cAMP-dependent protein kinase (PKA) complexes probed by complementary differential scanning fluorimetry and ion mobility–mass spectrometry

cAMP-dependent protein kinase (PKA) is an archetypal biological signaling module and a model for understanding the regulation of protein kinases. In the present study, we combine biochemistry with differential scanning fluorimetry (DSF) and ion mobility–mass spectrometry (IM–MS) to evaluate effects of phosphorylation and structure on the ligand binding, dynamics and stability of components of heteromeric PKA protein complexes in vitro. We uncover dynamic, conformationally distinct populations of the PKA catalytic subunit with distinct structural stability and susceptibility to the physiological protein inhibitor PKI. Native MS of reconstituted PKA R2C2 holoenzymes reveals variable subunit stoichiometry and holoenzyme ablation by PKI binding. Finally, we find that although a ‘kinase-dead’ PKA catalytic domain cannot bind to ATP in solution, it interacts with several prominent chemical kinase inhibitors. These data demonstrate the combined power of IM–MS and DSF to probe PKA dynamics and regulation, techniques that can be employed to evaluate other protein-ligand complexes, with broad implications for cellular signaling.


Introduction
Protein phosphorylation is a reversible post-translational modification (PTM) catalyzed by the protein kinase-mediated enzymatic transfer of ATP γ-phosphate to an appropriate side chain in a substrate [1]. A plethora of techniques, including X-ray crystallography, protein nuclear magnetic resonance (NMR), enzyme kinetics and (more recently) molecular dynamics, differential scanning fluorimetry (DSF) and mass spectrometry (MS), have been exploited to dissect enzymology, supramolecular structure, allosteric co-operativity and nucleotide/ligand-binding propensity within protein kinase complexes [2][3][4][5][6][7][8]. Concomitantly, many members of the protein kinase superfamily have become important therapeutic targets, in part due to their inherent druggability [9], but also because they play rate-limiting roles in many human diseases, including inflammation, heart disease and cancer [10]. Protein kinases are ubiquitous and evolutionary related, possessing canonical amino acid motifs that are positioned at the catalytic interface, trapping divalent metal cofactors and positioning protein substrate directly adjacent to the ATP phosphoryl donor, which is thought to bind prior to substrate in the catalytic cycle [11,12]. This conserved mode of ligand binding implies that the development of specific probe compounds to interrogate protein kinase catalytic function is a challenge, and much remains to be learnt about the chemical and protein ligand sensitivity of 'open' and 'closed' (lying between 'inactive' and 'active') conformational states that exist within dynamic populations of kinase signaling complexes [13]. Although considerable progress has been made in the synthesis and exploitation of kinase-specific ligands, some of which employ allosteric (non-ATP)-binding modes, the vast majority of clinically approved compounds interact with the nucleotide-binding site.
DSF of PKAc reveals discrimination of compounds and ligands with affinity for catalytically 'active' and 'inactive' PKAc variants. In addition, we provide IM-MS-based evidence for conformation-specific binding of PKI to PKAc, and demonstrate native stoichiometric and sub-stoichiometric PKA complexes. Finally, a dramatic ablation of R 2 C 2 tetrameric complexes by PKI, with preferential formation of a stable PKA:PKI heterodimer, is established for the first time. Together, our work represents a new approach to study macromolecular PKA complexes that can be readily extended to other kinase complexes and ligands in vitro.

Phosphopeptide enrichment and LC-MS/MS
Protein samples were reduced and alkylated prior to overnight digestion with trypsin (2% w/w) and subsequent phosphopeptide enrichment using Titansphere Phos-TiO 2 spin tips (GLSciences) with minor modification of the manufacturer's instructions (see Supplementary Material). Tryptic peptides were analyzed by LC-MS/MS before and after TiO 2 -based phosphopeptide enrichment. nLC-electrospray ionization (nESI)-MS/MS analysis was performed using either an Orbitrap Fusion Tribrid mass spectrometer (ThermoScientific) attached to an Ultimate 3000 nanoLC system (Dionex) or an AmaZon ETD ion trap (Bruker Daltonics) arranged in-line with a nanoAcquity n-UHPLC system (Waters). .mgf files were searched using MASCOT (version 2.1) against the E. coli IPI database (downloaded 24 March 2015) with the complete sequence of 6His-tagged PKAc included. Carbamidomethylation of Cys was set as a fixed modification; phosphorylation of Ser and Thr, oxidation of Met and deamidation of Asn and Gln were set as variable modifications. The tandem mass spectra for all identified phosphopeptides were interrogated manually.
Dephosphorylation of PKAc with GST-λPP and PKA kinase assay PKA (60 mg; ∼13 mM, 100 ml final volume in 50 mM NH 4 OAc buffer) was incubated at 37°C with 100 ng of λPP in the presence of 1 mM Mn(OAc) 2 for 2 h. The reaction was stopped by buffer exchange into 50 mM NH 4 OAc. For kinase assays, 1 ng of purified PKAc, or R133A PKAc was assayed in the presence and absence of recombinant full-length His-tagged PKI using a Caliper EZ Reader II enzyme assay platform in 25 mM Hepes ( pH 7.4), 1 mM ATP, 5 mM MgCl 2 , 1 mM DTT and 5 mM fluorescent 5-Fam-Leu-Arg-Arg-Ala-Ser-Leu-Gly-CONH 2 (Kemptide). Activity (kinetic mode) was reported as % peptide phosphorylation after ∼10 cycles.

Detection of PKAc and PKA RIIα binding by SEC
Equimolar concentrations (∼2.5 μM) of catalytic (C; WT or K72H) and RIIα (regulatory) PKA subunits were preincubated at 4°C for 2 h and loaded onto a Superdex 200 10/300 GL column (GE Healthcare) equilibrated in 50 mM Tris-HCl ( pH 7.4), 100 mM NaCl. Eluted fractions (0.5 ml) were collected, and complex formation was analyzed by SDS-PAGE and Coomassie blue staining. Binding was confirmed by co-elution of the C and RIIα subunits from the column.
Ion mobility-MS and collisional activation IM-MS analysis was performed on a Waters Synapt G2-Si instrument. All proteins were buffer exchanged into 50 mM NH 4 OAc (LC grade, Sigma) employing Amicon spin filter columns (10 kDa molecular cutoff ). Typically, 1-3 ml of 2-5 mM sample was analyzed using borosilicate emitters (Thermo ES 387). Spraying voltage was adjusted to 1.1-1.8 kV and the sampling cone was 20 V. Pressure in the travelling wave (T-wave) ion mobility cell was 2.78 mbar (nitrogen); wave height was kept at 30 V and wave velocity at 750 m/s. For collision-induced unfolding (CIU) experiments, a single charge state was quadrupole isolated and subjected to collisional activation by applying CID activation in the ion trap of the instrument. The activation voltage was varied between 25 and 41 V. The ion mobility wave height was 40 V and the wave velocity was 650 m/s. Data were processed with MassLynx 4.1 and OriginPro 9.0 and used to generate contour plots of the intensity of ion populations as a function of collision cross section (CCS) and collision voltage. Noncovalent interactions were studied by mixing the corresponding components in 200 mM NH 4 OAc and equilibrating for at least 10 min at room temperature prior to MS analysis.

CCS calibration and theoretical calculations
To experimentally determine CCS, the measured drift time through the T-wave mobility cell of β-lactoglobulin A (Sigma L7880), avidin (Sigma A9275), transthyretin (Sigma P1742), concanavalin A (Sigma C2010) and serum albumin (Sigma P7656) was calculated according to standard protocols [54]. The exact hard sphere scattering (EHSS) model implemented in the Mobcal software was used to calculate CCS values on the basis of X-ray structures [55].

Analysis of purified recombinant proteins
We expressed and purified to homogeneity a panel of proteins that can be used to reconstitute multimeric or holoenzyme (R 2 C 2 ) PKA signaling complexes in vitro. We also generated mutations in the catalytic subunit that have previously been shown to selectively prevent autophosphorylation and abrogate catalytic activity [56] ('kinase-dead' K72H PKAc mutant) or that specifically reduce binding to the RIIα subunit and the heat-stable inhibitor of PKA (PKI), while sparing autophosphorylation and catalytic activity [21,57] (R133A PKAc mutant). As shown in Figure 1, these proteins were pure, and SEC and DSF (see below) suggested that all proteins were folded and monomeric. CD also confirmed similar secondary structures in both WT and K72H PKAc proteins when analyzed in the presence and absence of ATP and Mg 2+ ions, the latter inducing no major changes in the spectra. A key goal of our study was to determine the effects of phosphorylation on PKA complex formation in vitro. Using peptide-based tandem MS analysis, we initially determined the sites of phosphorylation on PKAc proteins, confirming that it was highly phosphorylated on several known sites, including Ser10, the key T-loop (Thr197) and C-terminal tail motif (Ser338) sites, these being required to generate a stable, catalytically active PKAc [26,58,59]. Interestingly, our comprehensive sequence coverage revealed a total of 11 sites of phosphorylation in recombinant WT and 11 phosphosites on R133A PKAc, all of which were absent from K72H PKAc preparations (Table 1 and Supplementary Material), confirming that they are autophosphorylation sites.

Phosphorylation regulates PKA conformation
To assess effects of phosphorylation on the structure and intrinsic stability of the PKA catalytic subunit, recombinant intact autophosphorylated PKAc was analyzed by MS before and after treatment with λPP ( Figure 1). We observed an average of ∼7 autophosphorylation sites in WT PKAc ( Figure 1A), out of a total of 14 sites of phosphorylation that had been mapped (Table 1; Supplementary Tables S1-S3 and Figure S1); these identified sites of modification include three nonphysiological autophosphorylated residues found within the 6His affinity tag, whose nonspecific modification by basophilic kinases was noted in previous work [60]. Importantly, of the 11 phosphosites observed in the core PKAc polypeptide, we mapped phosphorylation at Ser34, only previously reported in a single high-throughput screen, and discovered two potential novel sites of modification in the kinase N-lobe (Thr37 and Ser65; Supplementary Figure S1). Previous work has demonstrated that recombinant PKAc is rather resistant to dephosphorylation at the T-loop (Thr 197) and C-terminal motif (Ser 338) phosphorylation sites [61,62]. Consistently, incubation with λPP, using either Mg 2+ or Mn 2+ as the divalent cofactor, generated hypophosphorylated forms of PKAc with a reduced average of ∼4 (data not shown) and ∼3 phosphosites per molecule, respectively, with 6 phosphosites mapped by peptide-based tandem MS (Table 1, Figure 1B,C, and Supplementary Table S3). Even after extensive incubation with λPP, complete dephosphorylation could not be achieved, providing us with an opportunity to analyze the conformation of differentially phosphorylated preparations of PKAc by native IM-MS. Crucially, K72H was observed in a completely nonphosphorylated form, consistent with the peptide-based analysis, and R133A yielded an intact phosphorylation phosphoprotein profile almost identical to WT protein ( Figure 1C).
When ESI-MS was performed under nondenaturing ('native' solution) conditions, autophosphorylated PKAc was observed over an extremely narrow charge state distribution, predominantly yielding 13+ and 14+ ions (Figure 2A), indicative of highly ordered tertiary protein structure [63]. IM-MS was subsequently used to determine the CCS distributions and, following calibration of the T-wave ion mobility cell, a TW CCS N2→He value of 29.4 nm 2 for the 12+/13+/14+ charge states of hyperphosphorylated PKAc was determined. As expected, charge-mediated unfolding of the 15+ species yielded multiple conformations with higher CCS values (Supplementary Figure S2). Comparison of CCS for PKAc before and after phosphatase treatment ( Figure 2B,C) revealed phosphorylation-dependent effects on both the absolute TW CCS N2→He value and the half-height width of the CCS distribution (CCSD); CCS (Ω) increased by 1.5% following treatment with λPP, whereas CCSD demonstrably reduced from 2.2 nm 2 (28.4-30.6 nm 2 ) to 1.8 nm 2 (28.8-30.6 nm 2 ). Autophosphorylation thus promotes compaction of PKAc while increasing conformational dynamics. Interestingly, hyperphosphorylated PKAc also exhibited a noticeable asymmetry in the CCSD, not apparent upon phosphatase treatment. This phosphorylation-dependent increase in conformational space is perhaps unsurprising given the number of modified residues identified (Table 1) and the likelihood of nonstoichiometric phosphosite occupancy.
The K72H and R133A variants of PKAc were employed as comparators for further structural investigation of the PKA complex. Of particular interest for understanding the role of phosphorylation on PKAc conformation, stability and ligand-binding capabilities is the K72H PKAc mutant, given its inability to autophosphorylate (Table 1 and Figure 1). Although the potential folding status of this purified mutant cannot be assessed as a function of catalytic output, the near-identical CD spectra obtained for WT and K72H PKAc ( Figure 1B) suggest similar secondary elements of structure in both proteins.
Gaussian fitting of the CCSDs determined by native IM-MS revealed two primary conformers for WT PKAc, with CCS values of 29.0 and 30.0 nm 2 , the larger of which overlaid with the predominant conformer apparent following phosphatase treatment ( Figure 2B). To better understand the conformational stability of these PKAc variants, we performed a series of CIU experiments evaluating the effect on CCSD. By elevating the collision energy (CE) applied to the native-solution derived conformers (but maintaining it below that required for covalent bond dissociation), it is possible to gradually unfold a protein and thereby gain an understanding of its (gasphase) stability and the structure of partially unfolded intermediates [64]. The CIU profile of hyperphosphorylated PKAc was markedly different from that of the phosphatase-treated form of PKAc ( Figure 2C and Supplementary Figure S3). WT-untreated phospho-PKAc is significantly more stable, requiring higher CE to initiate unfolding. While λPP-treated PKA starts to undergo conformational change at as low as ∼31 V, WT PKAc is conformationally stable under these conditions, not undergoing significant unfolding until ∼36 V.
Akin to data obtained at low CE (Figure 2), the conformational space (CCSD) adopted by hyperphosphorylated PKAc at high CE (>40 V) is larger than the phosphatase-treated form. Interestingly, both K72H and In agreement with the CIU profiles, WT PKAc was observed to have greater thermostability, as determined by DSF, than either R133A or K72H ( Figure 3A). Although the T m values at 50% were comparable for WT and

Small molecule and PKI binding to PKA is conformation-specific
In contrast with MS analysis, where electrospray ionization is suppressed by metal ions, a major strength of DSF is its ability to report nucleotide and metal interactions among active and inactive conformers present in a variety of ( pseudo)kinase domain preparations [47]. We therefore exploited DSF to investigate the binding of nucleotides to PKAc. As expected [65,66], thermostabilization was readily observed for WT PKAc using a broad range of nucleotides ( particularly ATP and ADP) in the presence of divalent cations ( Figure 3B), and this was recapitulated with the R133A variant. In contrast with the increase in T m observed for WT PKAc (ΔT m = 4.24 ± 0.09°C) and R133A (ΔT m = 3.49 ± 0.16°C) with 1 mM ATP and Mg 2+ , incubation of K72H PKAc (or a less stable K72R PKAc mutant, data not shown) with any combination of nucleotides or cations failed to promote any stabilization under these conditions (ΔT m for K72H = −0.49 ± 0.04°C upon incubation with ATP/ Mg 2+ ions), even at very high concentrations (up to 4 mM) of ATP. We therefore believe that our assay reveals either a loss of nucleotide binding in the K72H mutant or an inability to detect thermal stabilization upon nucleotide binding. These findings contrast with previous work, which employed fluorescent ATP displacement (mant-ATP) and CD-based thermal unfolding in urea, to demonstrate that K72H C-subunit could bind to ATP and the small-molecule inhibitor H89, despite a detectable lack of phosphorylation or catalytic activity [56].
We next studied the effects of exogenous protein binding to PKAc, taking advantage of the resistance of PKI to denaturation at temperatures above 70°C, at which temperature PKAc is fully unfolded ( Figure 3A). Binding of purified PKI inhibitor protein to all three PKAc proteins was assessed by both DSF and (IM-)MS. The highaffinity interaction of PKAc with PKI measured by DSF was confirmed by native MS with observation of the PKA:PKI heterodimer over multiple (15 + -17 + ) charge states ( Figure 4A). A ΔT m value of +1.97 ± 0.04°C was determined for WT PKAc upon PKI binding ( Figure 4B), while the IC 50 value for inhibition of PKAc activity was in the sub nM range (Supplementary Figure S4), consistent with previous data measuring a Kdapp of <0.2 nM [57]. However, neither K72H nor R133A formed a stable interaction with the inhibitor protein under the relatively mild conditions used for the MS analysis ( Figure 4A), and both had negligible shifts in thermal stability as determined by DSF: 0.29 ± 0.37°C and −0.36 ± 0.01°C upon PKI addition for K72H and R133A variants, respectively ( Figure 4B). Furthermore, R133A exhibited a nearly 200-fold higher IC 50 value for inhibition by PKI (Supplementary Figure S4), confirming loss of enzyme inhibition despite unchanged autophosphorylation and ATP affinity. Interestingly, the binding of both PKI ( pseudosubstrate inhibitor) and ATP/Mg (nucleotide) to the WT C-subunit was additive, or even synergistic, over a range of ATP concentrations ( Figure 4B), confirming that DSF detects co-operative binding at two separate sites on PKAc, consistent with recent work [41].
To assess PKAc conformational dependence of PKI binding by IM-MS, it was necessary to look indirectly, by comparing the CCS (Ω) of the non-PKI-bound form of PKAc remaining in the presence of excess PKI (Ω WT(PKI) ) with Ω of PKAc in the absence of PKI (Ω WT ). The direct effects of PKI binding on the structure of PKAc could not be evaluated due to the innate presence of ligand and the resultant increase in complex size. IM-MS analysis of the non-PKI-binding form of PKAc revealed a small, but highly reproducible 1.4% increase in Ω, with a Ω WT(PKI) value of 29.8 nm 2 compared with Ω WT of 29.4 nm 2 ( Figure 4). Interestingly, Ω WT(PKI) , that is the non PKI-bound form of PKAc, was essentially identical with that of the more elongated form of PKAc, suggesting conformer-specific binding of PKI to the more compact PKAc species. Our model of conformer-specific binding of PKI to PKAc is also supported by IM-MS analysis of PKA mutants; both K72H and R133A lack the ability to bind PKI ( Figure 4) and both exhibit similar Ω values to WT(PKI).
Published crystal structures of PKAc lacking (PDB 13JH) or containing (PDB 1APM) a PKI-derived peptide (PKI 5-24) [67][68][69], allowed us to compare our experimentally observed differences in PKAc conformation with those from published X-ray structures. The EHSS model, implemented in Mobcal, was used to calculate Ω EHSS for PKAc from PDB files. In agreement with our MS data, a decrease in PKAc Ω was observed upon PKI binding: Ω EHSS (PKAc 11-350) = 33.1 nm 2 and Ω EHSS (PKAc 11-350 with PKI) = 32.2 nm 2 . Although Ω cannot be compared directly between our studies and the computed Ω EHSS values from X-ray structures, primarily because the core PKA sequence is different in these two sets of experiments, the reduction in conformational space adopted by PKAc upon PKI binding is entirely consistent with our IM-MS data.
Interestingly, CIU profiles of the heterodimeric PKA-PKI complex ( Figure 4D) revealed distinct partially unfolded intermediates, unlike the gradual elongation observed for the non-PKI-bound species. In contrast with the complex formed between PKAc and the nM kinase inhibitor staurosporine, the PKA-PKI complex remained stable up to a CE of 50 V, confirming the strength of the inhibitory (noncovalent) PKA/PKI interaction, and consistent with the picomolar affinity of PKI for PKA [34,35,57].

The PKI-binding conformer of PKAc has enhanced structural stability
The preferential binding of PKI to the more compact PKAc conformer led us to question whether there was a difference in the structural stability of these species. The non-PKI-bound PKAc remaining in the presence of excess PKI, PKA(PKI), was thus subjected to CIU ( Figure 4) and compared with the unfolding profiles of WT, R133A and K72H PKAc (Figure 2). The CIU profile of PKA(PKI) was very similar to that observed for R133A; both demonstrate a significant reduction in gas-phase stability compared with WT PKAc, with collisionmediated unfolding starting at ∼32 V. These data suggest that it is the conformation, rather than phosphorylation status, of PKAc that regulates its structural stability and ability to bind to PKI.

PKI disrupts hetero-oligomers of PKA catalytic and regulatory subunits
To examine the disruptive effects of PKI on the physiological (catalytically inactive) hetero-tetrameric R 2 C 2 complex of PKAc and PKA RIIα regulatory subunits, we undertook a series of novel MS and biochemical analyses. As shown in Figure 5, under native ESI conditions, we clearly observe monomers and homodimers of PKAc, monomers, homodimers, trimers and tetramers of RIIα, and various combinations of multimers after reconstitution of the PKA holoenzyme from individually purified components at a 1:1 ratio. Importantly, we can readily detect the physiological R 2 C 2 holoenzyme complex. Dramatically, titration of equimolar PKI completely ablates observation of the R 2 C 2 complex, additionally disrupting R 2 C, CR and C 2 complexes, with preferential formation of the PKAc:PKI heterodimer. Disruption of PKAc, RIIα complexes by PKI, and formation of a PKAc:PKI heterodimer was also confirmed by SEC, the standard biochemical method for PKA complex analysis (Supplementary Figure S5). Consistent with previously published data with the RI subunit [62], there was no evidence for holoenzyme complex formation and RIIα binding with either the inactive K72H or the catalytically active R133A [57] variants either by SEC (Supplementary Figure S5) or native MS analysis.

Binding of small molecules to PKAc decreases conformational flexibility
The reversible binding of small-molecule ligands has revolutionized the study of protein kinases and is a powerful strategy for inhibiting (or activating) the catalytic activity of protein kinases in vitro and in vivo [10]. Using DSF, we assessed the effect of a panel of small-molecule PKA inhibitors on PKAc thermostability ( Figure 6). We also evaluated compound binding to the catalytically inactive K72H variant, which we demonstrated (Figures 3 and 4) could not interact with nucleotides. Remarkably, many ATP mimetics known to target the PKAc ATP site markedly increased thermostability of K72H PKAc. Staurosporine, an indolocarbazole that is a pan-protein kinase inhibitor, stabilizes PKAc WT and K72H equally well, as do the closely related (but more specific) compounds N-benzoyl staurosporine and K-252a. In contrast, and in a similar fashion to the pseudosubstrate PKI full-length protein or PKI (5-23) peptide, the AGC kinase sulfonylisoquinoline inhibitors H89, HA-100 and the oral pan AGC inhibitor AT13148 bind only to WT and R133A PKAc under these conditions, and not the catalytically inactive (unphosphorylated) K72H PKAc, suggesting conformer-specific selection by specific ligand classes. This raises the question as to whether catalytic activity is a useful indication of Figure 5. PKI disrupts non-covalent complexes between PKAc and RII.
Native ESI mass spectra of WT PKAc (C, blue dots), PKA regulatory subunit RIIα (R, green squares) and equimolar ratios of the two (C + R) in the absence or presence of PKI proteins (red triangles).  the potential ability of kinase domains to bind ATP-competitive molecules in vitro and in cells and whether the physiological effects of inhibitors may be mediated in a compound-specific manner through interactions with 'inactive' conformers found in both canonical protein kinases and the multiple different classes of protein pseudokinase [70].
To corroborate our DSF assays, binding of the PKA inhibitor H89, the oral multi-AGC kinase inhibitor AT13148 (which is in phase I clinical trials) [71] and staurosporine [72] was confirmed by native ESI-MS (Figure 7 and Supplementary Figure S6).
Given our discovery of preferential binding of PKI to a specific PKA conformer, we also assessed the effect of these small molecules on PKAc conformation by IM-MS. Interestingly, there was a reduction in the charge states observed upon incubation with all three inhibitors, and a slight reduction in CSD, potentially suggesting a more ordered structure upon inhibitor binding. Inhibitor-dependent changes in CCSD for the same charge state could not thus be compared directly. Average Ω could be analyzed between the lowest two charge states for vehicle control (12+, 13+) and inhibitor bound (11+, 12+), as these were deemed to be most representative of the stable folded PKAc structures. All the inhibitor-bound forms of PKAc exhibited reduced conformational flexibility, with CCSD decreasing over 20% from 2.1 nm 2 to <1.6 nm 2 . CCS also decreased upon inhibitor binding (in agreement with the reduction in CCSD), suggesting preferential adoption of a more compact PKAc conformer. Unlike PKI binding to PKAc, interaction with these inhibitors does not appear to be conformationdependent, although binding did promote structural compaction with reduced conformational flexibility.
Given the indiscriminate binding of staurosporine and K252a to both WT and K72H PKAc, and formation of selective complexes between AT13148 and WT PKAc, we reasoned that there must be a difference in the binding mode of these small-molecule ligands that could potentially regulate interaction between members of the PKA signaling module. We therefore evaluated whether preincubation with a small molecule influenced PKI:PKA binding or PKI disruption of the PKA holoenzyme. Unlike PKI, staurosporine exhibited conformerindependent binding to PKAc, since no unbound catalytic subunit was observed (Supplementary Figure S7), consistent with our solution DSF data. Increasing the CE applied to the PKA:small molecule:PKI complex preferentially resulted in dissociation of the small molecule (∼55 V), leaving the stable PKA:PKI heterodimer. Interestingly, there was no discernible effect of PKI disruption of the PKA holoenzyme when it was performed in the presence of any ATP-competitive inhibitors, demonstrating independent binding mechanisms (Figure 7), which are entirely consistent with DSF analysis ( Figure 4B).

Conclusions
The complementary techniques of DSF and IM-MS are exploited here to examine structure, stability and binding of proteins and inhibitors to PKA signaling components in vitro. Even though PKA is used as a benchmark for studying other protein kinases, MS-based structural evaluation of the PKA complexes has not previously been reported to the best of our knowledge. MS studies previously evaluated ligand and inhibitor sensitivity of type I cGMP-dependent protein kinase (PKG) subunits [73,74]. However, PKG differs from PKA in that its regulation involves sensing of the cyclic GMP nucleotide within a single polypeptide. In the present paper, we report the first MS-based characterization of the PKA holoenzyme and explore heteromeric PKA complexes using native MS and IM-MS. By building on previous studies of PKA autophosphorylation after isolation from bacteria [25,26], we identify three novel sites of C-subunit autophosphorylation. Future studies will evaluate whether these sites are also phosphorylated in cells and in vivo, and whether they contribute to catalytic activity, and the formation or stability of PKA protein:protein complexes or both. We provide new evidence by IM-MS that PKAc exists in two primary conformations that are differentially permissive for ligand binding at the PKI ( protein pseudosubstrate) site on the surface of the protein. Unfortunately, low MS resolution using ESI under native conditions prevents us from categorically determining if the phosphorylation status of the two C-subunit conformers are different. However, PKAc that has reduced (λPPase-treated) or completely lacks (K72H) phosphorylation preferentially exhibits an elongated conformation, similar to that of the elongated WT PKAc structure that does not bind to PKI. This predicts that these structural differences might, in part, be driven by altered phosphorylation status, similar to previous data that uncovered marked differences in structural stability between myristoylated and nonmyristoylated PKAc proteins [29,75]. However, our finding that the CCS value of normally phosphorylated R133A PKAc is identical with that of the larger (more elongated) WT PKAc conformer, and similar to that of unphosphorylated K72H, confirms that the structural diversity of PKAc is driven by factors in addition to the extent of phosphorylation. Modeling of R133A PKAc predicts an altered conformation, caused by ablation of the positive charge at Arg133 and loss of an electrostatic interaction with Glu230 [57]. Based on our work with R133A PKAc, we propose that this mutation can drive the phosphorylation-independent conformational differences observed by IM-MS. In this context, in the future, it will be interesting to compare these structural effects with those of stoichiometrically myristoylated PKAc, and a panel of PKAc phosphorylation site mutants, including the three novel sites identified in our in vitro study, whose cellular relevance is currently unknown.
Interestingly, we find that the smaller (less elongated) of the two PKAc conformers preferentially binds the pseudosubstrate inhibitor protein PKI, a pM affinity regulatory subunit with complex biological roles [37]. Consistent with this observation, the smaller conformer was absent upon IM-MS analysis of the K72H and R133A PKAc mutants, which are significantly impaired in their ability to bind PKI [21]. Distinct MS-unfolding profiles were observed for the non-PKI-binding form of PKA compared with the mixed conformer species. Of particular interest, the conformer incapable of binding PKI exhibited a markedly lower structural stability than the inhibitor-sensitive C-subunit population. Native MS analysis also reveals ternary formation of tetrameric PKA holoenzyme R 2 C 2 complexes, as well as mixed R 2 C and RC multimers in vitro. Furthermore, we provide unambiguous experimental evidence that PKI directly disrupts the formation of all possible R:C complexes (likely at the R:C interface), confirming the exclusive binding of PKI to structurally permissive PKAc subunits, even in the presence of RIIα. The resultant (noncovalent) PKA:PKI complex is extremely stable, exhibiting a multistage unfolding profile with populations of discrete stable (partially unfolded) intermediate structures. It will be interesting to translate our findings to the cellular environment, where subcellular PKAc complexes with differential affinity for PKI could exist, with inhibitor affinity regulated as a function of PKAc phosphorylation or heterodimerization. In this regard, the analysis of PKI binding to nonphosphorylated, myristoylated or mutated PKAc, or PKAc complexes disrupted by cell-permeable AKAP ligands [76], will be of particular interest.
Based on careful MS analysis, kinase-dead (K72H) PKAc does not autophosphorylate when expressed in bacteria, probably due to an inability to bind nucleotides (either in the presence or absence of divalent metal cations), as confirmed using a DSF assay previously validated to detect binding of a variety of ligands to a panel of catalytically inactive and ATP binding-deficient pseudokinases [47]. Interestingly, the K72H mutant behaves as a potent 'dominant negative' in vivo when injected into oocytes [21], and retains the ability to bind to the ATP-competitive small-molecule inhibitors staurosporine and K-252a, but not isoquiolinesulfonyl PKAc inhibitors [77] or the clinical candidate AT13148. The ability of the type I inhibitor staurosporine to stabilize an 'open' (inactive) conformation of phosphorylated PKI-bound PKAc is known from X-ray analysis, and a potential induced van der Waals interaction with Lys 72 has been postulated [78]. However, our DSF analysis confirms that K72H PKAc still binds with high affinity to staurosporines. Interestingly, the preferential targeting of 'type II' kinase inhibitor classes to inactive kinase conformers of tyrosine kinases is well known and clinically relevant [43,51]. However, the propensity of inactive Ser/Thr kinases, exemplified by K72H PKAc (the present study) or multiple pseudokinases [47], to bind to different classes of kinase inhibitor is an important new area of research focus [70,79,80]. Our work demonstrates that, for an unphosphorylated K72H PKAc mutant, an inability to bind to ATP does not preclude binding to bisindole-based ATP-competitive inhibitors whose lack of specificity precludes their rational exploitation as chemical biology probe compounds [19]. In the absence of a crystal structure, it is challenging to explain why we are unable to detect K72H nucleotide binding using standard procedures [65], especially given previous findings [56,62]. However, Lys72 of PKA is reported to interact with the αand β-phosphates of ATP and a critical C-helix Glu residue in the closed, active PKA structure [69]; mutation of the equivalent residue in ( pseudo)kinases can ablate ATP binding that can be quantified by solution DSF [47,81]. We therefore speculate that should ATP still bind to K72H PKAc, its conformation must be markedly different from that of WT and R133A PKAc, but nonetheless remain responsive to thermal stabilization by some kinase inhibitors. All the kinase inhibitors evaluated in the present study are formally ATP-competitive (type I) kinase inhibitors, and our work has ramifications for studying cellular small-molecule ligand effects on phosphorylated or inactive kinases, predicting that, for PKAc, phosphorylation might also be able to indirectly control kinase:smallmolecule interactions.
We conclude that analytical DSF complements IM-MS and CIU for interrogating ligand-binding and ligandmediated conformational dynamics of PKA signaling components. These strategies are quantitative, permit evaluation of specific and nonspecific interactions, and enhance our understanding of the structural effects pertaining to ligand:protein interactions in vitro. Indeed, we provide new insights into the PKA holoenzyme complex, a signaling system that continues to inspire, half a century after its discovery.