Molecular Interaction Map of a Macrophage

Identifying intracellular molecular interactions is the first step toward understanding complex dynamics and mechanisms of the cell. We have created a comprehensive map of molecular interactions in a macrophage based on published literature. The map was created to analyze the experimental data sets produced by the Alliance for Cellular Signaling (AfCS). Therefore, we have extensively focused on signal transduction pathways that are targeted by the AfCS and have included essential metabolism, transcription, and secretion pathways that may be relevant. The map was created using CellDesigner, a software we have developed for molecular interaction map construction, and represented in Systems Biology Markup Language (SBML). This report describes notations used in the interaction map and the main features of CellDesigner and provides a list of all published literature that has been referenced in creating version 1.0 of the map. 1. The Macrophage Map 1.1 General Features We created a comprehensive macrophage model of molecular interactions, including Ca2+ signaling via G protein-coupled receptors (GPCRs), cytokine signaling, phagocytosis of oxidized low-density lipoprotein (oxLDL), production of reactive oxygen species (ROS) as well as glucose and lipid metabolism, that can be used as a first step towards a model-based analysis of a signaling network (Fig. 1). The map data (PDF and XML files in SBML format) can be downloaded from the AfCS/Nature Signaling Gateway via this document and the Systems Biology Institute Web site (http://www.systems-biology.org/002/). To create this model, we collected legacy knowledge from 234 published manuscripts (see the list of references for the Molecular Interaction Map of a Macrophage) and also referred to the ligand descriptions on the AfCS/Nature Signaling Gateway Web site. Although there were some exceptions, the criteria used to include possible molecular interactions from the literature were 1) modules that were confirmed to exist in macrophage cell, in culture cell lines, or in vivo; 2) signals that were confirmed by at least two independent articles; and 3) data published in the last decade. 1.2 Modules A total of 506 reactions and 678 species were included in this map. A “species” is a term defined by SBML as “an entity that takes part in reactions,” and it distinguishes the different states that are caused by enzymatic modification, combination, dissociation, and translocation. To distinguish the different states of a component in the same compartment in detail, we only need to assign a unique name to each state using suitable subscripts. For example, states of the protein that translocates from cytosol to plasma membrane can be distinguishable by names such as “XXcyt” and “XXpm.” Since SBML is a machine-readable format, all the information can be used for a range of computational analysis, including computer simulation. The breakdown of the species shown on this map is as follows: 363 proteins, 15 ions, 135 simple molecules, 113 oligomers, and 39 genes. In the number of species, 11 degraded products and 2 unknown molecules are also included. The nucleotides, ROS, carbohydrates, lipids, coenzymes, peptides, and amino acids are all shown as “simple molecules” in this version. Among 363 protein species, we identified 281 molecules, that is, 6 G protein subunits, 121 enzymes (including 47 kinases), 40 receptors, 7 ion channels, 40 transcription factors and their cofactors, 7 transporters, 14 cytokines, and 46 adaptor proteins. 1.3 Inputs and Outputs As inputs, we included 22 out of 26 ligands that were selected and tested for the ligand screening project in RAW 264.7 cells by the AfCS. The list includes interleukin1â (IL-1â), interleukin-4 (IL-4), interleukin-6 (IL-6), interleukin-10 (IL-10), interferon-á (IFN-á), interferon-â (IFN-â ), interferon-ã (IFN-ã), macrophage colonystimulating factor (M-CSF), granulocyte-macrophage colonystimulating factor (GM-CSF), transforming growth factor-â (TGF-â), macrophage inflammatory protein-1á (MIP-1á), monocyte chemotactic protein 1 (MCP-1), platelet activating factor (PAF), sphingosine-1-phosphate (S1P), lysophosphatidic acid (LPA), prostaglandin E2 (PGE2), complement component 5a (C5a), uridine 5’-diphosphate (UDP), uridine 5’-triphosphate (UTP), adenosine 5’triphosphate (ATP) (in place of 2-methyl-thio-ATP), lipopolysaccharide (LPS), and immunoglobulin 2a (IgG2a). August 25, 2004 Vol. 2 No. 14 DA AfCS Research Reports www.signaling-gateway.org/reports/v2/DA0014/DA0014.htm 2 Fi g. 1 . M ol ec ul ar I nt er ac tio n M ap o f a M ac ro ph ag e. T hi s m ap w as c re at ed w ith C el lD es ig ne r v er si on 2 .0 (h ttp :// w w w .sy st em sbi ol og y. or g/ 00 2/ ). A to ta l n um be r o f 5 06 re ac tio ns a nd 6 78 s pe ci es w er e in cl ud ed . T hi s m ap is a va ila bl e in a hi gh er re so lu tio n PD F fil e vi a th e on lin e re po rt. August 25, 2004 Vol. 2 No. 14 DA AfCS Research Reports www.signaling-gateway.org/reports/v2/DA0014/DA0014.htm 3 The four other synthesized ligands (isoproterenol, PAM2CSK4, PAM3CSK4, and resiquimod [R-848]) were not included in this map. In addition, we also included tumor necrosis factor-á (TNF-á) and oxLDL as input signals because they have well-established roles in macrophages during immune response and formation of foam cells in atherosclerotic lesions, respectively. As outputs, we described the induction of second messengers (such as Ca2+ and cAMP) and the release of 9 cytokines, 10 biogenic lipids, and 4 ROS. 1.4 Ambiguity Issues and Updating During the construction of the map, there arose unclear cases for the specific expression of a gene in RAW 264.7 cells as well as for the specific occurance of proteinprotein interactions. For example, the cross talk between NFêB and PPARã was shown in pluripotent mesenchymal stem cells, but it is not clear if such cross talk also exists in macrophages. In addition, there are conflicts among published papers and possible alternative explanations for certain interactions because of the varied experimental systems studied. For legacy data that depend fully on published literature, there are no clear means for making decisions on such cases. Therefore, we have taken a heuristic approach. First, we ensure that we incorporate molecules and interactions that are certain to exist and well agreed upon in the community. This can be determined on the basis of consistency among research papers as well as numbers of review papers. In cases where several interacting partners for one protein are reported, priority is given to those with demonstrated biological activity in macrophages. For example, there are several reported ligands for PPARgamma, but only 9-HODE and 13-HODE are represented because they have known effects in macrophages. Selections of this kind are made because of space constraints on the map. However, when space permits, all possible interactions are included. Second, when ambiguity exists between papers based on in vivo and in vitro experiments, we opt for conclusions from in vivo experiments. When certain interactions are only ambiguously reported, or not reported but known to exist in a variety of different cell types, we look at reports that use the embryologically nearer cell types to monocyte-derived macrophages, such as bone marrow stem cells, to increase the reliability of the map. Although some interactions are incorporated hypothetically, such as the IRAK1-TRAF6-TAK1-TAB1-TAB2 cascade, molecules and interactions that do not meet the criteria described above are not included in version 1.0 of the map. In the future we hope to develop a consistent methodology to score the reliability of the map based on legacy data and recalibration based on controlled comprehensive measurements. The version 1.0 map is intended to be comprehensive but not necessarily exhaustive. To create an exhaustive map we need hard evidence on which proteins exist in RAW 264.7 cells as well as which genes are expressed. However, the presence of a protein can sometimes show a paradoxical relationship with gene expression, thus it is important that we directly assess the presence of signaling proteins using direct measurement and not just by inference. In addition, detailed time-course measurements of protein levels combined with phosphoprotein assays and shRNAi data generated by AfCS labs will help researchers to reproduce and analyze dynamics of the network. We will periodically update the map on the AfCS Web site based on the most current data available. 1.5 Future Plan Our next step will be a systems-level analysis using real data retrieved from quantitative experiments (1). To do this, we need a highly reliable data set that includes expression levels, time course information, and results of perturbation effects. Quantitative modeling, simulation, and analysis of a particular focused subset of the system, as seen in the FXM (Focus on X Module) project, will be the next step. Based on this map, we are going to create another smaller but more detailed model, including isoforms. Using the SBW-SBML platform, we can easily share and revise this map, which will facilitate sharing and exchange of views in this project. The other aspect of our future plan is to incorporate data derived from the RIKEN FANTOM3 project, which measures the expression profile of all transcription start sites, including that of noncoding RNA. Contrary to FXM, which is focused on specific cascades but measured in depth, FAMTON3-based analysis will be genome-wide but only with expression profiles. Our challenge is to create a system of model-based analysis methods that can accommodate these two extremes. 2. Graphical Notations 2.1 Rationale behind the Notations Most diagrams in published papers are drawn using informal notations with sets of arrows, bar-headed lines, and circles roughly representing activation, inhibition, and the proteins involved, respectively. Fig. 2 is a typical example of just such a diagram for a MAPK cascade in a mammalian cell. In this diagram, the arrows may implicate several different reactions. For example, the arrow from Ras to Raf (marked as 1 in Fig. 2) appears to indicate that Ras activates Raf. However, in reality, Ras enhances plasma membrane translocation of Raf. Thus, this arrow is more accurately read as recruitment or translocation, rather than activation. Two arrows originating from ERK to RSK and c-Myc (marked as 2 in Fig. 2) are interpreted as activation of RSK and c-Myc by ERK. However, the same representation could also be interpreted as one complex (ERK) that splits into two subcomponents (RSK and c-Myc). The reason that we exclude this interpretation is because we already know some of the properties of the components involved, not because of anything within the diagram itself. How should we interpret August 25, 2004 Vol. 2 No. 14 DA AfCS Research Reports www.signaling-gateway.org/reports/v2/DA0014/DA0014.htm 4 the arrow leading from RSK to RSK (marked as 3 in Fig. 2)? In this case, the arrow is meant to be read as the translocation of RSK from cytosol to nucleus, instead of activation of RSK by RSK itself. Therefore, among these simple examples, there are three possible interpretations of the same arrow symbol—activation, dissociation, and translocation. Not only do notations in Fig. 2 have multiple meanings, they are ambiguous and unable to represent essential information (and therefore not machine readable). Correct interpretation depends upon the reader’s foreknowledge. For example, two arrows leading to Raf from PKC and Src indicate the activation of Raf by these two kinases. However, it is unclear what the mechanisms are, which residues are phosphorylated, or which is the first modulator of Raf. Accompanying text can supplement missing information to explain otherwise ambiguous points; however, in some cases the text can be more ambiguous than the diagrams. Ras Raf MKK ERK RSK PDK-1