Probing Protein–Ligand Methyl−π Interaction Geometries through Chemical Shift Measurements of Selectively Labeled Methyl Groups

Fragment-based drug design is heavily dependent on the optimization of initial low-affinity binders. Herein we introduce an approach that uses selective labeling of methyl groups in leucine and isoleucine side chains to directly probe methyl−π contacts, one of the most prominent forms of interaction between proteins and small molecules. Using simple NMR chemical shift perturbation experiments with selected BRD4-BD1 binders, we find good agreement with a commonly used model of the ring-current effect as well as the overall interaction geometries extracted from the Protein Data Bank. By combining both interaction geometries and chemical shift calculations as fit quality criteria, we can position dummy aromatic rings into an AlphaFold model of the protein of interest. The proposed method can therefore provide medicinal chemists with important information about binding geometries of small molecules in fast and iterative matter, even in the absence of high-resolution experimental structures.

Supporting Figures and Tables Table S1 .

Figure S2 :
Figure S2: Assignment of selected Leucine and Valine resonances: A: Overlay of the spectrum of wildtype Brd4-BD1 in blue with the spectra of the four Brd4-BD1 mutants (V87A, V90A, L92A, L94A).The structure in (A) shows the 4 residues that are closest to the ligand binding site with V87, L92, L94 facing the ligand.B: Constant time 1 H-13 C HSQC of fractionally labeled Brd4-BD1 for stereospecific assignment of Leucine C-resonances and Valine C-resonances.Positive resonances (blue) are pro-S methyl groups, negative resonances (red) are pro-R methyl groups of Valine and Leucine residues.

Figure S7 :
Figure S7: Ring fit statistics for Ligand 1 (PDB: 6XUZ) and the respective shift and PDB probability density estimation components.A: Histogram of the fit score over all grid positions.B: Average fit score binned by the summed distance to the closest ring in the respective crystal structure (x-axis) and the sum of the absolute dot products between the respective ring normals in the fit and the experimental structure (y-axis).C: Histogram of the region indicated by a green box in C. D: Best (magenta) and worst (red) ring position based on fit values in the region indicated by a green box in C. Fitting 2 rings simultaneously and therefore evaluating expanding the single-ring grid by itself leads to a large increase in histogram counts compared to ligands 8 (Figure S8) and 12 (Figure S9)

Figure S9 :
Figure S9: Ring fit statistics for Ligand 12 (PDB: 9FXP) and the respective shift and PDB probability density estimation components.A: Histogram of the fit score over all grid positions.B: Average fit score binned by the distance to the closest ring in the respective crystal structure (x-axis) and the absolute dot products between the respective ring normals in the fit and the experimental structure (y_axis).C: Histogram of the region indicated by a green box in C. D: Best (magenta) and worst (red) ring position based on fit values in the region indicated by a green box in C.

Figure S10 :
Figure S10: Top and side view of the top 20 result of the ring fit based on overall fit value and its respective shift and PDB probability density estimation components for ligand 1, 8 and 12. Surfaces indicate variability in the ring position while red lines represent the fitted ring normal and therefore ring orientations.For ligand 1, two rings are fit simultaneously, and the green lines indicate connecting vectors between ring pairs.

Figure
Figure S11: A: Comparison of best ring fit results for the set of ligands.A: Difference in orientation (y-axis) vs distance of ring centers(x-axis)  between the best fit and the closest ring in the respective crystal structure.B: For ligand 10 we are missing shift data for Ile-C which leads to a large offset for the best fit.C: For ligand 3, there is an additional ring that has to be accounted for and leads to an intermediate position when fitting only 2 rings (left).When fitting 3 rings (right), the position of the 2 main fit rings improves accordingly, while the 3 rd ring covers a potential ring position of the 3 rd ligand ring after change of one dihedral angle.D: For ligand 4, the presence of the non-planar 7-membered ring hinders good orientational agreement (left).This is improved by including a second ring in the fit (right).

Table S2 :
B-factors for selected atoms and structures.Ring average is the average over all ring atoms of the double ring directly beneath

Table S3 :
Pople model parameters used in cgs units.
Figure S1: Interaction geometry parameters visualized on Ligand 8.For the pople model only θ and r are used as well as the ring size to select the correct diameter.