A dynamic switch in inactive p38γ leads to an excited state on pathway to an active kinase

Abstract

The inactive state of mitogen activated protein kinases (MAPKs) adopts an open conformation while the active state exists in a compact form stabilized by phosphorylation. In the active state, eukaryotic kinases undergo breathing motions related to substrate binding and product release that have not previously been detected in the inactive state. However, docking interactions of partner proteins with inactive MAPK kinases exhibit allostery in binding of activating kinases: interactions at a site distant from the activation loop are coupled to the configuration of the activation loop, suggesting that the inactive state may also undergo concerted dynamics. X-ray crystallographic studies of non-phosphorylated, inactive p38γ reveal differences in domain orientations and active site structure in the two molecules in the asymmetric unit. One molecule resembles an inactive kinase with an open active site. The second molecule has a rotation of the N-lobe that leads to partial compaction of the active site, resulting in a conformation that is intermediate between the inactive open state and the fully closed state of the activated kinase. Although the compact state of apo p38γ displays several of the features of the activated enzyme, it remains catalytically inert. In solution, the kinase fluctuates on a millisecond timescale between the open ground state and a weakly populated excited state that is similar in structure to the compact state observed in the crystal. The NMR and crystal structure data imply that interconversion between the open and compact states involves a molecular switch associated with the DFG loop.

Graphical Abstract

Introduction

Eukaryotic protein kinases, such as mitogen activated protein kinases (MAPKs), act in cellular signaling pathways as dynamic regulatory switches. Kinase activity and concomitant pathway activation is modulated by signal-responsive stabilization of an active conformation that is characterized by packing of spatially conserved hydrophobic residues in the active site and is associated with compaction of the bi-lobed kinase structure.¹ In the activated state, which is formed following phosphorylation of the activation loop, eukaryotic kinases fluctuate between open and closed conformations associated with substrate binding and product release.^2–5 In the inactive state, eukaryotic kinases adopt an open conformation with an outwardly oriented αC helix; activity is quenched by various mechanisms that include prevention of ATP/substrate binding, incorrect active site geometry, or a lack of conformational dynamics.^5,6

Here, we present a structural and dynamic study of the non-phosphorylated, inactive state of p38γ, a 43kDa MAPK involved in cellular signaling cascades related to stress response. Deregulation of p38 leads to pathologies related to inflammation and cellular proliferation such as arthritis, neurodegenerative disease, and cancer.^7,8 The p38 family of kinases is tightly regulated in homeostatic cells, requiring activation by double phosphorylation on the activation loop by upstream kinases or by partner-induced autophosphorylation and deactivation by various Tyr/Thr phosphatases.^9–12 Structures of inactive and active forms of various isoforms of p38 have been studied extensively by X-ray crystallography.^11,13–21 All exhibit a typical kinase fold with an N-terminal β-sheet domain and a C-terminal α-helical domain (Figure 1). Compared to other kinases, MAPKs incorporate an additional layer of regulation in the form of a docking site for binding of interacting proteins, a MAPK insert, and a C-terminal L16 loop and helix.^9,13,15,21 Although distant from each other, the configuration of the activation loop and docking site are allosterically linked in the inactive state.^13,15 Activation of MAPKs by phosphorylation in the activation loop leads to salt-bridge formation between the N lobe (via the αC helix), catalytic loop (via the HRD motif) and C lobe (via the APE motif) leading to compaction and rotation of the active site between the lobes and opening of the substrate binding site.¹⁴

Figure 1.

Structural model of active p38γ MAP kinase (1CM8,¹⁴) showing conserved features. The N-terminal lobe is colored blue and the C-terminal lobe is in pink. The ATP binding and active sites are formed by the αC helix, αC/β4 loop, Gly-rich loop, DFG (Asp-Phe-Gly), and HRD (His-Arg-Gly) motifs. Movement of the N-lobe relative to the C-lobe, indicated by the arrow, compacts or opens the active site resulting in an active or inactive state, respectively. The position of the αC helix plays an important role in kinase regulation and is stabilized in MAPKs by the αL16 helix and L16 loop. The substrate docking site and activation loop are on opposite sides of the C-lobe. The substrate binding site contains an APE (Ala-Pro-Glu) motif. The MAPK-insert is a distinctive feature of MAP kinases.

The structural dynamics of activated eukaryotic kinases are critical to function and regulation,^2–5,22 as observed for other dynamic enzymes.^23–29 Inactive eukaryotic kinases exhibit common as well as unique allosteric sites and various degrees of plasticity, depending on the mechanism for maintenance of inactivity; these differences confer regulatory specificity.⁶ Unlike Src-like kinases,^30–36 specific structural differences between inactive and active states of MAPKs are small, and activation involves rather subtle local rearrangements.^13–18 In addition, Src-like kinases are typically clamped in an inactive conformation by the presence of effector proteins or domains, and only transition into an active state when these “molecular brakes” are released. In contrast, MAPKs do not require additional effectors to remain inactive.

Without a requirement for external effectors to regulate the transition of MAPKs between inactive and active states, intramolecular processes are thought to control dynamics and thus activity. For example, in inactive ERK2, uncorrelated internal motions have been observed, particularly around the active site; these motions only become correlated upon activation, which is thought to unlock the hinge region between lobes.⁵ It is possible that differential dynamics may confer regulatory specificity among the MAPK family. Characterization of the conformational dynamics of inactive apo p38 provides valuable insight into how an enzyme that is inherently plastic, and necessarily so for function, remains properly regulated when inactive.

We describe a crystal structure at 2.55Å resolution that contains molecules of inactive p38γ in both a canonical inactive (open) and a novel active-like (compact) conformation. The active-like conformation is similar to other active MAPK structures but retains features of an inactive kinase. ¹⁵N and side chain methyl NMR relaxation experiments provide insights into the polypeptide dynamics and evidence that p38γ in solution undergoes conformational exchange between states corresponding to the open and compact structures. The data indicate that the interconversion between inactive open and compact conformations involves a molecular switch associated with the highly conserved DFG loop.

MATERIALS AND METHODS

Sample preparation

Expression plasmids for human p38γ (Uniprot P53778) were a gift from Dr. Peiqing Sun. A truncated construct was designed from which the flexible N and C-terminal regions, for which no electron density was observed in the X-ray structure of activated p38γ (1CM8¹⁴), were removed for crystallography. This construct is p38γ(7–352), where the first 6 N-terminal and the last 15 C-terminal amino acids of the full-length protein have been deleted, and was created by introducing a start codon, and NdeI restriction site at Pro6 and a stop codon and XhoI site at Pro353 by PCR cloning of the Homo sapiens p38γ gene into PET15b (Invitrogen). Both full length and shortened p38γ constructs contained an N-terminal His₆-tag and thrombin cleavage sequence.

The p38γ(7–352) and full-length wild-type p38γ constructs were overexpressed in Rosetta plysS BL21(DE3) (Invitrogen) Escherichia coli cells at 288K. Protein for crystallography was expressed in LB media. Perdeuterated NMR samples were expressed in M9 minimal media in D₂O with ¹⁵N ammonium sulfate. MILV ¹³C methyl labeled samples for relaxation dispersion experiments were grown in M9 minimal media in 100% D₂O, using perdeuterated glucose as the carbon source supplemented 1 hour prior to induction with α-keto-acid precursors (2-keto-3-d2–4-¹³C-butyrate for Ile-δ, and 2-keto-3methyl-d3–3-d1–4-¹³C-butyrate for Leu(δ1/δ2) and Val(γ1/γ2)) and ¹³C ε-methyl labeled methionine as described elsewhere.³⁷ Uniformly ¹³C labeled samples were also produced for ATP titration experiments by growth in M9 minimal media in H₂O using ¹³C glucose as the sole carbon source. Proteins were purified as described previously.³⁸ Following purification, samples for crystallization were treated with thrombin at room temperature to remove the N-terminal His-tag. Digestion was monitored by analytical HPLC on a C4 column. Thrombin was removed by a HiTrap benzamidine column (GE). Cleaved p38γ was separated from uncleaved kinase by Ni-NTA affinity chromatography. Samples for crystallography were dialyzed into 15mM Tris pH 7.5, 250mM NaCl, and 1mM TCEP to a final protein concentration of 10mg/ml. NMR samples were dialyzed into 25mM deuterated HEPES pH 6.8 (or pH 7.2 for methyl relaxation dispersion experiments), 250mM NaCl, 1mM TCEP, and 0.01% w/v NaN₃ in 90%/10% H₂O/D₂O. The protein concentration for NMR experiments was 200–300μM. The inhibitor BIRB796 (Selleck Chemicals) was resuspended in deuterated DMSO. Samples were incubated with a 2-fold molar excess of BIRB796 at room temperature. The pH was adjusted, as necessary, to pH 7.2 and the sample centrifuged to remove insoluble BIRB796.

For activation experiments, a constitutively active mutant of GST-tagged MKK6 (S207D, T211D) was expressed in Rosetta plysS BL21(DE3) (Invitrogen) E. coli cells at 293K in LB media. Cells with overexpressed GST-MKK6 were resuspended in 20mM Tris pH 8.0, 100mM NaCl, 5mM DTT, 0.1% Triton X-100, 1mM EDTA, and 10% glycerol and lysed by sonication. GST-MKK6 was purified by GST-Sepharose 4B (GE) affinity chromatography. Protein was dialyzed into a final buffer of 20mM HEPES pH 7.6, 100mM NaCl, and 2mM DTT. Activation of p38γ was achieved by incubation with MKK6 at room temperature in 75mM HEPES pH 7.5, 0.1mM PMSF, 1mM NaF, 0.1mM NaVO₄, 0.1mM β-glycerol phosphate, 10mM MgCl₂, and 1mM DTT. Activation was monitored by analytical HPLC on a C4 reverse-phase column. Activated p38γ exhibited a shift in elution time compared to inactive p38γ.

NMR spectroscopy

NMR backbone and methyl side-chain assignments for apo p38γ and in complex with BIRB796 were previously reported.^{38 15}N relaxation data were collected at 293K on Bruker Avance III 800MHz and Avance 500MHz spectrometers with triple resonance gradient cryoprobes. TROSY-detected R₁, R₂, and {¹H}−¹⁵N heteronuclear NOE experiments were performed in an interleaved fashion with 60% non-uniform sampling in the indirect dimension.^39–41 A total of 11 spectra were acquired with relaxation delays of 0, 0.016, 0.031, 0.047, 0.062, 0.078, 0.094, 0.109, 0.125 s for R₂, with 2 sets of duplicate points. A total of 10 spectra were acquired with delays of 0, 0.166, 0.332, 0.581, 0.913, 1.328, 1.826, 2.490 s for R₁, with 2 sets of duplicate points. Two sets of saturated and unsaturated {¹H}−¹⁵N heteronuclear NOE experiments were collected with a 16s ¹H saturation time and an 11ms delay between 180° pulses in the ¹H saturation block, as described previously.³⁹ Sinusoidal weighted Poisson gap sampling schemes were generated, and the optimal scheme was chosen.^41,42 Non-uniform sampling was used to reduce experiment time and also to enhance signal to noise. Data were reconstructed and processed with compressed sensing using MDDNMR and NMRpipe.^43–45

Multiple quantum ¹³C-¹H methyl relaxation dispersion experiments were performed at 500MHz and 800MHz spectrometer frequencies using published pulse sequences.⁴⁶ For each dispersion curve, 14 to 16 experiments were performed with different constant time CPMG pulse spacing and a CPMG block of length 20ms. Pulsing frequencies used were 100Hz*, 200Hz, 300Hz, 400Hz, 500Hz, 600Hz, 800Hz, 1000Hz, 1200Hz, 1400Hz, 1600Hz, 2000Hz*, where an asterisk indicates duplicate measurements. A reference experiment was also recorded, in which no CPMG element was applied. Relaxation dispersion data for apo p38γ were acquired at 283K and 293K. Samples of the complex with the BIRB796 inhibitor contained ~0.3% of residual DMSO. To ensure that changes in dispersion profiles were not due to the DMSO, a two-point relaxation dispersion experiment was performed as a control in the presence of an equivalent amount of DMSO. No difference was observed in the dispersion profiles in the presence of DMSO.

ATP titrations with wild-type p38γ were performed on a Bruker Avance 750MHz spectrometer with a triple resonance gradient probe. A 100mM stock of ATP and 1mM MgCl₂ was prepared in 25mM HEPES pH 6.8, 250mM NaCl and buffered to pH 6.8. Titrations with inactive p38γ at 200μM were performed at 295K. ¹H-¹³C HMQC spectra were recorded at each titration point and the average weighted ¹H and ¹³C chemical shift differences were calculated as described previously.⁴⁷ A one-site binding model was assumed and K_d was determined by fitting with in-house software, nmrKd. ⁴⁸

Relaxation data analysis

Spectra were analyzed with NMRView5.⁴⁹ R₁ and R₂ relaxation rates were obtained by fitting intensities to a single exponential decay curve (I(t) = I₀exp(−tR). Heteronuclear NOEs were obtained from the ratio of peak intensities of the saturated and unsaturated spectra (I_sat/I_unsat). Model-free analysis was performed using ¹⁵N R₁, R₂, and heteronuclear NOE data acquired at 500 and 800 MHz as previously described,50 using the extended Lipari-Szabo formalism.⁵¹ The apo p38γ X-ray structure solved here was used, with missing loops modeled, to optimize an axially symmetric rotational diffusion tensor (D_|| / D_⊥ = 1.562) using only relaxation data from residues with low flexibility (heteronuclear NOEs >0.65). A CSA of −170ppm and an amide bond length of 1.042Å was used in model-free analysis. The five models (model 1, S²; model 2, S² and τ_e; model 3, S² and R_ex; model 4, S², τ_e, and R_ex; model 5 partitions S² into fast and slow components, and also includes τ_e ) were fit with the in-house software eMF, which uses the Bayesian information criterion to select the best model with the fewest parameters.⁵⁰

Relaxation dispersion data were fit using in-house software Global and Local Optimization of Variable Expressions (GLOVE) that finds the global minima of local and global parameters. ⁵² GLOVE can be downloaded at scripps.edu/wright or is available for use on the NMRBox server (www.nmrbox.org). The fitted parameters included the kinetic rates (k_ex) for interconversion between states, the populations of the states, the carbon and proton chemical shift difference between states (Δω_C and Δω_H), and the transverse relaxation rate in the absence of exchange contributions (R₂₀). The error was determined from duplicate points or from the signal/noise ratio, whichever was greater. The Bloch-McConnell equations, describing multiple quantum magnetization and relaxation under an applied magnetic field at a frequency of (1/τ_cp), where τ_cp is the time between refocusing pulses, were used to fit the data to a two site exchange model.⁴⁶ F-statistics showed that inclusion of the Δω_H term did not improve fits even at a low 95% confidence level. All further analysis assumed Δω_H = 0. Exchanging residues, with R_ex > 1 at 500MHz, were clustered if the F-value between individual and global fits was < 2, using a fast exchange approximation⁵³ where appropriate. Global parameters were extracted from numerical fits with Δω_H fixed to 0.

Crystallization

Purified apo p38γ(7–352) was crystallized by sitting drop vapor diffusion at room temperature with a precipitant of 2.4M ammonium sulfate and 50mM Tris pH 7.7. Small needles were apparent within 1 week and larger crystals were obtained by streak seeding.⁵⁴ Crystals were cryopreserved in the well solution with sodium malonate at pH 7.0 added to a final concentration of 0.5M. Diffraction data were collected to 2.55Å, at the Advanced Photon Source at Argonne National Laboratory. Data were processed and scaled with the HKL-2000 software suite⁵⁵ in space group P1. The structure was determined by molecular replacement using PHASER⁵⁶ with PDB 1CM8 as a model. Two molecules of p38γ were found in the asymmetric unit. Regions of the model that did not initially fit the electron density well were improved by the morph module in Phenix⁵⁷ followed by rigid-body refinement with the N and C-terminal domains as separate bodies in Phenix. Iterative rounds of model rebuilding in Coot⁵⁸ followed by refinement in Phenix were performed. Ordered water and sulfate molecules were built in later stages of rebuilding. Crystallographic statistics are summarized in Table S1.

ATPase assays

A coumarin-labeled nucleoside diphosphate kinase (NDP) reporter system (Invitrogen) was used to measure ATPase activity by change in fluorescence at 487nm, with excitation at 440nm. The NDP system reports on the ATP/ADP ratio by reversible autophosphorylation of a histidine residue in the presence of ATP and dephosphorylation in the presence of ADP. The change in fluorescence due to ADP production from p38γ activity was measured over 5 minutes, and the linear portion of the curve, corresponding to the steady-state rate, was taken as the ATPase rate at a given ATP concentration. ATPase reactions were performed at 295K in 50mM HEPES pH 7.6, 500μM ATP, 10mM MgCl₂, and 1mM DTT.

Backbone dihedral angle analysis

Kinase structures in the RCSB⁵⁹ with resolution greater than 2.6A and having sequence identity ≥30% to p38γ (corresponding to most eukaryotic kinases) were used to form the structural dataset. Structures were classified as active, inactive, or DFG-out based on phosphorylation of threonine and the dihedral angles of Asp171 in the DFG loop. DFG-out structures were found to have Asp171 dihedral angles near ϕ = −120°, ψ = 100° or ϕ = −60°, ψ = −50°. Constitutively active kinases were also annotated manually. The dataset was analyzed by the Protein Structural Statistics (PSS) tool. ⁶⁰ The PSS tool aligned primary sequences using ClustalO⁶¹ prior to determining the ϕ/ψ dihedral angles for each structure. 117 active eukaryotic kinases and 118 inactive DFG-in p38 kinases were included in the analysis.

RESULTS

Crystal structure of inactive apo p38γ

Inactive p38γ (residues 7–352, with a spontaneous Gln31/Arg mutation) formed crystals that diffracted to 2.55Å (Table S1). The two molecules in the asymmetric unit have similar folds (Figure 2A) to published structures of inactive apo p38α.^16,17 A unique feature of the model is an N-terminal β-strand swap between the two molecules in the asymmetric unit (Figure S1). This strand swap does not alter the fold of the kinase and is not obligate; strand swapping is not observed in crystals of activated p38γ (1CM8¹⁴) where the N-terminal strand forms a β-hairpin (Figure S2A). Indeed, p38γ has been reported to be monomeric14 and gel filtration analysis of our construct, p38γ(7–352), shows no evidence of dimer formation in solution (Figure S3).

Figure 2.

A. Two conformations of p38γ(7–352) in the crystal asymmetric unit: open (green) and compact (yellow). B. Superposition of the open and closed structures with the crystal structure of activated p38γ (1CM8) (red), aligned on the C-domain. C. Close-up of the superposition showing the extent of the movement of the N domain towards the C domain. A domain closure of 2.4Å is observed in the compact conformation (compared to the open state), while the activated structure (1CM8) has a 4 Å domain closure, measured (dashed lines) between the top and bottom of the active site (as defined by the side chains of Lys56 and Asp153, shown in stick representation). Compaction occurs by pivoting around a hinge loop (arrow).

Although the two molecules in the asymmetric unit have the same overall fold, they are not identical (Figure 2A). Alignment of the C-terminal lobes (residues 116 to 280) reveals that the relative position of the domains in one molecule is typical of an inactive kinase (open conformation), while their position in the other molecule resembles an active kinase (compact conformation), formed by pivoting of the N-lobe around hinge residues 111 to 114 (Figure 2B, C). The overall Cα RMSD between the two molecules in the asymmetric unit is 0.67Å. However, the structural differences are amplified in the active site, where tilting of the N-lobe towards the C-lobe partially closes the active site, decreasing the distance between active-site residues in the N-lobe (Lys56) and C-lobe (Asp153) by 2.4 Å (Figure 2C). Comparison with the structure of activated p38γ shows that the domains of the compact molecule (yellow) occupy an intermediate position between the open state (green) and the active, phosphorylated conformation (red) (Figure 2B, C). Thus, the two conformations in our p38γ crystal structure represent the canonical, open inactive apo state and a more compact state that is closer to the activated conformation. Crystal contacts are similar for both molecules in the asymmetric unit (Figure S4), suggesting that the compact active-like conformation represents an accessible state within the conformational ensemble of inactive apo p38γ.

The compact conformer of inactive apo p38γ has many hallmarks of an active state. This includes reorientation of the αC helix (αC helix-in), which breaks the Arg73(αC) - Asp331(L16 loop) hydrogen bond (Figure 3A) and allows the αC helix to move towards the active site (Figure 3B). Interestingly, both the compact and open conformations can form a Lys56(β3)-Glu74(αC) salt bridge, but the movement of αC in the compact state results in relative side chain orientations that resemble more closely those of the activated form (Figure 3B). A conserved hydrogen bonding network between the His side chain of the HRD motif and the backbone of the DFG motif,⁶² found predominantly in active kinases, is also present in both compact and open conformations.

Figure 3.

A. αC-L16 interactions. A salt bridge formed between Asp331 and Arg73 in the open form of apo p38γ (green) results in the αC helix being further away from the core of the protein and opening of the active site. Breakage of the salt bridge is observed in activated p38γ (1CM8, red). In activated p38γ (1CM8, red), Arg73 forms a salt-bridge with phosphorylated Thr183 instead of Asp331, leading to closure of the active site. The Arg73 side chain of the compact state of apo p38γ (yellow) is oriented away from Asp331, similar to the activated structure. B. The Lys56/Glu74 salt-bridge links the β3-strand to the αC helix, stabilizing the N-lobe as a rigid body. Both open and compact conformations of p38γ can form the salt bridge.

p38γ dynamics by NMR relaxation

R₁ and R₂ relaxation rates and {¹H}−¹⁵N heteronuclear NOEs were measured for inactive p38γ labeled uniformly with ¹⁵N (Figure S5, Table S2). Model-free analysis gave a rotational correlation time of 31ns, in agreement with the estimated rotational correlation time, 31.5ns, obtained from hydroNMR⁶³ using a full-length p38γ model. The excellent agreement between these values confirms that p38γ is monomeric in solution. The generalized order parameter, S², a measure of the amplitude of amide-bond vector motion on a ps-ns timescale, the correlation time for internal motions of the bond vector, T_e, and exchange contributions, R_ex, were extracted and are shown in Figure 4A and Table S3. In the model-free formalism, S² values range from 0 to 1, where 0 indicates a fully disordered bond-vector and 1 represents a completely rigid bond-vector.⁶⁴ Model-free analysis for p38γ yields physically meaningful order parameters (S² < 1, with S² ~ 0.8 for ordered secondary structure elements and 0.5 < S² < 0.8 for loop regions) that are mapped onto the structure in Figure 4B. The high S² values (average S² = 0.94) for residues 10–16 confirm that the N-terminal β-strand, which is domain swapped in the crystal, is folded intramolecularly in monomeric p38γ in solution.

Figure 4.

Model free parameters extracted from ¹⁵N relaxation data. A. The squared generalized order parameter (S²), internal correlation time of the amide bond vector (τ_e), and chemical exchange contribution (R_ex) for p38γ. B. The squared generalized order parameter (S²) from model-free analysis of the ¹⁵N relaxation data is mapped on the structure of p38γ. Residues with S² values <0.6 are colored red, with S² between 0.6 and 0.8 in yellow, and with S² >0.8 in blue. Unassigned regions or residues that could not be fit, due to poor data or spectral overlap, are shown in gray.

Exchange between open and compact states measured by methyl relaxation dispersion

To investigate whether conformational fluctuations between the open and compact states of apo p38γ occur in solution, NMR relaxation dispersion measurements were performed. By fitting relaxation dispersion data to appropriate conformational exchange models, information is obtained on the differences in chemical shift (Δω), and hence differences in structure, of the exchanging conformers and on the kinetics and thermodynamics of the process. Due to exchange broadening of many backbone amide resonances, methyl ¹³C-¹H multiple quantum relaxation dispersion was measured at 283K and 293K to detect conformational sampling on a μs-ms timescale.⁴⁶ Curves that fit better to a model incorporating exchange contributions than to a flat line were identified using an F-statistic that gave a confidence level of 99.9% (Tables S4 and S5). Representative dispersion curves for a subset of Ile (δ), Leu (δ₁, δ₂), Val (γ₁, γ₂), and Met (ε) methyl groups are shown in Figure 5A (a complete set of curves is shown in Figure S6). All dispersion curves were fit to a two-site exchange model, where the protein undergoes conformational exchange at rate k_ex between its ground state and an excited state with population p_B and chemical shift differences Δω_H and Δω_C. The data were fitted with Δω_H = 0 since preliminary fits in which Δω_H was allowed to vary did not improve fits above a 95% confidence level. Since the dynamic process is in the fast exchange limit, where k_ex > Δω_C, fits of population (p_B) and chemical shift (Δω_C) become difficult to deconvolute and therefore ϕ^1/2 is reported, where ϕ^1/2 = (p_Ap_BΔω_C²)^1/2.

Figure 5.

Exchange process in apo p38γ. (A) Representative methyl ¹³C-¹H multiple quantum relaxation dispersion profiles at 800MHz (red) and 500MHz (black) at 283K. The solid lines show independent fits to a two-site exchange process for each residue. (B) Exchange contributions (R_ex) at 283K mapped onto a structure of p38γ in which the activation loop (residues 175–186) has been modeled. R_ex is plotted on a continuous scale from yellow (R_ex = 4 s⁻¹) to red (R_ex > 8 s⁻¹).

The R_ex values obtained from fits of the methyl relaxation dispersion curves at 283K are plotted on the structure of inactive p38γ in Figure 5B. Conformational fluctuations on the μs-ms timescale extend from the L16 loop and αC helix to the active site, F-helix, through the GHI subdomain and into the MAPK insert at the end of the C-terminal domain (Figures 5B and S7). This dynamic core spans the protein and is consistent with previously described allosteric information flow in p38γ.³⁸ Significantly, the δ₁ methyl of Leu77 in the αC helix, which reports on changes in the configuration of the DFG loop, and the δ methyl of Ile149 in the HRD loop, which reports on changes in the HRD, DFG, L16 loops and αC helix, both experience conformational exchange.

Residues exhibiting μs-ms conformational exchange, determined from methyl relaxation dispersion measurements, are distributed throughout p38γ (Figure 5B, Figure S7). We asked whether the measured methyl relaxation dispersion is reflecting a concerted conformational change. Assuming that exchange is between two predominant states, a ground state (A) and a transiently populated excited state (B), data were fitted using the Bloch-McConnell equations.⁴⁶ Exchange parameters and model selection are shown in Table S4. The exchange rates between states (k_AB+ k_BA ~ 1000 s⁻¹ at 283K) are very similar for all dispersive methyl resonances. Indeed, for most residues, the data can be fit to a global exchange rate of 960 ±30 s⁻¹ (Figure S6). The F-values from the global fit were never more than 2 times the F-values for the individual fits, indicating the global fit was no worse than the individual fit at a 95% confidence level. Globally fit exchange parameters are listed in Table 1. Residues that do not fit the global process, Leu225, Met182 and Met262, are in the αF-αG loop, activation loop and MAPK-insert respectively, and have independently fitted exchange rates of ~700 s^-1. Likewise, at 293K, a global fit of dispersive curves yields an exchange rate of 2300 s⁻¹ (±120) (Table S6 and Figure S6). At 293K, the dispersion curves for all residues can be fit to a single global process.

Table 1.

Global fit parameters obtained for apo p38γ methyl relaxation dispersion data at 283K

Methyl	kex(s−1)		Rex(s−1) 800MHz	Φ1/2 (s−1)	$\frac{{\chi }^2(\mathbf{\text{global}})}{{\chi }^2(\mathbf{\text{ind}})}$	location/interaction
Leu77-δ1	960 (30)		22.2	11.8 (1.1)	1.1	αC/ DFG
Val86-γ2	“	“	3.5	4.5 (0.2)	1.1	αC / αE / αC-β4 loop
lle87-δ	“	“	6.3	6.0 (0.4)	1.2	αC-β4 loop / gatekeeper / DFG
Met120-ε	“	“	3.2	4.2 (0.4)	1.2	αD/ αF/ docking
Met137-ε		“	7.7	6.7 (0.5)	1.3	αE/ αF
llel 49-δ	“	“	54.5	20.5 (1.4)	1.0	HRD/ αC
llel 50-δ	“	“	9.7	7.5 (0.4)	1.1	HRD/ αEF
lle209-δ	“	“	36.5	15.7 (1)	1.2	αF / GHI
Met219-ε	“	“	6.1	5.9 (0.5)	1.5	αD / αE/ αF
Leu288-δ1	“	“	29.1	13.7(0.8)	1.1	αF / αH
Leu288-δ2	“	“	17.3	10.2(0.7)	1.5	αF /αH
Met291-ε	“	“	5.8	5.8 (0.3)	1.6	αF/ αH/ αl
Leu292-δ1	“	“	17.5	10.3 (1.1)	1.5	αF/ αH/ αF-αG loop
Val293-γ1	“	“	3.2	4.2 (0.3)	1.0	αH-αl loop
Leu294-δ2	“	“	73.3	26.0 (2)	1.0	αG / αH-αl loop / MAPK insert
Val323-γl	“	“	3.7	4.6 (0.3)	1.9	αE/L16/αC-β4loop
Met182-ε	690(10)		28.8	28.2 (0.5)	-	activation loop
Leu225-δ1	660 (60)		150.2	35.1 (23.9)	-	αF / αl / αF-αG loop
Met262-ε	560 (40)		39.3	20.7 (0.8)	-	αG / αH-αl loop / MAPK insert

k_ex , Δω_C, and p_B were fit globally with a 2-site exchange model. Φ=p_Ap_BΔω_C². Methyls whose dispersion profiles could not be satisfactorily fit to the global process (where χ²(global)/χ²(ind) > 3.5) are highlighted in purple. These dispersion curves were fit individually.

Methyl relaxation dispersion data were also acquired for p38γ bound to the DFG-out inhibitor BIRB796 and fitted to a two-state conformational exchange process (Figure S8 and Table S7). The data were best fit by two clusters with different exchange rates. One corresponds to the previously described global fluctuation, but for a smaller cohort of residues within the C-lobe and with a slightly decreased exchange rate of 1900 s⁻¹ (±70) vs 2300 s⁻¹ (±120) for apo kinase at 293K (Figure S8, red). The second involves a cluster of residues (Met120, Leu119-δ2, and Met219) near the docking site that undergo slower conformational fluctuations with a rate of 310 s⁻¹ (±60) (Figure S8, yellow). Methyl groups of Leu77, Met109, and Ile169, located near the BIRB796 binding site, show altered dynamics; the CPMG dispersion observed for these methyls in apo p38γ is lost but large increases in R₂₀ indicate exchange broadening from conformational fluctuations on a different timescale. Decreased relaxation dispersion (flat dispersion profiles or decreased Rex values) is also observed for Ile149, Met182, Ile209, Leu265, and Leu294 (Figure S9), showing that the perturbations in apo p38γ dynamics caused by inhibitor extends from the active site into the C-lobe, following the previously described allosteric network³⁸ (Figure S9). However, other sites, which are mainly localized to loops in the C-terminal domain, show increases in R_ex, suggesting a decoupling of global dynamics.

ATP binding and enzymatic activity

The side-chain of Met112, in the hinge region between the N- and C-lobes, is rotated away from the ATP pocket in the crystal structures of the open, compact, and activated states of p38γ but projects into and blocks the ATP site¹⁹ in inactive p38α(Figure 6A). The outward position of Met112 suggests that inactive p38γ may have higher affinity for ATP than p38α. In addition, the glycine-rich loop (residues 34 to 39) of the compact state closes around the ATP binding pocket in a similar conformation to the activated structure^14,19 (Figure 6A).

Figure 6.

ATP binds to inactive p38γ. A. Region of the structures of open (green), compact (yellow), active (1CM8, red) and inactive-apo p38α(1WFC,¹⁷ magenta). In all of the p38γ structures, Met112 in the hinge is rotated out of the ATP binding pocket, whereas in p38α (1WFC, magenta) Met112 blocks the pocket. The Gly-rich loop in p38γ is also closed relative to inactive p38α, a requirement for nucleotide binding. B. Region of the ¹H- ¹³C methyl HMQC spectrum of inactive p38γ showing methionine methyl resonances and chemical shift perturbations due to ATP binding. Met81 and Met112 show the largest changes in chemical shift upon addition of ATP.

NMR was used to determine the affinity of ATP binding to apo p38γ. Chemical shift perturbations were observed for methyl groups of residues in the ATP binding site when p38γ was titrated with ATP, indicating specific binding (Figure 6B). The chemical shift titration curves were fit to a one-site binding model, showing that inactive p38γ binds ATP with sub-mM (770μM ± 140μM) affinity (Figure S10). This contrasts with the reported K_d (~10mM) for binding of ATP to inactive p38α under similar temperature and buffer conditions.⁶⁵ Inactive p38γ has affinity for ATP on the same order of magnitude as activated p38α(0.4mM).⁶⁵

The observation that inactive p38γ binds ATP and can adopt a compact conformation reminiscent of the activated state led us to measure its ATPase activity; however, we found no detectable hydrolysis of ATP (Figure S11). After activation of the p38γ by phosphorylation, ATP hydrolysis occurred with a k_cat/K_M of ~7000 M⁻¹ s⁻¹ (Figure S11), similar to previously reported values.⁶⁶

DISCUSSION

Organization and allostery in kinases

The organizing principles of an active kinase involve compaction of the active site by movement of the N and C lobes, αC helix, and Gly-rich loop, accompanied by reorientation of the catalytic side chains and packing of spatially conserved hydrophobic residues.¹ Some of the characteristic features of active kinases are present in both the open and compact conformations of inactive p38γ. These include a canonical positioning of the αC helix, packed into the active-site of the kinase (helix-in). In inactive kinases, the αC helix is typically partially disordered or oriented away from the core of the protein. The αC helix-in orientation is stabilized by the Lys56(β3)/Glu74(αC) salt bridge (Figure 3B) and is essential to organize the active site for catalysis. The αC helix-in orientation and Lys56(β3)/Glu74(αC) salt bridge are typically only found in active kinases and are believed to be required for the open-to-closed breathing motion observed during the catalytic cycle.¹ In addition, inactive apo p38γ has a pre-formed, spatially conserved network of closely packed residues (Figure 7, termed the R-spine¹), that reports on domain compaction and organization of the active-site. The R-spine, which is formed by hydrophobic interactions between Phe172 in the DFG motif, His151 of the HRD motif, Leu78 of the αC helix, and Leu89 of the αC-β4 loop, functions to transfer allosteric signals. ^38,67 The R-spine is almost always disrupted in inactive MAPKs, as shown for inactive p38α in Figure 7. However, in both the open and compact conformations of apo p38γ, the R-spine is completed through hydrophobic interactions between Phe172 and Leu78, enabled by re-orientation of the Phe172 side chain from its position in inactive p38α and by the Lys56/Glu74 salt bridge that compresses the active-site (Figure 7). In the activated state of p38γ, the side chain of Phe172 re-orients to pack more closely against Leu78, to accommodate greater compaction of the active site (Figure 7).

Figure 7.

A sterically connected, spatially conserved network of hydrophobic side chains (the R-spine), indicative of an activated kinase (1CM8, red), is formed in both the open and compact structures (green/yellow) of inactive p38γ. This is in contrast to the incomplete R-spine formation typically observed in inactive kinases (p38α:1WFC, magenta). Electron density for DFG-pucker, characteristic of active kinases, is only observed in the compact state of inactive p38γ.

Asn158 and the DFG loop act as a molecular switch

In the open and compact structures of apo p38γ, the conserved Asn158 side chain makes different interactions with the side chains of Asp171, a Mg₂ATP binding residue in the DFG loop, and Asp153, the catalytic base in the HRD motif. In the open structure, the side chains of Asn158 and Asp171 are within hydrogen bonding distance (2.9Å), stabilizing the Asp171 side chain in an inactive conformation that partially occludes the Mg₂ATP binding site (Figure 8). Upon activation, the Asn158 side chain is rotated ~180˚ about χ₂, which breaks the hydrogen bond to Asp171 and places the amide nitrogen within hydrogen bonding distance of the catalytic base, Asp153 (Figure 8A).¹⁴ This conformational change is facilitated by compaction of the active site, which decreases the distance between the side chain amide of Asn158 and the Asp153 carboxyl to 2.8Å versus 4.4Å in the open form. The breaking of the hydrogen bond between Asn158 and Asp171 leads to a shift in the Asp171 side chain that opens the ATP site (Figure 8B). As a result, the Asp171 carboxyl in activated p38γ forms a hydrogen bond with the backbone amide of Gly173(DFG), which stabilizes a puckered conformation of the DFG loop (Figure 8B, red). The structure of the compact state of apo p38γ has features of both the open (inactive) and activated forms. In fitting the electron density for the compact (active-like) molecule, we initially modeled an inactive DFG loop conformation. However, additional electron density is present in this region (Figure 8C), into which an active-like DFG loop pucker can be built. This extra electron density is not observed for the other molecule in the asymmetric unit, in the open conformation. In the compact structure, electron density for two χ₁ rotamers is observed for the Asp171 side chain, corresponding to inactive and active-like conformations. However, the distance between the Asp171 carboxyl group and the Gly173 amide in the latter conformation is longer than in the activated p38γ (4.3Å versus 3.0Å) and precludes formation of a hydrogen bond to stabilize a puckered DFG loop (Figure 8). The observation of additional electron density suggests that the Asp171 side chain and the entire DFG loop in the compact state samples both the open and active-like configurations, linked to the orientation of the Asn158 side chain. Thus, it appears that puckering of the DFG loop is associated with the N/C lobe compaction.

Figure 8.

Asn158 acts as a switch between inactive and active-like catalytic site configurations. The open (green), compact (yellow), and activated (1CM8, red) structures are shown for comparison. Hydrogen bonding (dashed lines) between Asn158 and either the catalytic base Asp153 (compact/active) or the ATP coordinating Asp171 (open) reorganizes side chains into positions conducive for or blocking catalysis, respectively. (A) In the open state (green), the Nδ of Asn158 is in hydrogen bonding distance (*, 2.9 Å) to the Asp171 carboxyl and positions the side-chain so that it blocks the ATP binding site. In this state, the active site is more open, with Asp153 not in position for catalysis. In the compact state (yellow), the side chain amide group of Asn158 is within hydrogen bonding distance to Asp153 (**, 3.2 Å versus 4.4 Å in the open state and 2.8 Å in activated p38γ), where the carboxyl moves towards the active site while also freeing the side-chain of Asp171 to move away. (B) The Asp171 side chain position in the active form opens the ATP pocket and correctly positions the sidechain for binding Mg₂ATP and hydrogen bonding with Gly173. The distance between the Asp171 carboxyl and the Gly173 backbone NH decreases in the compact state (yellow) but remains longer (•, 4.3 Å versus ••, 3.0 Å) than in activated p38γ (1CM8, red). (C) Asp171 in the compact state can be built in two χ₁ rotamer conformations. Only the inactive DFG conformation is built in the deposited compact structure (left, dark yellow), but a puckered conformation can be built into the empty density (right, bright yellow). Electron density for an active-like DFG-pucker is observed only in the compact state. The DFG-pucker as modeled in the alternate conformation of the compact state is shown in A and B.

Conserved structural mechanism for kinase activation.

High-resolution structures of eukaryotic kinases in the Protein Data Bank with 30% or greater sequence identity to p38γ (Table S6) were analyzed to determine whether puckering of the DFG loop is a common feature of activation. Active and inactive kinase structures have similar backbone dihedral angles, with exceptions predominantly in the DFG loop. The puckered conformation of the DFG motif (Figure S12) is characteristic of activated kinases and has distinctive backbone dihedral angles for Phe172(DFG) (active: ϕ=−94° ± 6, ψ=22° ± 11) and Gly173(DFG) (active: ϕ=−62° ± 12, ψ=−26° ± 8). The structures of inactive p38 have a wider distribution of DFG dihedral angles compared to active kinases but have a clustered set of dihedral angles for Phe172 (inactive: ϕ=−76° ± 23, ψ=143° ± 24) and Gly173 (inactive: ϕ=69° ± 8, ψ=11° ± 12) that are also observed in the broader family of inactive kinases (Figure S12). The open and compact p38γ structures, with the alternate DFG configuration modeled for the latter, have characteristic inactive and active state DFG dihedral angles, respectively (Figure S12). Other MAPKs, such as ERK2 (eg PDBID: 1ERK⁶⁸) and Fus3 (eg PDBID: 2B9F⁶⁹), exhibit DFG-pucker even in their inactive states, suggesting that a DFG unpuckered to puckered transition is not a conserved regulatory mechanism of activation even among the MAPK family.

Backbone dynamics

The ¹⁵N relaxation measurements reveal the presence of backbone dynamics in the switch and loop regions that mediate the conformational change between active and inactive forms of p38γ. The L16 loop (residues 316 to 323) and residues 328, 330, and 333 have overall low order parameters (average S² = 0.6) and internal correlation times (τ_e) that indicate motions on a ns timescale (Figure 4). Electron density is missing or poor for the L16 loop in our crystal structure of apo p38γ and that of the activated p38γ(1CM8). Destabilization of interactions between the L16 loop, specifically through Phe330, and the αC helix has been proposed to lead to reorientation of the αC helix into an active conformation.²⁰ Large amplitude fluctuations on the nanosecond timescale are observed in the hinge region [residues 111, 112 (S² =0.55), and 114] between the N/C lobes, suggesting that it may function as a pivot around which domain compaction occurs. In addition, the relaxation data for Gly173(DFG) show a large contribution from conformational exchange (R_ex = 10.7 s⁻¹) and a low S² (0.63), potentially reflecting fluctuations in dihedral angles required for the DFG pucker (Figure 4A & Table S2). Residues 181 and 182 in the activation loop are also highly dynamic, with low S² (0.55). Unfortunately, backbone resonance assignments are missing for most of the activation loop (residues 187–205) due to exchange broadening indicative of intermediate timescale dynamics. This region also has poor or missing electron density, with elevated B-values for residues 187–203. The high flexibility of the activation loop likely allows access for activating kinases.

Conformational fluctuations between open and compact structures

The methyl relaxation dispersion data for apo p38γ reveal the presence of μs-ms time scale conformational fluctuations in the active site and throughout the hydrophobic core of the C-lobe (Figure 5B, Figure S7). If the exchange process measured by relaxation dispersion is related to inter-conversion between the open and the compact, active-like conformation observed in the apo p38γ crystal structure, we would expect the chemical shift difference between exchanging states (Δω_C) determined from fits of the relaxation dispersion data to correlate to the chemical shift differences between the crystallographic conformations. Since the process approaches the fast exchange limit, we report ϕ^1/2 which is proportional to Δω_C (ϕ^1/2 = (p_Ap_BΔω_C²)^1/2). We used SHIFTX2 software⁷⁰ to predict methyl ¹³C chemical shifts from the open and compact structures, with the alternate conformation of the DFG loop modeled in the compact state. The predicted shift differences between open and compact forms of p38γ correlate well (R = 0.84) to values of ϕ^1/2 derived by fitting the relaxation dispersion data at 283K (Table 1) for residues involved in the global process (Figure S13). The correlation between the chemical shift differences predicted directly from the two apo p38γ structures obtained by crystallography and the chemical shift difference between the ground and excited states fitted from the NMR relaxation dispersion experiments is a strong indication that the alternate conformation of the compact crystal structure is representative of the excited state observed in the NMR experiments. We conclude that inactive p38γ in solution undergoes a concerted breathing motion between the canonical inactive conformation and a small population of an excited state with a conformation similar to the compact, active-like structure; fortuitously, both states are stabilized in different molecules in the asymmetric unit of the crystal.

Disruption of DFG-pucker and coupled breathing motions

To further investigate the interplay between dynamics of the hydrophobic core of apo p38γ, DFG pucker, and the open-closed breathing motion, we repeated the methyl relaxation dispersion measurements in the presence of a bound DFG-out inhibitor, BIRB796. DFG-out inhibitors prevent the activation of MAPKs by changing the conformation of the DFG loop to disrupt the active-site geometry in the activated state. ⁷¹ If the dynamic breathing of p38γ revealed by the relaxation dispersion experiments is coupled to DFG pucker, then binding of BIRB796 and stabilization of the DFG-out conformation would be expected to disrupt the global process observed in the apo state. This is indeed observed; binding of BIRB796 alters the timescale of conformational fluctuations in the inhibitor binding pocket (Leu77, Met109, and Ile169). The dynamic perturbations propagate from the active site through much of the C-lobe, resulting in loss of dispersion or decrease in the magnitude of the exchange contribution R_ex for many methyl groups. Binding of inhibitor suppresses dispersion of Met182 in the activation loop, likely reflecting the expected stabilization of a DFG-out/activation loop configuration by BIRB796, and decouples the fluctuations in the docking site on the C-lobe (Leu119, Met120, and Met219) from the global process. The partitioning of the global dynamic process into two distinct processes in the BIRB796 bound state, slow exchange in the docking site and faster exchange in the C-lobe, suggests that binding of the DFG-out inhibitor decouples motions in p38γ by disruption of the DFG-pucker. A nearly identical phenomenon was observed in network analysis of DFG-out p38γ chemical shift perturbations. ³⁸

Coupling of DFG loop pucker to domain compaction

The correlation between the chemical shift differences predicted from the crystal structures of the open and compact states and those derived from the NMR relaxation dispersion experiments shows that apo p38γ in solution fluctuates between an open ground state and a weakly populated, active-like structure that is compact but requires further stabilization of the closed conformation to achieve a fully active state (Figure 2B). The DFG-loop, in particular, samples a conformation that resembles the active structure, where the loop is puckered but the χ₁ rotamer of Phe172(DFG) remains in the open conformation, pointing directly towards Met81 at the end of the αC helix (Figure S14). After activation by phosphorylation of Thr183 and Tyr185, the χ₁ angle of Phe172 changes and reorients the side-chain so that it is parallel to the αC helix, allowing reorganization of the Met81 side-chain and leading to greater compaction of the lobes and re-positioning of the αC helix.¹⁴ Interestingly, no measurable relaxation dispersion is observed for Met81 in the inactive apo state of p38γ, which implies that the open to compact exchange process does not result in reorganization of the sidechain of Met81 to allow interaction with Phe172. However, Leu77 and Ile149, which report on the DFG loop structure and αC orientation, exhibit methyl relaxation dispersion that arises from a global process involving conformational exchange between open and compact states (Figures 5B, S13). In addition, backbone relaxation experiments show a low order parameter (S² = 0.63) and conformational exchange contribution (R_ex = 11 s⁻¹) for the amide of Gly173(DFG), which undergoes a large change in bond vector orientation and chemical environment upon DFG hairpin formation and H-bonding to Asp171 (Figure 4 and Table S2).

In receptor tyrosine kinases, the transition between DFG in and out states has been linked to the nucleotide binding kinetics.³¹ DFG-pucker, albeit more subtle, is also linked to ATP binding to apo p38γ. ATP binding to activated p38α is required to attain a catalytically active conformation.⁷² Although apo p38γ is able to bind ATP, with affinity on the same order of magnitude as reported for activated p38α,⁶⁵ the enzyme remains inactive and exhibits no ATPase activity (Figure S11). Stabilization of the compact active state, e.g. through phosphorylation, is therefore absolutely required for activity. Phosphorylation of the activation loop would shift the αC helix towards the active site and bring the N and C lobes closer together, priming p38γ for catalysis. The breathing motions of inactive p38γ, sampling a state that approaches the active state, facilitate ATP binding while still preventing activity. This suggests a process of conformational selection in which the protein dynamically samples a compact state that appears to be on-pathway to the fully active conformation.

The observation of concerted motions in inactive p38γ contrasts with reports of non-global dynamic processes in inactive ERK2 and p38α.^5,73 Interestingly, both conformers observed in the X-ray structure of apo p38γ show evidence of a salt bridge between Lys56(β3) and Glu74(αC) (Figure 3B), which is normally formed only in active kinases. In activated p38α, this salt-bridge is formed only after substrate binding.^13,19 The Lys56(β3) / Glu74(αC) salt bridge not only acts to compact the active site and position Lys56(β3) for interaction with nucleotide but also plays a role in rigidifying the N-terminal domain by linking the αC helix and N-lobe β-sheet. It has been proposed that rigid body movement of the N-lobe is required for the breathing motion observed in active kinases during the catalytic cycle.¹ The concerted dynamics are likely facilitated by a network of coupled hydrophobic residues. Network analysis of methyl chemical shift perturbations in inactive apo p38γ identified a hydrophobic allosteric network that extends from the N-lobe, through the active site, to the F-helix, and into the GHI subdomain and MAPK-insert. ³⁸ The results here show that methyl groups in this network experience a global conformational exchange process that occurs on a millisecond timescale: Leu77(αC), Ile149 and Ile150 in and around the R-spine, Met219 and Ile209 in the F-helix, Met291, Leu288, Leu292, and Leu294 in the GHI subdomain, Met120 near the docking site, and Met262 in the MAPK-insert (Figure S7). Disruption of the hydrophobic network by a DFG-out inhibitor leads to a corresponding loss of concerted motion (Figures S8, S9), resulting in clusters of residues with distinct dynamics that are identical to the isolated communities found previously by information flow analysis of chemical shifts.³⁸

Conclusion

This work shows that inactive apo p38γ has a globally dynamic core that exchanges between an inactive open conformation and a weakly populated compact state with an active-like conformation. The transition between these states occurs through reorganization of the DFG loop, which is coupled to rearrangement of catalytic residues into an active-like orientation and opening of the nucleotide binding pocket. The sidechain of Phe172 in the DFG loop of inactive apo p38γ adopts a conformation that connects the N and C lobes by a spatially conserved hydrophobic network, which is typically non-continuous in inactive kinases.¹ DFG loop reorganization and domain compaction dynamics are expected to propagate through this network. Fast dynamics in the L16 loop disrupt stabilizing interactions with the αC helix, thereby enabling the helix to reorient and compact the N-lobe. Even though apo p38γ dynamically samples a state that has structural features similar to those of the activated enzyme, the fully compact conformation associated with an active kinase remains inaccessible. The activated state requires tighter packing between the N- and C-lobes, mediated by interactions between the αC helix and the phosphorylated activation loop. Populating an active-like, but still inactive, state allows p38γ to bind ATP and potentially interact with upstream activators and partner proteins while still being incapable of ATPase activity. Phosphorylation of the activation loop, or lipid binding to the MAPK insert, is also required to open the substrate binding pocket.^11,19,74 The dynamic model described here is consistent with the requirement for p38γ to remain tightly regulated in an inactive state until an upstream signal induces activation. Compared to the lack of global dynamics in inactive ERK2 and in inactive and active p38α, the dynamics observed for apo p38γ suggest different mechanisms for regulation even within the same kinase family.

Data Deposition

The atomic coordinates and structure factors have been deposited in the Protein Data Bank with accession number 6UNA.