1. The history of the Protein Data Bank

The establishment of the Protein Data Bank (PDB) began in the 1970's as a grassroots effort. A group of (then) young crystallographers, including Edgar Meyer, Gerson Cohen and myself, began discussing the idea of establishing a central repository for coordinate data at an American Crystallographic Association (ACA) meeting in Ottawa, Canada, in 1970. Those conversations were continued with a larger group at the ACA meeting in Columbia, South Carolina, USA, in 1971. At that meeting, a petition was written and a proposal was submitted to the United States National Committee for Crystallography (USNCCr). Later that year, the Cold Spring Harbor (CSH) Symposium was held on `Structure and Function of Proteins at the Three Dimensional Level' (Cold Spring Laboratory Press, 1972). This meeting, characterized by David Phillips as a `coming of age', heralded a new era in biology. The discussions within the meeting room, on the lawn, and on the beach were exciting and intense. In an informal meeting convened by Max Perutz, protein crystallographers discussed how best to collect and distribute data. Until that point, coordinates for individual entries had only been exchanged among a few research laboratories using punched cards. Since each atom was represented by a single card, an exchange of a structure the size of myoglobin required more than 1000 cards. By providing a central repository for these data, the PDB would make such an exchange possible for anyone.

Walter Hamilton was also in attendance. A chemist at Brookhaven National Laboratory (BNL) and a leader in the crystallographic community, Hamilton had begun to focus on two new science and technology projects. In collaboration with postdoctoral fellow Tom Koetzle and others, he was working on the determination of the structures of all the amino acids using neutron diffraction (Lehmann et al., 1972). In another collaboration, he was developing new computer technologies for graphics and for remote computing with Edgar Meyer (Meyer, 1997). During the CSH meeting, Hamilton was approached with the idea that had been discussed within the ACA community – a public data bank of protein structures. At an ad hoc meeting of protein crystallographers attending the Symposium, it was proposed that there should be a repository with identical files in the United Kingdom and in the USA. Hamilton volunteered to set up the American data bank at Brookhaven.

When Max Perutz returned to England, he discussed this proposal with Olga Kennard, who was the founder of the Cambridge Crystallographic Data Centre (CCDC) (Kennard et al., 1972; Allen et al., 1973), and had wide experience in assembling and archiving crystallographic data. Walter Hamilton wrote to her with an offer of collaboration and proposed to meet and discuss some of the details of coordinating the activities. He visited England that summer and, by October 1971, the establishment of the Protein Data Bank archive, jointly operated by the CCDC and BNL, was announced in Nature New Biology (Protein Data Bank, 1971). After Hamilton's untimely death in 1973, Koetzle took over the direction of the PDB, and with the support of key members of the community – most especially Michael Rossmann and Fred Richards – the PDB was able to survive. In 1974, the first PDB Newsletter was distributed to describe the details of data deposition and remote access. At this point, thirteen structures were ready for distribution and four were pending.

According to the January 1976 report to the ACA Council, the PDB archive contained 23 structures and 375 data sets had been distributed to 31 laboratories in that year. A grant for USD 33000 from the National Science Foundation was awarded, and an Advisory Board consisting of David Davies, Fred Richards and Ken Neet had been established. The project, which began as a dream of a community, finally had all the components of a fully fledged international resource (Bernstein et al., 1977).

The PDB remained in Brookhaven until 1998. In 1999, the management changed to a consortium called the Research Collaboratory of Structural Bioinformatics (RCSB PDB) consisting of Rutgers, The State University of New Jersey, the San Diego Supercomputer Center at the University of California San Diego (UCSD) and the National Institute of Standards and Technology (Berman et al., 2000). In 2003, the Worldwide PDB (wwPDB) formalized the existing international collaborations and an agreement was made among the RCSB PDB, the Macromolecular Structure Database at the European Bioinformatics Institute (MSD-EBI) and PDB Japan (PDBj) at the Institute for Protein Research at Osaka University, Japan (Berman et al., 2003). This set the stage for the PDB to remain a single uniform global resource of structural biology data.

2. Impact of technology and community action

The first decade of the PDB was characterized by attempts to capture the interest of the community in depositing their structures. Because there were relatively few structural biology groups at that time, it was possible to know virtually every member of the structural biology community. Letters were written and phone calls made to authors of articles reporting structures requesting that they deposit their coordinates in the PDB archive.

The rate of structure determinations began to change as technology became more advanced. The 1980's saw rapid development in all aspects of the structure determination pipeline. Molecular biology made it possible to clone genes and express proteins. Crystallization methods began to evolve and sparse-matrix methods were introduced (Jancarik & Kim, 1991). Synchrotron beamlines made it possible to obtain extremely intense X-rays (Helliwell, 1983). Detection methods evolved from collecting one reflection at a time to multiwire detectors to CCDs. Computer speeds and storage capacities increased dramatically. Graphics computers made it possible to visualize molecules and electron density. New computational methods for all stages of structure determination were developed. Multiple anomalous dispersion (MAD) using synchrotron radiation enabled the direct determination of structures (Hendrickson, 1991). Programs such as FRODO (Jones, 1978) and O (Jones et al., 1991) allowed computerized electron-density fitting. Methods to refine structures evolved with the use of geometrical and energy restraints (Hendrickson & Konnert, 1981; Brünger, 1990).

As it became easier to determine structures, more and more structures were published. Soon it became commonplace to see pictures of structures on the covers of journals such as Nature and Science. New journals, including Acta Crystallographica Section D and Nature Structural Biology (now Nature Structural & Molecular Biology), were established to report the results of the analyses of biological macromolecules. Many now believed that structural biology could give definitive information that would be key to understanding the molecular basis of biology and medicine.

As the value of structural biology became more obvious to other biologists, several committees were formed to look at the new demands from the community for required data deposition. The IUCr Commission on Biological Macromolecules set up a committee to determine exact policies. The ACA and the USNCCr also set up a committee. Fred Richards created an ad hoc group of crystallographers who felt strongly about creating a policy that would require data deposition. These groups worked for a few years to hammer out the exact guidelines. The timing of deposition of coordinates and of structure factors was discussed at length. During this period, commentaries were published in journals, letters to the editor were written and there was intense debate in the community. By the end of the decade, a petition resulting from the action of the Richards committee was produced and, in 1989, a formal guideline for data deposition was published (International Union of Crystallography, 1989). The guidelines stated that coordinates should be deposited at the time of publication and released within one year; structure factors should be deposited and released within four years. Most major journals adopted these guidelines and the National Institute for General Medical Sciences made the bold step of saying that funding would be contingent on the open sharing of structural data.

3. The content of the PDB

By the end of the 1980's, the number of structures in the PDB began to increase dramatically (Fig. 1) and that growth continues to date. Nuclear magnetic resonance (NMR) methods began to be used to determine structures, thus providing additional types of information to be archived in the PDB. These structures now make up about 15% of the structures in the archive.

Figure 1 The growth of the number of structures in the PDB archive 1972–2006.

In addition, the complexity of the structures that could be determined grew. The use of flash freezing and highly intense synchrotron radiation during data collection combined with the sophisticated use of non-crystallographic symmetry for structure determination made it possible to solve virus structures (Arnold & Rossmann, 1988) (Fig. 2). Even larger structures, including molecular machines such as the ribosome particles, became amenable to the methods of X-ray crystallography (Moore, 2001). The ability to freeze single particles also allowed structures to be determined using cryoelectron microscopy (cryoEM) and models for these structures began to appear in the PDB archive (Fig. 3).

Figure 2 A selection of the more than 250 icosahedral virus structures currently available in the PDB archive. The representation of these entries has recently undergone a major facelift.

Figure 3 Growth of the number of cryoEM structures in the PDB and the number of related maps in the Electron Microscopy Database (EMDB) (Henrick et al., 2003).

By the year 2000, the diversity of molecules and complexes in the PDB archive (Fig. 4) was so great that the possibility of actually understanding biology and medicine at a molecular level was not just a far off dream but something that might actually be realized.

Figure 4 A look at the diversity of structures in the PDB archive. These images, shown to scale, were created by David S. Goodsell (The Scripps Research Institute), who also writes and illustrates the RCSB PDB's Molecule of the Month feature. An expanded version of this figure is available for download from http://www.pdb.org.

Could this be achieved even more rapidly? Enter structural genomics. Analysis of the PDB archive showed that, from the point of view of amino acid sequences, there was about a sevenfold sequence redundancy (Hobohm et al., 1992). The number of new protein folds deposited in the PDB archive was relatively low (10% since 2000). At the same time, bioinformatics analysis of the newly determined gene sequences showed that the coverage of all protein families was uneven and incomplete. Would it be possible to very carefully select target sequences that would be novel (<30% sequence similarity), determine their structures and, through homology modeling, obtain structures for the rest of the protein families? Several projects ensued which focused on high-throughput methods for structure determination of these unique targets (Nature Structure Biology, 2000; Levitt, 2007). To date, structural genomics centres worldwide have been responsible for adding more than 5500 new structures of which 50.5% are novel.

4. Challenges to the PDB

When the PDB began there were relatively few structures, the structures were small, and only X-ray crystallographic methods were used to determine the structures. Today, the PDB has structures with molecular weights of more than two million, and new structure-determination methods are waiting in the wings to join NMR and cryoEM in the PDB Exchange Dictionary. To meet the challenges of handling these data and making them accessible to a broad community of scientists, methods had to be developed for data representation, acquisition and processing, management and distribution. The information in the PDB must be tied to other data resources, such as GenBank (Benson et al., 2000), UniProt (Apweiler et al., 2004), model organism databases and many, many others. In addition, the global nature of science makes it essential to coordinate these data efforts at an international level.

5. Applications of PDB data in academia and industry

The PDB is very widely used. For example, an average of 211515 files are downloaded from the ftp site each day. To date, there are more than 5000 references to the first RCSB PDB publication (Berman et al., 2000), making it one of the most-cited papers in all of biology. While the early users of the PDB were mostly crystallographers who used the resource to store their data and to review other structures for comparison with their own, now more than half of the papers that cite this one publication alone describe bioinformatic and computational analyses of structural data. Enormous efforts are under way to be able to understand protein folding so that perhaps someday it will be possible to predict structure from sequence (Moult, 2005). While the PDB is considered an archival data resource that stores and distributes primary data, there are hundreds of derivative databases that catalog the data in different ways. For example, CATH (Orengo et al., 1997) and SCOP (Conte et al., 2000) provide classifications of the folds found in proteins. More recently, there have also been efforts to understand protein–protein interactions (Janin et al., 2003) and again to try to predict these. There are specialty databases such as the Nucleic Acid Database (Berman et al., 1992) and the HIV Protease Structural Database (Ravichandran et al., 2002) that create in-depth resources for researchers in nucleic acids and HIV, respectively.

When the PDB updates the ftp site each week, most pharmaceutical companies download the new data for inclusion in their own in-house databases. These structural data are used to aid the discovery of new pharmaceuticals. Indeed, the ready availability of the structure of HIV protease (Navia et al., 1989) enabled many companies to concentrate their efforts on the development of effective protease inhibitors that are now the basis of AIDS treatment.

6. Future of the PDB

The PDB archive is a key example of a community resource that has evolved over its 36 year history. Its evolution has been driven by changes in science and technology used to determine structures, the nature of the structures that are determined, community attitudes about data sharing, and the nature of the communities that are interested in structural data.

In the short term, there will be several new challenges. There needs to be better representation for disordered structures, for X-ray structures refined with multiple models (Furnham et al., 2006) and for very large macromolecular complexes. New annotations will be required to describe function. All of these changes must be done in consultation with depositors, software developers and users of the data. Annotation practices used by the different data centers will continue to be examined and standardized so as to keep the archive uniform. It is likely that the wwPDB sites will develop a single deposition system. In the long term, the wwPDB will be able to develop new joint services for analysis and browsing of the rich data contained within the archive.

As the PDB continues to evolve, in addition to being able to use these data to perhaps predict structure, an even greater challenge will be to determine function by knowing the structure. Once accomplished, the long-term vision of enabling a molecular view of biology and medicine will become a reality.