Towards definition of an ECM parts list: an advance on GO categories.

Those of us interested in the extracellular matrix (ECM) are faced with significant challenges of definition. ECM proteins are large, complex and assembled into crosslinked insoluble matrices. This has meant that defining the biochemical composition of ECMs has been difficult. Nonetheless, protein chemistry and molecular biology have defined many familiar ECM proteins — collagens, proteoglycans, laminins, thrombospondins, tenascins, fibronectins, etc. With the completion of many genomes it should now be possible to develop complete “parts lists” for the ECM. Such lists are needed for analyzing data from “omic” approaches such as expression arrays, latest-generation sequencing and proteomics. These approaches generate long lists and it is typically necessary to extract from those lists the genes/proteins of interest. Anyone who attempts to do this using the commonly used gene ontology (GO) categories soon discovers that they are largely useless for defining ECM proteins. Many ECM proteins are unannotated and those which are, are sorted, with little evidence of logic or consistency, into diverse categories such “extracellular matrix,” “basement membrane,” “cell surface” and many others. The human and mouse orthologs are often found in different categories and attempts to use GO categories to extract a complete list of ECM genes or proteins from a data set are unsatisfactory at best. Faced with this problem in the course of an ECM proteomics project, we decided we needed to develop a better list of ECM proteins (Naba et al., 2012). This turned out to be not entirely straightforward. While the familiar ECM glycoproteins, collagens and proteoglycans could be collected relatively easily, the standard procedures for collecting lists of homologs, using BLAST or domain-based searches, quickly run into problems. ECM proteins are characteristically formed from multiple domains and those domains are shared among different ECM proteins and also with non-ECM proteins (Hohenester and Engel, 2002; Adams and Engel, 2007). Two obvious examples among many are EGF and FN3 domains, both very ancient protein domains that predate the origins of extracellular matrix in metazoa. They are found in many ECM proteins but also in diverse membrane proteins and secreted factors. So a BLAST or domain search produces a confusing mix of “homologs.” It is the domain architecture (the entire domain composition, number and order) that defines families of ECM proteins. Nonetheless, the domain composition of ECM proteins does offer the route to defining an essentially complete list of ECM proteins and sorting them from homologous but non-ECM proteins (i.e., those which share one or more domains with ECM proteins but are clearly not matrix proteins — examples would include many tyrosine kinase or phosphatase receptors or adhesion receptors). We started with a list of characteristic ECM domains and used them to pull out all genes/proteins containing those domains from the human and mouse genomes/proteomes. We then culled those lists using a list