W54–W59 Nucleic Acids Research, 2020, Vol. 48, Web Server issue Published online 2 June 2020 doi: 10.1093/nar/gkaa440 HomolWat: a web server tool to incorporate ‘homologous’ water molecules into GPCR structures 1 Eduardo Mayol , Adrián Garcı́a-Recio 1 , Johanna K. S. Tiemann 2,3 , Peter 2,3,4 W. Hildebrand , Ramon Guixà-González 5,6 , Mireia Olivella 7,* and Arnau Cordomı́ 1,* 1 Unitat de Bioestadistica, Facultat de Medicina, Universitat Autònoma de Barcelona, Bellaterra 08193, Spain, 2 Institute of Medical Physics and Biophysics, Medical University Leipzig, Leipzig, Sachsen 04107, Germany, 3 Institute of Medical Physics and Biophysics, Charité Universitätsmedizin Berlin, Berlin 101179, Germany, 4 Berlin Insitute of Health (BIH), 10178 Berlin, Germany, 5 Laboratory of Biomolecular Research, Paul Scherrer Institute (PSI), 5232 Villigen PSI, Switzerland, 6 Condensed Matter Theory Group, Paul Scherrer Institute (PSI), 5232 Villigen PSI, Switzerland and 7 Bioinformatics and Medical Statistics Group, University of Vic-Central University of Catalonia, Barcelona 08500, Spain Received March 19, 2020; Revised April 29, 2020; Editorial Decision May 11, 2020; Accepted May 14, 2020 ABSTRACT are the largest family of membrane proteins with over 800 members in humans. Despite sharing a common seven ␣- Internal water molecules play an essential role in the helical transmembrane architecture and similar conforma- structure and function of membrane proteins includ- tion changes upon activation (1,6), they recognize a wide ing G protein-coupled receptors (GPCRs). However, diversity of extracellular signals like hormones, neurotrans- technical limitations severely influence the number mitters or entire proteins. As a result, they play a key role in and certainty of observed water molecules in 3D signal transduction and have become targets of 35% of the structures. This may compromise the accuracy of currently approved drugs (7). Recent studies have demon- further structural studies such as docking calcula- strated that the activation of these receptors involves a spe- tions or molecular dynamics simulations. Here we cific order of internal water molecules (8). Moreover, molec- present HomolWat, a web application for incorporat- ular dynamics simulations have shown that internal water ing water molecules into GPCR structures by us- molecules are highly conserved among GPCRs and par- ticipate in their common activation mechanism (9). Over- ing template-based modelling of homologous wa- all, this emphasizes the important role that water molecules ter molecules obtained from high-resolution struc- play in GPCR function. tures. While there are various tools available to pre- Sequence homology and phylogenetic analyses have clas- dict the positions of internal waters using energy- sified GPCRs into six families (or classes), namely A to F based methods, the approach of borrowing lacking (10). Recent advances in protein engineering and structural water molecules from homologous GPCR structures biology have led to a rapid growth in the number of GPCR makes HomolWat unique. The tool can incorporate structures deposited in the Protein Data Bank (PDB). Thus, water molecules into a protein structure in about 10 years ago only 4 GPCR structures of class A had been a minute with around 85% of water recovery. The deposited in the PDB whereas today this archive hosts 346 web server is freely available at http://lmc.uab.es/ structures of GPCRs, 64 of which belong to unique recep- homolwat. tor subtypes of four different GPCR classes (i.e. A, B1, C and F) (GPCRdb, http://gpcrdb.org/structure/statistics, 2020). However, technical limitations including resolution INTRODUCTION severely influence the number and certainty of solved water Water molecules confined inside cavities in a protein, named molecules in GPCR structures. In this context, molecular ordered or internal water molecules, play an essential role in modeling can help improve and maximize internal hydra- the structure and function of proteins, ligand binding mech- tion of these proteins. Current tools and methods to pre- anisms, and catalytic reactions. In several proteins like G dict water placement in proteins (reviewed in (11)), range protein-coupled receptors (GPCRs), water molecules also from knowledge-based or molecular mechanics methods to mediate the core mechanism of activation (1–5). GPCRs simulations approaches (e.g. molecular dynamics or Monte * To whom correspondence should be addressed. Tel: +34 945812344; Fax: +34 94581246; Email: arnau.cordomi@uab.cat Correspondence may also be addressed to Mireia Olivella. Email: mireia.olivella@uvic.cat Present address: Johanna K.S. Tiemann, Linderstrøm-Lang Centre for Protein Science, Department of Biology, University of Copenhagen, Kobenhavn, 2200, Den- mark.  C The Author(s) 2020. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Nucleic Acids Research, 2020, Vol. 48, Web Server issue W55 Carlo). One common pitfall of most of the former meth- RESULTS ods is the lack of experimental data supporting the mod- HomolWat pipeline eled position. Another group of tools such as PyWater (12) and ProBIS H2O (12) has focused on identifying conserved HomolWat protocol is schematically represented in Figure or homologous water molecules in proteins using experi- 1. The input of HomolWat is a file with the 3D structure mental data (12,13). The plethora of high-resolution GPCR in the PDB format (alternatively the use can select a PDB structures recently deposited in the PDB (14) have simulta- id and a specific chain) whereas the output is the same neously increased the number of solved waters, thus open- structure containing homologous internal water molecules ing a door to improving the placement of internal water obtained from the HomolWat reference database. This hy- molecules by structural homology. drated structure can be visualized interactively using NGL Here, we present HomolWat, a freely accessible web ap- viewer (19) and downloaded for further use. The overall plication (available at http://lmc.uab.cat/homolwat) aimed HomolWat protocol for water molecule placement contains at incorporating internal water molecules into GPCR struc- steps as follows: First, HomolWat uses Blast+ to perform a tures by using a molecular modeling method that borrows multiple sequence alignment of the input structure sequence lacking water molecules from homologous structures. Ho- against all GPCR sequences from receptors hosted in Ho- molWat relies on an up-to-date curated database of all in- molWatDB, resulting in a list of homologous structures ternal water molecules from high-resolution structures of with water molecules sorted according to their Blast+ score. GPCRs deposited in the PDB. The tool uses this informa- The user is then required to choose the GPCR functional tion to superpose water molecules from related structures state (i.e. inactive or active) in order to prioritize water in a hierarchical fashion. Water molecules that fit into re- molecules present in active or inactive structures. The user ceptor cavities not yet hydrated are incorporated into the can decide whether to try to incorporate or not a sodium model. Our method offers a novel, fast and reliable way to ion near the conserved residue D2.50 (numbering following place internal water molecules in GPCR structures. Ballesteros and Weinstein scheme (21)) present in most in- active structures (8,22) from the structure with the highest MATERIALS AND METHODS sequence identity and best resolution. The ion is introduced when there are no water molecules within a radius of 2.1 Å HomolWat reference database or protein atoms within 1.8 Å (23). Moreover, users have We have constructed an up-to-date reference database with the option to run popular energy-based method Dowser+ water molecules determined in all resolved GPCR struc- (24) alongside the main HomolWat protocol and incorpo- tures in the PDB (15). To ensure that coordinates have been rate predicted water molecules that do not overlap with ho- obtained using the latest experimental and computational mologous waters. This option becomes more useful when methods we have downloaded the structures from PDB- few homologous water molecules exist. In addition, using a REDO whenever possible (16). Any non-protein molecules Blast+ score threshold, the user can limit the incorporation other than water, or GPCR orthosteric/allosteric ligands of water molecules from close homologs only. Next, Ho- were removed from the structure. Additionally, auxiliary molwat performs a global structural alignment of homolo- proteins used to assist in structure stabilization were also re- gous structures to the query structure in descending order of moved. Water molecules with low order (B-factor > 45 Å2 ), Blast+ score and from highest to lowest resolution using the (17) were discarded. Water molecules were classified as in- align and super functions of the visualization software Py- ternal or external based on their circular variance, following Mol (v.2.0.5, (25)). Subsequently, the position of each wa- a previously reported method (18). We discarded external ter molecule is refined through a local structural alignment water molecules (those with circular variance < 0.6, com- of residues around 10 Å using the super function and only puted within a radius of 10 Å around the water oxygen) for those waters with a RMSD in the local structural alignment being incompatible with the membrane outside the crystal up to 2.0 Å are kept. Waters are assessed one by one in in- lattice. At the moment of writing HomolWat database con- creasing order of B-factor and incorporated into the model tains 191 receptor chains from 150 high-resolution struc- should they not clash with atoms from the query model tures and 44 unique receptors, totaling 2448 internal wa- or already incorporated water molecules (distance cutoff of ter molecules (see http://lmc.uab.cat/homolwat/gpcr table). 2.4 Å, (26)). The structure solvated with internal waters is The distribution of internal waters within GPCR classes shown interactively using an embedded NGL viewer (19) and subfamilies is shown in Supplementary Figure S1. that also shows the source structure (PDB id, chain and Uniprot entry name) of each water molecule. The solvated structure, a list with the incorporated water molecules and Implementation of the web service a PyMOL session with the structures used for water posi- HomolWat relies on a Python (v.3.7) backend that uses the tioning can be downloaded for further usage. Flask web framework (v1.1.1). Data for the internal wa- ter molecules is stored in a MySQL database (v.8.0.18). Test case The web server exploits the capabilities of the popular web- based viewer NGL (v.2.0.0, (19)) for structure visualization. To illustrate the use of HomolWat, we used our tool to Preprocessing of new GPCR structures and placement of place internal water molecules into the recently resolved water molecules is fully automated within a routine using crystal structure of the inactive serotonin 5-HT2A receptor Python and Bash scripts. Sequence alignment is performed at 2.9 Å resolution (PDB id 6A94, (27)) where no inter- using Blast+ (v2.6.0+, (20)). nal water molecules could be determined. We selected chain W56 Nucleic Acids Research, 2020, Vol. 48, Web Server issue Figure 1. Schematic representation of the HomolWat reference database (A) and overall pipeline (B). In (B), the user provides an input 3D structure file of a GPCR––that could contain ligands or water molecules––in PDB format and obtains an output PDB file of the same structure with additional homologous internal water molecules inserted. As a first step, HomolWat extracts the protein sequence from the PDB file and runs Blast+ on all receptor sequences in HomolWatDB. This provides a list of receptors sorted by Blast score from high to low sequence homology. Subsequently, HomolWat uses this list to assess if a water molecule fits in the receptor, considering previously inserted water molecules. This assessment is performed for every water molecule across all receptors within chosen Blast+ score, and across every structure for each receptor. A of this structure using the dropdown menu that loads spective water molecules of those structures included in the preprocessed structures without fusion protein, nanobod- test set. The results were compared to Dowser+ (24) predic- ies or other molecules used for crystallization purposes. tions, which we chose, among various alternatives (see (11)), We specified i) that the structure is in an inactive state, as a representative software able to find cavities in proteins ii) that we would like to add the conserved sodium, iii) and solvate them according to an energy criterion. Homol- that we do not want to use Dowser+ to predict additional Wat places a median of 86 molecules per receptor, consis- waters and that we will allow structures with a Blast+ tent with the number of experimentally determined water score threshold of 300 (i.e. include all amine receptors). In molecules within the protein core of rhodopsin (28). Figure about 15 seconds, HomolWat incorporated 25 water (Fig- 3 shows how many of the original internal water molecules ure 2A–B) molecules from the homologous 5-HT2B recep- for the tested 19 PDB structures were recovered by Homol- tor (61% sequence identity, 1 water), dopamine D2 recep- Wat using a cutoff radius <2 Å. The percentages of recovery tor (38%, 1 water), dopamine D4 receptor (37%, 3 waters), range between 41.2 and 100%, with a median of 84.6% (see ␤1 -adrenergic receptor (34%, 16 waters), the ␤2 -adrenergic Supplementary Table S1) and the average distance between receptor (34%, 2 waters) and the muscarinic M2 receptor original and recovered water is 0.68 Å (see Supplementary (29%, 2 waters). The hydrated model contains the con- Table S2). Larger recoveries are obtained for receptors with served water at the large proline-associated kink in trans- many homolog high-resolution structures, whereas poor re- membrane helix 6 (3) along with other known functional covery is obtained for receptors with a small number of de- water molecules (4,9) at regions that span from near the or- termined structures with water molecules and low homol- thosteric site to more cytoplasmic locations near residues ogy to other receptors. Our method outperforms previously N1.50, D2.50, N7.49 and Y7.53 (3,9) (Figure 2C–E). Inter- described knowledge-based methods like Dowser+ (24). estingly, HomolWat could not place the sodium ion as the mutated protein used for crystallization included a muta- tion (i.e. S3.39K) near residue D2.50, which introduces a CONCLUSIONS AND FURTHER DIRECTIONS large sidechain that hampers Na+ binding. HomolWat is a web application to introduce internal wa- ter molecules in GPCR structures using resolved water molecules from homologous structures. The tool uses a Validation database of GPCR structures containing internal water To validate the ability of HomolWat to introduce internal molecules to place homologous water positions into GPCR water molecules, we evaluated the percentage of recovery models or experimental structures with few or no inter- using a set of 19 Class A structures accounting for at least nal waters. Due to the foreseeable increase in the number one unique structure from any GPCR with resolution <2.8 of resolved high-resolution GPCR structures, HomolWat Å and more than five internal water molecules, where we will be able to use more templates, hence increasing its cur- had previously removed all water molecules. To avoid re- rent performance. Better GPCR models that explicitly in- dundancy, we excluded from the reference dataset the re- troduce internal water molecules may for instance improve Nucleic Acids Research, 2020, Vol. 48, Web Server issue W57 Figure 2. HomolWat test case: the crystal structure of the inactive serotonin 5-HT2A receptor (PDB id 6A94) without internal water molecules. (A) Screen- shot of the results for the Blast+ alignment to receptors in HomolWatDB. (B) Screenshot of the output PDB structure with the introduced internal water molecules displayed in NGL viewer. (C-E) Details of the clusters of water molecules associated to functionally important residues at different levels ranging from the orthosteric site to the G-protein binding crevice upon incorporation to its receptor core using HomolWat. The receptor is shown in light blue cartoons and water molecules are displayed as red spheres. Figure 3. Validation of HomolWat and comparison to Dowser+. Number of water molecules recovered using Homolwat (red squares) and Dowser+ (blue dots) upon removal of the originally resolved water molecules (black star) in each PDB structure (mapped to their Uniprot accession code). W58 Nucleic Acids Research, 2020, Vol. 48, Web Server issue docking calculations or molecular dynamics simulations. 5. Yuan,S., Filipek,S., Palczewski,K. and Vogel,H. (2014) Activation of HomolWat has been successfully applied as part of the G-protein-coupled receptors correlates with the formation of a continuous internal water pathway. Nat. Commun., 5, 4733. pipeline used in the GPCRmd project (http://gpcrmd.org), 6. Venkatakrishnan,A.J., Deupi,X., Lebon,G., Heydenreich,F.M., a community-driven effort to create the first open, interac- Flock,T., Miljus,T., Balaji,S., Bouvier,M., Veprintsev,D.B., Tate,C.G. tive and standardized database of GPCR molecular dynam- et al. (2016) Diverse activation pathways in class A GPCRs converge ics simulations (29). As shown in our test benchmark, Ho- near the G-protein-coupling region. Nature, 536, 484–487. molWat water placement pipeline showed a median recov- 7. Hauser,A.S., Attwood,M.M., Rask-Andersen,M., Schiöth,H.B. and Gloriam,D.E. (2017) Trends in GPCR drug discovery: new agents, ery of 83%. The fact that HomolWat employs experimen- targets and indications. Nat. Rev. Drug Discovery, 16, 829–842. tal knowledge to perform molecular modeling gives a clear 8. Liu,W., Chun,E., Thompson,A.A., Chubukov,P., Xu,F., Katritch,V., competitive advantage to our tool when compared to meth- Han,G.W., Roth,C.B., Heitman,L.H., IJzerman,A.P. et al. (2012) ods based on energy calculations such as Dowser+ (24), Wa- Structural basis for allosteric regulation of GPCRs by sodium ions. Science, 337, 232–236. terMap (30) or Waterdock (31). Due to its automated fash- 9. Venkatakrishnan,A.J., Ma,A.K., Fonseca,R., Latorraca,N.R., ion, this tool could expand to target water molecule place- Kelly,B., Betz,R.M., Asawa,C., Kobilka,B.K. and Dror,R.O. (2019) ment in other protein families. Diverse GPCRs exhibit conserved water networks for stabilization and activation. Proc. Natl. Acad. Sci. U.S.A., 116, 3288–3293. 10. Kolakowski,L.F. Jr (1994) GCRDb: a G-protein-coupled receptor DATA AVAILABILITY database. Receptors Channels, 2, 1–7. HomolWat is a web server freely accessible at http://lmc. 11. Nittinger,E., Flachsenberg,F., Bietz,S., Lange,G., Klein,R. and Rarey,M. (2018) Placement of water molecules in protein Structures: uab.es/homolwat. The reference database can be obtained From Large-Scale evaluations to Single-Case examples. J. Chem. Inf. on request or downloaded from the website. The source Model., 58, 1625–1637. code is available at https://github.com/EMayol/HomolWat. 12. Jukič,M., Konc,J., Gobec,S. and Janežič,D. (2017) Identification of conserved water sites in protein structures for drug design. J. Chem. Inf. Model., 57, 3094–3103. SUPPLEMENTARY DATA 13. Patel,H., Grüning,B.A., Günther,S. and Merfort,I. (2014) PyWATER: a PyMOL plug-in to find conserved water molecules in Supplementary Data are available at NAR Online. proteins by clustering. Bioinformatics, 30, 2978–2980. 14. Qu,X., Wang,D. and Wu,B. (2020) Progress in GPCR structure ACKNOWLEDGEMENTS determination. In: GPCRs. Elsevier, pp. 3–22. 15. Berman,H.M., Bhat,T.N., Bourne,P.E., Feng,Z., Gilliland,G., The authors thank current and former members of the Weissig,H. and Westbrook,J. (2000) The Protein Data Bank and the Laboratori de Medicina Computational for troubleshoot- challenge of structural genomics. Nat. Struct. Biol., 7(Suppl), 957–959. ing and discussion. 16. Joosten,R.P., Long,F., Murshudov,G.N. and Perrakis,A. (2014) The PDB REDO server for macromolecular structure model FUNDING optimization. IUCrJ, 1, 213–220. 17. Yuan,Z., Bailey,T.L. and Teasdale,R.D. (2005) Prediction of protein Spanish Ministerio de Ciencia e Innovación [SAF2015- B-factor profiles. Proteins, 58, 905–912. 74627-JIN to A.C., SAF2016-77830-R]; Deutsche 18. Mezei,M. (2003) A new method for mapping macromolecular topography. J. Mol. Graph. Model., 21, 463–472. Forschungsgemeinschaft (DFG) [SFB1423, project 19. Rose,A.S. and Hildebrand,P.W. (2015) NGL Viewer: a web number 421152132, subproject Z04, HI1502/12]; Stiftung application for molecular visualization. Nucleic Acids Res., 43, Charité, Einstein Center Digital Future (to P.W.H.); E.M. W576–W579. was a receiver of mobility grant [ECOST-STSM-CM1207- 20. Camacho,C., Coulouris,G., Avagyan,V., Ma,N., Papadopoulos,J., 220516-077546]. We acknowledge the support of COST Bealer,K. and Madden,T.L. (2009) BLAST+: architecture and applications. BMC Bioinformatics, 10, 421. Action CA18133 (ERNEST: the European Research Net- 21. Ballesteros,J.A. and Weinstein,H. (1995) [19]Integrated methods for work on Signal Transduction) and COST Action CM1207 the construction of three-dimensional models and computational (GLISTEN: GPCR-Ligand Interactions, Structures, and probing of structure-function relations in G protein-coupled Transmembrane Signalling: a European Research Net- receptors. Methods Neurosci., 25, 366–428. 22. Gutiérrez-de-Terán,H., Massink,A., Rodrı́guez,D., Liu,W., work). Funding for open access charge: COST Action Han,G.W., Joseph,J.S., Katritch,I., Heitman,L.H., Xia,L., CA18133 (ERNEST). IJzerman,A.P. et al. (2013) The role of a sodium ion binding site in Conflict of interest statement. None declared. the allosteric modulation of the A2A adenosine G protein-coupled receptor. Structure, 21, doi:10.1016/j.str.2013.09.020. 23. Mancinelli,R., Botti,A., Bruni,F., Ricci,M.A. and Soper,A.K. (2007) REFERENCES Hydration of sodium, potassium, and chloride ions in solution and 1. Tehan,B.G., Bortolato,A., Blaney,F.E., Weir,M.P. and Mason,J.S. the concept of structure maker/breaker. J. Phys. Chem. B, 111, (2014) Unifying family A GPCR theories of activation. Pharmacol. 13570–13577. Ther. Dent., 143, 51–60. 24. Morozenko,A., Leontyev,I.V. and Stuchebrukhov,A.A. (2014) 2. Yuan,S., Vogel,H. and Filipek,S. (2013) The role of water and sodium Dipole moment and binding energy of water in proteins from ions in the activation of the ␮-opioid receptor. Angew. Chem. Int. Ed., crystallographic analysis. J. Chem. Theory Comput., 10, 4618–4623. 52, 10112–10115. 25. Schrödinger,LLC. The PyMOL Molecular Graphics System, Version 3. Pardo,L., Deupi,X., Dölker,N., López-Rodrı́guez,M.L. and 2.0.5. Campillo,M. (2007) The role of internal water molecules in the 26. van Beusekom,B., Touw,W.G., Tatineni,M., Somani,S., Rajagopal,G., structure and function of the rhodopsin family of G protein-coupled Luo,J., Gilliland,G.L., Perrakis,A. and Joosten,R.P. (2018) receptors. ChemBioChem, 8, 19–24. Homology-based hydrogen bond information improves 4. Angel,T.E., Chance,M.R. and Palczewski,K. (2009) Conserved crystallographic structures in the PDB. Protein Sci., 27, 798–808. waters mediate structural and functional activation of family A 27. Kimura,K.T., Asada,H., Inoue,A., Kadji,F.M.N., Im,D., Mori,C., (rhodopsin-like) G protein-coupled receptors. Proc. Natl. Acad. Sci. Arakawa,T., Hirata,K., Nomura,Y., Nomura,N. et al. (2019) U.S.A., 106, 8555–8560. Nucleic Acids Research, 2020, Vol. 48, Web Server issue W59 Structures of the 5-HT2A receptor in complex with the antipsychotics (2020) GPCRmd uncovers the dynamics of the 3D-GPCRome. Nat. risperidone and zotepine. Nat. Struct. Mol. Biol., 26, 121–128. Methods, doi:10.1038/s41592-020-0884-y. 28. Fried,S.D.E., Eitel,A.R., Weerasinghe,N., Norris,C.E., Somers,J.D., 30. Abel,R., Young,T., Farid,R., Berne,B.J. and Friesner,R.A. (2008) Fitzwater,G.I., Pitman,M.C., Struts,A.V., Suchithranga M,D and Role of the active-site solvent in the thermodynamics of factor Xa Brown,M.F. (2019) G-protein-coupled receptor activation mediated ligand binding. J. Am. Chem. Soc., 130, 2817–2831. by internal hydration. Biophys. J., 116, 207a. 31. Sridhar,A., Ross,G.A. and Biggin,P.C. (2017) Waterdock 2.0: water 29. Rodrı́guez-Espigares,I., Torrens-Fontanals,M., Tiemann,J.K.S., placement prediction for Holo-structures with a pymol plugin. PLoS Aranda-Garcı́a,D., Ramı́rez-Anguita,J.M., Stepniewski,T.M., One, 12, e0172743. Worp,N., Varela-Rial,A., Morales-Pastor,A., Lacruz,B.M. et al.