NMR Characterization of Information Flow and Allosteric Communities in the MAP Kinase p38γ

The intramolecular network structure of a protein provides valuable insights into allosteric sites and communication pathways. However, a straightforward method to comprehensively map and characterize these pathways is not currently available. Here we present an approach to characterize intramolecular network structure using NMR chemical shift perturbations. We apply the method to the mitogen activated protein kinase (MAPK) p38γ. p38γ contains allosteric sites that are conserved among eukaryotic kinases as well as unique to the MAPK family. How these regulatory sites communicate with catalytic residues is not well understood. Using our method, we observe and characterize for the first time information flux between regulatory sites through a conserved kinase infrastructure. This network is accessed, reinforced, and broken in various states of p38γ, reflecting the functional state of the protein. We demonstrate that the approach detects critical junctions in the network corresponding to biologically significant allosteric sites and pathways.


NMR spectroscopy
Ile, Leu, and Val methyl side-chain resonances were assigned using methyl to backbone correlation experiments with specifically labeled methyl residues 2 . 3D side-chain correlation spectra were acquired at 293K using 30-50% non-uniform sampling (NUS) on a Bruker Avance III 800MHz spectrometer equipped with cryoprobe. Spectra for backbone assignments were acquired at 293K on Bruker Avance 750MHz and 900MHz spectrometers equipped with triple resonance gradient probes. Amide 1 H- 15 N, 13 C CO, Cα, and Cβ resonances were assigned using 3D TROSY HNCA, HNCACB, HNCOCA, and HNCO spectra acquired with 30-50% NUS.
Sinusoidal weighted Poisson gap sampling schemes were generated as in 3 and optimal schemes were selected 4 for use in the 13 C indirect dimension. The 15 N constant time dimension was acquired with uniform sampling. NUS was implemented on 750MHz and 900MHz spectrometers by modification of the pulse program to include explicit looping based on an increment list corresponding to the NUS scheme. Mutagenesis described in the main text was used to supplement assignments for residues that lacked backbone assignments or which had degenerate Cα and Cβ chemical shifts, and also to assign Met ε-methyls. A 13 C NOESY experiment was also acquired to aid in side chain assignment. Stereospecific methyl assignments were made with methyl CHD2 labeled samples using a constanttime 13 C HSQC experiment as described in 5,6 .
Chemical shift assignments Assignment of methyls by triple resonance requires prior assignment of Cα and Cβ chemical shifts. The 1 H-15 N TROSY spectrum of His-tagged apo p38γ contains 364 out of an expected 368 backbone resonances (Fig. S1a). Although a high percentage of expected cross peaks are observed, many peaks appear to be undergoing exchange broadening, and their cross peak intensities are near the noise and their linewidths are broad. As a consequence only 285 amide resonances or 77% of the backbone, excluding the His tag and prolines, could be assigned (Fig. S2). The backbone resonances of most regulatory regions of p38γ were assigned, including the activation loop, DFG loop, αC-helix, αF-helix, MAPK insert, and docking site.
Assignments for 121 or 73% of the methyl resonances were made via a tripartite method that combined triple resonance experiments correlating the assigned backbone chemical shift to the methyl resonances, 13 C NOESY spectra to establish through-space connectivities, and single site mutagenesis (Fig. S2). All 13 Met εmethyls, as well as all 13 Ile δ-methyls were assigned. One or both δ and γ methyls were assigned for 32 out of 41 leucines and 24 out of 28 valines, respectively.
Stereospecific assignment for 90% of the Leu and Val methyl assignments was accomplished. Assignments were transferred to the spectra of inactive ATP bound, inactive BIRB796 bound, and activated apo states.

Changes in highly connected residues
Most residues with a high number of network connections (in the top third) in inactive apo p38γ surround the active site and docking site (Fig. S6). To maintain signaling homeostasis, inactive p38γ must remain catalytically inert yet exist in a state that is primed for activation by phosphorylation. This is clearly reflected in the network; the active site is highly connected and is sensitive to interactions throughout the protein. The docking site is also highly connected, consistent with its role in recruiting upstream kinases and facilitation of the conformational rearrangement of the activation loop necessary for phosphorylation 7 . Other regions of high connectivity correspond to elements involved in non-canonical activation, such as the MAPK insert and the αC-helix [8][9][10] .
The overall network structure and flow suggests that the C-lobe in the inactive ATP bound and activated states has regulatory significance, as indicated by an increase in the number of highly connected residues in the C-lobe and a decrease in the Nlobe, relative to the inactive apo state (Fig. S6). The binding of ATP to inactive p38γ likely stabilizes a closed conformation, similar to that observed in activated kinases 11,12 . However, activation leads to proper packing of active site residues and further used to determine if an adequate number of mutants are being examined for the network analysis of a system.
To determine the minimum number of mutations required for accurate description of the networks of p38γ, datasets for individual mutations within the apo inactive state of p38γ were randomly removed from the analysis until the original network structure was no longer reproduced. With the 78 observed methyl perturbations at least 11 mutations were necessary to reproduce the principal features of the network structure described in the main text (which used 20 mutants). See Figure   S9 for details of the network structure derived from 11 mutations (L77V, M109L,  The 13 mutations used in the inactive ATP bound, BIRB796 bound, and activated states are shown as green triangles. Figure S3. Link communities of inactive apo p38γ. Overlapping communities and hierarchy were identified by the link community method 22 23 from two-point correlations of chemical shift perturbations resulting from conservative mutagenesis. a) Communities are grouped links (paired residues). The partition density function described in 22 was used to determine a cut-off (red dashed line) to obtain the 4 communities. b) Link communities in a) plotted by color on the structure of p38γ. The communities are similar to those in Fig. 3 and correspond to the MAPK-insert (red), N-lobe (green), entire protein (purple), and the activesite/docking-site (yellow). Some communities identified in Fig. 3 are not identified here, such as the R and C-spines, and the C-lobe, instead they are included in a single community encompassing the entire protein (purple), which was not identified in the results presented in Fig. 3. c) Network graph of link communities, demonstrating residue (node) connections and overlap. Nodes (circles) represent residues in p38γ. The community membership of each residue is indicated by piechart color corresponding to communities in a).  The size of the modules represents the amount of flux and connections between residues within the module. Thickness of arrows between communities represents the amount of flux between communities. Using a common set of 13 mutants of inactive apo p38γ (used also for ATP bound, BIRB796 bound, and activated states) in the network analysis yields clustered communities similar to the analysis with 20 mutants described in the main text (see Fig. 3). One noticeable difference is that the module most closely corresponding to the 'R-spine' residues is linked to the C-lobe and MAPK insert in this analysis, whereas the results from 20 mutants has the active site community linked to the C-lobe and MAPK insert. The R-spine and active site communities are overlapped so that this does not reflect a major change in the network structure. The remainder of the network connections are very similar for the analysis performed using data from 13 or 20 mutants. The relative flow between communities is also similar.    derived from a minimal set of mutants. Identified communities from chemical shift perturbation networks are colored based on tertiary structure and regulatory element: N-lobe, yellow; C-lobe, purple; active-site, green; C-spine, blue; R-spine, red; MAPK-insert, lavender. The size of the modules represents the amount of flux and connections between residues within the module. Thickness of arrows between communities represents the amount of flux between communities. Using a minimal set of 11 mutants of inactive apo p38γ in the network analysis yields clustered communities similar to the analysis with 20 mutants (and 13 mutants) described in the text (see Fig. 3 and Fig. S5). One noticeable difference is that the module most closely corresponding to the 'R-spine' residues is linked to the C-lobe and MAPK insert in this analysis, whereas the results from 20 mutants has the active site community linked to the C-lobe and MAPK insert. The R-spine and active site communities are overlapped so that this does not reflect a major change in the network structure. The overall network structure is the same but some communities are split (i.e. using a greater number of mutants the C-lobe is a single community, here it is represented by 3 highly overlapped communities). The relative flow between communities is also similar. Tables S8, S10, S12, S13: Pairwise flow is given as X X Y (pairwise between X and Y residues), three-residue flow is given as X Y Z.

First Res
Second Res

First Res
Second Res