C3HeB/FeJ mice infected with M. tuberculosis via aerosol exposure developed necrotizing granulomatous lesions in the lungs at 12 weeks postinfection (p.i.) ( Supplementary Figure 1 ), which is consistent with previous reports (30–34). We collected lung lesions with necrotic granulomas and prepared single-cell suspensions. To optimize the resolution of cellular transcriptomics, cell suspensions were further separated using Ficoll-Paque density gradient centrifugation, as necrotizing granulomas are primarily composed of abundant neutrophils and dead cell debris (30, 33, 35, 36). Cells collected from the interface layers of Ficoll-Paque solution were subjected to scRNA-seq ( Figures 1A, B ). Among 30,159 cells isolated from necrotizing granulomatous lesions, we manually annotated 11 major cell types. These included phagocytotic cells such as neutrophils and macrophages; dendritic cells (DCs) including conventional DCs (cDCs) and plasmacytoid DCs (pDCs); T cells including αβ T cell and γδ T cell subsets; natural killer (NK) cells; and B cells.

Next, we assessed the subclusters of T cells within the necrotizing granulomatous lesions ( Figure 2 , Supplementary Figure 2 ). Among CD4⁺ T cells, three clusters were identified: effector T cells, naïve T cells, and regulatory T cells (Tregs). For CD8⁺ T cells, two clusters were observed: cytotoxic and exhausted CD8⁺ T cells. Furthermore, we identified tissue-resident memory T cells (TRM) either expressing CD4 or CD8 within necrotizing granulomas.

During M. tuberculosis infection, γδ T cells participate in the immune response in the lungs (37). In particular, IL-17-mediated immunity is critical for γδ T cells to perform their function in host defense in TB murine models (38). In necrotizing granulomas, γδ T cells expressing Il17a were identified ( Figures 2C, D ). These γδ T cells also expressed Pdcd1, which encodes programmed cell death-1 (PD-1), an immune checkpoint molecule inhibiting immune responses (39).

Among other T cell types, we identified a unique population co-expressing Tcrg (T cell receptor gamma) and Klrd1 (encoding an NK cell receptor) ( Figures 2B, C ). This population also expressed Gata3, Tbx21, Il7r, Cd7, and Xcl1 ( Supplementary Figures 2A, C ), suggesting the presence of certain innate lymphoid cells (ILCs) expressing these markers.

Following the removal of doublet cells from the scRNA-seq data, we observed clusters expressing both T cell markers (such as CD3e) and characteristic genes of neutrophils or macrophages ( Figure 2B , Supplementary Figure 2C ). This observation may reflect biological processes such as the phagocytosis of T cell-derived material; however, further analyses are required to confirm this phenomenon.

To investigate the hallmark of macrophage populations within necrotizing granulomas, we analyzed their gene expression profiles. Given that Plin2 expression is significantly upregulated in foamy macrophages within necrotizing granulomas (29, 30), we identified the cluster exhibiting significantly elevated Plin2 expression compared to other clusters ( Figure 3A ). To characterize the remaining clusters, we analyzed their gene expression profiles and found that macrophages expressing Cxcl10, Clec4e, C1qc, Ighm, S100a9, or Siglecf also represented distinct macrophage populations within the necrotizing granulomatous lesions ( Figure 3A ). Moreover, we confirmed that the expression levels of the corresponding genes in their respective clusters were significantly higher than those in other clusters ( Figure 3B ).

We investigated the differentially expressed genes between the Plin2-expressing (Plin2 ⁺) cluster and other clusters within the macrophage population. Genes that were significantly upregulated in the Plin2⁺ cluster were selected ( Supplementary Table 1 ) and subjected to Gene Ontology (GO) enrichment analysis ( Figure 3C ). Genes related to fat digestion and absorption, as well as lipid and atherosclerosis, were found to be upregulated. To further investigate the gene expression profile of Plin2 ⁺ cluster, we performed gene set enrichment analysis (GSEA) ( Figure 3D ). GSEA revealed that genes associated with hypoxia, TNF-α signaling, glycolysis, and inflammatory responses were upregulated, whereas those associated with oxidative phosphorylation and type II interferon (IFN) responses were downregulated in the Plin2 ⁺ cluster compared to other macrophage clusters.

We investigated the gene expression profiles of additional macrophage clusters within necrotizing granulomatous lesions using GSEA ( Supplementary Figure 3 ). Clusters expressing Cxcl10 or Clec4e exhibited gene expression signatures characteristic of pro-inflammatory macrophages. In contrast, clusters expressing C1qc, Ighm, or Siglecf showed gene expression profiles associated with anti-inflammatory macrophages. A more detailed analysis revealed that Ighm ⁺ macrophages expressed several anti-inflammatory genes, including Cd244a, Nr4a1, Clec4a1, and Clec4a2 ( Supplementary Figure 4A ). Since Siglecf is a well-established marker of alveolar macrophages (40), the Siglecf ⁺ cluster was identified as alveolar macrophages. Siglecf-expressing macrophages expressed Alox5, a gene that promotes anti-inflammatory polarization and contributes to increased susceptibility to M. tuberculosis infection (41, 42) ( Supplementary Figure 4B ). Trem2 has been shown to act as a receptor for mycolic acid from mycobacteria and to limit anti-mycobacterial macrophage activation (43). The expression of Trem2 was mainly observed in the Siglecf ⁺ cluster ( Supplementary Figure 4B ). These results suggest that macrophages in the Siglecf ⁺ cluster may contribute to a permissive environment for M. tuberculosis infection within necrotizing granulomatous lesions.

Nos2 and Arg1, well-established markers of macrophage polarization, play key roles in regulating immune responses (44, 45). We have previously demonstrated that Plin2-expressing macrophages express Nos2 or Arg1 in necrotizing granulomas (30), suggesting that the polarization of foamy macrophages exhibits either pro-inflammatory or anti-inflammatory characteristics. Accordingly, scRNA-seq profiling revealed that Nos2 was highly expressed in pro-inflammatory macrophage clusters including the Plin2 ⁺ cluster ( Figure 4A ). Moreover, Arg1 expression was predominantly observed in the Plin2 ⁺ cluster. Feature plots revealed the presence of Plin2 ⁺ macrophages co-expressing Nos2 and Arg1 ( Figure 4B ). Further, we investigated the transcriptional factors associated with the regulation of Nos2 or Arg1 expression in macrophages within necrotizing granulomas. Among them, we found that Irf7 expression was correlated with Nos2 expression patterns in macrophage clusters ( Figure 4A ). Irf7 regulates macrophage polarization toward both pro-inflammatory and anti-inflammatory states (46). Fosl1 encodes a transcription factor that promotes pro-inflammatory polarization by repressing Arg1 expression (47). We observed Fosl1 expression in a subset of Arg1 ⁺ macrophages within the Plin2 ⁺ cluster ( Figure 4C ), suggesting that Plin2 ⁺ macrophages may consist of heterogeneous subpopulations with distinct or transitional polarization states. These findings suggest that Irf7 and Fosl1 may contribute to the dynamic regulation of macrophage polarization within necrotizing granulomas.

We further analyzed polarization states of macrophages derived from myeloid cell lineages in the whole lungs of M. tuberculosis-infected Sp140 knockout (KO) mice, using the data from Kotov et al. (28). In C3HeB/FeJ mice, the reduced gene expression of both Sp110 and Sp140 in Sst1 locus contributes to increased susceptibility to M. tuberculosis infection (48–50). In addition, similar to C3HeB/FeJ mice, Sp140 KO mice also exhibit increased susceptibility to M. tuberculosis infection, indicating that Sp140 is a key determinant of host vulnerability to the infection (51). Macrophage populations derived from M. tuberculosis-infected Sp140 KO mice were re-clustered based on the original annotations ( Figure 5A ). Subsequently, Plin2 expression was found to be elevated in the cluster of interferon-stimulated gene-positive (ISG⁺) interstitial macrophages (IMs) compared to other macrophage clusters. This result suggests that ISG⁺ IMs correspond to the Plin2 ⁺ cluster identified in our study ( Figure 5B ). This was further supported by a comparative analysis of macrophage populations between C3HeB/FeJ mice and Sp140 KO mice ( Supplementary Figures 5A, B ). Notably, Nos2 was expressed in ISG⁺ IM and IM populations, whereas Arg1 was dominantly expressed in ISGs⁺ IMs ( Figure 5B , Supplementary Figure 5C ). Moreover, macrophages co-expressing Nos2 and Arg1 were observed in a subset of the ISG⁺ IM cluster ( Figure 5C ). These results suggest the emergency of a novel macrophage polarization state in ISG ⁺ IMs derived from M. tuberculosis-infected Sp140 KO mice, corresponding to the Plin2 ⁺ cluster of C3HeB/FeJ mice identified in our study.

Type I IFNs and neutrophils contribute to TB exacerbation (35, 52). Accordingly, we investigated the expression of Ifnb1, a type I IFN gene, and Cxcl1, a chemokine gene involved in neutrophil recruitment, in macrophage populations ( Figure 6 ). Both genes were specifically expressed in the Plin2 ⁺ cluster, consistent with previous reports (28, 35).

We investigated novel gene expression signatures characteristic of the Plin2 ⁺ cluster within necrotizing granulomas. Among the genes expressed in the Plin2 ⁺ cluster, Flrt2, Hyal1, and Mmp13 were selected for further examination, due to their specific and enriched expression patterns ( Figures 7A, B ). IHC revealed that these proteins were localized to the rim regions of necrotizing granulomas, consistent with Plin2 localization ( Figure 7C ). Immunofluorescence microscopy (IFM) showed co-localization of all the three proteins with Plin2⁺ cells. Notably, the fluorescent signals of Flrt2, Hyal1, and Mmp13 displayed distinct subcellular localization patterns, whereas Plin2 was predominantly localized to the cytosolic regions of the same vacuolated cells ( Figure 7D ). These results suggest that Flrt2, Hyal1, and Mmp13 are novel molecular markers characterizing Plin2 ⁺ macrophages in necrotizing granulomas.

pDCs regulate viral infections by producing a large amount of type I IFNs (53). In the context of M. tuberculosis infection, the depletion of pDCs results in increased bacterial burdens in the lungs (28). Lee et al. demonstrated that M. tuberculosis-infected neutrophils can stimulate pDCs to produce type I IFNs (54). Therefore, we assessed the expression levels of Ifnb1 in pDCs ( Figures 8A, B ). However, Ifnb1 expression were predominantly detected in macrophages rather than pDCs. In contrast, pDCs expressed Il34 and Kmo. Il34 encodes an anti-inflammatory cytokine that regulates the expression of pro-inflammatory cytokines and promotes macrophage polarization toward an anti-inflammatory phenotype (55). Kmo encodes kynurenine 3-monooxygenase, which catalyzes the conversion of kynurenine to 3-hydroxykynurenine (3-HK), a metabolite inducing cellular damage and apoptosis via oxidative stress (56). These results suggest that pDCs are localized to necrotizing granulomatous lesions and contribute to the pathogenicity of M. tuberculosis infection via the expression of immunomodulatory and cytotoxic factors.

All animal experiments in this study were approved by the Animal Care and Use Committee of The Research Institute of Tuberculosis (RIT) (permission number ID 2022-02) and conducted in accordance with the RIT ethical guidelines for animal care and use.

C3HeB/FeJ mice were purchased from Jackson Laboratory and housed in a filtered-air, laminar-flow cabinet under specific pathogen-free conditions at the animal facility of the RIT. Mice were provided with sterile bedding, water, and mouse chow. Specific pathogen-free status was verified by monitoring sentinel mice housed within the colony. Mice aged 6–10 weeks were transferred to the biosafety level III animal facility of the RIT. For M. tuberculosis infection, frozen stocks of the M. tuberculosis Erdman strain stored at -80°C was used as previously described (65). Mice were infected with approximately 100 CFU of M. tuberculosis bacilli via aerosol using an infection exposure system (Glas-Col).

At 12 weeks p.i., infected mice were euthanized by exsanguination under anesthesia with 0.75 mg/kg medetomidine, 4.0 mg/kg midazolam, and 5.0 mg/kg butorphanol via the intraperitoneal route. The lungs were excised and subsequently dissected to collect infected lesions including necrotizing granulomas ( Supplementary Figure 1 ). For the preparation of single-cell suspensions, infected lungs from three or four mice were pooled to collect more than 10 lesions containing necrotizing granulomas. Necrotizing granulomatous lesions were minced and incubated in a collagenase/hyaluronidase/DNase I solution (Stemcell) with RNase inhibitor at 0.2 U/μL. Following red blood cell lysis, the dissociated cells were resuspended in PBS containing 2% fetal bovine serum and RNase inhibitor at 0.2 U/μL. To remove dead cells and a large proportion of neutrophils, cell suspensions were subjected to Ficoll-Paque density gradient centrifugation, followed by the collection of cells from the interface layer according to the manufacturer’s instructions (Cytiva).

Isolated cell suspensions from necrotizing granulomatous lesions were subjected to scRNA-seq library constructions using Chromium Fixed RNA Profiling Reagent Kit (10x Genomics). Cells were fixed with a fixation solution containing formaldehyde for 24 h at 4°C. Inactivation of M. tuberculosis in the fixed samples was confirmed by CFU assay. Fixed cells were further processed to construct scRNA-seq libraries according to the manufacturer’s protocol. The resulting libraries were sequenced on NextSeq 1000 (Illumina).

Raw sequencing reads were aligned against the mouse reference genome (mm10) using Cellranger version 7.0.1 (10x Genomics). Subsequent analyses were performed in R version 4.4.1 using the Seurat package version 5.2.1 (66). Data from four independent samples were filtered to include cells with 500–6000 genes and less than 20% mitochondrial reads. Data were normalized using the NormalizeData function, followed by the identification of highly variable features using the FindVariableFeatures function with the variance stabilizing transformation method, selecting the top 2000 most variable genes. Principal component analysis (PCA) was performed using the RunPCA function. To correct for batch effects, the Harmony algorithm was applied using the RunHarmony function. Following batch correction, uniform manifold approximation and projection (UMAP) was computed using the first 30 dimensions. A shared nearest neighbor (SNN) graph was then constructed using the FindNeighbors function with the same set of dimensions. Clustering was performed using the FindClusters function with a resolution parameter set to 0.7. The resulting dataset was processed with scDblFinder version 1.18.0 (67) to remove doublet cells. The filtrated data were reprocessed through data scaling, PCA, UMAP, followed by the construction of SNN graph and clustering. Cell clusters were manually annotated based on specific gene expression patterns of the respective cell types.

Differential expression analysis was performed using Seurat FindMarkers function. Gene expression analysis, GO analysis, and GSEA were performed using Nebulosa version 1.14.0 (68), ShinyGO version 0.82 (69), and, fGSEA version 1.30.00 (70), respectively.

IHC and IFM were performed as previously described (30, 36). Antibodies used in this study are listed in Supplementary Table 2 . IHC and IFM samples were visualized using a NanoZoomer S60 (Hamamatsu Photonics) and a FV4000 (Evident), respectively.