*3.1. Target Identification*

Understanding the mycobacterial transcriptome is key to connecting genome information to protein target expression, highlighting potentially druggable pathways and bacterial responses to drug exposure. With the development of whole-genome technologies, such as microarrays and more recently RNAseq, gene expression studies have been able to capture a snapshot of the total abundance and differential expression of transcripts present in an organism in various conditions. Transcriptomics has become an important tool for exploring the biology of *M.tb*, providing information about adaptive responses to understand mechanisms of pathogenicity, assign gene function, discover new drug targets and explore drug action. Transcriptomics in a discovery biology setting has uncovered induction of potentially druggable pathways involved in β-oxidation of fatty acids, the glyoxylate shunt and cholesterol metabolism in *M.tb* replicating intracellularly in macrophages [50,51] and in expectorated *M.tb* in patient sputa [52], alongside expression of metal detoxification systems [53,54] amongst others. Profiling in vitro models of persistence has revealed adaptations to respiratory and metabolic networks involved in the transition of *M.tb* between different growth states [55,56] highlighting target pathways for investigation. RNA signatures from animal models of TB infection [57,58] and human tissue [59] provide important information on the expression of targets in human disease, enhancing the prospect of cidal drug action by targeting pathways active in vivo. This is valuable evidence for drug discovery decision making, since the bactericidal or bacteriostatic inhibition of an essential target in vitro does not necessarily predict in vivo drug efficacy. This was demonstrated by Pethe et al. who identified pyrimidine-imidazoles as potent antimycobacterial agents in a whole cell screen against *M.tb*; lead compounds showed activity in vitro but failed to show any inhibition in a murine model of infection [60]. Compound efficacy was linked to the accumulation of glycerol phosphate and a reduction in ATP synthesis in the presence of glycerol. Glycerol metabolism is dispensable in vivo and thus inhibition of this pathway was not cidal in animal models of TB disease. Approaches combining transcriptomic and gene essentiality datasets offer a multi-omics solution to prioritizing pathways for further investigation.

#### *3.2. Mode of Action*

The transcriptional response of *M.tb* to antimicrobial drug exposure has improved the understanding of many drugs, providing new insights for antibiotics currently in use for TB treatment with known cidal mechanisms, as well as predicting mode of action and the targets of novel compounds. This unsupervised approach is especially useful for understanding the actions of lead compounds from high-throughput screens where the mechanism of *M.tb* killing is entirely mysterious. Comparison of the *M.tb* transcriptional response to a novel compound with mRNA signatures derived from drugs of known function allows broad mode of action to be revealed (Figure 2). In the first study of *M.tb* transcriptional adaptations to drug treatment, Wilson et al. used DNA microarrays to explore changes in gene expression in response to isoniazid [61]. The authors showed that drug exposure induced several genes relevant to isoniazid's known mode of action; drug treatment caused the cluster of five genes encoding type II fatty acid synthase enzymes (*fabD*-*acpM*-*kasA*-*kasB*-*accD6*) to be upregulated. Other induced genes, such as *efpA* and *ahpC* not in the biosynthetic pathway targeted by isoniazid, were linked to the toxic effects of the drug. In subsequent years, several studies have used DNA microarrays to correlate the mRNA signatures of *M.tb* exposed to antimicrobial drugs with predicted mode of action [62–64]. Boshoff et al. generated a dataset of 430 *M.tb* gene expression profiles to measure the effect of 75 different drugs, drug combinations and growth conditions [64]. The individual RNA drug signatures were classified into groups of agents with similar modes of action (protein synthesis inhibitors, transcriptional inhibitors, cell wall synthesis inhibitors and DNA damaging agents), which have been used to predict the mechanism of action of antimycobacterial compounds of unknown function derived from whole cell screening approaches [65–68].

**Figure 2.** The application of transcriptomics to define drug mode of action. Principle Component Analysis (PCA) of *M.tb* responses to seven different drugs (represented by shapes) derived from RNAseq of 2–3 biological replicates per drug. *M.tb* mRNA signatures from antimicrobial drugs with similar mechanisms of action will cluster together.

Transcriptomics has been applied to investigate the mode of action of drugs now in phase I/II studies or in the clinic [2]. The *M.tb* transcriptional signature to benzothiazinone exposure most resembled that of cell wall inhibitors and ethambutol in particular, providing an early indication that the compound was targeting cell wall biosynthesis and arabinogalactan synthesis. This was confirmed to be the case with benzothiazinones inhibiting decaprenyl-phosphoribose-2- -epimerase (DprE1) in the metabolism of D-arabinose, two steps upstream of the action of ethambutol [7]. RNA profiling provided fundamental insights to elucidate the molecular mechanism of mycobacterial killing by pretomanid (formerly PA-824), a nitroimidazole active against both replicating and non-replicating *M.tb*, by inhibiting cell wall synthesis and releasing nitric oxide [69,70]. Pretomanid exposure resulted in a dual signature indicating cell wall inhibition (similar to isoniazid) and respiratory chain disruption (similar to respiratory inhibitors such as cyanide). While the upregulation of genes such as *fasI*, the *fasII* operon, *efpA* and *iniBAC* signposted the aerobic killing mechanism targeting cell wall biosynthesis, drug efficacy in anaerobic conditions was marked by the induction of the *cyd* operon encoding the non-proton-pumping cytochrome bd oxidase, the nitrate reductase *narGHIJ* and other genes involved in respiration [71]. RNAseq analysis of the *M.tb* response to the recently approved nitroimidazole delamanid (formerly OPC-67683) showed many similarities to the pretomanid signature, highlighting that respiratory poisoning plays an important role in the bactericidal effect of these compounds in anaerobic conditions [72].

Boot et al. used RNAseq responses of *M.tb* and *Mycobacterium marinum* to subinhibitory concentrations of ciprofloxacin, ethambutol, isoniazid, streptomycin and rifampicin to select genes to act as drug mode of action-specific reporters [73]. Concordance between the expression levels of *M.tb* and *M. marinum* upon drug exposure were high for the orthologous genes; however, *M. marinum* showed a more distinct stress fingerprint that facilitated simple assays for quick mode of action determination. Ten drug-specific *M. marinum* genes were selected and their promoter regions were cloned into green fluorescent protein (GFP) reporter constructs. As proof of concept that these drug reporters could accelerate TB drug discovery by identifying the mode of action of hit compounds, the *MMAR\_4645*-ciprofloxacin reporter and *iniBAC*-isoniazid reporter were used to screen a library of 196 antimycobacterial compounds. The screening revealed two compounds to have a mode of action similar to isoniazid, likely targeting the mycobacterial cell wall, and one compound, similar to ciprofloxacin, that potentially inhibited DNA replication. Understanding mycobacterial responses to antimicrobial drugs may also offer new targets to enhance combination therapy. Peterson et al. used the Environment and Gene Regulatory Influence Network (EGRIN) model and Probabilistic Regulation of Metabolism (PROM) model of *M.tb* regulatory systems to demonstrate that bedaquiline pushed bacilli into a tolerant state that reduced bedaquiline killing [74]. Disruption of this network, by knocking out key transcription factors (*Rv0324* and *Rv0880*) predicted to mediate this response, significantly increased bedaquiline killing. Analysis of transcriptome data from *M.tb* exposed to antitubercular drugs identified molecules that significantly downregulated the expression of *Rv0324* or *Rv0880*, predicting synergism between bedaquiline and pretomanid through the inhibition of the *Rv0880* regulon by pretomanid. The in vitro combination of sub-inhibitory concentrations of both drugs showed an additive to mildly synergistic effect, while the effect was eliminated and a strong antagonism observed when the combinations were tested using a strain over-expressing *Rv0880*. Given the vast number of possible drug combinations, this strategy could complement other preclinical methods and accelerate the discovery of new drug regimens for TB by avoiding combinations with antagonistic interactions and prioritizing those with synergistic effects.

The interaction between drugs in combination were also explored in the computational model INDIGO-MTB, with the premise that drug synergy and antagonism occur due to coordinated, system-level molecular changes involving multiple cellular processes [75]. Using a compendium of publicly-available and in-house *M.tb* transcriptional responses to drug exposure in vitro as input, the model screened in silico more than 1 million potential combinations of 164 drugs, predicting synergistic and antagonistic regimens featuring 35 existing and potential anti-TB drugs. Combinations

containing chlorpromazine, a drug used to treat psychiatric disorders, and verapamil, used to treat hypertension, were highly enriched for synergistic interactions. In contrast, combinations featuring sutezolid, an oxazolidinone anti-TB drug in phase II trials, were observed to be antagonistic. Regimens featuring combinations of bacteriostatic and bactericidal drugs, and combinations of only bactericidal drugs, had significantly more antagonistic interactions than combinations of only bacteriostatic drugs. The predictions of INDIGO-MTB were validated experimentally in vitro using checkerboard assays and the high-throughput DiaMOND method and compared to a meta-analysis of data assembled from 57 phase II clinical trials. The authors found a significant correlation between INDIGO-MTB interaction scores for drug regimen synergy and sputum culture conversion rates after 8 weeks of treatment. The model also identified *Rv1353c* as a key transcriptional regulator mediating multiple drug interactions. The upregulation of *Rv1353c* in vitro reduced drug antagonism of the bedaquiline-streptomycin combination, suggesting this transcriptional factor might be targeted to enhance drug synergies.

Most transcriptional profiling studies map the response of log phase aerobically-respiring bacilli to drug exposure in vitro, this enables direct comparison between drug signatures from axenic culture, but misses the complexity of *M.tb* in vivo phenotypes. Studies by Walter et al. and Honeyborne et al. have characterized drug responses in patient sputa during standard therapy as a measure of in-patient drug efficacy, identifying, for example, an isoniazid signature in expectorated bacilli that disappears after only 3-4 days of the start of drug therapy [76,77].

### **4. Proteomics**
