Only a few decades ago, the elucidation of a single protein structure was a huge challenge. Today, the number of known structures is so large that the challenge has shifted to making sense of the great variety of tertiary structures in terms of their function and evolution. Recognition of the importance of domains as structural building blocks and potential elements of evolutionary change has led to the development of automated and semiautomated methods for the identification of domains and their classification. None of these methods is perfect because of enormous difficulties in identifying discontinuous domains (see Figure 1) or domains that are tightly associated. Nevertheless, the potential value of such methods for understanding protein function and evolution is undeniable. In the following, we shall describe one such method called CATH: Class, Architecture, Topology or fold, and Homologous superfamily. CATH is a semi-automated method; it has one strong competitor in SCOP, or Structural Classification of Proteins, which relies more heavily on human expertise and manual classification of domains.
Fig1. Two examples of multidomain proteins containing either distinct domains or closely related repetitive domains. (A) Pyruvate kinase, a three-domain protein. The central α/β barrel domain is present in numerous different enzymes that catalyze completely unrelated reactions. It is formed by different noncontiguous segments of the polypeptide chain coming in close, spatial proximity, hence the name discontinuous domain. The β-barrel regulatory domain is continuous, formed by a single stretch of the polypeptide chain. (B) A portion of the giant muscle protein titin which consists of repeating immunoglobulin-like domains. These domains are believed to confer elasticity to the muscle fibers. [B, adapted, courtesy of Wikimedia.]
CATH classifies protein domains at four levels: classes, which are the predominant secondary structure organization, α, β, or α/β; architecture, which is the overall shape; topology, which is the fold, spatial arrangement, and connectivity of secondary structural elements; and homology, which is the highest level of families and superfamilies (Figure 2). Figure 3 illustrates the distribution of domain structures among the three major protein classes. How are domains distributed overall among proteins? A comprehensive study in 2005 examined this question on a data base consisting of the total genome sequences of 150 organisms. Approximately 1 million protein-coding genes were identified, of which 850,000 could be classified into 50,000 families, each of unique domain organization. The remainder were mainly singletons and small proteins. By use of CATH and similar algorithms, 80% of the domains could be grouped into only 5,000 families, with the remainder falling into a category of new families.
Fig2. Structural classification of proteins by CATH. CATH is one of several existing databases that use algorithms to automatically assign domains in proteins with known sequences. The four major hierarchical levels for classifying proteins according to their structural features are Class, Architecture, Topology or fold, and Homologous superfamily. Seqid refers to sequence identity between the domains. [Adapted, courtesy of Christine Orengo and Ian Sillitoe, University College London, UK.]
Fig3. Structural classification of proteins by CATH. Distribution of domain structures among the three major classes: mainly α, mainly β, or mixed α/β, with representative examples in each class. α-Helices are colored in red; β-sheets are shown in yellow. [Adapted from Cuff AL, Sillitoe I, Lewis T et al. (2011) Nucleic Acids Res 39: D420–D426. With permission from Oxford University Press.]
Thus, a relatively small number of domains has been used to construct an enormous variety of proteins.
The next step is to predict protein function from the many sequences that we recognize in the genome but have not yet isolated as proteins. Unfortunately, the protein sequence does not in itself allow us, as yet, to predict the tertiary structure, except by analogy to proteins of very similar sequence. However, similarities between known protein tertiary structures provide better hints to function and evolution than do the sequences themselves.
