In silico screening for agonists and blockers of the β(2) adrenergic receptor: implications of inactive and activated state structures

Abstract

Ten crystal structures of the β(2) adrenergic receptor have been published, reflecting different signaling states. Here, through controlled-docking experiments, we examined the implications of using inactive or activated structures on the in silico screening for agonists and blockers of the receptor. Specifically, we targeted the crystal structures solved in complex with carazolol (2RH1), the neutral antagonist alprenalol, the irreversible agonist FAUC50 (3PDS), and the full agonist BI-167017 (3P0G). Our results indicate that activated structures favor agonists over blockers, whereas inactive structures favor blockers over agonists. This tendency is more marked for activated than for inactive structures. Additionally, agonists tend to receive more favorable docking scores when docked at activated rather than inactive structures, while blockers do the opposite. Hence, the difference between the docking scores attained with an activated and an inactive structure is an excellent means for the classification of ligands into agonists and blockers as we determined through receiver operating characteristic curves and linear discriminant analysis. With respect to virtual screening, all structures prioritized well agonists and blockers over nonbinders. However, inactive structures worked better for blockers and activated structures worked better for agonists, respectively. Notably, the combination of individual docking experiments through receptor ensemble docking resulted in an excellent performance in the retrieval of both agonists and blockers. Finally, we demonstrated that the induced-fit docking of agonists is a viable way of modifying an inactive crystal structure and bias it toward the in silico recognition of agonists rather than blockers.

Figure 1

Superimposition of two crystal structures of the β₂AR, one reflecting an inactive state of the receptor (PDB ID: 2RH1), the other reflecting an activated state of the receptor (PDB ID: 3P0G). The protein backbone is schematically represented as a cartoon, colored in pink for 2RH1 and with a continuum spectrum of colors ranging from red at the N-terminus to blue at the C-terminus for 3P0G. The orthosteric binding cavity, calculated for 3P0G, is represented as a turquoise colored surface. In the region lining the binding cavity, the most pronounced difference between 2RH1 and 3P0G is a displacement of a portion of TM5, centered on Ser207^5.46, which in 3P0G bulges towards the center of the receptor (circled in yellow). The Cα and the side chain of the serine residue are represented as balls and sticks and it are colored in red for 2RH1 and in green for 3P0G.

Figure 2

The chemical scaffolds on which the 60 β₂AR ligands object of this study are based. All agonists and 17% of the blockers (namely AH-3474A, dichloroisoproterenol, labetalol, pronethanol and sotalol) are based on scaffold a; the remaining 83% of the blockers is based on scaffold b. Due to a change in the priorities of the substituents on the β-carbon (indicated with an asterisk), the most potent stereoisomers of the compounds based on scaffold a and scaffold b feature this chiral center in the R and the S configuration, respectively.

Figure 3

Analysis of the prioritization of agonists and blockers in docking experiments targeting different crystal structures of the β₂AR. The receiver operating characteristic (ROC) curves were obtained treating agonists as positives and blockers as negatives. The area under each ROC curve (AUC) is indicated next to each of the curve. For an ideal docking experiment that prioritized all agonists versus blockers, the ROC analysis would yield a rectangular curve passing through the upper left corner of the plot, with an AUC of 1. Conversely, for an ideal docking experiment that prioritized all blockers versus agonists, the ROC analysis would yield a rectangular curve passing through the lower right corner of the plot, with an AUC of 0. Finally, an experiment incapable of distinguishing agonists from blockers would yield a diagonal curve with an AUC of 0.5.

Figure 4

Analysis of the difference between the docking scores obtained when targeting activated 3P0G structure or the inactive 3RH1 structure (ΔScore). (a) Receiver operating characteristic (ROC) curve obtained treating agonists as positives and blockers as negatives, after sorting the compounds on the basis of their ΔScore. The ROC analysis showed an excellent discriminatory power for ΔScore, with an area under the curve (AUC) of 0.98. (b) Plot of the ΔScore measured for agonists (in green) and blockers (in red). 97% of the agonists (29 out of 30) scored better when docked at 3P0G, while 93% of the blockers (28 out of 30) scored better when docked at 3RH1. The only outlier among the agonists is sulfonterol (1). The two outliers among the blockers are the compounds AH-3474A (2) and ICI-89406 (3). A linear discriminant analysis (LDA) placed the ideal watershed for the separation of agonists and blockers at 0.03 kcal/mol (Wilks’ lambda: 0.33; F (1,58): 119.33; p: 1.06*10⁻¹⁵).

Figure 5

Molecular weight is not a good parameter to separate agonists (in green) and blockers (in red). A linear discriminant analysis (LDA) placed the ideal watershed for the separation of agonists and blockers on the basis of their molecular weight at 290.06 Dalton (Wilks’ lambda: 0.80; F (1,58): 14.67; p: 0.0003). This yielded the correct classification of only 70% of agonists (21 out of 30) and 57% of blockers (17 out of 30).

Figure 6

Analysis of the prioritization of ligands versus decoys in docking experiments targeting four crystal structures of the β₂AR. Data points are colored in green when relative to agonists and in red when relative to blockers. The receiver operating characteristic (ROC) curves represented in the four panels were obtained treating ligands as positives and decoys as negatives. The area under each ROC curve (AUC) is reported within each panel. Additionally, AUC values calculated for ROC analyses relative to agonists versus decoys and blockers versus decoys are also reported.

Figure 7

Analysis of the prioritization of ligands versus decoys in receptor ensemble docking (RED) resulting from the combination of docking experiments conducted at four crystal structures of the β₂AR. (a) Receiver operating characteristic ROC analysis relative to the receptor ensemble docking, obtained treating ligands as positives and decoys as negatives. Data points are colored in green when relative to agonists and in red when relative to blockers. The area under the ROC curve (AUC) is reported. (b–d) Synoptic view of the ROC analysis relative to the RED experiment together with those relative to the docking experiments based on the four individual structures. The entire curve is represented in panel b, while enlargements of the top scoring portion of the screened database are represented in panels c and d. The area where RED performed better than any of its individual components is highlighted in pale yellow.

Figure 8

Superimposition of an induced fit model of the isoproterenol-bound β₂AR, with: (a) the crystal structure an inactive state of the receptor (PDB ID: 2RH1); and (b) the crystal structure an activated state of the receptor (PDB ID: 3P0G). The induced fit model, which was constructed docking isoproterenol to the 2RH1 structure, approximates very well the local conformational change that the segment of TM5 centered on Ser207^5.46 undergoes when transitioning from the inactive to the activated state (circled in yellow). The Cα and the side chain of the serine residue is represented as balls and sticks and it is colored in red for 2RH1, in green for 3P0G and in gray for the induced fit model. The protein backbone is schematically represented as a cartoon. The 2RH1 structure in panel a and the 3P0G structure in panel b are colored in pink. In both panels, the induced fit model of the isoproterenol-bound β₂AR is colored with a continuum spectrum of colors ranging from red at the N-terminus to blue at the C-terminus.

Figure 9

Analysis of the prioritization of agonists and blockers in docking experiments targeting three induced fit models of the agonist-bound β₂AR based on the inactive structure of the receptor (PDB ID: 2RH1). The receiver operating characteristic (ROC) curves were obtained treating agonists as positives and blockers as negatives. The area under each ROC curve (AUC) is indicated next to each of the curves. (a) The three ROC curves obtained for the induced fit models are compared with those obtained for the inactive 2RH1 and the activated 3P0G structure of the receptor. (b) The ROC curve resulting from the combination of the docking experiments targeting the three induced fit models through receptor ensemble docking (RED) is compared with those resulting from the three individual docking experiments.