Core Flipping in Lead Optimization: Rank Ordering Using λ-Dynamics

Abstract

In structure-based drug discovery, reliable structural models of ligands bound to their target receptors are critical for establishing the structure-activity relationship of the congeneric series. In such a series, substitutions on a common scaffold core might lead to different binding modes, ranging from slight changes of orientations to flipping or inversion of the core structure. Moreover, molecular docking might lead to alternative orientations within the top-ranked poses without being able to discriminate which is most likely. To determine the relative binding affinities between two alternative ligand poses, we propose a methodology based on relative binding free energy calculations using the λ-dynamics method. We used a dual-topology approach with distance-restraining schemes. We introduced a novel strategy using a one-step perturbation to calculate the contributions of the applied restraints. While using FEP/MBAR instead for that purpose led to smaller uncertainties, it suffered from convergence issues. We tested the validity and predictive power of our approach using two pharmaceutically relevant targets and eight small-molecule inhibitors from the experimentally characterized congeneric series. For each target, our approach correctly ranks the known X-ray poses as more favorable than alternative flipped poses. The proposed methodology can be easily extended to rank more than two poses and should also be applicable to the evaluation of alternative rotamers of target amino acids.

1

Thermodynamic cycle for calculations of the relative binding free energy of the flipped ligand pose to the protein compared to the X-ray ligand pose.

2

Thermodynamic cycle for evaluating the cost of (A) adding multiple distances to the X-ray pose and (B) removing multiple distance restraints on the flipped pose using the one step perturbation method. The reference state contains both ligand poses at λ = 0.5. ^∗∗ represents the full NOE force constant.

3

Terms of eq for the calculation of the free energy difference between the flipped and X-ray poses from MSλD with multiple distance restraints ( $\mathrm{\Delta }\mathrm{\Delta }{G}_{\mathrm{x}\mathrm{r}\mathrm{a}\mathrm{y}\to \mathrm{f}\mathrm{l}\mathrm{i}\mathrm{p}}$ ).

4

Selected dihydropyrimidinone noncovalent HNE inhibitors and the PDB IDs , corresponding to their X-ray structures with HNE.

5

Compound 2 bound to HNE in the (A) X-ray pose and (B) flipped pose.

6

LmNMT thiazolidinone inhibitors and the PDB IDs corresponding to their X-ray structures with LmNMT.

7

Compounds 1–3 bound to the LmNMT protein from the X-ray structures (top row) and flipped (bottom row) poses: Compound 1 (A, D), compound 2 (B, E) and compound 3 (C, F).

8

Workflow combining relative binding free energy calculations using MSλD and endpoint corrections for removing/adding restraints using the one-step perturbation (OSP) approach for accurately ranking the binding of flipped poses. Equilibrium MD consists of three steps; gradual heating, equilibration run, and production run.

9

Snapshots from λ-dynamics simulations for compound 1. (A) X-ray pose and (C) flipped pose using CHARMM-CGenFF force field combination. (B) X-ray pose and (D) flipped pose using the OPLS-AA force field.

10

Snapshots from λ-dynamics simulations for compound 2. (A) X-ray pose and (C) flipped pose using CHARMM-CGenFF force field combination. (B) X-ray pose and (D) flipped pose using the OPLS-AA force field.

11

Snapshots from λ-dynamics simulations for compound 3. (A) X-ray pose and (C) flipped pose using CHARMM-CGenFF force field combination. (B) X-ray pose and (D) flipped pose using the OPLS-AA force field.