Dynamic Equilibria in Protein Kinases

Abstract

Structural changes involved in protein kinase activation and ligand binding have been determined from a wealth of X-ray crystallographic evidence. Recent solution studies using NMR, EPR, HX-MS and fluorescence techniques have deepened this understanding, by highlighting the underlying energetics and dynamics of multistate conformational ensembles. This new research is showing how activation mechanisms and ligand binding alter the internal motions of kinases and enable allosteric coupling between distal regulatory regions and the active site.

Introduction

Eukaryotic protein kinases are essential mediators of cell signaling under normal and disease conditions. The success of kinase inhibitors in cancer therapeutics has intensified research into allosteric regulatory mechanisms for this enzyme class [1,2]. In protein kinases, conserved N-and C-terminal lobe residues place ATP and substrate in proximity for phosphoryltransfer (Figure 1) [3]. In the N-lobe, conserved Gly-rich loop, Asp-Phe-Gly (DFG) and Lys-Glu motifs coordinate Mg²⁺ and nucleotide phosphoryl oxygens, while hinge residues hydrogen bond with the adenine ring. The substrate contacts the C-lobe through an activation loop (“A-loop”), which contains regulatory phosphorylation site(s), and an adjacent “P+1” substrate recognition loop. A conserved Asp within a His-Arg-Asp (HRD) motif mediates base-assisted phosphoryltransfer to the substrate. Connecting these motifs are hydrophobic “R-spine” and “C-spine” structural elements that stabilize the positioning of active site residues [4].

Figure 1. Elements of the protein kinase active site.

(A) Crystal structure of phosphorylated PKA-C complexed with Mn₂ATP and the peptide inhibitor, PKI (PDB ID: 1ATP [50] from rcsb.org [51]). Protein kinases form contiguous hydrophobic residue interactions in the R-spine (light red) and C-spine (light blue), which help position active site residues and connect the N- and C-lobes. (B) Close-up view of key residues in the active site. These include the Lys-Glu salt bridge and the Asp residue in the DFG motif, which coordinate metal atoms (lime green) and nucleotide phosphoryl groups; residues in the hinge, which hydrogen bond with the adenine ring; and the Asp residue in the HRD motif, which serves as a catalytic base during phosphoryltransfer to the substrate hydroxyl acceptor.

Crystal structures show conformational changes that accompany the switch between active and inactive kinase states. In the N-lobe, activation often repositions helix αC from an “αC-out” to an “αC-in” conformation, forming a Lys-Glu salt bridge needed for nucleotide-phosphate interactions. Likewise, the DFG motif undergoes variable changes in dihedral angle from inactive “DFG-out” to an active “DFG-in” conformation, where the Asp is directed into the active site for metal coordination. Phosphorylation leads to dramatic movement of the A-loop/P+1 loop from closed to open conformations, exposing a surface for substrate binding. Crystallographic studies of kinase-inhibitor complexes have revealed a high level of plasticity in the kinase active site, allowing small molecules to trap different conformational states [5].

How these structural changes are coordinated dynamically is an evolving question in this field. In this review, we summarize recent investigations of protein kinases using NMR correlation and relaxation experiments, fluorescence and EPR spectroscopy, and hydrogen-deuterium exchange mass spectrometry (HX-MS) measurements. Breakthrough studies show that synchronous motions in kinases can be related to steps in enzymatic turnover, as well as allosteric responses to phosphorylation, mutations, and regulatory ligand or domain interactions. These findings help clarify the role of dynamics in kinase regulation and how they can be exploited by therapeutic inhibitors.

Linking dynamics to catalytic turnover in PKA

NMR relaxation studies of the cAMP-dependent protein kinase catalytic subunit (PKA-C) have revealed the control of dynamics by nucleotide and substrate binding. On slow (μs-ms) timescales, residues around the active site move asynchronously in the apoenzyme, but their dynamics become correlated when AMP-PNP binds to form the binary nucleotide complex. The correlated motions extend further into the N- and C-lobes following peptide substrate binding, revealing global exchange centered on the spine architecture in the ternary Michaelis complex [6–8]. Comparison of NMR chemical shifts to crystallographic observations of domain closure in binary and ternary complexes suggested that the global motion can be ascribed to opening and closing of the N- and C-lobes. Molecular dynamics (MD) simulations supported this model and further revealed an allosteric pathway between the ATP site and P+1 loop [9,10]. Remarkably, the rate constant for domain opening (31 s⁻¹) is comparable to k_cat (23 s⁻¹) suggesting that these movements are rate-limiting for the slowest step in turnover of product dissociation [7,8,11].

The connection between ligand-induced collective motions and catalytic function has been recently strengthened by studies of PKA-C mutants, including those found as drivers for human diseases. A mutation in Cushing’s syndrome, L205R, disrupted the allosteric network, abolished positive cooperativity between nucleotide and substrate, and suppressed k_cat/K_m [12••]. Similar results were found with a mutation at the adjacent residue, Y204, which structurally links the P+1 loop to the hydrophobic core [13], and with an oncogenic fusion between DNAJB1 and PKA-C from hepatocellular carcinoma [14•].

Ligand binding also affects motions on fast (ps-ns) timescales [6,7]. NMR order parameters and MD analysis indicate global effects on conformational entropy, which decreases in the binary PKA-C:AMP-PNP state but increases in the ternary substrate complex [15••]. This has been linked to positive cooperativity between ATP and substrate, which may be explained in part by nucleotide binding “prepaying” the entropic cost of substrate binding. Similarly, conformational entropy decreases in the ternary product complex, consistent with an observed negative cooperativity between ADP and phosphorylated product. The results suggest a role for conformational entropy in substrate binding and product release.

Taken together, the evidence reveals effects of nucleotide and substrate binding on both fast and slow motions throughout PKA-C. Slow, correlated motions of residue backbone and side chains extending throughout the protein create a dynamically committed state that couples domain movements to enzymatic turnover. Changes in fast side chain motions promote positive cooperativity between reactants and negative cooperativity between products. In this way, ligand binding regulates a global dynamic network in order to facilitate catalytic function.

Correlated motions in MAP kinases

Global dynamics have also been observed in the MAP kinases, ERK2 and p38α, following kinase activation by dual phosphorylation at the A-loop. NMR relaxation measurements of inactive, unphosphorylated ERK2 show local, asynchronous motions of residues throughout the active site and N- and C-lobes, which undergo a switch to correlated motions on a millisecond timescale in the active, phosphorylated enzyme [16] (Figure 2). The synchronous behavior can be modeled by global exchange between conformational states, including a new state that, based on HX-MS and X-ray evidence, enables proper alignment of Mg²⁺-ATP for catalysis [17,18••]. Although the structural differences that make up the new state are not readily detected crystallographically, MD simulations show small conformational changes in phosphorylated ERK2 that create a more compact active site, shortening the Lys-Glu distance and positioning Mg²⁺-ATP into a reactive configuration [19].

Figure 2. A network of residues with correlated motions in ERK2.

(A) Crystal structure of dual phosphorylated ERK2 (PDB ID: 2ERK [52]) illustrating the location of dynamically-correlated residues, observed using NMR Carr-Purcell-Meiboom-Gill (CPMG) relaxation dispersion experiments that report motions on slow (100-2000 s⁻¹) timescales. Blue spheres indicate (methyl-¹³C)-labeled residues that show global exchange behavior in the phosphorylated form of ERK2. In the unphosphorylated form, the methyl relaxation curves could not be fit globally, indicating asynchronous behavior. (B) CPMG dispersion plots of effective relaxation rate (R_2,eff) vs frequency (v_CPMG) for residues in the activation loop (L182, V186) and the N- and C-lobe core (I87, L155). Global exchange is indicated by the similar curvatures of each plot. Data collected at 600, 800, and 900 MHz are colored green, red, and purple, respectively, and can be fit to a global exchange rate constant (k_ex) of 300 s⁻¹. The results show that motions in the activation loop are coupled to those in the core. Adapted with permission from Iverson DB, Xiao Y, Jones DN, Eisenmesser EZ, Ahn NG.Biochemistry 2020, 59: 2698–2706. Copyright 2020 American Chemical Society.

Importantly, residues adjacent to the phosphorylation sites are part of the global network, indicating that the A-loop undergoes slow movements that are correlated to the motions within the kinase core [20•]. This occurs despite many ion-pair interactions formed between the phosphates and arginine residues. Coupling between the active site and A-loop was supported by the ability of A-loop mutations to disrupt global exchange within the core. Furthermore, different ATP-competitive inhibitors shifted the conformational equilibrium in opposite directions, and correspondingly altered HX behavior of the P+1 and A-loops [18••,20•,21]. A functional consequence of coupling was that these inhibitors could differentially alter the dephosphorylation of ERK2 by MAP kinase phosphatase [18••], explaining their opposite effects on the steady-state phosphorylation of ERK1/2 in cells [22,23].

As in ERK2, NMR relaxation studies of phosphorylated p38α revealed slow, correlated motions throughout the kinase, reflecting widespread communication across the N-lobe, C-lobe, A-loop, and hinge [24]. In p38α, however, the global effect also required ligand occupancy of the kinase interaction motif (KIM) docking site, located near the hinge [25]. KIM sequences are found in many MAP kinase substrates and effectors, and their binding to the docking site enhances ATP and substrate binding as well as the rate of phosphoryltransfer [26]. The global exchange behavior was disrupted by addition of Mg²⁺AMP-PNP to form a ternary complex, suggesting that a function of the collective motions induced by phosphorylation and KIM binding is to enhance nucleotide affinity [24]. MD simulations supported this model, by showing that the phosphorylated A-loop occludes the active site, blocking ATP binding until the KIM site is occupied [27].

Together, the evidence indicates that global motions in both ERK2 and p38α remodel the energy landscape to facilitate nucleotide binding and assembly of the catalytically competent state. Both enzymes require A-loop phosphorylation but differ in their need for ligand binding in order to coordinate motions over long distances.

Concerted movements in Aurora A and CDK2

Insight into conformational states in exchange has been gained from experiments measuring molecular distances. Recent applications have examined Aurora A (AurA), which is activated by A-loop phosphorylation and/or binding of its N-lobe to the allosteric subunit, TPX2 [28]. TPX2 aligns the R-spine through a residue in helix αC, which in turn coordinates interactions between the Lys-Glu bridge and the DFG motif via bound waters [29]. This forms an allosteric network linking the A-loop and TPX2 binding site.

The dynamics of A-loop remodeling were examined by placing fluorescent probes at pairs of residues to track different configurations observed by X-ray crystallography. Single molecule fluorescence quenching (Figure 3) and Förster resonance energy transfer (FRET) methods both revealed broad distance distributions, signifying heterogeneous populations of open and closed A-loop conformations corresponding to active and inactive states, respectively [29–33••]. EPR measurements of distances between nitroxide spin label probes confirmed these results [31,34]. The A-loop-open vs -closed states, assayed by FRET, were matched to DFG-in vs -out states, measured by absorbance of an infrared probe on helix αC [29]. These remained correlated in response to TPX2 and temperature, indicating that the movements of the DFG motif and A-loop are tightly coupled. The single molecule fluorescence data fit a two-state model with additive contributions from TPX2 and phosphorylation [33••]. FRET and EPR experiments, however, revealed more than two states as well as synergy between TPX2 and phosphorylation, perhaps due to the longer distance range of these measurements [29,31]. A survey of AurA inhibitors showed varying A-loop/DFG substates that correlated with the degree of binding cooperativity between inhibitor and TPX2. Given that TPX2 controls the function of AurA in spindle assembly but not centrosome maturation, the findings suggest that inhibitors could be designed to selectively block distinct functions of AurA [30].

Figure 3. Heterogeneity of A-loop conformations in AurA.

Single molecule measurements of fluorescence intensity were used to report locations of the A-loop based on distances between tetramethylrhodamine dye-labeled Cys residues. (A) Dye-labeled residue pairs in the A-loop (S283C) and N-lobe (K224C) fluoresce when the A-loop is open and quench when closed. Fluorescence intensity histograms show that TPX2 binding to unphosphorylated AurA shifts the equilibrium slightly towards an A-loop-open (active) conformation (~1.4-fold change in K_eq). Similar effects of TPX2 binding were observed with phosphorylated AurA (not shown). (B) Dye-labeled residue pairs in the A-loop and C-lobe (S283C, M373C) quench when the A-loop is open. This is enhanced by AurA phosphorylation, consistent with a 3.6-fold stabilization of the active A-loop conformation. Reproduced from Gilburt JAM, Girvan P, Blagg J, Ying L, Dodson CA. Chem Sci 2019, 10:4069–4076, with permission from The Royal Society of Chemistry.

While the elements of the AurA allosteric network are still incompletely understood, an intriguing study used ancestral sequence reconstruction (ASR) to identify an ensemble of residues involved in activation by TPX2. Fifteen amino acids in AurA were identified that rescued allosteric activation when all were swapped into a TPX2-unresponsive ancestral variant [35••]. Interestingly, the residues were distributed across the N- and C-lobes, without forming obvious clusters. They raise possibilities for discovering new features of the network contributing to allosteric communication in AurA.

Like AurA, CDK2 is activated by N-lobe binding to a regulatory subunit (Cyclin A), which aligns the R-spine and Lys-Glu bridge. But while AurA can be regulated by either TPX2 or phosphorylation alone, CDK2 requires both Cyclin A and A-loop phosphorylation for activation. EPR was used to measure distances between spin label pairs on CDK2 that separately reported conformations of αC and the A-loop [36••]. As seen with AurA, CDK2 showed heterogeneous populations of A-loop-open/closed and αC-in/out states. Crucially, paramagnetic relaxation enhancement NMR experiments revealed at least three different substates of the A-loop, only one of which matched an active conformation. Combining Cyclin A binding with phosphorylation shifted the equilibrium strongly towards the active A-loop-open/αC-in conformation, explaining why both events are required for activation [36••].

In summary, spectroscopic measurements of AurA and CDK2 showed how kinases can respond differently to similar allosteric mechanisms, via concerted movements that shift the conformational equilibria between heterogeneous populations. The studies revealed greater complexity in dynamics involving previously unrecognized substates of the A-loop. In fact, a recent Kincore database now classifies the DFG motif into eight substates [37,38•]. A more nuanced view will likely be needed to describe the multiple discrete states that make up the conformational ensemble.

Allosteric domain interactions in Abl

Src-module tyrosine kinases expand the understanding of dynamics in allostery to include intramolecular interactions with SH3 and SH2 domains, in a SH3-SH2-linker-kinase domain arrangement [39–41]. X-ray structures of autoinhibited Src-module kinases show an assembled configuration, with the linker segment bridging the SH3 domain and kinase N-lobe, and SH2 domain contacts with the kinase C-lobe. Activation involves disassembly of the regulatory SH3-SH2 module from the kinase domain, along with A-loop phosphorylation and opening. Such rearrangements release hydrophobic contacts between the linker and N-lobe, enabling alignment of the R-spine, helix αC, and Lys-Glu bridge into a catalytically competent configuration [42,43••].

Allosteric coupling between the SH3-SH2 interface and the active site has been investigated in many solution studies of Src-module kinases, as described in excellent reviews [39,41,44]. Abl provides a recent example and one of the most complete descriptions of a tyrosine kinase conformational landscape. Here, Type II inhibitors (e.g. imatinib) bind the ATP site in a DFG-out/A-loop-closed configuration, which triggers detachment of the SH3-SH2 domains as evidenced by increased radius of gyration and increased rotational correlation times [45]. In contrast, Type I inhibitors bind an assembled state with DFG-in/A-loop-open [46]. Asciminib, an allosteric inhibitor which binds an autoinhibitory pocket for the N-terminal myristate in Abl, reverses the imatinib-induced module disassembly [46]. Thus, conformational changes at the DFG/A-loop propagate to the N-lobe, disrupting or strengthening the SH3-linker interactions that maintain the assembled state.

NMR structural and relaxation studies reveal how the energy landscape of Abl is regulated by allosteric interactions [47,48••]. The unphosphorylated and unliganded Abl kinase domain (res. 248-534) was observed as three distinct conformers interconverting with slow kinetics (50-100 s⁻¹; Figure 4A) [48••]. The NMR structure of the major state showed an active-like structure with DFG-in, αC-in, and A-loop-open (“A”) (Figure 4B). By contrast, two minor states showed chemical shifts consistent with structures of inhibitor-bound complexes. Their NMR structures appeared inactive, with a broken R-spine, but one conformer (“I₂”) resembled a fully inactive conformation (DFG-out/αC-out/A-loop-closed), while the other (“I₁”) displayed mixed features (DFG-out/αC-in/A-loop-open) (Figure 4B). Therefore, the kinase domain in solution adopts a predominantly active conformation that exchanges slowly with two inactive conformers, revealing preorganization of the inactive states prior to inhibitor binding. Effects of phosphorylation or mutations on kinase activity were readily explained by their ability to shift the conformational equilibrium towards or away from the active state (Figure 4A).

Figure 4. A multi-state conformational landscape in Abl.

(A) NMR chemical exchange saturation transfer (CEST) experiments of (methyl-¹³C)-labeled residues in the Abl kinase domain (res. 248-534) report the relative populations of states found in slow conformational exchange. Dips in the intensity profiles reveal three distinct conformations corresponding to an active ground state, A, and two inactive excited states, I₁ and I₂. The top profile (residue L267) shows destabilization of the I₂ state in favor of I₁ in the imatinib resistance mutant, H415P, corresponding to a shift that disfavors imatinib binding. The bottom profile (residue L383) shows the disappearance of both inactive states following activation loop phosphorylation (pY412), corresponding to a shift towards the active state. Adapted from Xie T, Saleh T, Rossi P, Kalodimos CG. Science 2020, 370:eabc2754. Reprinted with permission from AAAS. (B) Solution NMR structures (red, green, blue) corresponding to the unliganded A, I₁, and I₂ conformational states observed by CEST NMR, overlaid with inhibitor-bound X-ray structures of the Abl kinase domain (tan). Left panel: The A state (red, PDBID:6XR6 [48]) shows active conformational motifs (DFG-in/A-loop-open/αC-in) similar to those described for dasatinib-bound Abl (tan, PDBID:2GQG [53]). Right top panel: The I₁ state (green, PDBID:6XR7 [48]) shows mixed active and inactive motifs (DFG-out/A-loop-open/αC-in), similar to the X-ray structure of PD173955-bound Abl (tan, PDBID:1M52 [54]). Right bottom panel: The I₂ state (blue, PDBID:6XRG [48]) shows fully inactive motifs (DFG-out/A-loop-closed/αC-out), similar to the X-ray structure of imatinib-bound Abl (tan, PDBID:1IEP [54]). The similarity between I₂ and the imatinib-Abl complex explains why destabilization of I₂ reduces affinity for imatinib in the H415P mutant (panel A). (C) ¹H-¹³C NMR correlation spectra (HMQC) for Abl residue M263, a sensitive probe for contacts between the SH2 domain and N-lobe. The overlaid spectra report relative populations of the extended, active state vs the assembled, autoinhibited state, and the effects of mutations and ligand binding on their equilibria. Adapted from Saleh T, Rossi P, Kalodimos CG. Nat Struct Mol Biol 2017 24:893–901. Reprinted by permission from Springer Nature Customer Service Centre GmbH: Nature Publishing Group, copyright 2017

Significantly, NMR structures of the Abl regulatory module (NCap-SH3-SH2-linker, res. 1-255) showed two distinct conformations in rapid exchange [47]. One conformer is consistent with the assembled, autoinhibited configuration, with contacts between the SH3 domain and the segment of the linker that interacts with the kinase N-lobe. In the other, the SH3 interface binds the NCap, displacing the linker segment that instead interacts with the SH2 domain. Strikingly, the new linker/SH2 interface matches that seen in a crystal structure of the regulatory module and kinase domain [49], where docking of the SH2 domain onto the kinase N-lobe locks helix αC in an active conformation. NMR analysis of the full-length enzyme (res. 1-534) revealed that adding the regulatory module shifted the kinase domain towards the I₂ state, explaining how the assembled module confers autoinhibition [48••]. Conversely, a mutant that stabilized the SH2/N-lobe interactions shifted the conformational equilibrium towards the A state, explaining its activation [48,49]. NMR chemical shifts of residue M263 in the N-lobe of Abl were used to model the relative populations of assembled vs extended states, which varied in response to mutations and ligand binding (Figure 4C) [47]. Together, the results describe a multi-state conformational landscape in Abl and show how cooperative transitions in the regulatory module and kinase domain can integrate allosteric signals to tune the level of enzyme activation.

Conclusions

Recent biophysical investigations of protein kinases in solution further the understanding of the role of dynamics in allosteric activation and catalytic turnover. Kinase activation promotes coordinated changes in dynamics that span the active site and extend into distal regions across the kinase core. Regulatory mechanisms that modulate kinase activity, such as phosphorylation, ligand binding, or subunit/domain interactions, can be explained by their ability to shift the equilibria between active, inactive, or intermediate/mixed states. Additional efforts are now needed to detail the conformations of these states, their relative populations and energetics, and their concerted movements. In many cases, differences between solution states match those observed crystallographically, for example in the conserved A-loop, αC, and DFG motifs. Other cases may involve only small conformational changes that are difficult to detect structurally. Importantly, new measurements are revealing landscapes with multiple substates, raising new questions about the heterogeneity of solution conformations. Future work is needed to describe the variations in discrete states formed by each protein kinase and how their correlated motions dictate the conformational landscape.