Streptococcus pneumoniae drives specific and lasting Natural Killer cell memory

NK cells are important mediators of innate immunity and play an essential role for host protection against infection, although their responses to bacteria are poorly understood. Recently NK cells were shown to display memory properties, as characterized by an epigenetic signature leading to a stronger secondary response. Although NK cell memory could be a promising mechanism to fight against infection, it has not been described upon bacterial infection. Using a mouse model, we reveal that NK cells develop specific and long-term memory following sub-lethal infection with the extracellular pathogen Streptococcus pneumoniae. Memory NK cells display intrinsic sensing and response to bacteria in vitro, in a manner that is enhanced post-bacterial infection. In addition, their transfer into naïve mice confers protection from lethal infection for at least 12 weeks. Interestingly, NK cells display enhanced cytotoxic molecule production upon secondary stimulation and their protective role is dependent on Perforin and independent of IFNγ. Thus, our study identifies a new role for NK cells during bacterial infection, opening the possibility to harness innate immune memory for therapeutic purposes.

1. It is unclear whether lung NK cells from 21d post-infection respond similarly to in vitro stimulation as the spleen NK cells. For example, do they produce IFN-gamma and granzyme B upon stimulation with IL-15/IL-18/SPN? We agree on the interest of this point. The main reason we use NK cells from spleen is because the spleen is the largest reservoir of NK cells in mice. Technically, a much higher number of mice is needed to purify enough NK cells from lungs. In addition, in vitro experiments require many more NK cells than in vivo transfers. Therefore, technically we have chosen to answer this point by performing in vivo protection experiments. In the revised version of the manuscript, we have now added an experiment where we have transferred purified Memory or Naïve NK cells from lung (rather than spleen as in the original manuscript) to recipient mice in vivo (see Figure 3F). Interestingly, these results show that Memory NK cells from lung have a similar protective effect than those from spleen and contribute to reduce the number of CFUs in recipient mice. This data is now included in Figure 3F. We present below the data for NK cell responses in spleen following sub-lethal infections ( Figure R1). We do not observe any transient response in NK cells percentage, number or activation in the spleen following sub-lethal infection, consistent with the fact that we also do not detect CFUs in this organ. These data are shown below but will not be included in the manuscript as it is all negative data. We hypothesize that NK cells enter in contact with SPN in the lung and only later circulate to other organs such as the spleen.   1. The study contains plenty of high quality data, derived from well-powered experimental analyses. However, a couple of important conclusions are reached based on data trends that don't reach statistical significance. The two sections where histone modification were assessed both suggest interesting (and potentially functionally important) changes that may be driven by infection. However, in both cases, the sample size for analysis is small (n=4 and n=5), which prevents firm conclusions from being drawn. I would suggest the authors attempt an extra few replicates, to pin down whether the mechanisms they describe are driven by epigenetic modifications in ifng and gzmb. We agree with the reviewer, but technically we do not believe statistical significance will be reached without sacrificing a very large number of animals. Indeed, for these experiments, every replicate is performed by pooling several mice to obtain enough NK cells and enough DNA to do ChIP-qPCR. One extra repeat of this experiment would imply the sacrifice of at least 12 mice and we are not certain about how many repeats would be needed to reach significance. We believe that the fact that we are looking at bulk NK cells, while expecting that the memory cells are only a sub population, contributes to the lack of significance in this experiment. However, we believe the consistent trend suggests long-term chromatin remodeling. Once the subpopulation of memory NK cells is characterized, we will be able to characterize chromatin features in a more convincing manner.
2. Similarly, the differences in mouse survival described in Figure 3C are based on experiments with n=4 per group. This is a very small number for a survival experiment and I would have more confidence in the conclusions reached, if the results were reproduced in a larger sample size.
We have added one more replicate to this experiment, which is now performed on 24 mice. This has reduced the p-value from 0.069 to 0.060 and is included in a new figure 2.
Considering the cruelty of this experiment and the elevated number of mice (two groups of donor mice and two groups of recipient mice), we believe that the current data is sufficient to show that there is a protective phenotype, especially given that significance is achieved when measuring CFU numbers in different organs.
3. The adoptive transfer experiments performed with the ifngr KO mice are nicely conceived, demonstrating that IFNg is not the basis of the protective mechanism at play in NK cell memory of pneumococcal infection. I found the experiments with the perforin KO less convincing. Why were the prf1 KO mice used as recipients in these experiments? Transfer of prf1 KO NK cells with a memory phenotype into WT mice might have been used to demonstrate that perforin production by memory cells was the basis of protection. Transfer into the prf1 KO leaves open the possibility that perforin production by nonmemory cells might contribute to protection against pneumococcal infection. What was the trajectory/outcome of infection in the prf1 KO animals? Do they experience worse outcomes or harbour higher bacterial burdens than WT?
To begin by answering the last questions, Perforin KO animals responded similarly to WT mice. Perforin KO NK cells could still acquire memory properties, as measured by increased levels of GzmB following stimulation in vitro, and also in lung supernatants following infection (see Figure S5B and S5C). For the first part of the question, when NK cells from Perforin KO animals are transferred to WT mice, the protective phenotype is maintained (see below Figure R2). This therefore suggests that Perforin from endogenous immune cells was sufficient for transferred NK cells to perform memory. In contrast, when the experiment is done with KO cells into KO animals, the complete absence of Perforin prevents protective functions. We hypothesize that the main actor is GzmB, which requires Perforin for full activity, and the optimal experiment would have been done with GzmB KO animals, which cannot be easily obtained. Altogether, we agree with the reviewer and want to make sure a clear message is included in the manuscript. We have changed the wording for the conclusion of this section to read "our data suggests that memory NK cell protection from lethal infection is abolished in the absence of Perforin". 1. The infection with SPN is intranasal, and the target organ for SPN is the lung. However, the authors used splenic NK cells for adoptive transfer experiments. Could similar results be obtained with NK cells from other organs, in particular lung?

WT NK
We agree with the reviewer that this is an important point. We have performed new experiments transferring purified Memory and Naïve NK cells from lung in vivo. The results are included in a new panel of figure 4 ( Figure 4F), which shows Memory NK from lung contribute to reduce the numbers of CFUs recipient mice and shows that that lung NK cells from 21d post-infection also provide a protective effect to recipient mice.
2. Fig. 1: Do NK cells also respond to SPN + IL-15 alone (no IL-18)? How important is an inflammatory environment, mimicked by IL-18, for the NK cell response in presence of SPN?
We have performed a pilot experiment (see Figure R3) in which we have stimulated NK cells with IL15+SPN alone. The results show that, NK cells are not as activated (measured by GzmB+ cells) in the absence of IL18, and although an increased activation trend might be present in memory cells, the activation is more efficient and pronounced when both IL18 and IL15 are present during stimulation with SPN.
3. Fig. 1: Do NK cells respond with a similar response if treated with another formaldehyde-inactivated bacterium?
These data were included in figure 6A and 6B, where we show that although the addition of inactivated SPN stimulated Granzyme B and Perforin specifically in memory NK cells, the incubation with inactivated L. monocytogenes or Streptococcus agalactiae (GBS) did not induce an increase of Granzyme B and Perforin ( Figure 6A-B). These data therefore support the specificity of the response.
4. Fig. 1: How do NK cells respond to live, noninactivated SPN? This is an interesting point to which we have not found a technical way to address it. Due to bacterial growth over the incubation time, we are unable to incubate NK cells with live bacteria for as long, as we do with inactivated bacteria (24h). The maximum possible incubation time with live bacteria is approximately 6 hours, at which point NK   Figure  6C). This result suggests that memory mechanisms are not mediated by TLR receptors. As a control, we have followed NK cells response to LPS or the TLR agonist by measuring Ifng positive cells (new Supplementary Figure 5B).
6. The authors suggest that the memory NK cell response is specific to the first pathogen the mice have been infected with (SPN). It would be highly relevant to identify the reason for this specificity. Do NK cells depend the on the same PRRs for responding to SPN and to L. monocytogenes? Furthermore, the main target organs for SPN and L. monocytogenes differ (lung vs liver). Can the same 'specific' NK cell memory response as e.g. in Fig. 1 be confirmed if mice were infected with L. monocytogenes instead of SPN and NK cells re-stimulated with the same pathogen? This would reveal important information and confirm the authors' statements concerning 'specificity' of the memory NK response.
We appreciate the interest of the reviewer, and we agree on the interest of the points mentioned. However, to induce memory with L. monocytogenes would imply setting up a different infection method and dose, and we believe that this is beyond the scope of the paper. However, we have added experiments in the new manuscript to begin to address whether detection of bacteria is through PRRs. In these experiments we asked whether memory NK cells generated by infection with S. pneumoniae were reactive to other TLR agonists. This is not the case, as shown in the new figure 6C. Therefore, we believe that memory responses are not driven by activation of PRRs, consistent with in vivo findings of specificity.
7. The authors should extend the phenotypic characterization of the responding NK cells (vs non-responding NK cells), e.g. NK cell differentiation/maturation, Ki67. Since only bulk NK cells are compared, it may be that the differences are hidden when looking at this level which may be revealed when gating further down on NK cell subsets based on expression of Ly6C+, CD27low/neg, CD11b+ for example (Sun et al., Nature. 2009., Schuster et al., Immunity. 2023. We have investigated the phenotypic properties of responding vs non-responding NK cells by gating GzmB+ (responding) or Gzmb-(non-responding) in both Memory and Naïve NK cells stimulated in vitro with SPN. We have then measured the percentage of positive cells for CD11b and CD27. We have observed that cells with enhanced cytotoxicity (Gzmb+) have a more mature phenotype as the percentage of CD11b+ cells is significantly increased. This result is expected as previously published in Hayakawa et al., 2006, Chiossone et al., 2009or Kim et al., 2002 In addition, we do not observe a difference in the percentage of CD11b+ NK cells between Memory and Naïve NK cells. We neither found significant differences in the percentage or MFI of CD27 in the analyzed conditions. This data are now presented in Supplementary Figure  5C.
8. Fig. S1C: A significant difference for the MFI of Ly49D is not visually clear from the data and difficult to believe -the authors should both increase the number of experiments and provide representative plots/histograms in order to confirm their statement. We believe that the difference on the MFI of Ly49D is not biologically significant. We provide here a zoom of the figure to appreciate the difference. In addition, at 21 days this difference is no longer present. We do not consider ethically valid to sacrifice more mice to repeat this control experiment.
9. Fig. S2: The percentage of CD69+ NK cells in the lung seems overall rather low, and after 72h, it decreases even further. The authors should provide representative stainings for CD69 as well as data for CD69 expression before infection -does the frequency increase after 24h compared to before (mock-)infection, or is CD69 expression rather decreased for some reason at 72h and 21d? Furthermore, since higher percentages of CD69+ NK cells at 24h are even present in the PBS-control mice, it seems unlikely that this is an effect due to the infection. These data are confusing, and the authors must be careful with their conclusion of a low-level immune response.
In relation to this, the authors should also reveal why granzyme B seems to increase in the control group (see also comment below).

✱✱
We thank the reviewer for bringing up this point. We have carefully checked our data and realized that the values for CD69+ cells at 24h were incorrect. We have modified the graph with the right values. Now, it can be observed that at 24h and 72h there is an increase of CD69+ cells following the sub-lethal infection to induce memory. In the PBS condition, the percentage of CD69+ cells remain low at all timepoints and in the memory condition values are back to basal levels.
10. Fig. 2 and S2: From Fig. S2, no clear granzyme B signal is detectable in comparison to the isotype control reflecting that gzmb is not expressed to significant levels at 24, 72 and 21D (the few 'positive' events are rather likely an effect by spillover from other channels since I assume that this was not a FMO ctrl?). While it might be possible that lung NK cells express less granzyme B, splenic NK cells were found to also express granzyme B according to Fig. 2D, which is not supported by the representative data in Fig. S2D. In particular at 21d, Fig. 2D shows a clear percentage of granzyme B-positive cells in lung and spleen, which is not confirmed by the representative overlays in Fig.  S2D. This is confusing, and the authors need to present more reliable data to support their statements concerning granzyme B expression. In relation to this, the authors also need to present representative data for perforin expression at the different timepoints and groups in order to support their statements. We thank the reviewer for the suggestions. We have now revised the gating of these experiments and added more replicates. Also, the representative data has been revised for GzmB levels at 24h, 72h and 21D, where the levels of GzmB are low in all cases and there is no difference between Naïve and Memory NK cells at any timepoint. In addition, as the reviewer suggests, we have added MFI and representative data for perforin.
11. It is unclear why only splenic NK cells have been analyzed in their phenotype (Fig. S2).
The authors should add analyses on NK cells from other organs, in particular the lungs. Figure S2 is dedicated to the phenotypic characterization of the lung. We think the reviewer switched the two and is probably asking for phenotypic characterization of splenic NK cells. We are including below NK cell responses in spleen following sublethal infections ( Figure R1). We do not observe any transient response in NK cells percentage, number or activation in the spleen following sub-lethal infection, consistent with the fact that we do not detect CFUs in this organ.
12. Fig. 4A: It would be interesting to see whether the percentage of IFN-g+ NK cells further increase e.g. at 72h, or whether the peak of the response of D21SPN NK cells is reached earlier.
We agree this would be interesting. Unfortunately, the proposed experiment is not technically feasible as following lethal infection, recipient mice need to be sacrificed at 48h and are too sick to carry out the experiment in later timepoints.
13. The number of experiments should be increased for several datasets throughout the manuscript. In general, more than just one experiment should be performed for each figure. It is also not entirely clear why the authors sometimes show the data as pooled experiments, and in other plots each individual mouse. This is confusing, and the paper would benefit from a more consistent data presentation.
We have now increased the number of replicate experiments throughout the paper Specially in Figures 2, Supplementary 2 and Figure 6. For the in vitro experiments, purified NK cells from several mice are pooled together and distributed in wells for stimulation with different conditions, as this reduces the mouse to mouse variability.
14. The authors state that 'transferred congenic CD45.2+ NK cells were circulating and detectable at similar percentages in lungs and blood, suggesting there is no preferential trafficking between D21PBS and D21SPN NK cells'. As the lungs are perfused extensively with blood, any differences in NK cell number/percentages could be completely diluted and therefore missed. The lungs could be flushed to clear the blood and then stained to determine NK cell numbers/ percentages. Alternatively, fluorescently labelled anti-CD45 could be given IV before analysis to show what is actually circulating and what may be resident in the parenchyma instead. Again, subset specification may also pull out more interesting results. We thank the reviewer for the suggestion. However, as we do not specifically investigate the residency and trafficking of memory NK cells, we do not support A B sacrificing more mice to test this in all our conditions. However, we will take this suggestion into consideration for future experiments.
15. Fig. 5D: The reduction of CXCL1 in the D21SPN NK cells indicates that this chemokine was e.g. consumed by cells infiltrating the lung, hence, this is not a clear indicator for the lack of NK cell infiltration. The authors should combine this analysis with the expression patterns of the respective chemokine receptors on NK cells. We thank the reviewer for the comment, it would be interesting to check this in a further study, however, we believe that it is outside of the scope of this paper.
16. Fig. 6: The authors here show a population of granzyme B+ as well as perforin+ NK cells. Does the phenotype differ to the respective negative NK population?
We have investigated the phenotypic properties of responding vs non-responding NK cells by gating GzmB+ (responding) or Gzmb-(non-responding) in both Memory and Naïve NK cells stimulated in vitro with SPN. We have then measured the percentage of positive cells for CD11b and CD27. We have observed that cells with enhanced cytotoxicity (Gzmb+) have a more mature phenotype as the percentage of Cd11b+ cells is significantly increased. This result is expected as previously published in Hayakawa et al., 2006, Chiossone et al., 2009or Kim et al., 2002. In addition, we do not observe a difference in the percentage of Cd11b+ NK cells between Memory and Naïve NK cells. We also found no significant differences in the percentage or MFI of CD27 in the analyzed conditions. This data is now presented in Supplementary Figure  5C.