Poster Abstracts

*Los Alamos National Laboratory, Los Alamos, NM 87545

Structural GenomiX,

10505 Roselle Street, San Diego CA 92121, USA

Tel: 00 1 858 558 4850, Fax: 001 858 558 4859

Email: tom@stromix.com

Structural GenomiX (SGX), a San Diego-based start-up company, was founded to capitalize on the value of protein structure information in drug and compound discovery. The company is developing a high-throughput platform to support the determination of hundreds of novel protein structures via x-ray crystallography. This platform integrates advances in genomics, x-ray crystallography and bioinformatics. The company’s approach to structure determination is genomic: families of target genes from a range of organisms are input into the platform, raising the odds of success and generating valuable information about family relationships. The company utilizes leading x-ray crystallography techniques to ensure fast structure solution, including the routine incorporation of selenium into proteins, the use of a third-generation synchrotron (the Advanced Photon Source in Illinois), and the use of MAD phasing. Bioinformatics tools enable target selection based on real-time information, and annotation that adds value for customers engaged in drug and compound discovery.

SGX plans to make its protein structures available to pharmaceutical, biotechnology, agricultural and industrial customers through subscriptions to an annotated database and in strategic alliances. The speed and quantity of structures generated by the SGX process will enable customers to access protein structure earlier in the compound discovery process, changing target selection and bringing structure-based drug design into more universal applications.

The SGX process of target selection to structure determination is presented along with an overview of the current process as compared to what is being constructed for higher throughput in the near future. Recent progress towards scale up and automation is presented along with results in terms of structures completed to date.

Biotechnology Research Institute, National Research Council of Canada, 6100 Royalmount Ave., Montréal, Quebéc, Canada H4P 2R2 and

Montréal Joint Centre for Structural Biology, Montréal, Quebéc, Canada

Several structural genomics pilot projects are now underway world-wide, with the eventual promise of high-throughput protein structure determination. An essential task within such projects is the development of effective methods to express, purify and crystallize large numbers of proteins efficiently. We have initiated a pilot scale project based on gene targets selected from the genome of E. coli. Thirty-six genes have been selected for cloning initially and most of these were successfully over-expressed as soluble proteins. Target genes have been cloned as N-terminal fusions with either glutathione-S-transferase or (His)₆ tags. Initial affinity purification is achieved using glutathione Sepharose, or Ni-NTA resins, followed by thrombin cleavage to remove the fusion tag. Further purification using conventional FPLC ion exchange or gel filtration chromatography is then performed. Using this approach, over 20 proteins have so far been purified to apparent homogeneity as assessed by SDS-PAGE. Purified proteins are further characterized for homogeneity and suitable solution properties using a combination of dynamic light scattering, electrophoretic methods, mass spectrometry and limited proteolysis. Purified protein samples are screened for initial crystallization conditions using a sparse-matrix approach. To follow the progress of various genes and to store all relevant experimental data required development of specialized database to store these data. Web-based software for displaying and searching the information from this local database and for cross-referencing with other genomics databases has been developed.

To date, some crystals have been obtained for more than half of the purified proteins. Of these, diffraction quality crystals were obtained for five proteins. Using SeMet substituted proteins we have at this time determined the structures of three of these proteins. Another group has reported the structure of the fourth protein in the meantime. We will present the statistics related to various steps of the process and summarize the current results.

Molecular Simulations, Inc.,

9685 Scranton Road, San Diego, CA 92121, USA

Email: lly@msi.com

With the vast amount of protein sequences determined from the genomic sequencing project, there is an emerging need to use high throughput method to predict structures and assign functions of the protein sequences. GeneAtlas is an automated, high throughput pipeline for the prediction of protein structure and function using sequence similarity detection, homology modeling, and fold recognition methods. It uses PSI-BLAST and SeqFold to search for homologous structures from PDB database and MODELER to build 3D models for the sequences based on the template structure. The quality of the 3D-model is validated using Profile-3D/verify score. The accepted model gives the correct fold and indicates the possible function of the genome sequence from the known template. Furthermore, protein 3D structure contains much richer information for its function than sequence along. Functional annotations based on the 3D-model using a suite of methods give further details of the protein function which are crucial for target discovery, protein engineering, and inhibitor design. Using a "virtual" genome, a subset of PDB structures from SCOP database which consists of protein structures of less than 40% sequence identity, as a benchmark, we demonstrate that GeneAtlas detects additional functional relationships by building 3D-models for genomic sequences in comparison with the widely used sequence searching method, PSI-BLAST. The method was applied to 22 publicly available genomes, including C. elegans, D. melanogaster, S. cerevisiae, H. sapiens, A. thaliana, etc. The modeling results of a small genome, M. genitalium, and the comparison with PSI-BLAST and Hidden Markov Model (HMMer/pfam) on function assignment will be discussed.

Torward the crystal structure of a lectin-like natural killer cell activator receptor bound to its MHC class I ligand.

Molecular Structural Biology Group. Medical Biochemistry and Biophysics.

Karolinska Institute. Tomtevodavägen 6 171 77 Stockholm. Sweden

Natural killer (NK) cell function is regulated by NK receptors that interact with MHC class I molecules on target cells. The murine NK receptor Ly-49D activates NK cells activity by interacting with H-2Dd through its C-type-lectin-like NK receptor domain. The activating Ly-49D receptor and the inhibitory Ly-49A receptor mediate opposing effects on NK cell cytotoxicity. The €missing self€ hypothesis predicts that all NK cells express at least one inhibitory receptor for a self MHC I antigen, allowing NK cells to delete cells with lost or altered self MHC expression. Thus, the very existance of NK activating receptors specific for MHC class I remains somewhat perplexing, and the exact physiological role of these receptors has not been clarified. We are trying to co-crystallizate activating receptor and ligand to elucidate those questions by providing basis for an analysis of the interaction in the activating function. We have already overexpressed and purified MHC class I (H-2Dd) and we are trying to obtain recombinant Ly-49D in a soluble form using a novel overexpression approach. In prokaryotes, the recently characterized TAT pathway can transport folded substrates across the inner membrane, and signal peptides specific for this pathway bears a twin arginine motif. I have demostrated that a folded large cytoplasmic domain of a non-TAT protein can be translocated by this machinery by fusion to a TAT signal sequence (Cristóbal et als, 1999) and we are using this strategy to overexpress and improve the solubility of the Ly-49D receptor.

Prediction of the 3D structure for proteins with tandem repeat sequences.

Center for Molecular Modeling

CIT, National Institutes of Health,

Bldg 12A, 12 South Drive MSC 5626, Bethesda, MD 20892, USA

The genome sequencing projects have revealed a considerable number of protein sequences with tandem arrays of 10-30 residue long repeats. Despite the established functional importance of many such proteins, only a few of their 3D structures are known. The lack of the structural information is explained by the fact that large molecular weight and elongated shape of these molecules hamper X-ray and NMR studies. On the other hand, inspection of the known structures suggests that structural prediction of such proteins can be more reliable than prediction of aperiodic globular proteins. The prediction can be facilitated by assuming repetitive spatial arrangements within the tandem repeat sequence and by more reliable distinguishing of structurally important residue positions in the repeats. Prediction and modeling of several proteins with 10 to 30-residue long repeats, such as leucine-rich repeat proteins, human involucrin, and bacterial surface-associated adhesins are described. The modeled structures incorporate constraints from electron-microscopy, circular dichroism and other indirect structural experiments. The approach used for prediction and modeling of these proteins can be applied to other proteins containing internal repeats and will lead to a valuable tool of structural bioinformatics.

[1] Laboratoire de RMN (CNRS URA 2185)

[2] Unite de Biochimie Cellulaire, Institut Pasteur,

28 rue du Dr. Roux, 75724 Paris Cedex 15, France.

Tyrosyl-tRNA synthetase (TyrRS) is a homodimeric protein that catalyzes both the activation of the amino acid through its reaction with ATP and the transfer of the aminoacyl-adenylate to the tRNA(tyr). In Bacillus stearothermophilus, each subunit of TyrRS comprises two structural domains, an N-terminal domain (residues 1-319), whose crystal structure is known, and a C-terminal domain (residues 320-419) which appears disorded in the crystals. The binding site of one tRNA-Tyr molecule encompasses both subunits of the TyrRS dimer. The folding state of the C-terminal fragment was characterized in solution by biophysical techniques and compared with those of full-length TyrRS. We present here the 3-dimensional structure of a recombinant protein TyrRS(de4) corresponding to the C-terminal domain of TyrRS solved by heteronuclear NMR spectroscopy. We suggest that the disorder observed in the crystal structure is due to a flexible linker between the N- and C-terminal regions. The TyrRS(de4) structure exhibits a novel fold among the anticodon binding domains of aminoacyl-tRNA synthetases, around two thirds of which appear to be shared with the ribosomal protein S4 and a heat shock protein HSP15. The common topology involves two _ helices packed against an antiparallel _ sheet. Of six basic residues identified by site directed mutagenesis as essential for tRNA binding, four are clustered in this domain and are likely to interact with the anticodon arm of tRNA.

Department of Mathematics, Rutgers University,

110 Frelinghuysen Rd. Piscataway, NJ, USA

Phone: (732) 445 34 78; Fax: (732) 445 55 30

Nicola D. Gold(1) and David R. Westhead(1).

1. School of Biochemistry and Molecular Biology, and 2. School of Computer Studies, University of Leeds, Leeds, W. Yorks. LS2 9JT.

e-mail: westhead@bmb.leeds.ac.uk

The relationship between biochemical function and protein structure is not straightforward. Sometimes function can be deduced directly from sequence and/or fold, but there are many examples of proteins with similar sequences and different functions, and examples of protein folds that support many unrelated functions. Even when proteins have related functions, small differences, for instance in enzyme specificity, are often only apparent when detailed structural information is available. Structural genomics initiatives are expected to generate structures for large numbers of proteins, the majority of which will be uncharacterised from a functional point of view. The aim of these initiatives is to provide structural information as a route to the determination of gene function. This will require the development of new bioinformatic tools able to predict functional characteristics from protein structure.

Many important biochemical processes, for example molecular interaction and recognition or catalysis, occur on or near protein surfaces. Surfaces with similar shape, chemical and physical properties are likely to perform similar functions. If a protein has an unknown function, a route to function prediction is therefore to locate similar surfaces in other proteins of better characterised function, and transfer the information. This process operates in direct analogy to the practice of predicting function from sequence by seeking to find related sequences of known function, for instance in a BLAST search.

Surface matching is a difficult computational problem, much studied in the field of computer vision. We report early progress in adapting methods from this field to the problem of protein surface matching. A multi-resolution approach views the surface in terms of differential properties, the curvedness and shape index, describing the degree of curvature and nature of the local surface shape (convex, concave, saddle) respectively. The surface matching algorithm uses a tree-structured search space where algorithmic efficiency is achieved through early pruning of branches corresponding to matches judged to be unreasonable by the above criteria. Preliminary results indicate that the method will be both efficient and useful. Future work will adapt the algorithms to include physico-chemical properties of the surface, with the aims of producing more reasonable matches from a biochemical point of view, and further increases in efficiency by increasing pruning of the search tree.

A database of surfaces for comparison will be generated and the algorithms will be used for similarity searches of this database. Other search algorithms will be implemented, including searches for user defined spatial arrangements of key functional residues. Ultimately these services will be made available on the World Wide Web, and for analysis of structural genomics data.

University of Oxford, Department of Biochemistry,

South Parks Road, OX1 3QU, UK

Arrays of antisense oligonucleotides corresponding to the first 120 nucleotides each of the cyclins B1, B4 and B5 were fabricated on the surface of aminated polypropylene. The arrays were hybridised with the appropriate radiolabelled transcript to assess the ability of the immobilised oligonucleotides to form heteroduplexes with their targets. Oligonucleotides that produced strong heteroduplex yield, as well as those that showed little annealing, were assayed for their effect on translation of endogenous cyclin mRNAs in Xenopus egg extracts and their ability to promote cleavage of cyclin mRNAs in oocytes by RNase H. Excellent correlation was found between the antisense potency and the affinities of the oligonucleotides for cyclin transcripts as measured by the arrays, despite the complexity of the cellular environment.

Schools of Chemistry and Biological Sciences, University of Exeter, Stocker Rd., EX4 4QD, UK

The hyperthermophilic archaeon Sulfolobus solfataricus lives at temperatures of up to 87ºC whilst Pyrococcus woesei survives temperatures in excess of 100ºC. The methods utilised in stabilising intracellular proteins have been studied using homology modelling techniques using phosphoglycerate kinase (PGK) as an example. The glycolytic enzyme PGK is a well studied enzyme with over 100 primary sequences currently available. The enzyme is normally a monomer, however S. solfataricus PGK is tetrameric whilst P. woesei PGK is dimeric. Homology models of the S. solfataricus, P. woesei, Methanococcus bryantii and Haloarcula vallismortis PGKs have been generated and compared to the existing X-ray structures of the enzyme from Bacillus stearothermophilus, Thermotoga maritima and Saccharomyces cerevisiae. These models have provided insights into the mechanisms of stabilisation employed by hyperthermophiles and the substrate and cofactor binding sites of archaeal PGKs. The modelled S. solfataricus PGK structure also reveals potential areas for hydrophobic subunit interaction. Examination of the archaeal PGK protein sequences has confirmed that the essential catalytic residues have been conserved. A possible gene duplication event evident in the S. solfataricus PGK has also been observed.

EMBL, Meyerhofstrasse 1, 69012 Heidelberg, Germany

We provide statistically reliable sequence evidence indicating that at least 12 of 23 scop (beta/alpha)8 (TIM) barrel superfamilies share a common origin. This includes all but one of the known and predicted TIM barrels found in central metabolism. The statistical evidence is complemented by an examination of the details of molecular function and protein structure, with certain structural locations favouring catalytic residues even though the nature of their molecular function may change. The combined analysis of sequence, structure and function also enables us to propose a phylogeny of TIM barrels. Based on these data, we are able to examine differing theories of pathway and enzyme evolution, by mapping known TIM barrel folds to the pathways of central metabolism. The results favour widespread recruitment of enzymes between pathways, rather than a 'backwards evolution' model, and support the idea that modern proteins may have arisen from common ancestors that bound key metabolites.

Stockholm Bioinformatics Center, Stockholm University, 106 91 Stockholm

Here we report results from two recent studies of different fold recognition methods. In the first study we have performed the first large (10000 pairs) test of alignment quality using several different alignment methods (local, global, profile alignment, hmmer, sam.t98, clustalW, sspsi) (Elofsson, 2000 submitted). We show that both evolutionary information and predicted secondary structure information improves the alignment quality. The best alignments are obtained from a method that combines a sequence profile obtained from psiblast with predicted secondary structures. In the second study. we present a novel, continuous approach aimed at the large-scale assessment of the performance of available fold-recognition servers (Bujnicki et al 2000 submitted). Six popular servers were investigated: PDB-Blast, FFAS, T98-lib, GenTHREADER, 3D-PSSM and INBGU. The assessment was carried out using as prediction targets a large number of selected protein structures released during October 1999 to April 2000. Overall, the servers were able to produce structurally similar models for one-half of the targets, but significantly accurate sequence-structure alignments were produced for only one-third of the targets. We further classified the targets into two sets: "easy" and "hard". We found that all servers were able to find the correct answer for the vast majority of the easy targets when a structurally similar fold was present in the server's fold libraries. However, among the hard targets - where standard methods such as PSI-BLAST fail - we found that the most sensitive fold-recognition servers were able to produce similar models for only 40% of the cases, half of which having a significantly accurate sequence-structure alignment. Unfortunately, the increased sensitivity of the fold-recognition servers over standard methods came with the cost of low specificity.

Probably the most interesting observation from these studies is that there is not a single method that always produce the best results (fold recognition or alignment). For instance we show that almost twice as many good models can be created using any method compared with the best method for fold related pairs and that each server had a number of cases with a correct assignment, where the assignments of all the other servers were wrong. This emphasizes the benefits of considering more than one method in difficult prediction tasks. And it also implies that it would be possible to improve fold recognition performance significantly if a combination of several methods could be done without loosing specificity.

In conclusion, we would like to encourage all protein structure predictors to take advantage of the variety of methods available. In both these studies we have used novel methods to measure the quality of a model generated from a fold recognition method. We will also discuss the advantages using these novel methods for measuring fold recognition capacity (Siew et al 2000, in press; Cristobal et al 2000, manuscript in preparation).

beta-Glucosyltransferase (BGT) is a DNA-modifying enzyme encoded by bacteriophage T4 which catalyses the transfer of glucose from uridine diphosphoglucose (UDPG) to 5-hydroxymethylcytosine (HMC) in double-stranded DNA. The glucosylation of T-4 phage DNA is part of a phage DNA protection system aimed at host nucleases. We previously reported the complete BGT co-crystal structure in the presence of UDPG¹ where the glucose is missing due to BGT cleavage. This BGT structure has provided us with a basis for detailed modelling of DNA bound to BGT. Furthermore, using the structural similarity between the catalytic core of glycogen phosphorylase and BGT, we have been able to model the position of the missing glucose moiety from UDPG.

We now report two BGT-UDP-Mg2+ structures from crystals grown in the same conditions except the concentration of magnesium ions. Crystal of BGT-UDP-Mg2+ at 20mM diffracts at 2.5Å resolution while crystal of BGT-UDP-Mg2+ at 40mM diffracts at 2Å resolution. Both crystals belong to P212121 space group but cell parameters are different. Both structures contain one magnesium ion in the UDPG binding site. The presence of a second Mg2+ ion far from the active site in the structure with 40mM Mg2+ could explain the difference of crystal packing between these two structures. Here, we present the metal site of BGT and from these two models, we propose a role of Glu163 in the catalytic mechanism of BGT.

1. Moréra, S., Imberty, A., Aschke-Sonnenborn, U., Rüger, W. and Freemont, P. (1999). "T4 phage beta-glucosyltransferase: substrate binding and proposed catalytic mechanism." J. Mol. Biol. 292, 717-730

Patrick Aloy ^§‡, Enrique Querol^§, F. Xavier Avilés^§ & Michael J.E. Sternberg ^‡

§ Institut de Biologia Fonamental. Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain.

‡ Biomolecular Modelling Laboratory. Imperial Cancer Research Fund, Lincoln´s Inn Fields, London WC2A 3PX, UK.

E-mail: patrick@luz.uab.es

An ultimate goal of genome analysis is to determine the biological function of all the gene products in a genome. Typically, prediction of function is based on sequence similarity with proteins of known function, but even in a simple organism with a small genome, like Mycoplasma Genitalium, more than 40% of the ORFs have no homologous sequences in the databases. Moreover, the new initiatives in Structural Genomics will lead to the determination of the three-dimensional structure of proteins prior to knowledge of their function. All this has created the need of new methodologies for the analysis and prediction of protein function from its sequence and structure analysis. Here, we present a method to predict the location of functionally important sites in proteins.

A non-redundant database (<25% homology) of 107 protein chains was build using all the entries with detailed information of the functional residues in the Brookhaven protein databank ( ~ 1800). An exhaustive analysis of residue type, secondary structure and solvent accessibility preferences of the active site residue was performed. The next step was to develop a fully automated method following the ideas given by the Evolutionary Trace method (Lichtarge et al., 1996). The standard database search method WU-BLAST (Altschul et al., 1990) was used to pick up all the proteins that matched our probe sequence and a multiple alignment was then carried out with the program CLUSTALW (Thompson et al., 1994). A hierarchical clustering algorithm was implemented to build up a phylogenetic tree and the consensus sequence was extracted for each branch (subfamily). If our probe protein, or at least one of the homologous, has known structure, the consensus sequence was mapped onto the protein structure and techniques of double spatial clustering were applied in order to identify those residues that are not only conserved but close in the space as well. A sphere containing all the residues that have clustered together was then build and defined as Functional Site.

The results show that it is possible to predict Functional Sites automatically in a given protein, with 60% accuracy and 99 % significance, when we can find enough homologous sequences in databases and, at least, one of them has known structure. The method also gives us a clue to identify cases of divergent evolution with still high homology levels. The information obtained from the functional sites was used to filter filter putative protein-protein and protein-DNA docking complexes generated by FTDock (Gabb et al., 1997) giving similar results than when experimental information from mutagenesis experiments is used.

Protein Stucture Function Group, School of Biosciences, University of Birmingham

A flavoprotein from E coli, with potential use for gene-directed enzyme prodrug therapy (GDEPT) has been purified and crystallized, with the eventual aim of protein engineering. X-ray diffraction data for three different crystal forms were collected and the structures solved by MAD, SAD and MR to a maximum resolution of 1.6 Angstroms. The overall fold of the protein and environment of the FMN cofactor with substrate analogue are presented.

Laboratory of Molecular Biophysics, Rex Richards Building

University of Oxford, South parks Road, Oxford OX1 3QU, UK

Glutamate receptors which are permeable to Na⁺, K⁺ and Ca2⁺ and potassium channels, selectively permeable to K⁺ are two important families of ion channels. The crystal structure of KcsA, a potassium channel, revealed the structure of the protein in the transmembrane region (Doyle et al, 1999). The glutamate receptor, Glur0, from Synechocystis PCC 6803 binds glutamate and forms potassium-selective channel (Chen et al, 1999). The transmembrane region of Glur0 was found to have a high sequence similarity to the KcsA protein from Streptomyces lividans, particularly in the selectivity filter region (TVGYG) and in the putative gating region. Hence, the KcsA structure was used as a template to develop a homology model for Glur0. The alignment was done such that the overlap is maximal between the selectivity filter region and the pore-lining residues of KcsA and Glur0. This model was then inserted in a fully solvated palmitoyl oleyl phosphatidylcholine (POPC) lipid bilayer. Three K⁺ ions were placed in the model protein, two in the selectivity filter and one in the cavity and a molecular dynamics simulation trajectory was generated for 1ns. The model is seen to be stable over the period of 1ns, in a membrane environment with no obvious collapse in any region. The ions were seen to enter into the intracellular side from the channel pore. The interaction of the ions in the selectivity filter and the cavity with the protein and water molecules, are compared to those of similar interactions in KcsA (Shrivastava and Sansom, 2000).

Doyle et al., (1998), Science 280:69-77

Chen et al. (1999) Nature, 402:817-821

Shrivastava & Sansom (2000), Biophys. J. 78:557-570

Centre for Molecular an Biomolecular Informatics

University of Nijmegen, The Netherlands

Many proteins have calcium ions as part of their structure. The number of oxygen atoms involved in binding this ion (coordination number) can vary. The most frequently found coordination numbers are 6 and 7. With different coordination numbers, the calcium binding site adapts different geometrical conformations. In case of a 7-coordination site, the geometry is a pentagonal bipyramid, while a 6-coordination site adapts a octahedral geometry. The bidentate ligand (capable of providing two oxygen atoms for calcium binding, only glutamic acid an aspartic acid are capable of doing this) is essential for the pentagonal bipyramidal structure. Detailed statistical analysis and superposition of sites that adapt the same geometry provide valuable information for homology modelling and structure validation.

Helix Research Institute, Yana 1532-3, Kisarazu, Chiba 292-0812, Japan

Tel +81-438-52-3951 Fax +81-438-52-3952

* Correspondence: cris@hri.co.jp

Protein structure determination has revealed an unexpected conservation and divergence of function both within and between families [Thornton et al, 1999], but shows the feasibility of function prediction of novel genes based on the tertiary structure of the coded proteins. We used THREADER [Jones et al, 1992] to analyze novel genes that had no sequence similarity to proteins with known functions. We focused on cytokine/growth factors because clones screened from a full-length cDNA library possessing signal-peptides were available and studies of cytokines have revealed they can be grouped into different structural families despite lack of sequence similarity [Rozwarski et al, 1994 ].

The THREADER program lists many relevant Z scores, but in the final selection we used the score of the weighted sum of pairwise and solvation energies from the structure search (Z-13), and the score for shuffling of the sequence on the candidate using pairwise and solvation energies (Z-7).

Among the full-length cDNAs analyzed by THREADER the human lung type-I cell membrane-associated protein hT1a-2 (160 aa) [Ma et al, 1998] had a very significant Z-7 score to the human growth hormone (PDB-ID: 1huw, 191 aa) and very significant Z-13 and Z-7 scores for proteinase A (PDB-ID: 2sga, 181 aa).

The first step of the experimental strategy consisted in sub cloning of the hT1a-2 cDNA gene in a pCDNA3.1(-) Myc/His expression vector. The 30 kDa hT1a-2 protein expressed in mammalian cells was subsequently purified using a Ni column. Bioassay of the hT1a-2 activity as a growth hormone was tested using Nb2 rat lymphoma cell line. Nb2 can not grow in FBS-free medium, but exogenous application of hT1a-2 (700 ng/ml) induced a 60% increase in the growth rate, in comparison to FBS-treated Nb2 cell. Proteinase activity was tested using 5 peptide-4 methylcoumarin amide (MCA) substrates and release of 7-amino-MCA was determined fluorometrically. The purified hT1a-2 was observed to cleave the factor Xa- and trypsin-specific substrate Boc-Ile-Glu-Gly-Arg-MCA.

The above results show that hT1a-2 possesses activity both as a growth hormone and as a proteinase, as has also been described for the growth factor from Spirometra mansonoides [Phares and Kubik, 1996]. In order to determine its main function in vivo, we used the phage display method and isolated a single phage expressing a 24 residue oligopeptide that specifically binds to the hT1a-2 protein. This oligopeptide exibited a 43% sequence identity to the signal peptide cleavage region of the human insulin-like growth factor precursor, indicating that the basic function of hT1a-2 in vivo is probably as a proteinase that cleaves specific sequences.

We have used a bioinformatics approach to make a priority of which genes to analyze in the laboratory, and this is the first work confirming the function of hT1a-2 as estimated from annotation to the structures obtained from the threading approach.

Sophie Quevillon-Cheruel (1), Sylvie Auxillien (1), Michel Desmadril (1), Philippe Minard (1), Joël Janin (2)

1. Laboratoire de Modélisation et d’Ingénierie des Protéines – Orsay, FranceBernard LABEDAN - Robert AUFRERE - Gilles HENCKESInstitut de Génétique et Microbiologie – Orsay, France

2. Laboratoire d’Enzymologie et Biochimie Structurales - CNRS - Gif-sur-Yvette, France

Among the about 6200 ORFs of the yeast genome, we have selected 282 unique proteins and 202 families of paralog proteins with size ranging from 100 to 500 residues. These sequences were also selected in such way that they correspond to cytoplasmic proteins. In this first list of proteins, we have chosen 20 ORFs to develop an efficient and cheap method for their cloning, overexpression in E. coli, purification and crystallization. This method will be further applied to the other selected ORFs.

The 20 ORFs were amplified by PCR using S. cerevisiae genome as template. Various constructions have been tested, by adding a 6His Tag either in C-terminus or N-terminus of the protein. The PCR products were then cloned into either pET9 or pET29 vector.

The proteins were overexpressed in B834(DE3)pLysS methionine auxotrophe strain, after induction by IPTG and using methionine. The optimized expression conditions and the solubility of the ORFs were tested. Under the selected conditions of expression induction, no significant difference was observed for the two vectors used: 1/3 of the ORFs are well expressed and soluble, 1/3 are expressed but not soluble and 1/3 are not expressed. The localization of the His-Tag has effect, neither on the level of expression, nor on the solubility of tested proteins.

The soluble proteins were purified by affinity chromatography (NiNTA) and gel filtration. The purity and integrity of the proteins were tested by SDS-PAGE and mass spectrometry. The conditions of crystallization for these proteins are in the course of development.

Medical Biochemistryu & Biophysics, Karolinska Institute, 17177 Stockholm Sweden

Mammalian thioredoxin reductase (mTRR) is a member of pyridine nucleotide -disulfide oxidoreductase (PNDO) family which contains glutathione reductase (GR), lipoamide dehydrogenase (LAD), trypanothione reductase, mercuric ion reductase and some other enzymes. All of them are flavoproteins and contain active disulfide which is reduced by NAD(P)H and then transfers the reducing equivalents to a substrate.

Thioredoxin reductase catalyses the reduction of the thioredoxin, a widely expressed 12-kDa protein that participates in many processes in the cells. Reduced thioredoxin is a hydrogen donor for ribonucleotide reductase and some other enzymes, it also controls the thiol-disulfide redox balance as well as involves in the regulation of various transcription factors. It was shown that TRR is overexpressed in certain tumor cells and down-regulated in apoptotic cells. In addition, TRR participates in ascorbate recycling; it is inhibited by auranofin – the therapeutics of rheumatoid arthritis; and it is involved in the protection of the skin from UV radiation.

Surprisingly, mTRR is more similar to human GR than bacterial or plants TRR if size, sequence and position of active cysteines are compared. TRR has broad substrate specificity, in addition to the reduction of thioredoxin it catalizes the reduction of lipoic acid, protein disulfide isomerase, and many other substrates but not a glutathione. Unlike all PNDO, mammalian TRR contains an essential selenocysteine residue. SeCys is penultimate residue of a 16 amino acid long extension of mTRR which is absent in GR or LAD. Mutation of SeCys498à Cys greatly decreases the activity of TRR, however, it allows to overexpress the protein. The recombinant SeCys498à Cys mutant of rat TRR was crystallised as a complex with NADP+. 3D structure of mTRR was solved and refined to 3Å (R/Rfree= 23.0/29.5%). It is the first 3D structure of mTRR. Here, the comparison of 3D structure of mTRR with other members of nucleotide-disulfide oxidoreductase family is presented.

The 3D structure of human GR is the most similar to that of TRR: the alignment of 3D structure shows that these two proteins have 153 identical residues at the corresponding positions, rmsd is 1.4Å for 407 residues of one subunit, or 1.7Å for 810 Ca atoms of the dimer. TOP server found other similar proteins, all of them belong to PNDO family: trypanothione reductase (1aog.pdb) with rmsd 1.4Å for 410 residues (145 of them are identical), and dihydrolipoamide dehydrogenase (1ebd.pdb) with rmsd 1.5Å for 387 residues (108 residues are identical). The main difference is 16-residues extension at C-terminus with essential Cys497-SeCys498 pair. The extension is located at the place, occupied by GSSG in GR; the distance between essential residue His472 and CysB498 is about 6.5A. The detailed comparison of the active site structure of TRR with all other members of NDO-family is presented.

1. University of Manchester, 2.205 Stopford Building, Oxford Road, M13 9XX

2. Michael Barbert Centre for Mass Spectrometry, Department of Chemistry, UMIST, Sackvill Street, Manchester

3.Department of Biomolecular Sciences, UMIST

The identification of individual protein species within Saccharomyces cervisiae proteome has ben optimised by increasing the information produced form mass spectra analysis through the chemical dervitisation of tryptic peptides. Matrix assisted laser desoprtion ionisation time-of-flight mass spectrometry (MALDI-TOF-MS) is commonly employed to analysis samples obtained from trypcti digestion; such protelytic enzyme allows formation of digest fragments, well-suited to be protonated in the mass spectrometer. Recent studies have shown the strong dominance of signals belonging to arginine-containing peptides in the peptide mass fingerprinting spectrum. This behaviour is readily explicable considering the different proton affinities of the two C-terminal residues. Conversion of lysine residues into homoarginine containing peptides increases the number of peptides in the spectra, rendering more detectable the lysine terminal peptides; such an improvement of signal response provides additional data for searching the database. Novel bioinformatic tools have been developed so as to exploit the further information obtained with guanidination in conjunction with other chemical dervatisation.

(1) Keck Graduate Institute of Applied Life Sciences, 535 Watson Drive, Claremont, CA 91711, USA.

david_wild@kgi.edu, alpan_raval@kgi.edu

(2) Gatsby Computational Neuroscience Unit, University College London , Queen's Square, London, UK

zoubin@gatsby.ucl.ac.uk

We describe Bayesian network models for protein folds and superfamilies which incorporate both primary sequence and structural information, with applications in the identification of remote homologues during the selection of potential targets for structure determination and in the classification of newly determined structures from structural genomics projects.

The Bayesian network approach is a framework which combines graphical representation and probability theory, which includes, as a special case, hidden Markov models (HMMs). Hidden Markov models trained on amino acid sequence or secondary structure data alone have been shown to have potential for addressing the problem of protein fold and superfamily classification. This poster describes a novel implementation of a Bayesian network which simultaneously learns amino acid sequence, secondary structure and residue accessibility for proteins of known three-dimensional structure. An awareness of the errors inherent in predicted secondary structure may be incorporated into the model by means of a confusion matrix. Training and validation data have been derived for a number of protein superfamilies from the Structural Classification of Proteins (SCOP) database. Results using posterior probability classification indicate that the Bayesian network performs better in classifying proteins of known structural superfamily than a hidden Markov model trained on amino acid sequences alone. These results will be compared to classifications obtained using predicted secondary structure and residue accessibility information, and to a Fisher kernel (Support Vector Machine) method of scoring.

aDepartment of Medical Microbiology, St. George’s Hospital Medical School, Cranmer Terrace, London SW17 0RE, UK; ^bDivision of Biochemistry & Molecular Biology, School of Biological Sciences, University of Southampton, Bassett Crescent East, Southampton SO16 7PX, UK

1. Horwich, A.L., Weber-Ban, E.U. & Finley, D. (1999) Proc. Natl. Acad. Sci. USA 96, 11033-11040.

Computational and Structural Sciences

SmithKline Beecham Pharmaceuticals R&D

New Frontiers Science Park (North)

3rd Avenue, Harlow, Essex, CM19 5AW

The goal of structural genomics is to understand the structure and function of proteins on a genomic scale. It is clear that many, and possibly even the majority of proteins exist or are active in the cell in multi-protein complexes. Thus it is vital that approaches are developed to understand protein protein interactions on a structural level in order to fully understand protein function.

Two major approaches to structure determination are NMR and X-ray crystallography. NMR has developed rapidly over the last few years and is no longer an approach that is limited to proteins of 20 KDa. The use of isotopic labelling, improved instrumentation and novel pulse sequences initially increased this limit to approximately 30KDa. Around this size structure determination dependent on large amounts of nOe data runs into difficulty. Protein - protein interactions are frequently also difficult to study due to the dearth of nOes in the interaction sites. Now a range of distance and orientation dependent NMR parameters promise to extend both the size and resolution of structure that can be determined and decrease dramatically the time that initial folds for proteins can be produced. Also these NMR approaches will increase the ability of NMR to understanding protein recognition at a molecular level.

One important class of these new approaches uses NMR parameters that become available when a paramagnetic ion (either metal or nitroxide spin label) is attached to a protein. In cases where a native metal binding site is present metal ions with appropriate properties may be substituted for the native metal. So far this has been limited to either Fe (often via heme) or Ca binding sites. In theory it could also be possible to produce generic metal-binding sites using recombinant techniques. It may also be possible that these approaches are not limited to the very strong metal-binding sites present in current published studies.

In this work we present data to discuss the use of paramagnetic lanthanides in a range of proteins with different affinities for metal ions to obtain distance and orientation data and probe how this can be extended to understand protein/ protein interactions by selection of the appropriate metal ion.

Benjamin J. Stapley & Michael J. E. Sternberg.

Biomolecular Modelling Lab, Imperial Cancer Research Fund. 44 Lincoln's Inn Field, London WC2A 3PX, U.K.

Functional annotation of proteins is an ever more pressing requirement for successful exploitation of genomic information. Here, we use textual information from Medlars to aid in the assignment of sequences to pfam alignments. From a pfam seed alignment, we extract Medlars documents that are cited in the SwissProt entries of sequences in the alignment. 'Log Entropy' term weighting is applied and Pfam documents are generated by concatenation of the relevant Medlars documents. We then attempt to assign remote homologues - not included in the orginal alignments - to their respective pfam alignments but measuring cosine similarities of their cited Medlars documents to the Pfam documents. Successful assignment to one of a subset of 100 Pfam alignments is achieved with up to 40% accuracy (25% recall). In addition to aiding in the correct functional assignment of sequences, generated Pfam documents allow textual information retrieval of pfam domains with much higher recall. We have also applied the method to annotating Pfam alignments of indeterminant function.

Biological Structure and Function Section, Division of Biomedical Sciences, Imperial College School of Medicine, London SW7 2AZ, UK

High-throughput screening crystallisation trials are already under way in several laboratories world-wide, but optimisation, which is the more difficult part, has yet to be adapted to cope with the volume of experiments required by the Genome Projects.

The first multiple experiments for both screening and optimisation were done as microbatch trials under oil[1]. This procedure lends itself for adaptation to high-throughput crystallisation.

The utilisation of oil has established a unique way of producing crystals, making the experiments more efficient and saving time and materials. Oil affects the accuracy, cleanliness and reproducibility of crystallisation experiments as well as providing a reliable environment for controlling nucleation and growth [2].

We have designed the following optimisation methods, which have resulted in production of better-ordered crystals compared to those grown by conventional methods:

• Containerless crystallisation, in which a crystallisation drop is suspended between two oils of different densities resulting in reduction of heterogeneous nucleation [3]. High-throughput is achieved by replacing one of the oils with a gelled hydrophobic surface onto which the drops are automatically dispensed.
• A means to slow down the equilibration rate and approach supersaturation more slowly, in order to avoid crystal "showers", is by placing a layer of oil as a barrier over the reservoir of a hanging or sitting drop trial [3]. The advantage of this technique is that no change is required to the crystallisation conditions nor to the method used - it can be applied in Linbro, VDX, Cryschem or any other vessel, and it can also be automated.
• Inducing nucleation in microbatch by evaporation through a thin oil layer, and then arresting it by variation of the oil layer thickness over time.
• Automatic dispensing of gelled microbatch drops.
• Special filtration of samples.
• Producing better ordered crystals by de-coupling nucleation and growth can be achieved in either microbatch [4] or hanging drops [5]. In the case of hanging drops, the coverslips holding the drops are transferred after incubation for some time at conditions normally giving many small crystals, over reservoirs at concentrations that normally yield clear drops. In microbatch, the drops are diluted by automated means after incubation.

Protein Design Group, CNB-CSIC

Cantoblanco, Madrid E-28049, Spain

Tel:+34 91 585 48 39 Fax:+34 91 585 45 06

Web: http://montblanc.cnb.uam.es/

The widening gap between sequences and functions has lead to the practice of assigning a potential function to an uncharacterised protein based on sequence similarity with other proteins of experimentaly investigated function. Even if the reliability of those homology based functional assignments is not well characterized, it represent common practises in whole genomes functional assignments. We propose here a systematic approach to the study of the margins of error in homology based functional prediction by analysing the conservation of the functional annotations in a large set of structural alignments. In particular, we analyze five aspects of protein function, commonly used in genome annotation, namely: i) PDB header line, ii) Enzymatic function classification: DE code, the standard definition of the chemical nature of the enzymatic function; iii) Functional annotations in the form of keywords, describing the biochemical function such as the interactions with compounds, cofactors, substrates, regulators and other cellular components; iv) Classes of cellular function, capturing the main types of cellular activities in which proteins participate, e.g. "carbon compound metabolism" or "DNA biosynthesis"; and v) Conservation of the type of amino acid in the binding site, related with the binding activity of the protein, and in many cases, the specificity of binding different substrates and cofactors. The screening of the full range of sequence functional similarities allows us to present an initial picture of the relation between sequence and functional similarity, and in particular, to derive a theoretical error rate for homology-based functionnal assignments (1). With those data, we estimate the theoretical error rates of predicted functions in different genomes. Indeed, it is particularly interesting to think in the consequences of this study for whole genome annotations carried out by automatic systems (2) and to compare the expected level of error with the different values published by different groups of expert annotators (3, 4, 5).

1.Devos and Valencia. PROTEINS in the press

2.Andrade et al.. Bioinformatics 1999; 15:391-412

3.Brenner. Trends Genet. 1999; 15: 132-133

4.Galperin and Koonin. In Silico Biol. 1998; 1:0007

5.Ouzounis et al.. Mol. Microbiol. 1996; 20: 985-900

(1) EMBL Outstation, European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge, United Kingdom.

(2) Biomolecular Structure and Modelling Unit, Biochemistry and Molecular Biology Department, University College London and Crystallography Department, Birkbeck College, London, United Kingdom.

The physiologically relevant macromolecular assembly of multimeric proteins can often not be derived without ambiguity from crystallographic studies of protein structure. An automatic tool is being developed to differentiate physiologically relevant intermolecular contacts from contacts that are artifacts of the crystalline state. We compare the performance of simple structural features in identifying the macromolecular assembly. The comparison is based on a non-redundant set of protein structures with known multimeric states prevalent in solution. Non-parametric statistical methods are used for error assessment.

Pazos F., Blaschke C., Oliveros J.C., and Valencia A.

Protein Design Group. CNB-CSIC. Cantoblanco, Madrid 28049. Spain

Tfn. +34-1-585 45 70, Fax. +34-1-585 45 06, email: valencia@cnb.uam.es

The increasing knowledge about individual protein components (genome sequences, structural genomic initiatives, high throughput functional genomics) is making clear the need of integrating information in superior orders of complexity. Protein-protein interaction is the obvious next step in this direction. We present here three complementary computational efforts for the study of protein-protein interactions.

The first approach is based on the study of the patterns of variation in multiple sequence alignments. The rational behind these approaches is that proteins that have evolved to form specific molecular complexes would have accumulated during evolution compensatory substitutions that can be detect in current protein families. We have previously demonstrated that the analysis of the patterns of variation is sufficient to single out the right inter-domain docking solution amongst many wrong alternatives in two-domain proteins (1) and tested the predictions of interacting regions in different experimental systems (2-3). The extension of this method to the detection of interacting partners in large collections of multiple-sequence alignments shows quite promising results in terms of coverage, related with the number of interactions predicted in complete genomes, and accuracy, defined as the quality of the predicted interactions when compared with known molecular complexes (4).

The second approach is based on the application of text retrieval techniques (5-6) to the extraction of information about protein interactions directly from the scientific literature (Medline abstracts). Our current system is able to detect automatically networks of functional interactions, by identifying protein names and the actions liking them (7). We will discuss the results of the application of the system to different complex biological systems.

Finally, the application of clustering techniques (8) and text retrieval methods (9) to the available expression array results leads to a new avenue for the discovery of relations between genes, that can be considered as complementary information to the predicted and detected protein interactions, and represent promising new technologies to be combined with other experimental approaches like yeast two hybrid systems.

9.- Blaschke et al., (2000) submitted

Bristol-Myers Squibb Pharmaceutical Research Insititute

Princeton, NJ 08543, USA

The integration of genomics information with drug discovery is expected to identify, in the next few years, thousands of novel protein targets. This presentation will describe how combining structural genomics methodologies and structure-based drug design can be used to prioritize drug discovery projects. The application of the following methods to disease targets allows for the rapid generation of potential lead compounds.

• Database Mining
• Protein Fold Recognition
• Protein Function Identification
• Protein Modeling
• Virtual Screening (structure- & pharmacophore- based

*Departamento de Biología Molecular, Facultad de Ciencias Biológicas, Universidad de Concepción, Chile:

# CIGB. La Habana, Cuban and EMBL-Heidelberg, Germany

Arginase (EC 3.5.3.1) and agmatinase (EC 3.5.3.11) catalyses the production of urea from closely related substrates. In fact, agmatine results from decarboxylation of arginine by arginine decarboxylase. On the other hand, several highly conserved residues are detected in the amino acid sequences of these enzymes. For these reasons, they are considered as members of the arginase family of proteins. The idea is that they diverged from a common evolutionary origin, to reach their particular substrate specifities. The crystal structures of rat liver and Bacillus caldovelox arginases are available, and specific roles has been asigned to several active site residues, including His101, His126, His141 and Asp128 (according to their positions in the rat liver sequence). These roles has been also validated by chemical modification and site-directed mutagenesis of the rat liver and human liver arginases. A critical role for His163 (His141 for rat liver arginase) has been also deduced from chemical modification and site-directed mutagenesis of Escherichia coli agmatinase.

At present, a crystal structure for agmatinase is not available. We have, therefore, used molecular modelling, by analogy with B. caldovelox arginase as a reference, to obtain a model for the structure of E. coli agmatinase. The model thus obtained gives an accurate description of the interaction of agmatinase with Mn2+ and the existence of a binuclear metal center in fully activated enzyme. It also suggest a significant role for a loop, that include C159, Y155 and F161, in agmatine binding to agmatinase. Since this loop differs from that for arginase, which is bigger, a knowledge of these regions would explain the diffences in specificity between arginase and agmatinase. To test the validity of these conclusions, site directed mutagenesis was used to introduce changes in the loop for agmatinase. Replacing these residues by the corresponding residues in the sequence of arginase, the single-mutants C159S, Y155N and F161N, the double-mutants C159S/Y155N and Y155N/F161N and the triple-mutant Y155N/C159S/F161N were constructed. Interestingly, alteration in the entire conformation of this region, produced in the triple-mutant, was required for total loss of agmatinase activity. The single-mutant C159S and the double-mutant C159S/Y155N were even more active than wild.type agmatinase (~ 2-4 fold) and the other species were almost equally active than wild.type enzyme. In conclusion, our results emphasizes the importance of one specific loop region in substrate recognition by agmatinase. For a better understanding of the significance of these loops regions for the specificity requirements of arginase and agmatinase, insertions are now being introduced in the agmatinase sequence.

Bioinformatics Group, Edward Jenner Institute for Vaccine Research,

Compton, Newbury, Berks, RG20 7NN, UK,