Bioinformatics Resources for the Study of Glycan-Mediated Protein Interactions

Podsumowanie

This protocol illustrates how to explore, compare, and interpret human protein glycomes with online resources.

Streszczenie

The Glyco@Expasy initiative was launched as a collection of interdependent databases and tools spanning several aspects of knowledge in glycobiology. In particular, it aims at highlighting interactions between glycoproteins (such as cell surface receptors) and carbohydrate-binding proteins mediated by glycans. Here, major resources of the collection are introduced through two illustrative examples centered on the N-glycome of the human Prostate Specific Antigen (PSA) and the O-glycome of human serum proteins. Through different database queries and with the help of visualization tools, this article shows how to explore and compare content in a continuum to gather and correlate otherwise scattered pieces of information. Collected data are destined to feed more elaborate scenarios of glycan function. Glycoinformatics introduced here is, therefore, proposed as a means to either strengthen, shape, or refute assumptions on the specificity of a protein glycome in a given context.

Wprowadzenie

Glycans, proteins to which they are attached (glycoproteins) and proteins to which they bind (lectins or carbohydrate-binding proteins) are the main molecular actors at the cell surface¹. Despite this central role in cell-cell communication, large-scale studies, including glycomics, glycoproteomics, or glycan-interactomics data are still scarce compared to their counterpart in genomics and proteomics.

Until recently, methods for characterizing the branching structures of complex carbohydrates while still being conjugated to the carrier protein had not been developed. The biosynthesis of glycoproteins is a non-template-driven process in which the monosaccharide donors, the accepting glycoprotein substrates, and the glycosyltransferases and glycosidases play an interactive role. The resulting glycoproteins can bear complex structures with multiple branching points where each monosaccharide component can be one of the several types present in nature¹. The non-template-driven process imposes biochemical analysis as the only option for generating oligosaccharide structural data. The analytical process of glycan structures attached to a native protein is often challenging as it requires sensitive, quantitative, and robust technologies to determine monosaccharide composition, linkages, and branching sequences².

In this context, mass spectrometry (MS) is the most widely used technique in glycomics and glycoproteomics experiments. As time goes, these are carried out in higher throughput settings and data is now accumulating in databases. Glycan structures in various formats³, populate GlyTouCan⁴, the universal glycan data repository where each structure is associated with a stable identifier irrespective of the level of precision with which the glycan is defined (e.g., possibly missing linkage type or ambiguous composition). Very similar structures are collected but their minor differences are clearly reported. Glycoproteins are described and curated in GlyConnect⁵ and GlyGen⁶, two databases cross-referencing each other. MS data supporting structural pieces of evidence are increasingly stored in GlycoPOST⁷. For a wider coverage of online resources, chapter 52 of the reference manual, Essentials of Glycobiology, is dedicated to glycoinformatics⁸. Interestingly, glycopeptide identification software has proliferated in recent years^9,10 though not to the benefit of reproducibility. The latter concern prompted the leaders of the HUPO GlycoProteomics Initiative (HGI) to set a software challenge in 2019. The MS data obtained from processing complex mixtures of N- and O-glycosylated human serum proteins in CID, ETD, and EThcD fragmentation modes, were made available to competitors whether software users or developers. The full report on the results of this challenge¹¹ is only outlined here. To begin with, a spread of identifications was observed. It was mainly interpreted as caused by the diversity of methods implemented in search engines, of their settings, and how outputs were filtered, and peptide "counted". The experimental design may also have put some software and approaches at a (dis)advantage. Importantly, participants using the same software reported inconsistent results, thereby highlighting serious reproducibility issues. It was concluded by comparing different submissions that some software solutions perform better than others and some search strategies yield better results. This feedback is likely to guide the improvement of automated glycopeptide data analysis methods and will in turn, impact database content.

The expansion of glycoinformatics led to creating web portals that provide information and access to multiple similar or complementing resources. The most recent and up-to-date are described in a chapter of the Comprehensive Glycoscience book series¹² and through cooperation, a solution to data sharing and information exchange is offered in an open access mode. One such portal was developed which was originally called Glycomics@ExPASy¹³ and renamed Glyco@Expasy, following the major overhaul of the Expasy platform¹⁴ that has hosted a large collection of tools and databases used across several -omics for decades, the most popular item being UniProt¹⁵-the universal protein knowledgebase. Glyco@Expasy offers a didactic discovery of the purpose and usage of databases and tools, based on a visual categorization and a display of their interdependencies. The following protocol illustrates procedures to explore glycomics and glycoproteomics data with a selection of resources from this portal that makes the connection between glycoproteomics and glycan-interactomics explicit via glycomics. As it is, glycomics experiments produce structures where monosaccharides are fully defined and linkages partially or fully determined, but their protein site attachment is poorly, if at all, characterized. In contrast, glycoproteomics experiments generate precise site attachment information but with a poor resolution of glycan structures, often limited to monosaccharide compositions. This information is pieced together in the GlyConnect database. Furthermore, search tools in GlyConnect can be used to detect potential glycan ligands which are described along with the proteins recognizing them in UniLectin¹⁶, linked to GlyConnect via glycans. The protocol presented here is divided into two sections to cover questions specific to N-linked and O-linked glycans and glycoproteins.

Protokół

NOTE: A device with an Internet connection (larger screen preferred) and an up-to-date Web browser such as Chrome or Firefox is required. Using Safari or Edge may not be as reliable.

1. From a protein N-glycome in GlyConnect to a lectin of UniLectin

1. Accessing resources from Glyco@Expasy
   NOTE: The procedure described here is to access GlyConnect but can be applied to accessing any resource recorded in the platform.
   1. Go to https://glycoproteome.expasy.org/glycomics-expasy and consider the bubble chart on the right showing different categories such as Glycoconjugates or Glycan Binding. In the leftmost menu that reflects the categories in the bubbles, check the Glycoproteins box so that the bubble chart on the right immediately zooms in the bubble matching that category.
      NOTE: Green bubbles are tools and yellow bubbles are databases. Clicking on either one zooms in again to provide details on the resource. Before doing so, the user may want to understand the dependencies of that resource to others.
   2. To get the information on dependencies, move from the Resource Thematic Classification tab to the Resource Dependency Wheel tab. Place the mouse on GlyConnect in the wheel to check its level of integration with other sources (Figure 1).
   3. Go back to the Resource Thematic Classification tab to reach the GlyConnect bubble as in step 1.1.1 and click on it (Supplementary Figure 1) to display the GlyConnect homepage in a new tab that shows the statistics of the content in the latest release of the database.
      NOTE: A color scheme detailed in Table 1 matches the different types of information stored in the database. This color code is valid in all entity pages in GlyConnect and is consistent throughout. The homepage also displays four sections dedicated to focused datasets such as those describing the glycosylation of the Sars-Cov-2 spike protein (COVID-19) or extensively detailing human milk oligosaccharides (HMO). These will not be explored in this protocol.
2. Exploring the contextual information of a protein N-glycome
   NOTE: All glycan structures in GlyConnect are displayed in three alternative and commonly used formats: (1) Symbol Nomenclature For Glycans (SNFG)¹⁷ (2) IUPAC condensed¹⁸, and (3) Oxford¹⁹. In contrast, there is no standard notation to express glycan composition. In GlyConnect, the following code is used: Hex for hexose, HexNAc for N-Acetylhexosamine, dHex for fucose, and NeuAc for sialic acids. For the sake of simplicity, visualizing tools rely on a condensed notation: H for hexose, N for N-Acetylhexosamine, F for fucose, and S for sialic acids. Additionally, small letters designate modifications such as "a" for acetylation, "p" for phosphorylation, and "s" for sulfation, for the most frequent of these so-called substituents.
   1. To view and explore the N-glycome of human Prostate Specific Antigen (PSA), from the GlyConnect homepage, proceed as follows.
      NOTE: The glycosylation of human PSA has been studied over the years, especially in the context of prostate cancer. The GlyConnect database stores three references^20,21,22, which combine glycomics and glycoproteomics data. Note that the results provided here were obtained with the September 2021 release of GlyConnect. Ulterior use of the database may yield slightly different statistics due to frequent data updates.
   2. Select the PROTEIN button to open the protein view of the database. In the protein view page, type prostate in the search window. Look for the two entries listed in the output distinguishing two isoforms of PSA with distinct pI values. Click on 790 (Id column) corresponding to the common isoform of PSA.
      NOTE: Look for the top multi-colored bar that shows summary information extracted from the published work in the scheme detailed above. Several options for navigation are possible as described below.
   3. On the top multi-colored bar, click on the SOURCE button in green to display the sample types from which the published data were processed: Urine and Seminal Fluid. To browse this information further, click on either of these sample types. The same applies to any item that appears when clicking on a colored button.
   4. To check the health-related content of the database, click on the DISEASE button, which contains two items, one of which is Prostate Cancer that links to the corresponding dedicated disease page in GlyConnect. The summary for that page shows that three large-scale studies have reported 319 compositions on 1,087 sites found in 308 human proteins.
   5. Click on the STRUCTURE button to view the full list of 135 structures associated with PSA from glycomics data. Click on the COMPOSITION button for the associated 78 compositions determined by glycoproteomics experiments. Click on any structure or composition to obtain further details.
      NOTE: Details such as the list of alternative proteins carrying the particular structure or the list of structures matching the composition can be obtained. PSA is known to have only one N-glycosylation site at Asn-69 (only one item counted for the brown SITE button).
   6. To reduce the ambiguity of compositions, click on SUGGESTED STRUCT below a selected composition (for example, Hex:6 HexNAc:3 NeuAc:1). A suggestion is made each time the monosaccharide count coincides with that of a structure listed above (Figure 2).
      NOTE: The Hex:6 HexNAc:3 NeuAc:1 composition generated by a glycoproteomics experiment is matched to four higher resolution structures from the glycomics data. In the case of PSA, there is no site ambiguity to resolve since only Asn-69 is glycosylated.
   7. To fully explore the protein page, view further details on the right side of the page (Figure 3).
      1. View the default 3QUM PDB (Protein Data Bank²³) entry for PSA that is shown with two complex glycans attached to each monomer (Figure 3) or the alternative 2ZCK entry, which is also available because of an attached carbohydrate. The second entry shows a single chain.
         NOTE: Both entries are visualized with the 3D LiteMol plugin²⁴ that displays glycans in SNFG-3D notation adopted in the PDB-RCSB.
      2. Click on the corresponding links of other cross-references to explore relevant functional information from major proteomics databases, such as UniProt (Figure 3).
3. Visualizing and correlating the contextual information of a protein N-glycome
   NOTE: As seen in the previous section, long lists of structures or compositions can be hard to apprehend as a whole and GlyConnect relies on two different tools to visualize key information, namely, GlyConnect Octopus and GlyConnect Compozitor (the first one expands the summary information captured in colored buttons and the second brings out structural dependencies in terms of a structure/composition being contained in another). As illustrated below, GlyConnect Octopus explores associations between the various entities stored in the database through highlighting multiple or single connections as a reflection of the database content.
   1. Perform a GlyConnect Octopus search to confirm the presence of common structural traits such as hybrid core structures and highly frequent sialic acid-containing structures in the diversity of glycans attached to PSA, as described below.
   2. Go to the Octopus homepage https://glyconnect.expasy.org/octopus/. Keep the N-linked tab selected by default. Move to the Cores subtab and click on the Hybrid icon. Move to the Properties subtab and click on the Sialylated icon. Click on the green Search button below.
      NOTE: The search results are graphically displayed as relationships between three categories of items. By default, the center list matches the query for compositions, the left collection spans related proteins, and the right one spans related glycans.
   3. In the displayed graph of relationships, hover over H6N4F1S1 to highlight links to six proteins and three structures. Contrast this by hovering over H6N4F2S1 that singles out the two isoforms of PSA (both referred to as UniProt ID: KLK3_HUMAN) and one structure (ID: 10996). Hover over the structure ID to show its SNFG representation and click on it to open the corresponding page (Supplementary Figure 2).
   4. Change the nodes of the Octopus to any other topic describing the context of glycosylation. The color code remains the same as the one described earlier (see Table 1).
      1. Change the Center Nodes to Tissues to display 15 options in the middle of the graph, many of which are body fluids. Look for all the associations between proteins and glycans matching the query depending on tissue information. Place the cursor on Urine or Seminal Fluid in the middle of the graph to view different associations (Figure 4A,B).
      2. Change the Center Nodes to Disease to display 13 options, one of which is Prostate Cancer. The only protein associated is PSA (KLK3_HUMAN) (Supplementary Figure 3).
         NOTE: A closer look at the PSA N-glycome shown in the protein page singles out the very high frequency of a terminal NeuAc(a?-?)Gal(b?-?)GlcNAc substructure in many cases on structures with two or three antennae. Another Octopus can be generated on that basis as described below.
   5. Click on the Clear button to refresh the search. Move to the Properties subtab and click on the Bi-antennary icon. Move to the Determinants subtab and click on the 3-Sialyl-LN (type 2) icon. Click on the green Search button below.
   6. Check the Octopus-retrieved associations with bi-antennary glycans containing a terminal 3-Sialyl-LN (type 2) motif, i.e., NeuAc(a1-3)Gal(b1-4)GlcNAc. Change the Center Nodes to Tissues for easier reading and hover over KLK3_HUMAN to directly connect Seminal Fluid with PSA common isoform and seven structures (Supplementary Figure 4).
      NOTE: The second visualization tool, GlyConnect Compozitor, performs the scan of potential relationships between each and every composition in a list thereof (see below). A relationship is defined as differing from only one monosaccharide between two compositions. These identified relationships plotted in a graph expose the (dis)continuity of a glycome.
   7. Use GlyConnect Compozitor to perform the scan of potential relationships between each and every composition in a list thereof, as illustrated below for the case of PSA.
      NOTE: GlyConnect Compozitor processes compositions in association with a context. It offers distinct tabs for querying GlyConnect, e.g., Proteins, Sources, Cell Lines, Diseases that are self-explanatory to qualify a context. This is illustrated here with PSA as follows.
   8. Go back to the protein page of PSA: https://glyconnect.expasy.org/browser/proteins/790. On the right side of the PSA entry page, click on the Compozitor link. Ensure that the Compozitor search fields are pre-filled with the details of the Id 790 entry in the Protein tab (Protein: Prostate-specific Antigen, Species: Homo sapiens, and Glycan Type: N-linked).
   9. Click on the Add to Selection button to retrieve data from the database and display the graph of connected compositions. Deselect the Include Virtual Nodes option. Click on the Compute Graph button to display a graph showing a well-connected set of 78 compositions representing the PSA N-glycome, and a bar plot showing the main characteristics of the glycans.
   10. Hover over the purple bar in the bar plot, which locates all sialylated structures in the graph to reveal an observable bias toward sialylated structures.
   11. Remain in the main Protein tab and select Prostate-specific antigen - high Pi isoform (psah) in the Protein (name) field.
       NOTE: The Glycan Type and Glycan Site fields are automatically filled.
   12. Click on the Add to Selection button to retrieve data from the database that amounts to 57 compositions. Click on the Compute Graph button to generate the superimposed graphs of both isoforms and assess the differences in glycomes of the two PSA isoforms. Hover over node labels to prompt the display of the number of structures corresponding to the compositions/labels (Figure 5).
4. Glycan-binding information in UniLectin
   NOTE: Recall the determinant tested in the Octopus, described as NeuAc(a2-3)Gal(b1-4). By definition, it is an established binding part of a glycan structure and, as such, can be searched in the UniLectin3D database²⁵.
   1. Go to https://www.unilectin.eu/ and click on the UniLectin3D button. Alternatively, directly go to the page: https://www.unilectin.eu/unilectin3D/.Click on the Glycan Search button to open this page: https://www.unilectin.eu/unilectin3D/glycan_search (​Supplementary Figure 6).
   2. Click on the purple diamond representing a sialic acid, which prompts the display of all glycan-binding motifs ending with a sialic acid stored in the database. The top part of that collection of motifs contains the NeuAc(a2-3)Gal(b1-4)GlcNAc motif investigated earlier (Supplementary Figure 7).
   3. Click on the NeuAc(a2-3)Gal(b1-4)GlcNAc motif to prompt the display of all lectins for which a 3D structure confirming the interaction with NeuAc(a2-3)Gal(b1-4)GlcNAc is known. The result by default shows lectins in all species. Use the Search by Field option to limit the view to human-centric information.
   4. Click on the Search by Field option. In the species field, type Homo sapiens. Click on the Explore X-ray Structures button to filter out the original list. Only one entry remains, i.e., the human galectin-8. Click on the View the 3D Structure and Information button on the upper-right corner of the listed item to display detailed information of human galectin-8 interacting with NeuAc(a2-3)Gal(b1-4)GlcNAc.
   5. Access the structural information on human galectin-8 displayed on the page with two different viewers.
      1. Hold the mouse to turn the molecule around and bring the ligand to the fore with the Litemol software²⁶ integrated to show the lectin 3D structure. Mouse over one of the listed interactions on the left to update the view on the right and locate where that particular interaction acts in the structure with the PLIP software²⁷ integrated to detail atomic interactions between the lectin and the ligand (Figure 6).
   6. Click on any green button that links to the corresponding entries in UniProt, PDB (European or American sites), and GlyConnect to explore these cross-references.

2. Exploring and comparing O-glycomes in GlyConnect

1. Browsing the HGI challenge high confidence dataset
   NOTE: The HGI dataset mentioned in the introduction is stored in the GlyConnect database. It contains 163 N- and 23 O-glycopeptides found in 37 glycoproteins considered as a high confidence list. GlyConnect Compozitor²⁸ is key to assessing glycome data consistency. Importantly, Compozitor allows for virtual nodes (shown in gray) when only one intermediary step is needed to connect the isolated nodes. In that way, virtual nodes tighten the graph and can be interpreted as structures potentially missed in the experimental results.
   1. Browse the HGI dataset from the GlyConnect homepage by going directly to the reference page of the article: https://glyconnect.expasy.org/browser/references/2943.
      NOTE: The summary in the colored buttons partially reflects the figures provided in the article. Yet, if only 69 unique peptides are listed, this reflects multiple associations between peptides and sites or structures. In the article, a glycopeptide is defined as a unique combination of a peptide and a composition. In GlyConnect, glycosites are first considered, and they are described as a combination of a peptide with structures. This explains the discrepancy in figures between GlyConnect and the above citation.
   2. Check the high frequency of occurrence of N-linked compositions, such as Hex:5 HexNAc:4 NeuAc:2, identified on 42 sites in 43 peptides as opposed to the frequent uniqueness of most O-linked compositions identified on 1 site in 1 peptide.
   3. Click on the Compozitor link on the right side of the reference entry page to assess the consistency of the dataset. Ensure that the Compozitor tool directly processes the DOI of the reference and fills the search field with reference=10.1101/2021.03.14.435332 in the Advanced tab of the tool. Type &glycan_type=O-linked after the DOI number to narrow down the search to O-linked glycans, so that the query becomes: reference=10.1101/2021.03.14.435332&glycan_
      type=O-linked
   4. Click on the Add to Selection button to retrieve data from the database (there are 20 O-linked compositions). Keep the Include Virtual Nodes option selected. Click on the Compute Graph button to display the graph of connected compositions. This result highlights several gaps in the expected continuity of glycan biosynthesis with nine virtual nodes required to complete the graph (Figure 7).
2. Comparing with the O-glycome of a selected serum protein in GlyConnect
   NOTE: To assess whether the gaps can be filled by data stored in GlyConnect, one O-glycosylated protein out of the 37 listed with the reference was selected. In the dataset, Inter-alpha-trypsin inhibitor heavy chain H4 (Q14624) is reported to be an O-glycosylated on Thr-725.
   1. Go to the Protein tab of GlyConnect Compozitor (see step 2.1.3). From the Protein list, select Inter-alpha-trypsin inhibitor heavy chain H4. Ensure that the Species selection is Homo sapiens by default. Deselect N-linked in the Glycan Type. Select only Thr-725 in the Site list by first clicking on the minus sign to the left of Site to deselect all the sites, and then selecting only Thr-725 from the list.
   2. Click on the Add to Selection button (note that six compositions are associated with Thr-725). Click on the Compute Graph button to display the graph of connected compositions (Supplementary Figure 8).
   3. Observe the displayed graph, which shows the 17 unique compositions out of the 20 O-linked compositions of the article dataset in blue and the three unique ones out of six in the database in red. In other words, the overlap between the two sources is present in three compositions that are represented in magenta. Note that a 45° rotation of the graph is generated automatically.
      NOTE: The number of virtual nodes is reduced by one. As it turns out, H2N2S1 missing in the 20 O-linked compositions of the article dataset and represented as a virtual node is now filled with an additional composition associated with Thr-725 of Inter-alpha-trypsin inhibitor heavy chain H4 in the database. This simplifies the topology of the graph because two other virtual nodes are rendered useless since they were alternative options for filling the gap between H1N2S1 and H2N2S2. Yet, a second composition imported from the database would be isolated if not for the creation of two new alternative virtual nodes H2N2F1S1 and H1N2F2S1.
   4. To make sense of the virtual nodes, check whether the corresponding compositions are present in GlyConnect. To do this, click on the Export button below the graph. Select Virtual only by deselecting all other options. Click on the clipboard icon to copy the selection of 8 compositions.
   5. Paste the selection in the query window of Compozitor's Custom tab. Select O-linked in the Glycan Type field. Set the Selection Label in the Compositions field to, for example, VN to name the list of 8 compositions.Click on the Add to Selection button, and then on the Compute Graph button. All virtual nodes are now displayed as green nodes (Figure 8).

Wyniki

The first part of the protocol (section 1) showed how to investigate the specificity or the commonality of N-glycans attached on Asn-69 of the human Prostate Specific Antigen (PSA) using the GlyConnect platform. Tissue-dependent (urine and seminal fluid), as well as isoform-dependent (normal and high pI) variations in glycan expression, were emphasized using two visualizing tools (Figure 4 and Figure 5).

First, GlyCon...

Dyskusje

GlyConnect Octopus as a tool for revealing unexpected correlations
GlyConnect Octopus was originally designed to query the database with a loose definition of glycans. Indeed, the literature often reports the main characteristics of glycans in a glycome such as being fucosylated or sialylated, being made of two or more antennae, etc. Furthermore, glycans whether N- or O-linked are classified in cores, as detailed in the reference manual Essentials of Glycobiology¹, that are ...

Materiały

Name	Company	Catalog Number	Comments
internet connection	user's choice
recent version of web browser	user's choice