Fragment-based design of selective GPCR ligands guided by free energy simulations

Fragment-based drug discovery relies on successful optimization of weakly binding ligands for affinity and selectivity. Herein, we explored strategies for structure-based evolution of fragments binding to a G protein-coupled receptor. Molecular dynamics simulations combined with rigorous free energy calculations guided synthesis of nanomolar ligands with up to >1000-fold improvements of binding affinity and close to 40-fold subtype selectivity.


Supplementary Figures
Functional assay for compound 22 S14 Functional assay compound 26 S15

Supplementary Tables
Experimental binding affinities of compounds 1-10 S16 Relative binding free energies for compounds 2-10 S17 Experimental binding affinities of compounds 11-26 S18 Relative binding free energies for compounds 11-26 S19 High affinity A 1 AR ligands with < 20 heavy atoms S20

NMR Spectra
Compounds synthesized at Enamine (2-6 and 9-10) S21 Compounds synthesized in-house ( respectively. Engineered mutations were reverted and non-protein atoms were removed prior to initiating the simulations. Each receptor was then placed in a pre-equilibrated 1palmitoyl-2-oleoyl-phosphatidylcholine (POPC) lipid bilayer using the GPCR-ModSim protocol 3 . The resulting systems were equilibrated at 310 K and atmospheric pressure for 40 ns using GROMACS 4 . During this step, protein heavy atoms were held rigid with tight positional restraints whereas membrane and water molecules were equilibrated. The OPLS 2005 all atom force field 5 , Berger lipids 6 and the TIP3P water model 7 were used to carry out these simulations. For each receptor-ligand complex, the studied compound was added and clashing waters were removed. The ligand binding modes in the A 1 AR binding site were based on a molecular docking pose of compound 1 8 and the core scaffold of the compound was superimposed with the coordinates of tozadenant for the A 2A AR. MD simulations were performed with the software Q 9 using its implementation of a newer version of the same protein force field 10 and the same lipid parameters and water model as in the equilibration of the system. For each compound, OPLSAA_2005 force field parameters were obtained from the hetgrp_ffgen program (Schrödinger, LLC, New York, NY, 2017). In these simulations, spherical boundary conditions were used with a 21 Å radius sphere centered on the ligand. All atoms outside the sphere were excluded from non-bonded interactions and ionizable residues at the sphere edge were set to their neutral form. In addition, atoms within 3 Å of the edge of the spherical system were restrained to their initial coordinates. The surface-constrained all atom solvent (SCAAS) model 11 was applied to water molecules at the sphere surface using radial and polarization restraints. Solvent bonds and angles were constrained with the SHAKE algorithm 12  Transformation of the partial charges, (ii,iii) annihilation of atoms by first introducing a soft-core potential and then removing the resulting Lennard-Jones term for these atoms, 14 and (iv) remaining Lennard-Jones parameters and bonded terms were changed.
These calculations were divided into 11, 11, 21 and 41 steps by using a mapping potential (U) that describes each transformation as a linear combination of the potential energy functions of the start (A) and end (B) states: where is varied from zero to one. The free energy difference between states was then obtained using the Zwanzig equation. 15 At each window, the receptor-ligand complex was equilibrated for 750 ps with harmonic positional restraints on solute heavy atoms, which were gradually released while the system was heated. Equilibration was followed by 500 ps production runs and energies were collected every 50 fs. The ligands were also prepared for simulations in aqueous solution using a water droplet of the same size. In these simulations, a weak harmonic restraint was applied to a central ligand atom to prevent it from approaching the sphere edge and the systems were equilibrated for 350 ps, which was followed by 100 ps productions. Each calculation was performed using three independent replicates and relative binding free energies were obtained based on a thermodynamic cycle, as described in a previous work. 8

Biological assays
Radioligand binding assays. HEK293 cells expressing the human A 1 -or A 2A ARs were cultured in DMEM supplemented with 10% fetal bovine serum, 100 Units/ml penicillin, 100 µg/ml streptomycin, and 2 µmol/ml glutamine. Cells were detached from plates by scraping into cold PBS and centrifuged at 250 g for 5 min. The pellets were resuspended in ice-cold Tris HCl buffer (50 mM, pH 7.4) and then homogenized. After homogenization and suspension, cells were centrifuged at 1000 g for 10 min, and the pellet was discarded.
The suspension was then re-centrifuged at 20,000 g for 60 min at 4 °C. The pellets were resuspended in buffer containing 3 Units/ml adenosine deaminase and incubated at 37°C