C57BL/6J (B6, #000664) and C3HeB/FeJ (C3H, #000658) mice were purchased from Jackson Laboratories (Bar Harbor, ME). Balb/c x 129/SvJ (C129F1) mice were a custom breeding order from Jackson Laboratories (Bar Harbor, ME). Female mice between the ages of 9–12 weeks were used. All animals were housed and maintained in specific-pathogen-free conditions at Seattle Children’s Research Institute (SCRI). All animal studies were performed in compliance with the SCRI Animal Care and Use Committee.

BCG-Pasteur was cultured in Middlebrook 7H9 with OADC supplement and 0.05% Tween-80 at 37°C with constant agitation for five days. BCG was back diluted in 7H9 for two days and grown to an OD of 0.2–0.5. Bacteria was diluted in PBS and mice were injected subcutaneously (SC) or intravenously (IV) with 200µl of 10⁶ CFU. After immunization, mice were rested for 8 weeks prior to Mtb infection.

The CoMtb model was established as described previously (7, 36). Briefly, mice were first anesthetized by intraperitoneal injection of ketamine and xylazine diluted in PBS. Mice were placed in a lateral recumbent position, and the ear pinna was flattened with forceps and pinned onto an elevated dissection board using a 22 G needle. H37Rv Mtb grown to an OD between 0.2–0.5 over a 48-hour period was diluted to 10⁶ CFU/ml in PBS, and 10 ul (10⁴ CFU) was administered into the dermis of the ear using a 26s G Hamilton syringe. Mice were then rested for 8 weeks prior to Mtb infection.

Infections were done with stocks of H37Rv, SA161 or CDC1551 as previously described (37). To perform aerosol Mtb infections, mice were placed in a Glas-Col aerosol infection chamber, and 50–100 CFU were deposited into their lungs. To confirm the infectious inoculum, two mice per infection were euthanized on the same day of infection, and their lungs were homogenized and plated onto 7H10 or 7H11 plates for determination of CFU.

Right lungs were homogenized in M tubes (Miltenyi) containing 1mL PBS+0.05% Tween-80 (PBS-T) using a GentleMACS machine (Miltenyi). Homogenates were then diluted in PBS-T and plated onto 7H10 plates. Plates were incubated at 37°C for at least 21 days before quantification of CFU.

Left lungs were fixed in 10% formalin for 24 hours, then dehydrated in 70% ethanol at 4°C for at least 24 hours. Samples were paraffin embedded and sectioned at the University of Washington Histology Core. Subsequently, slides were reviewed by a veterinary pathologist and scored in a blinded fashion based on the following metrics (see Supplementary Table 2 ): mixed granulomas (ill-formed granulomas with mixture of macrophages and lymphocytes), defined granulomas [well defined with increased separation of macrophages, epithelioid or multinucleated giant cells (MNGC) with lymphoid aggregates], perivascular lymphoid aggregates (PV LA), peribronchiolar lymphoid aggregates (PB LA), histiocytes, foamy macrophages, MNGC, alveolar hyperplasia, neutrophils, necrosis, cholesterol clefts, edema, extent 1 (percent involvement of the lung), extent 2 (percent involvement of the lung in the worst manner).

As an initial, quantitative measure of the degree and durability of protection afforded by each vaccine modality across the strain-strain combinations, we measured lung CFU burden at Day 28 and Day 98 following infection ( Figure 1 ). Among unvaccinated mice at Day 28, we observed a wide range of outcomes, with bacterial loads differing by over 100-fold. For example, infection with H37Rv resulted in the highest CFU burden among C129F1 mice, ( Figure 1A ), while infection with SA161 yielded the highest CFU burden among C3H mice ( Figure 1B ); however, these two strains of mice responded with similar CFU burdens after infection with CDC1551 ( Figure 1C ). Across Mtb strains, B6 mice developed the lowest CFU burdens among the host genotypes at Day 28 ( Figures 1A–C ), while C3H mice infected with SA161 developed the highest overall lung burden across both time points ( Figures 1B, E ). There was much less heterogeneity in burdens among unvaccinated mice at Day 98, with equivalent loads across all genotypes in response to H37Rv ( Figure 1D ) and SA161 ( Figure 1E ). Infection with CDC1551 resulted in significantly higher lung bacterial loads in C3H mice than in the other mouse strains at the late ( Figure 1F ) but not early ( Figure 1C ) timepoint, suggesting that, even at late time points, we can identify differential responses to Mtb infection across combinations.

As expected from prior reports of vaccine-mediated immunity in the B6 mouse model (6–8, 38), lung bacterial burdens in B6 mice were significantly reduced at Day 28 by all vaccine modalities following challenge with all bacterial strains ( Figures 1A–C ), with the exception of IV BCG with SA161 challenge, where a high variability in CFU burden was seen ( Figure 1B ). The C129F1 mice behaved similarly to B6 mice based on CFU reductions following all vaccine modalities in response to all challenge strains. Unvaccinated C129F1 mice had particularly high bacterial loads, but these burdens were reduced to similar levels as B6 mice following vaccination ( Figures 1A–C ). In contrast to both B6 and C129F1 mice, C3H mice displayed a wide range of vaccine-mediated responses at Day 28 ( Figures 1A–C ). Unexpectedly, these mice were not protected from CDC1551 challenge by any vaccine modality at this time point ( Figure 1C ), whereas all modalities resulted in approximately 0.6-log CFU reduction upon H37Rv challenge ( Figure 1A ) and up to 2-log CFU reduction upon SA161 challenge ( Figure 1B ) ( Supplementary Table 1 ). Overall, at Day 28, 22/27 (81.4%) strain-strain combinations resulted in significant decreases in lung CFU. At Day 98, however, CFU reduction in response to vaccination was only seen in 11/27 (40.7%) combinations. Some strain-strain combinations failed to display any vaccine-mediated reduction in bacterial load at Day 98, such as B6 and C3H mice infected with H37Rv ( Figure 1D ) and C3H mice infected with SA161 ( Figure 1E ). Among strain combinations that were protected at Day 98, there was heterogeneity associated with vaccine modality. For example, IV BCG mediated substantial protection in C129F1 mice infected with H37Rv ( Figure 1D ) and CDC1551 ( Figure 1F ), whereas SC BCG and CoMtb did not ( Figures 1D, F ).

We next plotted lung CFU in vaccinated vs unvaccinated mice within each strain-strain combination at each time point to visualize trends in the long-term durability of vaccine-mediated protection ( Figure 1G ). Among all strain-strain combinations, the magnitude of protection generally decreased over time [6/9 (B6-H37Rv, C129F1-H37Rv, C3H-H37Rv, C129F1-SA161, C3H-SA161, C129F1-CDC1551), 67%]; however, in some cases it stayed constant [2/9 (B6-SA161, B6-CDC1551), 22%] or even increased [1/9 (C3H-CDC1551), 11%]. While we noted heterogeneity in vaccine-mediated outcomes across strain-strain combinations and time points, as described in Figures 1A–F , there was a striking overlap in the behavior of the three protective modalities within most combinations. Thus, while bacterial and host strain diversity had a large overall impact on vaccine-mediated outcomes, the vaccines themselves contributed relatively little to the observed heterogeneity. For example, we observed similar initial log reductions in CFU followed by waning protection over time, as visualized by a positive slope, for B6 mice infected with H37Rv for all 3 modalities. Surprisingly, in the C3H/SA161 group, which was the most strongly protected combination at Day 28, none of the vaccines were able to mediate protective efficacy at the later time point. In contrast, durable protection across vaccine modalities, as visualized by a slope of ~0, was only seen in the B6/SA161 and B6/CDC1551 combinations. We found this was also true for almost all vaccines in C129F1 and C3H mice challenged with H37Rv ( Figure 1G ). The only example of improved protection over time, as indicated by a negative slope, was C3H mice challenged with CDC1551, although this combination failed to show evidence of vaccine efficacy at the early time point.

As previously noted, unvaccinated C3H mice infected with SA161 developed the highest bacterial burdens of any strain-strain combination at Day 28, peaking at loads approximately 1-log higher than in either B6 or C129F1 mice ( Figure 1B ). Furthermore, this combination demonstrated the strongest vaccine-mediated reduction in lung CFU at Day 28 of ~2 logs ( Supplementary Table 1 ). To address whether early bacterial load in unvaccinated mice correlated with vaccine-mediated CFU reduction across strain-strain combinations, we plotted the log fold-change of CFU upon vaccination against the lung CFU of unvaccinated mice ( Figure 2A ). This analysis confirmed a correlation (r=-0.74) between the initial bacterial load at Day 28 in unvaccinated mice and vaccine-induced reduction in bacterial load. This correlation was not observed at the late, Day 98 time point ( Figure 2B ), suggesting that the protective mechanisms induced by vaccination may be more effective at curbing rapid bacterial growth early during infection than in reducing bacterial burden in an established infection.

We next ranked the protective efficacy of each vaccine modality in each strain-strain combination in order of its lung CFU burden in vaccinated vs unvaccinated mice at Day 28 ( Figure 3A ). These data underscore the broad range of magnitudes of vaccine-mediated protection at early time points across strain-strain combinations, with bacterial load reductions ranging from 0 to >2 logs, highlighting the critical role of host and pathogen diversity in modulating vaccine outcomes ( Figure 3A ). In contrast, the combinations at Day 98 were largely restricted to between 0 and <1 log of protection, highlighting both the limited range of responses and overall waning of protection at later time points ( Figure 3B ). Importantly, the disordered nature of vaccine efficacy at Day 98 ( Figure 3B ) when ranked by vaccine efficacy at Day 28 ( Figure 3A ) illustrates the inability of early protective responses to predict late vaccine-mediated protection. This heterogeneity in vaccine outcomes that emerges upon incorporating host and genetic diversity underscores the limitations of traditional animal model vaccine testing that solely relies on the standard B6/H37Rv strain combination.

At the Day 98 time point, a time at which lesions have fully developed, the left lung was preserved in formalin for hematoxylin and eosin (H&E) staining and histopathologic analysis. The following features were assessed: extent, mixed granulomas, defined granulomas, perivascular lymphoid aggregates, peribronchial lymphoid aggregates, histiocytes, multinucleated giant cells (MNGC), neutrophils, alveolar hyperplasia, necrosis, cholesterol clefts, and edema ( Supplementary Table 2 ). Representative H&E images from unvaccinated mice that embody clear histological patterns are shown in Figure 4 . B6 mice infected with H37Rv had moderate involvement with disorganized lesions ( Figure 4A ), C3H mice infected with CDC1551 had extensive involvement with MNGC ( Figure 4B ), and C129F1 mice infected with H37Rv had a high degree of lung involvement driven by lymphoid aggregates ( Figure 4C ). Consistent with reports that C3H mice develop caseating granulomas more typical of primary TB granulomas seen in humans (19, 21, 39), C3H mice infected with SA161 displayed distinctive granuloma structures defined by overwhelming necrosis and neutrophil infiltration ( Figure 4D ). This strain combination reached the highest bacterial burdens at the Day 28 time point, nearly 1 log higher than all other combinations ( Figure 1B ), suggesting that early bacterial burdens can drive pathology at later time points. To gain insight into the pathological features driving lesion structure in each genotype, we performed a principal component analysis (PCA) using all scored pathology variables for all mice ( Supplementary Table 2 ), first showing unvaccinated mice for simplicity ( Figure 4E ). By examining the PCA loading scores, we found that individual pathological features tended to group into 4 distinct categories: Defined, Necrosis, Extent, and Lymphoid aggregates ( Figure 4F ). For unvaccinated mice, we found that variation in PCA scores was largely driven by mouse strain, and to a lesser extent by bacterial strain ( Figure 4E ). Overall, C3H mice clustered separately from B6 and C129F1 mice, typified by a high extent of involvement and the presence of neutrophils ( Figures 4B, E, F ). Conversely, C129F1 mice contained lymphocyte-rich lesions with prominent peribronchial and perivascular aggregates ( Figures 4C, E, F ). Lastly, B6 mice largely fell near the center of the PCA plot, consistent with the lack of prominent scored features and the presence of disorganized lesions, with histiocytes intermixed with loose lymphoid regions ( Figures 4A, E, F ).

To determine how the different vaccine modalities impacted pathology outcomes, we included data points from vaccinated mice onto the PCA plot, and separated them for each mouse strain to facilitate visualization ( Figure 5A ). For B6 mice, there was large overlap in pathology outcomes regardless of bacterial strain or vaccination regimen. In C3H mice, CDC1551 infection led to pathology scores that clustered in a region of the plot associated with higher lesion extent, regardless of vaccine strategy, while both H37Rv and SA161 infection resulted in a wide range in pathology scores depending on vaccination modality. In C129F1 mice, Mtb strain, but not vaccination modality, was the most significant driver of outcome along these principal components ( Figure 5A ). To gain insight into the vaccine-mediated changes in pathology in individual strain-strain combinations, we examined the relevant subset of data points from the PCA plot alongside representative H&E images from Day 98 ( Figures 5B–E ; Supplementary Figure 1 ; Supplementary Table 2 ). Several of the most striking combinations are shown in Figure 5 . For the B6/H37Rv combination, all vaccine modalities resulted in modest yet clear changes in pathology scores, largely driven by decreased extent of lung involvement and lesion definition, with the arrow indicating a shift from greater to lesser involvement with vaccination ( Figure 5B ). For B6 mice infected with CDC1551, histologic features were similar regardless of vaccination and were marked by poorly defined lesions and abundant myeloid aggregates ( Figure 5C ). In the C3H/H37Rv combination, there was an unexpected increase in involvement and degree of defined granulomas in vaccinated vs unvaccinated mice ( Figure 5D ). This aligns with the complete lack of protection from bacterial burden in this combination at Day 98, and even a trend toward increased CFU upon vaccination ( Figure 1D ). Finally, in the C3H/SA161 combination, there was dramatic overall improvement in pathology resulting from all vaccine modalities; the large necrotic granulomas observed in the unvaccinated group were absent in vaccinated mice ( Figure 5E ). CoMtb-inoculated C3H mice infected with SA161 had the largest reversal in degree of lesion extent, scoring lower than either BCG-immunized cohort ( Figure 5E ). The remaining strain-strain combinations and their respective PCA plots and H&E images are shown in Supplementary Figure 1 . Together, this shows that host and bacterial genetics play a significant role in determining the effect of vaccines on pathology, and that while vaccines appeared to improve pathology in most cases, they had no effect and potentially exacerbated pathology in certain strain-strain combinations.

Given the diverse histologic outcomes noted across strain-strain combinations, we sought to determine whether any of these patterns directly associated with lung bacterial burden outcomes. To address this, we overlaid a heatmap of lung CFU values at Day 98 for each mouse genotype onto their respective subset of the histopathology score PCA plot ( Figure 6 ). In B6 mice, there was an overall lack of correlation between bacterial burden and pathology features. In contrast, while high bacterial burdens in unvaccinated C3H/SA161 mice associated with necrosis, vaccination in this strain-strain combination led to both a decrease in CFU and a shift away from necrosis. No correlation was observed between histopathology PCA scores and either CFU or vaccination status for C3H/H37Rv mice, which otherwise overlapped with C3H/SA161 mice on the PCA plot. C3H/CDC1551 infection resulted in histopathology marked by high extent regardless of vaccination status. In C129F1 mice, infection with H37Rv was associated with both higher CFU and histopathology dominated by lymphoid aggregates, regardless of vaccination status or modality.

To more clearly identify the histopathologic features that correlate with CFU outcome among mouse strains across Mtb genotypes and vaccine modalities, we performed Spearman correlation analysis of each scored feature with either the mean CFU value for all mice with that genotype at Day 28 (because matching H&E images were not collected at Day 28) ( Figure 7A ) or the paired CFU at Day 98 (where individual CFU values had a matching histopathology score) ( Figure 7B ). Consistent with the CFU heatmap overlay in Figure 6 , B6 histopathology scores did not correlate with CFU at either time point, suggesting that the histological presentation of these mice is a poor indicator of bacterial control. Interestingly, in C3H mice, Day 28 CFU significantly correlated with several histological features, including necrosis, neutrophils, defined granulomas, and extent; however, these features did not correlate with Day 98 CFU burden. In C129F1 mice, on the other hand, most histopathology features correlated positively with CFU burden at both early and late time points, consistent with the observation that these mice retained a higher level of vaccine-induced protection at Day 98 ( Figures 1D–F ). Some correlative features were shared between C3H and C129F1 mice at Day 28, including mixed granulomas, MNGC, histiocytes, and extent 2, but certain features were uniquely correlated with bacterial load in each genotype. For example, burden in C3H mice uniquely correlated with neutrophils and necrosis, whereas burden in C129F1 uniquely correlated with both perivascular and peribronchial lymphocyte aggregates. Taken together, these data suggest that host genotypes display distinct pathological signatures, and high CFU burden can be associated with a range of unique pathologic indicators.