Author response:
The following is the authors’ response to the original reviews.
Reviewer #1:
The analysis of the dormancy rates is interesting and offers some intriguing questions related to the higher dormancy rate found for the L2 isolates and lower for the L3 ones. It will be interesting in the future to expand the data generated in this advanced in vitro plaAorm to in vivo studies.
Indeed, an increased dormancy propensity of L2 isolates was previously reported in broth culture and associated to specific genetic polymorphisms. The opposite phenotype observed in the L3 isolates is indeed particularly intriguing and was not described to date. Hence, we fully agree that it would be very interesting to find out whether these phenotypes are also observed in vivo.
The authors propose that ‘strains exhibiting greater proliferative capacity are more prone to induce macrophage apoptosis, thereby contributing to the extent of the granulomatous response.’ It would be interesting to know what happens if the macrophage apoptotic response is blocked.
This is an interesting suggestion that would deserve a dedicated comprehensive investigation covering other cell death pathways. Even though the trend is significant, the correlation coefficient is rather low in this interaction, which looks a fortiori due to substantial inter-host variability in the apoptotic propensity of macrophages from individual donors to a given strain. In addition, such blocking experiments may require performing isolated macrophage infections that would fall outside of the scope of this study, or considering the extent and the contribution of the apoptosis of other cell subsets.
In contrast to macrophage apoptosis, T cell activation correlated with less replicative bacteria. Are these two findings related, ie, are the granulomas showing more (apoptotic) macrophages the ones with a lower percentage of activated T cells? This would shed light on what distinguishes granulomas that are protective from those that support bacterial growth.
Indeed, a significant negative correlation between macrophage apoptosis induction and T cell activation can be observed, specifically with activated CD4 T cells expressing CD38 (rS = -0.36, p < 0.05) or CD69 (rS = -0.40, p < 0.01). We have added this additional result in the manuscript text (line 217).
It would also be interesting to know the functional impact of blocking early CXCL9 or IL1b on the outcome of granulomatous response/bacteria growth.
We have performed the suggested early blocking experiments and added the expected negative effect on granuloma formation upon neutralization of IL-1b (current Fig. 6E) in the revised version of the manuscript, and furthermore discussed the null effect on bacterial growth of the treatment with an anti-CXCL-9 specific antibody (current Fig. 6H).
The authors acknowledge the absence of neutrophils in this model. However, this could be discussed in more detail, as neutrophils play an important part in TB pathogenesis as shown in different models of infection and human TB.
We concur and have expanded the importance of neutrophils in TB pathogenesis (including references) in the discussion section (line 260).
Related to neutrophils and TB pathogenesis, another important player is type I IFN. The multiplex assay used included IFN-alpha, was this molecule detected? If so, was there any difference in the levels of type I IFN detected among the different infections?
We agree and that is why we had originally included IFN-α in our screen. However, this cytokine remained under the limit of quantification at both studied time points, preventing us to draw conclusions on the effect of Mtb strain diversity on the secretion of type I IFNs in in vitro granulomas.
Reviewer #2:
In Figure 1b/c, it is not clear what comparisons are being made to give the p-value annotations.
In Figure 2a/b, it is not clear what comparisons are being made to give the p-value annotations.
In Figure 3a, again it is not clear what comparisons are being made to give the p-value annotation.
The p-values formerly present on the upper le] corner of the panels were resulting from either Friedman (Figures 1C, 2A and 3A) or Kruskal-Wallis (Figures 1B and 2B) tests and indicated whether there was a significant difference between the analyzed groups overall. To avoid confusion, those values have been removed to only leave the post-test comparison between specific groups.
In the results narrative related to Figure 1 (lines 93-103), the authors refer to lineage heterogeneity without providing any objective quantification of this - I suggest they do so, by providing variance or standard deviations.
Thank you very much for this relevant suggestion, we have now included the coefficients of variation as a quantitative measure of the within-lineage heterogeneity in the manuscript (line 97).
I also suggest the authors explain what the data points actually represent in this figure - do I assume each data point = cfu from a well of 'granuloma'? Are they all from the same donor PBMC? What is the sample N for each lineage? If the data are not from the same donor PBMC, I think more informative to present the results of paired statistical analyses, stratified by donor cells. In addition, the authors should include a summary table of the demographic characteristics of the donors (at least sex, ethnicity, and age). If the data are derived from a single donor, I'd advocate providing data from at least one further donor.
In the new supplementary figure requested by Reviewer 3 Figure 1—figure supplement 1 (actual CFU data on days 1 and 8 p.i. used to calculate the growth rate) it is now indicated that bacterial load was quantified as CFU per well.
Regarding the number of donors used, as stated in the Material and Methods section (current line 418) and depicted by the four different shapes used when data are grouped by individual infecting strain, all figures in our manuscript have been generated using PBMCs from 4 independent donors. For greater clarity, “n = 4” has now been included in the figure legends. Regarding the statistical analyses, paired statistical analyses stratified by donor were already performed in the original version of the manuscript whenever appropriate.
As stated in the methods section, the buffy coats used for PBMC isolation are anonymized so demographic data are unavailable.
The premise of the analysis in Figure tic and the results narrative ("This finding suggests that an increased ability to enter dormancy is not necessarily associated with a more pronounced growth phenotype", line 132) is not clear to me. Why would increased dormancy relate to increased growth in the same context? I suggest this analysis be removed.
We apologize for the confusion in our original statement. We now rephrased it as “This finding suggests that an increased tendency to remain in a metabolically active state is not necessarily associated with a more pronounced growth phenotype”.
In Figure 3b, I think it may be more informative if the data points from the same donor were linked. Likewise in Figure 3c, I'd like to see a donor-paired statistical analysis.
For all figures, the choice of using individual symbols to identify data points from the same donor but not connecting lines was made to provide a neater image. Nevertheless, we have now modified the figure linking the data points from the same donor. The statistical analysis performed is always donor-paired whenever appropriate.
The casual inference suggested in the results narrative between ‘macrophage apoptosis’ and granulomatous response line 173-175) is not tested directly by the experiment – I suggest the authors exclude this statement.
Fair point, the statement has been removed.
To what extent have the authors considered whether variation in T cell responses between lineages may be confounded by variation in Mtb reactive T cell frequencies in donor PBMC. Can this be disentangled at all? This should be acknowledged as a potential limitation of the study.
We did characterize the presence of mycobacterial antigen-specific reactive T cells in the PBMCs from the investigated donors. To do so, we performed in vitro stimulations with purified protein derivative (PPD) or an ESAT-6/CFP-10 peptide pool and quantified the frequency of IFN-γ-positive CD4 T cells by flow cytometry. The percentage of IFN-γg-positive CD4 T cells recalled by PPD stimulation ranged from 0.02% to 0.13%, while no ESAT6/CFP-10 reactive T cells were detected. As such, we can akest that the PBMC donors never encountered Mtb even though some levels of memory recalled by PPD may be due to cross-reactivity with BCG or pre-exposure to non-tuberculous mycobacteria. We have now added a panel in Figure 5—figure supplement 2 representing the frequency of mycobacteria-specific CD4 T cells and, as suggested, discussed the impact on the extent of the T cell responses observed in granulomas in the revised version of the manuscript. Nevertheless, the observed MTBC strain-specific trends are consistent across the donors, as depicted in Figure 5B and Figure 5—figure supplement 2A-B.
Moreover, the experimental design does not really test cause and effect for the relationship between T cell proliferation/activation and bacterial growth. What is the impact of T-cell depletion from PBMC on bacterial growth?
The increased TB susceptibility of HIV patients demonstrated that T cells play a critical part in the control of Mtb infection. We agree and did envisage such a depletion experiment. However, depleting T cells from PBMCs would imply removing up to 70% of the cells present in the specimen, which would lead to a situation from which results cannot be compared to the original sample and therefore would not be interpretable.
Reviewer #3:
Data presentation:
- In Figure 1 (replication rate), actual cumulative CFU means from each strain for both days 1 and 8 with statistical analysis should be presented as panels in this figure.
Agreed. We are providing the requested representation of the data and the corresponding paired statistical analysis as supplementary material Figure 1—figure supplement 1.
- In Figure 2 (dormancy), a panel comparing the mean number of bacteria that are single positive for either Auramine-O, Nile Red, or are double positive should be included for each strain, with statistical analysis. Representative photomicrographs of phenotypes from the staining should also be included. Electron microscopy could be conducted to compare the presence of intermediate lipid inclusions within organoidbound mycobacteria.
As requested, percentages of single stained as well as double positive bacilli in each sample are now represented in Figure 2—figure supplement 1. In addition, we have now also followed the request and included a photomicrograph picturing representative Mtb staining phenotypes. Lastly, it would certainly be very elegant to visualize the presence of Mtb lipid inclusions within cellular aggregates by electron microscopy. However, we do not currently have the means for such investigations and the implementation of such a protocol under BSL3 conditions appears unrealistic in the context of this study.
- In Figure 3 (granulomatous response), the number, circularity, and size of immune aggregates are presented as "granuloma score" in which the mean ratio of size to circularity is divided by the number of inclusions. To their credit, in Supplementary Figure 2, the authors provide the data in a straighAorward manner. However, the granuloma score metric is reduced as the number of observed "granulomas" increases, which is counterintuitive. Additionally, circularity is not a definitive aspect of human granulomas (Wells et al., Am J Respir Crit Care Med, 2021, PMID: 34015247). I am skeptical that the "granuloma score" is an accurate predictor granulomatous inflammation. Is there precedent for this metric in the literature? If so, a reference should be provided. A high magnification inset of 1 representative granuloma from each strain should be included in Figure 3A.
As requested, insets of a representative average granuloma for each strain have been included in Figure 3A. The formulation of the “granuloma score” has no precedent and cannot be referenced. By doing so, we meant to integrate within one single parameter the visual differences represented in the current Figure 3— figure supplement 2. We intentionally sought to assign the highest score to the massive aggregation that some strains may promote unlike some that trigger several small, dispersed and diffused aggregates.
- In Figure 4 (macrophage apoptosis), a panel showing the percentage of dual Annexin V and 7-AAD positive cells should be included to provide the reader with the relative scope of ongoing apoptotic vs necrotic/secondary necrotic death in the model. If the data is readily available, including a control of uninfected PBMCs would also allow the reader to evaluate donor-dependent differences of in vitro cell death at baseline.
No significant differences were observed in the percentage of dual Annexin V- and 7-AAD-positive macrophages (necrosis/secondary necrosis) between the MTBC strains at this time-point. Nevertheless, we have disclosed this result in the revised manuscript as Figure 4—figure supplement 2.
- In Figures 5 and 6 (lymphocyte activation and soluble mediator secretion), panels showing unscaled data should be included. Panels depicting the unscaled immunoassay protein readings (pg/mL) by strain for CXCL9, granzyme B, and TNF with statistical analysis should be included in Figure 6.
As requested, unscaled lymphocyte activation and soluble mediator data have been included as Figure 5— figure supplement 2 and Figure 6—figure supplement 1, respectively (replacing former supplementary figures 5 and 7). In addition, updated Figure 6G panel now depicts correlation analysis with the unscaled cytokine concentrations.
The DosR-regulon:
The authors hypothesize that differences in the prevalence of the dormancy metrics (acid-fastness or lipid inclusion prevalence, are due to strain-specific increases in expression of the DosR regulon within the model's hypoxic conditions (lines 107-114, 126-127). The claim that their model is equipped to evaluate dosR-dependent mycobacterial phenotypes was also previously proposed (Arbués et el., 2021) and should be tested. A comparison of the dosR-dependent gene expression of each strain in PBMC aggregates and broth culture by qRT-PCR would test this idea at a very basic level.
We agree. Actually, a similar request was made during the revision of our first in vitro granuloma study for which such qPCR data were generated and presented in Fig. 1 D (PMID: 32069329). In addition, the work of Kapoor et al., who originally developed the in vitro granuloma model also demonstrated the induction of most of the DosR regulated genes by qPCR (PMID: 23308269). We trust that the reviewer will agree that this does not need to be repeated.
The modern Beijing lineage strain L2C:
The authors claim (Line 101-102) that the results of Figure 1 "confirm the higher virulence propensities of strains from modern lineages". From the data presented, it appears that strain L2C (Modern-Beijing) dominates the modern vs ancestral and inter/intra-lineage phenotypes of replication, dormancy, and apoptosis. Are significant differences between modern and ancestral lineages or between strains simply a facet of the distinct profile of L2C? Do the statistical differences disappear when the L2C group is excluded?
Indeed, among the modern lineages’ isolates, L2C exhibits a hypervirulent profile in terms of bacterial replication. However, the difference between modern and ancestral strains remains statistically significant when L2C is excluded from the analysis (p = 0.002). That is also the case when we analyze the proportion of dormant bacteria. Exclusion of L2C strain results in a Kruskal-Wallis overall p = 0.005, and p = 0.0002 when we compare L2 vs. L3. Lastly, regarding the percentage of apoptotic macrophages, if we use L2B (instead of L2C) to compare, the difference is still significant vs. L1A (p = 0.008) although there is no longer a trend for L2A (p = 0.1).
"Dormancy":
Dormancy is definitively a non-replicative state, where bacterial growth is absent. The authors' findings and claims appear to be incompatible with that definition, which they acknowledge (Lines 130-135). The lack of correlation between growth and dormancy in their model is supported with reference to Figure 2C, a Spearman's analysis of dormancy ratio with growth rate (inclusive of all strains under consideration). The figure supports a model where "dormancy" and "growth rate" are disjunct but also appears to show high "dormancy" accompanying increasing "growth" in the L2C group. How are strains able to grow if they are in a non-replicative state? Are the "growth rate" assays actually measures of survival? Are there different rates of infectivity? Are the bacteria growing cellularly in the serum-rich ECM, etc. etc? We need to see the hard CFU and Nile Red, and Auramine-O data to contextualize these findings. Alternatively, could the accumulation of inclusions in the model not be a reliable dormancy metric (Fines et al., BioRxiv [Preprint], 2023, PMID: 37609245)?
We fully agree. The Nile red profiles are always relative and only depict the proportion of the population that has entered a dormant state. Nevertheless, dormancy can be dynamic and bacteria may swi]ly resuscitate in that model. Furthermore, and as depicted in Figure 2—figure supplement 1, despite showing an increased tendency to enter a dormant-like state, a considerable population of lineage 2 bacilli still remains metabolically active and in a replicative state. The referred preprint is very interesting and we will follow it up closely.
Specificity of responses to PBMC aggregation:
The authors claim that their results "reveal a broad spectrum of granulomatous responses" (Line 73) but do not show any aggregation specificity of PBMC responses beyond the model's intrinsic metrics of area and circularity. To establish that their phenotypes such as lymphocyte activation, cytokine release, cell death, or mycobacterial acid-fastness/lipid inclusion prevalence, are aspects of the granulomatous response the authors could infect PBMCs from the same donors with the same strains and perform the same assays using established Mtb-PBMC models in which the cells do not aggregate. This would answer many important questions, for example, does the rate of macrophage infection account for variability in apoptosis percentage? Phagocytosis assay and quantification of stained intracellular mycobacteria within recently infected PBMCs could be conducted to determine if phenotypes are an aspect of granulomatous aggregation or due to strain-specific differences in cellintrinsic macrophage immunity. It would also be very informative to know what percentage of PBMCs and mycobacteria are granuloma-bound in the ECM.
We are not aware of Mtb-PBMC models in which the cells do not aggregate. We previously compared PBMC infection models in the presence or absence of the collagen matrix and cells also spontaneously coalesced around infection foci (PMID: 34603299). Regarding the last point, the melting step of the collagen matrix requires enzymatic digestion and pipetting that dislocate the aggregates. Accordingly, we cannot distinguish the bacteria that would remain within the matrix compared to those replicating within cellular aggregates. However, we did resolve this question by demonstrating that the bacteria were not able to grow in the absence of cells in this culture condition (Supplementary material, PMID: 34603299)
Minor recommendations
- The term TNF-a should be replaced with TNF throughout the manuscript.
We acknowledge that the term TNF-a can be interchangeable with TNF. However, we chose to use the TNFα terminology to differentiate it from lymphotoxin α, which is also referred to as TNF-β.
- The authors cite studies conducted in murine and NHP models to support the claim that "understanding of immune protective traits in TB remains insufficient and yet dominated by data from mouse and non-human primate studies" (Lines 63-64) but ignore an abundance of data from other in vivo and in vitro models that have provided numerous valuable insights in the field of TB immunology. This line should be revised or omired.
For us, the term “dominate” implies that these models are widely used, not that they are the only ones. Other models indeed provided additional relevant data. We are citing the lung-on-chip model of McKinney’lab and the in vitro granuloma model of Elkigton’s lab (line 66). We would be very happy to include more references upon further specifications even though we cannot build an extensive review here.
- The authors claim that their model "encompasses, with the exception of neutrophils, all immune cell types involved in TB" (Lines 67-68). To support this claim, they should provide additional references or data demonstrating that the PBMC aggregates include, eosinophils, mast cells, dendritic cells, yolk-sac-derived alveolar macrophages, and Langhan's giant cells.
With the aim of providing a more accurate and detailed information regarding the cell types present in the model, the sentence has been reformulated as: “The model encompasses all PBMC-derived cell types involved in TB immune responses, but lacks granulocytes (i.e. neutrophils, eosinophils, basophils and mast cells)” (line 260). Noteworthy, the presence of multinucleated giant cells was reported in Kapoor’s paper describing the in vitro granuloma model for the first time (PMID: 23308269).
- As an additional note, the title can be improved and made more broadly accessible by revising the use of the acronyms CXCL9, granzyme B, and TNF-α.
To render the title more broadly accessible we propose to replace the listed acronyms by “soluble immune mediators”, but we remain opened to more appropriate and specific suggestions.
Answers to the reviewers’ public comments
Reviewer #1:
First of all, we would like to thank the reviewers for their feedback and suggestions to improve our manuscript. To strengthen the findings of our study, we have performed and added results from IL-1b and CXCL9 blocking experiments evaluating the impact on the granulomatous response and bacterial load, respectively. In the revised version of the manuscript, while we discuss the null effect on bacterial growth of the treatment with an anti-CXCL-9 antibody and the potential reason behind it, we are now reporting a negative effect on the magnitude of granuloma formation upon neutralization of IL-1b that the correlation analysis had initially suggested.
Reviewer #2:
The revised version of our manuscript incorporates now all the points detailed in the private answers to the reviewer, including clarifications on the statistical tests performed, additional supplementary materials to transparently disclose the raw data behind the normalization approach, as well as flow cytometry data on the immune memory status of the blood donors. In addition, and as stated in the answer to reviewer #1, to test causal relationship between some host and pathogen traits, we have now performed and provided data and interpretation of IL-1b and CXCL9 blocking experiments.
Reviewer #3:
We are thankful and concur with these constructive comments and insights. We have now consistently revisited the statistics in the figures to improve clarity and included new supplementary figures reporting the raw data that were missing in the initial version of the manuscript. In addition, and as mentioned in the answers to reviewers #1 and #2, we have now performed and added IL-1β and CXCL9 blocking experiments to test causal relationship between specific host and pathogen traits. In particular, we are now reporting a negative effect on the magnitude of granuloma formation upon neutralization of IL-1β that the correlation analysis had initially suggested.
More specifically, regarding the point that our method for bacterial collection calls into question whether all Mtb plated for CFU assay resided within granulomatous aggregates, we previously reported that Mtb growth strictly required the presence of human cells in our culture conditions (Supplementary material, Arbués et al, 2021, PMID: 34603299). In the presence of cells, our microscopy read-out does allow us to observe extra-cellular growth if infections are carried on beyond an 8-day limit, which we applied in the current study to exclude this particular caveat.
Concerning the apparently conflicting observation that those strains displaying an increased tendency to enter a dormant-like state are the ones exhibiting the highest replication rates, we would like to point out that a considerable population of bacilli still remains metabolically active and in a replicative state. For instance, and as depicted in Figure 2—figure supplement 1, despite showing an increased tendency to enter a dormant-like state, a considerable population of lineage 2 bacilli does remain metabolically active. Moreover, dormancy can be dynamic and bacteria may swi]ly resuscitate.
Regarding the mentioned limitations of our study that we have discussed in the revised version of our manuscript, we fully concur that PBMC-based in vitro granuloma models lack tissue structure as well as some important stromal and immune cellular players. Nevertheless, we and others demonstrated the particular relevance of the 3-dimensional infection approach within a matrix of collagen and fibronectin by providing mechanistical insights into Mtb resuscitation previously associated to treatment with various immunomodulatory drugs (Arbués et al., 2020, PMID: 32069329; Tezera et al., 2020, PMID: 32091388).
