Limits...
Towards improved quality of GPCR models by usage of multiple templates and profile-profile comparison.

Latek D, Pasznik P, Carlomagno T, Filipek S - PLoS ONE (2013)

Bottom Line: Usage of multiple templates and generation of alignments based on sequence profiles may increase the rate of success in difficult cases of comparative modeling in which the sequence similarity between GPCRs is exceptionally low.In particular, GPCRM is the first GPCR structure predictor incorporating two distinct loop modeling techniques: Modeller and Rosetta together with the filtering of models based on the Z-coordinate.We also provide a database of precomputed GPCR models of the human receptors from that class.

View Article: PubMed Central - PubMed

Affiliation: International Institute of Molecular and Cell Biology, Warsaw, Poland. dlatek@iimcb.gov.pl

ABSTRACT

Unlabelled: G-protein coupled receptors (GPCRs) are targets of nearly one third of the drugs at the current pharmaceutical market. Despite their importance in many cellular processes the crystal structures are available for less than 20 unique GPCRs of the Rhodopsin-like class. Fortunately, even though involved in different signaling cascades, this large group of membrane proteins has preserved a uniform structure comprising seven transmembrane helices that allows quite reliable comparative modeling. Nevertheless, low sequence similarity between the GPCR family members is still a serious obstacle not only in template selection but also in providing theoretical models of acceptable quality. An additional level of difficulty is the prediction of kinks and bulges in transmembrane helices. Usage of multiple templates and generation of alignments based on sequence profiles may increase the rate of success in difficult cases of comparative modeling in which the sequence similarity between GPCRs is exceptionally low. Here, we present GPCRM, a novel method for fast and accurate generation of GPCR models using averaging of multiple template structures and profile-profile comparison. In particular, GPCRM is the first GPCR structure predictor incorporating two distinct loop modeling techniques: Modeller and Rosetta together with the filtering of models based on the Z-coordinate. We tested our approach on all unique GPCR structures determined to date and report its performance in comparison with other computational methods targeting the Rhodopsin-like class. We also provide a database of precomputed GPCR models of the human receptors from that class.

Availability: GPCRM SERVER AND DATABASE: http://gpcrm.biomodellab.eu.

Show MeSH
Antagonist docking to GPCRM-generated homology models versus self-docking: β1AR receptor (A) and D3R receptor (B).Structures of complexes with indicated polar contacts obtained by crystallography are shown in grey, while the docked structures are depicted in yellow. GPCRM-generated homology models are shown in green. Left panels show the best poses obtained in the docking to corresponding protein homology models. Right panels show results of self-docking to crystallographic structures (PDB id: 2VT4 and 3PBL). All polar contacts were preserved, except one hydrogen bond with Ser211 (A).
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3585245&req=5

pone-0056742-g005: Antagonist docking to GPCRM-generated homology models versus self-docking: β1AR receptor (A) and D3R receptor (B).Structures of complexes with indicated polar contacts obtained by crystallography are shown in grey, while the docked structures are depicted in yellow. GPCRM-generated homology models are shown in green. Left panels show the best poses obtained in the docking to corresponding protein homology models. Right panels show results of self-docking to crystallographic structures (PDB id: 2VT4 and 3PBL). All polar contacts were preserved, except one hydrogen bond with Ser211 (A).

Mentions: In a typical high-throughput virtual screening several thousand of various compounds are docked to a receptor structure. Such a large number imposes limitations on the docking precision and conformational sampling. Therefore, for testing the usefulness of GPCRM in drug design studies we have chosen fast, standard precision, flexible-ligand and rigid receptor docking in Glide with the default force field settings. The obtained results were compared to a self-docking test on the crystal structures of GPCRs performed by Glide with the same force field settings. The quality of the Rosetta-generated models seems to be sufficient to use them in virtual screening as the best (of the lowest RMSD) ligand poses (Table 3) contained properly oriented ligands in the binding site (Figure 5). In general the prediction of GPCR ligand binding modes is very challenging since even in the easy case of the self-docking to crystal structures not all the ligand rings are positioned properly (Figure S6–right panels in Supplementary Material S1). Most of rotamers of amino acids were properly predicted by GPCRM preserving polar contacts most important for the ligand binding. Nevertheless, falsely predicted rotamers of Thr112 (Figure S6–B in Supplementary Material S1) and Asp97 (Figure S6–C in Supplementary Material S1) caused a slight movement of ligands, yet preserving their proper orientation. In general, the quality of binding sites as well as the overall GPCR structures were much better (lower RMSD) in the case of loop modeling by Rosetta than by Modeller (Table 3). On average, the binding site area had been improved after the Rosetta step by 1–2 Å. Interestingly, the final Rosetta refinement slightly improved the rotamers in the TM region even though the protein backbone was restrained.


Towards improved quality of GPCR models by usage of multiple templates and profile-profile comparison.

Latek D, Pasznik P, Carlomagno T, Filipek S - PLoS ONE (2013)

Antagonist docking to GPCRM-generated homology models versus self-docking: β1AR receptor (A) and D3R receptor (B).Structures of complexes with indicated polar contacts obtained by crystallography are shown in grey, while the docked structures are depicted in yellow. GPCRM-generated homology models are shown in green. Left panels show the best poses obtained in the docking to corresponding protein homology models. Right panels show results of self-docking to crystallographic structures (PDB id: 2VT4 and 3PBL). All polar contacts were preserved, except one hydrogen bond with Ser211 (A).
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3585245&req=5

pone-0056742-g005: Antagonist docking to GPCRM-generated homology models versus self-docking: β1AR receptor (A) and D3R receptor (B).Structures of complexes with indicated polar contacts obtained by crystallography are shown in grey, while the docked structures are depicted in yellow. GPCRM-generated homology models are shown in green. Left panels show the best poses obtained in the docking to corresponding protein homology models. Right panels show results of self-docking to crystallographic structures (PDB id: 2VT4 and 3PBL). All polar contacts were preserved, except one hydrogen bond with Ser211 (A).
Mentions: In a typical high-throughput virtual screening several thousand of various compounds are docked to a receptor structure. Such a large number imposes limitations on the docking precision and conformational sampling. Therefore, for testing the usefulness of GPCRM in drug design studies we have chosen fast, standard precision, flexible-ligand and rigid receptor docking in Glide with the default force field settings. The obtained results were compared to a self-docking test on the crystal structures of GPCRs performed by Glide with the same force field settings. The quality of the Rosetta-generated models seems to be sufficient to use them in virtual screening as the best (of the lowest RMSD) ligand poses (Table 3) contained properly oriented ligands in the binding site (Figure 5). In general the prediction of GPCR ligand binding modes is very challenging since even in the easy case of the self-docking to crystal structures not all the ligand rings are positioned properly (Figure S6–right panels in Supplementary Material S1). Most of rotamers of amino acids were properly predicted by GPCRM preserving polar contacts most important for the ligand binding. Nevertheless, falsely predicted rotamers of Thr112 (Figure S6–B in Supplementary Material S1) and Asp97 (Figure S6–C in Supplementary Material S1) caused a slight movement of ligands, yet preserving their proper orientation. In general, the quality of binding sites as well as the overall GPCR structures were much better (lower RMSD) in the case of loop modeling by Rosetta than by Modeller (Table 3). On average, the binding site area had been improved after the Rosetta step by 1–2 Å. Interestingly, the final Rosetta refinement slightly improved the rotamers in the TM region even though the protein backbone was restrained.

Bottom Line: Usage of multiple templates and generation of alignments based on sequence profiles may increase the rate of success in difficult cases of comparative modeling in which the sequence similarity between GPCRs is exceptionally low.In particular, GPCRM is the first GPCR structure predictor incorporating two distinct loop modeling techniques: Modeller and Rosetta together with the filtering of models based on the Z-coordinate.We also provide a database of precomputed GPCR models of the human receptors from that class.

View Article: PubMed Central - PubMed

Affiliation: International Institute of Molecular and Cell Biology, Warsaw, Poland. dlatek@iimcb.gov.pl

ABSTRACT

Unlabelled: G-protein coupled receptors (GPCRs) are targets of nearly one third of the drugs at the current pharmaceutical market. Despite their importance in many cellular processes the crystal structures are available for less than 20 unique GPCRs of the Rhodopsin-like class. Fortunately, even though involved in different signaling cascades, this large group of membrane proteins has preserved a uniform structure comprising seven transmembrane helices that allows quite reliable comparative modeling. Nevertheless, low sequence similarity between the GPCR family members is still a serious obstacle not only in template selection but also in providing theoretical models of acceptable quality. An additional level of difficulty is the prediction of kinks and bulges in transmembrane helices. Usage of multiple templates and generation of alignments based on sequence profiles may increase the rate of success in difficult cases of comparative modeling in which the sequence similarity between GPCRs is exceptionally low. Here, we present GPCRM, a novel method for fast and accurate generation of GPCR models using averaging of multiple template structures and profile-profile comparison. In particular, GPCRM is the first GPCR structure predictor incorporating two distinct loop modeling techniques: Modeller and Rosetta together with the filtering of models based on the Z-coordinate. We tested our approach on all unique GPCR structures determined to date and report its performance in comparison with other computational methods targeting the Rhodopsin-like class. We also provide a database of precomputed GPCR models of the human receptors from that class.

Availability: GPCRM SERVER AND DATABASE: http://gpcrm.biomodellab.eu.

Show MeSH