SCOP database in 2002: refinements accommodate structural genomics

BACKGROUND

The SCOP (Structural Classification of Proteins) database (1) is a comprehensive ordering of all proteins of known structures, according to their evolutionary and structural relationships. The basic classification unit is a protein domain. A domain is defined as an evolutionary unit, in the sense that it is either observed in isolation in nature, or in more than one context in different multidomain proteins. Protein domains in SCOP are grouped according to species and hierarchically classified into families, superfamilies, folds and classes, whose meaning is described in the original articles (1,2).

A unique aspect of SCOP is that it embeds a theory of evolution as defined by a human expert, rather than the necessarily more limited set of rules implemented by a series of algorithms and automatic tools. Automation in SCOP is meant to support and extend human ability; to make expert knowledge broadly available to students and the scientific community; and, last but not least, to sustain the automation process itself, given increasing needs and constant number of persons.

The growth in content of the SCOP database closely tracks the corresponding growth in the number of deposited structures. With the exception of major software upgrades, like in the case of 1.55, in the last 2 years SCOP has been released every 4–6 months, and included the classification of all proteins whose coordinates were available in the PDB at the time we started working on the release (usually a few weeks before the release itself, a few months in the case of SCOP 1.55).

An early SCOP release in 1995 included 3179 domains clustered into 498 families, 366 superfamilies and 279 different folds (1). The first seven classes in the current release (1.55) comprise 30 403 domains, grouped into 605 different folds, 947 superfamilies and 1557 families. These domains correspond to 12 794 entries in the Protein Data Bank (3,4) and 39 references from the literature, for which the experimental coordinates are not available. The 10-fold increase in the number of domains since 1995 roughly produced twice as many folds, 2.5 times as many superfamilies and three times as many families.

Together with the ASTRAL Compendium (5), which provides sequences for all domains filtered according to different criteria, SCOP introduces a meaningful order in the huge amount of sequence and structural information coming out of the various genomics projects, and forms the basis upon which other services, like SUPERFAMILY (6), can be built.

This article briefly summarizes the new features introduced in SCOP starting with release 1.55. For a detailed description, the reader is referred to the online release notes (http://scop.mrc-lmb.cam.ac.uk/scop/release-notes.html).

NEW FEATURES IN SCOP

The original design of the SCOP database (7) proved to be flexible enough to accommodate in a relatively easy way not only the growth in number of experimentally determined protein structures in the last 7 years, but also deeper modifications of the database itself, like the recent introduction of unique identifiers.

In computational terms, although SCOP is essentially a hierarchy, a mechanism for cross-linking between nodes of the tree makes it a graph, which allows the representation of complicated biological relationships (8), as well as the more restrictive parent–child relationships in a tree.

Since the original implementation of SCOP is based on a description of the underlying data structure, rather than on the data structure itself, it is easy to introduce new classification levels, whose need was clear since the very beginning (2), but which have not been actually included so far. To introduce a new level, all that is required is to modify the description accordingly, and the rest will fall into place automatically.

HOW STABLE ARE STABLE IDENTIFIERS GOING TO BE?

Both SCOP unique identifiers (sunid) and SCOP concise classification strings (sccs) are expected to be stable across releases, with the caveat that a domain definition or its classification can change. A typical example is when a domain in a multidomain protein already classified in SCOP is observed for the first time either by itself, or in a different context, and therefore qualifies as a separate domain. Also, nodes in the hierarchy can merge or split as more evidence about evolutionary relationships becomes available from experimental data. In these cases, we will make corresponding identifiers obsolete, and introduce new ones. We will not reuse any identifier (in order to avoid identifiers with more than one meaning), and we will provide an easy-to-parse history tracking all changes.

A few examples can help explain what kind of changes should be expected to occur, and why.

It is not uncommon for only a fragment, or some of the domains of a protein to be crystallized. When the structure of the N-terminal domain of syntaxin 1-A was released (1br0) (13), it appeared to have a three-helix bundle fold with a topology similar to that of spectrin repeats, and therefore was classified in SCOP 1.50 as a new superfamily under the spectrin-like repeat fold. By the time of the SCOP 1.53 release, two new structures for the same protein became available (1ez3 and 1dn1) (14,15). While the three chains in 1ez3 are three-helix bundle fragments, the B chain in 1dn1 is a four-helix bundle with a peculiar arrangement which prompted the classification of all syntaxins as a separate superfamily under the STAT-like fold instead. A comment in SCOP indicates which of the structures are from a smaller fragment.

Perhaps more biologically interesting is the itinerary of the N-terminal domain of mukB, a protein implicated in ATP-dependent chromosome partitioning during cell division, and for which a putative nucleotide-binding region was predicted on the basis of a strong sequence signal for a P-loop motif. When the experimental structure of the N-terminal domain of mukB was determined (1qhl) (16), it came as a surprise. The predicted P-loop turned out to be located in between two helices on the protein’s surface. There was no structural similarity to known P-loop NTPase proteins. Accordingly, mukB was classified in SCOP on its own, as a new fold.

By the time of the next SCOP release, two new structures were solved: Rad50 ATPase (1f2t) (17) and SMC head domain (1e69) (18). The two ATPase proteins are closely related to each other and to mukB, despite a low sequence similarity. They show all the characteristic motifs of the ABC transporter ATPase domain, and a convincing structural similarity to other members of this family (17,18). The N-terminal domain of mukB corresponds to the N-terminal part of SMC, and the structural similarity suggests a similar mode of dimer complementation, in which the helix preceding the P-loop in mukB is rearranged in a β-strand when both the N- and C-terminal domains of mukB are translated as a complete head domain (18).

Because of this additional experimental evidence, the N-terminal domain of mukB is now classified under the ABC ATPase domain-like family, which also includes Rad50 and SMC.

This family is also a good example to illustrate another point, mentioned before, and source of possible changes in the near future. The ABC transporter ATPase family belongs to P-loop NTP hydrolases, a highly divergent superfamily that, should we decide to introduce new levels in SCOP, would classify as a hyperfamily. This would cause a rearrangement of the underlaying substructure, with the current superfamily becoming a hyperfamily, families within the superfamily turning into superfamilies, and proteins being regrouped into newly defined families.

In summary, stable identifiers will be as stable as the information assembled in SCOP is, and vary as a function of new evidence and new findings, or in order to accommodate new levels in SCOP. All variations will be recorded in an easy-to-track way and will constitute the history of the evolution not of the proteins themselves, but of how we come to know them, and we organize and structure this knowledge.

CONCLUDING REMARKS

Biological databases in all their diversity are an indispensable contribution to the scientific enterprise for interpreting and composing the huge amount of experimental data into a coherent understanding of life. The role played by these databases can only increase as the volume and intrinsic complexity of biological data rapidly expand.

The added value of a database is in the accuracy of its content, and in the structured knowledge that it embeds and makes accessible to a wider community. The SCOP database has been available on the web since 1994, and continously updated to include all newly determined protein structures since then, with the purpose of providing a coherent theory of how proteins have evolved. In this article, we have described some recent enhancements that will make its growth, usage and integration in the larger context of structured biological information available on the web a relatively easy and affordable task, even with the daunting perspective of 10 000 new proteins coming out of the various structural genomics projects in the next 10 years (19).