Multidomain assembled states of Hck tyrosine kinase in solution

Classes of Hck Assembly States.

The simplified CG model of Hck, though extremely simplified, incorporates the main generic features of the protein: (i) the native three-dimensional conformations of the SH3, SH2, and catalytic domains from the X-ray structures, (ii) the topological connectivity of the polypeptide chain via flexible linkers, and (iii) the excluded volume from the residue-residue core repulsion. The CG model also accounts for the intramolecular interactions involved in the stabilization of the down-regulated assembled form (3): the tyrosine-containing C-terminal “Ctail” of the kinase domain with the SH2 domain, and the polyproline SH2-kinase “linker” with the SH3 domain. As described in SI Appendix, both interactions were incorporated in the CG model, and adjusted to match the K_d from experiments for the SH2-binding Ctail peptide, and the SH3-binding polyproline peptide. The active and inactive conformations of the catalytic domain are also included in the multistate model as described previously (16, 17) to account for the ability of full-length Hck to adopt different conformations. To ensure proper sampling of various assembly conformations, simulations using different initial conditions were conducted, ranging from fully assembled to disassembled states, from increased to decreased binding interactions, and from destabilizing stable configurations to lowered transition barriers.

A large number of configurations were generated using MD simulations based on the CG model. A two-step clustering scheme was used to organize the large amount of information generated by the simulations of the CG model into a manageable form. Initially, 25 structural clusters were determined from the trajectories from a structural clustering based on residue-residue distances similarity criterion (16, 17). More detail on the clustering is given in SI Appendix: Fig. S1. Then, average scattering patterns were calculated for members of each cluster using the Fast-SAXS method (18), and the clusters were further regrouped into nine assembly states on the basis of a scattering pattern similarity criterion. The relationship between the nine states according to this metric is shown diagrammatically in SI Appendix: Fig. S1, and a representative conformation from each state is shown in Fig. 1. It should be noted that the configurations within those states are regrouped according to the similarity of their scattering patterns, and that they may display conformational differences. This set of nine assembly states, extracted from the simplified CG simulations, is then used as a basis-set for extracting the multidomain organization of Hck in solution from the experimentally observed SAXS patterns.

Bayesian-Based Monte Carlo Analysis of SAXS Data.

The calculated profile I_calc is an average over the N_s = 9 states, weighted by the fractional population {P_i} ≡ P₁,…,P_Ns of state i,

[1]

where I_i(q) is the theoretical scattering profile of state i (see examples in SI Appendix: Fig. S2). The goal of the analysis is to determine the set of N_s fractional population {P_i} from the SAXS data. To this end, we define a dimensionless scoring function, χ²[{P_i}], measuring the agreement between the calculated and experimental SAXS profiles,

[2]

where q_min and q_max are the lower and upper limit of q-range of the experimental profile of I_exp(q), respectively, and Δ is the offset between log I_calc and log I_exp(q) at q = q_min. The estimated experimental SAXS uncertainty, σ(q), is assumed to be [δ log I_exp(q)]^3/2. Assuming independent experimental errors and a uniform prior, the probability distribution exp(-χ²) is sampled in the space of the model parameters, {P_i}, according to a Bayesian-Based Monte Carlo (BMC) algorithm. The procedure yields the fractional population P_i for the N_s states, along with their uncertainties extracted the fluctuations.

Tests with the BSS-SAXS method indicate that the scattering pattern in the low-q regions, up to about 0.15 Å^-1 contains most of the information that controls the structural differences in Hck (see SI Appendix: Fig. S3A,C). Including higher-q SAXS data has little effect on the average values of P_i, although it reduces the uncertainty of the P_i (see SI Appendix: Fig. S3B, D). Additional tests with the BSS-SAXS/BMC simulations show that the procedure yields the correct population P_i (see SI Appendix: Fig. S3E, F), although large increases in the estimated SAXS uncertainty can lead to inaccuracies (see SI Appendix: Fig. S3G, H). The present approach has some similarities with other methods, such as EOM (19) and BILBOMD (20), which were also designed to go beyond standard strategies based on a single conformation to fit SAXS data. For example, all three methods primarily rely on a large set of random conformers. Both EOM and BILBOMD use a Monte Carlo genetic algorithm to select a small number of specific conformers out of the large set to best-fit the SAXS data, which are attributed equal statistical weight, whereas BSS-SAXS uses a BMC algorithm to determine the fractional population P_i for the states in a scattering basis-set. Overall, the tests show that the BSS-SAXS analysis can provide quantitative estimates of the fractional population P_i of Hck in solution.

BSS-SAXS Characterization of Hck Assembly.

SAXS data were collected for the wild-type Hck, and for the high-affinity Ctail mutant (Hck-YEEI) in the presence of two types of external peptide ligands: (p2) a high-affinity phosphorylated tyrosine-containing SH2-binding peptide (6); (p3) a high-affinity SH3-binding PPII peptide (21 and 22). All the experimental scattering profiles are shown in Figs. 2, 3. For the sake of completeness, a simple Guinier analysis was performed for all the SAXS data shown in Figs. 2 and 3, and the radii of gyration R_g were extracted by fitting the scattering patterns in the range up to 0.05 Å^-1. The results are reported in the Figure captions.

The experimental SAXS data from wild-type Hck in the absence of external peptides is shown in Fig. 2A. It is observed that the SAXS profile calculated from the population derived from the BMC analysis (red) accurately reproduces the experimental data (black). The fit is achieved with the population of assembly states determined via the BMC analysis. The latter indicates that the enzyme is predominantly in the assembled state 1, with the population of 82%. The more weakly populated states include the fully disassembled state 6 and the partially disassembled state 8 with SH2-dissociated but SH3-bound similar to the partially disassembled crystal structure of c-Src (10). As shown in Fig. 2A, this combination of states best fits the SAXS data. The experimental SAXS data from the high-affinity Ctail mutant YEEI is shown in Fig. 2B. In this case too, the assembled state 1 dominates the three-dimensional organization, with the population of 83%.

To investigate how ligand binding affects the equilibrium assembly of the multidomain complex, SAXS data were also collected from Hck with two types of peptide ligands. As shown in Fig. 3A, there is a substantial shift of wild-type Hck from a compact state to disassembled states in the presence of 250 μM of the high-affinity phosphorylated SH2-binding peptide p2. Comparison of Figs. 2A and 3A shows that the population of the compact state 1 decreases from 82% to 22%, while that of the disassembled state 6 increases from 8% to 29%. Repeating this experiment for the Hck-YEEI Ctail mutant yields a very different result. The BMC analysis given in Fig. 3B indicates that, for the Hck-YEEI mutant, the population of the assembled state is reduced only to 62% upon addition of the peptide p2. Here, the disassembled and compact state 6, where SH2 and SH3 are disassociated but the overall Hck architecture remains compact, is also populated.

The effect of the SH3-binding peptide p3 is shown in Fig. 3C. BMC analysis of the SAXS data indicates that p3 destabilizes the assembled down-regulated state 1 and promotes the disassembled compact state 6. Comparison of Figs. 2A and 3C shows that the population of state 1 decreases from 82% to 7% and that of state 6 increases from 8% to 50% in the presence of peptide p3. Also, the population of the SH3-displaced states 2 and 3 increases, reaching a total combined population of about 39%. Cross-correlation analysis of the population of states 2 and 3, however, suggests that they are highly correlated and that determining their relative weight is at the limit of resolution of the present BMC analysis.

Lastly, the spatial organization of Hck was examined in the presence of both the p2 and p3 peptides simultaneously. The results of the BMC analysis of SAXS data, shown in Fig. 3D, demonstrate that any remnant of the assembled down-regulated state is wiped out upon the binding of p2 and p3. Here, the fully disassembled states 5 and 6 are representative of the spatial organization. Both states are disassembled when Ctail is released from SH2 and SH3 is disassociated, but they differ in overall size: state 5 is extended (SI Appendix: Fig. S8) and state 6 exhibits a compact form (SI Appendix: Fig. S9). Consistent with the BSS-SAXS analysis, R_g reaches its largest value (31.7 Å) under these conditions.

Experiments.

Wild-type HCK3D or the HCK3D.YEEI mutant were coexpressed in bacteria with YopH phosphatase and purified as described by Seeliger et al. (30). The SH2 binding peptide (referred to as p2; sequence EPQ[pY]EEIPIKQ, where pY is phosphotyrosine) and the SH3 binding peptide (referred to as p3; sequence VSLARRPLPPLP) were synthesized in house. SAXS data were collected at the BioCAT-18ID from the Advanced Photon Source. Data were collected using a flow-cell kept at 10 °C with a series of 1 s exposures for the Hck protein sample, the buffer, and the empty capillary. SAXS profiles were determined by the subtraction of the buffer and the empty capillary scattering from the protein sample scattering. See more details in SI Appendix.

Multidomain Modeling of Hck.

Extensive configurational sampling was carried out using a multistate CG model of Hck following methods similar to previous studies (16, 31, 32). Models of inactive and active states of the catalytic domain were built from the crystal structures of Hck (PDB entry: 1QCF) and c-Src (PDB entry: 1Y57), respectively (4, 10), and models of the SH3 and SH2 were taken from the inactive state of the down-regulated Hck (4). The models are built at the residue level, where each residue is represented by its Cα atom. For postdata analysis, a clustering and mapping method as described previously was applied to simulation data (17). The resulting configurations were clustered into 25 structural clusters, where a list of 89 interdomain pair wise residue distances were used in a K-means clustering calculation. For each cluster, a number of 100 randomly selected configurations was used to compute the theoretical scattering profiles using the Fast-SAXS method (18), yielding 25 profiles (and their standard deviations) log I_i(q) ± δ log I_i(q) where i = 1,⋯,25. A second level K-means clustering procedure was performed using a metric based on the similarity between calculated scattering patterns. This second step assigns the 25 structural clusters to N_s = 9 scattering classes or states (see SI Appendix: Fig. S1). See the SI Appendix for more details.