Early cellular mechanism of type I interferon-driven susceptibility to tuberculosis

Mycobacterium tuberculosis (Mtb) causes 1.6 million deaths annually. Active tuberculosis correlates with a neutrophil-driven type I interferon (IFN) signature, but the cellular mechanisms underlying tuberculosis pathogenesis remain poorly understood. We found interstitial macrophages (IMs) and plasmacytoid dendritic cells (pDCs) are dominant producers of type I IFN during Mtb infection in mice and non-human primates, and pDCs localize near human Mtb granulomas. Depletion of pDCs reduces Mtb burdens, implicating pDCs in tuberculosis pathogenesis. During IFN-driven disease, we observe abundant DNA-containing neutrophil extracellular traps (NETs) known to activate pDCs. Cell type-specific disruption of the type I IFN receptor suggests IFNs act on IMs to inhibit Mtb control. Single cell RNA-seq indicates type I IFN-responsive cells are defective in their response to IFNγ, a cytokine critical for Mtb control. We propose pDC-derived type I IFNs act on IMs to drive bacterial replication, further neutrophil recruitment, and active tuberculosis disease.


Introduction
Mycobacterium tuberculosis (Mtb), the causative agent of tuberculosis disease, caused 1.6 million deaths in 2021 1 . Treatment requires a minimum 4-6 month course of antibiotics, or up to 2 years for increasingly prevalent multi-drug resistant strains. Moreover, the only approved vaccine for Mtb has variable or no efficacy in adults 2 . The pathophysiology of tuberculosis remains poorly understood. The mouse model has been used to discover most of the host factors known to control tuberculosis in humans, including tumor necrosis (TNF) factor and interferon-g [3][4][5] . Nevertheless, the use of mice has been criticized for poorly recapitulating key aspects of human disease 6 .
In humans, active tuberculosis disease is reproducibly associated with the induction of type I interferons (IFNs) 7-10 , a family of cytokines, including IFNα and IFNβ, that signal via the type I IFN receptor (IFNAR). Type I IFNs elicit a primarily anti-viral response that overlaps but is insufficient to recapitulate the protective anti-Mtb response elicited by IFNg. Infection with type I IFN-inducing viruses is associated with worse Mtb infection outcomes in humans. For example, influenza infection correlates with an increased risk of death among patients with pulmonary tuberculosis, and infants with cytomegalovirus have an increased risk of tuberculosis disease [11][12][13] . These human studies do not prove that the correlation between virus infection and Mtb disease is due to the induction of type I IFNs. However, a causal role for type I IFNs in driving human tuberculosis is supported by the finding that a partial loss-of-function mutation in IFNAR is associated with Mtb resistance in humans 14 . Moreover, a type I IFN response has the potential to be a driver of progression to active Mtb in humans, as type I IFNs can antagonize the critical, protective IFNg response during mycobacterial infection 5,15,16 .

Macrophages and neutrophils exhibit a variety of activation states during Mtb infection
To further characterize the Mtb-infected myeloid cells and dissect the cellular mechanism of type I IFN driven Mtb susceptibility, we compared B6 and Sp140 -/innate immune responses by performing scRNA-seq on myeloid cells from Mtb-infected or uninfected lungs 25 days after infection. For this experiment, CD64 + and Ly6G + cells were magnetically enriched, sort purified, and processed for library generation with the 10X Genomics platform ( Fig. 2A). To ensure proper cell clustering, mRNA transcripts and protein expression for select lineage markers were simultaneously measured by CITE-seq (Supplementary Fig. 3) 42 . Resulting datasets were analyzed with Seurat V4, and Weighted Nearest Neighbor (WNN) analysis was used to cluster cells based on mRNA and protein expression 43 . Analyzed datasets were then visualized by uniform manifold approximation and projection (UMAP) reduction on the WNN clustered data (wnnUMAP) 44 . The resulting dataset consists of 6,604 B6 and 13,668 Sp140 -/cells, almost exclusively consisting of myeloid cells (Fig. 2B). Major cell types were annotated based on protein or mRNA expression of lineage defining markers, such as Siglec F protein expression for identifying AMs (Supplementary Fig. 4) 45 . Individual clusters within a major cell type, like the 10 clusters of neutrophils, were annotated based on expression of maturation and activation markers (Supplementary Fig. 5) 46 . Each cluster is represented in the B6 and Sp140 -/datasets, however the proportions of some clusters are altered between genotypes. Most notably, the ratio of IFN stimulated gene (ISG) + IM to ISG -IM was higher in the Sp140 -/mice, as compared to B6, as expected from the exacerbated type I IFN response in Sp140 -/mice (Fig. 2B). The largest changes in composition are seen when comparing cells from naïve lungs to bystander and Mtb-infected cells from Mtb-infected lungs (Fig. 2B). For example, AMs are relatively abundant in naïve lungs but are rare among the Mtb-infected cells 25 days postinfection, as also seen by flow cytometry (Fig. 1F, 2B). To establish whether the differences in immune response between Sp140 -/and B6 mice occurred in response to Mtb infection, we compared naïve Sp140 -/and B6 lungs by scRNA-seq. Consistent with the normal cellular profile of naïve Sp140 -/mice (Supplementary Fig. 2), fewer than 10 differentially expressed genes were identified between the two genotypes in AMs, IMs, monocytes, and neutrophils (Supplementary Fig. 6). These results suggest that the immune compartment of B6 and Sp140 -/mice is highly similar at baseline, and the type I IFN-driven changes in the genotypes occur after Mtb infection.

Bystander pDCs and IMs are the primary sources of type I IFN during Mtb infection
To determine the cellular mechanism of type I IFN-driven Mtb susceptibility, we first sought to identify which cells produce type I IFN following infection. In general, our scRNAseq analysis revealed that very few cells were Ifnb1 positive, which may reflect a lack of sensitivity of scRNAseq, and/or the transient and stochastic expression pattern of this gene (Fig. 3A) [47][48][49][50] . Mtb infection resulted in increased expression of Ifnb1 in infected and bystander mononuclear phagocytes, with a slight bias towards Ifnb1 production by IMs compared to monocytes, and no production by AMs (Fig. 3A). While there was no major difference in the cell types producing Ifnb1 between B6 and Sp140 -/cells, a greater number and frequency of Sp140 -/cells expressed Ifnb1 compared to B6 cells (Fig. 3B). Additionally, Ifnb1 expressing cells in Sp140 -/mice trended towards a higher per cell expression of Ifnb1 than B6 cells, but this difference was not significant (Fig. 3C). Consistent with these findings, a prior scRNA-seq study of Mtb-infected and naïve lungs from non-human primates largely mirrors our findings in mice 51 . Our analyses of these data indicate that IMs were also the dominant IFNB1-expressing cells in non-human primates with active tuberculosis, and IMs did not express IFNB1 in naïve or latently infected lungs (Fig. 3D). These results suggest that mice faithfully recapitulate the Mtb-induced type I IFN production seen in non-human primates.
The type I IFN producers identified in the scRNA-seq datasets were validated using a genetic reporter of type I IFN production, called I-Tomcat mice (from Dan Stetson, manuscript in preparation). These mice express TdTomato and Cre downstream of Ifnb1; therefore, any cell that expresses Ifnb1 will also express TdTomato and Cre (Fig. 3E). While TdTomato expression was sufficient to identify Ifnb1 expression by I-Tomcat bone marrow-derived macrophages following in vitro stimulation with poly I:C (Supplementary Fig. 7A), TdTomato + cells were not detected 25 days after Mtb infection (Fig. 3F). The ability to detect TdTomato + cells was not improved by using I-Tomcat homozygous mice, examining an earlier timepoint of 19 days postinfection, or by gating on specific immune populations such as IMs (Supplementary Fig. 7B, 7C, 7D). Even though type I IFN drives the Mtb susceptibility of Sp140 -/mice 25 days postinfection, it is unclear when the type I IFN production occurs (Fig. 1A). It is possible that type I IFN is an early and/or transient event, which would be missed by analyzing a single timepoint with the I-Tomcat mice. To address this issue, we crossed I-Tomcat mice with the Ai6 Cre reporter mouse line to generate I-Tomcat Ai6 mice on both the B6 and Sp140 -/backgrounds ( Fig. 3E) 52 . In these mice, any cell that has ever expressed Ifnb1 will constitutively express ZsGreen, allowing for their sensitive detection; importantly, however, the levels of ZsGreen do not report the levels of Ifnb1 expression. Mtb-infected I-Tomcat Ai6 mice clearly contained populations of reporter-positive myeloid cells, while demonstrating low background among cell populations that are not expected to be reporter positive (e.g., ~0.1% of T cells were Ai6 + ) (Fig.  3G). Consistent with the scRNAseq analysis, IMs and monocytes were the primary Ifnb1 expressing cells in B6 and Sp140 -/mice (Fig. 3H, Supplementary Fig. 8). Interestingly, Sp140-deficient mice exhibited elevated Ai6 + expression frequency in all cell types, suggesting SP140 broadly modulates the sensitivity for inducing Ifnb1 expression (Fig. 3H). In addition to corroborating the scRNA-seq data, the I-Tomcat mice also identified pDCs as a major type I IFN producing cell population. Lung pDCs are very rare and were therefore not represented in our scRNA-seq dataset, demonstrating the power of using genetic reporters to study rare events and cell populations. Despite their scarcity, pDCs are known to be extremely robust producers of type I IFNs on a per cell basis 53 .
While we expected IMs to be a major type I IFN producing population given the scRNAseq results, we were surprised that the majority of the Ai6 + IMs were Mtband most Mtb + IMs were Ai6 - (Fig. 3G). These results suggest that direct infection of IMs is neither required nor sufficient for IFN-β production. To examine this phenomenon in greater detail, we performed confocal microscopy and histo-cytometry analysis of Mtb-infected I-Tomcat Ai6 and Sp140 -/-I-Tomcat Ai6 lungs 54,55 . While lesions of diseased tissue were clearly identifiable in I-Tomcat Ai6 mice, the size and myeloid cell influx into the diseased tissue were greatly exacerbated in Sp140 -/-I-Tomcat Ai6 (Fig. 4A). Additionally, Ai6 expressing cells were identifiable throughout the lungs, with an increased propensity to localize in diseased rather than healthy tissue (Fig.  4A, Supplementary Fig. 9). Within diseased tissue, Ai6 expressing cells were primarily located near Mtb harboring cells in I-Tomcat Ai6 and Sp140 -/-I-Tomcat Ai6 lungs (Fig. 4B). Similar to the flow cytometry results, SIRPɑ + cells, which are primarily macrophages in Mtb-infected lungs as they are ~100 fold more abundant than SIRPɑ expressing cDC2s, were a major Ai6 expressing cell population (Fig. 1D, 4B, 4C). The SIRPɑ + macrophages expressed Ai6 at a higher frequency than CD4 + T cells in the diseased tissue but not healthy tissue (Fig. 3H, 4B, 4C). Direct infection by Mtb was not a major driver of IFN-β expression, as ~2-3% of infected macrophages were Ai6 + and ~12-15% of Ai6 + cells were Mtb-infected, in line with the frequencies seen in IMs by flow cytometry (Fig. 3H, 4D). These results suggest that IM localization to Mtb rich regions provides the activating signals required for IFN-β expression, while direct infection of IMs is not required for IFN-β expression.
pDCs significantly contribute to the Mtb susceptibility of Sp140 -/animals While pDCs have a well-established role in anti-viral immunity in the lung, limited work has assessed their contribution during Mtb infection 56,57 . pDCs may have been previously overlooked because of their scarcity in the lung. Indeed, we observe only ~20,000 pDCs in the lungs of naïve B6 and Sp140 -/mice, but this number increases 10-fold following Mtb infection and is significantly higher in Mtb-infected Sp140 -/than B6 mice (Fig. 5A). Despite their scarcity, pDCs can have major effects due to the extremely high levels of interferons produced per cell 53 . Consistent with our finding that pDCs are a major type I IFN producer in Mtb-infected mouse lungs, Khader and colleagues described the presence of pDCs in lungs of non-human primates with active pulmonary TB 51 . However, the lack of genetic tools in non-human primates precluded functional studies of pDCs during TB. Therefore, we decided to take advantage of our experimentally tractable mouse model to assess whether pDCs affect Mtb control. The contribution of pDCs was initially tested by depleting pDCs using an anti-BST2 antibody [58][59][60] . This strategy efficiently depleted pDCs and resulted in a partial rescue of Mtb control in Sp140 -/mice (Fig. 5B, 5C). However, BST2 is known to be upregulated by cells other than pDCs in inflammatory environments, thus antibody depletion could have been protective against Mtb by depleting non-pDC cells 59 . We therefore also tested the contribution of pDCs by using a genetic pDC depletion strategy in which we crossed our Sp140 -/mice with mice expressing the diphtheria toxin receptor (DTR) downstream of the human BDCA2 promoter (pDC-DTR) 61 . We used Sp140 +/-pDC-DTR littermates as wild-type controls since a single copy of Sp140 is sufficient to restore wild-type control of Mtb. DT administration efficiently ablated pDCs in Sp140 -/and Sp140 +/mice, with the depletion specifically affecting pDCs (Fig. 5D,  Supplementary Fig. 10). pDC-DTR depletion of pDCs was able to fully rescue bacterial control in Sp140 -/mice, while depletion in Sp140-sufficient animals (that do not exhibit an exacerbated type I IFN response) did not affect lung bacterial burden, as expected (Fig. 5E). Additionally, pDC-DTR depletion of pDCs in Sp140 -/mice reduced expression of type I IFN stimulated genes to the level of Sp140-sufficient animals, while rescuing expression of the type II IFN stimulated gene H2-Ab1 (Fig. 5F, 5G). These results demonstrate a novel contribution by pDCs in limiting Mtb control in animals with a hyper type I IFN response.
While our results demonstrate a role for pDCs during Mtb infection in mice, and the Khader lab identified a correlation between pDCs and active Mtb in non-human primates, the role of pDCs during human Mtb infection has yet to be examined 51 . Therefore, we analyzed human lung and lymph node biopsies taken from Mtb culture-positive patients for the presence of pDCs near Mtb granulomas (Fig. 5H). Based on CD303 and CD123 staining, pDCs localized to the lymphocytic cuff surrounding Mtb granulomas in human lungs and lymph nodes (Fig. 5I,  5I, Supplementary Fig. 10). Of the 8 patient samples analyzed, 5 lung samples and 7 lymph node samples had pDCs in the same 400× field as an Mtb granuloma (Supplementary Table  1). The majority of the pDCs in the lung samples were distributed as individual cells, while lymph node pDCs were primarily grouped together in clusters of over 20 cells or scattered individually (Supplementary Table 1). These results demonstrate that pDCs, though generally an extremely rare cell population, are nevertheless located near Mtb-infected cells in granulomas in human lung and lymph nodes. These results, along with our results in mice and previous studies in non-human primates 51 , implicate pDCs as a plausible source of type I IFN that drives active tuberculosis in humans.
To understand why pDCs contribute to the susceptibility of Sp140 -/but not Sp140sufficient animals, we examined the availability of ligands that might potentially activate pDCs to produce type I IFNs. We focused on DNA-rich neutrophil extracellular traps (NETs) as a potential pDC-activating ligand because extracellular DNA is known to activate pDCs via TLR9, and NETs have been described as a stimulus for type I interferon production by pDCs in mice and humans in the context of autoimmunity 34,35,62 . Additionally, another Mtb susceptible mouse model with a hyper type I interferon response identified the presence of NETs in the lungs of susceptible mice and humans with active Mtb disease 63 . We first assessed whether neutrophils were enriched in Mtb-infected lungs and found a 5-fold increase in neutrophils in Sp140 -/relative to Sp140 -/-Ifnar1 -/mice (Fig. 6A). Additionally, neutrophil depletion partially rescued the susceptibility of Sp140 -/mice at lower and higher bacterial burdens (Fig. 6B, 6C,  Supplementary Fig. 11). Given the importance of neutrophils, we assessed NET production in Sp140 -/and B6 mice by staining for citrullinated H3 in the lungs of Mtb-infected mice (Fig. 6D). Sp140 -/mice had over a 100-fold increase in NET staining as compared to B6 animals, indicating that the lungs of Sp140 -/mice harbor substantially more ligand to activate type I interferon production by pDCs as compared to wild-type hosts (Fig. 6E). As NETs are a source of nucleic acids, we hypothesized that pDCs would sense the NETs through endosomal TLRs. In line with this prediction, deletion of Unc93b1, which is an essential chaperone required for TLR7 and TLR9 function, partially rescued the Mtb susceptibility of Sp140 -/mice (Fig. 6G). By contrast, deletion of Ticam1, which encodes for TRIF, an adapter molecule critical for type I IFN production downstream of TLR3 and TLR4, had no effect on bacterial control in Sp140 -/mice (Fig. 6F). Together, these data suggest a model in which extracellular nucleic acid, potentially from NETs, is sensed by endosomal TLRs triggering pDC production of type I IFNs (Fig. 6H).

Neutrophils and IMs are the major sensors of type I IFNs during Mtb infection
Having identified pDCs, IMs, and monocytes as the major type I IFN producers during Mtb infection, we next sought to identify which cells responded to this type I IFN. As expected, IFNAR was uniformly expressed by all lung myeloid cells, and therefore not informative for identifying IFN responsive cells (Fig. 7A) 64 . However, comparing differentially expressed genes in B6 and Sp140 -/neutrophils and IMs showed a clear induction of IFN stimulated genes in cells from Sp140 -/animals (Fig. 7B). A major complication, however, is that many genes induced by type I IFN are also induced by IFNg, and most existing studies do not distinguish the two. Therefore, we sought to develop type I IFN-specific and IFNg-specific transcriptional signatures. RNA-sequencing analysis of human macrophages and mouse bone marrow-derived macrophages stimulated with IFNg, IFN-β, TNF, transforming growth factor-β, or nothing were used to define cytokine induced genes (Supplementary Fig. 12, 13) 65 . Similar gene families were preferentially upregulated by type I or II IFNs in human and mouse macrophages and the resulting gene signatures for type I or II IFN responsiveness were highly unique to stimulation with IFN-β or IFNg (Supplementary Fig. 12). We next applied the mouse gene signatures to the mouse lung myeloid scRNA-seq dataset. Strength of signature expression in naïve mice was used to define the threshold for classifying cells as an IFNg or type I IFN responder (Supplementary Fig. 13). As expected, naïve mice had very few cells responding to either cytokine, while bystander and Mtb-infected cells responded strongly to type I and 2 IFNs (Fig.  7C, Supplementary Fig. 13). Interestingly, the type I IFN response was limited to IMs and neutrophils, even though monocytes and AMs were responsive to IFNg. Potentially, differences in the localization of these cells could explain their differences in cytokine responsiveness. As expected, Mtb-infected neutrophils and IMs from Sp140 -/mice exhibited a significant increase in type I IFN signaling relative to cells from B6 lungs (Fig. 7D). Consistent with considerable prior work demonstrating that type I IFNs impair responsiveness to IFNg 16,32,33,66 , the Sp140 -/mice harbored a distinct population of IMs that exhibited the signature of type I IFNresponsiveness but lacked the signature of IFNg responsiveness (note the distinct population of blue IMs among the Mtb-infected cells in Fig. 7C, 7D). The reduction in IFNg signaling in Mtbinfected IMs in Sp140 -/mice correlated with reduced IFNg receptor 1 expression on these cells in Sp140 -/relative to B6 mice (Supplementary Fig. 14). This reduced IFNg receptor 1 expression also correlated with increased induction of type I IFN stimulated genes and reduced expression of type II IFN stimulated genes (Supplementary Fig. 14). Since IFNg is critical for control of intracellular Mtb replication, these results suggest that type I IFN might impair Mtb control at least in part by opening a niche of susceptible IMs that fail to respond to IFNg.
As neutrophils and IMs were the primary sensors of type I IFN, these are the cell types we tested to determine which cell population mediates the type I IFN-driven susceptibility of Sp140 -/mice. Neutrophil-specific deficiency in type I IFN signaling was insufficient to rescue the Mtb susceptibility or the increase in lung neutrophils exhibited by Sp140 -/mice (Fig. 7E,  Supplementary Fig. 15A). Conversely, deletion of Ifnar1 expression on myeloid cells with LysM Cre Ifnar1 fl/fl mice or specifically in macrophages with CD64 Cre Ifnar1 fl/fl mice rescued bacterial control and reduced lung neutrophil numbers to the same extent as global Ifnar1 deletion (Fig. 7F, Supplementary Fig. 15B). Taken together, these results are consistent with a model in which type I IFNs act on IMs to inhibit IFNg signaling in these cells, thereby reducing their ability to restrict Mtb growth.

Discussion
The dominant gene signature identified in humans with active tuberculosis disease is a type I IFN signature [7][8][9][10] . Type I IFNs are critical for effective anti-viral immunity 8 but considerable data from humans and animal models indicate type I IFNs impair Mtb control 7,9,10,67 . For example, a range of viral co-infections or chemical interferon inducers exacerbate Mtb disease in mice [17][18][19][20][21] .
Interestingly, although type I IFN and IFNg (also called type II IFN) induce a highly overlapping set of target genes (Supplementary Fig. 12), the ability of type I IFNs to promote Mtb disease is not seen with IFNg, which is instead potently protective against mycobacterial infections in mice 5 and humans [68][69][70] . Indeed, type I IFNs can exacerbate bacterial infections by directly antagonizing IFNg signaling 16,32,33,66 . The mechanism of antagonism is poorly understood but seems to be due at least in part to the downregulation of surface expression of the IFNg receptor 32,33 . We see evidence for this type I IFN induced downregulation of IFNg receptor during Mtb infection, as Mtb-harboring IMs express lower levels of IFNg receptor (and IFNγ target genes) in Sp140 -/mice, which produce a strong type I IFN response, relative to those in B6 mice, which produce a weak type I IFN response (Supplementary Fig. 14). The antagonism of IFNg signaling by type I IFNs is conserved in humans, and has been shown to be relevant during Mycobacterium leprae infections 16 . Potentially, antagonism of IFNg signaling underlies the tuberculosis-susceptibility of humans that exhibit a type I IFN signature, as humans with active Mtb exhibit reduced IFNg receptor expression on their monocytes 15 . Type I IFNs can also impair IL-1 signaling, an additional pathway that is critical for Mtb control, through the induction of IL-1 receptor antagonist and eicosanoid imbalance 26,30,71,72 . Thus, the antagonism of protective IFNγ and IL-1 responses by type I interferons may be a key driver of progression to active tuberculosis. There is support in the literature for this hypothesis, as infants in their first year of life that acquire cytomegalovirus infection, a virus known to induce type I IFN, have a high risk of developing tuberculosis disease 11,12,[73][74][75] . Additionally, influenza infection, which also induces type I IFN, correlates with an increased risk of death in pulmonary tuberculosis patients 13 . While the correlation between viral infection and exacerbated tuberculosis disease is tantalizing, a major limitation of studying human disease is the difficulty in establishing the underlying causal mechanisms that account for the correlation. Thus, we sought a genetically tractable model of Mtb infection in order to establish the cellular mechanisms by which type I IFN drives Mtb susceptibility.
The commonly used B6 mouse model does not exhibit a strong type I IFN response after Mtb infection 24,30,31 . Consistent with the modest type I IFN response of B6 mice, Ifnar1 deletion on the B6 genetic background does not reliably impact survival or bacterial burdens in the lungs after Mtb infection [22][23][24]30 . Therefore, we sought a different mouse model that recapitulated two key aspects of human disease: the hyper type I IFN response, and the accompanying neutrophilic inflammation 7,76,77 . Previously, we identified B6.Sst1 S mice as a mouse model that exhibits type I IFN-driven susceptibility to Mtb infection 30 . We then demonstrated that the absence of Sp140 in B6.Sst1 S mice explains their susceptibility to Mtb 31 . SP140 is a member of the Speckled Protein family of epigenetic readers and is widely expressed in leukocytes 78 . In humans, polymorphisms in SP140 have been linked to Crohn's disease 79,80 , chronic lymphocytic leukemia 81 , and multiple sclerosis 82 by GWAS, but there is no known association between polymorphisms in SP140 and susceptibility to Mtb. It has been suggested that macrophages in Sp140-deficient mice exhibit inherent defects in bacterial control causing them to be more susceptible to dextran sulfate sodium-induced colitis 39,40 . However, an inherent defect in bacterial control is not evident during Mtb infection, as Sp140 -/mice lacking IFNAR restrict Mtb as well as B6 animals 25 days after infection 31 , a time point at which macrophages are critical for Mtb control 38 . This result suggests that the early susceptibility of Sp140 -/mice is due to their strong type I IFN response rather an inherent defect in bacterial killing by macrophages. In addition to their hyper type I IFN response, Sp140 -/mice also have more lung neutrophils after Mtb infection than Sp140 -/-Ifnar1 -/animals ( Fig. 1D, 7B, 7D). Therefore, Sp140 -/mice recapitulate the fundamental type I IFN and neutrophilic character of human active Mtb disease, and can serve as a platform for understanding the cellular mechanism of type I IFN-driven Mtb susceptibility.
Sp140 -/mice provide a unique model of the aberrant type I IFN response, as they are on a pure B6 genetic background, and do not require repeated administration of TLR agonists, viral co-infection, or other perturbations of the innate immune system for type I IFN production 17,18,25,26,63 . Other groups have also modeled the type I IFN response by infecting B6 mice with a lineage 2 clinical Mtb strain, such as HN878 83,84 . However, Ifnar1 deletion had no impact on survival or bacterial control at early time points in B6 mice infected with HN878, unlike Sp140 -/mice infected with Mtb Erdman 31,85,86 . Thus, we believe the Sp140 -/mouse model uniquely recapitulates the hyper type I IFN response exhibited by humans, and provides a tool to study the mechanistic basis of the aberrant type I IFN response.
We sought to use the Sp140 -/mice to address two key questions: (1) what cells produce type I IFN during Mtb infection? and (2) which cells respond to the type I IFN to mediate Mtb disease? Flow cytometry and imaging of I-Tomcat Ai6 Ifnb1 reporter mice identified IMs and pDCs as the major IFN-β producers during Mtb infection. Imaging provided insight into why these cells expressed type I IFN, as the frequency of type I IFN-expressing macrophages was enriched relative to CD4 + T cells in diseased tissue but not in healthy tissue. This result suggests that proximity to Mtb dictates access to activating signals required to induce IFN-β expression by macrophages. However, most IFN-β expressing IMs were not infected with Mtb, and most infected IMs were not Ifnb1 or reporter positive, indicating direct infection is insufficient and may not be the main driver of type I IFN expression in vivo. In vitro studies have shown that bone marrow-derived macrophages infected with Mtb induce type I IFN via the cGAS-STING pathway. However, this pathway does not seem to be of great importance in vivo 30,[87][88][89][90] . Instead, our results suggest that uninfected bystander cells may be the primary producers of type I IFN during Mtb infection. Consistent with this hypothesis, we found that mice lacking Unc93b1, a chaperone required for TLRs that sense extracellular nucleic acids, exhibit enhanced control of Mtb. Exogenous nucleic acids sensed by uninfected bystander cells may thus be an important pathway for type I IFN induction during Mtb infection in vivo. Of note, pDCs, which we found to be important type I IFN producers during Mtb infection, are robust producers of type I IFN after TLR7/9 sensing of exogenous nucleic acids 58,91,92 . Imaging demonstrated that B6 and Sp140 -/macrophages expressed IFN-β at similar frequencies in diseased tissue. This result appears to contrast with flow cytometry analysis of whole lungs, which demonstrates a higher frequency of IMs expressing IFN-β in Sp140 -/mice relative to B6 mice. The apparent discrepancy between the flow cytometry and imaging data is likely explained by the fact that Sp140 -/mice have considerably more diseased tissue than B6 mice. Additionally, the Ai6 signal from I-Tomcat Ai6 mice identifies cells that have expressed IFN-β, but does not indicate the level of IFN-β expression in these cells. Thus, even if macrophages are driven to express IFN-β at similar frequencies in B6 and Sp140 -/mice, the macrophages in Sp140 -/mice could produce more IFN-β per cell, as seen in the scRNA-seq data (Fig. 3C).
Expression of type I IFNs by pDCs during Mtb infection was particularly noteworthy as limited work exists on the effect of pDCs on Mtb control. pDCs are believed to be important for control of viral infections, with their production of type I IFN during the early phases of viral infections significantly contributing to viral control 56,57,61 . During bacterial infections, the potential for pDCs to impact control has become appreciated more recently, with pDCs demonstrating a protective function against Citrobacter rodentium, Chlamydia pneumoniae, and Klebsiella pneumoniae [93][94][95][96] . However, few studies have examined the contribution of pDCs during Mtb infection, and these studies have not yet led to a clear understanding. The number of pDCs in the blood was reduced in Mtb-infected humans, but lung pDC numbers or function were not assessed 97,98 . In non-human primates, there was a correlation between active pulmonary Mtb and the influx of pDCs and IFN-responsive macrophages into the lungs of the rhesus macaques 51 . Another group also identified pDCs in tuberculosis granulomas of a different nonhuman primate, cynomolgus macaques, but the frequency of pDCs among all cells in the granulomas did not correlate with bacterial burden of the granuloma 99 . A major issue in the studies using NHPs or humans is the difficulty in depleting or otherwise functionally assessing the role of pDCs. To address this limitation, we generated Sp140 -/-pDC-DTR mice to study the contribution of pDCs to bacterial control. These mice are the most specific tool currently available for deletion of pDCs during Mtb infection 59,61 and demonstrated that pDCs (or pDC-like cells) contribute significantly to the susceptibility of Sp140 -/mice. Additionally, we identified pDCs in the lymphocytic cuff surrounding Mtb granulomas in human lungs and lymph nodes. Together, these results suggest that disease progression driven by pDC production of type I IFN drives disease in mice, and is likely conserved across non-human primates and humans.
While pDC depletion rescued Sp140 -/mice, it had no impact on bacterial burden in B6 animals. This result was expected given that very few myeloid cells in B6 mice expressed a type I IFN signature and Ifnar1 deficiency also has only modest effects in the B6 background 24,30 . The specific lack of an effect of pDC depletion in Sp140-sufficient mice also suggests that the role of pDCs in exacerbating Mtb infection is related to their production of type I IFNs instead of an IFN-independent function. As pDCs are present in B6 and Sp140 -/mice, we speculated that the difference in pDC type I IFN production in these mouse strains could be due in part to differences in the availability of activating ligands. As seen in another Mtb susceptible mouse model, and in Mtb-infected human lungs 63 , Sp140 -/mice had a significant enrichment in NET production compared to B6 mice. Potentially, these NETs, which are DNA-rich products of neutrophils, act as ligands for TLR9 on the pDCs, as described in mouse and human autoimmunity 34,35,62,100 . In support of this hypothesis, neutrophil depletion rescued the susceptibility of Sp140 -/mice, as did deletion of Unc93b1 which is required for the function of endosomal TLRs, including TLR9 which senses DNA. Given that NET formation was also identified in Mtb granulomas in human lung sections 63 , it is possible that pDC sensing of NETs contributes to the type I IFN response detected in humans with active Mtb disease.
Having defined the cells producing type I IFNs in vivo after Mtb infection, we then sought to identify which cells responded to the type I IFN. To do this, we had to first develop transcriptional signatures that distinguish the response to type I IFN from the closely related response to IFNg. Applying these signatures to our scRNA-seq data, we identified neutrophils and IMs as type I IFN sensors. While both cell populations harbor Mtb, it is possible that type I IFN driven susceptibility to Mtb is caused by signaling in only one of these populations. In a GM-CSF blockade model of type I IFN-driven Mtb susceptibility, neutrophil-specific deletion of Ifnar1 rescued bacterial control 63 . We employed the same genetic strategy as this prior report, but were unable to detect any rescue of Sp140 -/mice when neutrophils lacked Ifnar1. This was surprising as we saw that Sp140 -/neutrophils expressed a type I IFN gene signature by scRNA-seq. However, GM-CSF is critical for maintaining lung alveolar macrophages and the responsiveness of lung monocytes and macrophages to infections, including Mtb infection 101-104 . Therefore, it is possible that impairing lung macrophages by GM-CSF blockade shifted the impact of type I IFN on Mtb control from macrophages to neutrophils. In support of this idea, we also identified IMs as a type I IFN-sensing population, and deletion of Ifnar1 on macrophages rescued Sp140 -/mouse bacterial control. Our scRNA-seq dataset only contains myeloid cells, therefore we would miss the contribution of type I IFN signaling in other cell types. However, it is likely that type I IFN largely acts through myeloid cells to drive Mtb susceptibility, as macrophage-specific deletion of Ifnar1 rescued Mtb control to a similar extent as global Ifnar1 deficiency. These results suggest that during a normal (GM-CSF sufficient) response, type I IFN signaling in macrophages reduces their ability to restrict Mtb. The mechanism by which type I IFNs impair Mtb clearance by IMs remains unknown. However, an observation to emerge from our studies was that infected IMs in Sp140 -/mice exhibit a reduced transcriptional signature of responsiveness to IFNγ, and instead primarily exhibited a type I IFN signature, consistent with the exacerbated type I IFN response in these mice. The transcriptional response of Sp140 -/-IMs contrasted dramatically with that of infected IMs in B6 mice, which control Mtb and which exhibited a uniform signature of responsiveness to IFNγ (Figure 7). Given the essential role of IFNγ in control of Mtb in mice and humans, our results suggest that one detrimental effect of type I IFNs may potentially be via inhibition of IFNg signaling in the macrophages 15,32,33 .
Taken together, our results identify the cell types that produce and respond to type I IFNs during IFN-driven tuberculosis disease. We propose that type I IFNs impair responsiveness to IFNg, leading to an initial loss of bacterial control. Bacterial replication then leads to neutrophil influx and NET production within the diseased tissue. We propose that DNArich NETs are sensed by pDCs, likely via endosomal TLRs, causing them to produce type I IFN, which further antagonizes IFNg signaling and reduces the IMs ability to restrict Mtb growth. Given the correlations between our results and findings in rhesus macaques and humans with active Mtb, we believe that our proposed mechanism of type I IFN driven loss of Mtb control is conserved across species. These findings open the door for the development of therapies targeting NET production or pDC function as host-directed strategies for treating active Mtb infection.

Limitations of the Study
While we see a strong correlation between NET formation and type I IFN-driven Mtb susceptibility, the current study does not directly test the contribution of NETs in this response. Additionally, the present study provides data suggesting that type I IFN signaling correlates with a loss of type II IFN signaling in IMs during Mtb infection, but does not directly examine whether Mtb-harboring IMs in Sp140 -/mice are unable to control Mtb infection because of a lack of response to IFNg. We also limited our studies to ~25 days post-infection, which is an early time point for Mtb infection. It is possible that the cellular sources and targets of type I IFN shift to other cell types at later time points in the infection.

Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Russell Vance (rvance@berkeley.edu).

Materials availability
Materials used in this study will be provided upon request and available upon publication.

Data and code availability
• Raw and processed bulk RNA-and single cell RNA-sequencing data is deposited in the NCBI Gene Expression Omnibus: GSE216023, GSE232827, GSE232922. This paper also analyzes existing, publicly available data, for which the accession numbers are listed in the key resources table.
• Code for bulk RNA-and scRNA-sequencing analysis is available on Github: https://github.com/dmitrikotov/Sp140-Type-I-Inteferon. • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Animals
Mice were maintained under specific pathogen-free conditions and housed at 23°C with a 12 hour light-dark cycle in accordance with the regulatory standards of the University of California Berkeley Institutional Animal Care and Use Committee. All mice were sex-and age-matched and were 6-12 weeks old at the start of infections. Male and female mice were used in all experiments. Littermate controls were used when possible, as indicated in the figure legends.

Bacterial strains
Mtb Following electroporation, bacteria were grown on 7H11 plates supplemented with 10% oleic acid, albumin, dextrose, and catalase, 0.5% glycerol, and either 200 µg / mL Hygromycin for Mtb-Wasabi or 50 µg / mL Kanamycin for Mtb-mCherry for 3-4 weeks at 37°C. Individual colonies where then propagated in 10 mL inkwell flask cultures using 7H9 medium supplemented with 10% albumin-dextrose-saline, 0.4% glycerol, 0.05% Tween-80, and either Hygromycin for Mtb-Wasabi or Kanamycin for Mtb-mCherry for 7 days at 37°C. The inkwell cultures were expanded into a 100 mL culture using the same 7H9 supplemented media with antibiotics and cultured for 4-5 days at 37°C. Once the bacteria were in log phase, the culture was filtered with a 5 µm syringe filter and frozen in 1 mL aliquots in 10% glycerol.

Mtb infections
For infection, a frozen aliquot of Mtb-Wasabi or Mtb-mCherry was diluted in distilled H2O, and 9 mL of diluted culture was loaded into the nebulizer of a inhalation exposure system (Glas-Col, Terre Haute, IN) to deliver ~20-100 bacteria per mouse as determined by measuring CFU in lungs 1 day post-infection.

Tissue Processing for CFU and Flow cytometry
Mice were harvested at various days post-infection (as described in figure legends) to measure CFUs by plating and innate immune populations by flow cytometry. All lung lobes were harvested into a gentleMACS C tube (Miltenyi Biotec) containing 3 mL of RPMI media with 70 µg / mL of Liberase TM (Roche) and 30 µg / mL of Dnase I (Roche). Samples were processed into chunks using the lung_01 setting on the gentleMACS (Miltenyi Biotec) and incubated for 30 minutes at 37°C. Tissue was then homogenized into a single cell suspension by running the samples on the lung_02 setting on the gentleMACS. The digestion was quenched by adding 2 mL of PBS with 20% Newborn Calf Serum (Thermo Fisher Scientific) and filtered through 70 µm SmartStrainers (Miltenyi Biotec).
For measuring plasmacytoid dendritic cell numbers, spleens were harvested into a 12 well plate with 1 mL of PBS with 2% Newborn Calf Serum and 0.05% sodium azide in each well. The spleens were sandwiched between 100 uM mesh filters and mashed into a single cell suspension with the back of a syringe plunger. The single cell suspensions were filtered through 70 µm SmartStrainers (Miltenyi Biotec).

Measuring Bacterial Burden
To measure CFU, 50 µL was taken from each single cell suspension and then serially diluted in phosphate-buffered saline (PBS) with 0.05% Tween-80. Serial dilutions were plated on 7H11 plates supplemented with 10% oleic acid, albumin, dextrose, and catalase and 0.5% glycerol. Colonies were counted after 3 weeks.

Flow Cytometry
For flow cytometry, lung single cell suspensions were pelleted and resuspended in 500 µL of PBS with 2% Newborn Calf Serum and 0.05% Sodium azide and 100-150 µL were stained with antibodies for analysis. Spleen single cell suspensions were pelleted and resuspended in 5 mL of PBS with 2% Newborn Calf Serum and 0.05% Sodium azide, of which 50 µL were stained with antibodies. Single cell suspensions were stained for 45 minutes to an hour at room temperature with the following antibodies: TruStain FcX PLUS (S17011E, BioLegend), BUV496-

Immunostaining human lymph nodes and lungs
Human lung and lymph node samples were acquired from the surgical pathology archives of Emory University Hospital with appropriate institutional approval. 8 lung samples and 8 lymph node samples were analyzed. Each sample had been previously culture verified for Mycobacterium tuberculosis infection. The samples were formalin-fixed and paraffin-embedded. Sections were cut and stained with anti-CD123 (6h6, Thermo Fisher Scientific), anti-CD303 (124B3.13, Dendritics), or hematoxylin and eosin 112,113 . Primary antibodies were detected by immunoperoxidase staining with the LSAB+ System and a standard DAB reaction following manufacturer's instructions (DakoCytomation). Sections were counterstained with hematoxlyin prior to mounting and microscopy. pDCs were assessed in multiple 400× fields for each section to calculate the frequency of samples containing pDCs and the clustering of pDCs within each sample, defined as either singe cells, loose clusters of 5-20 cells, or tight clusters of more than 20 cells.

Image processing and histo-cytometry analysis
Image analysis was performed using Chrysalis software 55 . Briefly, a compensation matrix was generated by automatic image-based spectral measurements on single color-stained controls in ImageJ by using Generate Compensation Matrix script. This compensation matrix was used to perform linear unmixing on three-dimensional images with Chrysalis. Chrysalis was also used for further image processing, including rescaling data and generating new channels by performing mathematical operations using existing channels. For histo-cytometry analysis, Imaris 9.9.1 (Bitplane) was used for surface creation to digitally identify cells in images based on protein expression 54 . Statistics for the identified cells were exported from Imaris and then imported into FlowJo version 10 (BD Biosciences) for quantitative image analysis.

Bulk RNA-seq sample preparation and analysis
RNA-seq of Mtb-infected Sp140-sufficient and -deficient mouse lungs genetically depleted of pDCs was performed on 20% of each lung single cell suspension, prepared as described for CFU and flow cytometry analysis. Single cell suspensions were preserved in Trizol LS (Thermo Fisher Scientific) and stored at -80°C. The samples were thawed at room temperature for 5 minutes, then 200 µL of chloroform (Thermo Fisher Scientific) was added per 0.75 mL of Trizol LS to each sample and the samples were removed from the BSL3. Samples were centrifuged in Phasemaker tubes (Thermo Fisher Scientific) to isolate total RNA, which was then purified following the RNeasy Micro (Qiagen) protocol starting at the ethanol addition step. Library preparation, sequencing, and read alignment to the mouse genome was performed by Azenta Life Sciences. Libraries were prepared from total RNA using an Illumina kit for Poly(A) selection. Samples were sequenced on an Illumina HiSeq sequencer with paired 150 bp reads to a depth of 20-30 million reads per sample. Sequence reads were trimmed of adapter sequences and low quality nucleotides with Trimmomatic v.0.36 114 and then mapped to the Mus musculus GRCm38 reference genome with STAR aligner v.2.5.2b 115 . The raw counts were used as input for DESeq2 116 analysis of differential gene expression.
RNA-seq of cytokine stimulated bone marrow-derived macrophages was performed by differentiating bone marrow from B6 mice in DMEM supplemented with 10% FBS, PenStrep, Glutamin, HEPES, and M-CSF for 7 days then reseeding the cells in a 6-well plate and incubating for 5 days at 37°C and 5% CO2. The macrophages were left untreated or stimulated with 10 ng / mL of IFN-β (BioLegend), IFNg (Abcam), TNF (Peprotech), or transforming growth factor-β (BioLegend) for 6 hours. Cells were lysed with TRK lysis buffer (Omega Bio-Tek) and 2mercaptoethanol (Thermo Fisher Scientific) followed by total RNA isolation using the E.Z.N.A Total RNA Kit I (Omega Bio-Tek). The library preparation, sequencing, and read alignment to the mouse genome was performed by Azenta Life Sciences as described for the Mtb-infected mouse lung samples. Raw counts were used as input for analysis with DESeq2 116 . The scRNA-sequencing libraries were generated using the v3.1 chemistry Chromium Single Cell 3' Reagent Kit (10X Genomics) largely following the manufacturer protocol with the following minor modifications. Cells were loaded into 3 different lanes on a Chromium Next GEM Chip. Lane 1 was loaded with Mtb-infected cells from all 3 B6 and 3 Sp140 -/lungs. Lane 2 was loaded with bystander myeloid cells from the 3 infected B6 lungs as well as the myeloid cell mixture from the 2 naïve B6 lungs. Lane 3 was loaded with bystander myeloid cells from the 3 infected Sp140 -/lungs as well as the myeloid cell mixture from the 2 naïve Sp140 -/lungs. All 3 lanes of the Chromium Next GEM Chip were super-loaded with 29000 cells with a target of 14800 single cells per lane, as hashtag barcoding allows for a lower effective multiplet rate due to the ability to identify most of the multiplets (https://satijalab.org/costpercell/) 117 . 0.5 U/µL RNaseOUT Recombinant Ribonuclease Inhibitor (Invitrogen) was added to single cell RT master mix during the loading step and 1 µL of ADT and HTO additive primers (0.2 µM stock) were added during the cDNA amplification, as recommended by the CITE-seq and Cell Hashing Protocol (https://cite-seq.com/protocols/) 42 . Following cDNA, ADT, and HTO purification, samples were decontaminated by 2 rounds of centrifugation through 0.2 µM filter microcentrifuge tubes and then removed from the BSL3. Library preparations were completed outside of the BSL3 following the 10X Genomics protocol for the cDNA and the CITE-seq and Cell Hashing Protocol for the ADT and HTO libraries. Quality control of the libraries was performed with a Fragment Analyzer (Agilent). The mRNA, ADT, and HTO libraries were pooled at the following proportions: 85% mRNA, 9% ADT, and 6% HTO. Libraries were sequenced on a NovaSeq 6000 (Illumina) using two lanes of a S1 flow cell and the following cycles read 1 (28 cycles), i7 index (10 cycles), i5 index (10 cycles), read 2 (90 cycles).

ScRNA-seq: data processing.
Raw sequencing reads for the mRNA libraries were processed into raw count matrices with CellRanger version 4.0.0 (10X Genomics). The ADT and HTO libraries were processed into raw count matrices with CITE-Seq-Count version 1.4.3 (https://hoohm.github.io/CITE-seq-Count/) 118 . The raw counts for mRNA, ADT, and HTO were analyzed in R 119 via the RStudio integrated development environment with Seurat v4.1.1 43 using default settings for normalizing the data, finding variable features, and scaling the data. HTO demultiplexing was performed with the HTODemux function. Data was filtered to only include single cells with between 200 and 4500 genes and less than 5% mitochondrial reads. The resulting datasets were integrated together using 30 dimensions for the FindIntegrationAnchors function and 30 dimensions for the IntegrateData function. The data was then scaled and analyzed by PCA with 30 principal components followed by UMAP analysis with 30 dimensions. Clustering was performed by using 30 dimensions with the FindNeighbors function and a resolution of 0.8 for the FindClusters function.
To improve resolution for clustering innate immune cells, weighted nearest neighbor analysis was used to combine the protein data (ADTs) and the mRNA data when clustering cells. For this analysis, variable ADT features were identified and then normalized using centered log ratio transformation and a margin of 2. The normalized ADT data was then scaled and analyzed by PCA. The ADT and mRNA data was then combined with the FindMultiModalNeighbors function using 30 dimensions for the mRNA and 10 for the protein. The resulting dataset was analyzed by UMAP and clusters were identified with the FindClusters function using algorithm 3 and a resolution of 1.5 44 . Tidyverse 120 , EnhancedVolcano 121 , and various Seurat functions were used for plotting the scRNA-seq data.

Type I IFN and IFNg Gene Signature Analysis
For generating the type I IFN and IFNg gene signatures, we utilized a published RNA-seq dataset (GEO: GSE20251) of primary human macrophages that were unstimulated or stimulated with 10 ng/mL of TNF, IFNg, IFN-β, transforming growth factor-β, or other ligands for 24 hours and then processed for RNA-sequencing 65 . We also generated an RNA-seq dataset mouse bone marrow-derived macrophages left unstimulated or stimulated with 10 ng/mL of TNF, IFNg, IFN-β, or transforming growth factor-β for 6 hours. For both datasets, counts were normalized by DESeq2's median of ratios method and biological replicates were averaged to generate average normalized counts for each cytokine condition. The average normalized counts were used to calculate a max to second-max ratio for the 4 cytokine conditions to determine gene specificity. Next, gene expression was compared across all stimulation conditions and filtered to only include genes that were induced relative to the untreated cells by a log2 fold change of 1 by IFNg or IFN-β and to exclude genes induced by a log2 fold change of 1.5 by TNF or transforming growth factor-β. Using this filtered gene list, a ratio was calculated of the log2 fold change following IFNg stimulation to the log2 fold change following IFN-β stimulation. For the human dataset, IFN-β specific genes were defined as those having a fold change ratio < 0, a log2 fold change upon IFN-β stimulation > 1, an average normalized count following IFN-β stimulation > 2000 and a max to second-max ratio > 2.5. Human IFNg specific genes were defined as those with a fold change ratio > 1.5, a log2 fold change upon IFNg stimulation > 2, an average normalized count following IFNg stimulation > 1000 and a max to second-max ratio > 2.5. Mouse macrophage IFN-β specific genes were defined as those with a fold change ratio < 0.66, a log2 fold change upon IFN-β stimulation > 4, an average normalized count following IFN-β stimulation > 4000 and a max to second-max ratio > 3. Mouse IFNg specific genes were defined as those that had a fold change ratio > 1.5, a log2 fold change upon IFNg stimulation > 2, an average normalized count following IFNg stimulation > 500 and a max to second-max ratio > 3.
Human IFN-β and IFNg gene signatures were validated by examining their induction following human macrophage stimulation with a variety of cytokines, such as IL-4, IL-6 and IL-10, using the published RNA-seq dataset (GEO: GSE20251). The specificity of the mouse macrophage IFNg signature was validated by stimulating bone marrow-derived macrophages for 6 hours with 10 ng/mL of IFNg, IFN-β, or nothing and then examining CXCL9 expression by flow cytometric analysis of intracellular staining with PE-labeled anti-mouse CXCL9 antibody (MIG-2F5.5; BioLegend). The mouse gene signatures were then used to score cells in the mouse myeloid scRNA-seq dataset based on their gene expression with the UCell R package 122 .

QUANTIFICATION AND STATISTICAL ANALYSIS
Group sizes were informed by the results of preliminary experiments and power calculations. The number of animals in each figure is indicated in the legends as n = x mice per group. Statistical significance was determined using Prism (GraphPad) software for unpaired one-tailed or two-tailed Student t test when comparing two populations, one-way or two-way ANOVA tests with Tukey's or Sidak's multiple comparisons test when comparing multiple groups. Prism (GraphPad) was also used to calculate linear correlations and R 2 . R was used to calculate statistical significance for bulk RNA-and scRNA-sequencing datasets using the Wald test with multiple testing correction by the Benjamini and Hochberg method and the Wilcoxon Rank-Sum test with Bonferroni correction, respectively. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, n.s. = not significant. See figure legends for more information on statistical tests. experiments are shown. Statistical significance was calculated by one-way ANOVA with Tukey's multiple comparison test. **p < 0.01, ***p < 0.001, ****p < 0.0001.

Supplementary Tables
Supplementary Table 1 pDC Distribution Tissue Cases Cases with CD303 + cells (% total)