Modular analysis.

Categorizing individual genes is a useful tool to obtain insights in relative representations of functional groups in the dataset. A number of independent studies have used the modular classification described by Damian Chaussabel [33], which is based on the assumption that the probability for multiple transcripts to follow a complex pattern of expression across dozens of conditions only by chance is low and such sets of genes should therefore constitute coherent and biologically meaningful transcriptional units. He used microarray based gene expression profiles from a number of diseases and grouped genes according to their concordance in expression profiles [33]. Pubmed searches were done to identify module functional associations, resulting in rather broad, cell type based allocations in particular as a consequence of using PBMCs as source material. Conceptually the modular framework is very elegant and expected to give insights in processes that are strongly enriched within a gene expression data set. However, the number of genes that can be grouped into the modules initially was limited. Recently, a novel version of these modules was published, which is now based on Illumina microarray platforms and contains a larger number of genes (publically available at: http://www.biir.net/public_wikis/module_annotation/G2_Trial_8_Modules). Also the number of modules was greatly expanded in this new version compared to the previous version [33]. All 409 genes in our current data set were annotated according to the latest version of the modules (web publication date August 2012) (Table S1). 371/409 (90.7%) genes in the combined dataset were retrieved and assigned to modules (Table S1). However, the majority of genes (195/371 or 52.6%) were allocated to modules without functional assignment (undetermined/not determined), resulting in only 176/409 (43.0%) of genes with a functional classification (Table S1, Figure 2). Two modules are highly represented in the dataset: 68/176 (36.6%) of the allocated genes belong to the inflammatory signature and 36/176 (20.5%) genes belong to the interferon related signature. Genes identified in more than 1 independent studies also predominantly represented patterns of inflammation (9/39; 23%) and interferon related pathways (7/39; 17.9%) (Figure 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Functional classification of individual genes identified by gene expression analysis on TB patients.

Categories have been based on combined output from Ingenuity and GSEA software modules and may include multiple canonical pathways or cell processes. Myeloid cells includes the following canonical pathways: role of macrophages, fibroblasts and endothelial cells in rheumatoid arthritis; Fcg Receptor mediated phagocytosis in macrophages and monocytes; role of pattern recognition receptors in recognition of bacteria and viruses; IL12 signaling and production in macrophages; Dendritic cell maturation; production of Nitrox Oxide and Reactive Oxygen Species in Macrophages; Toll like receptor signaling. T cells includes: T cell receptor signaling; CD28 signaling in T helper cells; iCOS-iCOSL signaling in T helper cells. B cells includes: B cell receptor signaling; PI3K signaling in B lymphocytes. Interferon related pathways include: Interferon signaling, role of jak1, jak2 and tyk2 in interferon signaling, role of PKR in interferon induction and antiviral response. Inflammation includes: IL-8 signaling; NF-kB signaling; altered T cell and B cell signaling in Rheumatoid Arthritis; systemic lupus erythematosus signaling; chemokine signaling; IL-6 signaling. TREM1 includes specifically TREM1 signaling and mitochondrial dysfunction also only contains mitochondrial dysfunction. Finally, hematopoiesis includes: erythropoietin signaling; IL-3 signaling; FLT3 signaling in hematopoietic progenitor cells; prolactin signaling; HGF signaling.

https://doi.org/10.1371/journal.pone.0073230.g002

The modular analysis as described above thus hints towards a prominent role for inflammation and interferon signaling in TB disease, as was also described by most individual studies. The disadvantage of the modular data analysis is the limited relation of gene sets within a module to specific functional capacities, the relative lack of sensitivity for small changes in gene expression profiles, mostly because the data sets used to annotate the modules were all derived from PBMCs, with their particular cellular composition, and did not include specialized cell types, or specifically activated pathways. For example, monocytes and macrophages only represent a low proportion of PBMCs (5–10%) and thus specific monocyte activation pathways may not be recognized by the modular annotation.

Molecular Interactions.

Ingenuity based pathway analysis revealed highly significant overlap with 92 canonical pathways with p<0.001 and a total of 196 canonical pathways p<0.05. To select the most important pathways within our dataset we defined and applied filter criteria on the 92 canonical pathways with the highest significance. First, pathways were excluded if they were smaller than 50 genes; secondly, pathways were excluded if fewer than 10 out of the total of 409 gene dataset mapped in that pathway; and thirdly, we only considered pathways in which our genes represented at least 10% of the total gene-set in that pathway. This resulted in 36 pathways that were significantly abundant in our data set of 409 genes (Table 3, Figure 2).

Biological pathways.

Gene Set Enrichment Analysis (GSEA) [34] determines whether there is significant overlap or enrichment of genes in published biological pathways with genes in a query list. To evaluate the degree of enrichment the GSEA method calculates an Enrichment score (ES) and False Discovery Rate (FDR). We tested our set of 409 genes against the Molecular signature Database C2 collection (4850 gene sets) collected from various sources such as online pathway databases, publications in PubMed, and knowledge of domain experts. In the GSEA analysis 230 gene sets were returned with a significant enrichment score (ES) although only 15 of these gene sets were significant at an FDR of <5% (Table 4, Figure 2).

TREM1 signaling pathway.

The Ingenuity canonical pathway overlapping the combined dataset with highest significance level was ‘TREM1 signaling’ (Table 3). Genes that comprise the TREM1 canonical pathway are depicted in Table 5, in addition genes identified in our dataset are indicated. The dataset overlaps with TREM1 signaling directly, but also has significant overlap with TLR and downstream TLR signalling as well as with effector cytokines secreted following TREM1 activation. This is a novel finding, not detected in any of the individual published studies. TREM1 or ‘Triggering receptor expressed on myeloid cells 1’ is a member of the Ig superfamily and is predominantly expressed on myeloid cells. Cell surface expression of TREM1 is increased upon cellular activation, e.g. by LPS and other microbial products [35]. Expression of TREM1 results in amplification of neutrophil- and monocyte-mediated inflammatory responses, by increased cytokine production and upregulation of cell activation markers, as seen in bacterial and fungal infections. TREM1 signaling pathways interact with signaling pathways downstream from the IFNaR1/2 and from TLRs [36], [37]. Cross-talk between TREM1 and TLR2 and TLR4 signaling cascades have been demonstrated, and potentially also other TLR (TLR 3, 5, 7, 8, and 9) and possibly NLR signaling pathways are influenced by TREM1 signaling [37]. TREM1 and TLR2 signaling pathways can synergize at the level of cytokine production [37]. TREM1 plays a crucial role in fine-tuning of the inflammatory response by amplifying or dampening TLR induced signals, and is known to tune the septic response in order to facilitate efficient clearance of the pathogen without damaging the host [36]. The ligand of TREM1 is currently unknown, although it has been speculated that it may be expressed on pathogens, thereby directly activating TREM1 signaling [37]. Peptidoglycans on gram positive bacteria and endotoxins on gram-negative bacteria are candidate ligands, and mycobacterial peptidoglycans might also serve as TREM1 ligands. This remains to be determined but would be an interesting explanation for the increased TREM1 signaling observed in patients with active TB disease.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. TREM1 canonical pathway.

https://doi.org/10.1371/journal.pone.0073230.t005

A soluble variant of TREM1 has also been identified (sTREM1). This results from the shedding of cell surface TREM1 by metalloproteinases, and is thought to negatively regulate TREM receptor signaling through neutralization of the ligands [35]. In patients with pulmonary TB sTREM1 has been detected in sputum [38], but levels could not discriminate pulmonary TB from community acquired pneumonia caused by other pathogens. In patients with pleural effusions, sTREM1 levels could discriminate infectious, including TB, from non-infectious causes and the levels of pleural fluid sTREM1 appeared a useful tool to discriminate Mtb infection from malignancies [39]. An independent study on pleural effusions showed a similar increase in sTREM1 levels in TB effusions compared to non-infectious effusions [40]. In addition, surface TREM1 was assessed on cells within the effusions and was undetectable in effusions from patients with TB pleuritis [40], however cell numbers and cellular composition were not described. Similarly, in BAL (bronchoalveolar lavage) cells from patients with pulmonary TB no significant increase in cell-surface TREM1 expression was observed compared to BAL cells from non-infectious controls, whereas BAL cells from patients with other pulmonary infections did express increased TREM1 levels on macrophages and neutrophils [41]. Thus, TREM1 related signaling seems important during active TB disease; although cellular TREM1 expression levels may not be changed, levels of soluble sTREM1 are increased at the site of disease.

Myeloid lineage cells.

In addition to TREM1 signaling, a number of other pathways associated with myeloid cell function or activation were prominently identified in the Ingenuity analysis (Table 3). These included: ‘Fcγ receptor-mediated phagocytosis in macrophages and monocytes’; ‘role of pattern recognition receptors in recognition of bacteria and viruses’; ‘role of macrophages, fibroblasts and endothelial cells in Rheumatoid Arthritis’; ‘Dendritic cell maturation’; Production of Nitric Oxide and reactive oxygen species in macrophages'; ‘Toll-like receptor signaling'; ‘IL-12 signaling and production in macrophages’. In Table 3 these are grouped together into a ‘myeloid’ category, which was then applied to allocate individual genes into functional categories in Table S1. Increased expression of genes associated with Toll-like receptor signaling [18], [19], [26] and Fcγ Receptor signaling [26], [29] were previously reported in patients with active TB disease. Grouping of genes according to the canonical pathway based groups resulted in categorization of 107 out of the total 409 genes in the dataset. The myeloid signature thus identified is strong and also more clearly represented than in any of the individual studies: 56/107 (52.3%) genes that were grouped into canonical pathways were in the myeloid category, although the majority overlaps with other categories as well, including TREM1 signaling.

As with the Ingenuity based pathway analysis, most of the GSEA gene sets we identified represented a myeloid biased gene signature and an inflammatory immune response. The gene sets SEKI_INFLAMMATORY_RESPONSE_LPS_UP, ICHIBA_GRAFT_VERSUS_HOST_DISEASE_35D_UPICHIBA_GRAFT_VERSUS_HOST_DISEASE_D7_UP, JISON_SICKLE_CELL_DISEASE_UP and REACTOME_IMMUNE_SYSTEM overlap with each other and comprise many genes associated with myeloid cells (Table 4). The dominance of these gene signatures reflect the findings of the studies the signatures were derived from. Cliff et al [31], Berry et al [17] and Ottenhoff et al [32] all highlight an increased myeloid cell inflammatory responses in active TB patients when compared to controls.

Network analysis revealed a strong contribution of genes associated with myeloid cells in the most dominant network (figure 1), however many genes overlapped with more than one process (Table S1, Figure 1). These comprised signaling pathways around Akt, MYD88 and the NFκB complex as central players in the network, indicating an important role for inflammation, phagocytes and professional antigen presenting cells including macrophages. Indeed the early microarray studies performed in TB patients indicated that genes with roles in inflammation and immunity were most abundantly expressed [18], [20], [26], [32] and most inflammation related genes differentially expressed were derived from monocytes [17], [19], [29], [32]. Interestingly, Cliff et al. speculated that some of the genes specifically upregulated during TB disease can be expressed by macrophages and DCs but not by monocytes, and thus may reflect activated APCs, possibly cells trafficking between lungs and lymphoid tissue via the blood [31]. This could also explain the relatively high myeloid gene expression observed in TB patients and in particular the involvement of TLR and Fc receptor induced signaling, since these processes would normally be expected to occur at the site of disease rather than in the circulation. The pathways associated with these myeloid cells point towards a strong role for direct pathogen related processes, even in the circulation (not the disease site in TB).

T-cells and B-cells.

Although the emerged profiles are dominated by inflammatory and myeloid gene signatures there was one gene set associated with CD4 T Cell/follicular helper T cell activation (PICCALUGA_ANGIOIMMUNOBLASTIC_LYMPHOMA_UP) [42] (Table 4). CD4 T cells are important for the control of Mtb infection and TB disease, proliferate and are activated in response to antigens from M. tuberculosis. It is possible that these follicular helper T cells are contributing to the formation of B cell follicles during active TB disease. Studies of human TB granulomas have identified B cell follicle structures which may contribute to the immune response in TB, and activated B cells have been found in granulomas of nonhuman primates infected with Mtb [43], [44].

As expected also canonical pathways associated with B cell function (as previously reported by [19], [31]), T cell function (previously reported by [31]) and more general inflammation processes were identified by Ingenuity analysis (Table 3). The potential functional implications of B cells in TB disease have been debated for decades, in particular because intracellular pathogens are considered to be sequestered from circulating antibodies. However, functional significance of B cells in TB disease has been demonstrated in mouse models: B cell deficient mice appear more susceptible to TB [45], and Fc-receptors play a role in protection. Furthermore, in addition to the presence of B cells in human TB lesions as described above, the expression of human FcγR1 is a consistent and strong component in TB biomarker signatures [19], [29], [30]. These observations suggest that B cells may play a hitherto unappreciated role in immunity in TB. This is further supported by recent results that suggest that also intracellular binding of antibody to pathogens can take place via a cytosolic Fc-receptor called TRIM21 [46], [47]. Pathogen bound antibody triggering of TRIM21 subsequently stimulated transcription factor pathways including NF-kB, IRF7 and others, resulting in immune activation and inflammatory signals. This may open up an interesting new angle for the potential role of B cells and in particular antibodies in the combat against TB, which should be explored in more detail. Interestingly, our network based analysis of genes expressed in peripheral blood from TB patients does include TRIM21 (Figure 1). Gene expression of TRIM21 was detected in cohorts of TB patients compared to healthy controls [19] (Table S1) and was grouped close to Fc Receptors and its defined signaling molecule IRF7 in the Ingenuity network analysis.

Hematopoiesis.

A process which was not expected but clearly represented in the canonical pathway analysis was hematopoiesis, represented by ‘Erythropoietin signaling’; ‘IL-3 signaling’; ‘FLT3 signaling in hematopoietic progenitor cells’; ‘prolactin signaling’ and ‘HGF signaling’. This indicates that in the blood of patients with TB disease active remodeling apparently is ongoing, either with renewal of hematopoietic cells at a possibly increased rate compared to non-TB controls or with cells emerging from the bone marrow and on their way to the site of inflammation (mostly the lung in TB patients). Recently it has been demonstrated that Mtb may hijack mesenchymal stem cells and may survive in the CD271+ stem cells for a long period after successful treatment of pulmonary TB, hiding from the immune system [48]. This suggests that stem cells in the bone marrow may be active players in TB disease, revealing a new cell type involved in TB pathogenesis. Another interesting link between mycobacterial infections and stem cells is the recent observation that Mycobacterium leprae is capable of reprogramming host gene expression in adult Schwann cells and induces a dedifferentiation program towards stem cell like cells (SLC) [49]. These SLC can subsequently redifferentiate into end stage tissue cells, and also release bacteria onto macrophages locally in tissues [49]. It remains unknown whether Mtb also possesses the capacity to reprogram host cells by activating pathways characteristic for hematopoiesis, but this may be an interesting possibility that fits with the observed changes in gene expression profiles. Interestingly, the vast majority of genes categorized as ‘hematopoiesis’ is also associated with several myeloid functions (Table 3).

Type I Interferon Signaling.

Berry et al [17] were the first to describe that Type I interferon signaling was increased in TB patients. The pathway MOSERLE_IFNA_RESPONSE was amongst the 15 most significant pathways identified by GSEA in active TB disease (Table 4). Surprisingly, interferon α/β related signaling pathways were not amongst the top pathways represented by our dataset in Ingenuity (Table 3), despite it being detected with high significance in several individual studies. Interferon signaling reached statistical significance (p = 0.000026), but was lost during application of gating criteria, due to the small number of genes in the pathway (<50) and the small number of genes overlapping with our dataset (<10). However, because of the interest in IFNα/β signaling in TB disease, we added a category IFNα/β to table 3 based on all 3 canonical pathway datasets related to type I interferon signaling in Ingenuity ‘interferon signaling’; ‘role of JAK1, JAK2 and TYK2 in interferon signaling’ and ‘role of PKR in interferon induction and antiviral response’. IFN signaling represented 13/107 (12.1%) genes in our dataset and was thereby not the most dominant pathway (figure 1, Table S1). Signaling through IFNα/β seems an important pathway associated with TB disease in most independent studies, although many molecules involved in downstream IFNα/β pathways are not uniquely involved in transmitting signals derived from IFNα/β receptors but may also be critical components in other pathways.

Type I interferons are classically known as early response molecules in the context of viral infections and are well known immune modulators. Type I interferon signaling in mycobacterial infections, as shown in progressive leprosy disease, may contribute to disease pathogenesis or protection, depending on the balance between IFNα/β and IFNγ [50], in which IFNα/β may actively inhibit IFNγ signalling. Interestingly in virus infection models the IFNα/β response mediated enhanced antiviral activity during acute infection but exerted strong immunomodulatory effects during chronic infection [51]. Both TB and leprosy are long term, chronic infections; in both diseases an increased expression of IFNβ induced genes has been observed, suggesting its potential involvement in tissue damage and inflammation, and in regulating adaptive IFNγ responses. It may be speculated that patients with latent Mtb infection have strong IFNγ induced responses which are downregulated by IFNα/β during reactivation from latent to active TB disease [4], [50].

Macrophages or monocytes infected with Mtb produce and secrete type I interferons [52]. Mtb strains with increased virulence induce increased IFNA mRNA in lungs of infected mice [53], and lower local levels of pro-inflammatory cytokine mRNAs (IL-6, IL-12, TNFα and IFNγ). The effect of IFNα/β was further investigated by administration of IFNα to mice following infection with Mtb, resulting in increased pulmonary bacterial loads and reduced survival of the mice [53]. In vitro, addition of type I interferons to BCG infected macrophages resulted in increased outgrowth of BCG, suggesting direct effects of type I interferons on infected cells in favor of the mycobacteria [54]. The increase in type I interferons and IFNα/β signaling during active TB disease may therefore reflect IFNα/β production by infected monocytes/macrophages and facilitate and maintain rather than clear chronic infection.

Fibrosis.

Statistically, using GSEA analysis, the greatest overlap of our gene-set was observed with the gene set MARKEY_RB1_ACUTE_LOF_UP, a gene set expressed in fibroblasts [55] (Table 4). Fibroblasts secrete extracellular matrix molecules, while extracellular matrix destruction is necessary for the growth and persistence of Mtb: Mtb is a potent inducer of the metalloproteinases (MMPs) which destroy extracellular matrix [56]. MMP9 and MMP25 are found among our gene list of 409. MMP expression is enhanced by G protein coupled receptor signalling and type 1 interferon – both of which are represented in our GSEA analysis (REACTOME_GPCR_LIGAND_BINDING and MOSERLE_IFNA_RESPONSE). MMP9 is the most abundant of the MMPs and has been found to correlate with disease severity in TB patients [57]. The neutrophil specific matrix metalloproteinase MMP25 degrades substrates found in fibroblasts and may enhance the phagocytic removal of neutrophils from inflammatory sites [58].