GPCRdb in 2018: adding GPCR structure models and ligands

Abstract

G protein-coupled receptors are the most abundant mediators of both human signalling processes and therapeutic effects. Herein, we report GPCRome-wide homology models of unprecedented quality, and roughly 150 000 GPCR ligands with data on biological activities and commercial availability. Based on the strategy of 'Less model - more Xtal', each model exploits both a main template and alternative local templates. This achieved higher similarity to new structures than any of the existing resources, and refined crystal structures with missing or distorted regions. Models are provided for inactive, intermediate and active states-except for classes C and F that so far only have inactive templates. The ligand database has separate browsers for: (i) target selection by receptor, family or class, (ii) ligand filtering based on cross-experiment activities (min, max and mean) or chemical properties, (iii) ligand source data and (iv) commercial availability. SMILES structures and activity spreadsheets can be downloaded for further processing. Furthermore, three recent landmark publications on GPCR drugs, G protein selectivity and genetic variants have been accompanied with resources that now let readers view and analyse the findings themselves in GPCRdb. Altogether, this update will enable scientific investigation for the wider GPCR community. GPCRdb is available at http://www.gpcrdb.org.

Figure 1.

The GPCRdb homology model pipeline builds on the principle of ‘Less model – more Xtal’ meaning that more of the target is covered by structural templates, as an incomplete or partly distorted main template is complemented with alternative local templates. All templates and models are automatically updated upon each database update. Models are produced for all the human non-olfactory GPCRs in the inactive, and where templates exist (classes A and B1), also intermediate and active states.

Figure 2.

RMSD (Å) values of database/server homology models calculated directly upon release of the first structure for four GPCRs (AA1R: 5UEN, CCR2: 5T1A, PAR2: 5NDD and APJ: 5VBL). Notably, the GPCRdb homology models have the best average RMSD values in all four categories. Superimposition was performed on backbone heavy atoms. For comparability, RMSD calculations were restricted to residues present in all of the models. SWISS-MODEL settings: default, used main template listed on top. GOMoDo settings: Blast – 2 search rounds, MODELLER options—two models, global alignment (no realign templates), no loop refinement, chose model with best normalized DOPE score (PAR2 model was sixth best, ones before were incomparable due to a shift in sequence numbering). GPCRM settings (kindly provided by the developers): beta version (under development), ECL2 disulphide bridge specified, Task mode Auto, Set of templates inactive, Lysozyme do not add from template, Rosetta loop-modeling yes, fast, additional options left default. Selected r01 model. GPCR-ModSim settings (kindly provided by the developers): group inactive, alignment was edited, number of models 10, selected one with best DOPE score, no Lenard-Jones restraints, select top templates – default, ECL2 disulphide bridge specified, short loops ICL1, ECL1, ICL2 and ECL3 were modeled with loop modeling, number of models 5, selected one with best DOPE score, no MD. GPCR-i-TASSER model selection from repository: Model 1. GPCR-SSFE model selection from repository: EntireModel1, loop versions 0.

Figure 3.

Example model of the inactive state GPR75. The colors indicate the use of 10 different backbone template to fill in lacking coordinates, elongate helices, add missing secondary structure segments to loops and to remove a main template helix bulge not shared by GPR75. Furthermore, the use of 27 sidechain templates from the GPCR position-specific rotamer library increased the percentage of residues that could be based on an identical amino acid from 15.7% to 58.2%.

Figure 4.

Trees showing the human GPCR classes (GRAFS family): A (Rhodopsin), B1 (Secretin), B2 (Adhesion), C (Glutamate), F (Frizzled) and T: Taste 2. Each tree is sorted alphabetically by ligand type (only class A) and receptor families which share the same physiological ligand. The tree branches are color-coded by ligand types, and the gray-scale circles before receptor names indicate their number of ligands (white: 0, light gray: >100, gray: >500 and black: >1000).