Fig 1.
Structural overview of DnaK in substrate-bound and -unbound states.
(A) Structure of DnaK in an unassociated (ADP-bound, substrate bound) conformation (PDB ID: 2KHO [9]). The nucleotide-binding domain (NBD) is shown in blue. The substrate-binding domain (SBD) has two subdomains: the α-helical lid is shown in red, and the βSBD is shown in teal. The substrate from the canonical NR peptide (PDB ID: 1DKZ [10]) is overlaid in the binding cleft of the βSBD in magenta. (B) Structure of DnaK in an associated (ATP-bound, no substrate) conformation (PDB ID: 4B9Q [11]). The ATP molecule is shown based on PDB ID: 3AY9 [12].
Fig 2.
Conserved βSBD-substrate binding conformation.
(A) Overlay of 18 bound substrate structures (S1 Table), aligned using the backbone atoms of the βSBD. The SBDs are show in gray cartoons, and the substrates in magenta backbone traces. Only the forward orientation substrates are included. (B) Overlay of two forward and reverse orientation peptide substrates (forward: NRLLLTG, PDB: 1DKZ [10]; reverse: NRLILTG, PDB: 4EZY [19]; underlines indicate residues at Site 0). The forward orientation is shown in thick bonds and the reverse in thin bonds. The backbone hydrogen bonds between βSBD and the substrate are shown in magenta dashed lines. Note that backbone hydrogen bonds involving M404, S427 and A429 completely overlap in two orientations. Side chains are shown only for the central L at Site 0. For more viewing angles of panel B, see S1 Movie.
Fig 3.
Substrate-interacting sites of DnaK.
The center image shows a NRLLLTG substrate (magenta sticks) bound to βSBD (gray cartoon) (PDB: 1DKX). Each zoom-in shows the surface of one of the five binding sites on DnaK, colored by atom type: O (red), N (blue), C/H (white), S (yellow). The side-chain conformations shown were selected to clearly illustrate the extent of each site. These additional side-chain and substrate backbone conformations are taken from: (Site -2) 4JWD, (Site -1) 1DKX, (Site 0) 1DKX (for more viewing angles, see S2 Movie), (Site +1) 1DKZ, 4F00 and (Site +2) 4JWI, 4EZN, 4EZO. The binding site surface was generated using DnaK residues within 5 Å of the substrate side chain(s) at each site, with some atoms excluded for clarity.
Fig 4.
Individual, unscaled physical interaction terms for all possible substrate amino acid side chains occupying the site 0 in the SBD.
Error bars are calculated as standard error of the mean. The residues are ordered by Wimley-White interfacial hydrophobicity scale to facilitate easy reading [37]. Error bars report standard error of the mean from the MD simulations.
Fig 5.
Receiver operating characteristic curves of various predictors.
The raw score was calculated using the energy terms with uniform weights. (A) Discrimination of both strong binders and binders from nonbinders, and (B) Discrimination of strong binders from nonbinders. All solid lines were derived from the training set and dotted lines from the validation set. The red dashed diagonal line represents a random prediction.
Table 1.
Term and site weights and reverse orientation penalty of the Paladin model.
Table 2.
Area under the ROC curves (Fig 5).
The curves were based on the discrimination of all binders from non-binders, and of only strong binders from non-binders.
Fig 6.
Registry predictions for forward-binding substrates.
For each substrate, the Paladin scores are shown for all possible binding registries that allow all five sites on DnaK to be occupied. Both backbone orientations are considered (forward: blue; reverse: orange). The registries as observed in the PDB structures are shown with blue stars. Note that the NRLLLTG peptide has been crystallized in two registries (PDB: 1DKZ, 4EZW).
Fig 7.
Prediction of substrate binding orientation by Paladin.
The difference between the lowest forward and reverse scores will be negative when the for5tward orientation is favored, and vice versa. Substrates that actually bind in the forward orientation are labeled on the left of the red dividing line, and those that actually bind reverse are on the right.