Preclinical Efficacy Testing of New Drug Candidates

ABSTRACT

This is a review of the preclinical efficacy testing of new antituberculosis drug candidates. It describes existing dynamic in vitro and in vivo models of antituberculosis chemotherapy and their utility in preclinical evaluations of promising new drugs and combination regimens, with an effort to highlight recent developments. Emphasis is given to the integration of quantitative pharmacokinetic/pharmacodynamic analyses and the impact of lesion pathology on drug efficacy. Discussion also includes in vivo models of chemotherapy of latent tuberculosis infection.

INTRODUCTION

Since the early 1950s, combination chemotherapy has remained the strongest line of defense against the ancient scourge of tuberculosis (TB). Between the years 2000 and 2015 alone, it was estimated that TB treatment averted 39 million deaths among people without HIV infection and, together with antiretroviral therapy, another 9.6 million deaths among people with HIV infection (1). Despite these successes, TB continues to exert a terrible toll on humanity. In 2015, TB was estimated to be the cause of 10.4 million new cases and 1.4 million deaths, making Mycobacterium tuberculosis the leading microbial cause of death in the world (1). The failure to achieve greater control of TB over the past half century is partly attributable to several important limitations of current chemotherapy regimens, including the prolonged treatment durations necessary to prevent relapse after treatment completion and the inability to effectively suppress resistance emergence when treatment is applied on a global scale. These deficiencies are especially notable for current second-line and salvage regimens used to treat drug-resistant TB (2–4), which are also complicated by excessive toxicity, poor tolerability, high cost, and the inconvenience of injections, multiple daily doses, and large pill burdens. As a result, shortening or otherwise simplifying regimens to treat TB without sacrificing efficacy is a major goal of TB drug development research (5).

Successful treatment of active TB requires the use of drug combinations to efficiently eradicate the diverse population of bacteria present in the infected host, including actively replicating bacilli, smaller subpopulations of persistent bacilli that are phenotypically tolerant to the bactericidal action of drugs, as well as small subpopulations of spontaneous drug-resistant mutants (6–9). Shortening the duration of TB treatment requires more rapid elimination of drug-tolerant persisters, while preventing the development of drug resistance requires more effective killing of spontaneous drug-resistant mutants by the remaining drugs in the regimen to which the mutants remain susceptible. Although promising results from recent phase 2 trials suggest that optimizing the dosing of existing rifamycin drugs may significantly shorten the duration of therapy for drug-susceptible TB (10, 11), a clear need exists for new drugs with treatment-shortening effects driven by novel mechanisms of action, especially against rifampin-resistant forms of TB. Such new drugs may also be expected to shorten the duration of treatment for latent TB infection, which would remove a major impediment to greater implementation of treatment strategies aimed at preventing development of active disease.

Fortunately, after a drought of over 40 years in which no new class of drugs was approved for use against TB, renewed discovery efforts and repurposing of existing antibacterial agents have produced a global portfolio of new drug candidates for TB. The portfolio of agents that are currently under evaluation in clinical trials is summarized in Table 1. It includes agents from six novel classes that are not approved or in development for other indications, including the newly approved drugs bedaquiline and delamanid. Other previously approved TB drugs are being re-examined to determine whether more effective dosing strategies will improve their contribution to TB treatment. Promising compounds also continue to emerge at earlier stages of discovery and preclinical development. A curated description of this preclinical portfolio is available at http://www.newtbdrugs.org/pipeline/discovery. The overarching objective of these efforts is to develop novel drug regimens containing one or more new agents capable of shortening or otherwise simplifying the treatment of drug-susceptible as well as drug-resistant forms of active and latent TB infection.

TABLE 1.

New drugs in clinical development for the treatment of active TB, including ongoing clinical trials and the preclinical evidence base that supports each trial^a

Drug (abbreviation)	Class	Mechanism(s) of action	Target	Trial objective (clinicaltrials.gov identifier)	Preclinical evidence base (references)
Existing TB drugs in dose optimization studies
Isoniazid (INH)	Nicotinamide analog	Inhibition of mycolic acid synthesis	2-trans-Enoyl-acyl carrier protein reductase (InhA)	Dose-ranging EBA vs. INH-resistant mutants (NCT01936831)	19, 150, 151
				EBA of high-dose INH vs. INH-resistant mutants (NCT02236078)
Rifampin (RIF)	Rifamycin	Inhibition of RNA synthesis	DNA-dependent RNA polymerase (RpoB)	Dose-ranging activity of RIF in the intensive phase (NCT00760149, NCT01408914, NCT02153528)	110, 152, 153
				Dose-ranging activity of RIF in a 4-month regimen (NCT02581527)
Rifapentine (RPT)	Rifamycin	Inhibition of RNA synthesis	DNA-dependent RNA polymerase (RpoB)	Efficacy of 4-month regimens based on high-dose RPT (NCT02410772)	153–155
Repurposing of anti-infectives for TB treatment
Moxifloxacin (MXF)	Fluoroquinolone	Inhibition of DNA synthesis	DNA gyrase (GyrA)	Efficacy of 4-month regimens based on high-dose RPT with or without MXF (NCT02410772)	155, 156
				Efficacy of MXF in place of EMB in retreatment of TB (NCT02114684)	73, 74, 157, 158
				Efficacy of regimens containing BDQ, PA-824, and PZA ± MXF (NCT02193776)	77, 78
				Efficacy of 4- and 6-month regimens containing PA-824, MXF, and PZA (NCT02342886)	75, 77, 109
				EBA of MXF vs. ofloxacin-resistant mutants (NCT02236078)	159–161
				Efficacy of MDR-TB regimens containing BDQ, PA-824, LZD ± MXF, or CFZ (NCT02589782)	E. Nuermberger, unpublished; 70*, 76
Levofloxacin (LVFX)	Fluoroquinolone	Inhibition of DNA synthesis	DNA gyrase (GyrA)	Dose-ranging activity of LVFX in MDR-TB (NCT01918397)	161–163
				Efficacy of MDR-TB regimen containing DLM, LZD, LVFX, and PZA (NCT02619994)	76
				Efficacy of a 4.5-month regimen containing 1st-line drugs + LVFX (NCT02901288)	157
Clofazimine (CFZ)	Riminophenazine	Redox cycling, production of reactive oxygen species	N/A	Efficacy of a short-course CFZ-containing regimen for MDR-TB (NCT02409290)	164
				Efficacy of MDR-TB regimens containing BDQ, PA-824, LZD ± MXF, or CFZ (NCT02589782)	Nuermberger, unpublished; 70, 76
Linezolid (LZD)	Oxazolidinone	Inhibition of protein synthesis	Ribosomal initiation complex	Dose-ranging EBA of LZD (NCT02279875)	29, 33, 40, 165
				Efficacy of a short-course regimen of BDQ, PA-824, and LZD for MDR/XDR-TB (NCT02333799)	76
				Efficacy of MDR-TB regimen containing DLM, LZD, LVFX, and PZA (NCT02619994)	76*
				Efficacy of MDR-TB regimens containing BDQ, PA-824, LZD ± MXF, or CFZ (NCT02589782)	Nuermberger, unpublished; 70*, 76
New chemical entities in clinical development for TB
Bedaquiline (BDQ)	Diarylquinoline	Inhibition of ATP synthesis	F1F0 proton ATP synthase (AtpE)	Efficacy of a short-course regimen of BDQ, PA-824, and LZD for MDR/XDR-TB (NCT02333799)	76
				Efficacy of regimens containing BDQ, PA-824, and PZA ± MXF (NCT02193776)	77, 78
				Efficacy of MDR-TB regimens containing BDQ, PA-824, LZD ± MXF, or CFZ (NCT02589782)
				Efficacy of a short-course CFZ-containing regimen for MDR-TB (NCT02409290)	Nuermberger, unpublished; 70*, 76
Delamanid (DLM)	Nitroimidazo-oxazole	Inhibition of mycolic acid synthesis, production of reactive nitrogen species	Unknown	Efficacy of MDR-TB regimen containing DLM, LZD, LVFX, and PZA (NCT02619994)	76*
Pretomanid (PA-824)	Nitroimidazo-oxazine	Inhibition of mycolic acid synthesis, production of reactive nitrogen species	Unknown	Efficacy of a short-course regimen of BDQ, PA-824, and LZD for MDR/XDR-TB (NCT02333799)	76
				Efficacy of regimens containing BDQ, PA-824, and PZA ± MXF (NCT02193776)	77, 78
				Efficacy of MDR-TB regimens containing BDQ, PA-824, LZD ± MXF, or CFZ (NCT02589782)	Nuermberger, unpublished; 70*, 76
SQ109	Diethylamine	Dissipation of proton motive force, inhibition of menaquinone synthesis	Mycobacterial membrane protein-large 3 (MmpL3), MenA, MenG	Completed phase 1
				No active trials**
Sutezolid (SZD)	Oxazolidinone	Inhibition of protein synthesis	Ribosomal initiation complex	Completed phase 1
				No active trials**
PBTZ169	Benzothiazinone	Inhibition of arabinogalactan synthesis	Decaprenylphosphoryl-b-d-ribose	Phase 1
			2′-epimerase (DprE1)
OPC-167832	3,4-Carbostyril derivative	Inhibition of arabinogalactan synthesis	Decaprenylphosphoryl-b-d-ribose	Phase 1
			2′-epimerase (DprE1)
Q203	Imidazopyridine amide	Inhibition of electron transport chain	Cytochrome bc1 complex (QcrB)	Phase 1

^a *, regimen(s) tested in pre-clinical study and clinical trial may differ by at least one drug in the same class. **, no active trials registered on clinicaltrials.gov (last confirmed on May 18, 2017). EBA, early bactericidal activity; PZA, pyrazinamide.

As for most infectious disease indications, drug and regimen developers focused on TB rely on preclinical models to provide evidence of efficacy suitable for selecting and advancing new drugs and drug regimens and for informing clinical trial designs. Indeed, given the many challenges inherent in the clinical development pathway for new TB therapies, including limited financial resources for phase 3 trials and lack of reliable biomarkers predictive of long-term efficacy in such trials (12), TB drug development efforts may rely more on preclinical models than many other infectious indications. This dependency and the recent availability of new investigational and repurposed drugs have brought increased scrutiny to preclinical efficacy testing and its clinical translation (12–14). Currently, there is no consensus on what preclinical tools and studies might constitute a “critical path” for development of new TB drugs or regimens (15). On the contrary, substantial gaps in knowledge exist regarding the predictive accuracy of commonly used preclinical models and the most efficient and effective manner in which to utilize the outputs of these models to optimally inform key decisions, including the design and interpretation of clinical trials (16). This review will describe existing dynamic in vitro and in vivo models of TB chemotherapy and their utility in preclinical evaluations of promising new drugs and combination regimens, with an effort to highlight recent developments.

GOALS OF PRECLINICAL EFFICACY STUDIES

Simply stated, the primary goal of preclinical efficacy studies is to evaluate the potential of a new candidate drug or drug regimen to shorten or otherwise improve the treatment of drug-susceptible and/or drug-resistant TB. Initial studies focus on comparing the efficacy of new compounds to existing TB drugs and identifying drug exposures expected to produce optimal anti-TB effects while minimizing the risk of toxicity and selective amplification of drug-resistant mutants. More advanced studies seek to identify the best strategies for combining a new drug candidate with other new and existing drugs to produce superior new regimens. Once potentially superior regimens are identified, further studies may seek to refine the optimal dose and dose schedule for each drug and estimate the duration of treatment needed to produce outcomes equivalent or superior to existing comparator regimens.

The more specific objectives of preclinical efficacy studies are to characterize the drug’s (or regimen’s) activity against actively and nonactively multiplying M. tuberculosis, its potential to selectively amplify resistant bacterial subpopulations, and the pharmacokinetic and pharmacodynamic (PK/PD) relationships that govern these activities. Knowledge of the drug’s ability to reach M. tuberculosis and exert its effect in various sublesional locations and microenvironments characteristic of TB lesions (e.g., necrotic [caseous] foci and cavities) is expected to further inform dose and regimen selection (17).

While dose optimization is nothing new to TB drug development efforts, the importance of quantitative PK/PD-based dose selection is increasingly recognized (18). Indeed, there are compelling arguments for further dose optimization of three of the four first-line TB drugs (i.e., isoniazid, rifampin, and pyrazinamide) (11, 19–21) and the two most important second-line drug classes (i.e., fluoroquinolones and aminoglycosides) (22–24), all of which have been used to treat TB patients for at least 3 decades and, in most cases, longer. The most important example is the rifamycin class, where continued use of suboptimal doses and dramatic interpatient variability in PK and drug exposure, despite receiving the same dose, likely interact to drive the current length of TB therapy and the emergence of multidrug-resistant (MDR) TB (10, 25–27). As with other anti-infectives, both the magnitude of drug concentrations achievable at the site of the infection as well as the shape of the concentration-time curve are important determinants of activity for some anti-TB drugs (21, 28–33). PK/PD relationships established in simpler, more tractable in vitro models are likely to be relevant for more complex in vivo disease models (18), although it will be necessary to account for factors that modify the concentrations of biologically active drug that are achievable at the site of infection in vivo (e.g., plasma protein binding, intracellular penetration, diffusion through caseum or blood-brain barrier) and the influence of microenvironmental conditions or the host immune response (13, 17, 34, 35).

No single preclinical in vitro or animal model recapitulates all aspects of human TB within a cost-effective and ethically acceptable framework (16). Therefore, it is unlikely that a critical path for development of novel drugs and drug regimens can rely on a single model to provide data sufficient to inform all key development decisions (15). Instead, sound knowledge of the utility of a variety of models and methods will be necessary to employ all available tools in the most integrative and complementary fashion to guide drug and regimen selection and optimization. In the ideal critical path, the predictive accuracy of each model and its fitness for each designated purpose would be well established through a set of validating experiments and quantitative analyses (15).

PRECLINICAL EFFICACY MODELS

In Vitro Models

In general, there are two categories of in vitro models used to study the activity of anti-TB drugs: static models in which drug concentrations remain fixed over time and dynamic models in which drug concentrations change over time.

Static drug concentration models

Static models evaluate the bacterial response to drug concentrations that are fixed over time. Response measures include colony-forming unit (CFU) counts or surrogates such as optical density or other markers of bacterial growth or viability in liquid culture systems (36). The most common output from static models is the minimum inhibitory concentration (MIC). The greatest value of MIC may be its relative simplicity as a quantitative measure of potency for ranking compounds and for indexing microbial susceptibility to drug exposure in PK/PD analyses. However, by definition, it measures only growth inhibitory effects and cannot be used under conditions in which there is little or no growth. Bactericidal effects, as measured by the minimum bactericidal concentration and quantitative time-kill curves, are more informative but more costly and time-consuming to measure. Classically, MIC and minimum bactericidal concentration measurements are obtained against actively multiplying bacteria in optimal growth conditions. However, given the important role of persistent bacterial subpopulations in the treatment of TB, a wide variety of alternative models have been developed to test static drug concentrations against bacteria under stressful conditions that alter bacterial growth and metabolism in a manner that may reproduce mechanisms that drive bacterial persistence in vivo (8, 36). To date, there is limited evidence based on which to favor one particular persister model over another, and a detailed discussion of this topic is outside the scope of this review.

Because the shape of the concentration-time curve can have a profound effect on drug activity (37), PK/PD parameters associated with optimal efficacy in static concentration models may be significantly different from those derived from models that expose bacteria to changing drug concentrations over time.

Dynamic drug concentration models, including the hollow fiber system model of tuberculosis

Hollow fiber systems (HFSs) and related dynamic in vitro models are useful to study the bacterial response to fluctuating drug concentrations over time. They have been employed to derive information on PK/PD relationships for antibacterial agents and used in regulatory filings for decades, but HFS-TB models emerged only in the past 15 years (23, 38). HFS units (Fig. 1) consist of a bioreactor cartridge, the interior of which is traversed by multiple hollow fiber capillary tubes to create two compartments: an extracapillary compartment for culturing M. tuberculosis and an intracapillary compartment through which drug-containing media is supplied to the cartridge. The semipermeable nature of the hollow fibers allows small molecules (e.g., drugs, nutrients, bacterial metabolites) to diffuse between the two compartments but prevents the movement of bacteria into the intracapillary compartment. The intracapillary compartment of each cartridge is linked to a vascular system of tubing through which the flow of media is controlled by an adjustable perfusion pump. Drugs are introduced into the system using a programmable drug delivery pump and cleared by dilution in a central compartment at a rate controlled by additional pumps linked to input (fresh media) and output (waste media) reservoirs.

FIGURE 1.

Diagram of a hollow fiber system model of TB (HFS-TB) (49).

In many respects, the HFS-TB model is an ideal tool for developing a quantitative understanding of PK/PD relationships to inform drug and regimen development. As in static concentration models, bacteria in HFS units can be cultivated under a variety of environmental conditions by manipulating aspects of the media (e.g., pH, nutritional content) or the external environment (e.g., HFS enclosed in an anaerobic chamber) (21, 39). Even the intracellular niche of M. tuberculosis can be studied by cultivating infected macrophage-like cells (e.g., THP-1 or J774 cells) in the extracapillary space (40). Tight control of the culture conditions enables the study of PK/PD relationships against specific phenotypic subpopulations of bacteria that may be similar to in vivo subpopulations and relevant to clinical outcomes. Various culture conditions have been studied, including log-phase growth in nutrient-rich media, slower growth under acidic conditions (e.g., pH 5.8), nonreplicating persistence under low oxygen tension (e.g., ≤10 parts per billion), and intracellular infection (41). Another advantage of the HFS-TB over animal models is that its compartments can be serially sampled to measure viable bacterial counts, bacterial metabolites, and drug concentrations over time. This facilitates time-to-event and repeated event analyses that increase statistical power and enable construction of more dynamic and robust systems pharmacology models (41).

Perhaps the greatest advantage of the HFS-TB over static drug concentration models (and most animal models) is the precise control of the drug concentration-time profiles to which the bacteria are exposed. Concentration-time profiles may be designed to mimic in vivo (human or animal) PK profiles or to produce concentration-time profiles that are unattainable in animal models but have clinical or experimental value. For example, the activity of carbapenems and related β-lactams correlates best with the proportion of the dosing interval for which drug concentrations exceed the MIC (37). Because they are cleared much faster in mice than in humans, it is difficult to attain clinically relevant exposures in mice for many drugs in this class (42). In contrast, producing human-like exposures is quite straightforward in the HFS (29).

With appropriate attention to sterile technique to prevent contamination of the HFS, treatment durations of 28 to 56 days are feasible, and treatment durations as long as 6 months have been studied. Experimental designs have included dose-ranging and dose fractionation studies of single agents and comparisons of drug combinations, including drug sequencing studies (41). Data output, typically in the form of CFU counts, has been used to identify PK/PD parameters correlated with antibacterial effect and to derive target drug exposures associated with optimal microbial kill (e.g., as defined by the exposure associated with 80% or 90% of the maximal effect [EC₈₀ or EC₉₀, respectively]) and suppression of drug-resistant mutants (41, 43). Output from HFS-TB models also has been used to perform computer-aided clinical trial simulations based on Monte Carlo analyses (19, 23, 43, 44). In some studies, key PK/PD parameters and target exposures derived from HFS-TB experiments were used together with population PK data and wild-type MIC distributions to predict the probability of different clinical antibiotic doses achieving optimal drug exposures and rates of sterilizing effect in sputum in patient populations and to predict the proportion of patients that would develop acquired drug resistance despite receiving combination therapy as a result of population-level PK variability (26, 43). Finally, although more prospective study is required, HFS-TB data have been used to derive PK/PD-based breakpoints for drug susceptibility testing that could distinguish clinical scenarios in which treatment response is likely or unlikely (45, 46). In a formal statistical analysis, kill rates in sputum of patients, PK/PD parameters, and targets associated with optimal effect and even probability and time to emergence of drug resistance were found to be similar between patients and results of analyses based on HFS-TB outputs (43, 44).

The demonstrated predictive accuracy of the HFS-TB model should give confidence to drug developers and regulatory agencies that they are a valuable tool for PK/PD profiling to support regulatory activities (47, 48). Indeed, the European Medicines Agency’s Committee for Medicinal Products for Human Use recently rendered a positive qualification opinion that the HFS-TB model may be used in anti-TB drug development programs as an additional and complementary tool to existing methodologies (including animal models) to inform dose and regimen selection, including combinations of two or more drugs, to maximize bactericidal effects and minimize emergence of drug resistance (Table 2) (49). Specifically, the group endorsed the HFS-TB as a model fit to provide preliminary proof of concept for developing a specific drug or combination to treat TB, to select the PK/PD target for optimal effect, and to provide data to support PK/PD analyses leading to initial dose selection for preclinical and clinical studies. As a tractable dynamic in vitro PK/PD model, the HFS-TB model is expected to reduce the complexity and size of dose-finding studies in animal models and clinical trials and thereby shorten the duration of TB drug development programs. The HFS-TB model is also expected to assist in confirming dose regimens for later clinical trials taking into account any accumulated human PK data and available information on exposure-response relationships.

TABLE 2.

Qualification opinion of the European Medicines Agency’s Committee for Medicinal Products for Human Use regarding the HFS-TB (48)

The HFS-TB is qualified to be used in anti-TB drug development programs as an additional and complementary tool to existing methodology to inform selection of dose and treatment regimen, to maximize bactericidal effects and minimize emergence of resistance. More specifically, the HFS-TB may be useful as follows: To provide preliminary proof of concept for developing a specific drug or combination to treat tuberculosis To select the pharmacodynamic target (e.g., T/MIC, AUC/MIC) To provide data to support PK/PD analyses leading to initial dose selection for nonclinical and clinical studies, with the aim of limiting the number of regimens that are to be tested in vivo; it is anticipated that HFS-TB may be used to limit doses tested both in single-drug and combination regimen studies in vivo To assist in confirming dose regimens for later clinical trials, taking into account the accumulated human PK data in healthy volunteers and then patients as well as available information on exposure-response relationships

Despite its advantages, the HFS-TB model cannot replace preclinical animal efficacy studies or clinical trials (16, 47). Moreover, because most studies included in the predictive accuracy analysis of the model were performed after the clinical trials that the HFS-TB output was compared to, it is important to prospectively collect and analyze data on the performance of the model (16, 47). Finally, the use of the HFS-TB model has thus far been limited to a small number of laboratories. It will be important to study the reproducibility of the method and any operational issues related to model performance as it is taken up by other investigators.

Animal Infection Models of Active TB Suitable for Efficacy Testing

While drug exposure-response relationships derived in the HFS-TB or other dynamic in vitro models are expected to translate to in vivo models and to the clinic, in vivo drug efficacy occurs in the complex milieu of the infected host, where a variety of factors may be encountered that modify the relationship between plasma drug exposures and drug effect at the site of infection. These factors include various host defense mechanisms, protein binding, drug diffusion through caseum, unique environmental determinants of drug effect, and more diverse bacterial heterogeneity under conditions that cannot be fully reproduced in vitro. Moreover, tuberculous lesions are three-dimensional structures that introduce temporal and spatial gradients of these modifying factors that influence drug distribution and effect. Used appropriately, in vivo models enable study of the antibacterial effects of dynamic drug concentration-time profiles to derive PK/PD relationships or to confirm those derived in in vitro models. Many animal models also enable simultaneous study of multiple subpopulations, perhaps in clinically relevant proportions, to better understand the sterilizing potential of drug regimens and thus their potential to shorten TB treatment. Mice are by far the most commonly used preclinical efficacy model. However, because they do not form caseating lung lesions, the most commonly used mouse strains do not mimic the three-dimensional structure and heterogeneity of TB lesions. As a result, emerging mouse models and larger non-mouse species may have a key role to play in TB drug development.

The major premise behind the use of non-mouse species is that tuberculous lesions in species such as guinea pigs, rabbits, and non-human primates more closely represent the pathological hallmarks of human TB (e.g., caseation necrosis, and in rabbits and non-human primates, cavitation). If such lesions or the microenvironmental conditions found therein are important determinants of drug effect, by altering drug distribution, the mechanism or kinetics of drug action, or the susceptibility of the pathogen to the drug effect, then the presence or absence of such pathology could have an important influence on drug efficacy and the PK/PD relationships that govern it. Within the architecture of active caseous lung lesions, whether they are organized caseous granulomas, more disorganized areas of caseous pneumonia, or cavities, most bacilli are found extracellularly in the caseum, and only a minority reside intracellularly in neutrophils, epithelioid macrophages, and foamy macrophages in the cellular borders of the caseous lesions and in other cellular lesions (50–54). Detailed descriptions of caseating graulomas in some models have noted relatively small numbers of extracellular bacilli in caseum (55). However, such lesions are frequently sterilized by host immune mechanisms and resolve without treatment (56) and should not be confused with more poorly contained caseous lesions (e.g., caseous pneumonia and cavities) where host immune mechanisms are less effective and extracellular bacillary populations are disproportionately large (54, 56, 57). The latter progressive caseating lesions better reflect those that determine treatment outcomes in patients presenting with active TB.

Bacilli in different types of tuberculous lesions and in different compartments of caseous lesions experience different microenvironmental conditions that may affect a drug’s concentration-response relationship. Bacterial responses to stress conditions such as hypoxia, low pH, and oxidant stress have long been known to cause phenotypic tolerance and persistence in the face of drug exposure, affecting some drugs more than others. On the other hand, certain drugs require specific microenvironmental conditions for optimal effect. For example, pyrazinamide’s anti-TB effect is inversely proportional to pH (58). Metronidazole requires very low oxygen tension for bioactivation and activity against M. tuberculosis (59). These drugs may exhibit very different effects in different models and in different lesion compartments despite achieving similar concentrations at the site (35, 53, 60–63). Finally, although some drugs (e.g., small polar molecules such as pyrazinamide and isoniazid) distribute quite evenly through the various lesion types and sublesional compartments, other drugs partition differently into cells or into acellular caseum, creating the potential for markedly different drug exposures for bacilli in different lesion compartments (17, 34, 64, 65). Drugs exhibiting high lipophilicity as measured by logP and high protein binding tend to accumulate inside cells and to diffuse poorly through caseum. Examples include bedaquiline and clofazimine (34, 65). When compared to plasma concentrations, these drugs may achieve much higher concentrations where bacilli reside in macrophages but lower concentrations against extracellular bacilli in caseous lesions. Other interesting drug distribution profiles have recently been revealed. For example, unlike pyrazinamide and isoniazid, which diffuse rapidly in and out of caseous regions, rifampin diffuses relatively slowly into caseum but accumulates there with repeated dosing (65). The heterogeneous nature of human TB lesions and the various influences on drug concentration and concentration-response relationships illustrate the importance of understanding how a given drug candidate may be affected so that it can be taken into account in drug and regimen development.

Animal species used as preclinical efficacy models

Mice

Mice are the most commonly used species in TB drug development owing to their ease of handling and relatively low procurement and housing costs (36, 66). Particularly in the early phases of drug development, their ubiquity, tractability, cost, and small compound requirements make mice the most practical model. Nevertheless, the absence of caseating lung lesions and cavities is a potential limitation of commonly used mouse strains.

Instead of developing caseating lung lesions, commonly used mouse strains (e.g., Swiss, BALB/c, C57BL/6) contain the infection in nonnecrotic cellular granulomas composed predominantly of lymphocytes, epithelioid macrophages, and foamy macrophages. Here, virtually all tubercle bacilli reside in macrophages and may not be exposed to all of the microenvironmental conditions found in caseous lesions that are closed to the airways (e.g., hypoxia) (60, 63, 67). Thus, at least for some drugs that partition very differently in various lesion compartments, commonly used mouse strains may best represent the intracellular compartment and not other lesion compartments found in caseous disease models. The utility of these mouse strains for evaluating new drugs and drug regimens may therefore depend on several factors. Of perhaps the greatest importance is whether a drug requires conditions found only in caseous lesions to be active. This is an unusual scenario but is exemplified by metronidazole, which requires very low oxygen tension for activation and has demonstrable anti-TB activity in closed caseous foci of rabbits and macaques but not in mice (53, 60, 62, 63, 68). A second factor is the extent to which a drug partitions differently into cells and caseum. Drugs that diffuse evenly into these compartments are likely to be well represented in mice. On the other hand, the activity of drugs that show markedly greater partitioning into cells compared to caseum, such as clofazimine and bedaquiline (34, 65), could be overestimated in mice that do not develop caseous lesions (69, 70). However, it is important to emphasize that the treatment-shortening potential of such drugs demonstrated in mice may yet be relevant to human TB, provided that the intracellular bacterial populations modeled in mice are similar to intracellular bacilli found in human TB lesions and that these populations, even if they are minority subpopulations, persist there and play a role in determining the duration of treatment needed to cure human TB.

Despite their limitations, mice have been instrumental in drug and regimen development for TB. The greater use of mouse models relative to any other species means that there is a larger evidence base on which to judge their utility. To date, no other animal model has demonstrated the same or superior predictive value for drug regimens in clinical use (16). Despite developing rather homogeneous intracellular infections, commonly used mouse strains such as Swiss and BALB/c mice have a good track record in predicting the clinical utility of new TB drugs and regimens. For example, rifampin and pyrazinamide are the only drugs with a clinically validated ability to shorten the duration of TB treatment to 12 months or less. These treatment-shortening effects were first demonstrated in mice (71). More recently, a large phase 3 trial (72) sought to shorten the treatment duration from 6 months to 4 months by replacing isoniazid or ethambutol with the late-generation fluoroquinolone moxifloxacin, based in part on results from murine models (73, 74). Despite shortening the time to sputum culture conversion, the 4-month moxifloxacin-containing regimens were not as effective as the 6-month control regimen (72). Although this result prompted criticism of mouse models as well as surrogate markers used in phase 2 clinical trials leading up to the phase 3 trial, an objective appraisal of the available mouse model and clinical data suggests that a treatment-shortening effect of less than 2 months is entirely consistent with both the murine and clinical data (14). More novel regimens with new and repurposed drugs (e.g., bedaquiline, pretomanid, linezolid) have also demonstrated treatment-shortening potential in mice (75–78). Although early clinical data are promising (79–81), confirmation of their treatment-shortening effects in patients and comparisons to results in mouse models will require phase 3 trials using relapse as an endpoint.

Experimental infection of mice

Experimental infection of mice is generally via the respiratory tract (e.g., aerosol or intratracheal inoculation) or intravenous injection. In either case, virulent M. tuberculosis multiplies exponentially in the lungs and spleen of naive mice for the first 2 to 3 weeks postinfection. Thereafter, in immunocompetent mice, the adaptive immune response reduces the net bacterial multiplication rate (82). In the absence of treatment, the outcome of the infection depends on the virulence of the M. tuberculosis strain, the size of the infectious dose, and the susceptibility of the mouse strain. High-dose infection (e.g., implantation of approximately 5 × 10³ or more CFU in the lungs by any route) with a virulent M. tuberculosis strain typically results in overwhelming infection and death within 4 to 6 weeks postinfection in any mouse strain. Lower infectious doses produce the same fate in immunodeficient mice (e.g., athymic nude mice or gamma-interferon knockout mice) (36). In contrast, similar low-dose infections (e.g., 10 to 500 CFU implanted in the lungs) in immunocompetent strains such as BALB/c and C57BL/6 mice result in containment of infection by the adaptive immune response, a plateau in lung CFU counts beginning at approximately 4 weeks postinfection, and long-term survival of the animal with chronic infection (36, 66). Among inbred mouse strains, there are clear differences in susceptibility to infection (83) that may translate into differences in disease susceptibility and treatment outcomes. However, there are unlikely to be distinct differences in drug efficacy between commonly used immunocompetent mouse strains, especially with respect to ranking drug regimens (36).

Common experimental designs

The natural history of infection in mice enables three basic experimental designs for initial drug efficacy studies (Fig. 2). The first is an acute infection model in which treatment is initiated within the first week (often the first 1 to 3 days) postinfection, and drug effects against logarithmically multiplying bacteria are measured over 1 to 4 weeks of treatment. Use of “real-time” efficacy measures such as mouse survival, body weight, prevention of macroscopic lung lesions and splenomegaly, and in vivo or ex vivo biomarkers of bacterial viability other than culture-based outcomes (e.g., CFU counts) may permit relatively high throughput but may not reliably discriminate bactericidal and bacteriostatic effects (36). Whether non-culture-based or culture-based readouts are used, it is clear that assessments using acute infection models favor compounds with rapid onset of antibacterial effects and bactericidal mechanisms of action and may misrepresent the sterilizing activity of compounds or doses having a slower onset of action or activity predominantly against slowly replicating or nonreplicating bacilli (84). One has to look no farther than the performance of isoniazid versus rifampin and pyrazinamide for an example. In acute infection models, isoniazid, which has celebrated early bactericidal activity and limited sterilizing activity in TB patients (6, 7), markedly outperforms rifampin and pyrazinamide, the two first-line drugs with bona fide treatment-shortening effects (84, 85). Therefore, such acute infection models are most useful for rapid screening for in vivo anti-TB activity and perhaps some measure of a drug’s ability to kill mutants resistant to companion agents in the absence of a suitable host response.

FIGURE 2.

Schematic representation of various experimental models.

A chronic infection model, in which treatment is initiated at least 3 weeks, and often 4 to 6 weeks, after low-dose infection, is more effective for demonstrating the superior sterilizing efficacy of compounds such as rifampin and pyrazinamide that are most active against the slowly replicating or nonreplicating bacilli that predominate during the plateau in lung CFU counts brought about by the adaptive immune response (84). Thus, although it takes longer to set up compared to an acute infection model, a chronic infection model is more likely to differentiate between compounds or doses of the same compound on the basis of their potential sterilizing activity. However, since there is no net multiplication in untreated animals (82, 86), the chronic model is unable to distinguish bacteriostatic effects from no effect.

Both the acute and chronic models defined above have the disadvantage of significantly lower bacterial burdens at the initiation of treatment relative to the bacterial burden in patients with cavitary lesions, where estimates suggest 10⁷ to 10⁹ cultivable bacteria reside in the cavity wall alone (87). Larger bacterial populations are more diverse, have higher numbers of spontaneous drug-resistant mutants, and require longer treatment durations to attain cure with combination chemotherapy. Thus, a subacute infection model, in which treatment begins 2 weeks after high-dose infection (e.g., 10⁴ CFU inoculated via aerosol or 5 × 10⁶ CFU inoculated intravenously), may be preferable for evaluating multidrug regimens for their effects on actively and nonactively replicating bacilli and drug-resistant subpopulations, and for estimation of the treatment duration necessary for cure (88). Whether the route of infection is via the aerosol or intravenous route, the bacterial burden at the initiation of treatment is close to 10⁸ CFU in the lungs in this subacute model, and the modern short-course regimen of 2 months of rifampin, isoniazid, pyrazinamide and ethambutol followed by rifampin and isoniazid requires approximately 6 months of treatment to prevent relapse in the great majority of mice, as it does in TB patients (66, 88).

C3HeB/FeJ mice

C3HeB/FeJ mice have garnered significant interest recently because, unlike commonly used mouse strains, they develop caseating lesions after M. tuberculosis infection under the proper experimental conditions (52, 53, 89). Development of the caseating lesions is primarily determined by a natural deficiency in the Ipr1 (for intracellular pathogen resistance 1) gene in the sst1 (for supersusceptibility to TB 1) locus, which promotes necrosis, rather than apoptosis, of infected macrophages (89–91). The closest human homolog of Ipr1 is SP110b, an interferon-inducible protein that normally downregulates proinflammatory cytokines, including tumor necrosis factor-alpha (TNF-α), and limits excessive tissue damage and development of necrotic lung lesions (92). The increased susceptibility of C3HeB/FeJ mice appears limited to virulent intracellular pathogens (93). However, they do not appear to have difficulty containing infections with nontuberculous mycobacteria, Mycobacterium bovis bacille Calmette-Guérin (BCG), and even less virulent isolates of M. tuberculosis (52, 89, 94). Although the role of SP110 in human susceptibility to TB is debated, evidence is mounting that variations in this locus do contribute (92).

Considerable heterogeneity in lung pathology may be observed in C3HeB/FeJ mice, even among those infected in the same experiment (52, 53). After infecting mice via aerosol with 50 to 75 CFU of virulent M. tuberculosis, Irwin et al. (52) performed detailed histopathological analyses at various time points over the course of infection. They described three distinct types of lung lesions: fulminant necrotizing alveolitis, well-organized caseating granulomas, and cellular granulomas typical of those predominating in other commonly used mouse strains. Although they may develop features of caseous necrosis, mice exhibiting the fulminant granulocytic alveolitis typically succumb to infection within 5 to 6 weeks of infection and are therefore of limited utility for drug efficacy studies. In contrast, the well-organized caseating granulomas developing in one-third to two-thirds of mice display central caseation, fibrous encapsulation, and hypoxia characteristic of larger species, including humans (52, 67). Some large caseating granulomas or foci of caseous pneumonia contain more than 10⁹ bacilli and eventually cavitate, although this tends to be a late event (53, 60, 95). Cavities in association with a more stable clinical condition can be promoted specifically by allowing caseous granulomas to form, then treating with a noncurative course of combination chemotherapy (e.g., 8 weeks for the first-line regimen) before allowing the disease to recur (95). Mice with cavities in this scenario can be identified by serial computed tomography (CT) and may live for months once cavities are found (95), thereby enabling drug distribution studies and efficacy testing against this lesion type.

Given the potential impact of caseous lesions on the PK/PD of TB drugs, there is substantial optimism that the histopathological changes observed in C3HeB/FeJ mice will enable better representation of human drug efficacy in mice. The potential value of C3HeB/FeJ mice may be indicated by the differential response to some anti-TB drugs in this strain compared to commonly used mouse strains, as indicated by recent experiences with pyrazinamide and clofazimine. As in humans and larger species (64, 65), pyrazinamide distributes evenly through various caseous lesion compartments, including acellular caseum, in C3HeB/FeJ mice (34, 35). However, unlike in BALB/c mice and in C3HeB/FeJ mice harboring only cellular granulomas, where it exerts a significant bactericidal effect, pyrazinamide has only limited effect against the numerous extracellular bacilli in the caseum of large caseous granulomas in C3HeB/FeJ mice (35). This discrepant activity was ultimately shown to be due to the elevated pH of the caseum, which prevented pyrazinamide from exerting a bactericidal effect (34, 35). Nevertheless, pyrazinamide does exert a sterilizing effect even in C3HeB/FeJ mice with large caseous lesions (96), an effect that is presumably restricted to bacteria engulfed by activated macrophages, where the pH falls to a level sufficient for pyrazinamide effect. This result suggests that the bacterial persisters targeted by the treatment-shortening effects of pyrazinamide in the first-line regimen reside intracellularly.

Unlike pyrazinamide, clofazimine partitions very differently into cellular compartments compared to acellular caseum, a trait evident in C3HeB/FeJ mice as well as rabbits and humans (65). This provides at least one explanation for the poor initial activity of clofazimine observed in C3HeB/FeJ mice relative to BALB/c mice, but the hypoxic conditions observed in caseous granulomas of the former are another potential explanation (69). Whether clofazimine, like pyrazinamide, exerts bactericidal activity against intracellular bacilli and contributes sterilizing activity to regimens in C3HeB/FeJ mice remains to be seen. Likewise, several novel regimens now in clinical development based on results in BALB/c mice have been evaluated in C3HeB/FeJ mice (67, 76). The clinical outcomes observed relative to the standard of care should allow further evaluation of the potential utility of the latter strain.

Rats

Although commonly utilized in the pharmaceutical industry for PK and toxicology studies, rats were long considered too resistant to M. tuberculosis infection to be a suitable model for TB drug development. However, aerosol and intratracheal infections of outbred Wistar rats produce pathological features similar to those observed in commonly used mouse strains other than C3HeB/FeJ (97, 98). Advantages of working with this species include the ability to study efficacious drug exposures in a commonly used toxicity model, greater feasibility to control drug exposures through long-term vascular catheters or implantable infusion pumps, and sometimes lower compound clearance compared to mice (99). For example, a low-dose aerosol infection model was employed at AstraZeneca to test a series of DprE1 inhibitors alone in part because the compounds were cleared too rapidly in mice to attain optimal drug exposures (100). The cotton rat (Sigmodon hispidus) can develop caseating lung lesions more characteristic of human TB. However, the presence of conflicting reports (101, 102) suggests that such lesions are dependent on experimental conditions that will require further optimization.

Guinea pigs

Once the favored in vivo model for TB drug development due to their marked susceptibility to M. tuberculosis infection, guinea pigs are now used infrequently for this purpose. However, they remain an alternative or complementary model to mice given the occurrence of caseating primary lung lesions. The histopathology of M. tuberculosis infection is well characterized. Moreover, elegant staining studies during chemotherapy have mapped the location of the most numerous persistently acid-fast bacilli to the necrotic core of such lesions, where the organisms reside extracellularly (51, 103), suggesting that this is an important persister population. Smaller numbers of extracellular bacilli are found in the acellular rim of these necrotic lesions. Still smaller numbers of intracellular bacilli are found in the cellular rim of necrotic primary lesions and in the secondary disseminated lesions that do not typically undergo necrosis (51).

Guinea pigs are typically used in chronic infection schemes to allow time for caseous lesions to develop. Infection is most often via the aerosol route, although parenteral routes are possible. Recent experiments have recapitulated the combined treatment-shortening effects of rifampin and pyrazinamide compared to streptomycin and isoniazid (104–106). Remarkably, however, cure without relapse of disease is obtained with shorter durations of treatment in guinea pigs compared to BALB/c mice, despite similar bacterial population sizes and drug exposures in the two models, suggesting that infections in BALB/c mice harbor more persistent bacteria (104, 107). Another interesting observation of guinea pigs is the apparent difficulty in selecting isoniazid-resistant mutants harboring mutations in the mycobacterial catalase-peroxidase katG (105). This suggests greater fitness costs associated with these mutations in guinea pigs, compared to mice and possibly cavitary disease in humans, perhaps a result of enhanced oxidative stress experienced by M. tuberculosis in guinea pigs.

Several novel drugs and drug regimens now in clinical trials have been compared to standard therapies in guinea pigs, and results seem generally consistent with those in BALB/c mice (61, 103, 105, 106, 108–111), although the guinea pig may have better demonstrated the rifapentine doses needed to achieve efficacy superior to that of rifampin in phase 2 trials (10, 109, 112). Although this may be attributable to the caseous lesions occurring in guinea pigs and not in mice, further study is required. Compared to rifampin, rifapentine accumulates to a greater extent inside cells and diffuses and accumulates less readily in caseum compared to rifampin (65, 113). Thus, for a given plasma drug exposure, rifapentine may exhibit superior activity relative to rifampin against intracellular bacilli in the lesions of BALB/c mice and the cellular regions of caseous lesions but inferior activity in the caseous regions (13, 113, 114). Pharmacometric analyses of ongoing phase 3 trials of high-dose rifamycin-containing regimens and more advanced modeling of drug distribution and compartmentalization of drug effect may shed further light on the predictive accuracy and most complementary applications of guinea pig and mouse models for preclinical drug development.

Despite their potential advantages relative to non-C3HeB/FeJ mouse strains, guinea pigs only infrequently develop cavitary lesions and have other disadvantages that limit their use. They are both more expensive to maintain and more fragile than mice, being especially susceptible to the adverse effects of broad-spectrum antimicrobials on the normal intestinal flora. Surprisingly, guinea pigs also clear a number of drugs, including first-line TB drugs as well as moxifloxacin and pretomanid, faster than mice do (107, 109). Combined with their larger size, higher clearance in guinea pigs leads to large compound requirements that discourage their use in early drug development. Nevertheless, guinea pigs may be useful as a relatively economical model for confirming or revising results obtained in mice, especially when microenvironmental conditions such as hypoxia (51, 63) or other aspects of larger caseous lesions are expected to influence drug distribution or drug action.

Rabbits

Rabbits are highly susceptible to virulent M. bovis strains (115), whereas infection with common laboratory strains of M. tuberculosis, such as H37Rv and Erdman, does not result in progressive infection (116). In contrast, aerosol infection of outbred New Zealand white rabbits with the HN878 strain of the M. tuberculosis Beijing subfamily was recently shown to cause progressive infection with caseating lung lesions by 8 weeks and cavitary lung disease by 16 weeks postinfection (117). Cavitary disease also can be produced with a wider range of M. tuberculosis strains using sensitization with heat-killed bacilli followed by bronchoscopic instillation of viable cultures (118). The creation of lung cavities is the major potential advantage of rabbit models for preclinical drug efficacy testing because cavitary disease is associated with slower response to TB treatment, relapse, and emergence of drug resistance and because cavities are not a feature of non-C3HeB/FeJ mouse or guinea pig models. Although cavity formation may not occur uniformly in all animals at the same time, CT imaging is useful for predicting when and where cavities will occur (119, 120). CT also allows for serial quantification of lesion/cavity volume and cavity surface area as biomarkers of treatment response (120).

The larger size of rabbits and their caseous lung lesions makes them well suited for studying the impact of specific lesion types on drug distribution and efficacy. Recent lesion-specific drug distribution and PK studies demonstrating similarities between results in rabbits and humans are promising in this regard (64, 65, 113, 121–123). To date, however, studies of drug efficacy in rabbits are very limited. One tantalizing finding is that large, closed caseous lesions in rabbits are likely to harbor more extensive zones where M. tuberculosis encounters severe hypoxia, as demonstrated by the anti-TB activity of metronidazole in rabbits but not in mice (53, 60, 63, 68). Unfortunately, evidence confirming that metronidazole is useful in human TB remains elusive (vide infra) (124). Combination therapy has been studied more rarely in rabbits, although a recent study characterized the response to therapy with first-line TB drugs, including evaluation of positron emission tomography (PET)-CT as a potential surrogate marker for antimicrobial effects (125).

Disadvantages of rabbits include the high acquisition and housing costs, the limited number of reagents for characterizing immunological responses, and greater pathogen containment concerns. Too expensive to be a “workhorse” model for preclinical efficacy studies, rabbits will likely remain most useful in drug development for informing PK/PD models incorporating lesion-specific drug distribution, exploring hypotheses related to the effects of microenvironmental conditions on drug efficacy, and perhaps bridging efficacy studies using surrogate markers that can be used in clinical trials such as PET-CT.

Non-human primates

Non-human primates, primarily macaques, were used as preclinical TB drug efficacy models as early as the 1950s and have recently garnered renewed interest. Unlike other animal models considered here, they may represent the full spectrum of pathology and outcomes of M. tuberculosis infection in humans.

Infection of cynomolgus macaques via intrabronchial instillation of approximately 25 CFU of M. tuberculosis produces active TB or subclinical/latent infection in roughly equal proportions over the ensuing 6 to 8 months (126, 127). Serial PET-CT imaging beginning as early as 3 weeks postinfection appears useful for identifying animals destined to develop active versus latent disease (128). A higher proportion of animals with active disease can be obtained more rapidly (e.g., within 10 weeks) with instillation of approximately 10³ CFU (129). Active disease is evidenced by symptoms (e.g., cough, weight loss), elevated erythrocyte sedimentation rate, abnormal chest radiograph, and positive cultures from gastric aspirate and bronchoalveolar lavage specimens (126, 127). Potential endpoints for drug efficacy studies include serial measures of viable bacterial counts from gastric aspirates or bronchoalveolar lavages, serial PET-CT imaging endpoints, and terminal measures such as organ CFU scores and gross pathology scores (62, 128). To date, however, these endpoints have not been proven to be useful in discriminating between active regimens with differing potencies. The responses to monotherapy with the oxazolidinones linezolid and AZD5847 were superior to no treatment but indistinguishable from each other after 1 to 2 months of treatment in macaques using various scores based on bacterial burden and PET-CT imaging (130). However, linezolid has demonstrably greater bactericidal activity in human sputum over the first 2 weeks of treatment (131) than AZD5847 (132), despite achieving higher AZD5847 plasma exposures in patients than in macaques (130, 132). Metronidazole was also evaluated in macaques with active disease, where it did not appear to increase the activity of a rifampin-isoniazid combination (62). In one small clinical trial with MDR-TB, addition of metronidazole to a background regimen of second-line drugs was associated with higher rates of sputum smear conversion and culture conversion in liquid, but not solid, media at 1 month, but these differences did not persist after the second month (124). Clearly, further study is needed to determine the predictive accuracy of endpoints measured in macaques for clinical outcomes.

Many challenges are associated with the use of macaques for preclinical drug efficacy models, including limited availability, great expense, requirements for special husbandry and enrichment, and ethical concerns. With respect to drug development, the challenge to the field is to demonstrate whether the potential advantages of working with these primates are unique to this model or whether similar information can be gained from more economical and more ethically acceptable models.

Common marmosets (Callithrix jacchus) are New World primates that, at 250 to 500 grams as adults, are much smaller than macaques (and even rabbits) yet develop similar caseous lesions, including cavities, in response to M. tuberculosis infection (133). In addition to being much more economical, marmosets have the advantage of frequent twinning, which provides opportunities for ideal experimental controls (133). Despite these promising features, there are currently few published data affirming the utility of the marmoset in drug efficacy studies. The first description of experimental M. tuberculosis infection in marmosets was published only recently (133). Since then, a study compared the efficacy of the modern, four-drug short-course regimen to that of a more primitive streptomycin-isoniazid combination over 6 weeks of treatment (134). Intriguingly, the former regimen showed greater effect in cavitary lesions, significantly reducing the proportion of animals with cultivable bacilli and the total number of bacilli in cavities. No significant difference was noted in noncavitary lesions or other body sites, except that the streptomycin-isoniazid combination was more active in the spleen. These results suggest that the marmosets are a more economically viable alternative to macaques for identifying the impact of drug distribution and microenvironmental conditions on drug effect at the site of action in caseous lesions. As with C3HeB/FeJ mice, additional studies are warranted to demonstrate the relative value of marmosets as alternative or complementary models to more commonly used mouse strains.

MODELING THE CHEMOTHERAPY OF LATENT TB INFECTION

Treatment of latent TB infection (LTBI) is critical to improving TB control and essential to TB elimination with available tools (135). Thus, it is a key component of the WHO End TB Strategy. LTBI is defined by evidence of M. tuberculosis infection in the absence of clinical symptoms. Only approximately 5 to 10% of immunocompetent individuals with LTBI will ultimately develop active TB, with half of the risk occurring within the first 2 years after infection. The principal objective of drug development for treatment of LTBI is to develop shorter or otherwise simpler regimens to promote adherence and treatment completion. This is especially true for treatment of LTBI among people exposed to MDR and extensively drug-resistant (XDR) TB cases, for which existing rifamycin- and isoniazid-based LTBI regimens are expected to be less effective and second-line drug regimens are longer and less well tolerated.

Although the pathogenesis of LTBI and reactivation remains incompletely understood, there is growing appreciation that LTBI represents a spectrum of conditions resulting from different outcomes of the host-pathogen interaction (136). These conditions range from cleared infection (and thus no risk of reactivation) with residual immunological memory, to controlled or dormant infection with viable bacilli (with intermediate risk of progression), to percolating or incipient disease (with the highest risk of progression to active disease). With respect to evaluating experimental models of LTBI chemotherapy, an important question is whether it is necessary to model the entire spectrum of viable infectious states including dormant, possibly noncultivable, bacilli or whether most people with LTBI who will subsequently reactivate (especially in the timeframe studied in clinical trials) harbor chronic, low-level infection with cultivable bacilli that can be modeled successfully in a variety of species.

Mice are not generally considered to develop LTBI because they are incapable, on their own, of preventing progression to disease after experimental infection. However, mice have still proven useful as models of LTBI treatment and contributed directly to the development of treatment-shortening LTBI regimens used in the clinic. Grosset et al. developed a mouse model of LTBI chemotherapy in which mice are immunized with M. bovis BCG 4 to 6 weeks before infection with virulent M. tuberculosis (137). The enhanced immune response promoted by BCG immunization limits multiplication of M. tuberculosis, leading to a lower-burden infection with stable CFU counts over time. Unlike the more commonly used Cornell model, in which the paucibacillary state is created by treatment of mice with high doses of first-line TB drugs, the paucibacillary state of the BCG immunization model is created by the host immune response and is therefore more likely to represent the phenotypic state of bacteria in people with LTBI (138).

The BCG immunization model garnered attention after being used to demonstrate that treatment with 2 months of rifampin plus pyrazinamide (2RZ) had sterilizing activity superior to 6 months of isoniazid (139). This finding gave rise to clinical trials (140, 141) that subsequently established the efficacy of this short-course LTBI regimen that enjoyed clinical usage until being abandoned due to toxicity concerns (142). Once-weekly administration of isoniazid and rifapentine for 3 months (3HP_1/7) was later studied and shown to be at least as effective as daily administration of isoniazid for 6 months (143), a finding that was also confirmed in a large clinical trial (144).

Use of a more immunogenic recombinant BCG strain overexpressing antigen 85B and a very low aerosol challenge dose with M. tuberculosis produces a stable paucibacillary state (≤10⁴ CFU) contained within compact cellular granulomas that may be even more representative of LTBI (145). This refined paucibacillary model has been validated by comparing the efficacy of five clinically recommended LTBI regimens (145, 146). The ranking of the regimens in order of increasing activity and the decreasing duration needed to cure the infection in this model are consistent with the decreasing treatment durations recommended when these regimens are used in humans (i.e., 6 to 9 months of daily isoniazid, 4 months of daily rifampin, 3 months of daily rifampin-isoniazid, 3 months of once-weekly isoniazid-rifapentine, and 2 months of daily rifampin-pyrazinamide (Table 3). Moreover, the duration of treatment necessary to prevent relapse in approximately 50% or more of mice appears to give a good general estimate of the effective treatment duration in clinical use. No other LTBI chemotherapy model, including the Cornell model, is validated in this way. Recent studies identified other effective regimens containing new drugs in development that may be capable of treating LTBI caused by both drug-susceptible and MDR/XDR M. tuberculosis (146–148).

TABLE 3.

Ranking of regimens to treat latent TB infection in the paucibacillary mouse model and correspondence with clinical guidelines^a

Regimen	Clinically recommended duration for treatment of latent TB infection (166–168)	% of mice with positive M. tuberculosis cultures 3 months after completing the indicated treatment duration (146, 148)
		2 months	3 months	4 months	6 months
INH	9 months	100		100	100
RIF	3–4 months	100	87	30–46
RIF+INH	3 months	93	54
RPT+INH (1/7)b	3 months	87	47
RIF+PZAc	2 months	60		0

^a For comparison purposes, treatment durations producing treatment success in ∼50% of mice are in bold font. Abbreviations: INH, isoniazid; PZA, pyrazinamide; RIF, rifampin; RPT, rifapentine.

^b (1/7) indicates once weekly treatment; all other regimens are daily (5 to 7 days per week).

^c The RIF+PZA regimen is no longer clinically recommended due to excessive hepatotoxicity (142).

As species relatively resistant to M. tuberculosis infection, rabbits and non-human primates may also prove to be useful models of LTBI treatment. However, there is limited relevant experience with chemotherapy to date. Infection of rabbits with many commonly used M. tuberculosis strains results in apparent clearance of the infection (116). Infection of New Zealand white rabbits with the CDC1551 strain under conditions similar to those that cause progressive caseating disease and cavitation after HN878 strain infection also resulted in an apparently abortive infection in which the bacterial burden peaked around 4 weeks and subsequently declined, leaving few or no cultivable bacilli and only rare organized granulomas by 16 to 20 weeks postinfection (149). However, viable bacteria persisted and reactivated to cause disease upon treatment with corticosteroids (149). In contrast to the murine model, which arguably best represents the percolating or incipient disease portion of the LTBI spectrum, this rabbit model may better represent the more dormant range of the spectrum. To the author’s knowledge, this rabbit model has not yet been used to study the chemotherapy of LTBI. Given the different states of infection in the murine and rabbit models, a fascinating experiment would be to compare the efficacy of isoniazid versus one or more rifamycin-containing regimens to determine how results compare to the outcomes of clinical trials. In addition to validating the model, such a comparison might shed light on the persistent question of how isoniazid, which is not believed to be active against dormant bacilli, is as effective as it is in preventing reactivation of LTBI. Beyond the absence of data indicating its predictive value, the key challenge facing the rabbit LTBI model is the exponentially higher cost required to run the model.

As the most pathologically similar animal models of M. tuberculosis infection, non-human primates also could be considered the most relevant LTBI models. Roughly half of cynomolgus macaques infected with approximately 25 CFU of M. tuberculosis Erdman develop lung lesions representing the full LTBI spectrum within 6 to 8 months of infection. The presentations range from sterilized infection and dormant infection with no cultivable bacilli (and predominantly fibrocalcific or sclerotic granulomas), to quasi-stable, low-burden infections with 10¹ to 10⁴ detectable CFU per lesion, to percolating disease with a higher bacterial load (126, 127). Spontaneous reactivation of the more dormant end of the spectrum was rare. In contrast, some percolators manifested pathologic features of active disease. Lower-dose infection may be useful to enrich for more latently infected animals, with or without as many percolators. Compared to no treatment, treatment of latently infected macaques with isoniazid for 6 months (n = 5 animals) or rifampin-isoniazid for 2 months (n = 7) prevented progression of disease after challenge with an inhibitor of TNF-α, as measured by gross pathology, bacterial burden, and dissemination scores (62).

In an interesting departure from negative results in the murine Cornell model (68), metronidazole appeared to reduce the bacterial burden after TNF-α inhibitor challenge in the same experiment (62). This result indicates the potential value of the macaque model in presenting conditions not found in mice. Unfortunately, due to toxicity concerns, it is unlikely that the efficacy of metronidazole will ever be evaluated against human LTBI to further validate the model. More recently, once-weekly isoniazid-rifapentine for 3 months appeared efficacious in a similar model using rhesus macaques, although only five animals were studied (169). As discussed for rabbits, further study of isoniazid versus various rifamycin-containing regimens in dormant versus percolating infections would again help to understand the reactivation potential of these two states and the relative merits of focusing experimental LTBI chemotherapy work on one state or the other. Study of responses to LTBI treatment also may be especially valuable for biomarker discovery to facilitate early phase 2 clinical trials evaluating new LTBI treatments since phase 3 trials require thousands of patients and many years to complete. It would also be of interest to determine whether a paucibacillary infection in marmosets could be produced by BCG immunization followed by low-dose challenge.