BioMed Research International. 2014;2014:713071.

A Derived Network-Based Interferon-Related Signature of Human Macrophages Responding to Mycobacterium tuberculosis.

Kang Wu,1,2,3 Hai Fang,3 Liang-Dong Lyu,1 Douglas B. Lowrie,1,2 Ka-Wing Wong,1,2 and Xiao-Yong Fan1,2


1 Shanghai Public Health Clinical Center, Fudan University, 2901 Caolang Road, Shanghai 201508, China

2 Key Laboratory of Medical Molecular Virology of MOE/MOH, Shanghai Medical College, Fudan University, 138 Yixueyuan Road, Shanghai 200032, China

3 State Key Laboratory of Medical Genomics, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, 197 Ruijin Road II, Shanghai 200025, China



Network analysis of transcriptional signature typically relies on direct interaction between two highly expressed genes. However, this approach misses indirect and biological relevant interactions through a third factor (hub). Here we determine whether a hub-based network analysis can select an improved signature subset that correlates with a biological change in a stronger manner than the original signature. We have previously reported an interferon-related transcriptional signature (THP1r2Mtb-induced) from Mycobacterium tuberculosis (M. tb)-infected THP-1 human macrophage. We selected hub-connected THP1r2Mtb-induced genes into the refined network signature TMtb-iNet and grouped the excluded genes into the excluded signature TMtb-iEx. TMtb-iNet retained the enrichment of binding sites of interferon-related transcription factors and contained relatively more interferon-related interacting genes when compared to THP1r2Mtb-induced signature. TMtb-iNet correlated as strongly as THP1r2Mtb-induced signature on a public transcriptional dataset of patients with pulmonary tuberculosis (PTB). TMtb-iNet correlated more strongly in CD4+ and CD8+ T cells from PTB patients than THP1r2Mtb-induced signature and TMtb-iEx. When TMtb-iNet was applied to data during clinical therapy of tuberculosis, it resulted in the most pronounced response and the weakest correlation. Correlation on dataset from patients with AIDS or malaria was stronger for TMtb-iNet, indicating an involvement of TMtb-iNet in these chronic human infections. Collectively, the significance of this work is twofold: (1) we disseminate a hub-based approach in generating a biologically meaningful and clinically useful signature; (2) using this approach we introduce a new network-based signature and demonstrate its promising applications in understanding host responses to infections.

PMID: 25371902



Genome-wide transcriptional profiling is a powerful way to probe how a biological system responses to changes in environments. In context of pathogen infections of human cells measuring transcriptional changes in human cells can provide insights into pathogenesis.

Transcriptional signatures are commonly generated through network analysis based on known protein-protein interaction between the products of two highly-expressed genes [1]. This approach usually produces a signature containing hundreds of highly-expressed genes. A refined signature with smaller size may represent a core transcriptional response. We were therefore motivated to seek an unbiased method to extract a refined transcriptional signature within a given transcriptional signature so that the refined subset could still reflect the overall transcriptional responses as efficient as the original signature.

We made one key assumption in our approach. We reasoned that related gene products are co-regulated by a third factor during a transcriptional response. We refer such third factor as hub. This indirect relationship is not given any importance in a network analysis based solely on direct interaction. Yet, a few key regulator factors can co-regulate multiple target genes with major consequences during a biological process. We therefore hypothesized that hub-connected highly-expressed genes form a core transcriptional response (Fig. 1).

We have previously identified a common transcriptional signature of human macrophages after infections of different clinical strains of Mycobacterium tuberculosis, the causative agent of human tuberculosis [2]. The signature contains 369 highly-expressed genes. Our approach produces a core subset of 165 genes, all of which are highly connected by a total of 56 hubs. In fact, each hub makes at least 14 direct physical or functional interactions to the 165 high-expressed genes or their gene products. When this hub-connected signature (TMtb-iNet) is applied to public transcriptional dataset of pulmonary tuberculosis it produces correlation as strongly as the parent signature (THP1r2Mtb-induced) (Fig. 2). More strikingly, when applied to data during treatment of tuberculosis TMtb-iNet produces the most pronounced responses during the course of treatment and shows the weakest correlation at the end of treatment in comparison to the parent THP1r2Mtb-induced signature (Fig. 2). On the other hand, the set of genes that is excluded from the hub-connected refined signature (TMtb-iEx) produces the poorest correlation on multiple transcriptional datasets (Fig. 2). Thus, our hub-based approach selects a refined subset from the original signature that correlates at least as strongly as the original signature against clinical transcriptional datasets.

Our approach is limited to identify genes connected by direct protein-protein interactions. But in reality, transcriptional interactions are also relevant during transcriptional responses. Promoter analysis on genes from the original signature identifies common transcriptional binding sites for STAT1/2, IRF-1, IRF-7 and Oct-1 [2]. Interestingly, transcriptional binding site for IRF-7 is enriched specifically in the refined signature and is absent in the excluded signature. Thus, it is possible to incorporate other interaction database and identify key regulator within the refined signature.

IRF-7 is a type I interferon-inducible protein and controls all type-I interferon-dependent immune responses [3]. Consistent to the specific enrichment of IRF-7 in the refined signature there are more interferon-related interacting genes in the refined signature than in the original signature. Excessive type I interferon production is linked to exacerbation of tuberculosis [4]. Early work indicates type I interferon induction by M. tuberculosis involves the ESX-1 secretion system, a major determinant of M. tuberculosis virulence [5]. It is now shown that ESX-1-mediated permeabilization of phagosomal membrane allows cytosolic recognition of Mtb DNA by the cytosolic DNA sensor cyclic GMP-AMP synthase, which then initiates type I interferon production [6-8]. Our refined signature contains AIM2, which may also function as a DNA sensor and regulate macrophage responses to Mtb [9].

In summary, our simple hub-based strategy is able to generate a refined signature that is clinically relevant and biologically meaningful. We envision our method can be applied to other biological processes and yield potentially useful information.


kww fig1

Figure 1. A hub-based strategy to refine an expression signature. The original (input) signature is the expression signature of THP-1 macrophages in response to Mtb infection (THP1r2Mtb-induced) (5). We first identify proteins (hubs in grey) from public-domain STRING protein-protein interaction database that made a minimum i number of direct connections with gene products of genes in the THP1r2Mtb-induced signature. We group all interacting genes to each hub into a subset known as TMtb-iNet (refined signature). All other genes from the original signature are grouped into TMtb-iEx (excluded signature). We choose to use hubs making at least 14 direct connections because the resultant refined signature from such hubs produce the most transcriptionally active signature. The refined signature, the original signature and the cognate excluded signature genes (i.e. TMtb-Ex) are then analyzed for their correlations with other patient-derived transcriptome datasets using a standard transcriptional data comparison tool (gene set enrichment analysis).


kww fig2

Figure 2. Correlation of signatures on patient-derived transcriptome datasets. Gene set enrichment analysis is performed to analyze correlation of signatures on public domain datasets. The result is reported as normalized enrichment score (NES) and false discovery rate (FDR). A positive NES value indicates positive correlation between two datasets and a negative NES indicates negative correlation. FDR value above 0.05 indicates non-statistical significance of NES. The refined signature TMtb-iNet correlates as strongly as the original signature. TMtb-iNet produces the most pronounced responses over the course of tuberculosis treatment and produces a non-significance correlation at the end of the treatment. In the case of dataset from CD4+ and CD8+ T cells, TMtb-iNet produces the strongest correlation. The significance of this strong correlation is unclear. Training set and test set: all donors were from London, UK. PTB, pulmonary TB; LTB, latent TB; HC, healthy controls. Validation set: all donors were from Cape Town, South Africa [10]. Neut, purified neutrophils; Mono, purified monocytes; CD4, purified CD4+ T cells; CD8, purified CD8+ T cells [10]. PTB_0m, PTB patients before drug treatment; PTB_2m, 2 months after initiation of drug treatment; PTB_12m, 12 months after initiation of drug treatment [10].



This work was supported in part by grants from Chinese National Mega Science & Technology Program on Infectious Diseases (2013ZX10003007-003), National Science Foundation of China (81301407, 81273328, 30901276, 81371777), Shanghai Rising-Star Program (12QH1401900), Shanghai Health Bureau (20114013), Shanghai Science and Technology Commission (134119a5200, 114119a3100), and Shanghai Natural Science Fund for Youth Scholars (12ZR1448200).



Ka-Wing Wong, Ph.D.

Xiao-Yong Fan, Ph.D.

Shanghai Public Health Clinical Center, Fudan University

2901 Caolang Road, Shanghai 201508, China

Key Laboratory of Medical Molecular Virology of MOE/MOH, Shanghai Medical College, Fudan University

138 Yixueyuan Road, Shanghai 200032, China



  1. Barabasi, A.L., N. Gulbahce, and J. Loscalzo, Network medicine: a network-based approach to human disease. Nat Rev Genet, 2011. 12(1): p. 56-68.
  2. Wu, K., et al., An interferon-related signature in the transcriptional core response of human macrophages to Mycobacterium tuberculosis infection. PLoS One, 2012. 7(6): p. e38367.
  3. Honda, K., et al., IRF-7 is the master regulator of type-I interferon-dependent immune responses. Nature, 2005. 434(7034): p. 772-7.
  4. Mayer-Barber, K.D., et al., Host-directed therapy of tuberculosis based on interleukin-1 and type I interferon crosstalk. Nature, 2014. 511(7507): p. 99-103.
  5. Stanley, S.A., et al., The Type I IFN response to infection with Mycobacterium tuberculosis requires ESX-1-mediated secretion and contributes to pathogenesis. J Immunol, 2007. 178(5): p. 3143-52.
  6. Collins, A.C., et al., Cyclic GMP-AMP Synthase Is an Innate Immune DNA Sensor for Mycobacterium tuberculosis. Cell Host Microbe, 2015. 17(6): p. 820-8.
  7. Wassermann, R., et al., Mycobacterium tuberculosis Differentially Activates cGAS- and Inflammasome-Dependent Intracellular Immune Responses through ESX-1. Cell Host Microbe, 2015. 17(6): p. 799-810.
  8. Watson, R.O., et al., The Cytosolic Sensor cGAS Detects Mycobacterium tuberculosis DNA to Induce Type I Interferons and Activate Autophagy. Cell Host Microbe, 2015. 17(6): p. 811-9.
  9. Saiga, H., et al., Critical role of AIM2 in Mycobacterium tuberculosis infection. Int Immunol, 2012. 24(10): p. 637-44.
  10. Berry, M.P., et al., An interferon-inducible neutrophil-driven blood transcriptional signature in human tuberculosis. Nature, 2010. 466(7309): p. 973-7.




Multiselect Ultimate Query Plugin by InoPlugs Web Design Vienna | Webdesign Wien and Juwelier SchönmannMultiselect Ultimate Query Plugin by InoPlugs Web Design Vienna | Webdesign Wien and Juwelier Schönmann