Studying latent Mycobacterium tuberculosis (Mtb) infection has been limited by the lack of a suitable mouse model. We discovered that transient depletion of biotin protein ligase (BPL) and thioredoxin reductase (TrxB2) results in latent infections during which Mtb cannot be detected but that relapse in a subset of mice. The immune requirements for Mtb control during latency, and the frequency of relapse, were strikingly different depending on how latency was established. TrxB2 depletion resulted in a latent infection that required adaptive immunity for control and reactivated with high frequency, whereas latent infection after BPL depletion was independent of adaptive immunity and rarely reactivated. We identified immune signatures of T cells indicative of relapse and demonstrated that BCG vaccination failed to protect mice from TB relapse. These reproducible genetic latency models allow investigation of the host immunological determinants that control the latent state and offer opportunities to evaluate therapeutic strategies in settings that mimic aspects of latency and TB relapse in humans.
Infection with Mycobacterium tuberculosis (Mtb) most frequently results in latent tuberculosis infection (LTBI), a clinically asymptomatic state in which the only sign of infection is a positive test for immune reactivity to TB antigens, either a tuberculin skin test or IFNγ release assay. Individuals with LTBI are at risk for developing active TB, and many active TB cases that occur in someone with LTBI, especially in areas of low active TB prevalence, arise from reactivation of the Mtb strain that originally infected the individual. Although most active TB cases may represent progression of a primary Mtb infection that occurred recently, within 1–5 yr (Behr et al., 2018, 2019), even these cases proceed through a paucibacillary, clinically asymptomatic state and therefore fit within the definition of LTBI.
TB chemotherapy often fails to sterilize Mtb infections, resulting in individuals at risk of relapse TB, even in the absence of bacterial drug resistance (Malherbe et al., 2016). In fact, 5% of TB patients who have been treated with standard TB chemotherapy develop recurrent TB (Romanowski et al., 2019; Colangeli et al., 2018; Merle et al., 2014), and recurrent TB may account for up to 10–30% of all TB cases (Chaisson and Churchyard, 2010). In patients with recurrent TB, the Mtb strain that caused the primary infection regrows to lead to a second episode of active TB (relapse), or infection with a new strain causes the second episode (reinfection; McIvor et al., 2017). The proportion of relapse and reinfection among recurrent TB cases varies with the overall TB incidence rate and HIV status (Cardona, 2016). Posttreatment persistent infections, in which drug-susceptible Mtb survives despite chemotherapy and can cause disease relapse, contributing to the spread of TB.
The standard TB mouse model exhibits some features of human TB and has been informative about TB immunopathogenesis, the role of host genetics, the efficacy of antimicrobial therapy, and host–pathogen interactions (Flynn, 2006). However, it fails to mimic human LTBI and persistent infections following TB chemotherapy. Infection of C57BL/6 or BALB/C mice by aerosol with ∼100 CFU of Mtb follows a reproducible disease course consisting of a 3–4-wk-long period during which the bacteria replicate until Mtb antigen-specific T cell responses prevent replication and maintain the infection at a stable bacterial load for many months, until eventual death of the animals (Wolf et al., 2008; Urdahl et al., 2011). During this chronic, persistent infection, the typical mycobacterial load is ∼106 CFU in the lung, consisting of bacilli that likely exist in different subpopulations, which are either killed by the immune system, replicating slowly, or maintain viability in a nonreplicating state (Gill et al., 2009; Wang et al., 2019). The overall bacterial burden remains nearly constant, indicating a balance between the host immune response restricting the infection and Mtb resisting eradication. Although during LTBI in humans, the bacilli are likely in a similar standoff with the immune system, a healthy human immune system restricts bacterial replication much more vigorously and is therefore capable of driving Mtb into a state of latency in which Mtb is believed to be contained at low numbers within granulomas (Barry et al., 2009; Colangeli et al., 2020). In contrast to the high-titer persistence that occurs during the chronic phase of the conventional mouse model, latent infections in humans are defined by paucibacillary stages, which are characterized by low, often undetectable numbers of Mtb and the corresponding lack of clinical symptoms. This is supported by studies in which pairs of guinea pigs were inoculated with lung and lymph node lesion tissues from people without TB symptoms, and a significant number of lesions caused infection in only one of two guinea pigs, implying a low viable burden (Opie and Aronson, 1927). Nonhuman primates with latent Mtb infection had significantly fewer CFU in the lungs than animals with active disease (Lin et al., 2014). However, even animals with active disease displayed significant lesion heterogeneity.
Recently, infection of mice with an ultralow dose of Mtb was shown to cause outcomes that more closely model human TB (Plumlee et al., 2021). Infection with 1–3 CFU resulted in heterogeneous bacterial burdens in lungs, lymph nodes, and spleens, and mice established pulmonary lesions that shared features with human granulomas. Some ultralow dose–infected mice developed a single granuloma with little or no dissemination to the contralateral lung. Despite the variability in bacterial burden, in all infected mice, the bacterial titers increased steadily in lungs; on day 83 after infection, Mtb had disseminated to the spleen, reaching titers of ∼10,000 CFU. Thus, while ultralow-dose infection of mice models aspects of human TB, it does not mimic LTBI and relapse TB.
We sought to develop a mouse model that is characterized by low numbers of persisting bacteria and allows study of posttreatment latent infection. That Mtb can establish latent infection in mice was first demonstrated in the Cornell model, developed by McCune and colleagues at Cornell University (McCune and Tompsett, 1956; McCune et al., 1956, 1966a, 1966b). In this model, mice infected intravenously with Mtb were treated with isoniazid (isonicotinic acid hydrazide [INH]) and pyrazinamide (PZA) until no bacteria could be detected by microscopy and culture on agar plates, even when homogenates of whole lungs, spleens, livers, and every organ including bones from several mice were cultured. After cessation of drug treatment, the infection relapsed spontaneously in ∼30% of animals, and relapse could be induced in the remaining mice by immunosuppression with corticosteroids. The Cornell model has been used to study various aspects of Mtb persistence and latency (Flynn et al., 1998; van Pinxteren et al., 2000; Dhillon and Mitchison, 1994). However, this model is notorious for its varying relapse rates, and it did not result in reproducible relapse when we tried to implement it. To overcome its limitations, particularly the lack of reproducibility but also the need for prolonged antibiotic therapy, which can induce resistance-conferring mutations or drug tolerance (Scanga et al., 1999), we developed models in which latent infection is established via transient genetic inactivation of essential Mtb genes. After Mtb can no longer be detected, target gene expression is restored, and the timing and frequency of spontaneous or corticosteroid-induced relapse indicate the number of mice in which Mtb persisted in the form of a paucibacillary infection. In these mouse models, Mtb persists for months below the limit of detection and reproducibly causes TB relapse in a fraction of infected animals, mimicking aspects of posttreatment latency in humans. Unexpectedly, we discovered that the immunological requirements for control of latent Mtb in these models differ depending on the protein that was transiently depleted to induce latent infection.
Chemotherapy of Mtb-infected mice did not result in consistent latent infection
In an effort to implement the Cornell model, we infected C57BL/6 mice with virulent Mtb, strain H37Rv. Starting 4 wk after infection, we treated the infected mice with INH and PZA for 14 wk (Fig. S1 A). CFU in lungs, spleens, livers, lymph nodes, bone marrow, and kidneys in groups of 5 mice declined with drug treatment, and on day 126, when no CFU were detected in 10 mice, chemotherapy was terminated. On day 182 after infection, 8 wk after the end of chemotherapy, a group of 10 mice were analyzed, and no Mtb was detectable in any of the mouse lungs or other organs. A group of 10 mice received dexamethasone (dexa) for 4 wk. 8 wk later, these 10 immunosuppressed mice and 10 mice that did not receive dexa were analyzed for CFU, but all mice had been sterilized by the drug treatment. In a repeat experiment, we increased the number of mice in each treatment group. But again, in none of the organs from 15 mice that did not receive dexa was any Mtb CFU detectable, and only 1 of 14 mice that had received dexa relapsed with CFU in lung and spleen (Fig. S1 B). In past studies, the frequency of relapse also varied substantially, and sometimes relapse could not be observed at all, even with immunosuppression (Scanga et al., 1999; Lenaerts et al., 2004; Jeon et al., 2012). Therefore, we sought a different approach to reproducibly establish latent Mtb infection in mice.
Genetic depletion of essential proteins reproducibly results in LTBI in mice
We have previously developed genetic switches to control gene expression in Mtb during mouse infection and have demonstrated that the depletion of essential proteins kills Mtb (Ehrt et al., 2005; Kim et al., 2013; Puckett et al., 2017; Lin et al., 2016; Tiwari et al., 2018). A dual-control (DUC) switch that combines transcriptional repression with inducible proteolysis can generate >99.9% repression, quickly inactivates gene activities in replicating and nonreplicating mycobacteria, and provides increased robustness against phenotypic reversion (Kim et al., 2013). We hypothesized that killing of Mtb following transient depletion of essential proteins does not fully sterilize the infection but may lead to paucibacillary infection states during which low numbers of viable bacteria persist and could reactivate to cause TB relapse. We further expected that the depletion of different targets will have different consequences for establishing a latent state and result in reproducibly distinct relapse rates. We selected DUC mutants for biotin protein ligase (BPL) and thioredoxin reductase (TrxB2) to test these hypotheses.
BPL covalently attaches biotin to acyl-CoA carboxylases, and inactivation of BPL inhibits all lipid biosynthesis (Tiwari et al., 2018). Inactivating BPL is thus expected to cause pleiotropic effects, including profound changes in Mtb’s envelope structure, which might make the pathogen exceptionally vulnerable to host immunity and only rarely allow Mtb to establish a persistent, paucibacillary infection. To test this, we infected C57BL/6 mice with a BPL-DUC mutant and, after the mutant had established a chronic infection, eradicated it via doxycycline (doxy)-mediated BPL depletion (Fig. 1 A). After apparent sterilization, doxy was withdrawn to restore BPL expression and reestablish a wild-type paucibacillary infection. At multiple time points after doxy treatment, groups of 10 mice were sacrificed and analyzed for Mtb CFU. We could not detect Mtb CFU in lungs or spleens of the majority of mice. In three experiments, the frequency of relapse varied from 0 to 10% (mean, 6.25%). In two of these experiments, we treated additional groups of mice with dexa for 8 or 4 wk. 11 of 35 mice that were treated with dexa relapsed; thus, the frequency of mice with regrowth in lung and/or spleen increased ∼5-fold from an average of 6.25 to 31.4%. Table 1 summarizes relapse data from multiple experiments with BPL-DUC and reveals that targeting Mtb’s BPL results in a low-frequency-relapse latency mouse model, in which almost 70% of relapse depends on immunosuppression. On average, treatment with dexa increased the relapse rate in mice infected with BPL-DUC by 25%. A Bayesian analysis of these data (see Materials and methods) defines the 95% highest-probability density interval (HDI) for the incremental effect of dexa on relapse as 9.5 to 40.5%, which is statistically significant.
TrxB2 is essential for thiol redox homeostasis and has been considered as a drug target (Lin et al., 2016; Harbut et al., 2015). Depletion of TrxB2 has pleiotropic effects and kills Mtb in vitro and during acute and chronic high-titer mouse infections. However, the importance of redox homeostasis for establishing paucibacillary infection was unclear. We analyzed the consequences of targeting TrxB2 using the TrxB2-DUC mutant in six independent experiments (Fig. 1 B shows data from four experiments). The experimental design was similar to the one used to analyze BPL-DUC. In contrast to BPL-DUC, which caused on average 6.25% relapse without immunosuppression, TrxB2-DUC regrew on average in lungs and/or spleens of 75% of immunocompetent mice after a period of ≥8 wk during which no CFU could be recovered from lungs or spleens in the majority of mice. In two experiments, we included dexa treatment, which caused relapse in 83.3% of the mice (Fig. 1 B and Table 1). Targeting Mtb’s TrxB2 thus results in a high-frequency-relapse (>80% relapse) latency mouse model in which relapse occurs frequently even without immunosuppression. On average, treatment with dexa increased the relapse rate of mice infected with TrxB2-DUC by 10%. A Bayesian analysis defines the 95% HDI for the incremental effect of dexa on relapse for the TrxB2-DUC infected mice as −8.1 to 23.0%, which is not statistically significant.
To simultaneously compare the effect of strain and dexa treatment, we analyzed the combined data using a generalized linear model with a negative binomial likelihood function. The model included independent terms for strain and dexa treatment to predict relapse rates, along with an interaction term between them. We found, via a Wald test on coefficients, a significant difference between the reactivation rates for the two mutants (TrxB2-DUC being ∼79%, and BPL-DUC, only ∼19%, when averaged over dexa treatments). Furthermore, while a generalized effect of dexa on relapse rates was not significant (P = 0.36) when averaged over the two strains, the model did indicate a significant interaction between dexa and BPL-DUC (P = 0.0033 for the interaction term), supporting that treatment with dexa significantly increased relapse rates of mice infected with BPL-DUC (consistent with the results of the Bayesian analysis above).
For both the TrxB2-DUC and BPL-DUC models, we confirmed that in the bacteria that were recovered following relapse, the targeted genes were still regulated in an anhydrotetracycline (atc)/doxy–dependent manner and thus did not represent escape mutants. Both mutants were killed with similar kinetics in mice during doxy treatment, and we confirmed that the minimal inhibitory concentration of atc and doxy required to suppress growth in vitro was not significantly different for the two mutants (Fig. S2).
To investigate if viable Mtb was detectable in organs other than lung and spleen, we analyzed multiple organs and found Mtb in the lymph nodes, liver, kidney, bone marrow, and thymus during acute infection (Fig. 1, C and D). These bacteria were eliminated with doxy treatment similarly to those in lungs and spleens, and we did not detect Mtb in any of the surveyed organs during latency. We previously demonstrated that depletion of BPL and TrxB2 is accompanied by progressive healing of lesions in the lungs of infected mice (Tiwari et al., 2018; Lin et al., 2016) Previous work also demonstrated that doxy penetrates every mouse tissue and organ to the extent needed to silence Mtb genes controlled by the DUC system (Gengenbacher et al., 2020). Thus, it is unlikely that the bacilli survived in a doxy-protected site.
We found that Mtb reactivated most frequently in lungs, followed by spleens, lymph nodes, and thymus (Fig. 1, C and D), suggesting that the primary site of paucibacillary persistence in aerosol-infected mice is the lung. We cannot distinguish whether TB relapse is caused by very few surviving organisms that may be below the limit of detection by culture on agar plates or is due to a larger population of bacilli in a metabolically altered state that does not support colony formation on agar plates. We tested if Mtb was detectable with a liquid resuscitation assay (Fig. S3, A and B), which resuscitated Mtb from sputa from TB patients, chronically infected mice, and in vitro stressed cultures (Chengalroyen et al., 2016; Hu et al., 2015; Saito et al., 2017; Imperial et al., 2018). However, this method also failed to identify viable bacilli during latency, and during active infection, it yielded numbers similar to those obtained by CFU determinations of infected organ homogenates on agar plates. In summary, the data presented here demonstrate that conditional depletion of an essential Mtb protein can induce latency of Mtb in mice, which is defined by a period of time lasting from weeks to months during which Mtb CFU are not detectable by culturing organ homogenates and that is followed by spontaneous or immune suppression–induced regrowth of Mtb in a significant proportion of mice.
Control of TrxB2-DUC requires adaptive immunity
These new models of TB latency provided an opportunity to interrogate the immune factors that control the Mtb latent state. T cells are crucial for the control of chronic Mtb infections (Jasenosky et al., 2015), and we therefore analyzed the pulmonary T cell populations in BPL-DUC– and TrxB2-DUC–infected mice at different time points of relapse experiments. As expected, the proportion of CD4, and to some degree CD8, T cells in the lung increased during acute infection and decreased with the decline of bacterial titers (Fig. 2, A–C). This included antigen 85–specific CD4 T cells (Fig. 2 D). There were no significant differences in the proportions of cytokine producing T cells between mice infected with either mutant (Fig. S3 C). We interrogated whether lung-infiltrating T cells resembled effector memory T cells (CD44hiCD62Llo) and resident memory T cells (CD44hiCD62LloCD69+). This revealed that latent infection with TrxB2-DUC was accompanied by an increased fraction of CD4 lung-resident memory T cells and CD4 effector memory T cells compared with mice infected with BPL-DUC 5 and 7 mo after infection (Fig. 2). Memory CD8 T cell populations were not significantly different between mice infected with the two mutants, except that at 7 mo after infection, the proportion of CD8 lung-resident memory T cells was higher in TrxB2-DUC–infected mice compared with BPL-DUC–infected mice.
To assess whether adaptive immunity is required to control latent Mtb, we infected severe combined immunodeficiency (SCID) mice, which lack functional B and T cells, with both mutants. Because SCID mice are highly susceptible to Mtb, we started doxy treatment at 2 wk after infection, kept the mice on doxy for 8 wk, which cleared the infection, and then monitored relapse (Fig. 3). BPL-DUC reactivated only in 1 of 14 mice and, in this mouse, could be isolated from multiple organs (Fig. 3, A and B). In contrast, TrxB2-DUC did not establish latent infection in SCID mice, as it regrew almost immediately after removal of doxy and disseminated to all organs we cultured (Fig. 3, C and D). Already 8 wk after doxy treatment, 87.5% of SCID mice (7 of 8) had relapsed with CFU in the lungs (Fig. 3 C). In contrast, only 13% of C57BL/6 mice (2 of 15) had relapsed at 8 wk after doxy treatment, and it took ∼20 wk for the majority of C57BL/6 mice to relapse (Fig. 1 B). Thus, the majority of reactivation occurred <8 wk after doxy in SCID mice, and >12 wk after doxy in C57BL/6 mice. Using a binomial statistical model, the P value for relapse occurring at the same time in the two mouse models was 5.8 × 10−11, strongly supporting that relapse occurred earlier in SCID than in C57BL/6 mice. This identifies the adaptive immune response as essential to establish latent Mtb infection following depletion of TrxB2. Control of BPL-DUC occurred even in the absence of adaptive immunity. The reactivation frequency of BPL-DUC in SCID mice was not significantly different from that in C57BL/6 mice (Fig. 1 A; P = 0.742 by χ2 test), indicating that adaptive immune responses are not crucial to establish latency after inactivation of BPL. However, the differences in experimental setup, including the earlier start (2 wk instead of 4 wk after infection) and shorter period of doxy treatment (66 d instead of 112 d) of the SCID mice complicates the direct comparison of these experiments. Future work with identical experimental parameters is required to substantiate these findings and their interpretation.
BCG vaccination does not prevent TrxB2-DUC reactivation
Vaccination with Mycobacterium bovis Bacille Calmette–Guérin (BCG) protects infants from extrapulmonary manifestations of Mtb infection and against progression from infection to active disease (Roy et al., 2014; Trunz et al., 2006). In contrast, its effectiveness in preventing pulmonary TB disease in adults is highly variable (Mangtani et al., 2014). We sought to determine if BCG vaccination reduces TB relapse in our genetic latency model. To assess this, we vaccinated mice that were latently infected with TrxB2-DUC with BCG Pasteur. We vaccinated via the i.v. route because in nonhuman primates, high-dose i.v. BCG vaccination was recently demonstrated to prevent or substantially limit Mtb infection (Darrah et al., 2020). We analyzed immune responses in BCG-vaccinated mice and PBS-treated controls 25 d after vaccination and determined reactivation of Mtb in lungs and spleens of both groups on day 280 after infection, 20 wk after vaccination (Fig. 4). BCG vaccination did not affect reactivation of latent Mtb in the lungs of mice; almost all mice in the vaccinated and control groups had CFU in the lungs (Fig. 4 A). In the spleen, however, Mtb regrew in only 5 of 21 vaccinated mice, whereas 17 of 21 PBS-treated control mice yielded CFU from spleen (Fig. 4 B). BCG vaccination induced more cytokine-expressing CD4 T cells in the spleen than in the lung; in particular, the number of IL-2–producing CD4 T cells only increased in the spleen of vaccinated mice (Fig. 4, C and D). This might be due to the higher BCG titers in spleens compared with lungs of vaccinated mice (Fig. 4 E). BCG titers declined in the lungs ∼10-fold from day 1 to day 24 and then remained stable until day 40 after immunization. In spleen, BCG CFU did not significantly decline during that period and remained ∼1.5 log10 higher than in lungs. The proportion of effector memory CD4 and CD8 T cells increased in both lungs and spleen in response to BCG vaccination (Fig. 4, F and G). In summary, these data demonstrate that BCG-induced immune responses are not sufficient to prevent reactivation of latent TrxB2-DUC infection in this mouse model.
To address the need for mouse models of LTBI and relapse TB, we developed models of paucibacillary infection in mice, which cause reproducible rates of relapse. We demonstrated that Mtb persists for months below the limit of detection and causes relapse TB in a proportion of infected animals. Although these mouse models do not represent the equivalent of human latency, they can likely inform our understanding of human TB, via comparison, of analogous human TB disease states, such as reactivation from latency and relapse after TB treatment. These are both states of paucibacillary infection in which bacterial and host immune factors interact to determine whether Mtb reappears or remains clinically silent.
Depletion of two essential proteins, BPL and TrxB2, resulted in two models of LTBI, followed by strikingly different relapse frequencies and different immune requirements for control. Adaptive immunity was crucial for control of LTBI following TrxB2 depletion, whereas it was dispensable for preventing relapse of LTBI after BPL-DUC depletion. Of note, after the end of doxy treatment, both mutants can express the depleted protein and are phenotypically identical and indistinguishable from wild-type Mtb. Thus, the way LTBI is established in these models seems to determine how infection is controlled. Most likely, BPL depletion, which interferes with cell envelope biosynthesis (Tiwari et al., 2018), synergizes more effectively with host immunity in controlling Mtb than depletion of TrxB2, which perturbs growth-essential pathways, including sulfur metabolism, DNA replication, and antioxidant defense systems (Lin et al., 2016). Although the population size of viable Mtb in the two LTBI models is unknown, the increased proportion of CD4 memory T cells in the lungs of TrxB2-DUC infected mice might reflect a larger latent TrxB2-DUC population that eventually causes relapse than the persisting BPL-DUC population, which the mice are able to control better. Currently, it cannot be excluded that the infrequent reactivation of BPL-DUC in C57BL/6 and SCID mice compared with TrxB2-DUC might be attributed to differences in mutant fitness or tissue distribution. Future experiments that explore these questions as well as the impact of innate immune cells, such as NK cells, which have been associated with LTBI in humans and nonhuman primates (Esaulova et al., 2021), will help shed light on the mechanisms that lead to control of the BPL-DUC mutant.
Systemic BCG vaccination did not prevent relapse in our LTBI TxrB2-DUC model. It elicited T cell responses and reduced Mtb regrowth in spleens but had no measurable impact on regrowth in lungs. Vaccination-induced immune responses might reduce dissemination of reactivating Mtb if reactivation primarily starts in the lungs, as our data suggest (Fig. 1, C and D). Postexposure vaccines that prevent reactivation of LTBI have been debated for many years, but there has been no consistent success in using prophylactic vaccine candidates as therapeutic vaccines (reviewed in Andersen, 2007; Andersen and Woodworth, 2014). The TrxB2-DUC latency model can be used to test therapeutic vaccine candidates or strategies for their efficacy in preventing TB relapse.
The mouse latency models described here allow the study of aspects of TB biology that are not accessible in the conventional TB mouse model. They provide an opportunity to define the host cells and molecules that prevent transition from LTBI to acute TB. They could help prioritize drug candidates that eradicate latent populations, and thus shorten chemotherapy, and they could facilitate the evaluation of TB postexposure vaccine candidates.
Materials and methods
Bacterial strains and culturing conditions
Mtb H37Rv was used as virulent wild-type strain. BPL-DUC and TrxB2-DUC have been previously described (Tiwari et al., 2018; Lin et al., 2016). BCG Pasteur was from ATCC strain TMC 1011 (#35734). Strains were cultured in liquid Middlebrook 7H9 medium supplemented with 0.2% glycerol, 0.05% Tween 80, and ADN (0.5% BSA, 0.2% dextrose, and 0.085% NaCl) and on Middlebrook 7H10 agar plates supplemented with 0.2% glycerol and Middlebrook OADC enrichment (Becton Dickinson). Antibiotics were added for selection of genetically modified strains at the following concentrations: hygromycin (50 µg/ml), kanamycin (25 µg/ml), and zeocin (25 µg/ml).
Female 8–10-wk-old C57BL/6 (#000664; Jackson Laboratory) or SCID (#001913; Jackson Laboratory) mice were infected with the indicated Mtb strains using an inhalation exposure system (Glas-Col) with a mid-log phase Mtb culture to deliver ∼200 bacilli per mouse. To vaccinate, mice were injected in the lateral tail vein with 100 µl BCG suspension in PBS at a concentration of 107/ml. Mice received doxy-containing mouse food (2,000 ppm; Research Diets) for the indicated periods. For the Cornell model, mice received INH (125 µg/ml) and PZA (15 g/liter) in the drinking water for 14 wk. To enumerate CFU, organs were homogenized in PBS and cultured on 7H10 agar. Charcoal (0.4%, wt/vol) was added to the plates that were used to culture homogenates from doxy- and antibiotic-treated mice. Agar plates were incubated for 3–4 wk at 37°C. We identified suppressor mutants by culturing half of each organ homogenate on 7H10 plates containing atc (500 ng/ml). When suppressor mutants grew up, which was very rare, the infected mice were excluded from the experiment. Mice were housed in a BSL3 vivarium. All mouse experiments were approved by and performed in accordance with requirements of the Weill Cornell Medicine Institutional Animal Care and Use Committee.
Limited dilution assay and most probable number (MPN) calculations
Half of each lung homogenate (three right lung lobes resuspended in 2 ml PBS supplemented with 2% [vol/vol] BD MGIT PANTA antibiotic mixture; #245114; BD) was used to set up the limited dilution series as described previously (Saito et al., 2017). Mtb H37Rv culture filtrate was prepared, and MPN calculations were performed as described (Saito et al., 2017).
Mouse lungs were isolated and placed in RPMI 1640 containing Liberase Blendzyme 3 (70 µg/ml; Roche) and DNase I (50 µg/ml; Sigma-Aldrich). Lungs were then cut into small pieces and incubated at 37°C for 1 h. The cells were filtered using cell strainers, collected by centrifugation, resuspended in ammonium-chloride-potassium hemolysis buffer (Thermo Fisher Scientific), and incubated for 10 min at room temperature. Cells were then washed with PBS and resuspended in splenocyte medium (RPMI 1640 supplemented with 10% FBS, 2 mM GlutaMax, 10 mM Hepes, and 50 µM 2-mercaptoethanol). For intracellular staining samples, cells were stimulated with purified protein derivative (PPD; 20 µg/ml) in the presence of anti-CD28 antibody (37.51; BioLegend) for 1.5 h, and 10 µg/ml Brefeldin A (Sigma-Aldrich) and monesin were added and incubated at 37°C for another 3 h. Samples were kept on ice in a refrigerator overnight. Cells were stained with Zombie-UV (BioLegend) to discriminate live and dead cells. Purified anti-CD16/32 antibody (93; BioLegend) was used to block Fc receptor before staining. PerCp-Cy5.5 anti-CD62L (MEL-14; Thermo Fisher Scientific), Alexa Fluor 700 anti-CD45 (30-F11; BioLegend), BV605 anti-CD4 (RM4-5; BioLegend), BV711 anti-CD8 (53-6.7; BioLegend), APC Cy7 anti-CD69 (H1.2F3; BioLegend), and BUV395 anti-CD44 (IM7; BD Biosciences) were used to stain cells for 30 min at room temperature. PE-conjugated I-A(b) Mtb Ag85B precursor 280–294 (FQDAYNAAGGHNAVF) from the National Institutes of Health tetramer core facility (Atlanta, GA) was used to stain for Ag85-specific T cells. The cells were fixed in fixation buffer (BioLegend) for 30 min, taken out of BSL3, and incubated for 20 min in permeabilization buffer (eBioscience) before intracellular cytokine staining. FITC anti-TNF (MP6-XT22 BioLegend), BUV737 anti-IFNγ (XMG1.2; BD Biosciences), APC anti-IL-2 (JES6-5H4; BioLegend), and BV785 anti-CD3 (17A2; BioLegend) were used to stain cells for 30 min. Cells were washed and resuspended in cell staining buffer (BioLegend). Flow cytometry data were acquired on a cytometer (LSR Fortessa; BD Biosciences) and analyzed with FlowJo v10. Gating strategies are depicted in Fig. S3.
Flow cytometry data were analyzed by one-way ANOVA with Tukey’s multiple comparison test or by unpaired, two-tailed t test, and differences in cell frequencies between TrxB2-DUC– and BPL-DUC–infected animals are reported using Prism 9.0.1 (GraphPad Software).
Effect of dexa on relapse rates for BPL-DUC and TrxB2-DUC
To determine a posterior distribution on the Δs (dexa effect for each strain), we first generated a representative sample from the posterior of each relapse rate variable by Monte Carlo sampling (10,000 random draws from the updated beta posterior distributions, using α = 1, β = 1 for the hyperparameters of the beta prior), and then took the difference of samples to generate the distribution over the Δs: δs ~ ϕs,d = yes − ϕs,d = no.
To quantify the statistical significance, the 95% HDI was extracted from the Monte Carlo sample representing the posterior distribution for each Δ. The 95% HDI for the incremental effect of dexa on reactivation of BPL-DUC is 9.5–40.5%, which is statistically significant (because it does not overlap 0). The 95% HDI for the incremental effect of dexa on relapse for the TrxB2-DUC–infected mice is −8.1 to 23.0%, which is not statistically significant (because it overlaps 0). This analysis supports that dexa treatment has a significant effect on the relapse rate for BPL-DUC but not for TrxB2-DUC.
Linear model for evaluating combined effects of strain and dexamethasone treatments
In the resulting model, the difference between the relapse rates for the two mutants (TrxB2 being ∼79% and BPL only ∼19% when averaged over dexa treatments) was found to be weakly statistically significant (P = 0.0664 by Wald’s test). Furthermore, while a generalized effect of dexa on relapse rates was not significant when averaged over the two strains (P = 0.36), the model did indicate a significant interaction between dexa and the BPL mutant (P = 0.0033 for the interaction term), supporting that treatment with dexa significantly increased relapse rates of mice infected with the BPL mutant (consistent with the results of the Bayesian analysis above).
Analysis of difference in timing of relapse of TrxB2-DUC–infected C57BL/6 vs. SCID mice
TrxB2-DUC–infected SCID mice relapsed with high frequency (∼90%, similar to the relapse rate in C57BL/6 mice at ∼75%; no dexa), and relapse occurred early. Data in Figs. 1 B and 3 C show that the majority of SCID mice had relapsed by 8 wk after doxy treatment, whereas relapse of the majority of C57BL/6 mice occurred >12 wk after doxy treatment. To assess the statistical significance of this difference in timing of relapse, we tested the null hypothesis that relapse occurred in C57BL/6 mice as early as in SCID mice (<8 wk) and that the low proportion of mice with CFU in C57BL/6 mice at weeks 8 and 12 after doxy was just due to sampling error. The maximum likelihood estimate of the longer relapse rate of C57BL/6 mice was 75.0% (60/80, based on pooling the mice relapsed at 20 and 23 wk after doxy). Assuming reactivation had occurred by week 8 after doxy in C57BL/6 mice, the P value for the outcomes observed at weeks 8 and 12 after doxy (combined, 3 of 25 mice relapsed) is the cumulative of the binomial for r ≤ 3 given n = 25 and P = 0.75, which is P = 5.8 × 10−11, strongly supporting that relapse occurred later in C57BL/6 than SCID mice.
Online supplemental material
Fig. S1 displays the data from the Cornell mouse model experiments. Fig. S2 displays data demonstrating that BPL-DUC and TrxB2-DUC are equally susceptible to atc- and doxy-mediated growth inhibition. Fig. S3 shows data revealing that there is no significant difference in detecting TrxB2-DUC by CFU and liquid outgrowth assay and no difference in cytokine-producing CD4+ T cells in lungs from mice infected with BPL-DUC and TrxB2 DUC. The figure also shows the gating strategies for CD4+ and CD8+ T cells and representative flow cytometry plots identifying memory T cells.
We thank J. Andres, N. Betancourt, I. DaSilva, J. Kim, J. McConnell, R. Aguilera Olvera, N. Song, and P. Pino Tamayo for technical assistance; K. Saito, T. Warrier, K. Burns, C. Nathan, M. Glickman, K. Rhee, and V. Dartois for advice and helpful discussions; and M. Glickman, C. Nathan, and K. Rhee for critical reviewing of this manuscript and constructive suggestions.
This work was supported by the Tri-Institutional TB Research Unit (National Institutes of Health grant U19 AI111143).
Author contributions: H. Su, K. Lin, D. Tiwari, C. Healy, C. Trujillo, Y. Liu, D. Schnappinger, and S. Ehrt conceived the study and analyzed the data; H. Su, K. Lin, D. Tiwari, C. Healy, C. Trujillo, and Y. Liu performed the experiments; T.R. Ioerger performed statistical analyses; S. Ehrt wrote the manuscript with input from the other authors.
Disclosures: The authors declare no competing interests exist.
D. Tiwari’s present address is Centre for Global Health, Barts, and The London School of Medicine and Dentistry, Queen Mary University of London, Whitechapel, London, UK.
C. Healy’s present address is School of Medicine, Trinity College Dublin, Dublin, Ireland.