Structure-based discovery of positive allosteric modulators of the A1 adenosine receptor

Abstract

Allosteric modulation of G protein-coupled receptors (GPCRs) is an exciting strategy for developing new therapeutic agents, and it has several advantages over more commonly used orthosteric drugs. Recently determined GPCR structures have revealed allosteric pockets facing the lipid bilayer, enabling rational drug design. Here, we develop a virtual screening strategy to discover ligands of extrahelical binding pockets and apply this approach to the adenosine A₁ receptor (A₁R). The A₁R is a high-value therapeutic target for ischemia-reperfusion injury and chronic neuropathic pain. Developing effective A₁R therapeutics remains challenging due to high structural conservation across orthosteric binding sites and on-target unwanted effects stimulated by prototypical A₁R agonists, such as bradycardia and atrioventricular block. However, A₁R positive allosteric modulators (PAMs) acting through spatially distinct allosteric sites can fine-tune A₁R activity with high subtype selectivity and spatiotemporal specificity, thereby overcoming current limitations. A chemical library of 160 million compounds was computationally docked to the allosteric pocket identified in a cryo-EM structure of the A₁R, and a set of 26 top-ranked compounds were selected for experimental evaluation. Pharmacological evaluation of these, and structure-guided hit optimization, led to the discovery of subtype-selective A₁R PAMs. These compounds demonstrated minimal allosteric agonism and negligible impact on A₁R-mediated beat rate of an orthosteric agonist. The discovered PAMs pave the way for potential treatments for neuropathic pain and ischemia-reperfusion injury without accompanying side effects. Our results demonstrate the utility of a synergistic computational and experimental approach in GPCR drug discovery.

Keywords: A1 adenosine receptor; G protein–coupled receptor; allosteric modulation; drug discovery; structure-based virtual screening.

Fig. 1.

Molecular docking at GPCR extrahelical binding sites. (A) Allosteric binding site location of A₁R (PDB accession code: 7LD3), P2Y₁ (PDB accession code: 4XNV), and FFA₁ (PDB accession code: 5TZY). The receptors are represented as gray cartoons and the ligands as sticks. The phospholipid bilayer location is shown by the gray square. (B) Virtual screening performance of ensemble docking using implicit and explicit membrane modeling, shown in blue and orange, respectively. In the implicit membrane model, the dielectric constant for the region surrounding the binding site was reduced from ε = 80 to ε = 2 in order to represent the hydrophobic core of the membrane. The explicit membrane model used lipid bilayer conformations (1 to 200 snapshots) obtained from MD simulations. Ligand enrichment over property-matched decoys was assessed using logAUC values. The logAUC values obtained with the top-enriching structures are shown with triangles. Error bars correspond to the SD. For FFA1, conformations obtained from two structures (PDB accession codes 5TZY and 5KW2) were combined. (C) Structure of A₁R in complex with the PAM 1 (MIPS521). 1 is shown as sticks with orange carbon atoms. Side chains interacting with the PAM are shown as gray sticks (PDB accession code: 7LD3). ROC curve for the enrichment of A₁R PAMs over property-matched decoys (D) and experimentally confirmed inactive compounds (E). Enrichment curves obtained from docking to the experimental structure or an ensemble of five MD snapshots that included the lipid bilayer are shown in blue and orange, respectively. The dashed black line represents the enrichment expected from random selection.

Fig. 2.

Influence of virtual screening hits on NECA A₁R pharmacology. (A) The change in NECA affinity (ΔpK_I; black) and potency (ΔpEC₅₀; red) in the presence of 30 μM virtual screening hit compounds in A₁R-FlpInCHO cells was assessed using [³H]DPCPX competition binding or cAMP inhibition, respectively. (B) Predicted binding mode for virtual screening compound 12 (orange carbon sticks). A₁R is shown as a gray cartoon with key residues in sticks. Hydrogen bonds are indicated with yellow dashed lines. Interaction profile of NECA and 12 in [³H]DPCPX competition binding (C) and inhibition of forskolin-stimulated cAMP accumulation (D) in the absence and presence of 12 (3, 10, and 30 μM) in A₁R-FlpInCHO cells. Data represent the mean ± SEM from n = 3 to 4 individual replicates performed in duplicate. *P < 0.05, one-sample t test compared to a hypothetical value of 0. ND denotes not determined due to near complete inhibition of [³H]DPCPX binding.

Fig. 3.

Structure-guided optimization of A₁R PAMs. (A) The change in NECA affinity (ΔpK_I; black) and potency (pEC₅₀; red) in the presence of 30 μM 50–56 in A₁R-FlpInCHO cells, assessed in [³H]DPCPX competition binding or inhibition of forskolin-stimulated cAMP accumulation, respectively. Data represent the mean ± SEM from n = 3 to 4 individual replicates performed in duplicate. ^∗P < 0.05, one-sample t test compared to a hypothetical value of 0. The predicted binding mode of A₁R PAMs 54 (B) and 56 (C). Ligands are shown as sticks with carbon atoms in orange, A₁R as a gray cartoon with key residues as sticks, and hydrogen bonds in yellow dashed lines. [³H]DPCPX interaction binding with NECA and 54 (D) and 56 (E), and NECA inhibition of forskolin-stimulated cAMP accumulation with 54 (F) and 56 (G) in A₁R-FlpInCHO cells. Data represent the mean ± SEM from n = 3 to 4 individual replicates performed in duplicate. (H) The influence of 56 on the NECA-mediated decrease in the contractile rate of isolated rat atria. (I) The influence of 56 on NECA-mediated inhibition of forskolin-stimulated cAMP accumulation in primary mouse cortical neurons. Data represent the mean ± SEM from n = 3 to 4 individual replicates performed in duplicate.