Received date: 30/11/2015; Accepted date: 05/08/2016; Published date: 15/08/2016
Visit for more related articles at Research & Reviews: Journal of Microbiology and Biotechnology
The cell wall of Mycobacterium tuberculosis plays an important role in pathogenesis. It is impossible to analyze the cell wall-associated genes one by one since lacking of functional annotation. Here, we performed clustering analysis of gene microarray expression data and acquired 33 co-expression modules by the construction of three nodes co-regulated networks. A total of 555 cell wall-related genes were predicted in the modules using a multi-factor logistic regression model and motif predicting analysis. The module analysis identified 15 genes without annotation that were also associated with the cell wall. Twentyfive modules contained significant motifs, and genes in 10 of these 25 modules shared a common motif. The methodological approach utilized herein may be applied to identification and description of other function-associated genes in the M. tuberculosis genome. The results of this study might improve understanding of the M. tuberculosis cell wall, and in finding new targets for anti-TB drugs.
Mycobacterium tuberculosis, Cell wallrelated modules, Motif analysis, Cluster analysis
Tuberculosis (TB) is one of the most serious global health issues, the pathogenesis of which is still not clearly understood . Emergence of multidrug resistant (MDR) and, more recently, extremely drug-resistant (XDR) Mycobacterium tuberculosis strains, along with TB-HIV co-infection, have become the new major challenge for TB therapy and control . Anti-TB drugs, such as rifampin and isoniazid, were discovered in 1963, and since then, there have not been any discovery of novel, efficient anti-TB drugs . The number of effective therapeutic targets for TB treatment is insufficient, especially for treatment of MDR-TB. Thus, high-throughput screening for therapeutic targets is the first and most important step for the development of novel anti-TB drugs and improved control of TB .
The cell wall of M. tuberculosis is mainly composed of capsule, mycolic acid, peptidoglycan, arabinose and intima . These components play an important role in the processes that maintain the integrity of M. tuberculosis cell morphology, act against erosion by chemicals, escape host immune response, and lead to development of drug resistance and on a whole, increase the pathogenicity of M. tuberculosis. In this view, the M. tuberculosis cell wall-related components and biosynthesis pathways could be used as targets of anti-TB drugs. Traditional anti-TB drugs, such as Isonicotinic acid hydrazide (INH), target the mycolic acid synthesis pathway . However, inadequate or interrupted treatment with INH results in INH resistance through acquisition of mutations in inhA, ahpC, nadh, katG, or KasA in M. tuberculosis clinical isolates [7-9]. Hence, it is prudent to screen new M. tuberculosis cell wall-relevant genes. In addition, the components and functions of M. tuberculosis cell walls are quite complex and various. Traditional methods for screening M. tuberculosis cell wall genes are expensive and inefficient since they use the technologies of gene knockout and RNA interference to screen the potential cell wall synthesis genes or the target molecules in metabolism. Along with the limitations of experimental methodology, there is still no efficient technology to systematically screen molecules relevant to cell wall synthesis. Thus, the development of a novel approach for panoramic scanning and screening of the M. tuberculosis genes related to cell wall synthesis is necessary. For the above reasons, in our study, we retrieved all published gene microarray data for M. tuberculosis H37Rv, and established the co-regulatory networks of M. tuberculosis genes associated with cell wall synthesis, unknown genes and other genes by means of integration and clustering . The module analysis and the high-throughput annotation of M. tuberculosis genes associated with cell walls provide a molecular basis for the research and development of novel efficient and sensitive anti-TB drugs with less harmful effects.
Cell wall-related modules in M. tuberculosis
Microarray data for M. tuberculosis H37Rv were downloaded from the NCBI database (as of May 2013), which totaled 2863 microarrays of 43 series. These microarrays were related to DNA methylation (n=15), Drug action on tuberculosis (n=1076), Growth and Growth condition to M. tuberculosis (n=758), infection (n=910), gene mutation (n=78) and regulating factors (n=21) (Table S1). A total of 727 genes that were annotated in the Gene Ontology database (www.geneontology.org) were utilized as “seed” genes, and processed for cluster analysis by means of hierarchical clustering, K-means clustering and integrated clustering (Table 1). Based on the results of hierarchical clustering, every module contained a large number of genes, and the largest module contained 342 genes. This method decreased the discrimination of gene functions, although hierarchical clustering revealed the interaction between genes . On the contrary, based on K-means clustering, the genes associated with cell walls were scattered in 201 modules. This method had higher discrimination, but was less clear regarding the interaction between genes . Considering the outcomes of these two methods, integrated clustering was performed based on both K-means clustering and hierarchical clustering (detailed in Materials and Methods). Using this integrated cluster analysis, we identified 163 modules, to which all the known cell wall-associated genes were allocated. Statistical analysis demonstrated that 33 of the 163 modules were closely associated with cell wall synthesis, which contained a total of 555 genes. The correlation of these modules was calculated by Pearson correlation coefficient, and the integrated clustering results were visualized with Cytoscape software version 3.0.2. The results of GO and Pfam analysis illustrated that these genes correlated significantly with several biological processes, such as pathogenesis and response to stimulation (Figure 1), which were consistent with the function of the cell wall in M. tuberculosis pathogenesis. These results showed a better discrimination of the integrated cluster analysis since at most 49 genes were contained in one module and at most 16 genes were associated with the cell wall.
|Total number of modules||No. of modules containing cell wall-associated genes||Maximum No. of genes in one module||Significantly related modules||Maximum No. of cell wall-associated genes in one module|
Table 1. Cluster analysis of Mycobacterium tuberculosis H37Rv gene expression
Multi-factor logistic regression analysis of cell wall related modules
A multi-factor logistic regression was used to analyze the correlation between observations and several factors due to polygenic co-effect of M. tuberculosis cell wall functions . Each factor was evaluated by means of multi-factor linear regression equations . After that, we analyzed the significance of GO-results using multi-factor logistic regression analysis (Figure 2), and constructed a predictive model (detailed in Materials and Methods).
We randomly chose 126 genes from the 727 known cell wall-associated genes, and 128 genes from the 1703 known genes not associated with the cell wall to perform verification. Our predictive model identified 125 of the 126 cell wall-associated genes with an accuracy 99.2%, while 122 of the 128 non-cell-wall-associated genes were correctly determined, the accuracy of which was about 95.3%. So we utilized this predictive model to identify the genes in the 33 cell wall-associated modules described above; 120 of the 197 annotated genes were found associated with cell wall (Table S2).
Motif analysis of cell wall-related modules
The 33 modules associated with the cell wall were subjected to motif analysis using BOBRO software . As a result, 25 modules were found to contain significant motifs whereas genes in 10 of these 25 modules shared a common motif (Figure 3). In addition, these motifs were located at several sites, and the number of motifs before operons was different, which was in accord with the general law of the distribution of motifs.
According to the results of motif prediction, 15 genes without annotation in the 10 modules were associated with cell wall. In particular, module 5 included three cell wall-associated genes (Rv2005c, Rv3132c, Rv3133c) and one unknown gene (Rv0082). Using DOOR2 (http://csbl.bmb.uga.edu/DOOR/), we found that the four genes in module 5 were regulated by three operons (NO.7810, NO.8253, NO.8521), which had a common motif sequence of CGGCGTCG. This revealed the relationship between Rc0082 and the function of the cell wall. In addition, CAAT-box located at 70~100 bps, demonstrated the accuracy of the upstream motif prediction. The cell wall-related genes, Rv1440 and Rv0702, in module 13 were also identified as cell wall-associated genes through the model analysis. These two genes shared common transcription factors with other cell wallassociated genes in module 13. Therefore, we assumed that Rv1440 and Rv0702 were closely related to cell wall function. It is remarkable that Rv0702 is a part of the Rv0702~Rv0710 gene cluster, whose expression was associated with ribosome proteins as we showed by using the KEGG pathway analysis tool (http://www.genome.jp/kegg/) (Figure 4).
Previous studies demonstrated that genes involved in the synthesis of ribosome proteins also participate in the synthesis of cell wall , and the restriction of the function of genes associated with cell wall synthesis can change the structure of ribosomes . Hence, genes associated with cell wall synthesis also play an important role in ribosome synthesis. In addition to the above findings, the genes, Rv0702, Rv0706, Rv0707, and Rv0709, in module 13 not only were close in proximity, but also shared the same operon, No. 7948. Other genes regulated by operon No. 7948 contained 33 modules in all, including Rv0703, Rv0704, Rv0705, Rv0708, and Rv0710. Furthermore, it was verified that Rv0706 and Rv0707 were associated with M. tuberculosis cell wall. Together with the characteristics of prokaryotic genic expression, we believe that genes in the modules regulated by operon No. 7948 might be associated with M. tuberculosis cell wall.
Collection and mining of data
Gene microarray data for strain H37Rv of M. tuberculosis were downloaded from NCBI (www.ncbi.nlm.nih.gov/gds). After selection, 2710 H37Rv microarrays of 43 series were retained. Owing to the different probe number of several microarrays from different companies, we unified the genes in different microarrays according to the M. tuberculosis gene number and names published by KEGG database in 2013; missing items were represented as NA. We further standardized the microarray data by means of min-max procedure as follows: the 5% largest and 5% smallest data in every microarray were given the same maximum value or minimum value, respectively, to remove the effect of extreme values. The values ranged from 1 (min) to 100 (max). A total of 727 M. tuberculosis genes were selected through GO analysis.
A 3994 × 2710 matrix was built using the standardized microarray data. The rows of the matrix represented the expression of each gene under 2710 conditions, and the columns represented the expression of each condition for every gene. The cluster analysis was performed using the bioinformatics toolbox of Matlab software. During hierarchical clustering, we used the Spearman function to calculate correlation. The complete function was best for calculating the linkages within clusters. Finally, the correlation function was used to calculate correlation in K-means clustering.
Construction of co-expression networks and predicitive model
Every gene in the same module had a certain correlation. To display the correlation, we calculated Pearson correlation coefficients for each pair of genes in every module. A positive co-expression was one with an R-value greater than 0.90, while negative co-expression was indicated by an R-value less than -0.90 (Usadel et al.). A co-expression network was constructed using Cytoscape software version 3.0.2.
Multi-factor logistic regression analysis
Highly significant cell wall-associated items (P< 0.01) were selected in the GO database, and a multi-factor logistic regression analysis model was constructed using those items as dependent variables. To verify the sensitivity and specificity of the multifactor logistic regression model, 126 genes were randomly chosen among 727 cell wall-associated genes and 128 genes of 1703 genes not associated with cell wall. We marked the items, which had the dependent variables in the annotations, with 1. Then, by the theory of logic regression, the gene was not associated with M. tuberculosis cell wall if all of the results were 0. This model was used to predict cell wall-associated the genes in the 33 modules.
A total of 33 modules associated with cell walls were analyzed using BOBRO software . BOBRO is accurate motif prediction software focused on the features of prokaryotic genomes, which uses the algorithms of motif closures and graph theory. It is based on the hypothesis that the internal genes might be regulated by the same transcription factors and prokaryotic regulatory regions . Upstream 300-bp DNA sequences were selected as regulatory regions, and the conservative sites were searched upon the regulatory regions, used as candidate regulatory motifs. The significant motifs were selected utilizing the characteristics of transcriptional regulation motif sequences.
The significances were analysed using hypergeometric distribution, and P-values less than 0.01 were considered statistically significant. Correlation was calculated using the Pearson correlation coefficient, and Chi-square test was used to analyze the discrepancy between the three clusters.
This work was supported by National Natural Science Foundation of China (61303084 and 81271897), Specialized Research Fund for the Doctoral Program of Higher Education of China (20110061120093), China Postdoctoral Science Foundation (20110491311 and 2012T50285, Foundation of Jilin Provincial Health Department (2011Z049), Foundation of Jilin Province Science and Technology Department (20130522013JH and 20140414048GH) and the Norman Bethune Program of Jilin University (No. 2012219). We also thank the Medjaden Bioscience Limited for editing and proofreading this manuscript.