Multi‐state modeling of G‐protein coupled receptors at experimental accuracy

Abstract

The family of G-protein coupled receptors (GPCRs) is one of the largest protein families in the human genome. GPCRs transduct chemical signals from extracellular to intracellular regions via a conformational switch between active and inactive states upon ligand binding. While experimental structures of GPCRs remain limited, highaccuracy computational predictions are now possible with AlphaFold2. However, AlphaFold2 only predicts one state and is biased toward either the active or inactive conformation depending on the GPCR class. Here, a multi-state prediction protocol is introduced that extends AlphaFold2 to predict either active or inactive states at very high accuracy using state-annotated templated GPCR databases. The predicted models accurately capture the main structural changes upon activation of the GPCR at the atomic level. For most of the benchmarked GPCRs (10 out of 15), models in the active and inactive states were closer to their corresponding activation state structures. Median RMSDs of the transmembrane regions were 1.12 Å and 1.41 Å for the active and inactive state models, respectively. The models were more suitable for protein-ligand docking than the original AlphaFold2 models and template-based models. Finally, our prediction protocol predicted accurate GPCR structures and GPCR-peptide complex structures in GPCR Dock 2021, a blind GPCR-ligand complex modeling competition. We expect that high accuracy GPCR models in both activation states will promote understanding in GPCR activation mechanisms and drug discovery for GPCRs. At the time, the new protocol paves the way towards capturing the dynamics of proteins at high-accuracy via machine-learning methods.

Structure Prediction, Evaluation, and Validation of GPR18 Lipid Receptor Using Free Programs

“…However, this simple analogy might not be straightforward in protein folding methods. As it has been shown by recent studies of Heo [ 27 ], folding methods, with a strong indication to AlphaFold protocols, tend to show bias in the models they generate towards inactive and intermediate states vs. active states. Therefore, we incorporated the average distance measurement between Cα atoms of the last five cytoplasmic residues in TM3 (from Y120 3.51 to V124 3.55 ) and TM6 (from K227 6.30 to I231 6.34 ) as an additional activity state measure [ 29 , 30 ].…”

“…Additionally, an AlphaFold2 model (DeepMind, denoted as AF-DM) GPR18 [ 26 ] was used as a reference model. However, when performing the planned calculations, a recent study by Heo and Feig was delivered [ 27 ], which suggested that AlphaFold2 might only predict one state biased towards the inactive conformation. Therefore, we decided to incorporate in this study a multistate prediction protocol that extends AlphaFold2 to predict active or inactive states with very high confidence, thus allowing the usage of two additional inactive state models: a Collab-script-generated model (denoted as CF-DM) and a model downloaded from a preprepared repository (denoted as PF-DM).…”

“…Therefore, we incorporated the average distance measurement between Cα atoms of the last five cytoplasmic residues in TM3 (from Y120 3.51 to V124 3.55 ) and TM6 (from K227 6.30 to I231 6.34 ) as an additional activity state measure [ 29 , 30 ]. The values obtained from the generated models were then compared with those of active-state models either generated by CollabFold script or obtained from the repository described in [ 27 ], with average values of 15.28 Å and 15.97 Å, respectively. Figure 5 shows the changes of the average distances during the simulations for unbound and CB148-complexed protein models.…”

“…Although several homology models [ 21 , 22 , 23 , 24 , 25 ] have been described in recent years (including one from our group), since then, many more accurate methods have been developed to predict and validate protein structures. This includes the thriving of methods based on artificial intelligence/machine learning (AI/ML), which were extensively used in the recent edition of the CASP and CAMEO experiments [ 26 , 27 , 28 , 29 , 30 ].…”

Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery

“…Sometimes this is inherently related to the function of the protein, such as the activation of GPCRs and the gating mechanism of ion channels, or may serve as a regulatory mechanism such as DFG-in and out conformations of kinases. In our data set the MET target was in the DFG-in conformation in the AF2 to predict active and inactive state GPCR structures using annotated structure templates 43 . This, however, relies on ample characterization of conformational states of the protein family, which is still lacking for many drug target classes.…”