|Home | About | Journals | Submit | Contact Us | Français|
While the immune system is credited with averting tuberculosis in billions of individuals exposed to Mycobacterium tuberculosis, the immune system is also culpable for tempering the ability of antibiotics to deliver swift and durable cure of disease. In individuals afflicted with tuberculosis, host immunity produces diverse microenvironmental niches that support suboptimal growth, or complete growth arrest, of M. tuberculosis. The physiological state of nonreplication in bacteria is associated with phenotypic drug tolerance. Many of these host microenvironments, when modeled in vitro by carbon starvation, complete nutrient starvation, stationary phase, acidic pH, reactive nitrogen intermediates, hypoxia, biofilms, and withholding streptomycin from the streptomycin-addicted strain SS18b, render M. tuberculosis profoundly tolerant to many of the antibiotics that are given to tuberculosis patients in a clinical setting. Targeting nonreplicating persisters is anticipated to reduce the duration of antibiotic treatment and rate of post-treatment relapse. Some promising drugs to treat tuberculosis, such as rifampicin and bedaquiline, only kill nonreplicating M. tuberculosis in vitro at concentrations far greater than their minimal inhibitory concentrations against replicating bacilli. There is an urgent demand to identify which of the currently used antibiotics, and which of the molecules in academic and corporate screening collections, have potent bactericidal action on nonreplicating M. tuberculosis. With this goal, we review methods of high throughput screening to target nonreplicating M. tuberculosis and methods to progress candidate molecules. A classification based on structures and putative targets of molecules that have been reported to kill nonreplicating M. tuberculosis revealed a rich diversity in pharmacophores. However, few of these compounds were tested under conditions that would exclude the impact of adsorbed compound acting during the recovery phase of the assay, and few were tested under more than one condition imposing nonreplication. That nonreplicating mycobacteria are metabolically active was corroborated by their susceptibility to several antibiotics and tool compounds that target the synthesis of lipids, RNA, DNA, proteins, and peptidoglycan.
Two parallel revolutions were born in the golden era of antibiotics (~1940–1960). One was a revolution in medicine, as physicians went to war with microbes. The second was a revolution in biology, as microbiologists and geneticists used anti-infectives as tools to reveal how microbes functioned on a molecular level. Scientists converged on a surprisingly short list of essential biological processes that appeared to make up an Achilles’ heel shared by diverse bacterial pathogens– the biosynthesis of nucleic acids (DNA and RNA), protein, cell walls (peptidoglycan and lipids), and folate. Only later were the far wider dimensions of potential target space appreciated (1). The discovery of targets led to the development of methods to improve existing antibiotics and find new ones.
The success of chemical biology at advancing antibiotic development was spectacular but short-lived. New antibiotics quickly encountered genetically-encoded drug resistance (2). The selective pressure imposed by antibiotics presented bacteria with a seemingly impossible task of becoming drug resistant by modifying the antibiotic’s target, modifying the antibiotic’s structure, effluxing the antibiotic, or altering their cell wall’s permeability to the drug without a major fitness cost. Yet bacteria solve this problem routinely in laboratories, the environment, animal models of disease, and patients. Resistant mutants distribute drug resistance by vertical transmission (passing chromosomal DNA to their progeny) and horizontal transmission (via phages and plasmids). Antibiotic research concentrated on understanding the basis of genetically encoded drug resistance and medicinal chemistry campaigns focused on bypassing it.
However, genetic drug resistance was not the only hurdle. Pioneering observations published by Hobby, Meyer and Chaffee in 1942 (3), and Bigger in 1944 (4) cast an ominous cloud over the remnants of optimism that antibiotics could eradicate diseases of bacterial origin. Hobby and her colleagues observed that about 1 streptococcus of 106 in a replicating culture survived exposure to penicillin, while in a culture whose replication was halted by cold, nearly all the cocci survived (3). Bigger made the same observation with staphylococcus and additionally noted that the cocci became tolerant to penicillin when their replication was halted by acidification or hypotonicity of the medium (4). Microbiologists had long assumed that logarithmically growing bacterial cultures were uniform. Use of penicillin as a tool allowed Hobby, Meyer, Chafee and Bigger to discover that the assumption of bacterial homogeneity was incorrect. Moreover, they demonstrated that bacteria could resist killing by antibiotics through a non-heritable mechanism. The penicillin-resistant cells were as sensitive to penicillin as the population from which they came when they were expanded in fresh medium and exposed to penicillin a second time. Bigger used the term “persisters” for bacteria that survived antibiotics without heritable resistance. The property allowing persisters to survive was later termed “phenotypic drug resistance”, or, “phenotypic tolerance”. These historic studies have important implications for anti-infective discovery paradigms today (3, 4).
Two decades later, Hobby and Lenerts extended the observation of phenotypic tolerance to a different organism, Mycobacterium tuberculosis, and additional drugs, isoniazid and para-aminosalicylate (5). Isoniazid targets the synthesis of mycolic acids, para-aminosalicylate targets the synthesis of folate and penicillin targets the synthesis of peptidoglycan. Thus, the phenomenon of phenotypic tolerance was independent of the chemical class of antibiotics and of the pathways they inhibit.
The problem of persisters is central to the chemotherapy of tuberculosis. It is believed to be a major reason why the current WHO-approved treatment regimen for drug-sensitive tuberculosis takes 6 months to achieve cure in ~95% of participants in formal studies; the cure rate is about 86% in routine practice. Drug-resistant tuberculosis generally requires treatment for over two years and cure is often not achieved (6). In the “Cornell model”, mice with drug-sensitive tuberculosis that are treated with isoniazid and pyrazinamide for two months harbor no detectable colony forming units of M. tuberculosis when their organ homogenates are spread on bacteriologic agar. However, about one-third of the remaining members of the same cohort of mice relapse spontaneously some months later, and nearly all of them relapse if immunosuppressed with corticosteroids, anti-IFNγ, anti-TNF, or inhibitors of inducible nitric oxide synthase (7–9). The M. tuberculosis recovered at relapse is as sensitive to isoniazid and pyrazinamide as the population used for inoculation. These observations indicate the presence of drug tolerant persister populations after antibiotic treatment, even if they are temporarily undetectable by standard microbiologic methods. Likewise, sputa from about 80% of treatment-naïve tuberculosis patients contained M. tuberculosis that were not quantifiable by CFU analysis (10, 11)
Experience with metronidazole illustrates the challenge of translating the foregoing knowledge into a faster and more effective treatment of tuberculosis. In some animal models, M. tuberculosis encounters hypoxia in necrotic granulomas (Table 1). In vitro, hypoxia causes mycobacteria to cease replicating and become phenotypically tolerant to most drugs. In contrast, the anti-bacterial and anti-parasitic drug metronidazole kills hypoxic mycobacteria in vitro. Thus metronidazole seemed well suited to kill nonreplicating M. tuberculosis. However, metronidazole’s activity in animal models of tuberculosis correlated imperfectly with hypoxia in granulomas (Table 1) (12–19). Metronidazole improved the proportion of patients whose sputum became smear- or culture-negative at one month of treatment, but did not impact treatment outcome at 6 months, other than contributing to peripheral neuropathy (14). In retrospect, the ability of metronidazole to kill hypoxic M. tuberculosis in vitro was studied in the absence of an alternate electron acceptor, putting the organism at a greater disadvantage than it is likely to face in vivo. M. tuberculosis is not restricted to using oxygen as an electron acceptor; it can also use nitrate or fumarate (20–22). Nitrate is a physiologic constituent of human body fluid. Inclusion of nitrate markedly diminished the in vitro efficacy of pyrazinamide (23).
The experience with metronidazole suggests that it may not be enough to find antibiotics with the exceptional property of killing bacteria that are phenotypically tolerant to most other antibiotics; it matters how the bacteria are rendered phenotypically tolerant. If phenotypic tolerance is achieved by using conditions that prevent the bacteria from replicating, it matters how they are prevented from replicating. The more the conditions resemble those in the host, the more likely that drugs that work under those conditions may also work in the host.
The foregoing statements are a hypothesis whose testing is just beginning. After Bigger’s report (4), it took another 40 years until Coates proposed large-scale screening to target nonreplicating M. tuberculosis (24). His proposal came at time when many pharmaceutical companies were scaling back or abandoning anti-infective discovery. Other firms stuck to the industry’s standard practice of seeking broad-spectrum agents that could cure infections prevalent in economically advantaged regions. Only after 1999 did a new funding landscape emerge that supported academic–industrial partnerships for drug discovery for infectious diseases that are prevalent chiefly in economically disadvantaged regions (1, 25–27). Only about 10 years ago did pharmaceutical companies and their academic partners begin large-scale screens for drugs targeting phenotypically tolerant mycobacteria (28, 29).
This chapter describes and categorizes approximately one hundred compounds that have been reported to kill mycobacteria rendered nonreplicating in one or another in vitro model. We also offer comments about the biology of drug tolerance, strategies for screening compounds against phenotypically tolerant mycobacteria and progressing the actives through secondary assays, and pitfalls in data interpretation.
Sensitivity to an antibiotic is conventionally defined under replicating conditions and reported as a minimum inhibitory concentration (MIC), typically meaning a concentration that restricts growth by at least 90% compared to a culture under the same conditions that is exposed to the vehicle alone for the same period of time. Phenotypically tolerant bacteria of class I are those rare cells that survive exposure to the antibiotic at or above its MIC when tested under these standard, replicating conditions. In contrast, phenotypically tolerant bacteria of class II are the majority of a population that survives exposure to the antibiotic at or above its MIC under different conditions, typically those that impose nonreplication. To distinguish the nonreplication imposed by the test conditions from death imposed by the antibiotic usually requires removing the antibiotic by washing or dilution and reversing the conditions that impose nonreplication, then detecting recovery or the lack of recovery of the surviving bacteria by allowing survivors to replicate. The hallmark of both class I and class II phenotypic tolerance is that the survivors, when tested again under the standard conditions, display the same MIC as the original population (25).
To fully appreciate the diversity exhibited by nonreplicating cells, it will be useful to start by correcting several misconceptions. First, when Hobby et al. (3) and Bigger (4) discovered what we now call class I persistence, they attributed it to nonreplication of about 1 bacterium in a million in an otherwise replicating population. They had no direct evidence for this. More than half a century later, it became possible to test this notion, and the results have been mixed. In short, class I phenotypic tolerance sometimes is and sometimes is not associated with nonreplication of a minority of cells in a replicating population. By definition, class II persisters are nonreplicating. Therefore, class I and class II persisters should not be grouped together as “nonreplicating cells”. Second, just because a bacterial population has stopped changing in number over a period of time does not exclude the occurrence of balanced replication and death. For simplicity, we use the term “nonreplication” to describe a population of static size, but without implication as to the degree of turnover. Third, just because one population of bacteria has entered a nonreplicating state in response to one condition does not mean that it has the same phenotype as another population that has entered a nonreplicating state in response to another condition.
Single-cell analyses of persister populations are now feasible. For example, cell division can be monitored by dilution of a fluorescent signal from a chromosomal copy of mCherry (35, 36), and metabolism monitored using redox sensor green (RSG), which generates a fluorescent signal upon reduction by bacterial reductases. The fates of individual cells can be tracked over time by microfluidics and time-lapse microscopy (36–39). Replicating and nonreplicating cells, and metabolically active and metabolically inactive cells, may be sorted using FACS. For example, Brynildsen and colleagues found that while non-growing cells were enriched for class I persisters, 20% of the persisters were replicating, and slow metabolism correlated with, but was not required, for persistence (36).
Persister diversity may result from heterogeneity in such bacterial processes as maintenance of membrane potential, DNA replication, and ribosomal translation (40, 41), or in host environments, where bacteria may be extracellular in connective tissue or caseum, or intracellular in phagosomes or cytosol (42). The transcriptome of M. tuberculosis class I persisters enriched by D-cycloserine treatment to kill replicating cells overlapped very little with the transcriptomes of class II phenotypically tolerant cells generated by incubating M. tuberculosis under conditions of hypoxia (43, 44), stationary phase (12, 44, 45), or nutrient starvation (12). Moreover, only 5 genes were identified as commonly upregulated in the 4 nonreplicating models (46) and there was little overlap of M. tuberculosis’s differentially regulated genes in the 3 class II nonreplicating models (43, 44, 46). Another comprehensive comparison found a poor correlation of transcriptomes of M. tuberculosis rendered nonreplicating in multiple models, including removing streptomycin from the streptomycin-addicted strain SS18b, exposing wild type M. tuberculosis to reactive nitrogen intermediates, depriving it of phosphate, nutrients or oxygen, or combining a variety of stresses (47). On the other hand, Voskuil et al. found a correlation among the transcriptomes of M. tuberculosis exposed to hypoxia, the nitric oxide donor DETA-NO, and cyanide (48), and the transcriptional changes were similar to those seen during M. tuberculosis’s infection of IFNγ-activated bone-marrow derived macrophages (49). While these transcriptomics experiments were insightful, we do not know the relevance of transcriptional regulation of individual genes to the survival of mycobacteria as class I or class II phenotypically tolerant. Transcriptomics profiles are time-dependent, making it difficult to compare transcriptomes studied at different times. Moreover, many key regulatory steps are post-translational.
Another indication of the diversity of nonreplicating mycobacteria is that the same compounds are differentially active against M. tuberculosis rendered nonreplicating in different ways, such as nutrient starvation, stationary phase, hypoxia, and a combination of acidic pH and reactive nitrogen intermediates (50–52). In the multi-stress model (acidic pH, reactive nitrogen intermediates, hypoxia, and a fatty acid carbon source), some compounds specifically required reactive nitrogen intermediates for their activity (28, 53). Grant et al. found that of 52 molecules active against M. tuberculosis in a carbon starvation model, only 33% were also active against bacilli rendered nonreplicating by hypoxia (54). The same study found diversity of the activity profiles of 4 compounds, from 3 chemical classes, in a class I persister model and three class II models: hypoxia, starvation, and removal of streptomycin from the addicted strain, SS18b (54, 55).
Very few compounds have been demonstrated to kill M. tuberculosis rendered nonreplicating in more than one way. This may be because such “pan-actives” are rare in chemical space or because investigators do not routinely test actives from one model in other models. We describe pan-actives in the section “Proof of concept molecules”.
Long before the work of Bigger, Hobby and colleagues was rediscovered (3, 4), and the term “phenotypic tolerance” became widely used (56), researchers had observed evidence of class I persisters in vitro and in vivo. Kill curves, in which the X-axis of the graph represents time, and the Y-axis represents the number of viable bacteria recovered on agar plates using a CFU-based assay, often have a biphasic, or “hockey stick”, shape (25). Following a sharp, logarithmic decrease in viable CFU at early time points, the CFUs plateau, or decrease at a reduced rate. Notably, compounds fail to reduce CFU below the limit of detection at any concentration tested (46, 55). Put differently, the CFU assay reveals a small population of cells that is refractory to killing by the antibiotic. Biphasic kill curves have been observed for M. tuberculosis and other mycobacterial species treated with dapsone, ciprofloxacin, isoniazid, D-cycloserine, rifampicin, streptomycin, and various combinations of these antibiotics (30, 46, 54, 55, 57, 58). Class I persisters appear to play a role in phenotypic drug tolerance during human infections caused by Pseudomonas aeruginosa, Escherichia coli, and Candida albicans (57, 59–61). Evidence of class I persisters was observed in murine and guinea pig models of tuberculosis (62–64) and in the human disease (57).
By definition, class I mycobacterial persisters are reversibly tolerant to one or another of the standard antibiotics (46, 55, 65) but not necessarily to their combinations. There is no reason to expect class I phenotypically tolerant bacteria to be more resistant than their siblings to molecules that have multiple targets, such as hydroxyl radicals (55, 66). Unlike in E. coli (67), there is conflicting evidence whether class I mycobacterial persisters are cross-tolerant to other antibiotics. In one study, M. smegmatis and M. tuberculosis persisters that survived exposure to a combination of ciprofloxacin and isoniazid were tolerant to a bactericidal concentration of rifampicin (55). However, a different study found persister populations of ~ 1.7×10-5 to isoniazid, ~ 7.0×10-4 to rifampicin, and > 10-1 to pyrazinamide (65). The persister population resistant to the combination of isoniazid, rifampicin, and pyrazinamide was ~ 2.8×10-7, indicating that individual persisters were not broadly resistant to other antibiotics (65). While strategies to target class I persisters have been proposed (40, 41, 68), to our knowledge, high throughput screens targeting class I mycobacterial persisters have not been undertaken. Conversely, most compounds known to have activity against mycobacteria have not been tested for activity against mycobacteria displaying class I phenotypic tolerance to other compounds. Compound 57, identified in a class II phenotypic screen against carbon-starved M. tuberculosis, serves as an illustrative example of a compound whose ability to additionally kill class I phenotypically tolerant M. tuberculosis was discovered post-screening (54). Compound 57 is described in more detail in the section “Screening hits: carbon starvation”(54).
In vitro, genetic mutations in hipA and hipB (high persister genes) lead to approximately 10–10,000-fold more class I drug tolerant persisters in E. coli, Salmonella, and other species (57, 69–73). Use of hip mutants permitted the observation of persisters by time-lapse studies in microfluidic devices (39). High persister mutants that survived treatment with streptomycin and rifampicin were recently identified in an ethyl methanesulfonate-mutagenized auxotrophic strain of M. tuberculosis, and characterized by genome re-sequencing and transcriptomics (57). Genetic control over the size of a class I phenotypically tolerant population should not be confused with heritable resistance. The survivors, when grown up without antibiotic and exposed again, have the same MIC as the population from which they were derived. Even a mutant strain of E. coli with a 10,000-fold increase in the wild type proportion of class I phenotypically tolerant persisters to ampicillin will display a 99% reduction in survival at the same concentration of ampicillin as the wild type strain if the proportion of class I persisters has increased from 1×10-6 to 1×10-2.
Numerous mechanisms can impel a cell to display class I phenotypic tolerance (39, 74–79). Mycobacterial asymmetric division results in differential antibiotic sensitivity of daughter cells (80, 81). As in E. coli (82, 83), mycobacteria may depend on toxin-antitoxin genes (46, 65, 84) to induce class I tolerance. Javid and colleagues found that mistranslation of two amino acids, glutamate for glutamine, and aspartate for asparagine, resulted in modified RNA polymerase (RpoB, encoded by rv0667) that was more resistant to rifampicin (85). Only a minority of cells in a wild type population accumulated enough mutant copies of RpoB with Asp in place of Asn at position 434 to survive rifampin at its MIC (85, 86). When these cells were expanded, the MIC remained the same. Analogous to the situation with hip genes in E. coli, mutation in the GatCAB aminotransferase that normally corrects mistranslation of the Asn codon increased the frequency of these class I phenotypically tolerant mycobacteria, but the MIC was no greater in progeny of these cells than in the population from which they were selected (86). Rendering M. smegmatis nonreplicating by acidic pH or nutrient starvation led to protein mistranslation and phenotypic kanamycin resistance (85). Isoniazid is a pro-drug that requires oxidation by a catalase-peroxidase (KatG, encoded by rv1908c) and forms an NAD-isoniazid adduct that targets NADH-dependent enoyl-ACP reductase (InhA, encoded by rv1484). Isoniazid kills multiple log10 CFU of replicating mycobacteria within days; yet, isoniazid dosed by itself takes weeks to months to achieve a modest reduction in the M. tuberculosis bacterial burden in mice (87). Stochastic gene expression has been described in eukaryotes and prokaryotes (88, 89) and provides one potential explanation for the appearance of class I phenotypic tolerance in a small subpopulation of bacteria. For example, Wakamoto et al. found that stochastic expression of katG explains some mycobacterial tolerance to isoniazid (90). In E. coli, fluoroquinolones can damage DNA and induce an SOS response protein, TisB, which transforms cells to a persister phenotype by depolarizing the membrane and depleting ATP (91–93).
Class II persisters are defined as a population of cells displaying phenotypic drug tolerance under externally applied conditions that halt net replication. As noted, nonreplication in this sense is a terminologic simplification that encompasses balanced bacterial growth and death. In some models of nonreplication, there is a slow reduction in viable bacteria over the period of observation that may be difficult to detect by a CFU assay (28, 94). Conditions that arrest growth are associated with resistance to a large number of antibiotics. Some investigators have assumed that failure to grow is synonymous with shutdown of the bacterial machinery that synthesizes macromolecules and that lack of need for macromolecules explains lack of sensitivity to drugs that inhibit their synthesis (30, 50–52, 95, 96). However, M. tuberculosis adapts to the stresses that impose nonreplication with a robust transcriptional response (46, 49, 97) and cell wall remodeling (98, 99) and maintains metabolic activity, although with a different profile of metabolites than during replication (K. Rhee, personal communication). Non-reliance on biosynthetic processes is an unsatisfactory explanation for class II phenotypic tolerance.
The rate at which M. tuberculosis achieves stasis may impact the bacilli’s biology and sensitivity to certain compounds. For example, some models of nonreplication, such as hypoxia or starvation, require pre-adaptation periods of 1–2 weeks or more (12, 54, 95). In contrast, reactive nitrogen intermediates cause immediate growth arrest (48, 100). In addition, exogenously applied stresses may be perceived at different rates by mycobacteria at different locations within a clump.
There have been numerous whole cell screens to identify small molecules in academic and industrial collections that kill nonreplicating mycobacteria (29, 53, 54, 94, 96, 101–105). Compounds arising from whole cell screening are presumably taken up into the cell to exert bactericidal activity, without any preconceptions about suitable targets. An alternate approach is to postulate which enzymes play a role in nonreplicating persistence based on informatics or biochemical or genetic studies, set up relevant biochemical assays, identify inhibitors, and then assay these inhibitors for whole-cell activity in nonreplicating models (100, 106–115). The limitation of biochemical screening, however, is that the majority of enzyme inhibitors so identified lack activity against intact M. tuberculosis due to poor uptake, the sufficiency of residual enzyme activity for cell survival, intracellular metabolism, or redundant pathways (116). Translating biochemical screening hits to whole cell activity is hampered by using a binary readout of life/death of a bacterial cell as a surrogate to monitor target engagement (117).
In screens carried out against nonreplicating bacteria, a failure to increase in optical density over time cannot be used as a measure of anti-bacterial activity since, by definition, the optical density does not change for the duration of a nonreplicating experiment. There are limited examples of screening by recording fluorescence from nonreplicating mycobacteria (29, 118, 119). While most replicating assays use an inoculum of ~ A580 of 0.01 or lower, use of a fluorescent readout can require a larger inoculum (upwards of ~ 50-fold) to achieve a sufficient signal (120). Using a high inoculum of cells may preclude identifying active molecules from compound classes such as beta-lactams, which are highly sensitive to inoculum effects (121). Moreover, nonreplicating screens employing fluorescent readouts often depend on subtle differences in the fluorescence of compound-treated versus vehicle-treated cells (often less than 2-fold), which in turn requires exceptional Z’ scores (29). In some nonreplicating assays, the drug-exposure phase of the assay is coupled to a drug-free secondary phase that permits bacterial growth, and allows one to make a semi-quantitative estimation of the number of surviving cells (Figure 1) (53, 54, 96). The two-stage assay, while effective, can take 14 to 17 days (a 7-day drug exposure, and a 7–10 day outgrowth), and runs a risk of evaporation, edge effects, and contamination with mold (30, 94).
One must carefully weigh the relative importance of potential variables when designing a high-throughput screen against nonreplicating mycobacteria (Figure 2). Table 2 provides a non-exhaustive list of potential microenvironments encountered by M. tuberculosis during infections that may lead to sub-optimal growth or nonreplication. The numerous nonreplicating models and technical variables lead to a staggering number of possible combinations.
The most commonly used models for nonreplicating mycobacteria are: hypoxia (the “Wayne model”) and the “low oxygen recovery assay” (“LORA”) (29, 95, 96); carbon starvation (54, 122); nutrient starvation (12, 52); stationary phase (105); maintenance of intrabacterial pH under acidic culture conditions (120, 123–127); biofilms (102, 128–131); depleting strain SS18b of streptomycin (103, 104, 132, 133); and a multi-stress model that combines acidic pH (pH 5.0), mild hypoxia (1% O2), nitric oxide and other reactive nitrogen intermediates (0.5 mM NaNO2), and a fatty acid carbon source (0.05% butyrate) (53, 94, 100, 134, 135). There are variations of these models, including an “acidic Wayne model” that combines hypoxia with mild acidity (131), and a nutrient-poor, multi-stress model in which cells are cultured at low pH (pH 5.0) under mild hypoxia or tissue-level normoxia (5% O2) and supra-physiologic levels of CO2 (10% CO2) (136). Sub-lethal doses of antibiotics targeting translation can also arrest growth (137). Potassium starvation has been reported to lead to the formation of differentially detectable mycobacteria (also called, “viable but not culturable”) (138, 139).
High throughput screening typically identifies many molecules with properties unsuitable for further progression, including those whose structures contain toxicophores and/or metabolic liabilities (140, 141). Comprehensive post-screening characterization of compounds from primary screens is extremely expensive in terms of time and resources. In Table 3, we summarize post-screening assays that are suitable for molecules with activity against replicating and/or nonreplicating mycobacteria. Table 4 summarizes assays used to characterize the action of compounds on nonreplicating mycobacteria. A hit progression flowchart for a nonreplicating active compound should attempt to include the assays described in both Tables 3 and and44.
As molecules progress from the initial high-throughput screen to in vivo models, there are mounting risks of wasting progressively larger amounts of time and money. Meticulous analysis of physicochemical and metabolic properties (140) of compounds is the norm in pharmaceutical companies, and unfortunately, often pursued insufficiently in academia due to lack of experience, personnel, funds, or access to expert chemistry advice, experimental analysis and synthesis (141).
Chemical structures can be misidentified as a result of inaccurate assembly of the compound library, erroneous dispensing of compounds, incorrect structure assignment, splashing of compounds between microtiter wells, compound degradation, and incomplete removal of reagents or catalysts used to synthesize the original compound, such as organotin, which can have antiseptic properties (142). For these reasons, it is critical to validate a molecule’s structure after cherry-pick confirmation and prior to initiating downstream hit characterization. Validation studies include testing a subset of screening hits for correct molecular mass, structure and purity by LC-MS and NMR. Prioritized molecules should be re-synthesized and re-evaluated in the original assay to confirm that they recapitulate the activity of the original hits. A surprisingly large number of screening compounds fail to meet these criteria.
Compound solubility is a problem at the forefront of high-throughput screening. The real and assumed concentrations of drug stocks can differ by several orders of magnitude (143–145). DMSO is hygroscopic and can absorb water from room air, leading to compound precipitation of water-insoluble compounds. Some DMSO-soluble compounds precipitate immediately, or over time, when transferred to assay plates containing aqueous media or buffers. Antimycobacterial compounds often have high logP values (146) that favor their precipitation in aqueous solution. Compound precipitation can lead to false-negative activity or to false-positive activity in optical density-based assays.
As scientists explore more diverse bacteriologic media for whole-cell screening to mimic in vivo microenvironments and stresses, another issue arises – the chemical stability of the compounds in the assay conditions. Careful determination of the structure of a molecule under the nonreplicating assay conditions is a critical, and often overlooked, step. As illustrated in Figure 3a, structures may be transformed by conditions found in nonreplicating assays, including acidic pH, reactive oxygen intermediates and reactive nitrogen species. If specific transformation products can be identified, they should be tested for activity in the original model of nonreplication and for their potential toxicity to eukaryotic cells. For example, oxyphenbutazone was chemically transformed in cell-free medium used in the multi-stress assay of nonreplication (Figure 3b) (53). In acidic medium containing reactive nitrogen species, the carbon on which the butyl chain attaches to the pyrazolidinedione ring was hydroxylated to form 4-hydroxy-oxyphenbutazone. This oxidation was followed by the pyrazolidinedione ring opening and formation of a quinoneimine. 4-Hydroxy-oxyphenbutazone’s quinoneimine, a Michael acceptor, reacted in vitro with glutathione and mycothiol (Figure 3b). In live M. tuberculosis, intracellular covalent adducts formed between 4-hydroxy-oxyphenbutazone and mycothiol, N-acetyl cysteine, and other uncharacterized metabolites. 4-Hydroxy-oxyphenbutazone killed replicating M. tuberculosis and mediated some of oxyphenbutazone’s activity against nonreplicating mycobacteria. In another example, three cephalosporin analogues (one of which was compound 68) with equipotent activity against M. tuberculosis rendered nonreplicating in the multi-stress model had different stability profiles: 2 were stable, and one was unstable (147). These results suggested that their uptake into M. tuberculosis occurred more rapidly than their extracellular transformation. As this example indicates, the relevance of cell-free stability should be evaluated on a case-by-case basis.
Replicating bacterial cultures fail to increase in biomass when treated with effective concentrations of either bacteriostatic or bactericidal molecules. This makes it relatively straightforward to recognize when compounds are active on replicating mycobacteria. Assays can be short in duration; there are many ways to assess viability; and false-positives are unlikely.
In contrast, determining the impact of compounds on nonreplicating cells is technically challenging. Nonreplicating conditions are themselves bacteriostatic, precluding the detection of viability using methods suitable for replicating cells. By coupling nonreplicating assays to a recovery phase under replicating conditions, one only obtains a rough estimation of a compound’s activity (Figure 1) (28, 53, 54, 94, 96). In the case of dual active molecules, one must ensure bona fide activity against the nonreplicating cells, or, alternatively, determine if the nonreplicating activity is an artifact of compound carry-over from the nonreplicating phase of the assay into the replicating phase of the assay (Figure 1) (30). Carry-over need not be via the fluid phase; compounds can absorb to mycobacterial components and be carried over to the replicating phase of the assay (30). As noted, drug carry-over was shown to be particularly troublesome for the extremely potent and extremely hydrophobic compound, TMC207 (30, 32, 148). This is not to deny that TMC207 has utility in animal (148, 149) and human (150) tuberculosis, and in fact, drug adsorption to mycobacteria may be a useful property. For example, compounds that associate with the bacterial cell wall may deliver a potent post-antibiotic effect (30) as they are slowly released into the intrabacterial cytosol. In vitro assays have arbitrary time points that are far shorter than clinical regimens and as such may grossly underestimate a drug’s bactericidal potential.
We and others have attempted to minimize carry-over effects from enumerating bacilli from in vitro assays or from organs harvested from antibiotic-treated, M. tuberculosis-infected animals (31–33). One solution was to include 0.4% (w/v) activated charcoal in bacteriologic agar plates to rapidly and completely bind the majority of first and second line anti-mycobacterial antibiotics (30).
Many compounds with nonreplicating activity were originally identified as highly potent replicating actives. Only a few studies confirmed their nonreplicating activity with a CFU assay, and almost never in the presence of activated charcoal or BSA in the agar plates. Given the challenge of testing large numbers of candidate dual active molecules by the CFU assay, the charcoal agar resazurin assay (CARA) was developed to rapidly categorize molecules as replicating-bacteriostatic, replicating-bactericidal, nonreplicating-bactericidal, or dual-active (that is, replicating bacteriostatic or bactericidal and nonreplicating bactericidal) (30). The CARA helps indicate which compounds should be explored by CFU assays and at which concentrations.
Molecules that reportedly kill class II phenotypically tolerant mycobacteria are structurally diverse. We have grouped approximately 100 such compounds according to core structure, targets, and/or method of discovery (Figures 4, ,66–11, and and1313).
The list of compounds was assembled with the intent of demonstrating both diversity and common themes. However, this is by no means a complete catalog. Additional actives can be found in databases such as SciFinder (https://scifinder.cas.org), PubChem (https://pubchem.ncbi.nlm.nih.gov/), and Collaborative Drug Discovery (https://www.collaborativedrug.com/) (151). Many other actives found in screening do not have their structures in scientific reports or deposited in public databases.
A relatively small number of compounds described to have bacteriostatic or bactericidal activity against replicating mycobacteria have also been tested for activity against nonreplicating bacteria. At best, most compounds were tested against mycobacteria in a single model of nonreplication. Many studies did not test the activity of the reported compounds with the gold standard CFU-based assay to determine viability.
Highly potent replicating actives may register as false-positives in nonreplicating assays due to compound carry-over from the nonreplicating phase to a replicating phase (see section “Evaluating bactericidal action against nonreplicating mycobacteria”). TMC207, which is active in multiple nonreplicating models, is an example of compound that has a high propensity for carry-over, clouding interpretation of results (30, 152). Some molecules are listed under more than one classification. For example, TMC207 is in three figures depicting structures: “Proof-of-concept molecules”, “Quinolines”, and “Membrane depolarizers”.
The distinction between bactericidal and bacteriostatic varies significantly in the literature. For this review, we define bacteriostatic as < 99% bacterial kill (< 2 log10 CFU) in ≤ 15 days (30, 153).
A major hurdle for large-scale commitment of resources towards identifying compounds that target nonreplicating bacteria is the paucity of examples that demonstrate the success of this approach. ADEP4 (compound 48), a synthetic acyldepsipeptide that dysregulates ClpP proteolysis, may serve as a prototype (154). ADEP4 killed both S. aureus class I persisters surviving ciprofloxacin exposure, and bacteria in three class II models of nonreplication: stationary phase, chemically defined minimal medium, and biofilms (155). ADEP4, when dosed with rifampicin, eradicated S. aureus in a mouse thigh infection. It is unknown to what extent the activity of ADEP4 against the replicating and nonreplicating populations contributes to its in vivo efficacy (155). In the following sections we explore examples of proof of concept molecules in mycobacteria.
There are few examples of molecules that fail to kill replicating mycobacteria, and kill those rendered nonreplicating in one or more in vitro models of nonreplication (Figure 4a). Although a limited number of molecules with pan-activity against nonreplicating mycobacteria have been identified, it is likely that more would emerge if the appropriate tests were performed. Since there is a paucity of common transcriptional responses among M. tuberculosis populations rendered nonreplicating in different in vitro models, such experiments are critical. The first example of a molecule with pan-activity against nonreplicating mycobacteria was published in 2008 (100). Bryk et al. identified a rhodanine, D157070 (compound 1) from a structure activity relationship campaign to develop a pro-drug inhibitor of dihydrolipoamide acyltransferase, DlaT (100, 156, 157). D157070 selectively killed M. bovis BCG and M. tuberculosis rendered nonreplicating by acidic pH and reactive nitrogen intermediates (34), hypoxia (95), a multi-stress model of nonreplication combining acidic pH, reactive nitrogen intermediates, hypoxia and restriction of the carbon source to a fatty acid (53, 94), human tissue culture medium (Dulbecco’s modified Eagle medium containing 10% FBS), and infection of bone marrow derived macrophages activated with IFNγ (100, 158).
Dual active molecules are defined as possessing bacteriostatic or bactericidal activity against replicating bacteria and bactericidal activity against nonreplicating bacteria. Of the dual actives, moxifloxacin (compound 2 (30, 50, 51, 96, 132, 159), PA-824 (compound 3 (30, 50, 51, 96, 132, 152, 160), rifampicin (compound 4 (30, 50, 51, 95, 96, 132, 152, 159–164) and TCM207 (compound 5, bedaquiline (30, 50, 51, 132, 150, 152, 159, 165, 166) are reported to kill mycobacteria rendered nonreplicating in diverse ways, including hypoxia, nutrient starvation, stationary phase, multi-stress, and deprivation of streptomycin from an addicted strain (Figure 4b). Moxifloxacin, PA-824, rifampicin and TMC207 target DNA gyrase, lipid/protein biosynthesis, RNA polymerase, and ATP synthase, respectively.
While none of the available in vitro or in vivo assays (including those described in Tables 3 and and4)4) can predict the efficacy of a compound in human tuberculosis, it is reasonable to prioritize compounds that have dual activity and potency against mycobacteria infecting macrophages and mice. In fact, a combination of three compounds in Figure 4b, PA-824, moxifloxacin, and PZA (PaMZ), showed promise in the NC001 clinical trial in tuberculosis patients (167). The 14-day NC001 EBA study was too brief to evaluate the impact of PaMZ on eradicating persisters and decreasing relapse rates (167). The Nix-TB clinical trial is evaluating the combination of TMC207, PZA, and linezolid on MDR- and XDR- tuberculosis, and may shed light on this question by increasing the duration of treatment up to 6–9 months (http://www.tballiance.org/portfolio/trials).
Of the molecules with selective activity against slowly replicating or nonreplicating mycobacteria, only metronidazole (compound 6) (95) and pyrazinamide (compound 7) (168) have been shown to be effective in animal models of tuberculosis (Figure 4c). As described in the introduction, metronidazole is bactericidal to hypoxic mycobacteria but this was only demonstrated under conditions in which no alternate electron acceptor was provided; killing was attenuated by inclusion of nitrate, a physiologic electron acceptor used by hypoxic M. tuberculosis and present in body fluids (169, 170). Activity of metronidazole did not correlate with evidence of lesion hypoxia (Table 1). Pyrazinamide is the sole representative of the nonreplicating-active class of compounds that is known to kill M. tuberculosis in humans, but pyrazinamide also kills slowly replicating mycobacteria in vitro. Pyrazinamide’s activity on M. tuberculosis under acidic conditions was enhanced by additionally including hypoxia (23) or a 3–10 day period of pre-adaptation to nutrient starvation in PBS (171). While its complete set of targets is still under investigation, pyrazinamide inhibits fatty acid biosynthesis by targeting FAS-I (172–174), trans-translation by targeting ribosomal protein S1 (RpsA, encoded by rv1630) (175) and pantothenate and coenzyme A by targeting aspartate decarboxylase (PanD, encoded by rv2601c) (176, 177).
The potency of many dual active molecules in Figure 4 was lower against nonreplicating bacilli than against mycobacteria replicating in standard bacteriologic medium. This could be due to less reliance on these processes during nonreplication, decreased uptake of the compounds due to a change in membrane composition and/or permeability (178), compound modification by the assay conditions (Figure 3) (28, 53), increased or altered intrabacterial metabolism of the compound (116), or sequestration of compound into lipid bodies that accumulate in mycobacteria in some in vitro models of nonreplication (28, 53, 136). Moreover, dual actives may kill nonreplicating mycobacteria by engaging non-canonical targets, or have a non-specific mechanism of action (Figure 5).
The canonical targets of many compounds that kill nonreplicating mycobacteria are in pathways for the biosynthesis of cell wall, lipids, RNA, DNA, protein or peptidoglycan (Figure 6). These compounds build a compelling case that nonreplicating mycobacteria engage in turnover of macromolecules. It is particularly encouraging that numerous antibiotic classes, some of whose members are approved for use in humans, including fluoroquinolones, rifamycins, macrolides, tetracyclines, and beta-lactams, have representatives that kill nonreplicating M. tuberculosis. This suggests that the anti-mycobacterial members of these families may likewise be tailored to display the pharmacokinetic and pharmacodynamic properties and low toxicities that allowed approval of the family members in clinical use (179).
Compounds that generate reactive oxygen species or reactive nitrogen species likely impact the function of numerous targets, including lipids, DNA, and the membrane. For example, PA-824 donates reactive nitrogen species (160). Sub-lethal nitric oxide induced a specific set of genes in the dos regulon, but higher concentrations of nitric oxide induced the expression of hundreds of other genes and implicated reactive nitrogen species in interfering with numerous processes (48, 180). In addition to engaging high affinity targets, compounds like PA-824 are probably promiscuous when they achieve higher intrabacterial concentrations.
High-affinity targets may exist that have an essential function unique to mycobacteria in a nonreplicating state. However, to date, we know of no instance in which differential expression or differential essentiality of a target has been shown to explain how a compound selectively kills nonreplicating mycobacteria.
A tetrahydrobenzothienopyrimidine (compound 8) targeting InhA killed M. tuberculosis rendered nonreplicating by hypoxia (181) (Figure 6a). That InhA might be an essential target during hypoxia is surprising. Isoniazid, which targets InhA, does not kill hypoxic M. tuberculosis or M. tuberculosis rendered nonreplicating in other conditions. Isoniazid is even used experimentally as a control compound to confirm that cells have achieved a state of nonreplication. This raises an important question why isoniazid fails to kill hypoxic, nonreplicating mycobacteria. The structure of isoniazid (likely the hydrazide moiety) may be unstable in hypoxia and/or other nonreplicating conditions (Figure 3a). Perhaps KatG fails to activate isoniazid under nonreplicating conditions. To test this hypothesis, InhA inhibitors that do not require KatG activation could be tested against nonreplicating bacilli (182). Another possibility is that compound 8, like isoniazid itself, may have more than one target, but unlike isoniazid, one of the alternate targets of compound 8 may be essential in hypoxia.
PA-824 is an inhibitor of lipid and protein synthesis (183, 184). While it has multiple targets, pyrazinamide is an inhibitor of lipid biosynthesis (172–174). Both PA-824 and pyrazinamide are described in the section, “Proof of concept molecules”.
Nonreplicating conditions may lead to oxidative stress, as may antibiotics with diverse primary targets (55, 66). DNA damage may result and survival may require DNA repair. Compounds that target DNA synthesis and kill nonreplicating mycobacteria are shown in Figure 6b. Inhibitors of topoisomerase I (TopA, encoded by rv3646) (compound 9) (113) and DNA gyrase B (GyrB, encoded by rv0005) (compounds 10 and 11) (108, 112) killed M. tuberculosis rendered nonreplicating by nutrient starvation, redox stress or hypoxia. Cyclohexyl griselimycin (compound 12), which targets the DnaN (encoded by rv0002) sliding clamp of DNA polymerase, is anticipated to kill nonreplicating M. tuberculosis due to its ability to reduce CFU during the persistent phase of murine tuberculosis (185–187). To our knowledge, the activity of cyclohexyl griselimycin against M. tuberculosis rendered nonreplicating by in vitro models has not been explored. Multiple fluoroquinolones, whose canonical targets are DNA gyrase and/or topoisomerase IV (188), including ciprofloxacin (compound 13) (51), gatifloxacin (compound 14) (51), levofloxacin (compound 15) (51), moxifloxacin (30, 50, 51, 96, 132, 159), and sparfloxacin (compound 16) (51), killed nonreplicating M. tuberculosis also described in the section “Quinolones”).
The transcriptomic adaptations of mycobacteria in nonreplicating conditions imply a requirement for RNA synthesis for their survival. Inhibitors of RNA polymerase (RpoB, encoded by rv0667 (Figure 6c), including rifampicin (described in “proof of concept molecules”) (30, 50, 51, 95, 96, 132, 152, 159–161), rifabutin (compound 17) and rifapentine (compound 18), killed nutrient starved and hypoxic M. tuberculosis (51, 96).
Mycobacteria are killed by a large number of compounds belonging to different structural classes and targeting diverse steps in protein biosynthesis (Figure 6d). The newly synthesized proteins may help mycobacteria detoxify or compensate for the stresses imposed by nonreplication. In mycobacteria, protein translation is vastly decreased, but not abrogated, during the first 40 days of nonreplication (189). Numerous compounds targeting the 30S and 50S components of the ribosomal machinery killed M. tuberculosis in multiple models of nonreplication, including deprivation of strain SS18b for streptomycin, hypoxia, and nutrient starvation. The protein synthesis inhibitors included the oxazolidinones linezolid (compound 19) (103) and sutezolid (compound 20) (103); the tetracycline minocycline (compound 21) (96); the aminoglycosides amikacin (compound 22) (51, 96), streptomycin (compound 23) (50, 51, 96), and kanamycin (compound 24) (51); an aminocyclitol antibiotic, the spectinamycin analogue 1599 (compound 25) (190); the cyclic peptide antibiotic capreomycin (compound 26) (50, 51, 96); the quinoline macrolide RU66252 (compound 27) (96); and fusidic acid, which prevents elongation factor G turnover and translocation (compound 28) (96). M. tuberculosis rendered nonreplicating by incubation for 2 months in stationary phase and then acidified to pH 5.5 under mild hypoxia (1% O2) was susceptible to methionine aminopeptidase inhibitors (compounds 29 (114, 115) and 30 (114)). PA-824, previously described as disrupting lipid biosynthesis, was also shown to inhibit protein synthesis (183).
Until recently, the dogma in the tuberculosis field was that M. tuberculosis was naturally resistant to beta-lactams. The two leading hypotheses were that beta-lactams failed to cross the mycobacterial outer membrane, and that beta-lactams were susceptible to beta-lactamase cleavage (191–195). Unexpectedly, there are now multiple examples of beta-lactams and other molecules targeting steps in peptidoglycan biosynthesis that kill nonreplicating mycobacteria (Figure 6e).
The canonical targets of beta-lactam antibiotics are enzymes catalyzing steps in peptidoglycan biosynthesis. The correlation between the bacterial replication rate and beta-lactam activity fostered the assumption that beta-lactams selectively target replicating bacteria (196). The choice of compounds for these studies led to the belief that activity was restricted to replicating cells, although some beta-lactams were identified that killed both replicating and nonreplicating Streptococcus pneumoniae and E. coli (56). The basis of dual activity remained a mystery for many years. Most bacteria contain murein predominantly composed of 4 → 3 transpeptides (197, 198). M. tuberculosis in stationary phase, or in hypoxia, had peptidoglycan enriched for 80% and 68% 3 → 3 crosslinks, respectively (98, 99). These studies suggest that the peptidoglycan layer in nonreplicating cells may be different than that of replicating cells, and if so, offers an underexplored set of target enzymes, such as the L,D-transpeptidases (199). A caveat to this conclusion, however, is that 3 → 3 crosslinks were also enriched in replicating M. tuberculosis (~ 62%) and may not be unique to nonreplicating mycobacteria (99).
A landmark paper in 2009 by Hugonnet et al. demonstrated that meropenem (compound 31), when paired the beta-lactamase inhibitor clavulanic acid (compound 32), killed replicating M. tuberculosis (195). The Hugonnet study made a critical, and unanticipated, discovery – that the combination of meropenem and clavulanate also killed hypoxic, nonreplicating M. tuberculosis (195). A 14-day trial demonstrated that meropenem, amoxicillin and clavulanate had marked early bactericidal activity in human tuberculosis (200).
A number of other molecules have been reported to target nonreplicating mycobacteria by disrupting steps of peptidoglycan biosynthesis, including: L,D-transpeptidases by faropenem (compound 33) (38, 201); UDP-galactopyranose mutase (compound 34) (202); phospho-N-acetylmuramoyl-pentapeptidetransferase (MurX) by CPZEN-45 (compound 35) (203); and the capuramycin analogue UT-01320 (compound 36) (204). CPZEN-45 may additionally target decaprenyl-phosphate-GlcNAc-1-transferase (WecA, encoded by rv1302), which has a role in synthesizing teichoic acid in Bacillus subtilis and mycoylarabinogalactan in mycobacteria (205). UT-01320 (compound 36) was bactericidal to both hypoxic and nutrient-starved M. tuberculosis (204). The structure of capuramycin (compound 37) illustrates an example in which replacing a hydroxyl group with an O-methyl (UT-01320, yellow highlighted carbon atom) confers nonreplicating activity on a molecule whose activity was restricted to replicating mycobacteria. However, UT-01320’s activity profile change was accompanied by the failure to inhibit MurX in vitro and suggests that its ability to kill nonreplicating M. tuberculosis may have been due to engaging a different target (204).
We did not find reports of nonreplicating mycobacteria being killed by inhibitors of folate biosynthesis.
There are numerous examples of quinolines and their derivatives that kill nonreplicating mycobacteria (Figure 7).
Maintaining ATP levels is critical for mycobacteria to survive the nonreplicating state (166). TMC207, which has a quinoline core, is bacteriostatic to replicating mycobacteria, and has been reported to kill those rendered nonreplicating by hypoxia, nutrient starvation, and streptomycin removal from the addicted strain SS18b (30, 50, 51, 132, 150, 152, 159, 165, 166). Due to its hydrophobic nature (logP of 7.3) and nanomolar potency, TMC207 is subject to carry-over artefacts that make it challenging to evaluate its activity in nonreplicating models (30). To our knowledge, TMC207’s activity against mycobacteria in nonreplicating models has not been established by CFU enumeration under conditions that prevent drug carry-over, such as the presence of 0.4% (w/v) activated charcoal in the bacteriologic agar (described in “Evaluating bactericidal action against nonreplicating mycobacteria”). In our own studies, use of activated charcoal eliminated most of the apparent activity of TMC207 against M. tuberculosis in a multi-stress model of nonreplication (30).
Another ATP synthase inhibitor, the substituted chloroquinoline compound 38 (161, 206) was reported to be bactericidal to hypoxic M. tuberculosis. The antimalarial drug mefloquine (compound 39) was reported to kill M. tuberculosis rendered nonreplicating by hypoxia and nutrient starvation (51). Mefloquine targets ATP synthase in Streptococcus pneumoniae (207, 208), and its target in M. tuberculosis is currently not known.
8-hydroxyquinolines have antibacterial activities on M. tuberculosis in vitro and in vivo (209, 210). The unsubstituted 8-hydroxyquinoline (compound 40) killed replicating M. tuberculosis and was bactericidal to M. tuberculosis rendered nonreplicating by mild hypoxia (1% O2), acid (pH 5.5), and acidic nitrosative conditions (pH 5.5 containing 0.5 mM NaNO2) (101). Compound 30, a 5-chloro, 7-bromo-substituted 8-hydroxyquinoline inhibitor of methionine aminopeptidase, killed M. tuberculosis that had been in stationary phase for two months (114). Some 8-hydroxyquinolines may mediate toxicity by chelating metals essential for mycobacterial survival, or by forming metal-quinoline complexes that engage target enzymes (211).
Phenotypic screening led to the discovery of mefloquine-like molecules with a 4-hydroxyquinoline core (compounds 41 and 42) that killed both replicating and hypoxic, nonreplicating M. tuberculosis 212–215). Nontoxic isoxazolecarboxylic acid ethyl esters analogues of Compounds 41 and 42, represented by compound 43, killed replicating and nonreplicating hypoxic M. tuberculosis (215, 216). 2-substituted 4-hydroxyquinolines (compounds 44 and 45 inhibitors of triacylglycerol lipase LipY (encoded by rv3097c) killed hypoxic M. tuberculosis (217). The activity of compounds 44 and 45 was specific to recovery of dormant M. tuberculosis, in which LipY plays an essential role in triacylglycerol breakdown (218). With our colleagues at Sanofi, U. North Carolina, Memorial Sloan Kettering, and the Lankenau Institute, we identified and characterized 4-hydroxyquinolines with bactericidal activity against mycobacteria in the multi-stress model of nonreplication (data not shown).
Numerous fluoroquinolones kill nonreplicating M. tuberculosis (Figure 8). As mentioned, moxifloxacin kills mycobacteria in numerous nonreplicating models (30, 50, 51, 96, 132, 159). Similar to moxifloxacin, other fluoroquinolones possess activity against hypoxic, nonreplicating M. tuberculosis, including ciprofloxacin (51), gatifloxacin (51), levofloxacin (51) and sparfloxacin (51).
Inhibitors of the mycobacterial proteolysis and proteostasis pathways are selectively active against bacilli that are nonreplicating or have dual activity (Figure 9). In addition to turnover of undamaged proteins, the proteolysis pathway also degrades damaged or misfolded proteins that result from exposure to stresses such as reactive nitrogen intermediates or oxidative damage (55, 219). The proteostasis pathway helps insure proper folding of nascent, misfolded, or damaged proteins with the aid of chaperone proteins and refoldases (220). Some stresses that contribute to nonreplicating persistence engage the proteolysis and proteostasis pathways (Table 2) (55, 66, 219). As described previously, the acyldepsipeptide ADEP4 (compound 49) forces a bacterium to eat itself by unregulated digestion of intrabacterial proteins (155). Unlike the PrcBA proteasome (encoded by rv2109c and rv2110c), the ClpP1P2 proteolytic machinery (encoded by rv2461c and rv2460c) is essential for M. tuberculosis to survive under replicating conditions (221). The dual active molecules cyclomarin A (compound 46) (222), ecumicin (compound 47) (51, 132, 159, 166, 222, 223), and lassomycin (compound 48[cyclic(GLRRLFAD)]-QLVGRRNI-CO2CH3) (223–225), target the ClpP1P2 pathway by engaging ClpC, and kill M. tuberculosis rendered nonreplicating by hypoxia or stationary phase (222, 223, 225, 226). Ecumicin stimulates ClpC’s ATPase and impairs its activator activity, while sparing ClpP1P2’s proteolytic activity (223, 226). This results in depletion of cellular ATP (see, “Membrane depolarizers”). While not essential for logarithmic growth, the M. tuberculosis proteasome is required to survive stationary phase and long-term nutrient starvation (227, 228). Small molecule inhibitors targeting the proteasome, such as GL-5 (compound 50, an irreversible inhibitor) (106) and DPLG-2 (compound 52, a reversible inhibitor) (229), killed M. tuberculosis rendered nonreplicating with acidified nitrite. Compound 51, an analogue of GL-5, had bactericidal activity against a 2-week nutrient starved culture of M. tuberculosis (230). These studies indicate that targeting the ClpP1P2 and PrcAB proteolysis pathways are good strategies to kill nonreplicating persisters.
There have been a limited number of phenotypic high-throughput screens to discover molecules that kill nonreplicating mycobacteria (Figure 10). The following non-exhaustive set of examples illustrates screening methods and the molecular diversity of the hits.
Grant et al. identified compounds that killed carbon-starved, nonreplicating M. tuberculosis (54) (Figure 10a). The screen was notable for testing a set of confirmed actives against M. tuberculosis in different class I and class II models of phenotypic tolerance. This strategy uncovered that the hits had diverse activities: some molecules were active only in the carbon starvation model used for the screen (compounds 53 and 54); others were active in carbon starvation and hypoxia models (compounds 55, 56, and 57; and compound 58 was bactericidal to M. tuberculosis rendered nonreplicating by carbon starvation or hypoxia as well as to M. tuberculosis in a class I persister model.
As noted, metronidazole kills hypoxic M. tuberculosis in vitro and can reduce CFU in some animal models of tuberculosis Table 1). Mak et al. set out to identify more molecules that killed nonreplicating, hypoxic M. bovis BCG (29) (Figure 10b). The hypoxia screen was performed using a recombinant M. bovis BCG strain expressing the M. tuberculosis narGHJI operon. This strategy artificially increased ATP levels by augmenting BCG’s use of nitrate as an electron acceptor and made it easier to identify active molecules by their ability to decrease cellular ATP content. Active molecules of the benzamidazole, imidazopyridine and thiophene classes were confirmed by a CFU assay (compounds 59, 60 and 61).
One strategy to identify molecules that kill nonreplicating mycobacteria is to combine various stresses encountered by M. tuberculosis during infections of macrophages, animals, and humans (49, 53, 125, 231–247). A combination of stresses is anticipated to capture a greater diversity of screening hits than a single stress. This strategy requires a secondary hit deconvolution to identify which stress(es) are essential for a given compound’s activity. Our laboratory used this strategy to develop a multi-stress assay of nonreplication, in which M. tuberculosis’s replication was halted by mild acidity (pH 5.0), nitric oxide and other reactive nitrogen intermediates generated by NaNO2 at acidic pH, a fatty acid carbon source (butyrate), and mild hypoxia (1% O2). Examples of molecules active in the multi-stress model of nonreplication (Figure 10c) include compounds 62, 63, 64, 65, and 66 (28), oxyphenbutazone (compound 67) (53), nitrofuranylcalanolide (compound 81) (134) (described further in “nitro-containing compounds”), and cephalosporins 68 and 69 (147). The cephalosporins, compounds 68 and 69, bearing a C-2 alkyl ester or oxadiazole had clavulanate-independent bactericidal activity against M. tuberculosis in a multi-stress model of nonreplication (147) and killed M. tuberculosis infecting IFNγ-activated bone marrow mouse macrophages. The unusual structures and activity profiles of compounds 68 and 69 suggests that they may engage non-canonical targets (248–251).
The intraphagosomal environment of an M. tuberculosis-infected, IFNγ-activated mouse macrophage decreases to approximately pH 4.5 (233). Furthermore, M. tuberculosis may create local acidic environments by secreting succinate when cultured under hypoxic conditions (22, 170). Using an intrabacterial pH-sensitive green fluorescent protein, Darby et al. screened for compounds that disrupted M. tuberculosis’s intrabacterial pH during incubation in growth-restricting conditions of phosphate-citrate pH 4.5 buffer (119, 123, 125) (Figure 10d). Compounds 70, 71, 72, 40 and 73, identified in this screen, had acidic-pH dependent bactericidal activity. High throughput antibacterial screens, in which bacteria are incubated in an acidic pH, often identify protonophores as nonspecific disrupters of the membrane proton gradient. Monensin (compound 88 (120)) is one such example and is described in more detail in the section “Membrane depolarizers”.
Many pathogenic bacteria form biofilms as a survival strategy to evade host immunity and antibiotics (252, 253). Mycobacteria form drug tolerant biofilms in vitro (129, 254), although the relevance of mycobacterial biofilm formation in human disease is unknown. Wang et al. screened for compounds that killed nonreplicating M. smegmatis that presented as a pellicle at the interface of the bacteriologic medium and air. They identified a replicating-active molecule, TCA1 (compound 74) (Figure 10e), that potently inhibited biofilm formation by M. smegmatisM. tuberculosis, and M. bovis BCG, and killed M. tuberculosis in a nutrient starvation model (102). TCA1 targets both decaprenyl-phosphoryl-β-D-ribose 2-oxidase (DprE1, encoded by rv3790), which synthesizes decaprenylphosphoryl-β-D- arabinose, and MoeW (encoded by rv2338c), an enzyme involved in the synthesis of a molybdenum-containing prosthetic group, the molybdenum cofactor. Under nutrient starvation conditions, overexpression of MoeW conferred partial resistance to TCA1 (102)
A number of dual- and nonreplicating-active molecules contain a nitro functional group (Figure 11). The strongly electronegative nitro moiety has varying degrees of reactivity that depend on its local environment within a molecule. Mycobacterial nitroreductases may catalyze the conversion of the nitro group (-NO2), by first reducing it to a nitroso (-NO), then to a hydroxylamine (-NHOH), and finally to an amine (-NH2) (255, 256). Nitrofurans and nitroimidazoles undergo intrabacterial bioactivation to metabolites that can redox-cycle and cause oxidative damage, and/or dismutate to an electrophilic nitroso intermediate that reacts with intracellular thiols (257, 258). Reactive molecules such as nitrofurantoin may have specific, high affinity targets such as ribosomal proteins, yet become promiscuous at higher intrabacterial concentrations (258). The nitroso radical anion causes DNA oxidation and cell death due to DNA strand breaks (257). PA-824 donates a nitric oxide-like species that may have numerous intrabacterial targets (160, 180).
Examples of nitro-containing molecules that kill nonreplicating mycobacteria include metronidazole (19, 159, 161, 259); furaltadone (compound 75) (259); nitrofurantoin (compound 76) (96, 259); nitrofurazone (compound 77) (259); furazolidone (compound 78) (96); nitazoxanide (compound 79) (34, 50, 260); niclosamide (compound 80) (96); compound 9 (113); nitrofuranylcalanolide (compound 81) (134); OPC-67683 (compound 82) (261, 262); PA-824 (50, 51, 96, 132, 152, 160, 263); TBA-354 (compound 83) (262); and compound 84 (264).
While nitroimidazole-containing molecules such as PA-824 and OPC-67683 are in clinical trials, molecules containing the structurally similar nitrofuran moiety are considered hazardous due to the risk of DNA mutagenicity (265). Even if a molecule is non-toxic to eukaryotic cells, nitroimidazoles, nitrofurans, and nitrothiophenes should be assayed for mutagenicity (Table 4). The nitrofuranylcalanolide (compound 81) has potent MIC90 values against replicating and nonreplicating M. tuberculosis, and kills M. tuberculosis infecting human macrophages (134). However, the nitrofuranylcalanolide was genotoxic in vitro (134).
Some nitro-containing molecules, such as nitazoxanide and niclosamide, are described in the “Membrane depolarizers” section.
Nonreplicating mycobacteria maintain an energized membrane (20). During hypoxia, depolarizing the mycobacterial membrane results in the inability to synthesize ATP by disrupting the intracellular to extracellular proton gradient (159, 166). Collapsing the proton motive force, and depleting intrabacterial ATP levels, may be a common feature of molecules that target replicating and nonreplicating mycobacteria (Figure 12) (266). The structure of a compound may not provide obvious clues that its anti-infective activity results from non-specific depolarization of the bacterial membrane. For example, some salicylanilides such as niclosamide shuttle protons by forming a stable 6-membered ring via hydrogen bonding (Figure 13) (267–269). The hydrogen-bonded form of salicylanilide has a higher logP than its non-hydrogen bonded structure, which is predicted to improve penetration across the lipophilic bacterial membrane. After passing into the bacterial cytosol and releasing its proton, the negatively charged salicylanilide anion, which is internally hydrogen bond stabilized, can return across the membrane again to pick up another proton (268).
Compounds that kill nonreplicating mycobacteria by depolarizing the membrane include the valinomycin (compound 85) (29, 159); nigericin (compound 86) (29, 132, 159); boromycin (compound 87) (270); nitazoxanide (34, 50, 260); monensin (compound 88) (120); salinomycin (compound 89) (132); niclosamide (96); a derivative of α-mangostin, A-0016 (compound 90) (271); N,N’-dicyclohexylcarbodiimine (DCCD, compound 91) (159); clofazimine (compound 92) (50, 55, 96, 272); and thioridazine (compound 93) (29, 51, 159, 161). The pro-drug clofazimine is reduced by type 2 NADH-quinone oxidoreductase. Clofazimine likely kills by interfering with mycobacterial respiration by generation of hydroxyl radicals (55, 273). The bactericidal action of clofazimine is reversed by addition of menaquinone-4 (MK-4) (273, 274). The phenothioazines thioridazine and trifluoperazine (compound 94) target the electron transport chain, disrupt membrane potential, and ultimately decrease ATP levels (275). Trifluoperazine was bactericidal to M. tuberculosis that was nonreplicating when cultured at acidic pH, in nutrient starved cultures, or at neutral pH in the presence of the nitric oxide donor DETA-NO (275). As described in the section on quinolines, TMC207 targets ATP synthase (AtpE, encoded by rv1305). Compound 95, which kills hypoxic M. tuberculosis, targets MenA (encoded by rv0534c) (110). Inhibition of MenA, which catalyzes the prenylation of 1,4-dihydroxy-2-naphthoate to demethylmenaquinone, is anticipated to impact membrane potential and decrease ATP levels (110). N-octanesulphonylacetamide (OSA, compound 96) killed both replicating and hypoxic, nonreplicating M. tuberculosis (276–279). The rapid depletion in ATP levels after treating mycobacteria with compound 97 was consistent with inhibition of the ATP synthase and/or the respiratory chain. The compound n-decanesulfonylacetamide (DSA, similar to compound 96) inhibits ATP synthase and lipid biosynthesis (276, 277, 279), killed hypoxic M. bovis BCG, and had bactericidal activity against class I rifampicin tolerant persisters (278). Q203 (compound 97) targets the cytochrome bc1 complex (QcrB, encoded by rv2196), depletes ATP levels, kills replicating and hypoxic M. tuberculosis, and has activity in a mouse model of tuberculosis (280).
Dual active protonophores boast low frequencies of resistance. Membrane depolarization, and ensuing depletion of ATP, has an amplified downstream impact on the cell that cannot be easily compensated for by mutating a single gene. In the cases of nitazoxanide, boromycin, and AM-0016, the frequencies of resistance were estimated as < 1×10-12 (34), < 1×10-9 (270), and < 1×10-8 (271), respectively. Low frequencies of resistance are consistent with a non-specific mechanism of action or multiple specific mechanisms of action (281).
ATP depletion can lead to persister formation in S. aureus (282). It will be of interest to determine if clinical use of protonophores and membrane-targeting molecules selects for live, ATP-depleted, persister populations that are tolerant to other antibiotics.
Many compounds that are reported to kill nonreplicating mycobacteria do not fall into the categories described above (Figure 14). Auranofin (compound 98), a drug used to treat rheumatoid arthritis, was recently described as a potent inhibitor of mycobacterial thioredoxin reductase TrxB2 (encoded by rv3913) and had bactericidal activity against nutrient starved M. tuberculosis (283). While auranofin probably forms an Au(I) adduct with the cysteine active site of thioredoxin reductase, it has additional thiol targets (276, 283, 284). Compound 99 inhibits CysM (encoded by rv1336), which has a critical role in mycobacterial cysteine biosynthesis, and killed nutrient-starved M. tuberculosis (107). DevR (also called DosR, encoded by rv3133c) controls expression of genes in response to hypoxia (109, 285). A DevR inhibitor, compound 100, killed M. tuberculosis that was cultured under hypoxic conditions, but spared the bacilli when they were nonreplicating during nutrient starvation (286). A 2-thiopyridine, compound 101, killed both replicating and nonreplicating hypoxic and nutrient-starved M. tuberculosis (287). The 2-thiopyridines were bactericidal to viable-but-not-culturable M. tuberculosis in a potassium starvation model (287). Compound 102, an inhibitor of pantothenate synthase (PanC, encoded by rv3602c) killed M. tuberculosis in a nutrient starvation model, and killed M. marinum infecting zebrafish (111). Compound 103 killed hypoxic M. tuberculosis (288). A quinoxaline 1,4-di-N-oxide, compound 104, displayed potent activity against replicating and hypoxic nonreplicating M. tuberculosis and was active in an acute mouse model (289).
Phenotypic tolerance is complex. We must refrain from projecting the biology deciphered in one persistence model to other persistence models, for fear that we oversimplify phenotypic tolerance, much as Jacques Monod oversimplified biochemical unity when he declared, “anything found to be true of E. coli must also be true of elephants” (290). We need to focus time and resources on developing antibiotics that target mycobacteria in in vitro states that have relevance to persisters found during human tuberculosis, whose structures are accurately understood under the conditions of the assays, that exert bona fide bactericidal activity against nonreplicating mycobacteria, and that have a structure-activity relationship which enables at least one of its derivative compounds to make it through the gauntlet of drug development, or at least, to become an informative tool for chemical biology to guide our understanding of bacterial persistence (291, 292).
The challenges are formidable. Our confidence in in vitro screens is shaken by the dissociation between the high value of pyrazinamide in the clinic and its lack of activity in vitro at the highest concentrations typically considered appropriate for screening. We know many ways to put M. tuberculosis into a nonreplicating state in vitro but are not confident which states model the class II phenotypic tolerance that is hypothesized to contribute to the inefficiency of conventional drugs against tuberculosis in the human host. We need to validate compounds in animal models but are not sure which animal models (if any) qualitatively and quantitatively mimic which aspects of the human disease (42, 293). It is commonly noted that all current tuberculosis drugs work in mice, but compounds inactive in mouse models of tuberculosis have, to our knowledge, not been tested in primates or humans. A number of compounds active against replicating or nonreplicating M. tuberculosis in vitro and displaying suitable pharmacokinetics for studies in mice have failed to show efficacy in mice, but unless they were toxic, we almost never learn why they failed, or if the question is answered, the answer is almost never published. Such failures fail again when little or no insight is gained from the investment that led to the in vivo tests.
Two recent developments nonetheless justify optimism. First, academic microbiologists, immunologists, biochemists, structural biologists and physicians interested in these challenges are now working alongside professional medicinal chemists and pharmacologists. Second, these partnerships involve multiple academic institutions and multiple pharmaceutical companies, and the participants share their questions, hypotheses, failures and successes as they go along.
We are thankful to Drs. Thulasi Warrier, Selin Somersan, Landys Lopez-Quezada, Kristin Burns-Huang, and Kyu Rhee (Weill Cornell Medicine, New York, NY) and Drs. Christine Roubert, Laurent Gouilleux, Laurent Fraisse, and Cédric Couturier (Sanofi, Lyon, France) and Alfonso Mendoza-Losana, David Barros and Robert Bates (GlaxoSmithKline, Tres Cantos, Spain) for insightful discussions and ideas. We are grateful to Drs. Steven Brickner (SJ Bricker Consulting, LLC), Jeff Aubé (U. North Carolina), and Ouathek Ouerfelli (Memorial Sloan Kettering, NY, NY) for invaluable discussions, ideas, and chemistry expertise. We thank SB for bringing the chemistry of salicylanilides to our attention. We thank TW, SS, KBH, LLQ, KR and SB for careful editing of the manuscript. We thank our colleagues (CR, LG, LF and CC at Sanofi (Lyon, France), JA at U. North Carolina (Chapel Hill, North Carolina, USA), OO and GY at Memorial Sloan Kettering (New York, NY), and Mel Reichman at the Lankenau Institute (Wynnewood, PA)) for sharing preliminary data on 4-hydroxyquinolines. The work fostering these ideas was supported by the TB Drug Accelerator of the Bill and Melinda Gates Foundation, the Abby and Howard P. Milstein Program in Chemical Biology and Translational Medicine, and NIH TB Research Unit grant U19 AI111143. The Department of Microbiology and Immunology is supported by the William Randolph Hearst Foundation.