MANORAA (Mapping Analogous Nuclei Onto Residue And Affinity) for identifying protein-ligand fragment interaction, pathways and SNPs.

Protein-ligand interaction analysis is an important step of drug design and protein engineering in order to predict the binding affinity and selectivity between ligands to the target proteins. To date, there are more than 100 000 structures available in the Protein Data Bank (PDB), of which ∼30% are protein-ligand (MW below 1000 Da) complexes. We have developed the integrative web server MANORAA (Mapping Analogous Nuclei Onto Residue And Affinity) with the aim of providing a user-friendly web interface to assist structural study and design of protein-ligand interactions. In brief, the server allows the users to input the chemical fragments and present all the unique molecular interactions to the target proteins with available three-dimensional structures in the PDB. The users can also link the ligands of interest to assess possible off-target proteins, human variants and pathway information using our all-in-one integrated tools. Taken together, we envisage that the server will facilitate and improve the study of protein-ligand interactions by allowing observation and comparison of ligand interactions with multiple proteins at the same time. (http://manoraa.org).

INTRODUCTION

Understanding protein–ligand interaction is crucial for drug discovery research, as it defines the binding affinity, steric complementarity of the surface and pharmacophoric patterns of the compound to the target protein. Favorable ligand interactions with protein such as suitable polar groups counterparts and proper hydrogen bonding partners are crucial for the ligand design process and the imperfect fit between the protein and the ligand will result in decreased binding affinity (1). A number of tools is available for visualizing and analyzing protein–ligand interaction; however, only few can provide comprehensive information such as verified binding affinity, and couple the results with the ligand interaction visualization available for multiple protein comparison in the same place (2). By understanding the favorable interactions between the target protein and a ligand of interest, one can start to rationalize drug design strategy and make the protein engineering possible by strengthening preferred interactions for instance.

To date, there are more than 100 000 structures in the Protein Data Bank (PDB) (3). However, it is not always straightforward to harness all the relevant information from the PDB. Querying the substructure of the ligands to return multiple molecular interactions that are available in the PDB can take a considerable amount of time as one normally goes through a series of non-intuitive steps. After multiple protein–ligand structures are retrieved, the comparison can be complicated and time-consuming, especially when the structures contain a large amount of protein–ligand interactions from multiple contacts points, which normally have to be investigated individually and manually. To the best of our knowledge, there is no existing tool specifically designed for comparative analysis of protein–ligand interactions in multiple structures at the time.

Two of the most popular tools for searching molecular interactions in the binding sites are Relibase (4) and PDBeMotif (5). Both tools are restricted to the structures in the PDB and are often used to show the distribution of protein–ligand binding patterns in the PDB as a whole. Other tools such as PLIP (6) are also available for investigating protein ligand interactions and visualization; however, the users cannot obtain knowledge of preferred interactions easily because it is dedicated to visualization and does not allow sorting by binding affinities, or viewing multiple proteins structures that bind to the same ligand in the same panel. The PLIC database (7) provides protein–ligand interaction clusters and also other related binding site information, and also has a superposition panel based on the clustering of similar binding sites. However, the ligand superposition is performed as a whole molecule, not based on the equivalent substructures, and hence, it is difficult to directly relate that information to the change in the binding affinities. WONKA (8) on the other hand, can offer observation from multiple structures but it requires the users to supply the set of superposed proteins with their equivalent amino acids renamed to the corresponding residue numbers. PoSSuM (9) aims to detect similar small molecule binding pockets; however, the overall similarity between pockets do not guarantee the same ligand binding pattern. A tool such as PLI (10) can also be used to find a particular ligand binding to a list of homologous proteins.

A direct query of ligand to the RCSB Protein Data Bank (3,11) returns the data retrieval in the form of PDB files and Jmol applet but does not provide ligand substructure analyses for multiple structures. The databases BindingDB (12) and Binding MOAD (13) emphasize the binding affinities data for further use such as for QSAR analysis (14), which does not offer the structural analysis of the binding site or the trend of binding affinity. In addition, these databases do not provide links from fragments to pathways or known human variants such as SNPs, a feature that will be useful for the drug design in the personalized medicine era.

To this end, we have developed the integrative web server MANORAA (Mapping Analogous Nuclei Onto Residue And Affinity) to facilitate understanding of ligand selectivity and promiscuity through the analysis of multiple protein structures on the web interface. It enables researchers to retrieve multiple chemical compounds and their binding partner proteins from the PDB, and compare and visualize the ligand-residue contact interactions all at the same time. Other useful functionalities include sorting of binding affinities of multiple proteins, as well as obtaining additional information such as protein functions, the species that a particular ligand is found in a complex, and the pathways that the ligand is found to take part in by linking to a pathway map such as KEGG (15), all in one place.

MANORAA: rationales, input and output

We built MANORAA with an aim to provide a user-friendly, one-stop service for ligand–protein interaction investigation. MANORAA was developed on top of CREDO, a database devoted specially to the protein–ligand interaction, which provides all pairwise atomic interaction contacts between ligand and proteins from the PDB in the form of a relational database (16,17). By filtering and ranking the interaction types in a systematic manner, the ligand contacts that are most important can be shown and can be related to the change in the binding affinities. This server provides integrated information about the target and off-target proteins interacting with the query ligand from the latest publicly available mirror of CREDO. MANORAA also provides protein–ligand binding affinity values from the Binding MOAD database (13) where the high quality binding affinity data are collected from literature. Importantly, the users can observe and compare the interaction by processing the ligand contact with multiple protein structures based on the complexes deposited to the PDB all at the same time. All queries to MANORAA start from one simple input page, and the results are provided in two output steps, as described here. We have extensively tested MANORAA on several common operating systems and web browsers (see Supplementary Data for details), and the most compatible browsers that we recommend are Windows version 7 or higher, and OSX Maverick or higher, and on Chrome 49 or higher, and Safari 9 or higher.

Input: chemical structure

The users can start with a ligand or part of a ligand of interest by providing one of the following as an input: (i) chemical name, (ii) SMILES expression, (iii) PDB ligand's 3-letter code or (iv) chemical structure (Figure 1). To facilitate generation of a SMILES expression, the MANORAA provides the SMILES lookup and then exports the SMILES to the chemical sketch panel as shown in Figure 1. The users can also select to create or edit a SMILES expression by drawing a chemical structure, or modify some parts and then import that sketch to a SMILES string before submission. From SMILES, the users can link to extended ligand names and compound bioactivity information via ChEMBL (18). The web server employs a JavaScript library called MarvinJS from ChemAxon to achieve the task. MarvinJS provides an HTML5-based user interface for chemical drawing, which allows the users to have an interactive interface without the need to install any additional plug-in.

Figure 1.

Input panel for ligand fragment via chemical names, SMILES expression, PDB ligand's three-letter code or chemical structure drawing.

Output 1: list of proteins interacting with a queried ligand

Once the users submit a ligand, a list of PDB entries that contain the submitted ligand will be returned. Ligand(s) of similar chemical structure and their target proteins will be returned if any part of the molecule matches with the SMILES input fragment. For example, Figure 2 shows a table of PDB entries interacting with a ligand ‘STU’ (Staurosporine). The binding affinity values, taken from Binding MOAD (13), are provided to help prioritize target proteins as they imply the binding strengths between the query ligand and the targets. For each entry, the following external information is also provided: (i) the pathway information from KEGG (15), (ii) the protein information from UniProt (19), (iii) the amino acid variants from SAMUL (20) and (iv) variants, isoforms and genomic context, protein/RNA baseline expression, gene ontology from the Centre for Therapeutic Target Validation (Open Targets, https://www.targetvalidation.org/). The server links crystallographic structures of protein–ligand interaction to related biochemical pathways via UniProt ID to KEGG ID mapping. The user can also link the proteins of interest to known human variants such as SNPs in the coding regions via the SAMUL web server (20). This essentially allows researchers to predict whether the candidate ligands will have a tendency to bind proteins with different annotated SNPs in the coding regions, an important step of drug design in the personalized medicine era. These results can be exported as a CSV table.

Figure 2.

MANORAA provides integrated analyses for ligand–protein interactions, linking structural biology to genomics, pathways and target information. Middle inset: the ligand panel on the left shows a chemical fragment with information of their protein binding partners on the right. This allows users to make a query on their molecular interactions with available binding affinity, and obtain additional information about the target proteins/genes. Top left: target gene information such as baseline expression is provided via the Open Targets project, using UniProt name. Top right: link to the protein structure information via PDBe. Bottom left: SNP information via SAMUL. Bottom right: KEGG pathway where the protein/gene of interest is highlighted. The results can be sorted by the proteins’ name, resolution, binding affinity and can be saved to a CSV file on the top of the page as needed.

This result page serves as an input form for the next step, which is to visualize three-dimensional (3D) structures of protein–ligand interactions based on the selection of (i) atom of interest in the ligand, and (ii) the target PDB chains. To aid this process, the MANORAA web server shows the ligand chemical structure so that users can pick up atoms of their interest interactively. By default, all the heteroatoms of ligands and PDB chains where the binding affinity are available are pre-selected.

Output 2: visualization of ligand–protein interactions

Once the users select ligand atoms and associated protein chains of interest, MANORAA will connect to CREDO to obtain interacting partner proteins of each ligand and PDB pair, grouped by nine interaction types and highlight them by different colors. The available interaction types are aromatic, hydrogen bond, ionic interaction, covalent bond, metal complex, carbonyl interaction, halogen bond, hydrophobic interaction and van der Waals clash (as shown in Figure 3 with criteria in Supplementary Data). The server will then rank the most important contact based on the shortest distances of unique interaction found for every atom per amino acid residue. JSmol (21), a JavaScript framework based on HTML5 for displaying interactive 3D molecular structures, has been employed into the user interface to enable the users to toggle display of the interaction partners in the 3D viewing panel. Display of ligand–protein interaction at each residue can be obtained by clicking the loading button of each PDB IDs, then choosing the residue name of interest. This step allows the user to have full control on what part of the chemical structure that they want to focus on. The results can be revisited using a unique URL provided. The list of target proteins and contact residues can be printed as PDF file together with the protein structures which can be saved from JSmol. Note that additional technical details of the web server can be found as Supplementary Data.

Figure 3.

Display of the ligand three-letter code STU (staurosporine) interacting with several proteins in the kinase family. The top row shows interaction with the lowest binding affinity value, which means staurosporine can tightly bind (0.00033 μM) via aspartate, and there are at most two hydrogen bonds (cyan) or ionic interaction (magenta) in combination. The second group has binding affinity values between 0.0065 to 0.010 μM and has three interactions that are either ionic or hydrogen bonds. All the others with binding affinity value more than 0.010 μM have one hydrogen bond, or ionic interaction, or no interaction at all.

To assist the first-time users, we have provided comprehensive step-by-step tutorial, demo video and sample pages on the web server. Here, we also provide two examples of how MANORAA can be employed to assist real-world drug design and protein engineering research.

Making use of MANORAA in ligand–protein interaction studies

Case study 1: the trend of interaction observed in N4 of STU interacting with the kinase family

To illustrate the use of MANORAA and its features, here we use our previous comprehensive study on staurosporine's binding strength as an example (22). The study demonstrated that staurosporine's strength of interaction with kinase depends on the number and the orientation of hydrogen bonds and ionic interactions made around the N4 atom of staurosporine. For the worst STU binding to kinase cases, N4 from PDB IDs: 1XBC, 1U59, 3HMO have only one bond in the non-preferable orientation while the one from 1Q3D does not have any hydrogen bonding or ionic interaction at all. These two structural observations imply poor binding affinity and hence they are present in the group of large binding affinity values. In contrast, 1OKY, 1NVR, 1STC and 1YHS have better binding affinities due to better hydrogen bonding and ionic interaction in both the structural orientation sense and also the amount of bonds they made which are at least two hydrogen bonds plus one ionic interaction. Note that binding via aspartate makes the interaction tighter than binding via arginine as seen in 1XJD and 2Z7R and hence shown with the lowest Ki or tightest binding affinities from the Binding MOAD database. These types of analysis will be beneficial to drug design because we know which part of the ligand is the major determinant of the binding affinities and what amino acids facilitate those preferred interactions. With MANORAA, these processes can be performed all in one place (Figure 3).

Case study 2: how MANORAA can be powerful for analyzing ligand and its binding protein with known SNPs that can be linked to diseases

Trifluoperazine (PDB ligand code TFP) was originally identified as an antipsychotic drug used in the treatment of schizophrenia, via blocking D2 dopaminergic receptors in the brain. However, trifluoperazine has recently been repurposed to inhibit the growth of cancer stem cells via its function as a calmodulin inhibitor, but the inhibitory mechanism is unclear (23,24). Using MANORAA, we can demonstrate that this ligand can also bind with many target proteins such as the placental calcium binding protein (S100-A4), calmodulin and troponin C (in a way, ‘off-targets’ to dopaminergic receptors). The result shows that the N3 atom of TFP interact with the placental calcium binding protein (S100-A4) via one hydrogen bond, or one optional ionic interaction in the cases of bovine calmodulin and human troponin C, suggesting the importance of this atom for all the proteins that bind to this small molecule. On the other hand, the N2 atom of TFP forms two ionic interactions with both bovine calmodulin structures but does not make any significant interaction for either S100-A4 protein or human troponin C. This kind of information can be useful for designing the selectivity of the drug.

Furthermore, MANORAA (via SAMUL) also reveals two SNPs in the human calmodulin gene that have been associated with ventricular tachycardia, a common side effect from trifluoperazine use. MANORAA also provides a list of multiple bovine calmodulin with crystallographic structures that harbor the ligand, enabling researchers to explore the effect of amino acid changes to affect ligand–protein interaction. This demonstrates how MANORAA can be used for an initial assessment of drug repurposing results. It should be noted that our data relies on crystallographic structures deposited to the PDB at the time. Saying that, MANORAA provides another way to make use of the growing PDB by linking the structures to human genome variations.

DISCUSSION AND FUTURE DIRECTIONS

Thanks to technological advances in crystallography and other methods for determining structures of biological molecules, the bottleneck of structural biology is now shifting from obtaining the structures to interpreting and linking them to other biological information such as pathways and genomic variants. With the wealth of information on ligand–protein interactions from publicly available databases such as PDB, it is now possible to perform a comparative study of multiple ligands and proteins (or drug candidate compounds and target proteins) at the same time. MANORAA has been established to facilitate these processes all in one place.

The web server has a number of useful features that assist the investigation of ligand–protein binding specificity, biological pathways the proteins are involved in and known human variants in the coding regions of the proteins. As depicted by Böhm, ligand with poor binding affinity is caused by missing crucial active site interactions in comparison with other tight binding ligands (1). Our service allows the user to compare the ligand's binding affinities with the numbers and the types of interactions that the ligand makes with multiple proteins, which should be useful for users to identify the key residues of proteins and the atoms of ligands in order to manipulate the interaction strength.

Existing protein–ligand contact and interaction databases are required in order to expedite the process of calculating and classifying the molecular interaction on the fly. To this end, we make use of the CREDO backend, and provide links to SAMUL (20), UniProt (19), PDBe (5), PDBsum (25) and KEGG (15). Note that the total size of calculated interaction for CREDO databases alone, including all the structures in the PDB, is very large (72 GB), but that would allow the interactions to be observed almost instantaneously. The graphical representation on JSmol (21) allows the users to view multiple structures with the fragment in the same window. The color highlight of the protein–ligand interaction that links to the JSmol structure visualization panel in real-time allows robust ligand interaction identification so that the researchers can relate the knowledge of the binding affinity value to the missing or occurring interaction by themselves. MANORAA employs a support responsive design, which means the output structures can be visualized without distortion even on a tablet or mobile phone. Another unique feature of MANORAA is its multiple-structures visualization panel with multiple loading buttons. The users can observe multiple structures one at a time and progress to each one to get an impression of the whole set of proteins that interact with this particular ligand fragment, and could identify the amino acid residue or atoms of a ligand that can be modified to fine-tune the ligand–protein binding interaction.

The main strengths of MANORAA over other previously available web servers aforementioned includes its flexibility of analyzing multiple experimentally verified ligand–protein interactions at the time, using its user-friendly and fast responsive interface. PLIP (6), for instance, focuses on the visualization of one structure at a time, while PoSSuM (9) provides the superposition and makes a comparison between residues surrounding a protein pair, rather than giving details of the type of residues in contact. Even though there are a number of tools that allow multiple structure observation, the interactions are not dissected to different chemical interaction types and do not provide visualization panel for the users to drill down to the level of ligand substructure interactions.

Looking ahead, we aim to routinely maintain the server and add new functionalities, which will be managed by a programmer dedicated to MANORAA's development and a multidisciplinary team. For instance, we have been developing a new algorithm to show gradient color of atomic position conservation. This will allow us to show position-specific interaction by highlighting the active site based on the percent conservation of the atomic position surrounding the ligand substructure. For the time being, we have implemented this for 18 staurosporine superposed complexes as an example from our sample page (see Supplementary Data). In addition, protein ligand contacts will be updated for every newer release of CREDO. In addition, PDBe (5), PDBsum (25), CACTVS (26,27), ChEMBL (18), KEGG (15), Open Targets (https://www.targetvalidation.org/), UniProt (19) and SAMUL (20) are accessed in real-time through their websites, hence the results shown will always be the most updated.

With MANORAA, the chemical fragments that have an influence on different pathways in different organisms can open the door for a more robust and insightful analysis for the study of multi-target drug design, species selectivity, off-target inhibition causing drug side effect problems. We envisage that MANORAA will provide a missing link between structural biology, systems biology and genetics information by one central concept surrounding the ligand's chemical structure to assist drug discovery and the probe molecules community.