CYCLONET—an integrated database on cell cycle regulation and carcinogenesis

Abstract

Computational modelling of mammalian cell cycle regulation is a challenging task, which requires comprehensive knowledge on many interrelated processes in the cell. We have developed a web-based integrated database on cell cycle regulation in mammals in normal and pathological states (Cyclonet database). It integrates data obtained by ‘omics’ sciences and chemoinformatics on the basis of systems biology approach. Cyclonet is a specialized resource, which enables researchers working in the field of anticancer drug discovery to analyze the wealth of currently available information in a systematic way. Cyclonet contains information on relevant genes and molecules; diagrams and models of cell cycle regulation and results of their simulation; microarray data on cell cycle and on various types of cancer, information on drug targets and their ligands, as well as extensive bibliography on modelling of cell cycle and cancer-related gene expression data. The Cyclonet database is also accessible through the BioUML workbench, which allows flexible querying, analyzing and editing the data by means of visual modelling. Cyclonet aims to predict promising anticancer targets and their agents by application of Prediction of Activity Spectra for Substances. The Cyclonet database is available at Author Webpage.

INTRODUCTION

The main goal of the Cyclonet database is to integrate information from genomics, proteomics, chemoinformatics and systems biology on mammalian cell cycle regulation in normal and pathological states. This will help molecular biologists working in the field of anticancer drug development to analyze systematically all these data and generate experimentally testable hypotheses (Figure 1).

Cyclonet incorporates data on various carcinogenesis related topics, such as: cell cycle control in mammals (Figure 2), cell survival programs (e.g. NF-κB pathway), regulation of covalent histone modifications and chromatin remodelling in cell cycle, DNA methylation and other epigenetic mechanisms of cell growth and differentiation. Biological pathways, computer models of cell cycle, microarray data coming from studies of cell cycle and analysis of cancer-related materials are also systematically collected in this database (1) (Author Webpage).

Cyclonet supports discovery of novel drug targets and development of effective anticancer therapies by collecting all available data related to the control of cell cycle in normal and pathological states and providing a system biology platform for knowledge-based anticancer drug discovery.

Novel software technologies were used for the database development:

• the BioUML workbench [Author Webpage, (2,3)] was used for formal description and visual modelling of biological pathways and processes related to the cell cycle regulation and cancer (Figure 2). It also allows to simulate the behaviour of the described systems using Java or MATLAB simulation engines;

• BeanExplorer Enterprise Edition (Author Webpage) was used to develop web interface for the Cyclonet database (Figure 3).

THE CYCLONET DATABASE STRUCTURE AND CONTENT

The Cyclonet database consists of three main components (see Table 1):

• diagrams and models of biological pathways (metabolic pathways, signal transduction pathways and gene networks) involved in cell cycle regulation and carcinogenesis;

• microarray original data and results of their statistical analysis;

• chemoinformatics data—drug targets, ligands and pharmacological activities for cancer treatment.

The Cyclonet database is organized as a relational database (MySQL DBMS). All sections contain a number of tables that are highly interconnected through crossreferences. Such elaborated relational schema enables complex queries combining various types of information.

Data in Cyclonet are compiled mainly by manual literature annotation. Links to the public databases, such as, GeneOntology (4), RefSeq (5) and Ensembl (6) are provided from genes, proteins and other respective entries. Cyclonet also contains a vast body of literature references that are arranged by categories.

Biological pathways

We use BioUML for formal description of signal transduction pathways and gene networks involved in cell cycle regulation and carcinogenesis (2,3,7). Cyclonet pathways section allows to store the detailed description of biological pathways, their components, models as well as the results of simulation.

We are using several diagram types of BioUML to describe cell cycle regulation and carcinogenesis:

• semantic networks describing relationships between the main concepts (for example, G1 phase, G1/S transition, mitotic checkpoints) and components of cell cycle regulation;

• pathways describing structure of cell cycle regulatory networks as compartmentalized graphs. We have classified the annotated networks into a number of categories that describe different parts of cell cycle regulatory networks in details (for example, a network that provides G1/S transition, NF-κB signal transduction pathway and its influence on apoptosis and others).

Models

The BioUML technology was also used for visual modelling of cell cycle regulation. Known cell cycle models were imported from SBML (8) and CellML (9) model repositories. We added into Cyclonet several new recent models by manual annotation of respective literature sources. We also created our own novel model of regulation of G1/S transition of cell cycle.

Currently, Cyclonet contains 37 models of cell cycle regulation. All models can be classified into two groups: (i) general models that simulate behaviour of rather small systems including abstract objects that reflect real biological components in the cell; (ii) ‘portrait’ models that try to simulate different sub-processes in cell cycle and include real genes, proteins and other cellular components. We validated each model by using the BioUML simulation engine and comparing the results with the published results. The results of such simulations were then stored in the Cyclonet database. These data can be displayed as graphs by the BioUML workbench (Figure 2) or web isnterface generated by BeanExplorer EE.

Microarray data

Cyclonet contains a comprehensive list of human genes which is composed from the genes described in HGNC (10) and UniGene (11) databases. Cyclonet also contains all assignment of cDNA clones to the corresponding human genes.

We analysed 41 microarray resources [mainly, Standford Microarray Database (12), GEO (13), Oncomine (14) and published articles, for example, (15)] and obtained 354 links to microarray experimental data related to the cell cycle and cancer. These links to microarray data were classified according to cancer types.

Currently data for five microarray experiments related to breast cancer and five experiments with cell cycle time series were loaded into the Cyclonet database and analysed. We did a statistical analysis as well as meta-analysis of the data (see Supplementary Data) and obtained 33 gene lists (IDs GL0001–GL0033 in ‘Microarray data and results’ of Cyclonet) that belong to several categories:

• lists of genes periodically expressed during cell cycle (GL0007, GL0020 and GL0021) (15,16);

• lists of genes whose expression is changing monotonically during cell cycle (GL0022) (15,16);

• breast cancer gene lists:

  • up- and down-regulated genes in each of the five experiments (GL0001–GL0006, GL0008–GL0018) (17–21);

  • up- and down-regulated genes revealed on the basis of meta-analysis (GL0019, GL0023–GL0033) (22);

• lists of genes obtained by other authors during microarray analysis of breast cancer (18–22) and pancreatic cancer (23).

Such lists of differentially expressed genes are very good resources for selecting cancer biomarkers as well as perspective targets for further experimental and bioinformatic analysis. Statistical methods used in this analysis are described in Supplementary Data.

Chemoinformatics data

Chemoinformatics section summarizes the current knowledge about known anticancer targets, anticancer agents, mechanisms of their action and conditions where those compounds are applied. For this purpose we are collecting the following information as it is represented in Supplementary Figure 1S:

• names of anticancer agents (generic name, brand name) and its synonyms;

• chemical name;

• CAS number;

• structural formulae;

• class (activity)—includes information about molecular mechanisms of action (e.g. Topoisomerase II inhibitor) and pharmacotherapeutic action (e.g. Antimetabolite);

• literature references where the data were obtained for the respective anticancer agent.

Semantic networks provide a reasonable formalism to describe the relationships between the anticancer agents and their targets, activities and cancer types (or other conditions) where these agents are generally applied (Figure 4). Summary statistics of the chemoinformatics section is shown in Table 1.

INTEGRATION BETWEEN COMPONENTS OF CYCLONET

Integration between all three components of the Cyclonet database, namely, biological pathways and models, microarray data and chemoinformatics data, is provided by the following mechanisms:

• All data are stored in the same relational database. This allows us to develop the complex SQL queries to integrate data from different components. A number of predefined SQL queries are provided through the web interface for the Cyclonet database.

• The web interface provides detailed representation (view) of components of biological pathways, microarray and chemoinformatics data with a number of crossreferences between the components. For example, a view for an anticancer agent contains links to its activities, cancer types, conditions of its application for anticancer therapy, components of biological pathways (genes and proteins) that are targets for this agent. These targets, in turn, can be linked to diagrams and dynamic models of cell cycle. Another example is a gene view that contains links to cDNA clones used for this gene in microarray experiments, microarray experiments where expression level of this gene was measured, gene lists where this gene was revealed as result of microarray analyses, anticancer agents for which this gene is a target, diagrams and models where this gene participates.

• The BioUML search engine allows to find the relationships between the anticancer agent and biological pathway components and display these results as an editable graph. As a starting point user can select the anticancer agent (small molecule), concept, gene or protein.

APPLICATION OF THE CYCLONET DATABASE

Prediction of new anticancer agents for known targets/mechanisms of action

All anticancer agents are grouped in the Cyclonet database according to their targets/mechanisms of action and chemical structure. This information is used for the training of computer program PASS (Prediction of Activity Spectra for Substances) (24). As a result of the training procedure, PASScan predict if new molecules from databases of commercially available samples may have activities related to the regulation of cell cycle. Three commercially available chemical compounds' sample databases were analysed, provided by ASINEX, ChemBridge and InterBioScreen (IBS). They contain totally the structures of 1 445 018 compounds. We predicted a number of compounds as potential cell cycle regulators using probability threshold Pa > 70%. By increasing the Pa threshold, e.g. to 90%, one can select highly specific compounds only. The results of this analysis are stored in the Cyclonet database (see the statistics in Table 2). One may conclude that commercially available chemical compounds databases contain a plethora of ligands acting on different targets related to the cell cycle regulation.

Application of Cyclonet to model the cell cycle

Computer simulation methods have been applied to study the dynamics of gene networks regulating the cell cycle of vertebrates. The data on the regulation of the key genes obtained from the Cyclonet database have been used as a basis to construct gene networks of different degrees of complexity controlling the G1/S transition, one of the most important stages of the cell cycle. The behaviour dynamics of the model has been analysed. Two qualitatively different functional modes of the system have been obtained. It has been shown that the transition between these modes depends on the duration of the proliferation signal. It has also been demonstrated that the additional feedback from factor E2F to genes c-fos and c-jun, which was predicted earlier based on the computer analysis of promoters (25), plays an important role in the transition of the cell to the S phase (see Supplementary Figure 2S) as it is documented in gene expression databases TRANSFAC (26) and TRANSPATH (27).

Application of Cyclonet for searching of new targets for anticancer therapy

The Cyclonet database can be applied for searching of new targets for anticancer therapy. For this purpose we have revealed genes whose expression are significantly deregulated during breast cancer and created a set of diagrams in the Cyclonet database (diagrams DGR0228–DGR0240) and mapped information about gene expression into the diagrams. An example of gene expression data mapping is shown in Supplementary Figure 3S for a fragment of a diagram of the proapoptotic network (DGR240).

FURTHER DEVELOPMENT

Now we are developing a set of plug-ins in the BioUML workbench for visual modelling of integration between the biological pathways and microarray data that will provide: coloring of diagrams for biological pathways to display data on gene expression levels, reconstruction of gene networks and fitting the model parameters in accordance with the microarray data. Also, a new information arising from both ‘omic’-sciences and chemoinformatics is added periodically to the Cyclonet database, to update its content.

SUPPLEMENTARY DATA

Supplementary data are available at NAR online.

Authors are grateful to V. Komashko and V. Valuev for microarray data annotation, E. Cheremushkina for annotation of a number of pathway diagrams and V. Zhvaleyev for technical assistance. This work was supported by INTAS grant No. 03-51-5218, MIUR-FIRB grant No. RBLA0332RH Laboratory for Interdisciplinary Technologies in Bioinformatics, by European Commission under FP6-‘Life sciences, genomics and biotechnology for health’ contract LSHG-CT-2004-503568 ‘COMBIO’ and under ‘Marie Curie research training networks’ contract MRTN-CT-2004-512285 ‘TRANSISTOR’ and BIOINFOGRID No. 026808. Funding to pay the Open Access publication charges for this article was provided by the European Commission project No. 037590 from the call FP6-2005-LIFESCIHEALTH-7.

Conflict of interest statement. None declared.

REFERENCES