SBC seminars 2007

The first and main part of the talk concerns bioinformatic studies of G protein-coupled receptors (GPCRs) performed during a PhD at Uppsala University and the department of Neuroscience. It includes results from searches for new human receptor sequences, analyses of the human, mouse and rat GPCRomes and a new classification system for GPCRs. Experiences from genome and protein database mining, phylogenetic analysis, sequence motifs and the usage of expressed sequence tags (ESTs) for gene sequence curation and derivation of preliminary expression profiles are also to be discussed.

The second part will briefly describe the development of (XML) standards for antibody and protein array databases and database exchange. Standards are currently implemented, in a post doctoral project at the European Bioinformatics Institute, by extending the functionality of the IntAct database (www.ebi.ac.uk/intact) and HUPOs standard for exhange of molecular interaction data (http://psidev.sourceforge.net/mi/rel25). The effort is part of a European consortium (www.ProteomeBinders.org) which proposes to establish a resource of binding molecules against the entire human proteome.

Post-transcriptional control of eukaryotic gene expression is more general and elaborate than previously thought. Analysis of sequence data and splicing-sensitive arrays suggests that al ternative splicing regulates half of all human protein-encoding genes and numerous studies ha ve shown that different transcript isoforms (splice variants) have different, sometimes antag onistic, function. For example, Fas exon 6 can be included or skipped to generate mRNAs encod ing, respectively, a membrane bound form of the receptor that promotes apoptosis or a soluble isoform that prevents programmed cell death (Cascino, Fiucci et al. 1995).

We have used Affymetrix exon arrays to study transcriptional and post-transcriptional responses that occur in primary mouse T-cells after activation. The exon arrays (promise to) enable unbiased detection of the expression level of individual exons and may therefore be used to identify alternatively spliced exons and alternative promoters in extension to the gene expression levels. To this end, we developed a novel algorithm to identify alternatively regulated exons, which enabled us to characterize both the post-transcriptional and transcriptional regulation. Our algorithm was validated by real-time PCR analysis of a set of 16 exons, inferred from the exon array data, which included "skipped exons", shifts in promoter and 3'UTR usage. In addition, we are in the process of determining the conservation pattern of the post-transcriptional events, by analyzing exon array data from activated and resting human cord blood derived T-cells. Finally, we would like to use this model system to investigate the coupling between different modes of gene regulation (e.g. miRNAs and alternative 3'UTR usage) as well as the functional importance of specific isoforms.

Since its conception a little over 10 years ago, the primary goal of the Protein FAMily database (Pfam) has been the classification of protein domains into distinct families. Over the past two years Pfam has increased dramatically in size, both in terms of the number of families and also the breadth of information available for each family.Our goal is to expand the data within Pfam so as to allow it to be understood in a truly biological context, rather than as simply a collection of families.

As part of this work we have started to classify Pfam entries hierarchically, gathering them, where possible, into groups of related families termed Clans. In this talk I will outline the tools we employ to identify distantly related Pfam families and discuss in detail one such tool, known as SCOOP (Simple Comparison Of OutPuts), which was developed within the group. SCOOP takes the outputs from a set of profile HMM searches and looks for sequences that match multiple HMMs, scoring each match according to the likelihood that it may be expected to occur purely by chance. A higher score, indicating the likelihood that a given match is statistically unlikely by chance alone, is taken to indicated a potential distant similarity. Having identified these distant relationships we are now able to use our understanding of well categorised families to improve our understanding of other, less well characterised ones.

As well as continuing to improve our techniques for annotating domains, we have also begun to increase the number of annotations in Pfam which are based on sequence features, using a number of different approaches. The first approach which will be described here, relies on the development of a methodology for the transfer of experimentally defined active site between sequences within Pfam. The second approach that I will mention uses the Distributed Annotation System (DAS) to retrieve data from disparate resources around the world and integrate them in a single visualisation. We hope to build on these and other data-mining techniques to improve and increase the number of annotations across the Pfam data.

Finally, I will touch on a completely different area of work within the group, which focusses on identifying domain-domain interactions using known three-dimensional structures obtained from the Protein DataBank (PDB). These data are contained in a subsection of Pfam termed iPfam (UK site only). I will describe some of the applications of the iPfam data and describe how we hope to use them to improve our understanding of protein networks at the domain level, as well as in terms of disease-causing mutations.

Identifying gene families is an old problem that many researchers have looked at and there are probably as many solutions as there are researchers. Now there is one more. I will describe some recent work by myself and Jens Lagergren that includes a formal definition of the term gene family and some practical results.

By using a database of highly refined protein X-ray structures, it is possible to build to a high degree of accuracy the stucture of any protein containing only some of its atoms. The database is broken into a set of short segments, which are then fitted onto the framework of the target structure. In the process, three criteria are used for filtering out good segment matches: amino acid residue similarity, conformational similarity (rms deviation) and segment compatibility with the target structure (van der Waals interactions).

For a test set of proteins, that has between 46 and 323 residues, the all-atom rms deviation of the modeled stuctures is between 0.93 and 1.73 Å, values comparable with differences in refining of X-ray structures.

Also, this approach is very suitable for building models using multiple templates. Using multiple template sequences derived from consensus modelling approaches, it is possible to improve upon the single best template in a majority of cases.

Proteomics provides description of cancer-related changes in cells.Richness of these datasets requires tools to unveil systemic properties of carcinogenesis.

We perform proteome profiling of tumor and non-cancerous tissues, primary and established human breast epithelial cells. Analysis of the identified cancer-specific proteins showed that multiple regulatory pathways are affected, e.g. metabolism, apoptosis, stress response, serine/threonine and tyrosine kinase signalling. Systemic exploration of the identified proteins indicated that the status of signalling pathways is as important for cancer development, as expression of particular oncogenes or tumor suppressors. Moreover, analysis of systemic properties of cell responsiveness to chemotherapeutics provides the basis for individualized treatment of patients.

Daley, D.O., Rapp, M., Granseth, E., Melén, K., Drew, D., and von Heijne, G. (2005) Global topology analysis of the Escherichia coli inner membrane proteome. Science 308, 1321-1323.

Rapp, M., Seppälä, S., Granseth, E., and von Heijne, G. (2006) Identification and evolution of dual topology membrane proteins. Nature Struct.Mol.Biol. 13, 112-116.

Rapp, M., Susanna Seppälä, S., Granseth, E., and von Heijne, G. (2007) Emulating membrane protein evolution by rational design. Science 315, 1282-1284.

Phobius has been benchmarked as the most accurate transmembrane topology predictor on single sequences, in particular when signal peptides are treated as unknowns. In the 2004 Phobius publication it was found to be 10-23% more accurate than TMHMM, previously considered the most accurate program. Despite this, the awareness that TMHMM is deprecated and superseded by Phobius is poor. With this talk I hope that at least scientists at SBC will be informed, and hopefully will pass it on to external colleagues.

The novelty of Phobius was that it can predict both signal peptides and transmembrane segments. This allows it to discriminate between these two often confused features, and the accuracy improvement stems from this ability.

To illustrate how important this discrimination is, we carried out the following study: TMHMM and SignalP were applied to five complete proteomes. 30-65% of all SignalP predicted signal peptides and 25-35% of all TMHMM predicted transmembrane topologies overlapped. This casts doubt over the predictions for 5-10% of each proteome, see following article. Phobius resolves these conflicts by making an optimal choice between transmembrane segments and signal peptides. It also allows constrained and homology-enriched predictions.

To make Phobius accessible beyond the web servers phobius.cgb.ki.se and phobius.binf.ku.dk we have also set up a DAS (Distributed Annotation System) server at SBC. This service can be queried with the DAS protocol using any Uniprot accession number, and a topology prediction will be returned. I will describe the DAS system briefly and demonstrate the DAS registry at www.dasregistry.org which currently holds 264 services (62 for

DAS is a simple protocol designed for easy sharing of biological data. In this talk I will introduce you to DAS, show some examples of data sources that offer their information via DAS, and describe my work on DASher, a viewer application that collects DAS-format annotations and displays them along a protein sequence.

Expression of the Epstein-Barr virus (EBV) latent membrane protein (LMP1) is regulated by virus- and host cell-specific factors and plays an important role in switching between different types of EBV. The current thinking is that LMP1 is involved in host networks and the expression of LMP1 is controlled by host cell transcription factors (TF). Here we present the results of in silico prediction TF binding sites within the LMP1 promoter region in EBV sequence.

Humans host complex microbial ecosystems, which are postulated to contribute to both health maintenance and the development of chronic inflammatory diseases. Alterations in the human gut microbiota have recently been associated with diseases such as cancer, type II diabetes, inflammatory bowel disorder, allergy and obesity. Determining the microbial composition in patients and healthy controls may provide novel therapeutic targets, but is currently a time-consuming and expensive process. Large-scale studies have therefore been prohibited. However, new high-throughput culture-independent molecular tools are now developed, allowing the scientific community to characterize and understand the microbial communities underpinning biological processes in unprecendented ways.

Modern technology provide us with large amounts of biological data. This opens possibilities to study the large-scale organization and function of a biological system. The information is, however, typically not comprehensive enough to use the same tools for studying large-scale systems (e.g. the metabolism) as small subsystems (e.g. the citric acid cycle). I will discuss network theory in general, and how statistical graph theory can be used to study the organization of metabolism. Specifically I will talk about the role, definition and detection of "currency metabolites" -- ubiquitous substances (like water, carbon dioxide, ATP, etc.) that occur in a multitude of reactions.

We present a comprehensive computational study of some 900 possible lambda-lac mutants of the lysogeny maintenance switch in phage lambda, of which up to date only 19 have been studied experimentally (Atsumi & Little, PNAS 103: 4558-4563, (2006)). We clarify that these mutants realise regulatory schemes quite different from wild-type lambda, and can therefore be expected to behave differently, within the conventional mechanistic setting in which this problem has traditionally been framed. We verify that indeed, with reasonable modelling assumptions and across this wide selection of mutants, the lambda-lac mutants for the most part either have no stable lytic states, or should only be inducible with difficulty. In particular, the computational results contradicts the experimental finding that four lambda-lac mutants both show stable lysogeny and are inducible. This work hence suggests either that the four out of 900 mutants are special, or that lambda lysogeny and inducibility are holistic effects involving other molecular players or other mechanisms, or both. The approach illustrates the power and versatility of computational systems biology to systematically and quickly test a wide variety of examples and alternative hypotheses for future closer experimental study.

It is of principal importance to understand how different signalling pathways are changed in relation to different disorders. In this seminar, the JAK-STAT-SOCS signalling pathway will be discussed. This pathway is known to be of relevance for immune functions, metabolic control and cancer. In the seminar, the components of this pathway will be briefly reviewed as will the principles how the pathway is activated and inactivated. A particular emphasis will be made on STAT5 and SOCS2 - these factors seem to be important in growth regulation and in prostate cancer. The potential of using high through-put technologies and predictive methods to probe the JAK-STAT pathway will be exemplified.

Non-coding (nc) RNAs are emerging as important and surprisingly ubiquitous players in gene regulation across all kingdoms, adding new layers of complexity to the regulatory machinery. Beside challenging views about the junk content of higher organisms' DNA and opening possible explainations to the riddle of high complexity vs low gene number observed in many organisms.

I present the sequel to a genome wide ncRNA search in the genome of enterococci which is eventually applicable to other prokaryotes and to some extent to eukaryotic and viral genomes.

DAS is a simple protocol designed for easy sharing of biological data. In this talk I will introduce you to DAS, show some examples of data sources that offer their information via DAS, and describe my work on DASher, a viewer application that collects DAS-format annotations and displays them along a protein sequence.

With the human genome sequenced much effort is put into elucidating the interactions and functions of the encoded proteins. This has been done focusing on both the proteome and the transcriptome with analysis of e.g. co-expression, co-localization and tissue profiling. How mRNA expression correlates with relative protein amounts could be of utmost importance e.g. when using protein co-expression as an indicator of protein functional coupling. Pairwise correlations at the gene level were generated for two datasets. One data set of antibody-based tissue profiling of proteins, from the human protein atlas, and one of tissue-specific patterns of mRNA expression, from the GNF transcriptome atlas. This was done using Pearson- and Spearman correlations as well as mutual information. Finally the correlation between the two datasets was calculated, revealing a moderate correlation which was nevertheless significantly higher than random simulations.

Several bioinformatic approaches have previously been used to find novel sites of Adar mediated A-to-I editing in human. These studies have resulted in the discovery of thousands of genes hyper-edited in their non-coding regions but very few substrates that are site selectively edited. We have compiled a screen to search for new sites of selective editing primarily in coding sequences. To avoid hyper-edited repeat regions we have applied our screen to the alu-free mouse genome. First we construct an explorative screen based on RNA structure and genomic sequence conservation. We evaluate the explorative screen by means of enrichment of A-G mismatch, that is, the discrepancy between the expressed sequence and the genomic template for A-to-I edited sites. The enrichment and the corresponding p-values implicate A-to-I editing as a key mechanisms in fine tuning proteome diversity. Known substrates suggest that A-to-I editing is particularly important for normal brain development in mammals. Subsequently, we extend the explorative screen by including A-G mismatch as well as a specific scoring scheme based on characteristics for known A-to-I edited sites. The result of applying our extended screen to the mouse genome gives a substantial number of novel putative substrates of which 63 are currently experimentally validated.

Understanding the dynamics behind domain architecture evolution is of great importance to unravel the functions of proteins. Complex architectures have been created throughout evolution by rearrangement and duplication events. An interesting question is how many times a particular architecture has been created, a form of convergent evolution or domain architecture reinvention. Previous studies have approached this issue by comparing architectures found in different species. We wanted to achieve a finer-grained analysis by reconstructing protein architectures on complete domain trees.

The prevalence of domain architecture reinvention in 96 genomes was investigated with a novel domain tree based method that uses maximum parsimony for inferring ancestral protein architectures. Domain architectures were taken from Pfam. To ensure robustness, we applied the method to bootstrap trees, and only considered results with strong statistical support.

We detected multiple origins for 12.4% of the scored architectures. In a much smaller dataset, the subset of completely domain-assigned proteins, the figure was 5.6%. These results indicate that domain architecture reinvention is a much more common phenomenon than previously thought. We also determined which domains are most frequent in multiply created architectures, and assessed whether specific functions could be attributed to them. However, no strong functional bias was found in architectures with multiple origins.