SSEThread: Integrative threading of the DNA-PKcs sequence based on data from chemical cross-linking and hydrogen deuterium exchange

Abstract

X-ray crystallography and electron microscopy maps resolved to 3-8 Å are generally sufficient for tracing the path of the polypeptide chain in space, while often insufficient for unambiguously registering the sequence on the path (i.e., threading). Frequently, however, additional information is available from other biophysical experiments, physical principles, statistical analyses, and other prior models. Here, we formulate an integrative approach for sequence assignment to a partial backbone model as an optimization problem, which requires three main components: the representation of the system, the scoring function, and the optimization method. The method is implemented in the open source Integrative Modeling Platform (IMP) (https://integrativemodeling.org), allowing a number of different terms in the scoring function. We apply this method to localizing the sequence assignment within a 199-residue disordered region of three structured and sequence unassigned helices in the DNA-PKcs crystallographic structure, using chemical crosslinks, hydrogen deuterium exchange, and sequence connectivity. The resulting ensemble of threading models provides two major solutions, one of which suggests that the crucial ABCDE cluster of phosphorylation sites cannot undergo intra-molecular autophosphorylation without a conformational rearrangement. The ensemble of solutions embodies the most accurate and precise sequence threading given the available information.

Keywords: DNA-PKcs; Electron microscopy; Integrative modeling; Threading; X-ray crystallography.

Figure 1:. Structure of DNAPKcs and identification of unassigned helices.

A) The crystal structure of human DNA-PKcs (5LUQ) is shown in blue cartoon. Red cartoon identified the residues with only C_α coordinates assigned. The structure has been fitted to the 4.4 Å resolution EM map, (grey volume), (Sharif et al., 2017) which shows no density in the unassigned region. The yellow star indicates the position of the kinase active site. B) Zoom-in of unassigned region highlighting the three helices identified by DSSP as red, green and orange. The pink and yellow cartoon identifies the structured and assigned residues immediately N- and C-terminal of the disordered domain. C) Blue bars identify residues with both structure and sequence identification in the crystal structure. The inset highlights the 199-residue disordered region of interest in this study. The sector corresponding to the ABCDE cluster of phosphorylation sites (residues 2609–2638) is identified in pink. D) Cartoon representation of DNA-PKcs with the kinase domain in light blue and kinase active site as a yellow star. Potential threading arrangements that include a long linker between the ABCDE cluster (pink) and unassigned helices (colored bars) could allow an intra-molecular autophosphorylation event (left), while localization on or near the ordered helices would require a conformational change or inter-molecular autophosphorylation mechanism (right). Panels A and B were generated in part using VMD. (Humphrey et al., 1996)

Figure 2:. Four stages of integrative threading of DNA-PKcs using IMP:SSEThread.

The full integrative threading protocol proceeds through four stages. In stage 1, we gather all information about the system that we wish to use and decide at which stage of modeling we will apply it. In stage 2, we define a representation that includes the degrees of freedom we wish to assess and translate the information from stage 1 into spatial restraints [Section 2.2.2]. In stage 3, we sample alternative threading models using a Monte Carlo approach, using the scoring function from stage 2 as a guide [Section 2.2.3]. Finally, in stage 4 we assess the set of models generated in stage 3 by filtering those models that satisfy the input information, estimating the sampling and modeling precision as well as validating the models by both data used for modeling and data not used for modeling (orange boxes in stage 1).

Figure 3:. Relationship between structure element keys and threading model.

A SE defines a secondary structure designation, a set of C_α coordinates and four keys that map these coordinates to residues in the primary sequence. SE1 defines the C_α coordinates of ten residues of a helix and the set of four keys map the six blue coordinates onto sequence. The start residue, 4, denotes that the threaded sequence begins at residue four, the length, 6, means six total coordinates from the SE are assigned, a polarity of 1 assigns the coordinates in advancing order and an offset of 2 begins from coordinate 3 in the structure element. SE2 shows a similar assignment, beginning at residue 21; however, the polarity of −1 flips the assignment, such that the last assigned coordinate in the SE is threaded to the sequence at residue 21 and the remainder of the SE is assigned backwards.

Figure 4:. Formulation of crosslinking restraint for unstructured residues.

Schematic of the evaluation of the model crosslinking distance, XL_M, for a single unstructured residue between residues with coordinates at X₀ and X₁ and a structured residue at coordinate X_A. See Section 2.2.2.3 for a complete description of the evaluation of XL_M.

Figure 5:. Simulated benchmark results.

A) C_α trace of the human MLL5 PHD domain (PDB 2LV9) showing the three identified SEs (red, green, orange), distance restraints used (grey dashed lines connecting black residues), and SeMet restraints (yellow atoms are anomalous peaks and blue atoms are the corresponding Methionine C_αs). B) Table of restraints for SE localization using IMP:SSEThread. C) Residue occupancy of the top 5000 threading models following enumeration of all possible states. Each box represents the mapping of a residue in sequence (X-axis) to a coordinate in a structure element (Y-axis). A black box indicates that 100% of the top models map the corresponding residue to the structure element coordinate. SE3 shows multiple threading possibilities that are equally likely. The correct threading solution is indicated by the red outline of the boxes.

Figure 6:. Results of DNA-PKcs threading.

A) Cartoon models of the three unassigned helices assigned as SEs in the DNA-PKcs crystal structure in space relative to each other (Figure 1A). B) Table of restraints utilized for SE localization using IMP:SSEThread. C) Residue occupancy of the two clusters of models identified from the top 500 threading models following enumeration of all possible start residues values. Each box represents the mapping of a residue in sequence (X-axis) to a coordinate in a structure element (Y-axis). A black box indicates that near 100% of the top models map that residue in sequence to that structure element coordinate. The green shadow highlights residues identified by hydrogen exchange as being partially protected. Cluster 1 shows a high specificity for SE3 near residue 2740 with SE1 and SE2 highly variable. Cluster 2 localizes SE1 and SE2 with high precision and has SE3 disordered. The threading solutions for SE1 and SE2 also match the HDX data, which suggest some local order in these regions. The ABCDE cluster (pink) is unlocalized in Cluster 1, while it occurs at the N-terminal end of SE2 in Cluster 2, highly constraining its position in space. D) Cartoon models of potential autophosphorylation mechanisms for the ABCDE cluster of DNA-PKcs based on Cluster 2 models. The intra-molecular mechanism (left) is not supported by this model, as the ABCDE cluster (pink) in SE2 (green) cannot interact with the kinase site (yellow star). The helices could potentially unravel (center), allowing the cluster to interact with the kinase site on the same chain. Alternatively, the other DNA-PKcs molecule in the synaptic complex could position itself to perform an inter-molecular autophosphorylation (right).