• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. May 15, 2012; 109(20): 7853–7858.
Published online Apr 30, 2012. doi:  10.1073/pnas.1121072109
PMCID: PMC3356621
Medical Sciences

Common patterns and disease-related signatures in tuberculosis and sarcoidosis

Jeroen Maertzdorf,a,1 January Weiner, 3rd,a Hans-Joachim Mollenkopf,a TBornotTB Network,2 Torsten Bauer,b Antje Prasse,c Joachim Müller-Quernheim,c and Stefan H. E. Kaufmanna,1
aDepartment of Immunology, Max Planck Institute for Infection Biology, 10117 Berlin, Germany;
bLung Clinic Heckeshorn, Helios Klinikum Emil von Behring, 14165 Berlin, Germany; and
cDepartment of Pneumology, University Medical Center, 79106 Freiburg, Germany


In light of the marked global health impact of tuberculosis (TB), strong focus has been on identifying biosignatures. Gene expression profiles in blood cells identified so far are indicative of a persistent activation of the immune system and chronic inflammatory pathology in active TB. Definition of a biosignature with unique specificity for TB demands that identified profiles can differentiate diseases with similar pathology, like sarcoidosis (SARC). Here, we present a detailed comparison between pulmonary TB and SARC, including whole-blood gene expression profiling, microRNA expression, and multiplex serum analytes. Our analysis reveals that previously disclosed gene expression signatures in TB show highly similar patterns in SARC, with a common up-regulation of proinflammatory pathways and IFN signaling and close similarity to TB-related signatures. microRNA expression also presented a highly similar pattern in both diseases, whereas cytokines in the serum of TB patients revealed a slightly elevated proinflammatory pattern compared with SARC and controls. Our results indicate several differences in expression between the two diseases, with increased metabolic activity and significantly higher antimicrobial defense responses in TB. However, matrix metallopeptidase 14 was identified as the most distinctive marker of SARC. Described communalities as well as unique signatures in blood profiles of two distinct inflammatory pulmonary diseases not only have considerable implications for the design of TB biosignatures and future diagnosis, but they also provide insights into biological processes underlying chronic inflammatory disease entities of different etiology.

Gene expression in peripheral blood cells from tuberculosis (TB) patients and healthy controls, both latently Mycobacterium tuberculosis-infected and uninfected, has been profiled by several groups in the recent past (17). Identified expression profiles indicate chronic activation of the immune system, with a marked activation of IFN signaling (3), proinflammatory signaling through the JAK-STAT pathway (5, 6), and elevated expression of Fc γ-receptors and their downstream response elements (2, 4). Although these biosignatures have been identified by several independent groups and possess the potential to discriminate latently M. tuberculosis-infected healthy individuals from active TB patients, the question remains whether these gene expression signatures are specific for TB or shared, at least in part, with diseases of similar pathology but distinct etiology.

To address this question, we have conducted a comparative analysis of blood profiles in patients with active pulmonary disease manifestation of TB and sarcoidosis (SARC). The rational for choosing pulmonary SARC for such a comparison is the remarkable similarity in immune activation with active TB, suggesting a shared underlying pathophysiology (8). Although SARC is generally considered a noncommunicable disease of unknown etiology, patients with pulmonary involvement present with histological and clinical symptoms highly similar to TB, including granulomatous structures in the lung (8, 9). In this study, we compared global gene and microRNA (miRNA) expression in blood cells and cytokine abundance in serum between TB and SARC patients as well as healthy individuals (Dataset S1).

The communalities in blood profiles from pulmonary TB and SARC patients as described here underline the preponderance of common inflammatory processes in similar but distinct disease conditions. In particular, we consider the striking similarity in expression profiles between both diseases of critical importance for understanding underlying mechanisms of pathology. This consideration could greatly benefit the definition of true biosignatures for TB and SARC and the development of new array-based diagnostic tools, which discriminate not only disease status from healthy individuals but also between different diseases of similar pathology. We conclude that biomarker profiles not only contain disease-specific signatures but also provide insight into common biological processes shared by different diseases.


Differential Gene Expression.

We have interrogated the transcriptomes of peripheral blood cells from active TB and SARC patients with the transcriptomes of healthy individuals as comparators. A large number of transcripts were differentially expressed between patients of both disease groups and healthy individuals (at significance level q < 0.01, where q is the P value adjusted for multiple testing), with a remarkably identical direction of expression (up- or down-regulated in diseased vs. healthy individuals). Only four significantly regulated genes showed a different direction between the patient groups. Intriguingly, other than these common expression patterns, a significant number of disease-specific genes were identified for TB as well as SARC (Fig. 1).

Fig. 1.
Common and differential expression of genes, miRNAs, and serum analytes. Venn diagrams show the number of significantly (q < 0.01) differentially expressed genes, miRNAs, and cytokines (P < 0.01) between the disease groups and healthy ...

Common Signatures.

Common profiles of differentially expressed genes between diseased and healthy individuals were analyzed and compared with previously identified TB and SARC signatures. The recently published TB signature identified in the work by Berry et al. (3), which includes a dominant IFN-inducible gene profile, showed a remarkable similarity to differentially expressed genes in TB patients in this study. However, virtually the same set of genes was observed to be significantly regulated in SARC patients, indicating that this signature is not specific for TB. Similarly, the TB-specific 86-transcript profile published in the work by Berry et al. (3) showed a 77% overlap of significantly differentially expressed genes in our TB cohort, whereas the group of SARC patients showed an even higher 84% overlap with this profile (Table 1). Similarly, in our cohort, TB and SARC patients showed a highly similar up-regulation of IFN-signaling and -inducible genes, emphasizing the profound similarities in blood expression profiles between TB and SARC.

Table 1.
Expression overlap with previously identified TB-related signatures

Particularly telling is the comparison with the TB pathway, which was recently included in the KEGG database. The genes in this Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway (KEGG ID hsa05152; i.e., genes with established significant relation to TB) revealed highly similar differential expression patterns in both diseases compared with healthy individuals. The same was observed for genes involved in several other KEGG pathways that were previously shown to include a significant number of genes that are differentially regulated in TB compared with healthy individuals (6). These pathways include genes in systemic lupus erythematosus, complement and coagulation cascades, toll-like receptor signaling, and Fc γ-receptor–mediated phagocytosis (Table 2). The striking similarities in these biological processes underline the shared inflammatory processes underlying both disease conditions.

Table 2.
Expression overlap with previously identified biological pathways

Disease-Specific Genes.

Our analysis revealed a total of 1,442 known genes that were differentially expressed between TB patients and healthy controls but not between SARC and controls. Likewise, we identified 2,451 SARC-specific genes that were different from TB patients and healthy controls (Fig. 1) (all significant at q < 0.01). Gene ontology (GO) analysis of TB-specific genes revealed a highly significant enrichment of genes involved in translational activity of structural proteins (Dataset S2, sheet 1). This enrichment was mainly accounted for by genes that showed increased expression over healthy controls, suggesting an increased protein turnover in blood cells from TB patients, which was not observed or less significant in SARC patients. For the SARC-specific genes, GO enrichment was less obvious. Among the strongest differentially expressed genes (log2 fold change > 0.75 or < −0.75), an association with increased regulation of mast cell activation was observed (Dataset S2, sheet 2), which was indicated by decreased expression of negative regulators cannabiniod receptor (CNR) 2 and heme oxygenase (HMOX) 1 and increased expression of positive regulators Fc epsilon receptor (FCER) 1A and membrane-spanning 4-domains (MS4) A2. Other striking features within the SARC-specific gene list were the pronounced up-regulation of gene matrix metallopeptidase 14 (MMP14) and a decreased expression of genes cluster of differentiation (CD) 247, CD8A, and CD8B, which may reflect abnormally low T-cell counts in a substantial number of SARC patients as described previously (11).

Differential Expression Between TB and SARC.

In the previous section, we compared the differential expression profiles of patients with both TB and SARC with healthy controls. Zooming in on differences between TB and SARC, we also evaluated differentially expressed genes between both disease groups directly, irrespective of expression levels in healthy individuals. This analysis revealed a highly significant enrichment of genes involved in the electron transport chain within the inner mitochondrial membrane. Genes in this functionally enriched group showed an overall elevated expression in blood cells from TB patients that manifested stronger than in SARC patients (Fig. S1). Other GO clusters with significant enrichments included genes involved in translation and cellular responses to reactive oxygen species (Dataset S3, sheet 1).

Among the top-ranking genes with a more than twofold higher expression in TB over SARC, enrichment analysis revealed a strong involvement in defense response genes (Dataset S3, sheet 2). These genes include the microbicidal defensins alpha (DEFA) -3 and -4, the hypoxia-activated genes lactotransferrin and lipocalin 2, and the serine protease cathepsin G (CTSG). The markers likely underlie the disease-specific clinical manifestations of TB.

Classification of Patients.

Having shown that, despite high similarities in gene expression, distinctive patterns between TB and SARC can be singled out, we went on to further verify genes that contribute to the distinctive patterns. To this end, we determined the discriminative power of such genes using the supervised machine learning algorithm random forest (RF) analysis (12). RF allows us to rank the predictor variables (i.e., genes) by their relative importance in successful discrimination of study groups. The predicted genes possessed high accuracy in discriminating each donor sample, with only one TB patient and one SARC patient each falsely predicted to belong to the counter group (Table 3 and Dataset S4, the most discriminative genes are shown). A heat map based on the most discriminative genes illustrates the differential expression profiles between both disease groups and healthy controls (Fig. 2). The color patterns in this heat map also reflect the variability of gene expression between individuals in each study group. It is this variation in expression levels between individuals that can lead to misclassifications of patients as belonging to the other disease group by RF analysis. Fig. S2 shows the different error rates in the classification depending on the number of discriminatory genes used in the computational models.

Table 3.
RF analysis
Fig. 2.
Expression differences between disease groups and controls. Hierarchical clustering illustrating differences in gene expression between TB, SARC, and healthy controls (CTRL) based on the top discriminating genes as identified by RF analysis. Color coding ...

Serum Analytes.

Circulating cytokines in the serum reflect overall immune activity of white blood cells. We measured a total of 48 serum analytes in the sera of the TB and SARC patients and healthy controls to define the general immune activity in each patient and differences between the diseased and healthy study groups (Fig. 1). Sera from TB patients showed an increase in proinflammatory cytokines (IFNγ and IL-12) compared with both SARC and control individuals, and they had a higher abundance in factors with chemotactic and activating effects on monocytes, macrophages, and neutrophils [GM-CSF, macrophage migration inhibitory factor (MIF), and VEGF]. In contrast, sera from SARC patients showed decreased levels of factors with chemotactic and stimulatory activity on granulocytes [chemokine (C-C motif) ligand (CCL) 27, RANTES (CCL5), stem cell growth factor beta (SCGFb), and leukemia inhibitory factor (LIF)]. Decreased levels of MCP-1 in both disease groups may reflect an antiinflammatory feedback to dampen recruitment of proinflammatory immune cells (Fig. 3). An analysis of cytokine levels and expression of their respective genes did not reveal a significant correlation between gene and protein expression patterns in blood. This finding could be because of the fact that cells in granulomas as sites of disease manifestation (and not blood leukocytes) are a major source of cytokines.

Fig. 3.
Serum analytes in diseased and healthy individuals. Cytokine and chemokine levels in serum from diseased and healthy individuals. Bars indicate mean with SEM. *Significant difference between indicated groups at P < 0.01 (one-way ANOVA with Tukey ...

miRNA Profiles in TB and SARC.

Gene transcription and translation is a complex and tightly regulated process. Among such regulatory processes, miRNAs play a decisive role by silencing or destabilizing their target RNAs. We, therefore, also measured expression levels of miRNAs in the blood of eight TB and eight SARC patients from the study group and eight healthy controls. Similar to the gene expression patterns, differential expressions of miRNAs between the two diseases and the healthy controls were comparable, with an overlap of 145 miRNAs between the TB and SARC groups (Fig. 1). Differential miRNA levels between TB and SARC were limited to only four miRNAs at significance q < 0.01 (i.e., miRNAs hsa-miR-182, miR-355, miR-15b*, and miR-340) (Dataset S5 has the full list of miRNAs). The most strongly up-regulated miRNA in both diseases was miR-144, an miRNA that has been identified as a central regulator of cellular response to oxidative stress (13). Its passenger strand miRNA-144*, which was also significantly elevated in both TB and SARC in this study, was recently found to be elevated in TB and modulate T-cell cytokine production (14).

For the identification of differential expression of predicted target genes, we focused on the most strongly up- (log2 fold change > 2) and down-regulated (log2 fold change < −1.2) miRNAs in both diseases. To limit the number of predicted target genes to the most likely candidates, we used an intersection of predicted targets from three different prediction programs (targetScan, picTar, and miRanda) available through the starBase platform (15) (http://starbase.sysu.edu.cn/). Considering the destabilizing action of miRNAs on mRNA molecules, we expected that up-regulated miRNAs would correspond to decreased expression levels of their respective target genes. In contrast, a large proportion of predicted target genes of the most up-regulated miRNAs showed the same direction of change (i.e., predicted targets from up-regulated miRNAs showed increased gene expression in disease vs. control). Analysis of predicted target genes with a negative expression correlation with miRNAs revealed no significant enrichment in GO functions.

Correlations in Blood Profiles.

As a final step, we created clusters of genes and miRNAs with correlating expression levels, which could indicate a possible functional relationship and reveal particular biological processes involved in pulmonary diseases. Clusters of correlating genes and miRNA were calculated on expression levels of TB and SARC patients to identify such disease-related processes, exploiting the intragroup variability of expression levels of both gene transcripts and miRNAs.

Different correlation settings were tested to obtain a manageable number of clusters, with considerable numbers of genes and miRNAs per cluster. Lower cutoff values resulted in many more clusters with very few genes each, whereas a higher cutoff generated few clusters with many genes, which in our view, does not accurately reflect possible biological correlations. One of the resulting clusters, containing the highly up-regulated miRNA-144, is shown in Fig. 4. The results reveal that, within this cluster, miRNAs and genes do display a negative expression correlation. Whereas in the previous section, no clear link between up-regulated miRNAs and down-regulation of their predicted target genes could be identified, this approach revealed a cluster of up-regulated miRNAs (including miRNA-144) and mostly down-regulated genes for TB and SARC. Using a higher cutoff to increase cluster size resulted in an enrichment of genes involved in lymphocyte migration and apoptosis. GO enrichments of genes within this cluster at cutoff = 0.95 are listed in Dataset S6.

Fig. 4.
Correlation clustering of genes and miRNAs. The figure illustrates the correlations in expression levels between gene transcripts (Top) and miRNAs (Middle) in the cluster containing miRNA-144. Bottom show the in-between correlations of miRNAs and genes. ...


Blood cell transcriptional profiling has provided insights into pathological conditions in TB (37). As shown here, however, the identified profiles—at least in large part—more likely reflect activation of common proinflammatory pathways.

In an attempt to identify commonalities and differences in blood signatures between TB and other pulmonary inflammatory diseases, we describe here a direct comparison of gene expression profiles in the blood cells of active pulmonary TB and SARC patients. Based on profound similarities in clinical symptoms and histopathologic patterns (8, 9), we argued that pulmonary SARC would be the closest matching disease to active pulmonary TB.

Transcriptomes in peripheral blood cells from both disease groups showed remarkably high similarity in differential expression compared with healthy controls. This resemblance in expression reflects the critical role of general inflammatory building blocks in the two diseases. Comparison with the KEGG TB pathway is instructive, because it shows very similar differential expression patterns not only in TB but also SARC patients. The same was observed for genes within several other pathways, showing similar patterns in both TB and SARC.

Thus, the TB signature published in the work by Berry et al. (3), which includes a dominant IFN-inducible gene profile, showed a comparable pattern of differentially expressed genes in TB as well as SARC patients. Although the IFN signature described may be characteristic for TB, it is not unique to this disease. Similarly, the TB-specific 86-transcript profile presented in the same work (3) showed a highly similar pattern within both TB and SARC patients. In agreement with this observation, an indirect comparison recently described (16) indicated resemblance of the TB-specific signature with findings in SARC patients.

Other than these common expression patterns, however, a significant number of independent disease-specific gene patterns were identified for both TB and SARC. TB profiles showed increased metabolic activity and protein turnover compared with SARC patients, comprising an enrichment of up-regulated genes involved in translation and a higher up-regulation of genes involved in mitochondrial oxidative phosphorylation and translational activity. Another functional group of genes enriched in TB, including Kruppel-like factor 2 (KLF2), is involved in responses to reactive oxygen species. Hypoxia and/or bacterial products reduce the expression of KLF2, an essential regulator of the innate immune system, while simultaneously inducing hypoxia-inducing factor 1α (17). In concordance with these findings, we observed a profound down-regulation of KLF2 and highly significant up-regulation of hypoxia-inducing factor 1α (Dataset S3). Both of these effects were most pronounced in the TB group. TB patients, furthermore, displayed significantly higher antimicrobial defense responses, consistent with the notion that TB disease is caused by an actively replicating pathogen, whereas SARC is considered a noncontagious disease.

However, SARC patients also displayed a significant number of unique features, with the most pronounced being increased expression of MMP14. This report links MMP14 to SARC, which also is distinct from its expression in TB and healthy individuals. Hence, the membrane-anchored metallopeptidase MMP14, with important roles in tissue development, tumor invasion, and angiogenesis (18), is a potential marker in differential diagnosis of TB and SARC. We consider it likely that these disease-specific patterns impact on the distinct clinical manifestations, whereas the shared signatures dictate the common pathways underlying inflammatory conditions shared by both diseases.

Classification of active TB patients and healthy individuals (with or without M. tuberculosis infection) based on gene expression in blood cells has been described in previous publications (17). The very high overlap of differentially expressed genes between TB and SARC within these published TB-related signatures, as shown in Tables 1 and and2,2, indicates that these signatures do not reliably distinguish between the two diseases. Given this strong similarity between the expression profiles of TB and SARC described here, we tested whether similar classification methods would be able to discriminate the two types of patients based on a distinct set of genes. Applying random forest analysis in these data revealed that study participants could, indeed, be classified as belonging to either of the disease groups with a high accuracy, despite their similar gene expression profiles.

On the level of miRNA expression, significant differences in expression between healthy and diseased individuals were observed. However, both TB and SARC revealed highly similar profiles. This finding is consistent with the view that miRNAs are primarily responsible for fine tuning of responses rather than on/off switch signals (19). Differential miRNA levels have been described recently in peripheral blood mononuclear cells (PBMC) (20) and serum (21) from TB patients. As key regulators in gene expression, miRNAs may play an important role in modulating biological pathways that are affected by or participate in inflammatory conditions. Because of our limited knowledge about the contribution of particular miRNAs to biological processes and disease, analytical tools are still missing that allow for direct enrichment or pathway analysis of miRNAs. In an attempt to link miRNAs to certain biological processes, we clustered miRNAs and genes based on their expression levels. Clusters of genes and miRNAs with correlating expression patterns can reveal functional relations to biological processes underlying disease pathology. As illustrated in Fig. 4, the clustering approach applied herein revealed several strong correlations between expression levels of miRNAs and genes, thus emphasizing functional interactions in general biological pathways including transcription, translation, and pulmonary inflammation underlying TB and SARC. This finding allowed linkage of differentially expressed miRNAs with genes involved in immune-related processes in both TB and SARC (Dataset S6).

Although we observed a broad variation in serum cytokine levels, elevated concentrations of several mediators in TB point to slightly higher proinflammatory responses compared with SARC and controls. However, SARC patients presented with lower abundance of T cell-derived migration factors acting on monocytes and granulocytes compared with both TB patients and healthy controls. Previous reports have shown increased levels of sIL-2Ra in SARC (22). Although our data suggest a similar trend in both TB and SARC, differences were not statistically significant. We identified significantly lower serum levels of RANTES in SARC patients, which contrasts a recent publication (23) that reports the opposite. Increased abundance of IFNγ in TB patients and almost undetectable levels of IL-2 were in agreement with previous findings (24), although increased levels of IL-2 in active TB have also been reported (25).

In conclusion, our data show both common and disease-related blood profiles in TB and SARC. The complex interplay between genes, miRNAs, and serum analytes involved in disease phenotypes needs to be elucidated in more depth to gain deeper insight into pathologic mechanisms underlying disease (26). Our study describes a first attempt to gain a detailed insight into similarities and differences between the two chronic inflammatory lung diseases TB and SARC. The common disease-related signatures identified in these blood profiles underline contribution of both general inflammatory processes and factors unique to disease manifestations. Such patterns in pathophysiologically similar diseases as described here suggest a potential future use in differential diagnosis of different lung diseases. Common patterns between diseases can herein be just as informative as disease-specific ones. Common patterns, which reveal immune processes that point to a similar underlying pathology, may be used for identification of clusters of diseases, whereas others allow for differential diagnosis of distinct types of diseases (e.g., infectious or noninfectious etiologies). We realize that biomarker-based differential diagnosis requires simultaneous measurement of a number of analytes that is not cost-effective currently for point of care diagnosis of diseases (such as TB) prevalent in resource-poor regions. Thus, by not only focusing on differential signatures but also taking into account shared ones, an algorithm can be designed for triaging patients with pulmonary symptoms. Such a design provides a more promising approach to biomarker-based differential disease diagnosis.

Materials and Methods

Ethics Statement.

The study was approved by the Ethical Committee 1 of the Charité University Medicine, Campus Mitte in Berlin, the University of Luebeck, and the University of Freiburg in Germany. Written informed consent was obtained from all study participants.

Subject Enrolment and Sample Collection.

TB patients included in this study were recruited at the Helios Clinic Emil von Behring in Berlin, Germany. Patients with pulmonary SARC were recruited at the Department of Pneumology, University Medical Center, Freiburg, Germany. All subjects were HIV−, and samples from TB patients were obtained before chemotherapy. Samples from healthy control subjects were obtained from a larger cohort recruited by the TB or Not TB Consortium in Germany. General characteristics of subjects included in this study are given in Dataset S1. Peripheral whole blood (2.5 mL) was collected from every donor in PAXgene tubes (PreAnalytiX) and stored at −80 °C before processing. Serum samples were stored at −20 °C.

Serum Cytokine Multiplex Assay.

Cytokine and chemokine levels in serum were measured on a Luminex 100/200 machine using the human groups I (27-plex) and II (21-plex) Bio-Plex assays (Bio-Rad) according to the manufacturer’s instructions, except that beads and detection antibodies were used at one-half of the concentration indicated. A total of 18 control sera from unrelated donors served to determine background levels of analytes in healthy individuals.

RNA Extraction and Microarray Procedure.

RNA was isolated from PAXgene tubes (PreAnalytix). One-half of the content was used to isolate RNA according to the standard PAXgene blood RNA kit as described previously (6); the other one-half was isolated using the miRNeasy Kit (Qiagen) to include small RNA molecules as well. After two rounds of centrifugation and washing with RNase-free water, the pellet was resuspended in 700 μL Qiazol lysis reagent, and RNA was further isolated according to the standard kit procedure. RNA was labeled with the Fluorescent Linear Amplification Kit (Agilent Technologies) according to manufacturer’s instructions. Quantity and labeling efficiency were verified before hybridization of the samples to whole-genome 4 × 44-k human expression arrays (Agilent) and scanning at 5 μm using an Agilent scanner. Agilent human 8 × 15-k miRNA microarray chips were used to evaluate the expression of 795 different miRNAs (including 719 human miRNAs). Healthy control samples were from the same cohort but different individuals than the RNA preparations for gene expression analysis. Microarray data comply with MIAME (minimal information about a microarray experiment) guidelines and have been deposited in the Gene Expression Omnibus database under accession no. GSE34608.

Data and Statistical Analysis.

Analysis of the scanned images was performed with Feature Extraction software (version 10.5.1; Agilent Technologies). Data were analyzed using the R package Limma (27). Data were log-transformed, and differentially expressed genes were identified based on log twofold changes (M values) in average gene expression with a q value < 0.01 (q equals the P value corrected for multiple testing) (28).

Enrichment analysis of differentially expressed genes based on GO terms was performed using the web-based tool GOrilla (http://cbl-gorilla.cs.technion.ac.il) (29).

To build clusters of genes and miRNAs and make the data manageable, genes were first filtered on an interquartile range of >0.75 to limit the number of genes to those genes with the highest levels of variation; thus, they were more likely to be regulated (<9.000 genes). Results from one of the miRNA samples were omitted from the analysis because of insufficient array signal quality after normalization. For the construction of a coexpression network, we have applied hierarchical clustering using the sample profiles from both miRNA and gene expression data of TB and SARC. To construct the network, a group-wise normalization procedure was first applied. For each gene and each experimental group (TB + SARC), the log intensities were corrected by mean and SD. Hierarchical average clustering with an absolute Spearman correlation coefficient was used as distance metrics. Correlations between the different miRNAs were not considered, because functional links between miRNAs and genes were the focus of the analysis. Results from one of the miRNA samples were omitted from the analysis because of insufficient array signal quality after normalization. Clusters were then explored visually and analyzed for enrichment of GO terms.

Supplementary Material

Supporting Information:


We acknowledge the contributions made by Tom Schaberg, Roland Diel, Oswald Bellinger, and Stefan Ehlers as well as Popgen and all medical officers involved in sample collection and biobanking. This study was funded in part with support from European Commission Seventh Framework Programme Grants NewTBVAC (new TB vaccines) and ADITEC (Advanced Immunization Technologies) and the BMBF (German Federal Ministry of Education and Research) through the TBornotTB Network (Pulmonary Tuberculosis—Host and Pathogen Determinants of Resistance and Susceptibility to Tuberculosis Grant 01KI784).


The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE34608).

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1121072109/-/DCSupplemental.


1. Mistry R, et al. Gene-expression patterns in whole blood identify subjects at risk for recurrent tuberculosis. J Infect Dis. 2007;195:357–365. [PubMed]
2. Jacobsen M, et al. Candidate biomarkers for discrimination between infection and disease caused by Mycobacterium tuberculosis. J Mol Med (Berl) 2007;85:613–621. [PubMed]
3. Berry MP, et al. An interferon-inducible neutrophil-driven blood transcriptional signature in human tuberculosis. Nature. 2010;466:973–977. [PMC free article] [PubMed]
4. Maertzdorf J, et al. Human gene expression profiles of susceptibility and resistance in tuberculosis. Genes Immun. 2011;12:15–22. [PubMed]
5. Jacobsen M, et al. Suppressor of cytokine signaling-3 is affected in T-cells from tuberculosisTB patients. Clin Microbiol Infect. 2011;17:1323–1331. [PubMed]
6. Maertzdorf J, et al. Functional correlations of pathogenesis-driven gene expression signatures in tuberculosis. PLoS One. 2011;6:e26938. [PMC free article] [PubMed]
7. Lesho E, et al. Transcriptional responses of host peripheral blood cells to tuberculosis infection. Tuberculosis (Edinb) 2011;91:390–399. [PubMed]
8. Prince JE, Kheradmand F, Corry DB. 16. Immunologic lung disease. J Allergy Clin Immunol. 2003;111(2 Suppl):S613–S623. [PubMed]
9. Marchiori E, Zanetti G, Barreto MM, de Andrade FT, Rodrigues RS. Atypical distribution of small nodules on high resolution CT studies: Patterns and differentials. Respir Med. 2011;105:1263–1267. [PubMed]
10. Chaussabel D, et al. A modular analysis framework for blood genomics studies: Application to systemic lupus erythematosus. Immunity. 2008;29:150–164. [PMC free article] [PubMed]
11. Sweiss NJ, et al. Significant CD4, CD8, and CD19 lymphopenia in peripheral blood of sarcoidosis patients correlates with severe disease manifestations. PLoS One. 2010;5:e9088. [PMC free article] [PubMed]
12. Breiman L. Random forests. Machine Learning. 2001;45:5–32.
13. Sangokoya C, Telen MJ, Chi JT. microRNA miR-144 modulates oxidative stress tolerance and associates with anemia severity in sickle cell disease. Blood. 2010;116:4338–4348. [PMC free article] [PubMed]
14. Liu Y, et al. Modulation of T cell cytokine production by miR-144* with elevated expression in patients with pulmonary tuberculosis. Mol Immunol. 2011;48:1084–1090. [PubMed]
15. Yang JH, et al. starBase: A database for exploring microRNA-mRNA interaction maps from Argonaute CLIP-Seq and Degradome-Seq data. Nucleic Acids Res. 2011;39:D202–D209. [PMC free article] [PubMed]
16. Koth LL, et al. Sarcoidosis blood transcriptome reflects lung inflammation and overlaps with tuberculosis. Am J Respir Crit Care Med. 2011;184:1153–1163. [PMC free article] [PubMed]
17. Mahabeleshwar GH, et al. The myeloid transcription factor KLF2 regulates the host response to polymicrobial infection and endotoxic shock. Immunity. 2011;34:715–728. [PMC free article] [PubMed]
18. Gonzalo P, Moreno V, Gálvez BG, Arroyo AG. MT1-MMP and integrins: Hand-to-hand in cell communication. Biofactors. 2010;36:248–254. [PubMed]
19. O’Connell RM, Rao DS, Baltimore D. microRNA Regulation of Inflammatory Responses. Annu Rev Immunol. 2012;30:295–312. [PubMed]
20. Wang C, et al. Comparative miRNA expression profiles in individuals with latent and active tuberculosis. PLoS One. 2011;6:e25832. [PMC free article] [PubMed]
21. Fu Y, Yi Z, Wu X, Li J, Xu F. Circulating microRNAs in patients with active pulmonary tuberculosis. J Clin Microbiol. 2011;49:4246–4251. [PMC free article] [PubMed]
22. Miyoshi S, et al. Comparative evaluation of serum markers in pulmonary sarcoidosis. Chest. 2010;137:1391–1397. [PubMed]
23. Loza MJ, et al. Inflammatory profile and response to anti-tumor necrosis factor therapy in patients with chronic pulmonary sarcoidosis. Clin Vaccine Immunol. 2011;18:931–939. [PMC free article] [PubMed]
24. Katti MK. Assessment of serum IL-1, IL-2 and IFN-γ levels in untreated pulmonary tuberculosis patients: Role in pathogenesis. Arch Med Res. 2011;42:199–201. [PubMed]
25. Turgut T, Akbulut H, Deveci F, Kacar C, Muz MH. Serum interleukin-2 and neopterin levels as useful markers for treatment of active pulmonary tuberculosis. Tohoku J Exp Med. 2006;209:321–328. [PubMed]
26. Schadt EE. Molecular networks as sensors and drivers of common human diseases. Nature. 2009;461:218–223. [PubMed]
27. Smyth GK. In: Bioinformatics and Computational Biology Solutions Using R and Bioconductor. Gentleman R, Carey V, Dudoit S, Irizarry R, Huber W, editors. New York: Springer; 2005. pp. 397–420.
28. Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Statist Soc B. 1995;57:289–300.
29. Eden E, Navon R, Steinfeld I, Lipson D, Yakhini Z. GOrilla: A tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinformatics. 2009;10:48. [PMC free article] [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...