An Interactive Macrophage Signal Transduction Map Facilitates Comparative Analyses of High-Throughput Data

Macrophages (Mϕs) use a broad range of receptors to recognize and interact with pathogens and to initiate intrinsic signal cascades. These cascades can be modulated by extrinsic signals such as chemokines, drugs, and virulence factors. This modulation influences the balance between activating and inhibiting signals. The sum of events shapes the ensuing macrophage phenotype, which can contribute either to perpetuation or resolution of inflammation (1, 2). These key phenotypes are governed by a large and highly interconnected biochemical network containing a large array of transcriptional, signaling, and metabolic circuits. The cross-talk between circuits allows the Mϕs to fine-tune responses and to adapt to the tissue microenvironment (3). The multilayer composition of Mϕ regulation makes it inevitable to use advanced computational tools that incorporate network and systems biology approaches (4, 5). Analyses of transcriptome, proteome, and other types of high-throughput data can thus lead to more precise insights into the regulation of immune responses to infections.

Bacterial lung infections remain one of the most widespread infectious diseases and are one of the leading causes of death. They are associated with multiple bacterial species, especially Streptococcus pneumoniae, Legionella pneumophila, or Mycobacterium tuberculosis (6, 7). Apart from the alveolar epithelium and neutrophils, resident Mϕs and recruited monocyte-derived Mϕs essentially contribute to the immune response, mainly through the production of inflammatory cytokines and the phagocytosis of bacteria (1, 7) after their recognition by specific receptors. For example, S. pneumoniae peptidoglycan is recognized by nucleotide-binding oligomerization domain (NOD)2 (8), L. pneumophila flagellin by TLR5 (9), and M. tuberculosis trehalose dimycolate by macrophage-inducible Ca²⁺-dependent lectin (10). As a countermeasure against recognition, pathogens produce virulence factors that actively manipulate the signal transduction in Mϕs. The directed modulation by pathogens may lead to Mϕ dysfunction and thus contribute to disease progression (11). For example, L. pneumophila and M. tuberculosis perturb their phagolysosomal processing to establish their intracellular niche (12, 13). Similarly, effectors of the extracellular pathogen S. pneumoniae can reach the cytoplasm of Mϕs, promote cell death, and thus contribute to the pathogenesis of pneumonia (14). The specific molecular targets of virulence factors are, however, mostly unknown.

Severe S. pneumoniae– or L. pneumophila–associated pneumonias are treated with antibiotics such as β-lactams (e.g., amoxicillin), macrolides (e.g., clarithromycin), fluoroquinolones (e.g., levofloxacin), or tetracyclines (e.g., doxycycline) (15, 16). M. tuberculosis–associated lung infections (tuberculosis) require a combination therapy of isoniazid, rifampicin, ethambutol, and pyrazinamide (17). Recently, specific immunomodulation of Mϕ signal transduction has gained interest in the therapy of pneumonia (18, 19). Adjuvants or supportive immunomodulating drugs are required to further promote the efficacy and beneficial effects of antibiotics, and to reduce harmful effects of the host immune reaction (e.g., tissue damage, organ failure, or even systemic inflammation, which is associated with high mortality). Taken together, the detailed understanding of Mϕ signaling and the identification of key access points are required for the development of immunomodulating drugs and the successful treatment of bacterial lung infections.

In this study, we provide a comprehensive, fully annotated, and expert-validated map of inflammation-related signal transduction pathways in Mϕs. We introduce an interactive Web platform that can be used to browse the map as well as analyze and compare high-throughput data. Our systems level case study on the druggability of a pathogen-modulated pathway network revealed vantage points for immunomodulation that are shared across three infection scenarios. In this manner, we highlight potential interference between drugs, pathogens, and Mϕ signaling, which might support the development of Mϕ-directed immunomodulating therapeutic strategies against bacterial lung infections.

For reconstruction of the map, the National Center for Biotechnology Information MEDLINE archive was queried for “macrophage,” “signaling,” and “infection.” The obtained literature was browsed for pathways and proteins described to be important for macrophage activation and the subsequent immune response (e.g., pathogen recognition and cytokine production). This information was then graphically reconstructed using CellDesigner (v4.3 and v4.4) (20).

Furthermore, previously published models for macrophage signaling pathways were searched in the databases BioModels (release of database version [r]27) (21) and PANTHER Pathway (r9.0) (22). The following models were fully or partially reused: BioModels MODEL1203220000 and MODEL2463683119; Panther P00048, P00052, P00054, P00035, P00036, P00046, and P05918. Additionally, the following pathways were downloaded in May 2014 from http://www.macrophages.com and partially integrated: “Interleukin-4 -10 & -13 Pathways,” “Macrophage Activation Extended,” “NF-kappa-B Signaling,” “Non-TLR Pathogen Detection,” “p53 Signaling Pathway,” “Toll Like Receptor & Kinase Pathways,” “VEGF and TGFB Pathways.” The map furthermore contains excerpts of maps previously published by Oda et al. (23), Oda and Kitano (24), and by Raza et al. (25, 26).

Next, all reactions in the curated map were verified by searching experimental evidence for their relevance in mouse or human cells. The acquired publications were used to annotate the reactions in the map using CellDesigner’s minimum information requested in the annotation of models (27) support. Additionally, genes were annotated with Ensembl IDs (r75 and the following ones) (28, 29), microRNAs (miRNAs) with miRBase IDs (r20 and the following ones) (30), proteins with UniProt IDs (r2014_7 and the following ones) (31), complexes with BioGRID IDs (r3.2.110 and the following ones) (32), and simple molecules and ions with ChEBI IDs (r118 and the following ones) (33). If information could not be annotated or graphically presented, text notes were added. Because most of the experimental evidence in immunology is obtained in, with, or starts from murine cells and models, most molecular annotations refer to the species mouse.

To validate the map’s quality, a random sample of 25 reactions with modifiers (∼3% of all reactions) was selected from the final version. Two postdoctoral researchers each with 5 y of experience in molecular immunology independently graded the correspondence between each depicted reaction and its referenced literature, paying attention to errors and unsupported details. The degree of correspondence was classified into one of six groups and summary statistics were calculated.

The signaling map was transformed into a Graph Modeling Language (GML) (34) representation by an algorithm implemented in the programming language Python (v2.7) (35) (available from https://github.com/marteber/miRNexpander). This automated process removed factor duplicates, pruned some factor types (genes, mRNA, ions, small molecules, and phenotypes) and merged or discarded some reaction types (transcription, translation, transport) (see Fig. 1A for an example and https://vcells.net/macrophage/uploads/Extended-Methods.pdf for details). All annotations were extracted and transferred together with their respective molecules and reactions.

For expansion of the network, initially missing annotations for molecules, particularly in the case of complex subunits, were automatically retrieved from the databases National Center for Biotechnology Information Gene and UniProt, manually curated, and finally introduced computationally into the network. Furthermore, the databases miRTarBase (r4.5) (36) plus miRecords (r4) (37), TRANSFAC (r2015.1) (38), and RegPhos (r2.0) (39) were queried automatically for miRNA–mRNA, transcription factor (TF)–gene, and kinase–phosphoprotein interactions, respectively. The results were then filtered to discard all genes not associated with a predefined set of gene ontology terms. This set was assembled from those gene ontology terms associated with genes in the curated map and lowest in the ontology hierarchy (via The Jackson Laboratory’s bioinformatics resources and a database from http://www.geneontology.org), and those conceptionally associated with macrophages and infections (selected manually through QuickGO queries) (40). After filtering, the miRNA–mRNA and kinase–phosphoprotein interactions were manually curated by evaluating the referenced publications, whereas this step was skipped for the filtered TF–gene interactions due to their large number. Finally, the correct and relevant database entries were introduced into the network together with initially absent corresponding molecules, and the generated network was written to a GML file.

The converted and expanded network (GML file) was imported into and analyzed with Cytoscape (v3.2.1) (41). For the graphical representation of the expanded network based on cellular compartments, the Cytoscape plug-in Cerebral (v1.2) (42) was used.

The databases Gene Expression Omnibus (43) and ArrayExpress (44) were queried for expression data sets that meet the following criteria: 1) cell type: human Mϕs, monocytes, or PBMCs; 2) context: bacterial lung infection; 3) time point: early, nonchronic phase of infection; and 4) data: publicly available; processing and analysis technically feasible. The data sets GSE61535 (human monocyte-derived Mϕs infected with L. pneumophila strain AA100 for 1 h at a multiplicity of infection [MOI] of 20:1 and incubated for another 8 h) and E-MEXP-3805 (human monocyte-derived Mϕs infected with M. tuberculosis strain H37Rv for 2 h at a MOI of 1:1, then incubated with fresh medium for another 18 h) were the best matches. Furthermore, we used an expression data set (GSE77506) of human monocyte-derived Mϕs infected for 16 h with S. pneumoniae strain D39 (MOI of 1:10).

Only human data sets were used for the case study to improve its clinical relevance. Additionally, six data sets related to LPS stimulation under different experimental conditions (GSE4712, GSE8621, GSE46903, GSE50542, GSE28880, and E-MEXP-3469) as well as one related to reactive oxygen species (ROS) challenge (GSE15457) were added to the Web platform.

All data sets were preprocessed and analyzed with the statistics tool R (45) with the package limma (46). For all unprocessed sets, data were background corrected, quantile normalized and log₂ transformed before feature selection with control of the false discovery rate (47) was performed. All resulting expression-related values were assigned to the molecules in the network based on the latter’s gene symbol annotation.

For the identification of perturbed subnetworks, the Cytoscape v3.2 plug-in jActiveModules (v3.1) (48) was employed. jActiveModules combines a statistical scoring system with a search algorithm based on simulated annealing and was applied to each of the 44 data sets separately. The number of modules to detect was set to 3, the overlap threshold (i.e., maximal fraction of overlap between detected network regions) was set to 0.1, the search depth (i.e., distance of module growth steps starting from differentially regulated factors) to 1, and the maximal depth was set to 1 (i.e., maximal number of iterations). The three obtained modules were merged into one subnetwork that is representative for the corresponding data set.

For the case study on L. pneumophila, M. tuberculosis, and S. pneumoniae, we relaxed the above parameter settings to allow for some factor overlap between the scenarios. To achieve this, the module count per scenario was increased to 10, the overlap threshold to 0.5, and the maximal depth to 3. The 10 modules were then combined, yielding one subnetwork per infection scenario. From these three, the shared factors and reactions were extracted, and the resulting network was termed regulatory core. The Venn diagram of the three subnetworks was produced in R with the package VennDiagram (49).

The Cytoscape plug-in CyTargetLinker (v3.0.1) (50) was used to attach drugs to their target proteins based on information listed in the database DrugBank 4.2 (51).

We collected and evaluated scientific literature to obtain a resource for the analysis of macrophage signal transduction (Fig. 1A, left). The reconstructed map of signal transduction pathways in inflammatory Mϕs is shown in Fig. 2A. It is organized in 16 areas roughly depending on functional relations. We included primary signaling induced by ligand–receptor interactions through IFN (box 1 in Fig. 2A), Ras/Raf/ ERK (box 2), protein kinase C/PI3K/protein kinase B (AKT) (box 3), Toll-/NOD-/ retinoic acid–inducible gene–like receptors (box 4), growth factors (box 5), hypoxia (box 6), chemokines (box 7), complement/integrins (box 8), ILs (box 9), or pathogen molecules (box 10). Additionally, we incorporated subsequent downstream regulatory circuits and the phenotypes they regulate, including entire blocks accounting for caspase activation (box 11), transcriptional regulation (box 12), translational regulation (box 13), production of reactive nitrogen species/ROS/antimicrobial peptides (box 14), production of lipid mediators (box 15), and triggering of cytokines/chemokines/Mϕ phenotypes (box 16).

The reconstruction procedure incorporated information from 479 publications and several previously published maps (see 2Materials and Methods), yielding a network of 842 molecule species (i.e., genes, mRNAs, miRNAs, proteins, complexes) and 814 reactions (Table I). The cellular compartments relevant for infections (i.e., plasma membrane, nucleus, endosome, and phagosome) are represented. Cell surface receptors are situated at the top and lateral margins, with their transduction pathways directed inward and converging on the nucleus in the bottom middle. The effects of signal modulation are depicted as conceptual phenotypes at the very bottom.

The map was reconstructed by a postdoctoral scientist with experimental expertise on macrophage immunology and cross-validated by having 25 of the incorporated reactions checked independently by two postdoctoral researchers. Out of the combined 50 checks, 32 (64%) were categorized as correct, 10 (20%) were supported with secondary sources or weak evidence, 5 (10%) were incomplete, and only 3 (6%) were partially wrongly or wrongly attributed (see Supplemental Table I). Although the great majority of the sample thus turned out to be in good agreement with the literature, the result shows that further curation by independent reviewers and a comprehensive list of criteria for deciding on ambiguous cases may increase the accuracy.

The map was converted into a network as described in 2Materials and Methods and illustrated in Fig. 1A. Sixteen (42%), 115 (10%), and 1331 (62%) of the miRNA–mRNA, kinase–phosphoprotein, and TF–gene interactions extracted from databases (miRecords plus miRTarBase, RegPhos, and TRANSFAC, respectively) were added in an expansion step (for detailed statistics of the evaluation, see Supplemental Table II), after which a network with 1122 factors and 2705 reactions was obtained (Fig. 2B, Table I). Due to incomplete biological knowledge, 33 factors in the Cytoscape network were disconnected from the main network and were disregarded in the analysis.

To make the map accessible as a research tool, we have set up a Web page at https://vcells.net/macrophage that is free of charge and requires no registration. Visitors can interactively browse both the curated and the expanded map, or download them in several standard compliant formats for import into desktop analysis tools.

The maps come preloaded with expression data sets focusing on the stimulation of Mϕs with LPS, and the Web platform offers the option to mount additional, user-supplied data sets. The current version of the Web platform allows the browsing and mining of 44 data sets, accounting for different experimental conditions, which were grouped in eight categories depending on the biological context of the experiments: 1) infection of primary human macrophages by S. pneumoniae, L. pneumophila, and M. tuberculosis (termed “Lung pathogens” and obtained from GSE77561, GSE61535, and E-MEXP-3805); 2) exposure of macrophages to LPS for different time spans (“LPS stimulation,” GSE4712); 3) LPS stimulation of macrophages after a state of tolerance has been induced with LPS (“LPS tolerance induction,” GSE8621); 4) activation of macrophages with different types of LPS and combinations of LPS with cytokines and immune complexes (“LPS with co-stimulants,” GSE46903); 5) stimulation of macrophages with LPS after knock-down of JunB, a transcriptional modulator of macrophage activation (“LPS stimulation in JunB deficiency,” GSE50542); 6) stimulation of macrophages with LPS after knock-down of JunD, another transcriptional modulator (“LPS stimulation in JunD deficiency,” E-MEXP-3469); 7) activation of macrophages with LPS in the context of tristetraprolin (TTP)-mediated degradation of inflammation-induced mRNA transcripts (“TTP-related mRNA stability after LPS,” GSE28880); and 8) induction of oxidative stress in macrophages via hypochlorous acid (“ROS stress,” GSE15457). To provide a starting point for customized analyses in the expanded map section, each data set is accompanied by a regulatory network that we extracted in Cytoscape with jActiveModules (see 2Materials and Methods for details).

The browser interface (Fig. 3) permits quick access to molecule and reaction annotations, displays expression data with the option to perform side-by-side comparisons, can traverse and manipulate the network to focus on regions of interest, and makes snapshots available as portable network graphics images. Besides data projection, the platform also facilitates ontology highlighting and subnetwork exploration (Fig. 1B). Users are thus able to customize many aspects of the map to generate or illustrate hypotheses about Mϕ signal transduction. The secondary resources (ontologies, regulatory subnetworks) were selected with the goal of maximum relevance for users. Please see the Web platform’s help section for more information.

The example in Fig. 3 elucidates differences in the regulation of IL-1β expression in response to infection with three common lung pathogens. In contrast to Mycobacterium tuberculosis and Legionella pneumophila, Streptococcus pneumoniae strongly increased the IL-1β expression. Through such findings, the interactive map supports researchers in comparing different scenarios and coming up with hypotheses on how pathogen-modulated signaling pathways in Mϕs may lead to the macroscopic phenotype.

In this section, we demonstrate in a case study with the desktop application Cytoscape how the maps can be exploited. Our goal here is to investigate common lung pathogens to find shared vantage points for immunomodulating drugs in Mϕs.

Above the level of individual molecules, small connected regions in the Mϕ signal transduction network are perturbed during an immune response to bacterial infections. To identify those regions, we made use of gene expression data of human primary cells infected with S. pneumoniae, L. pneumophila, or M. tuberculosis as described in 2Materials and Methods. We obtained expression values for 959 of the 1089 factors in the expanded Mϕ network. Upon infection with S. pneumoniae, L. pneumophila, or M. tuberculosis, respectively, 114 (12%), 15 (1.6%), or 84 (8.8%) of the genes in the expanded map were found to be differentially expressed (α = 0.05; data not shown). We separately extracted one significantly perturbed module (i.e., a connected region in the network that shows significant changes in expression; see 2Materials and Methods for details) per infection scenario by using jActiveModules (48). In these subnetworks (data not shown), S. pneumoniae, L. pneumophila, and M. tuberculosis manipulated the surrounding of 396, 239, and 228 molecules, respectively (Fig. 4A). A comparison of the three subnetworks showed that 41 molecules were perturbed in all three investigated scenarios. We extracted them and their interactions and termed the obtained network the regulatory core (Fig. 4B). The molecules in the core are only partially connected because they represent regulated regions that are not necessarily adjacent in the original map.

The visualization of the binary logarithm of the fold change allowed the qualitative comparison of expression level changes between the three scenarios (Fig. 4B). Overall, infection with S. pneumoniae and L. pneumophila upregulated about one half of the factors (24 and 23, respectively) and downregulated the other half, with a similar pattern across factors. For example, the cytokines TNF, CCL5, IL-12 p40, and CXCL10, which are engaged in the amplification of the immune response via autocrine loops or the recruitment of other immune cells (52, 53), show upregulation under both S. pneumoniae and L. pneumophila infection. M. tuberculosis, alternatively, downregulated the expression of most molecules (32 factors) in the core, albeit with much smaller changes compared with the other two bacteria (cf. color legend range).

Therapeutic compounds such as antibiotics can modulate Mϕ signaling (54–59) and thereby potentially antagonize or foster the manipulation by pathogens. Therefore, we investigated the druggability of the pathogen-perturbed regulatory core by employing CyTargetLinker to select interactions from DrugBank. In total, we found 140 drugs targeting 16 of the 41 molecules in the regulatory core (Fig. 5). The targeted molecules are listed in Supplemental Table III together with the respective drugs.

The distribution of drug targets in the regulatory core is not uniform. There are three highly druggable targets (the estrogen receptor [Esr1], TNF, and 5-lipoxygenase) and four intermediate ones (inhibitor of κB kinase B [IKKb], spleen tyrosine kinase [Syk], myeloperoxidase [Mpo], and the IL-12 p40 subunit). The remaining nine targets can be affected by one or two drugs, and 25 molecules were not targeted at all.

Some of the druggable molecules might represent interesting targets for drug repurposing efforts in lung infections. For instance, the estrogen receptor has been implicated as bystander target in anti-inflammatory glucocorticoid therapy (60). Conversely, selective inhibition of myeloperoxidase can control tissue damage in some (61) but not all scenarios (62). In conclusion, the pathogen-perturbed regulatory core can be used to select druggable molecules that represent targets with potential clinical benefit.

This work presents a fully validated and annotated model of Mϕ signal transduction. The signaling map was used to set up a Web platform that offers a network-based analysis tool for the research community, and to demonstrate in a case study how regulated network regions and their druggability can be investigated across experimental conditions.

The widespread lack of comprehensive pathway diagrams has led to efforts providing maps of signaling pathways in dendritic cells (63), of the influenza virus replication cycle (64), and of signal transduction pathways in activated Mϕs in this study. In contrast to previously published Mϕ maps (23, 26), we 1) manually curated literature and database knowledge, 2) stringently annotated all molecules and molecular interactions in a computationally accessible manner, and 3) focused on a specific process, that is, Mϕ activation in the context of immunological processes. A major difference between our study and previous publications is that we used transcriptomic data for the refinement of a pre-existing network, not for its reconstruction (3, 65). Our approach is thus closely related to the way experimental researchers think.

To unravel details of the Mϕ signaling through systems-level approaches, previous studies, for example, analyzed the cross-talk between pathways (66) or created a semantic network of Mϕs for mathematical modeling (67). We followed a similar objective but chose another approach by integrating database knowledge and expression data into our reconstructed network. Compared to those studies, our approach was narrower, as it focused on pathogens associated with lung infection. At the same time, we assume that our results mirror the physiological situation more closely, because we used data derived from primary human cells infected in vitro with bacteria instead of data from cells stimulated with single pathogen-derived factors or host cytokines. However, during construction of the map we strived to find a compromise between pathway details and clarity of presentation. As an example, NF-κB and AP-1 are represented as monomeric proteins instead of p65/p50 and c-Fos/c-Jun heterodimers, respectively. We also want to emphasize that we rely on database knowledge with less stringent curation in the expanded map.

To disseminate the curated and the expanded Mϕ maps, and to offer additional customization and analytical options, we created a Web platform. Its goal is to accommodate both biomedical researchers and bioinformaticians and support data inspection and hypothesis building with visual clues, a strategy that is under active exploration (68). Questions about pathway engagement, signaling cross-talk, and expression differences can be answered quickly and comparatively between experimental conditions. Also, the platform’s scope is not limited to transcriptomics because it can be used to mine any kind of data, including proteomics, phosphoproteomics, enzyme activities, quantitative receptor engagement, and others. Therefore, it provides researchers with a rich environment to investigate and draw conclusions from obtained high-throughput data. We are not aware of any other free Web platform for inflammatory macrophages that offers a comparable range of capabilities.

In the case study, we extracted pathogen-perturbed subnetworks on the basis of transcriptomic data from human Mϕs infected with S. pneumoniae, L. pneumophila, or M. tuberculosis. This shows how combining expression data from different scenarios and integrating it with existing knowledge can promote hypothesis building. Neither the tools—Cytoscape, jActiveModules (69), CyTargetLinker, and DrugBank—nor their combination into a workflow (70, 71) are novel. Still, employed on the new Mϕ map that consolidates current knowledge, our workflow allowed us to pinpoint perturbed network regions. By comparing three data sets from infected Mϕs, we found common points of interference even though the three pathogens differ significantly in terms of their life cycle, cell morphology, and effector molecules. The shared features, which we termed regulatory core, might arise from similarities in the pathogen recognition process, from directed bacterial manipulation of signal transduction, or from both. In particular, the response of regulatory core factors to M. tuberculosis was found to be different in both direction and magnitude from that to L. pneumophila and S. pneumoniae. A thorough comparison will have to take into account the different experimental conditions to segregate true biological distinctiveness from consequences of the particular infection strategy.

The identified regulatory core contains central modulators such as PU.1, or NF-κB and its processing machinery. This finding is in line with the idea that one infection strategy of bacteria is to interfere with highly central host proteins (72) to facilitate the manipulation of host cell functions. For example, the effector proteins LegK1 of L. pneumophila and EstA of S. pneumoniae phosphorylate the NF-κB inhibitor IκB (73) and lead to degradation of IκBα (74), respectively, which both lead to activation of NF-κB. In addition to those central factors, the regulatory core contains molecules of the afferent and efferent signaling branches (e.g., Grb2/SHC2/SOS, the IL-7 receptor, or cytokines such as TNF and IL-12). This reflects the involvement of retrograde feedback in the network and probably complementary bacterial strategies to interfere with host defense mechanisms. We found that this proposition also embodied an entire array of cytokines with upregulation in S. pneumoniae and L. pneumophila infection but downregulation in M. tuberculosis, for example, TNF, CCL5, CXCL10, IL-18, and IL-12 p40.

Besides identifying a common subnetwork of anti-infection response, we illustrated in the case study which factors of the regulatory core are targeted by drugs. The estrogen receptor, a protein that was cloned in 1986 and has a well-studied pharmacological profile, appeared as a prominent druggable target in the regulatory core. It would seem surprising, then, that there are very few studies on the effect and putative benefit of estrogenic compounds in lung infections (75–78). Disregarding the outcomes of these studies, the analysis facilitated by our signaling map highlighted targets whose investigation in the context of lung infections seems plausible (61). Furthermore, the drugs we identified can be assumed to modulate key Mϕ functions in lung infections, even if they have not yet been linked to this disease. Thus, they have the potential to improve the response of Mϕs to pathogens and counterbalance bacterial manipulation, constituting candidates for repurposing and potential administration in the supplemental treatment of bacterial lung infections (79). This would be of particular interest because the bacterial species causing acute pneumonia is rarely identified in current clinical practice. Through coadministration, such therapeutic substances might reduce the required dose of antibiotics or improve their efficacy.

Taken together, we present in this study a fully curated and annotated map of the immunologically relevant signal transduction pathways in inflammatory macrophages. To facilitate customized analyses of Mϕ signaling pathways and comparison of high-throughput data, we provide a Web platform that supports researchers in generating new hypotheses. Using our novel resources, we identified a regulatory core in the Mϕ signal transduction network that is perturbed through infection with L. pneumophila, S. pneumoniae, and M. tuberculosis and thus of importance during the immune response. Drugs targeting these factors can be further investigated as specific immunomodulating drugs that may counteract the pathogenic interference with Mϕ activation.

We thank Ulf Schmitz, Faiz Muhammad Khan, and Shailendra Gupta for helpful discussions. We also thank Roland Lang and Aline Bozec for supplying us with data set candidates for the Web platform.

The authors have no financial conflicts of interest.