2.1. The SPINE website

The SPINE website, https://www.spineurope.org/ , aims to provide an information portal for all activities relating to the SPINE project and addresses three key audiences. Firstly, it provides information to the general public about all public aspects of the project including its background, partner contact details, project workpackages, an up-to-date scoreboard and list of structures, key publications, forthcoming meetings, data and tools generated by the project, project-derived protocols and news. Secondly, it provides a repository of information for members of SPINE to store and share private information such as meeting presentations, private protocols, unpublished structures and administrative information. Thirdly, it provides extended controlled access to members of the SPINE industrial platform. The website is built around the free open-source bulletin board system PHPBB (https://www.phpbb.com/ ) and it uses the common internet portal framework provided by PHP (Zandstra, 2004; https://www.php.net ) and MySQL (DuBois, 1999), upon which a custom-built content-management system was built.

2.2. The SPINE targets database

SPINE targets are held in a database at the EBI and are available in a format defined by an XML document-type definition file (target.dtd; https://www.ebi.ac.uk/msd-srv/msdtarget/ ). The database is populated with data provided by SPINE partners either using a Java loader (the EBI target manager, described below) or online through a target register (https://www.ebi.ac.uk/msd-srv/msdtarget/cgi-bin/spnreg/login.pl ) which can be run in-house for ORACLE or PostgreSQL (Matthew & Stones, 2001) databases. SPINE targets are annotated with data-integration tools that access public databases such as Medline (Macleod, 2002), EMBL (Hamm & Cameron, 1986), GO (The Gene Ontology Consortium, 2001) and PFAM (Bateman et al., 2004). Information is retrieved through web-based tools for searching and tracking, including SpineSearch (https://www.ebi.ac.uk/msd-srv/msdtarget/spine/SPINEindexp.html ), SpineStatus (https://www.ebi.ac.uk/msd-srv/msdtarget/status.html ), SpineScoreboard (https://www.spineurope.org/page.php?page=scoreboard ), Spine­StructuralGalleries (https://www.spineurope.org/page.php?page=structures ) and Spine­Alert (https://www.ebi.ac.uk/msd-srv/msdtarget/MSDtargetsAlert/ ).

A customizable target scoreboard service (https://www.ebi.ac.uk/msd-srv/msdtarget/cgi-bin/spnreg/login.pl ) is provided to enable multiple plots of target statistics against target status and species data. The EBI target-tracking system offers an efficient means of tracking the status of targets that are products of a single gene, following the lead of TargetDB (Chen et al., 2004). However, this framework cannot handle protein complexes, which form a significant proportion of SPINE targets. To complement the SPINE schema (https://www.ebi.ac.uk/msd-srv/msdtarget/target.dtd ), an extended schema (https://www.ebi.ac.uk/msd-srv/msdtarget/spine/new_SPINE_schema/ ) was developed to deal with not only simple targets but also complexes and targets that are either cocrystallized or co-expressed and allows referencing between targets. The Amsterdam partners have designed a database and data-capture system that allows individual targets to be combined as complexes. A more extensive overview of that work is reported by Romier et al. (2006).

The EBI target-manager application was developed using WebObjects, a system that uses Java to process queries and build webpages. The application was designed to allow for easy data entry by small laboratories. The system allows laboratory personnel to store and update data relevant to their proteins and the status of their projects at any time during their work, including dates, access privileges and contact information. The protein data are submitted via standard XML forms from the UniProt database (Bairoch et al., 2005). All data are initially stored in a local database and a scheduled weekly job builds an XML file in the EBI SPINE standard, which is automatically uploaded by the EBI to keep the status of targets up to date.

The targets database currently tracks information on 2260 SPINE targets distributed 66 and 34% between workpackages on pathogens and human proteins. A detailed breakdown of progress in these areas is reported in Alzari et al. (2006) and Banci et al. (2006). Of the 20 SPINE partners contributing targets, some have automated or semi-automated procedures, but the majority load targets manually with the EBI target manager.

2.3. Target selection and construct design

This activity divides into two distinct phases. Firstly, in order to understand the potential biomedical role of a candidate protein and hence to decide if the candidate actually warrants study, diverse data including the source organism, predicted domain organization, data on splicing variants, structures of related sequences and the significance of any known mutations must be organized into an information network and presented to the experimentalist. Secondly, for each particular protein (or complex) that appears interesting it is then important to determine whether one or more constructs can be defined that are likely to prove amenable to structural analysis (e.g. do they contain likely transmembrane helices or significant regions of disorder?). Within SPINE, several partners have built on their experience in these areas to provide tools addressing both of these phases. Although there is some commonality between the requirements of different laboratories and there was open exchange of ideas, no attempt was made to construct a single platform incorporating all these requirements. Nevertheless, the powerful tools described below are publicly available and the SPINE-adopted protein-production data model (Pajon et al., 2005) may provide a framework for assembling the next generation of more unified software.

2.3.2. Oxford

The Oxford node has developed a single resource for protein and DNA analysis, the Oxford Protein Analysis Linker (OPAL), under which sits the Oxford Protein Target Information Collection (OPTIC), a database for storing the results of these analyses. OPAL incorporates both publicly available analysis tools and bespoke tools developed in-house. These tools include the calculation of various parameters from the sequence such as pI, GRAVY (grand average of hydropathy; Kyte & Doolittle, 1982), extinction coefficient and rare-codon usage, the prediction of signal sequence and cellular location predictions, potential glycosylation sites, secondary-structure predictions, transmembrane region predictions and potential protease-cleavage sites. They also include BLAST-based (Altschul et al., 1990) sequence comparisons against publicly available databases, including CDD (Marchler-Bauer et al., 2005), PDB (Berman et al., 2000), SCOP (Andreeva et al., 2004) and SWISS-PROT (Boeckmann et al., 2003). The interface to OPAL is through a publicly available web form (https://www.oppf.ox.ac.uk/opal/OPAL.php ). One or more sequences can be entered in FASTA format or specified by GenBank accession No. (Benson et al., 2005) and the choice of analyses is made through a set of tick boxes. The results are returned as a web page that users can save locally. Where hits are found in external databases, the page includes links to the relevant web site. The results can also be presented graphically using SEView, a Java applet for browsing molecular-sequence data (Junier & Bucher, 1998; see Fig. 2). For authorized users, there is an alternative input form which uses locally installed (and licenced) versions of the tools and local databases to create and store a full annotation in the OPTIC database. Not only is this process considerably faster, the stored annotations are also checked regularly and can generate user alerts (by e-mail). Visualization of OPTIC entries are based on the SEView applet. For the whole-genome scale the process of annotation can be trivially scripted.

Figure 2 Presentation of automated sequence analysis performed using the integrated bioinformatics suite developed at Oxford (OPAL annotation tool and OPTIC database). The schematic shows the SEView-based display for one of the proteins annotated as part of the joint Oxford/York study of proteins from B. anthracis. The target protein sequence runs across the top of the diagram and coloured bars indicate regions of the target sequence that show significant sequence similarity to sequences in bioinformatics/structural databases. The width of each bar is related to the degree of sequence similarity. The different colours indicate different sources of information, e.g. yellow, PDB entries; orange, domain definitions; magenta, SCOP definitions; white, TargetDB entries. Clicking on any bar will bring up further information. The example indicated by the red arrow shows the detailed comparison of the target sequence to a COG family.

The final phase of target selection is construct design, which has been largely automated in the OPTIC-linked OPINE application and is based on the Primer3 tool (https://www-genome.wi.mit.edu/cgi-bin/primer/primer3_www.cgi ) customized into a Windows DLL. OPINE provides the link, at two levels, between the bioinformatics-based exercise of target selection and experimental work. Firstly, based on its knowledge of which expression strategies are available in the laboratory, OPINE provides a simple-to-use tool for the design of primers. Secondly, at the data-management level it transfers data on the designed constructs into the LIMS to make it available for authorization (where the primer orders are created in appropriate 96-well format) and experimental work. The construct-design strategy has proven its reliability in the work on Bacillus anthracis (see Au et al., 2006), with only two PCR failures for 114 Gateway (Invitrogen) pDEST14 and In-Fusion (Clontech) constructs, even when cloning out of long (>5 Mbp) genomic DNA. These construct-design tools are routinely used to define a set of constructs of high-value targets (as exemplified by the study of MICAL; Siebold et al., 2005). The possibility of using data analysis to inform construct design is still at a very early stage. Within SPINE, this activity is exemplified by the Oxford/York analysis of the B. anthracis targets (Au et al., 2006), which reinforces earlier findings from the Joint Center for Structural Genomics (JCSG; Page et al., 2003; Stevens, 2004).

2.3.3. Weizmann Institute

The Weizmann node has developed, tested and opened for public use several web-based tools for automatic annotation.

(i) FoldIndex (https://bioportal.weizmann.ac.il/fldbin/findex ) is a dynamic and interactive process that estimates the local and general probabilities that any query sequence will fold under specified conditions (Prilusky et al., 2005; also considered further in Esnouf et al., 2006). (ii) SeqAlert (https://bioportal.weizmann.ac.il/salertb/main ) is a sequence-alerting service that will periodically compare submitted sequences against sequences for known structures deposited with the PDB (Berman et al., 2000) and targets recorded in TargetDB (Chen et al., 2004). It also reports the PubMed IDs of papers on proteins published over the previous 60 d that are related to the target sequence. One useful feature of SeqAlert is that it allows both bulk and e-mail-based requests. (iii) SeqFacts (https://bip.weizmann.ac.il/sqfbin/seqfacts ) is a tool for sequence identification, analysis, characterization and annotation.

A description of the interface between these tools and a wet laboratory LIMS is given in §2.4. As with other partners, there is the possibility of performing primer design as part of the LIMS, in this case using the bestPrimer server.

2.4. Wet-laboratory LIMS

High-throughput structural projects have the potential to create a vast amount of diverse data. The number and variety of the target proteins and macromolecular assemblies, the diversity of experimental processes and the abundance of new laboratory practices and protocols make it very hard to follow and analyze experiments without the use of informatics. The stages involved in the production of proteins for structural studies include selection and design of targets, PCR, cloning, recombinant expression (both small scale and scale-up, both incubation and fermentation and using a variety of expression protocols) to produce native and labelled samples, purification and analytical characterization/quality assurance (QA). Furthermore, with high-value targets it is common to follow a strategy of multiple constructs and expression protocols. This approach generates a cascade of experimental results (Fig. 3) for which automated tracking becomes effectively essential. Informatics solutions to laboratory data-management problems are collectively referred to as laboratory information-management systems (LIMS). Commercial LIMS can be expensive and are usually highly tuned to specific (and unchanging) industrial processes, typically analytical and QA processes. This specificity makes them particularly poorly adapted to the fluid processes that characterize academic research. Thus, there is a clear need for an easy-to-use LIMS covering the processes of structural proteomics that is well adapted to use in the rapidly changing world of academic research.

Figure 3 An indication of the scale of results generated by high-thoughput studies. The example considers the data generated from a single plate containing DNA for studying 96 constructs through PCR, the Gateway cloning protocol (Invitrogen) and small-scale expression screens using two cell lines with each of two experimental protocols. The top half of the figure shows the usage (and therefore requirement to record in a LIMS) of 96-well and 24-well plates, the bottom half shows the number of 96-lane agarose gels, 24-well colony-plate images and 26-lane SDS–PAGE gels that also need recording. In total, a set of 96 constructs uses 34 96-well plates and 36 24-well plates and generates 480 images of colony wells, 1536 lanes on agarose gels and 416 lanes on SDS–PAGE gels.

In the absence of useful academic software and without in-house experience, the Oxford partners decided in 2001, prior to the start of SPINE, to use as a first implementation an adaptation of a commercially available LIMS, Nautilus (Thermo Electron Corporation). This has now been developed as far as practical linking back to target selection and primer design and covering plate layout, PCR, cloning, small-scale expression trials and crystallization trials. The middle stages, expression scale-up and purification, have proved very difficult to integrate for technical and design-based reasons and these are currently covered by fixed-format worksheets. The software is in everyday use and automated routes exist for the reporting of progress to the OPPF webpages and the SPINE web site. Nautilus is a two/three-tier LIMS underpinned by the Oracle RDBMS. Interaction with the database is via `thick' clients (most calculations are performed locally on each client, with only database access to the server). The basic Nautilus schema is fixed and is built around a simple hierarchy of Sample-Group > Sample > Aliquot > Test > Result. Flexibility is introduced by allowing Aliquots to be split, pooled and linked recursively to other Aliquots in parent–child relationships. In the context of work in Oxford, any physically separate amount of any reagent is viewed as an Aliquot. Tests correspond to experiments, either characterizing an Aliquot or (as an Oxford-specific extension outside the original scope of Nautilus as an analytic LIMS) transforming one Aliquot into another, thereby allowing complex workflows to be modelled. Other useful features of Nautilus include a mechanism to represent experiments on plates of Aliquots, a patented workflow technology, a method for parsing information from instrument log files, a background processor for automatic execution of tasks, a way of incorporating custom-written code and a complex mechanism for maintaining database integrity and security. An example of the Nautilus user interface is shown in Fig. 4. When a construct design is authorized for laboratory work, its details are passed on to Nautilus. Through Nautilus, the construct designs are laid out in micro-titre plates, instructions to laboratory staff for the creation of PCR template plates are produced and correctly formatted orders for forward and reverse primer synthesis are generated. The PCR stage is fully integrated with the LIMS, with output from Mastercycler instruments (Eppendorf) being parsed automatically. The molecular-biology stages relevant to the Gateway system (BP and LR reactions; Invitrogen) as well as the steps involved in small-scale (E. coli) expression trials have also been modelled. Gels are a crucial part of the data accumulation in these stages and it is possible to incorporate data from scanning 96-well format gels into the database and to associate these scans directly with experiments (although this is not fully automated). As noted above, tracking of expression scale-up and purification has proved impractical to implement and development effort has now shifted to building this into a new framework (PIMS; see below).

Figure 4 An example of the Oxford interface to the Nautilus LIMS (Thermo Electron Corporation). The back window shows a tree-based view onto the LIMS database showing all the plates known to the system. The middle window shows (some of) the properties for a particular plate (following user-interface standards, this view is obtained by right-clicking over the plate name and then selecting `Properties') describing graphically the contents of each well in the plate. The front window displays (some of) the properties for an individual well (the window is again obtained with the expected mouse clicks), in this case the Oxford-developed tab displaying the sequence of the DNA sample in that particular well.

The Weizmann node has developed and implemented a LIMS to cover its current needs. The system consists of a set of tools for automatic annotation of large-scale data set covering all aspects of going from gene to three-dimensional protein structure. This LIMS is intended as a `laboratory notebook' and for tracking and evaluation of different methods. In close collaboration with Anne Poupon (University Paris-Sud, Orsay), the HalX LIMS has been extended (Prilusky et al., 2005). For example, HalX is now capable of querying and retrieving information from remote servers. The first implementation of this web services feature was for the design of primers and is carried out routinely at the Weizmann by the bestPrimers server. In addition, Xtrack (Harris & Jones, 2002) is used to track crystallization and data-reduction steps. A custom-developed Perl application programming interface (API) for the protein-production data model (Pajon et al., 2005) provides an open path to integrate with the wide selection of Perl-based bioinformatics tools and libraries that are available worldwide.

To develop LIMS further, three SPINE partners (Oxford, York and the EBI) have agreed to joint development of a free-to-use academic LIMS specially suited to protein production (the Protein Information Management System; PIMS; https://www.pims-lims.org ). This development is starting from the protein-production data model, itself generated with input from several SPINE partners. The data model is emerging as a standardized method for experimental information interchange in the structural biology field and it forms the basis of HalX developments. Also building on this model, and in collaboration with several projects in this area (especially the UK BBSRC-funded eHTPX project), a web service portal (at Daresbury Laboratory, UK) has been created to allow secure exchange of information between sites, including SPINE nodes. Following test exchanges, the first exchange of real data in September 2005 transferred crystal-shipping information between Oxford and the UK-funded beamline at the ESRF, BM14. This has been followed by similar transactions involving Oxford, York, one further academic site and one industrial user. The same information-interchange standards are being adopted at other European synchrotrons. Although agreements on standards can be difficult to achieve, especially for standards as complex as data models, progress is being made and real benefits are in sight. These advances have depended to a large extent on the impetus given to collaboration by projects such as SPINE.

2.5. Crystallization and data tools

The Oxford partners have integrated the crystallization process into their protein production LIMS, Nautilus. The LIMS is used to record the details of all proteins going into the crystallization facility and stores the layout of the standard and optimization crystallization screens. The liquid-handling robots used in the set-up of crystallization trials have been interfaced to Nautilus via custom-written control software. Custom-written graphical user interfaces and the use of barcode scanners keep manual data entry to a minimum. Once the crystallization trial has been set up, the various data-management aspects of imaging and crystal identification are managed through the PlateDB and Vault software (Mayo et al., 2005; see also Berry et al., 2006). The system automatically detects and processes new images of the crystallization drops, recording information in a PostgreSQL database. All images acquired in Oxford are automatically evaluated on a local small Linux-based cluster using the ALICE software developed by the York partner (Wilson, 2002, 2004; Berry et al., 2006). A web-based image-viewing facility allows users to monitor and annotate their experiments remotely. This system is freely available to academic laboratories and is being incorporated into PIMS to encourage uptake. Access to a demonstration account on the system is available at https://www.oppf.ox.ac.uk/vault/demo.php . Currently, 210 users have taken 31.6 million images from 13 500 plates created for 857 projects. The value of this annotated database is exemplified by the design of a 96-condition screen optimized for SPINE viral and human protein targets (see Berry et al., 2006).

2.6. Structure determination

Software developments within SPINE have addressed structure determination by both NMR and crystallography. The NMR developments are detailed in AB et al. (2006) and we here concentrate on those for crystallography. SPINE partners, in collaboration with the BioXHIT project, have worked on the automated model-building package ARP/wARP and by utilizing structural bioinformatics to feed data into newly developed pattern-recognition software have extended the useful resolution range of this software (Morris et al., 2004). Thus, recent versions of ARP/wARP can deal with lower resolution data and produce more complete models. These improvements were driven by a careful analysis of a limited number of experimental electron-density maps to improve the `prior knowledge' of protein structure (Morris et al., 2004; Cohen et al., 2004). A more comprehensive statistical analysis is in progress using a larger number of data sets, collected in a database developed as a SPINE workpackage 7 initiative. This database captures, along with the final PDB model of a structure, additional crucial information which is not available from the PDB; namely, the experimental phase set derived from MIR/MAD/SAD experiments or molecular replacement. This database can be accessed at https://xtal.nki.nl/Depot and an interface for data searching and retrieval will be made available to other software method developers. All deposited data are stored in a relational database and submitted for curation and validation before being made available for statistical analysis. During the curation step, the data are standardized so that each data set can be run in the different analysis stages as simply as possible. The development of standard procedures for this task relates to the use of automated structure-solution pipelines developed within SPINE (see Bahar et al., 2006). A long-term objective is to capture all SPINE-related structure solutions in this database.

2.7. Structure annotation

The ProFunc annotation server (Laskowski, Watson et al., 2005; https://www.ebi.ac.uk/thornton-srv/databases/ProFunc/ ) has been developed to help identify the likely biochemical function of a protein from its three-dimensional structure. ProFunc makes use of both existing and novel methods to analyse protein sequence and structure, identifying functional motifs or close relationships to functionally characterized proteins including, for example, a full analysis against the latest release of the Superfamily database. Often, where one method fails to provide any functional insight, another may be more helpful. A single-page summary of the analyses provides an at-a-glance view of what each of the different methods has found, whereas more detailed results are available on separate pages. The system has been made available to SPINE members and a large amount of testing has been performed, leading to the inclusion of new algorithms for finding structural motifs and similarities, better default parameters for many of the individual tasks and a clearer graphical presentation of the results. The system has been extended to ligand annotation since ligand prediction will play a crucial role in elucidating function from structure. Additional annotation is available through the new PDBsum service, https://www.ebi.ac.uk/thornton-srv/databases/pdbsum , where structure files can be uploaded and pages are produced providing at-a-glance views of the structure plus detailed structural analyses of each protein chain, any bound ligands and metals (Laskowski, Chistyakov et al., 2005).

A second approach to docking aimed at high-throughput virtual screening for protein-function prediction has been developed at the EBI (Morris, 2004; Morris, Najmanovich et al., 2005). This method aims at narrowing down ligand space by excluding those ligands whose shape does not match the predicted binding pocket. A new protein structure is analyzed with the program SURFNET (Laskowski, 1995) which picks up potential binding pockets by identifying surface clefts and indentations. Residue-conservation scores are computed with a maximum-likelihood phylogenetic approach and employed to reduce the size and enhance the quality of predicted binding pockets. Expansion is performed to describe the resulting three-dimensional shapes of the predicted binding pockets. Heuristics have been developed for three-dimensional object registration based on the first three moments of the Cartesian atomic coordinates, thus enabling the expansion coefficients to be used directly for shape comparison. This results in an extremely fast comparison metric based on the Euclidean distance in coefficient space.