Fig 1.
(A) The protocol is divided into docking and filter sections. During the docking (or prediction phase) we use centroid (TransformMover) and full-atom (HighResMover) representations of the protein structure and lipid conformers. Protein and lipid conformer moves are repeated based on the accept/reject Metropolis Monte Carlo ratio. In the filtering section, the evolutionary lipid accessibility and lipid specificity matches are used to determine the specificity score. The specificity score determines the probability of a protein-cholesterol binding site being specific. The output is an individual protein-lipid complex and a csv file. Highlighted in gray are the additional steps added to the current RosettaLigand protocol that compose the RC protocol. (B) Cartoon representation of each benchmark setup. Protein A and CLR A (self-dock); Protein A and Flip-CLR A (flip-dock); Protein A and CLR B (cross-dock); Protein A and CLR (global-dock). A dashed line depicts the center of the membrane. Protein A is shown in cartoon representation, while CLR is shown in stick representation. An example of the protein area cholesterol is docked into for each benchmark is colored black.
Fig 2.
Comparing protein-cholesterol binding site `nativeness`in the self-dock, flip-dock, and cross-dock benchmarks.
(A) Histogram of PNear values for each benchmark. PNear values range from 0 (non-native) to 1 (native). The docking protocol is labeled in the upper-left corner of each plot. (B-C) Scatter plot of the PNear values of the three different methods: RC (lipid) versus AutoDock and RC (lipid) versus RosettaLigand. Improved RC PNear values are highlighted in gray and quantified in the upper-left corner of each plot. (D) Distribution of RMSDs. RMSD values range from 0 to 18 Å. AutoDock models are gray, RosettaLigand models are blue, and RC (lipid) models are cyan. The red dotted lines mark 2 Å.
Fig 3.
Cholesterol tilt angle in the self-dock, flip-dock, and cross-dock benchmarks.
The tilt angle θ is defined as the angle between the sterol plane (C3 and C17 atoms) and the membrane normal (z axis). A probability density quantifies the tilt angle distribution. AutoDock is shown in gray, RosettaLigand in blue, and RC (lipid) in cyan. The tilt angle distribution found for cholesterol in experimental X-ray and cryo-EM structures is shown as a black dotted line. The red dotted line marks the average tilt angle of the native structure (25 degrees).
Fig 4.
Illustrative examples of the best-scoring model in the self-dock, flip-dock, and cross-dock benchmarks.
(A,B) RC (lipid) succeeds. (C,D) RC (lipid) fails. The co-crystal structure is shown (gold) aligned with the results from AutoDock (gray), RosettaLigand (blue), and RC (lipid, cyan). A black dotted line denotes the RMSD of 2 Å, and a red dotted line with a red star denotes the best-scoring model.
Fig 5.
Cholesterol interaction site specificity score calculation.
(A) Global docking energy landscape. (B) Cartoon representations of top query sites in each cluster. (C) Rate of evolution calculation of query binding sites. Surface residues are colored gray, and the query pockets are colored based on the stars in (A). (D) Buried Area (A2) of query pockets. (E) Cartoon illustration of query site pockets. (F) Residue-cholesterol 1-D fingerprints of the query pockets and percent similarity quantification. (G) Hydrophobicity of query pockets with percent similarity quantification. (H) Bulkiness of query pockets with percent similarity quantification.
Table 1.
Summary of specificity score predictions.