg 209. A network of conserved residues from cAMP Antagonist Promotes Selection and Preferentially Binds Inactive of Protein Kinase A the PBC that have been demonstrated to be important for cAMP-mediated allostery of PKA are those A) between the hydrophobic core region of the PBC and the hinge region formed by the highly important ␣:B-C helices (46) and B) between Arg 209 of PBC with Asp 170 from strands ␤1-␤2 (19); (40) (Fig. 4). The hydrophobic core of the PBC includes conserved residues, Leu 201, Ala 202, Leu 203 and Ile 204 responsible for communication with the ␣:B-C region via the Tyr 229 residue. This has been demonstrated in Epac2 by mutagenesis experiments to be an important conformational switch between the inactive H and active B-conformations (46) as well as NMR (47). The extensive conformational changes in this region in our structure of Rp-bound R A and differences in amide exchange between cAMP-bound and Rp-bound states clearly validates the previous results.
Clear differences in the position of Asp 170, critical for a second relay via Arg 209, are seen between our structures of the cAMP-bound and Rp-bound R A (Figs. 1 and 4). Previous mutagenesis (48) as well as structural comparison of the B (15) and H-conformational states (8) are consistent with our results.
Our findings confirm that the equatorial oxygen is critical for cAMP action and this is borne out by the important contacts that this atom mediates with residues of the PBC. This oxygen is positioned precisely to coordinate and bridge the interactions of the 2ЈOH of the ribose moiety together with the axial oxygen that are essential for binding, with the allosteric relays listed above. The basis for cAMP action is therefore a combination of stabilization of the PBC followed by allosteric relays propagated from the equatorial oxygen of the cyclic phosphate.
The S-P bond in phosphorothioates is more electronegative than the O-P bond of cAMP, making Rp a stronger acid relative to cAMP (49 -51). This reduces the strength of Hbonding of the sulfur with both Ala 202 as well as Arg 209 (Table II, Fig. 4) leading to disruption of both allosteric relays. Substitution of the axial oxygen in Sp on the other hand has no effects on allostery. This is completely consistent from both amide exchange and x-ray crystallography, whereby increased deuterium exchange together with the increased interatomic distances seen in the crystal structure confirm that Rp binding to PBC is more dynamic than Sp/cAMP. Rp thus preferentially binds the inactive-H conformation through conformational selection while satisfying all the binding constraints associated with the "active-B" conformation.
Identification of Highly Selective Allosteric Inhibitors that Specifically Bind and Stabilize "Inactive" Conformations-Our results provide the following major insights into cAMP-dependent regulation of PKA. First, we demonstrate that the apo R-subunit is equally capable of binding cAMP or C-subunit through conformational selection although the dominant conformation is likely to be the H-form (42). This is because of the apo state existing as an ensemble of multiple conformations. This also explains its higher affinity for the C-subunit com-pared with cAMP-bound R-subunit. Second, the molecular basis for cAMP action is now better understood as occurring via binding interactions with PBC followed by specific allosteric conformational changes mediated solely by the equatorial oxygen of the cyclic phosphate. Thirdly, our results demonstrate the basis for antagonism of Rp. Substitution of the equatorial oxygen by sulfur alters the preferred conformation for binding to the H-form and disrupts all the allosteric conformational relays while retaining binding interactions necessary for docking to the PBC.
The above studies of the R-subunit provide insights into "rational" drug design. First it is important to obtain structural information of more than a single end point conformation of the target protein of interest. Complementary solution studies are equally essential to define the conformational ensemble behavior of the target in solution in the absence of effector proteins/ligands. Because the goal of inhibitor selection is to design ligands that maintain the target in an inactive conformation, it is necessary to distinguish between contacts on the protein necessary for binding from those important for inducing allosteric conformational changes. The ideal inhibitors are therefore those that satisfy all "binding" constraints without inducing conformational changes necessary for activation. This principle is evident from the synthesis of specific inhibitors such as Gleevec for Abl-kinase which specifically recognize and stabilize the inactive conformation and is a classic example of conformational selection. These classes of conformation-specific inhibitors show better target-specificity and are therefore more desirable (52). In our study, the key to how Rp functions as a mechanistic inhibitor of PKA, is that it too preferentially binds the inactive H conformation of the R-subunit as it contains all the important binding moieties yet it fails to induce allosteric conformational changes.
Our analyses also highlight and support increasing evidence emerging for the importance of protein flexibility (which allows conformational selection) in rational drug design. In addition to crystallographic snapshots, conformational dynamics of proteins in solution lead to better identification of small molecule disruptors of protein-protein interfaces, especially for proteins that tend to exist as conformational ensembles.
ACCESSION NUMBER: Coordinates of apo-R A , cAMP-R A and Rp-cAMPS-R A were deposited in the Protein Data Bank (http://www. pdb.org) under accession codes 3IIA, 3PNA and 3PLQ, respectively.