MatrixDB, a database focused on extracellular protein-protein and protein-carbohydrate interactions.

SUMMARY: MatrixDB (http://matrixdb.ibcp.fr) is a database reporting mammalian protein-protein and protein-carbohydrate interactions involving extracellular molecules. It takes into account the full interaction repertoire of the extracellular matrix involving full-length molecules, fragments and multimers. The current version of MatrixDB contains 1972 interactions corresponding to 4412 experiments and involving 259 extracellular biomolecules. AVAILABILITY: MatrixDB is freely available at http://matrixdb.ibcp.fr

1 INTRODUCTION

Most of the cells in multicellular organisms are surrounded by an extracellular matrix (ECM), which is comprised of proteins (∼500 genes) and complex polysaccharides termed glycosaminoglycans (GAGs). GAGs play crucial roles either within the ECM or at the cell surface (Rodgers et al., 2008). A global view of the extracellular network is required to understand how the ECM conveys information, and how its assembly is regulated. Interactions involving extracellular biomolecules are under-represented in interaction databases, which mainly focus on interactions occurring inside the cell and report few protein–carbohydrate interactions. Furthermore a number of extracellular proteins are multimeric, and are able to form supramolecular assemblies either alone (Ricard-Blum et al., 2005) or in association with other molecules (Kielty, 2006). This is important because the oligomerization state may modulate the ability of proteins to interact with other molecules (Jokinen et al., 2004). Enzymatic cleavage of full-length extracellular biomolecules also affects the interaction network by releasing fragments with specific molecular recognition properties referred to as matricryptins (Davis et al., 2000, Ricard-Blum et al., 2005). Matricryptins are enzymatic fragments of ECM containing exposed biologically active cryptic sites (matricryptic sites) that are revealed after structural or conformational alteration of these molecules. Matricryptic sites and matricryptins have been reported within protein components of the ECM as well as in GAGs (Davis et al., 2000). We have built a database to report protein–protein and protein–carbohydrate interactions involving extracellular molecules. MatrixDB takes into account all the above features and mainly focuses on mammalian molecules. It integrates data from four major interaction databases, in-house literature curation and binding experiments performed with protein and polysaccharide arrays. The current version of MatrixDB contains 1972 interactions involving 259 extracellular biomolecules. MatrixDB is a member of the International Molecular Exchange (IMEx) consortium.

2 BUILDING MATRIXDB

2.1 Biomolecule sources

Protein data were imported from the UniProtKB/Swiss-Prot database (Bairoch et al., 2005) and identified by UniProtKB/Swiss-Prot accession numbers. In order to list all the partners of a protein, interactions are associated by default to the accession number of the human protein. The actual source species used in experiments is indicated in the page reporting interaction data. Intracellular and membrane proteins were included to obtain a comprehensive network of the partners of extracellular molecules. Indeed, ECM proteins and GAGs bind to a number of membrane proteins or cell-associated proteoglycans and some of them interact with intracellular partners upon internalization (Dixelius et al., 2000). ECM proteins were identified by the UniProtKB/Swiss-Prot keyword ‘extracellular matrix’ and by the GO terms ‘extracellular matrix’, ‘proteinaceous extracellular matrix’ and their child terms. The proteins annotated with the GO terms ‘extracellular region’ and ‘extracellular space’, which are used for proteins found in biological fluids, were not included because circulating molecules do not directly contribute to the extracellular scaffold. Additionally, 96 proteins were manually (re-)annotated through literature curation. These new or corrected annotations will be submitted to UniProtKB/Swiss-Prot. Glycan-binding proteins were annotated with cross-references to the KEGG GLYCAN database when available (Hashimoto et al., 2006). We created specific identifiers for fragments (PFRAG_number), multimers (MULT_number), cations (CAT_number), lipids (LIP_number) and GAGs (GAG_number), with cross-references to the ChEBI database when available (Degtyarenko et al., 2008).

2.2 Interaction sources

MatrixDB integrates 1378 interactions from the Human Protein Reference Database (HPRD, Prasad et al., 2009), 211 interactions from the Molecular INTeraction database (MINT, Chatr-Aryamontri et al., 2007), 46 interactions from the Database of Interacting Proteins (DIP, Salwinski et al., 2004), 232 interactions from IntAct (Kerrien et al., 2007a) and 839 from BioGRID (Breitkreutz et al., 2008) involving at least one extracellular biomolecule of mammalian origin. We added 283 interactions from manual literature curation and 65 interactions from protein and GAG array experiments. All interaction data, except those imported from HPRD and BioGRID, are reported according to the MIMIx standard (Minimum Information required for reporting a Molecular Interaction experiment, Orchard et al., 2007). The coverage of the extracellular interaction network is not complete and MatrixDB is regularly updated.

2.3 Architecture of the database

We used AceDB, a database system developed by Durbin and Thierry-Mieg (1994). SwissKnife (Hermjakob et al., 1999) was used to extract data from UniProtKB/Swiss-Prot. HPRD, DIP, MINT, IntAct and BioGRID data, available in PSI-MI XML 2.5 (Kerrien et al., 2007b), were transformed into MatrixDB format. AcePerl (Stein and Thierry-Mieg, 1998) and AceBrowser, a set of Perl-CGI scripts, provided a customizable and straightforward browsable interface to MatrixDB. An Apache HTTP server hosts MatrixDB, which is freely available at http://matrixdb.ibcp.fr

3 QUERYING MATRIXDB

All members of a category (i.e. all the extracellular molecules, proteins, fragments, multimers, GAGs, lipids or cations) can be displayed. Biomolecules can be searched by their common name, UniProtKB/Swiss-Prot (or CheBI for GAGs) accession number or gene name. GO terms, UniProtKB/Swiss-Prot keywords, PubMed IDentifiers or author names can be used to query the database. Interactions from each database can be retrieved. All the binding partners of a biomolecule are listed on the ‘Biomolecule Report’ page. The list of experiments reporting an interaction is accessible on the ‘Association Report’ page. Experimental data (binding site, kinetic and affinity constants) are displayed on the ‘Experiment Report’ page. Selected molecules can be saved in a cart for building the corresponding interaction network. A tutorial is available on-line and interaction data can be downloaded from the MatrixDB website in PSI-MI XML 2.5 and MITAB 2.5.

4 VISUALIZATION OF INTERACTION NETWORKS

We have developed a set of scripts based on Cytoscape (Shannon et al., 2003) to visualize the extracellular interaction network or sub-networks. Biomolecule types are shape- and color-coded. Besides non-covalent physical interactions, links are displayed between a multimer and its constitutive monomers, a fragment and its parent molecule, and the protein and GAG moieties of proteoglycans (Fig. 1, Supplementary Material). Sub-cellular localization of the biomolecules can be visualized. Biomolecule and interaction data are imported into Cytoscape and displayed when clicking on a node (biomolecule) or on an edge (link) of the network. Alternatively, interaction networks can be visualized on MatrixDB website using Medusa (Hooper and Bork, 2005).

5 CONCLUSIONS

The architecture of MatrixDB takes into account the structural and functional complexity of the ECM organization, including protein–carbohydrate interactions. The analysis of the extracellular interaction network will be used to investigate the mechanisms of ECM assembly and homeostasis, and to determine how genetic and acquired diseases interfere with these processes.