Male and female C57BL/6 mice, 4–6 weeks of age, were purchased from The Jackson Laboratory. Mice were housed in a biosafety level 3 animal facility at Colorado State University (CSU), and all experimental protocols and procedures were performed in accordance with the CSU Institutional Animal Care and Use Committee under protocol number 1278. Mice were euthanized by CO2 inhalation at 56 or 114 days after exposure to Mtb.

Mice were mock-vaccinated with 50 μl of phosphate buffered saline (PBS) or vaccinated with the human clinical-grade Danish-1331 BCG vaccine, Lot #120005-C (AJ Vaccines), by intradermal injection at the tail base. Assuming a stock bacterial concentration of 2 x 10⁶ BCG organism per ml, as provided by the manufacturer, a targeted dose of 1 x 10⁴ CFU was administered in a 50 μl injection volume to each mouse.

150 days post-vaccination with BCG, mice were infected by low-dose aerosol with Mtb Erdman (BEI Resources) using a Glas-Col aerosol chamber targeting an exposure dose of 50–100 CFU, as previously described (32). Immediately post-exposure, three mice were euthanized and whole lung homogenate was plated on 7H11 agar plates to confirm aerosol delivery of expected exposure dose, achieving a mean exposure dose of 54 CFU per mouse.

At necropsy, the right caudal lung lobe of each mouse was collected and homogenized in PBS using a Bullet Blender Blue (Next Advance) device at speed 8 setting for 4 minutes, as previously described (33). Tissue homogenate was plated at serial 1:5 dilutions on 7H11 agar plates and final CFU counts were enumerated following 8 weeks of incubation.

One third of the left lung lobe was collected at necropsy and immediately fixed in 4% paraformaldehyde for 48 hours before being paraffin embedded. 5μm tissue sections were cut and mounted on charged microscope slides, stained with hematoxylin and eosin (H&E), and scanned at 20X magnification using an Olympus VS120 scanning microscope, Hamamatsu ORCA-R2 camera, and Olympus VS-ASW 2.9 software available through the Experimental Pathology Facility at Colorado State University.

Visiopharm software was used for image analysis. Entire lung section regions of interest (ROI) were identified at 5X magnification with a custom algorithm that utilizes decision forest training to differentiate tissue from microscope slide based on color and area. An additional custom-made algorithm based on decision forest training and classification based on staining intensity, area, and morphological features at 20X magnification was created to automate granuloma detection. A final algorithm was used for quantification of area of both granulomas and healthy tissue within tissue section ROIs to calculate percent granuloma lesion burden. Identified granulomas and calculations received a final review by a pathologist (Podell BK) (34).

Slides were deparaffinized and stained using a Leica Bond RXm automated slide stainer and associated Leica Bond reagents ( Supplementary Table 1 ). Heat induced epitope retrieval (HIER) was performed on tissues with ER1 solution at 97°C for 30 minutes prior to quenching endogenous peroxidase activity with 3% hydrogen peroxide. Tissues were blocked using 2.5% Normal Goat Serum Blocking Solution (Vector Laboratories) then incubated with the following rat monoclonal primary antibodies: CD4 (1:100, clone 4SM95, catalog number 14–9766-82, eBioscience), CD8a (1:100, clone 4SM15, catalog number 14–0808-82, eBioscience), and CD45R/B220 (1:100, clone RA3–6B2, catalog number 103201 BD Biosciences). Sections were incubated with undiluted ImmPRESS mouse adsorbed goat anti-rat IgG-HRP polymer (Vector Laboratories) for primary antibody detection. Visualization of cell surface markers was accomplished using Opal 520 (1:200), Opal 570 (1:200), and TSA plus Cyanine 5 (1:1000) in 1X TSA Plus Automation Amplification Diluent (Akoya Biosciences). Antibody stripping was performed using HIER at 95°C and ER1 solution for 20 minutes (Leica Biosystems) between each antibody application in the following order: CD4 detected with Opal 520, CD8 detected with Opal 570, and B220 detected with Cyanine 5. Finally, nuclei were stained with Spectral DAPI (Akoya Biosciences). Whole stained lung sections were scanned at 20X magnification using an Olympus VS120 microscope, Hamamatsu ORCA-R2 camera, and Olympus VS-ASW 2.9 software at the Experimental Pathology Facility at Colorado State University. Exposure times were determined using the autoexposure function and adjusted to distinguish antibody detection of cell surface markers from erythrocyte background.

Slides cut in serial to those used for multiplex IHC were deparaffinized and targets hybridized using the RNAscope LS multiplex fluorescent reagents (Advanced Cell Diagnostics) according to the manufacturer’s recommendations for use on the Leica Bond RXm automated slide stainer and respective Leica Bond reagents ( Supplementary Table 1 ). HIER was performed on tissues with ER2 solution (Leica Biosystems) at 97°C for 30 minutes prior to quenching endogenous peroxidase activity with 3% hydrogen peroxide. Mtb was identified using RNAscope 2.5 LS Probes B-MTB-23SrRNA-1-C1 (Advanced Cell Diagnostics #471658) and B-MTB-pre-rRNA-O1-C2 (Advanced Cell Diagnostics # 507548-C2) diluted 1:50 in the C1 probe (31, 35). ISH probes were visualized using TSA plus tetramethylrhodamine (Akoya Biosciences) at a dilution of 1:750, TSA plus fluorescein (Akoya Biosciences) at a dilution of 1:1000, and TSA plus cyanine 5 (Akoya Biosciences) at a dilution of 1:750 in 1X TSA Plus Automation Amplification Diluent (Akoya Biosciences). Nuclei were counterstained with DAPI (Advanced Cell Diagnostics). Whole stained lung sections were scanned at 40X magnification using an Olympus VS120 microscope, Hamamatsu ORCA-R2 camera, and Olympus VS-ASW 2.9 software at the Experimental Pathology Facility at CSU. Exposure times were determined using the autoexposure function and adjusted to distinguish specific probe hybridization signal from erythrocyte background.

Whole slide images were imported into Visiopharm image analysis software. Granulomas were identified using a decision forest algorithm trained on intensity and density of all fluorophores. Granulomas identified were confirmed or corrected based on annotations acquired from detection on H&E-stained slides. Each granuloma was then automatically assigned a unique ROI number within each lobe using the enumeration function in Visiopharm. Next, using a threshold-based algorithm, nuclei were detected and annotated based on DAPI signal intensity and given margins to allow for membrane analysis. For IHC, membranes were detected based on a threshold intensity of each fluorophore and assigned to their corresponding marker. For ISH, bacilli were detected based on a threshold intensity of the fluorophore assigned to the 23S rRNA probe, and further designated as replicating or non-replicating based on intensity of the corresponding fluorophore assigned to the pre-rRNA probe. In both IHC and ISH, red blood cell autofluorescence was accounted for based on low DAPI signal intensity combined with inherent combined signals of both FITC and tetramethylrhodamine. Thresholds for all markers were adjusted for any differences in staining or imaging quality and then reviewed for accuracy. Following review, counts for each cell type were quantified using an additional Visiopharm algorithm and exported as .csv files for downstream data analysis.

Statistical data analysis was performed in R (version 4.3.1). Data, code, and R packages from this study can be accessed through GitHub (https://github.com/PodellLab/Granuloma_RSRatio_ISH). Data exported from Visiopharm were pre-processed to prepare for downstream analysis in R. Preliminary exploratory statistical and graphical testing were performed to evaluate data distribution and differences between treatment groups. An unpaired t-test was used as a preliminary approach to calculate differences in granuloma burden, CFUs, and ISH RS ratios across treatment groups and infection endpoints.

Next, we fit a regression model to investigate the degree to which the immune cell populations in a granuloma are associated with the number of replicating bacteria in that granuloma, adjusting for the total number of bacteria in the granuloma. To do this, we fit a regression model with the number of replicating bacteria in a granuloma as the outcome variable and with three immune cell populations as potential explanatory variables, while controlling for the timepoint and vaccination status and offsetting by the total number of bacteria. To make model coefficients more interpretable, we scaled immune cell counts by dividing by 500, based on the approximate interquartile range of each of these cell populations across the full study data. Therefore, the estimated coefficients represent an expected change per approximately one interquartile-range change in an immune cell population. Specifically, we fit the following quasi-Poisson regression model, adjusting for potential overdispersion of observed count data from each granuloma and including as an offset the log transformation of total bacterial counts measured by ISH:

Where, for each granuloma i :

Lung lobes were collected at necropsy at days 56 and 114 post-infection to assess Mtb burden at the level of the whole lung tissue. Colony forming units (CFU) of culturable bacteria in lung homogenate increased between day 56 and day 114 post-infection in both unvaccinated and BCG-vaccinated mice. However, BCG-vaccinated mice displayed reduced CFU burden at both days 56 and 114 post-infection compared to unvaccinated mice ( Figure 1A ). To establish whether the BCG vaccination offers protection from progressive pulmonary pathology at days 56 and 114 post-infection, the granuloma lesion burden was quantified histologically across the whole lung lobe using image analysis software ( Figure 1B ). Consistent with CFU at day 56, a significant decrease in granuloma lesion burden was observed in BCG-vaccinated mice compared to unvaccinated mice at day 56. Between day 56 and day 114 post-infection, the overall granuloma lesion burden increased in both groups. A decreased granuloma lesion burden was observed in BCG-vaccinated mice at day 114 post-infection although not statistically significant ( Figure 1C ).

Regardless of differences across pulmonary pathology burden, heterogeneity could be observed in morphological structure and cell composition among individual granulomas within an individual mouse, within vaccination status, and within infection endpoints ( Figures 1D-I ). H&E-stained lungs demonstrated granulomas with morphology of multiple and variable subtypes as observed by the reviewing pathologist (Podell BK). Granulomas were observed that consist of large centralized lymphoid aggregates resembling de novo tertiary lymphoid structures, some of which contain follicular centers composed of cells with macrophage-type morphology. Lymphoid structures are surrounded by either dense and eosinophilic epithelioid macrophages that replace pulmonary architecture ( Figures 1D, E ), or alveolitis consisting of loose aggregates of heavily vacuolated macrophages contained within remnant alveolar spaces ( Figures 1H, I ). Other granulomas consisted of multiple smaller and less organized lymphoid aggregates surrounded by heavily vacuolated macrophages or have a paucity of lymphocyte cells and consist almost entirely of vacuolated macrophages. Among a proportion of the granulomas with poorly organized lymphoid structures, the heavily vacuolated macrophages were surrounded by alveolar interstitial fibrosis embedding small foci of necrosis with a high frequency of neutrophils ( Figures 1F, G ). While measures of granuloma lesion burden offer an indication of immune protection across the whole lung, this fails to encompass the heterogeneity of histological morphology observed across granulomas. This degree of histological heterogeneity suggests that distinct and diverse immune responses exist within and between mice, irrespective of vaccination, and that granuloma targeted analysis could offer greater insight into correlates of protection that cannot be observed through bulk tissue analyses.

Divergent features of granuloma histomorphology suggest a more granular measure of bacterial replication may distinguish differences between granulomas. To better understand each granuloma’s ability to control bacterial replication, multiplexed fluorescent in situ hybridization (ISH), using targets analogous to the RS ratio dPCR (31, 35), was employed to visualize individual replicating and nonreplicating Mtb organisms, referred to as RS ratio ISH. In this approach, a replicating bacillus was defined by colocalized detection of both ETS1/ITS1 pre-rRNA sequence and mature 23S rRNA sequence, while a non-replicating Mtb bacillus was defined by detection of 23S rRNA only. 191 granulomas identified on H&E-stained slides were evaluated ( Table 1 ). The number of granulomas per mouse tended to be higher in unvaccinated mice at day 56 post-infection compared to BCG-vaccinated mice, but increased in number among BCG-vaccinated mice between day 56 and day 114 post-infection. 23S rRNA and ETS1/ITS1 pre-rRNA detection was quantified using Visiopharm image analysis software and the proportion of replicating bacteria relative to total bacteria calculated ( Figures 2A–D ). At both days 56 and day 114 post-infection, granulomas from unvaccinated mice displayed higher proportions of replicating Mtb compared to BCG-vaccinated animals ( Figure 2E ). Collectively, the proportions of replicating Mtb decreased across granulomas between day 56 and day 114, irrespective of vaccination status. Regardless of timepoint or vaccination status, a wide range of total bacterial burden and replication status was observed between and within groups and within individual animals.

Considering the heterogeneity of replicating proportions of Mtb observed across the granuloma, a direct comparison was made to the whole lung CFU per mouse that, as a stand-alone metric, supports vaccine-mediated protection. In comparing these metrics, it is evident that regardless of CFU value, granulomas per mouse vary in number. Within and between mice, these granulomas display an expansive range of total Mtb burden and proportion of replicating Mtb by ISH. There was no evident directional association between replicating Mtb by RS ratio ISH and culturable Mtb burden by CFU ( Figure 2F ). Rather, CFUs appear to more linked to the total number of granulomas in the mouse. These data suggest that total CFU burden reflects a collective contribution of diverse granuloma responses with differing capacities to control Mtb replication. Overall, the results from granuloma-targeted RS ratio ISH highlights that a wide range of capacity for immune control of Mtb replication exists across granulomas, which cannot be effectively captured by CFU. Thus, RS ratio ISH may distinguish granulomas with immune cell composition that provide greater control of Mtb replication.

We investigated underlying spontaneous and BCG-induced adaptive immunity by multiplexed IHC in conjunction with the granuloma heterogeneity observed both histologically and in Mtb replication rate by RS ratio ISH. Essential adaptive immune cell phenotypes were considered because previous work shows a reduction in total lung RS ratio at the onset of adaptive immunity (31). Accordingly, CD4 cells, CD8 cells, B cells, and other host cells were targeted within granulomas in which the RS ratio ISH was quantified ( Table 1 ). Within each granuloma designated as an independent ROI, each cell subset was identified and quantified using Visiopharm image analysis software ( Figures 3A-D ). Granulomas displayed heterogeneous cellular composition and organization across all groups and endpoints evaluated, where total cell counts of any measured phenotype are unable to distinguish vaccine status or day of infection ( Figure 3E ). All measured immune cells increase in median number per granuloma between day 56 and day 114, and this increase remained similar across B cells, CD4, and CD8 cells, but these changes between infection endpoints are small in comparison to the large variability observed within each group. The most prevalent cell type at either day 56 or day 114 are those that do not belong to any of the three evaluated phenotypes and are designated by the term, “other cells”. Overall, a wide spread of cell counts was observed independent of timepoint or vaccination status, exemplifying differences across granulomas that may provide context for ability to control bacterial replication.

Having established granuloma heterogeneity in the C57BL/6 mouse model using simultaneous measures of host and bacterial response, we then performed regression analysis to identify any association with immune environments capable of controlling Mtb replication. To do so, a generalized linear regression model was utilized to determine associations between immune cell composition and counts of replicating Mtb, while accounting for total Mtb burden through a model offset. Based on differences between interquartile ranges (IQR) among immune cells, counts of immune cells were scaled down by a factor of 500 to approximate the IQR, which facilitates interpretation of respective comparisons. Controlling for all variables, irrespective of vaccine status or timepoint, an inverse relationship between CD4 and CD8 cells was observed with changes in the relative rate of bacterial replication. A 17.7% decrease in estimated relative rate of Mtb replication was associated with each 500-cell increase in CD4, and a 25.8% increase in relative rate of Mtb replication was associated with each 500-cell increase in CD8, while B cell count showed no significant association with Mtb replication ( Figure 4 ). Controlling for differences in these host cell populations, BCG vaccination was associated with a 35.4% decrease in estimated relative rate of replication compared to unvaccinated, while progression to day 114 post-infection was associated with a 54.7% decrease over time compared to day 56 post-infection. Here we identify a significant association between granuloma immune cell composition and control of Mtb replication. Data obtained from all evaluated granulomas, including immune cell counts and replication rates are available at https://github.com/PodellLab/Granuloma_RSratio_ISH.

Based on predictions using this model, we back-translated associations between cell composition and RS ratio ISH replication rates to evaluate granuloma morphology associated with immune control of bacterial replication. Selection of two granulomas from the lung lobe of a single BCG-vaccinated mouse at day 56 post-infection highlights the observation that different degrees of bacterial replication existing between granulomas is associated with immune cell composition unique to each granuloma ( Figure 5 ). Between these two granulomas within this single mouse, CD8 cells differ by 547 cells and Mtb replication is increased by 28.9% ( Table 2 ). Consistent with the multiple morphologies described in Figure 1 , these granulomas contain either higher proportions of vacuolated macrophages or epithelioid macrophages and B cell-rich de novo follicular structures. In these granulomas, the higher Mtb replication rate is localized to macrophages with vacuolated morphology, which is overrepresented in the granuloma with the higher CD8 count. These findings indicate that cell populations with a designated morphology may align with poor control of Mtb replication. Furthermore, granuloma morphology may indicate an immune environment with capacity to successfully limit, or fail to control, Mtb replication.