Attenuated Mycobacterium tuberculosis vaccine protection in a low-dose murine challenge model

Summary Bacillus Calmette–Guérin (BCG) remains the only approved tuberculosis (TB) vaccine despite limited efficacy. Preclinical studies of next-generation TB vaccines typically use a murine aerosol model with a supraphysiologic challenge dose. Here, we show that the protective efficacy of a live attenuated Mycobacterium tuberculosis (Mtb) vaccine ΔLprG markedly exceeds that of BCG in a low-dose murine aerosol challenge model. BCG reduced bacterial loads but did not prevent establishment or dissemination of infection in this model. In contrast, ΔLprG prevented detectable infection in 61% of mice and resulted in anatomic containment of 100% breakthrough infections to a single lung. Protection was partially abrogated in a repeated low-dose challenge model, which showed serum IL-17A, IL-6, CXCL2, CCL2, IFN-γ, and CXCL1 as correlates of protection. These data demonstrate that ΔLprG provides increased protection compared to BCG, including reduced detectable infection and anatomic containment, in a low-dose murine challenge model.


INTRODUCTION
Tuberculosis (TB) is a leading cause of mortality from infectious disease worldwide with more than 1.5 million deaths in 2020. 1 Bacillus Calmette-Gué rin (BCG) demonstrates high efficacy against disseminated infection in children, but lifetime efficacy in adults ranges widely between 0% and 80%. 2 Despite this limited efficacy, BCG has remained the sole clinically approved vaccine for nearly a century. 3 Accordingly, the development of nextgeneration vaccines with favorable safety and improved efficacy profiles relative to BCG is an urgent global health priority. 4 We recently developed a live attenuated vaccine derived from the virulent Mycobacterium tuberculosis (Mtb) H37Rv strain termed DLprG with a deletion in rv1411c-rv1410c, an operon that encodes a lipoprotein (LprG) and transmembrane efflux pump (Rv1410) that function as a lipid transporter whose disruption results in altered lipid content of the Mtb cell wall as well as its metabolic state. 5 DLprG was well tolerated in immunocompromised mice and demonstrated greater immunogenicity and reductions in bacterial burdens than BCG after 100 colony-forming unit (CFU) aerosol challenge. 5,6 The murine challenge model has proved valuable for studying TB pathophysiology and vaccine development. For example, the essential contributions of CD4 T cells 7 interferon gamma (IFN-g) 8,9 and tumor necrosis factor alpha (TNF-a) 10 were described in mice. Furthermore, the antigens comprising the M72/AS01 E TB vaccine candidate were characterized in mice. 11,12 However, Mtb is likely transmitted by small respiratory droplets containing few bacilli, 13,14 and the widely employed 100 CFU murine aerosol challenge model generally fails to recapitulate key hallmarks of human disease including granulomatous inflammation 15 and heterogeneous infections. 16 Moreover, the protective efficacy of vaccines is limited to moderate reductions in bacterial burdens, complicating the preclinical interpretation of results in this model. Accordingly, the development of preclinical models that are both experimentally tractable and more reminiscent of human disease is important for next-generation vaccine development. 17 Recently, a lower dose 1-3 CFU murine aerosol model showed heterogeneous bacterial loads as well as a subset of mice with unilateral infection, and barcoding experiments showed that most cases of bilateral infection were driven by dissemination of a single infecting bacterium to the contralateral lung. 18 Moreover, additional studies in this model found that BCG immunization improved long-term control of bacterial loads, prevented dissemination of barcoded bacilli to contralateral lung, and reduced the proportion of mice with detectable infection, although these studies required large numbers of mice due to the low protective efficacy of BCG. 19,20 In this study, we evaluated the protective efficacy of BCG and the more potent live attenuated DLprG vaccine candidate in the low-dose model. 5,6 Compared to BCG, DLprG yielded protection from detectable infection in a substantial subset of mice and anatomic containment in all animals with breakthrough infection.

DLprG is immunogenic and protective after 100 CFU challenge
We first sought to confirm the immunogenicity and protective efficacy of the live attenuated DLprG vaccine strain against a 100 CFU challenge. We focused these studies in C3HeB/FeJ mice, which exhibit susceptibility to Mtb infection and show granulomas with central caseous necrosis that are reminiscent of human disease. 21 Mice were vaccinated at week 0 with BCG or DLprG, and we measured purified protein derivative-specific T cell responses in peripheral blood mononuclear cells (PBMCs) at week 2 by intracellular cytokine staining. BCG did not elicit significant CD4 IFN-g-secreting ( Figure 1A), CD4 TNF-a-secreting ( Figure 1B), or CD8 IFN-g-secreting ( Figure 1C) responses in PBMCs at this time point. In contrast, DLprG stimulated significant cytokine-secreting T cells responses of all three phenotypes ( Figures 1A-1C). CD4 IL-2 responses were not detected as previously reported. 5 Next, we performed multiplexed serum cytokine analysis in naive and vaccinated mice at week 2.5. Compared to naive mice, BCG-vaccinated mice showed upregulation of serum cytokines including IFN-g, TNF-a, and interleukin-17A (IL-17A), among others (Figures S1A and S1B). DLprG also Figure 1. DLprG is more immunogenic and protective than BCG following 100 CFU H37Rv challenge in C3HeB/ FeJ mice (A-C) Groups of C3HeB/FeJ mice were immunized with BCG (n = 5) or DLprG (n = 5) at week 0 followed by PBMC ICS following stimulation with purified protein derivative (PPD) at week 2 to quantify subsets including CD4 IFN-g+ T cells (A), CD4 TNF-a+ T cells (B), and CD8 IFN-g+ T cells (C). (D) Challenge study design (D). (E) Groups of C3HeB/FeJ mice were immunized with BCG (n = 5) or DLprG (n = 5) at week 0 followed by 100 CFU H37Rv aerosol challenge at week 8 and lung and spleen harvesting at week 12 for bacterial load quantification. Lung CFU from the challenge study (E). (E) Bottom dotted line represents assay LOD of 5,000 CFU. Spleen CFU from the challenge study (F). Bottom dotted line represents assay LOD of 500 CFU. For (A)-(C), data are represented as median +/À interquartile range. For (E) and (F), bars represent group medians. For all panels, p values represent pairwise Mann Whitney U tests. For all panels, * represents p < 0.05 and ** represents p < 0.01. iScience Article stimulated upregulation of serum cytokines including IFN-g, TNF-a, and IL-17A, among others (Figures S1A and S1B). There were no cytokines differentially detected between BCG and DLprG, although DLprG showed a trend toward greater IL-17A levels ( Figure S1C) as previously reported. 5 We next assessed the protective efficacy of BCG and DLprG against a 100 CFU H37Rv aerosol challenge. Groups of C3HeB/FeJ mice (n = 5 per group) were vaccinated at week 0 with BCG or DLprG, challenged at week 8 with 100 CFU of H37Rv by the aerosol route, and lungs and spleens were harvested at week 12 for bacterial load quantification. DLprG yielded a greater reduction in bacterial loads in the lung relative to BCG, as we previously reported, 5 and both vaccines reduced bacterial loads in the spleen ( Figures 1E and 1F). Thus, DLprG showed a more favorable immunogenicity and protective efficacy profile against 100 CFU challenge.

Characterization of 1 MID50 challenge in C3HeB/FeJ mice
To adapt the low-dose challenge model, we first performed a log 10 in vivo titration of a single-cell suspension H37Rv challenge stock to determine a dose that produced a 60%-70% infection rate 18 (data not shown). We next performed a more focused log 2 -scale in vivo dose-finding study, harvesting separately dissected right and left lung lobes 4 weeks after challenge of 30 C3HeB/FeJ mice (n = 10 mice per group) with the singlecell suspension H37Rv challenge stock. This study yielded infection rates of 10%, 60%, and 80% with a limit of detection (LOD) of 5 CFU per lung lobe (Figures 2A-2D). These studies demonstrated heterogeneous bacterial loads spanning an approximately 4 log 10 range and showed that subsets of mice demonstrated unilateral infection with Mtb CFU detected solely in the right or left lung lobe ( Figures 2E and 2F).
In order to formulate a quantitative nomenclature for challenge doses, and given that the intermediate challenge dose yielded a 60% infection rate ( Figure 2B), we termed this challenge dose 1 median infectious dose 50 (1 MID50). We further calculated that the bacterial inoculum required for 100 CFU challenge was approximately 2 log 10 higher than for the 1 MID50 challenge, and we therefore termed this challenge dose 100 MID50.

DLprG reduces infection and dissemination after 1 MID50 challenge compared to BCG
We next performed vaccine studies with BCG and DLprG using 1 MID50 challenge ( Figure 3A). Three cohorts of C3HeB/FeJ mice (n = 18 mice per cohort, n = 54 total mice) were divided equally into three groups including naive, BCG, and DLprG. Mice were vaccinated at week 0 and underwent a 1 MID50 H37Rv aerosol  18 Among infected naive animals, there were 4/13 (31%) unilateral infections ( Figure 3B). As before, we observed a broad distribution of bacterial loads with a mean lung lobe bacterial burden of 4.90 log 10 CFU ( Figure 3B).
In the BCG group, we observed an infection rate of 16/17 (95%) including a unilateral infection rate of 6/16 (38%, Figure 3B). We used an exact logistic regression model to compare both rates of mouse infection (in either one or both lungs) and rates of dissemination (among infected mice to both lungs) between vaccine groups. We observed no difference in the rate of infection or dissemination between the naive and BCG groups (p = 0.12 for both comparisons, exact logistic regression, Figures 3C and 3D). In addition, we used an ordinal logistic regression model to compare a composite outcome including both infection and dissemination between vaccine groups. We observed no difference in this composite outcome between the naive and BCG groups (p = 0.29, ordinal logistic regression). However, BCG did yield a 0.58 log 10 reduction in mean lung lobe CFU relative to the naive group (p < 0.001, mixed effects negative binomial, Figure 3E). Thus, BCG reduced bacterial burdens but failed to abort establishment or dissemination of infection after 1 MID50 challenge.
In contrast to BCG, DLprG yielded an infection rate of 7/18 (39%), and all breakthrough mice (7/7, 100%) demonstrated unilateral infection ( Figure 3B). The infection rate in the DLprG group was lower compared to both the naive and BCG groups (p = 0.049 and p = 0.005, respectively, exact logistic regression, Figure 3C), and the unilateral infection rate was higher compared to both the naive and BCG groups (p = 0.009 and p = 0.014, respectively, exact logistic regression, Figure 3D). Moreover, the composite outcome of infection and dissemination was lower in the DLprG group compared to both the naive and BCG groups (p = 0.002 and p = 0.001, respectively, ordinal logistic regression). Finally, the DLprG group showed a 1.3 log 10 reduction in mean lung lobe CFU relative to the naive group, which also represented a 0.67 log 10 reduction relative to the BCG group (p < 0.001 and p = 0.023, respectively, mixed effects negative binomial, Figure 3E). These data demonstrate that DLprG vaccination resulted in a striking reduction in both the establishment and dissemination of infection in this model.

Repeated 1 MID50 challenge infects most mice and shows greater stringency
In order to better model real-world dynamics including repeated exposure, 22,23 we designed studies incorporating repeated 1 MID50 challenge ( Figure 4A). Two cohorts of C3HeB/FeJ mice (n = 18 mice per cohort, n = 36 total mice) were divided equally into three groups including naive, BCG, and DLprG. Each cohort was vaccinated at week 0, underwent four consecutive 1 MID50 H37Rv challenges at weeks 8, 9, 10, and 11, and at week 15, right and left lungs were dissected separately for bacterial load quantification. For a third cohort (n = 18 mice), lungs were fixed in formalin at week 15 for histopathological studies.
As expected, we observed an increased proportion of infected mice and a reduced proportion of mice with unilateral infection in the repeated challenge model ( Figure 4B). Specifically, in the naive group, we observed that 11/12 (92%) of mice became infected, and only 1/11 (9%) mice demonstrated unilateral infection. Similar to single 1 MID50 challenge, there was a wide distribution of bacterial loads. In the naive group, repeated challenge also yielded increased average lung lobe bacterial loads relative to single low-dose challenge (5.69 log 10 CFU vs. 4.90 log 10 CFU, respectively, p = 0.002, Mann-Whitney U test). Histopathologic studies showed that repeated 1 MID50 challenge granulomas (up to 7 weeks post-initial challenge) in naive mice were indistinguishable in size and composition from those observed after 100 MID50 challenge (4 weeks post-challenge), consistent with early establishment of infection and prolonged bacterial replication ( Figures S2 and S3A-S3C).
In the BCG group, we observed an infection rate of 12/12 (100%) including a unilateral infection rate of 2/12 (17%, Figure 4B). Thus, BCG vaccination did not reduce the rate of infection or increase the rate of unilateral infection ( Figures 4C and 4D). In contrast, BCG showed a substantial 0.99 log 10 reduction in lung lobe bacterial loads compared to the naive group ( Figures 4B and 4E, p < 0.001, mixed effects negative binomial). In the DLprG group, we observed an infection rate of 9/9 (100%) including a unilateral infection rate of 4/19 (44%, Figure 4B). In contrast to single 1 MID50 challenge, following repeated challenge DLprG did not reduce infection rates relative to the naive and BCG groups ( Figure 4C). However, we observed a trend iScience Article toward an increased proportion of mice with unilateral infection relative to the naive (44% vs. 11%) and BCG (44% vs. 17%) groups ( Figure 4D). Finally, the DLprG group showed a 0.85 log 10 reduction in average lung lobe bacterial loads relative to the naive group ( Figures 4B and 4E, p = 0.016, mixed effects negative binomial model). Mice vaccinated with either BCG or DLprG had fewer and smaller granulomas than naive mice after repeated low-dose challenge, consistent with a delay in acquisition of disease or greater control of bacterial replication ( Figure S3). Increased numbers of lymphocytes as well as increased perivascular lymphocytic cuffing were observed in granulomas from vaccinated mice.

Repeated 1 MID50 challenge facilitates correlates of protection analyses
We evaluated post-vaccination, pre-challenge serum cytokine levels among one cohort of repeated lowdose challenge mice in which all (6/6, 100%) naive animals became infected and compared them with post-challenge whole-lung CFU on a per-mouse basis. We observed that 7 of 35 assayed post-vaccination serum cytokines were negatively correlated with post-challenge bacterial loads ( Figure 5). Notably, the iScience Article serum cytokine most correlated with reductions in bacterial load was IL-17A, a molecule that we and others have found to be correlated with protection in both mouse 5,24,25 and macaque 26 vaccine studies as well as natural infection studies in macaques 27 and humans. 28 Additional correlates included IL-6 and CXCL2 as we previously described in the 100 MID50 challenge model 5 as well as others including CCL2 and CXCL1 which to our knowledge have not previously been described as correlates of TB vaccine protection.

DISCUSSION
Preclinical TB vaccine studies have primarily relied on a 100 CFU murine aerosol challenge that does not model the low-dose dynamics that are thought to mediate transmission in humans. Moreover, BCG and iScience Article other preclinical vaccine candidates only show moderate reductions in bacterial loads, complicating the interpretation of vaccine protection in this model. In this study, we demonstrate striking protective efficacy of the live attenuated DLprG preclinical vaccine candidate against a low-dose challenge including prevention of detectable infection among a substantial subset of mice and anatomic containment to a single lung among all mice that are infected. These data provide an experimentally tractable model of substantive vaccine protection against TB and corroborate use of the low-dose model for preclinical vaccine development.
We show that DLprG reduces bacterial loads as well as two additional measures of vaccine protection that are not captured by the 100 CFU challenge model: prevention of detectable infection and prevention of dissemination to the contralateral lung. These data corroborate other recent work showing that BCG reduces detectable infection and disseminated infection in addition to traditional bacterial load measurements, although these studies required substantially larger numbers of mice due to the lower protective efficacy of BCG. 19,20 Therefore, the capacity of DLprG to abrogate detectable infection and disseminated infection with relatively small group sizes may provide an experimentally tractable model for assessing potent vaccines such as DLprG, including facilitating mechanistic studies evaluating immune molecules and subsets in mediating defined aspects of protection.
Mtb transmission typically occurs after repeated exposures, for example in the setting of infected household contacts. 22,23 Indeed, preclinical nonhuman primate (NHP) studies are increasingly performed with repeated low-dose Mtb challenge. 26 Similarly, repeated low-dose exposure is commonly used in the HIV field to test prophylactic interventions. 29,30 In both of these models, repeated challenge creates unique opportunities for measuring correlates of protection. 26,29,30 We therefore developed a repeated low-dose Mtb challenge model that resulted in increased challenge stringency and reduced vaccine efficacy compared with the single low-dose Mtb challenge model. Correlates of protection analysis suggested that serum IL-17A was a key correlate of vaccine protection as previously described 5,24,26 and identified additional biomarkers of protection including CCL2 and CXCL1 that to our knowledge have not previously been described in this context. The reduced capacity of DLprG to mediate prevention of detectable infection in the repeated challenge model may reflect a narrow protective window related to the limitations of murine immune responses to Mtb. 31 Several aspects of vaccine protection mediated by DLprG in the low-dose model will be important to evaluate in future studies. We did not assess mice after four weeks, and data from later time points and survival studies could provide measures of longer term vaccine protection. Indeed, BCG provides long-term bacterial control in the low-dose model, in contrast to the short-lived protection observed after 100 CFU challenge. 20 Use of challenge strain barcoding 18,20 and analysis of spleen and other sites of dissemination like draining lymph nodes could provide a more granular understanding of vaccine protection. In addition, future studies could assess the contributions of myeloid and T cell subsets as well as the cytokines identified in our correlates analysis for their potential roles as regulators of DLprG vaccine protection in the lowdose model. Finally, intravenous administration of DLprG could be explored in the low-dose model, as recently reported for BCG in the 100 CFU mouse 32 and NHP 33 models.
In summary, we demonstrate that the attenuated TB vaccine candidate DLprG provides strikingly greater protection than BCG against low-dose challenge in mice, including prevention of detectable infection in the majority of animals and anatomic containment for all breakthrough infections. We further developed a repeated low-dose challenge model that was more stringent and identified biomarkers of vaccine protection. These data support a compelling framework for the preclinical evaluation of next-generation TB vaccine candidates that are more protective than BCG as well as studies dissecting the immunologic mechanisms that modulate Mtb infection and dissemination in addition to control of bacterial burden.

Limitations of the study
Our study of the protective efficacy of DLprG in the low-dose murine aerosol challenge has several limitations. First, our data are limited to the C3HeB/FeJ mouse and H37Rv challenge strains, and it will be important to show generalizability of our findings to additional mouse strains and heterologous challenge strains. Moreover, our necropsies were performed at four weeks and we did not perform longer term protection studies. In addition, we did not examine the status of dissemination to spleen. Finally, we did not perform immunophenotyping or mechanistic studies examining the roles of immune cells and pathways in mediating protection in the low-dose model.  iScience Article manufacturer's instruction. The assay plates were read by MESO QUICKPLEX SQ 120 instrument and data were analyzed by Discovery Workbench 4.0 software.

Low dose challenge stock preparation and in vivo titration studies
H37Rv challenge strain was obtained and propagated as detailed above. Low dose challenge stock was generated as previously described. 18 Briefly, H37Rv was grown to an OD of 0.7-0.8. Cultures were passed through a 5 mm filter in order to generate a single cell suspension prior to aliquoting, freezing at À80 C, and tittering. For in vivo titration, challenges were performed as detailed above. We first estimated a 100-fold reduction in challenge dose relative to 50-100 CFU challenge and challenged mice with three log 10 dilutions centered around the predicted low dose challenge dose. This identified a challenge dose with a 44% infection rate and was followed by a secondary dose titration study with three log 2 dilutions centered around the predicted low dose challenge dose.

Lung processing and CFU quantification
Mice were euthanized 4 weeks following single 50-100 CFU challenge, 4 weeks following single low dose challenge, or 4 weeks following the final weekly low dose challenge in the repeated low dose challenge model. For 50-100 CFU challenge, both lung lobes were dissected en bloc whereas for low dose challenge right and left lung lobes were dissected separately. Tissues were placed into gentleMACS M Tubes (Miltenyi Biotec 130-096-335) containing 5 mL of PBS and mechanically dissociated using a gentleMACS Dissociator (Miltenyi Biotec). Lysates were then plated in serial log 10 dilutions onto 100 3 15mm Middlebrook 7H10 plates (Hardy Diagnosatics). In order to achieve an LOD of 5 CFU in low dose challenge studies, we additionally plated 1 mL of lysate onto to two 150 3 15mm plates containing Middlebrook 7H10 agar (BD Difco), 10% Middlebrook OADC (BD BBL), 0.5% glycerol (Sigma Aldrich), and cycloheximide (Sigma Aldrich) at 100 mg/mL. CFU were counted after a 3 weeks incubation at 37 C.

Histopathological studies
Lungs were insufflated with 10% neutral buffer formalin for 48 h then transferred to 70% ethanol and processed routinely into paraffin blocks for hematoxylin and eosin or Ziehl-Neelsen acid-fast staining. Whole slide scanning (20x) was performed using a Midi II Pannoramic scanner (Epredia) and images evaluated by a boarded veterinary pathologist (AJM) using HALO (Indicalabs). Immune subsets including lymphocytes and macrophages were identified by hematoxylin and eosin staining by a board-certified veterinary pathologist (AJM) based on morphology.

QUANTIFICATION AND STATISTICAL ANALYSIS
Pairwise tests on immunogenicity and 50-100 CFU challenge data were performed using GraphPad Prism 9.4.0 software. Heatmaps were generated using the R package pheatmap. Cytokines plasma levels were normalized using the Z score method implemented in the pheatmap package. The correlation of cytokines with lung CFU was performed using the R package corrplot and Spearman's method. Statistical evaluation was assessed using a t-test distribution implemented in the R cor.test function. Rates of animal level infection (any lobe vs. none) were compared using exact logistic regression models. The composite infection outcomes (none vs. one lobe vs. both lobes) were analyzed using ordinal logistic regression models under the assumption of proportional odds. Between-group differences in bacterial loads were analyzed using mixed effects negative binomial regression models while controlling for lobe side/size (right vs. left). Two lobes of the same animal were considered a cluster, and animals were considered independent from each other.