Survey of solution dynamics in Src kinase reveals allosteric cross talk between the ligand binding and regulatory sites

The catalytic domain of protein tyrosine kinases can interconvert between active and inactive conformations in response to regulatory inputs. We recently demonstrated that Src kinase features an allosteric network that couples substrate-binding sites. However, the extent of conformational and dynamic changes that are propagated throughout the kinase domain remains poorly understood. Here, we monitor by NMR the effect of conformationally selective inhibitors on kinase backbone dynamics. We find that inhibitor binding and activation loop autophosphorylation induces dynamic changes across the entire kinase. We identify a highly conserved amino acid, Gly449, that is necessary for Src activation. Finally, we show for the first time how the SH3–SH2 domains perturb the dynamics of the kinase domain in the context of the full length protein. We provide experimental support for long-range communication in Src kinase that leads to the relative stabilization of active or inactive conformations and modulation of substrate affinity.

corresponds to the Gly residue denoted by the arrow. This Gly has the largest height in the stack indicating that it is highly conserved in human protein kinases at this position in the sequence alignment. The height of the letter stack corresponds to a measure of the invariance, whilst the height of each letter within the stack reflects the frequency of that amino acid for that position in the sequence alignment. Numbers below the stacks correspond to the occupancy, which describes the probability of observing an amino acid shown from the stack. The HMM logo was generated using R 1 , R 2 , R 2 :R 1 ratio, and heteronuclear NOE ratio are plotted along with the backbone secondary structure. The R 2 :R 1 ratio analysis is an indicator of the global tumbling. Data points colored in red and red circles under the secondary structure schematic indicate residues which directly interact with dasatinib. Residues which reflect R 1 , R 2 , and R 2 :R 1 values within one standard deviation of the trimmed mean suggests that they behave rigidly and only have significant motion attributed to global tumbling. Hallmarks of enhanced fast internal motion include: R 1, one standard deviation greater than the trimmed mean, and R 2 and R 2 :R 1 values one standard deviation less than the trimmed mean.

Molecular weight evaluation and NMR sample optimization
To ensure that the CSPs and IPs observed in SrcKD could be attributed to input signals affecting  relaxation, promote resonance broadening 10 . Thus, an increase in the conformational exchange contribution relative to the fast internal motion will result in an intensity ratio < 1 (broadening). An increase in the fast internal motion contribution relative to conformational exchange will result in an intensity ratio > 1 (sharpening). Thus, perturbations which cause the intensity ratio for a backbone resonance to deviate from a value of 1 imply such dynamic changes. A ratio of 1 suggests no change in conformational exchange and fast internal motion contributions or that their differential contributions cancel each other out so there is no relative change. Broadening and sharpening can also be attributed to an increase (slow tumbling) or decrease (fast tumbling) in the global correlation time (τ c ). Because τ c is dependent on the molecular weight and hydrodynamic radius of the protein, resonance intensities are sensitive to aggregation/oligomerisation. However, we find that the global correlation time τ c of SrcKD corresponds to that of a monomer under all experimental conditions tested here ( Supplementary Fig. 3, Supplementary Table 2).

Backbone relaxation experiments
T 1 relaxation delays were set to 0.01, 0.05, 0.2, 1, 1.5, 2, 2.5, 3, 4 s, whereas T 2 relaxation delays were set to 8, 16, 24, 32, 40, 48, 56, 64 ms. All delay times were collected in duplicate. The T 1 and T 2 datasets were fitted to monoexponential decay curves described by I(t) = I(t0) x e(-t/T 1 ), and I(t) = I(t0) x e(-t/T 2 ), where I is intensity, in order to determine the T 1 , T 2 time constants. Errors for the T 1 , T 2 time constants were derived from the error associated with nonlinear fitting. A relaxation delay of 5 s was used for the ( 1 H)-15 N heteronuclear NOE unsaturated and saturated experiments. NOE ratios were determined in which the NOE ratio for each resonance = saturated I/unsaturated I. Errors in the ( 1 H)-15 N Heteronuclear NOE experiment were calculated through error propagation using the root mean square noise of the spectra. All spectra were processed using Topspin, with resonance intensities picked, analyzed, and fitted accordingly using CcpNmr Analysis. Backbone amide order parameters S 2 were determined using the RCI server 11 . Model-free formalism from 15 N relaxation data was performed using Fast-modelfree 12 and modelfree4 13 , and their initial inputs were determined from pdbinteria, r2r1_tm, and quadric diffusion. Residues which deviated from the mean R 2 :R 1 ratio by >1.5 S.D, or had large amplitude fast internal motions (NOE values <0.6) were excluded from the input residues used in quadric diffusion.

Hydronmr7c calculations
The τ c for pSrcKD•dasatinib and the diffusion tensor model determined from FAST-modelfree was validated using hydronmr7c 14,15 with the input PDB model 3G5D 2 .

Foldx calculations
The effect of the G449A mutation on the stability of different Src structures was determined by using Foldx 4 to calculate the change in free energy of stability (ΔΔG = ΔΔGmut-ΔGwt (kcal mol -1 )). The input PDBs used were: 1Y57, 2SRC, 3G5D, 4YBJ, 4YBK 1, 2, 16 . Foldx error bars were based on the s.d of performing three runs.