Mycobacterium tuberculosis, the causative agent of tuberculosis (TB), kills more than 2 million people per year and has infected an estimated 2 billion people worldwide. It is the leading cause of mortality due to infectious disease . The host immune response to aerosol infection is to quarantine M. tuberculosis in a structure called a granuloma which halts replication of the bacillus and suppresses the immediate threat of active infection . However, granuloma associated M. tuberculosis bacterium can switch to a dormant or non-replicative state and successfully evade any immune response for decades post-initial infection . As the host immune system falters, M. tuberculosis returns to replication mode, which leads to the recurrence of active infection. Thus current TB therapy to fight off active disease requires a strictly monitored treatment period or DOTS (directly observed treatment, short course) lasting up to six months and involving four different drugs: isoniazid, rifampicin, pyrazinamide, and ethambutol [4-6]. Patient compliance with this prolonged therapeutic regime is an important concern. Moreover, prolonged exposure to drugs has likely been an important factor behind increasing reports of anti-biotic resistant bacterium . The lack of well-defined targets specific to dormancy phase M. tuberculosis has been a major obstacle in the development of effective short-course therapies.
A number of studies have attempted to develop in vitro and in vivo models of non-replicating, dormant phase M. tuberculosis. These can be grouped into four main types: hypoxia, starvation, macrophages, and murine infection. Each system mimics some of, but not the entire, clinical situation. The Wayne model (slowly stirred, sealed cultures with a defined air-space to medium ratio)  captures the presumed hypoxic nature of the granuloma, but lacks the effect of the immune response, macrophage phagocytocis, and eventual release to the extracellular milieu. The starvation models are not hypoxic and may not capture the unbalanced diet in the granuloma. The macrophage phagocytosis experiments show the early adaptation to the host immune response, but do not address the long term metabolic changes. While the murine model replicates many facets of the human immune response , mice do not show well-formed granulomas and lack the caseous, necrotic centers characteristic of human infection . None of these models fully capture the heterogeneity of the granuloma, with a gradient of active and inactive/dead immune cells, oxygenation, and nutrients.
DNA microarrays have been used to determine the complete transcriptional response of M. tuberculosis cultured in each of these experimental models. In addition to the dormancy models, DNA microarrays have also been employed to do genome-scale knockout experiments using saturating transposon insertion mutagenesis. Mutants have been profiled for the ability to grow in vitro, in mouse macrophages, and in vivo mouse models [11-14]. In these experiments genes containing (presumably inactivating) insertions are selected for the ability to grow. A disrupted gene that inhibits growth yields a decreased signal on the microarray compared to the genomic control. In the absence of suitable gene inactivation studies for the dormant phase, the phenotypic effects of gene knock-outs on growth phase M. tuberculosis seems to be the best indicator of gene essentiality from a drug target perspective.
Here we present a novel and comprehensive meta-analysis of the M. tuberculosis gene expression and gene disruption microarray data sets. In the absence of a perfect experimental model we chose a consensus approach and combined the different analyses together into a single database. Recently, Hasan et al. of the pharmaceutical company, Novartis, described an in-house software tool for in silico prioritizing of genomic drug targets in pathogens and illustrated its use on M. tuberculosis . Using this tool, they provided three lists based on different prioritization criteria for: 1) genes associated with critical metabolic reactions (chokepoints) unique to M. tuberculosis; 2) genes highly specific to the Actinobacter (the taxonomic class of bacteria inclusive of M. tuberculosis) and absent from other gut flora and 3) genes potentially important in maintaining persistence. The latter criterion is the most critical clinical need for TB treatment, and both Hasan's et al. study and ours use several similar published gene expression and essential data-sets to approach the identification of M. tuberculosis persistence targets. However, GlaxoSmithKline experiences in developing novel antibacterials from a genomics-driven target-based approach have shown the importance of multiple analysis perspectives and the need to substantiate initial bioinformatics target identifications with additional biological rationale . Therefore, while encouraged that several potential targets presented here are also substantiated by earlier computational studies using different approaches, we feel it is important to further extend and rationalize in silico target validation with current knowledge on the biology, biochemistry and disease pathology of M. tuberculosis. Thus, we have tried to provide such additional rationale when evaluating those targets from our prioritized list that are the most promising candidates for concerted drug development programs.