In this study, we compared the responses of the novel C3HeB/FeJ Kramnik mouse model to the more conventional BALB/c mouse model following aerosol infection with M. tuberculosis and monotherapy using a panel of five clinical drugs. This study was driven by the need for a mouse model of TB infection that more closely mimics the pathology of naturally occurring TB in humans, to facilitate more realistic in vivo drug testing against nonreplicating or slowly replicating persistent TB.
The most unique feature of the lung lesions that developed in Kramnik mice was the formation of multiple discrete encapsulated lesions with central liquefactive necrosis as well as areas of necrotizing alveolitis. In contrast, the more widely used BALB/c strain of mice developed less extensive foci of mixed inflammatory cells with minimal cellular necrosis. Unlike the BALB/c mouse strain, the Kramnik mouse model showed evidence of lesion hypoxia, fibrosis, liquefactive necrosis, and occasional cavity formation. The use of immunohistochemistry to detect adducts of pimonidazole as an indicator of hypoxia showed that residual viable cells within necrotic lesions were under hypoxic stress at the time points tested. For most late time points, pimonidazole staining was most often concentrated in clusters of foamy macrophages rather than areas of liquefactive necrosis. This can be explained either by a lack of viable cells capable of metabolizing pimonidazole or the regions of liquefactive necrosis being anoxic rather than hypoxic.
A critical component of lung lesions of TB-infected Kramnik mice was the deposition of collagen that delineated foci of liquefactive necrosis from more normal lung parenchyma. The picrosirius red staining showed a mixture of mature and immature collagen fibrils, suggesting that there was continuous remodeling of the capsule and that the granuloma is therefore a highly dynamic structure, which is in contrast to common belief.
Another important feature of the encapsulated lesions in the Kramnik model was the accumulation of macrophages with abundant foamy cytoplasm that formed the innermost layer of viable cells (located directly on the inside the fibrous capsule). Although the role foamy macrophages play is unknown, studies suggest that they contribute to the deposition of collagen via the production of granulocyte-macrophage colony-stimulating factor (GM-CSF) (35
). In both Kramnik and BALB/c mice, the accumulation of foamy macrophages was often accompanied by cholesterol deposition derived from the cell membrane of necrotic host cells. It has been shown primarily using in vitro
cultures that bacilli in microenvironments of hypoxia and nutrient deprivation adapt by altering their metabolism to utilize lipids (28
). Studies have shown upregulation of genes for lipid sequestration and metabolism in human patients infected with TB, representing pathogen-mediated dysregulation of host lipid metabolism (28
Although commonly used as an alternative to the mouse model, evaluation of drug therapy in the guinea pig model that develops necrotic granulomas has presented some challenges. Guinea pigs clear the first-line TB drugs (22
), as well as moxifloxacin and PA-824 (4
), very quickly (even more rapidly than mice), resulting in the need for much larger drug doses in order to achieve similar drug exposures over time. This leads to substantial drug intolerability, as guinea pigs are especially susceptible to perturbations in the normal intestinal flora caused by broad-spectrum antibiotics. Although many methods have been attempted to add vitamin supplements and probiotics to their diet (43
), adverse effects may still occur and could potentially skew study results. Interestingly though, several studies to date have shown that guinea pigs, compared directly with mice, have greater and faster bacterial killing after drug treatment (3
). This is surprising as one would expect the bacteria within the caseous necrotic lesions of infected guinea pigs to be more drug refractory (due to the slowed metabolism of the bacteria and/or drug penetration issues) versus the bacilli in the inflammatory lesions of the classical mouse model.
Healing of lesions by dystrophic mineralization, which is a prominent pathological feature in human TB and in the guinea pig TB model (7
), is not a response observed readily in the Kramnik mouse model. There are species differences in the rate of mineral deposition, and the shortened survival time may explain why mineralization is not seen in the Kramnik and other mouse models of TB.
To visualize the TB bacilli in the lung lesions of the Kramnik mice during disease progression, different acid-fast stains were used. In an earlier report, we described a novel triple-staining method which identified TB bacilli as well as the surrounding tissues in the mice and guinea pigs (23
). While the vast majority of bacilli identified in pulmonary lesions of BALB/c mice were intracellular, the majority of bacilli present in the Kramnik mice were extracellular. In addition, comparison of the auramine-rhodamine and Kinyoun carbol-fuchsin staining methods revealed differences in the number of stained bacteria identified in the Kramnik mouse model over the course of infection. In the Kramnik mice, the bacilli appear to lose their AR acid fastness as the disease progressed into large hypoxic, liquefactive, necrotic granulomas. The decrease in AR-positive bacilli is noticeable starting from the center of the necrotic lesion and continuing outward throughout disease progression until eventually all bacteria within the lesion core become AR negative. This is in sharp contrast with the results obtained after Kinyoun staining, where the bacteria within the granulomas remain positively stained over time. Although the exact target of both staining methods is still being debated (53
), these results suggest that the lack of detection with AR over time is due to a change in the bacterial phenotype. As AR still stains bacteria under hypoxic conditions in vitro
), as well as in guinea pig lung lesions (23
), it is unclear why the particular bacterial population in the Kramnik mice causes these negative staining results. Genetic approaches will be required to examine the changes in the bacterial phenotype in the lungs of the Kramnik mice.
To assess the suitability of the Kramnik mouse model for evaluation of new TB drugs and treatment regimens, the efficacies of a panel of five single drugs were compared in Kramnik mice that form necrotic lesions versus BALB/c mice solely forming nonnecrotic or inflammatory lesions. Examination of the bacterial load in the lungs and the spleens revealed that Kramnik mice were significantly more refractory to drug treatment using INH, RIF, LZD, or PZA monotherapy than BALB/c mice. This refractory effect in the Kramnik mice was most pronounced for PZA, as PZA treatment was highly effective in the BALB/c mice (99% of bacteria were eradicated in 4 weeks), whereas no decrease in bacterial CFU was observed in the lungs of the Kramnik mice from the start of treatment. At 150 mg/kg, PZA contained the bacterial load in the lungs of the Kramnik mouse model over the course of treatment, whereas in the untreated Kramnik controls, there was a slowed but continuous growth observed (1 log10
over 7 weeks). This is in contrast to what we expected from previous in vitro
data describing PZA as primarily active against persistent bacilli (21
). Earlier in vitro
data identified some of the conditions required for potentiation of PZA activity. Anaerobic conditions have been shown to enhance the activity of PZA, and older cultures of TB are somewhat more sensitive to PZA (63
). PZA is known to show better in vitro
activity at lower pH (68
) and to have better efficacy against intracellular bacilli in the macrophage infection model (11
). To date, limited information is available on what bacterial population PZA may target in vivo
. A few earlier reports showed activity of PZA against tuberculosis using mouse models infected via different routes and with different inoculum sizes (reviewed in reference 58
). The reason for the failure to observe any cidal activity in the Kramnik mouse model cannot merely be attributed to the extracellular nature of the bacteria or penetration across the granuloma as PZA has been shown to have activity in the guinea pig model of infection (2
). It is remarkable that the mechanism of action for PZA, one of the three pillar TB drugs, remains largely a mystery after many years. This study was limited as only a single PZA drug dose was tested, and therefore it is possible that a higher drug dose might show clear bactericidal activity in vivo
. As the lack of PZA activity in the Kramnik mouse model might have significant clinical implications, it will be crucial to investigate a higher dose range in this mouse model.
Metronidazole, on the other hand, did not show any activity in either mouse infection model, even though excessively high doses were administered. MET, a drug that requires anaerobic conditions for activation, has been previously reported to lack demonstrable activity in mice (8
), whose lesions fail to become hypoxic (60
), but has shown demonstrable activity in rabbits against M. bovis
) and is currently being evaluated in nonhuman primates. Metronidazole also lacked activity in guinea pigs, but the conclusions were limited due to the increased inflammation in the lung and toxicity of the drug (22
). While there are currently no clinical data confirming the activity of MET in TB patients, the additional evaluation in other animal models might indicate which conditions are required in order to observe MET activity in vivo
In the drug studies described here, the bacillary burden in the lungs of both mouse models was high at the start of treatment (6.85 log10 CFU in Kramnik mice and 7.45 log10 CFU in BALB/c mice), which increases the likelihood of drug resistance following single-drug treatment. Although the bacterial load at the start of treatment in BALB/c mice was approximately four times higher than in the Kramnik model, only few single-drug-resistant mutants in a minority of BALB/c mice were observed for INH and RIF after 6 and 8 weeks. This is in great contrast with the drug resistance observed in the Kramnik mouse model, where we observed a substantial increase in the number of bacilli that were drug resistant after 7 weeks of treatment with INH (up to 4.47 log10 resistant CFU) or RIF (up to 5.72 log10 resistant CFU). The exact reason for the higher number of drug-resistant bacteria following treatment in the Kramnik mouse model is not clear, but it is crucial to understand the mechanism as this could have significant clinical implications.
We speculate that the high prevalence of resistant mutants might be due to multiple causes; such as having subtherapeutic drug exposure levels, a weakened immune response, or an altered bacterial mutation rate in the Kramnik mice. First, due to the highly organized and encapsulated nature of the granulomas observed in the Kramnik mice, it is possible that bacteria within the granulomas received subtherapeutic exposure levels of drug due to poor drug penetration. This could facilitate survival and selection of mutants with low-level drug resistance in Kramnik mice, which would not be observed in the primarily inflammatory lesions of BALB/c mice. For moxifloxacin, a recent elegant study measured an increased drug distribution in cellular regions of granulomas and uptake by macrophages, but decreased penetration into the cores of necrotic rabbit lesions (49
). The same was predicted for the standard clinical drugs via a recent mathematical modeling study and confirmed through plasma and tissue pharmacokinetic studies by the same group (29
). A second explanation for the increased frequency of drug resistance in Kramnik mice might be that the precursor mutation frequencies in both mouse models are in fact the same, but the immune pressure in the BALB/c mice selectively kills drug-resistant mutants that are less fit for in vivo
survival than their drug-susceptible counterparts due to its fully activated immune response. Although the C3HeB/FeJ mice do not have any broadly systemic immune deficiencies, an altered or weakened immune response as the result of inactivation of the Ipr1
gene may selectively decrease immune-mediated pressure on drug-resistant mutants. Recent studies with athymic nude mice support this finding, revealing a remarkable propensity to select drug-resistant mutants. Daily treatment with human-equivalent doses of RIF-INH-PZA for 2 months followed by RIF-INH selected INH-resistant mutants in the majority of mice, and this was only prevented by the addition of ethambutol (EMB) (67
). In support of this hypothesis, clinical studies of individuals coinfected with HIV and TB and treated with INH and rifapentine (61
) or rifabutin and INH (9
) were significantly more likely to develop rifamycin monoresistance, especially in patients with the lowest CD4 T cell counts. Rifamycin resistance is rarely observed in immunocompetent individuals, even under conditions of poor treatment compliance, which facilitates the emergence of drug resistance, implicating a positive role for host immunity in suppression of acquired drug resistance. A third explanation, and perhaps the most intriguing one, is the occurrence of a heightened bacterial mutation frequency within the hypoxic regions of necrotic granulomas found in the Kramnik mouse model. Recently, a study using latently infected nonhuman primates described that hypoxic conditions can suppress the expression of DNA repair enzymes, thereby resulting in higher mutation frequencies (20
). Further studies examining the specific mutations of the drug-resistant colonies and the mutation frequencies at different times following treatment are necessary to answer these questions and are under way. Nevertheless, the increased resistance observed in the Kramnik mouse model, which shows clinical relevance, might be an additional benefit of the model to assess resistance of single drugs or drug regimens before entering in clinical trials.
It is important to note that the Kramnik mouse is a highly susceptible mouse model and that there is a natural variability and narrow margin for an infectious inoculum to establish a chronic infection within the animals themselves. As shown by our data, a higher infectious inoculum (~100 CFU) can result in a faster development of disease, while a lower inoculum (~50 CFU) will allow for a slower progression of chronic disease. This adaptability can be useful depending on the experimental aim.
In these studies, we showed that the Kramnik mouse responded to aerosol infection with the formation of hypoxic, encapsulated lung lesions with liquefactive necrosis, which contained numerous extracellular bacilli that were less susceptible to drug therapy than those in BALB/c mice. In contrast, BALB/c mice developed less extensive foci of mixed inflammatory cells that harbored primarily intracellular bacilli and had minimal individual cellular necrosis. Using these two models, we examined the effect lesion morphology had on the response to treatment with a panel of clinically relevant TB drugs that differed in their modes of action. Overall, the results from the experiments presented here demonstrate that the Kramnik mouse model has promise to evaluate novel preclinical drugs in a more stringent animal model possessing liquefactive necrotic granulomas containing persisting bacteria.