Immune cell landscape in sarcoid and non-sarcoid skin granulomas. We collected 4-mm skin biopsies (affected and unaffected skin) from 18 patients with sarcoid and 10 with non-sarcoid skin granulomas with clinically active and histologically validated skin disease (Figure 1A and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/JCI178711DS1). Unaffected skin biopsies from the same donor served as control samples. Non-sarcoidosis patients were diagnosed with granuloma annulare, xanthogranuloma, multicentric reticulohistiocytosis, annular elastolytic giant cell granuloma, or rubella virus–associated skin granulomatous dermatitis (Supplemental Table 1). In our sarcoidosis cohort, we had 9 male and 9 female patients, with a mean age of 54 years; 56% of the patients were Black/African American. Most patients exhibited disease in multiple organs (Supplemental Table 1). In our non-sarcoidosis cohort, we had 3 male and 7 female patients, with a mean age of 68 years. We generated 492,200 high-quality single-cell RNA-sequencing (scRNA-seq) profiles (Supplemental Table 2). Unsupervised clustering of scRNA-seq profiles identified 59 cell clusters, which were annotated to 10 cell types based on marker gene identification, lineage marker genes, and mapping to single-cell databases (Figure 1B, Supplemental Figure 1, B–E, and Supplemental Table 3). The identified cell types were shared among affected and unaffected samples (Supplemental Figure 1, B–F).

Immune cell landscape in sarcoidosis and non-sarcoidosis skin granulomas. (A) Overview of sample collection. (B) Identification of cell clusters from patients with sarcoidosis (n = 18 for affected and n = 18 unaffected skin) and non-sarcoidosis skin granuloma patients (n = 10 for affected and n = 9 unaffected skin). VE, vascular endothelium; LE, lymphatic endothelium. (C) UMAP depicting subclustering of immune cells. pDC, plasmacytoid DC; cDC, conventional DC. (D) Marker genes defining immune subsets. Dot size reflects percentage cells expressing the gene, and color illustrates level of gene expression. (E) Box-and-whisker plot shows relative contribution of immune cells as percentage of total cells. In the box-and-whisker plot, the box extends from the 25th to 75th percentile. The line in the middle of the box represents the median, and the whiskers represent the minimum and maximum. All data points are covered, no outlying values. Statistical significance was calculated using a 2-tailed Student’s t test. (F) Dot plot depicting gene activation in different immune clusters. Dot size reflects percentage cells expressing the gene, and color illustrates level of gene expression. Data depicted as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001.

We tracked the immune cell contributions to non-sarcoid and sarcoid granulomas. We subclustered the immune cell populations and identified 19 different subtypes (Figure 1C, marker genes for each subtype are listed in Figure 1D, Supplemental Figure 2, A–D, and Supplemental Table 3). Within the myeloid lineage, we identified 4 distinct subsets of macrophages in all samples, with Mac 1 and Mac 2 being the dominant populations. Mac 1 adopted an inflammatory phenotype, including expression of inflammatory marker genes: CCL3, CSTB, FN1, and LYZ. Mac 2 exhibited an antiinflammatory phenotype, including expression of antiinflammatory marker genes: CD163, MRC1, MS4A7, and SELENOP. Both sarcoidosis- and non–sarcoidosis-affected skin recruited more myeloid cells compared with unaffected skin (Figure 1E). However, their activation states differed. Mac 1 and Mac 2 in non-sarcoid granulomas maintained inflammatory and antiinflammatory activation states, respectively (Figure 1F, green rows). In contrast, both Mac 1 and Mac 2 in sarcoid granulomas adopted an inflammatory state with sarcoidosis-specific gene activation, including CHIT1, PTGDS, and PLA2G7. PLA2G7-expressing macrophages have been shown to be highly immunosuppressive and impede T cell activation (18). Taken together, the results show that macrophages are recruited to similar levels in sarcoidosis and non-sarcoidosis skin granulomatous diseases, but exhibit different activation states. To be comprehensive, we also identified a population of mature DCs expressing LAMP3 (Figure 1, D and E).

Within the lymphoid lineage, we identified 6 subtypes of T cells, B cells, and NK cells (Figure 1D and Supplemental Figure 2, C and D). The T subtypes were CD8+ cytotoxic (Tc) cells, central memory (Tcm) cells, follicular helper-like (Tfh-like) cells, CD4+ Th1 cells, Th17.1 cells, and regulatory (Treg) cells (Figure 1D and Supplemental Figure 2, C and D). Within sarcoidosis-affected skin, Th1 and Th17.1 T cell populations were the dominant subtypes; Th17.1 cells produced IFNG, but not IL17A (Supplemental Figure 4A) (19, 20). Overall, sarcoidosis-affected skin recruited more lymphoid cells across all subtypes compared with non–sarcoidosis-affected skin (Figure 1E and Supplemental Figure 3). Th1 and Th17.1 cells in sarcoidosis-affected skin induced NF-κB–associated inflammatory genes, including LTB, TNF, CCR6, and IL6. Similar to myeloid cells, sarcoidosis-activated lymphoid cells adopted a more inflammatory phenotype compared with non–sarcoidosis-activated lymphoid cells (Figure 1F). Surprisingly, lymphoid cells in non–sarcoidosis-affected skin specifically induced immune checkpoint genes, including TIGIT, LAG3, and PDCD1. LAG3 and PDCD1 are well known for negatively regulating T cell expansion and their effector functions such as cytokine secretion (21, 22). Thus, the well-known paradoxical immune response in sarcoidosis, characterized by both T cell anergy and T cell expansion, can potentially be explained by the involvement of immunosuppressive macrophages coupled with the lack of immune checkpoint gene induction in T cells. Finally, sarcoidosis-affected skin was enriched in B cells that expressed CDA and BCL6, 2 genes typically induced during late-stage changes in germinal center differentiation and isotype switching (Supplemental Figure 4B) (23, 24).

Sarcoid granulomas specifically recruit ILC1s. Next, we wanted to take advantage of our unbiased approach and identify sarcoidosis-specific immune cell populations. Within affected sarcoid skin, we identified a population matching ILCs by using previously identified marker genes (HSPA1A, HSPA1B, IFNG, TNF, CXCR4, and TNFSF11) (Figure 1D, Supplemental Figure 2C, and Supplemental Figure 3) (25, 26). Differential gene expression analysis also confirmed the absence of T cell receptor (TCR) genes in this ILC cluster (TRA, CD3E, and CD247) (Supplemental Tables 3–5). 3D uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbor embedding (tSNE) clustering further confirmed that this population was distinct from other T cell populations (Supplemental Figure 4, C and D). Affected sarcoid skin contained on average approximately 5-fold more ILCs compared with non–sarcoidosis-affected skin (Figure 1E).

We confirmed the identity of this ILC population using 4 additional methods. (a) Due to lack of a human skin immune cell reference atlas, we overlaid the gene expression profiles from our identified ILC subcluster with the mouse Immunological Genome Project (ImmGen) database (https://www.immgen.org/ Accessed April 30, 2023.), and this expression pattern matched their ILC profile (Figure 2A). (b) Given the mouse-human comparison, we also compared our gene expression profiles against published human data sets of flow cytometry–purified ILC subtypes (27). Our sarcoidosis ILCs most closely matched the ILC1 gene profile (Figure 2B). (c) We adapted a published flow cytometry gating strategy for all individual ILC subtypes, including ILC1, ILC2, ILC3, and NK cells (12, 28, 29) (Figure 2C). We performed flow cytometry on affected and unaffected skin from 4 patients with sarcoid and 4 with non-sarcoid skin granulomas. Affected sarcoid skin contained approximately 10-fold more ILCs, specifically ILC1s, compared with non–sarcoidosis-affected skin (Figure 2D). Non–sarcoidosis-affected skin did not display any enrichment for ILCs. (d) Finally, we used the same flow cytometry gating scheme to purify ILC1s from affected sarcoid skin and performed bulk RNA-seq (Supplemental Figure 4E). Using this data set as a reference, 72% of the top 100 genes within the ILC population identified in our scRNA-seq data set overlapped, including 8 of the top 10 expressed genes (Figure 2E). Taken together, our results show that sarcoidosis-affected skin specifically recruits more ILC1s.

Group 1 ILCs are specifically recruited to sarcoid granulomas. (A) Gene expression profile of single-cell-identified ILC cluster best matched the ILC definition in the Immunological Genome Project (ImmGen) database. (B) Dot plot comparing gene expression profiles of sarcoidosis immune cells and ILCs to published purified ILC subpopulations. Sarcoidosis ILCs most closely matched ILC1. Dot size reflects percentage cells expressing the gene, and color illustrates level of gene expression. (C) Flow cytometry gating scheme for human ILCs, including NK cells. (D) Flow cytometry analysis of ILC subtypes as percentage of total sorted cells in affected and unaffected sarcoidosis and non-sarcoidosis granuloma skin (n = 3). One-way ANOVA revealed statistical significance for ILC1. *P < 0.05. (E) Comparison of gene expression from flow cytometry–purified sarcoidosis skin ILC1s (n = 3) to sarcoidosis ILCs identified in our scRNA-seq data sets. Eight out of top 10 genes matched and are listed.

Two other groups have generated scRNA-seq data sets from sarcoidosis-affected skin, and we found similar enrichment of ILCs in their data sets (Supplemental Figure 4, F–H) (30, 31). ILCs have been previously implicated in other inflammatory skin diseases, including psoriasis and atopic dermatitis. We compared ILCs from our sarcoidosis and non-sarcoidosis samples with ILCs identified from 6 other skin inflammatory conditions (32). ILCs from sarcoidosis-affected skin formed a distinct cluster, underscoring their unique nature and distinguishing them from ILCs in other skin inflammatory conditions (Supplemental Figure 5).

In summary, sarcoid and non-sarcoid skin granulomas exhibit disease-specific immune cell recruitment. Non-sarcoid granulomas displayed a macrophage-dominant response, and sarcoid granulomas exhibited a more complex immune response with Th1 cells, B cells, and ILC1s.

Spatial organization of sarcoid and non-sarcoid granulomas. We next assessed the spatial organization of B cells and ILCs within sarcoid skin granulomas using spatial transcriptomics and immunohistology. Visium spatial transcriptomics (10× Genomics) quantifies RNA transcripts within a 55-μm spot size, and sarcoid granulomas are often 200–400 μm in size, which enables determining whether different immune cell populations are located at tissue granulomas. Spatial transcriptomes generated from affected and unaffected skin from 2 patients with sarcoidosis and 1 non-sarcoidosis patient yielded 9,272 spots at an average depth of approximately 1,159 spots per sample and approximately 1,127 genes/spot (Supplemental Table 2). These patients were part of our skin scRNA-seq cohort. We performed UMAP dimensionality reduction and unsupervised clustering to identify 10 cell clusters. These clusters were annotated to cell types based on marker gene identification, lineage marker genes, and mapping of spot gene signature to single-cell databases (Supplemental Figure 6, A and B, and Supplemental Table 3).

We performed spatial deconvolution of individual spots using gene signatures derived from our scRNA-seq data to investigate the distribution of immune cell populations (33). Sarcoid granulomas specifically contained ILCs, B cells, Th1 cells, Th17.1 cells, and mature DCs (Figure 3A and Supplemental Figure 6C). We noted that transcripts of mature differentiated B cell markers, CR2, FCER2, and AICDA, localized specifically within sarcoid granulomas (Supplemental Figure 6D). In contrast, non-sarcoid granulomas exhibited fewer T cells, lacked ILCs completely, and did not express mature differentiated B cell markers (Figure 3A and Supplemental Figure 6D). We performed ligand-receptor analysis using our spatial data sets. Non-sarcoid granulomas exhibited osteopontin (also known as SPP1) signaling as the major interaction, and sarcoid granulomas demonstrated prominent CXCL12/CXCR4 and CCL19/CCR7 signaling, which are 2 important signaling pathways promoting TLS formation (Figure 3, B and C, and Supplemental Figure 6D). Indeed, CCL19, CCR7, and CXCR4 transcripts all localized within sarcoid granulomas (Supplemental Figure 6D).

Sarcoid granulomas contain ILC1s and B cell aggregates. (A) Deconvoluted cell type identification from spatial transcriptomics of patients with sarcoidosis (n = 2) and non-sarcoidosis granuloma (n = 1) patients. Each spot is represented as a pie chart displaying relative cell proportions. The middle panel highlights individual immune cell populations and the right panel highlights ILCs specifically. (B and C) Ligand-receptor analysis of spatial transcriptomics data sets for patients with sarcoidosis and non-sarcoidosis patients. Color represents signaling intensity. (D) Representative immunohistochemistry (n = 7 patients) depicting localization of B cells (CD20, PAX5), T cells (CD3), and macrophages (CD68) in sarcoidosis-affected skin. Dotted box outlines sarcoid granuloma. (E) Representative histology (n = 3 patients) depicting localization of mature germinal center–like B cells (CD3–CD20+CD23+, white arrows). Scale bars: 50 μm and 5 μm (yellow insets). (F) Representative histology (n = 3 patients) depicting localization of ILC1 (Lin–IL7R+Tbet+, white arrowheads). Lineage– = CD3–CD16–CD19–CD20–CD56–CD68– (labeled in green). Scale bars: 100 μm and 10 μm (insets).

To complement our spatial transcriptomics data sets, we performed protein immunofluorescence on sarcoidosis-affected skin samples taken from 7 patients (4 patients from our scRNA-seq cohort) (Figure 3D and Supplemental Figure 7). Sarcoid granulomas from all 7 patients displayed distinct aggregates of B cells (Pax5+ or CD20+) that colocalized with T cells (CD3+) and were distinct from macrophages (CD68+ or CD163+) (Figure 3D and Supplemental Figure 7A). Moreover, many of these B cells were also positive for CD23, an established marker of B cell and TLS maturation (Figure 3E and Supplemental Figure 7B) (34). Finally, ILCs (Lineage–CD127+Tbet+) localized to the periphery of sarcoid granulomas (Figure 3F). Taken together, these results show that B cells and ILCs localized to sarcoid skin granulomas to form mature TLSs.

Sarcoid granulomas exhibit molecular features resembling mature TLSs. Next, we performed ligand-receptor analysis on our non-sarcoidosis and sarcoidosis single-cell data sets to study the cross-talk among immune cells. Globally, non-sarcoid skin granulomas were dominated by SPP1 signaling, and sarcoid granulomas were enriched in TNF and CCL family signaling, including B cell–activating factor (BAFF), lymphotoxin (LT), LIGHT (TNFSF14), and CCL19 (Figure 4A and Supplemental Figure 8A). Cell-specific ligand-receptor analysis revealed that most, if not all, interactions within non-sarcoid granulomas were conserved in sarcoid granulomas, but the reverse was not observed (Supplemental Figure 8, B and C, and Supplemental Table 6). We subtracted the common interaction pairs and focused on 5 sarcoidosis-specific ligand-receptor interactions that promote TLS formation (Figure 4B). (a) A subpopulation of fibroblasts and mature DCs both induced CCL19, while mature DCs and B cells both received the signal via CCR7 (Figure 4, B and C, and Supplemental Figure 8D). (b) Notably, the same subpopulation of fibroblasts also expressed CXCL13, which was received via CXCR5 on B cells specifically (Figure 4, B–D, and Supplemental Figure 8, D and E). (c) ILC1s strongly induced LIGHT and LTB, individual cytokines previously shown to be sufficient to induce TLS formation (35). LIGHT signals through TNFRSF14 expressed on mature DCs, Mac 1, and Mac 2 cells, and Mac 1 cells induced IL1B (Figure 4, B and C). The same subpopulation of fibroblasts received both IL1B and LTB signals. (d) In addition to LIGHT, sarcoidosis-associated ILC1s also expressed high levels of CD40L, BAFF, and ICOS, which are well-established B cell–activating cytokines (Figure 4E). In fact, prior work demonstrated that ILCs directly activate B cells through BAFF and CD40L (36). Sarcoidosis-associated ILC1s also expressed CCL20, which may signal to T helper and B cells through CCR6. (e) Finally, sarcoidosis-associated B cells expressed MS4A1 (also known as CD20), FCER2 (also known as CD23), BCL6, and AICDA, which are all markers of late-stage differentiated B cells in mature TLSs (37) (Figure 4F and Supplemental Figure 4B).

Sarcoid granulomas exhibit molecular features resembling mature tertiary lymphoid structures. (A) Global analysis of ligand-receptor pathways between sarcoid and non-sarcoid granulomas. Arrows highlight TNF family and CCL signaling. (B) Cell-specific ligand-receptor analysis. Bottom half of circle depicts secreting cell types, and upper half of circle depicts receiving cell types. Inner bottom circle is a summary of the receiving cell types. (C) Dot plot demonstrating average expression (color) and percentage of cells (dot size) expressing specific cytokines. (D) Volcano plot of differential gene expression for sarcoidosis and non-sarcoidosis fibroblasts depicting increased CCL19 and CXCL13 expression. (E) Volcano plot of differential gene expression from bulk RNA-seq of isolated ILC1s from sarcoidosis skin (n = 2) vs. sarcoidosis blood (n = 2), demonstrating increased expression of CCL19 and CXCR4 in skin ILCs. (F) Density plot demonstrating that sarcoidosis-specific B cells express MS4A1 (CD20) and FCER2 (CD23) markers. (G) Summary of unique ligand-receptor interactions found in sarcoid granulomas.

Finally, CXCL12/CXCR4, a third signaling pathway important for TLS formation, was strongly induced in both non–sarcoidosis- and sarcoidosis-affected skin (Supplemental Figure 8A) (38). Fibroblasts and macrophages induced CXCL12, and multiple immune cell types expressed CXCR4 (Supplemental Figure 8F). Taken together, these results show that skin granulomas from patients with sarcoidosis and non-sarcoidosis patients exhibited unique molecular differences (Figure 4G). Non-sarcoid granulomas are driven by SPP1 signaling. In contrast, sarcoid granulomas induce cytokines involved in TLS formation (37).

Patients with sarcoidosis have increased levels of circulating blood ILC1s. We wanted to assess whether tissue-specific ILC1 changes may also be reflected in circulating blood. We collected peripheral blood mononuclear cells (PBMCs) from 7 patients with sarcoidosis for single-cell sequencing and created a combined approximately 116,767 cell data set (Supplemental Table 1). We also used publicly available data for 6 healthy adult controls (Supplemental Table 2) (Figure 5A and Supplemental Figure 9A). In contrast to skin, we identified cell populations by using a comprehensive single-cell reference atlas created by CITE-seq that is available for human PBMCs (39, 40). Our ILCs overlaid on the ILC coordinates depicted in the atlas (Supplemental Figure 9, B–D). Blood from sarcoidosis patients had an approximately 2.5-fold increase in ILCs compared with healthy volunteers. ILC differential gene expression profile showed a similar induction of inflammatory genes to that seen in ILCs from sarcoid skin (CCL5, GZMA, and GZMK) (Figure 5B, Supplemental Figure 9E, and Supplemental Table 4).

Blood from patients with sarcoidosis contains increased circulating levels of ILC1s. (A) Identification of cell clusters from blood of patients with sarcoidosis (S, n = 7) and healthy volunteers (HV, n = 6). (B) Scatter plot shows relative contribution of ILCs as percentage of total cells. Statistical significance was calculated using 2-tailed, unpaired Student’s t test. (C) Flow cytometry analysis of ILC subtypes as percentage of total CD45+ cells in healthy volunteers (n = 17), non-sarcoidosis granuloma patients (G, n = 5), and sarcoidosis patient blood (n = 13). One-way ANOVA revealed statistical significance for ILC1: F(2,27) = 19, *P < 0.05, ***P < 0.001; ILC2 and ILC3 were not significant among groups. (D) Receiver operating characteristic (ROC) curves for ILC1 (red) and ILC3 (black) from sarcoidosis patient PBMCs. Area under the curve (AUC) values are listed. (E) Flow cytometry analysis of ILC subtypes as percentage of total CD45+ cells in blood from no-treatment patients (n = 7), treated sarcoid patients (n = 6), and healthy volunteers (n = 17). One-way ANOVA revealed statistical significance for ILC1: F(2,27) = 55, ***P < 0.001; ILC2 and ILC3 are not significant. (F) Comparison of gene expression from bulk RNA-seq analysis of flow cytometry–purified sarcoidosis skin ILC1s (n = 2), sarcoidosis blood ILC1s (n = 2), and healthy volunteer ILC1s (n = 3). (G) Volcano plot of pathway analysis comparing flow cytometry–purified ILC1s from sarcoidosis blood and healthy volunteer blood. Data represented as mean ± SEM. NS, not significant. UPR, unfolded protein response; Ox phos, oxidative phosphorylation.

In parallel, we used the same flow cytometry gating strategy to assess ILC subtypes in blood from 13 patients with sarcoidosis, 5 non–sarcoidosis granuloma patients, and 17 healthy controls (Figure 5C and Supplemental Table 1). ILC1 was the major subtype, and healthy volunteers and sarcoidosis patient blood contained, on average, 0.05% and 0.45% of CD45+ cells, respectively (Figure 5C and Supplemental Figure 9F). Blood from non–sarcoidosis granuloma patients did not display enrichment of circulating blood ILCs of any subtype. Thus, patients with sarcoidosis exhibited a greater than 8-fold increase in circulating ILC1s. The receiver operating characteristic (ROC) curve assesses the accuracy of a diagnostic test and represents the relationship between the true positive rate (TPR, or sensitivity) of the test and its false positive rate (FPR). The area under the curve (AUC) index for ILC1 and ILC3 was 0.90 and 0.56, respectively (Figure 5D), where a score above 0.7 is considered a reliable biomarker (41). We noted that patients with sarcoidosis exhibited higher and lower populations of circulating ILC1s (Figure 5C). While this difference may reflect the known clinical heterogeneity of sarcoidosis, we subgrouped our sarcoidosis patients based on treatment status (Figure 5E). Patients with active disease that were not receiving any treatment (no-treatment) exhibited 12-fold more circulating ILC1s compared with healthy controls. Patients being actively treated with methotrexate, TNF inhibitors, or hydroxychloroquine had approximately 4-fold less circulating ILC1s compared with no-treatment sarcoidosis patients. The lone patient in the treated group with high ILC1s had active skin lesions despite taking hydroxychloroquine and a PDE4 inhibitor. The difference in circulating ILC1s between treated sarcoidosis patients and healthy controls was not statistically significant. Taken together, these data show that circulating ILC1s at a threshold of 0.45% of CD45+ cells may serve as a reliable biomarker for sarcoidosis diagnosis. Circulating ILCs levels may also reflect treatment status.

We also used flow sorting to purify ILC1s from the blood of healthy volunteers and sarcoidosis patients and performed bulk RNA-seq. Using the sarcoidosis blood ILC1 data set as a reference, 75% of the genes overlapped with our sarcoidosis skin ILC1 data set. In contrast, only 47% of the genes overlapped with ILC1s from healthy volunteer blood (Figure 5F). Sarcoidosis blood ILC1s induced genes involved in TNF signaling and the JAK/STAT pathway (Figure 5G). Thus, circulating ILC1s in patients with sarcoidosis showed both increased abundance and signs of increased activation.

Lung sarcoid tissue also exhibits increased B cells and ILCs. We wanted to determine whether sarcoid granulomas in other tissues also recruit B cells and ILCs. Similar to skin granulomas, sarcoid lung tissue displayed distinct aggregates of B cells (CD20+) (Figure 6A and Supplemental Figure 10A). Many of these B cells were also positive for CD23, suggesting these B cells were forming mature TLSs (34). Moreover, we found that ILCs (Lineage–CD127+Tbet+) localized to the periphery of sarcoid granulomas (Figure 6A). Taken together, these results show that lung sarcoid granulomas also exhibited increased recruitment of B cells and ILCs.

ILCs are necessary for noninfectious granuloma formation in mice. (A) Representative histology (n = 2 patients) demonstrating ILCs (Lineage–IL7R+Tbet+) and recruitment of B cells (CD20+CD23+) and to human lung sarcoid granulomas. Lineage– = CD3–CD16–CD19–CD20–CD56–CD68– (labeled in green). Scale bars: 100 μm and 10 μm (insets). (B) Representative H&E staining depicting granuloma formation in whole lung sections from cadmium nanoparticle–treated WT, Rag2–/–, and Rag2–/– Il2rg–/– (ILC-KO) mice. Inset images show higher magnification. Scale bars: 100 μm. (C) Quantification of lung granulomas (WT, n = 6; Rag2–/–, n = 7; ILC-KO; n = 8). (D) Representative immunofluorescence depicting macrophage (F4/80+) accumulation in lung tissue. Two-tailed, unpaired Student’s t test. Scale bars: 100 μm. The WT macrophages are also shown in Supplemental Figure 10C. (E) Flow cytometry analysis of ILC1 and NK cells as percentage of live CD45+ cells in different mouse genotypes before and after treatment with cadmium nanoparticles (QDOT). WT, n = 6; QDOT-treated, WT, n = 4; Rag2–/–, n = 7; ILC-KO, n = 3. Two-tailed, unpaired Student’s t test, comparing WT mice before and after treatment. Data represented as mean ± SEM. **P < 0.01; ***P < 0.001.

ILCs are necessary for noninfectious granuloma formation in mice. Next, we tested whether ILCs are necessary for noninfectious granuloma formation. There is no broadly accepted mouse model for sarcoidosis. We used the established cadmium nanoparticle (QDOT) mouse model, where pulmonary exposure to QDOT induces formation of lung granulomas (42). Similar environmental exposure to other heavy metals (i.e., beryllium) in humans is also linked to sarcoidosis-like diseases (43). After 30 days, we harvested lung tissue and counted granuloma formation per high-power field across the mid-coronal lung section. We verified that QDOT induced discrete granulomas in WT mice (WT, n = 6; QDOT-treated WT, n = 4; Figure 6, B–D, and Supplemental Figure 10B). Importantly, these granulomas recruit multiple immune cell types, including T cells, B cells and ILCs, similar to human sarcoid granulomas (Figure 6, D and E, and Supplemental Figure 10, C–F). We performed the QDOT model on mice lacking all ILCs (Rag2–/– Il2rg–/–, henceforth known as ILC-KO) as well as control Rag2–/– mice (8, 44, 45). Rag2–/– animals may lack mature B and T cells, but these mice remained capable of forming discrete tissue granulomas composed primarily of macrophages (Figure 6, C–E) (46). Notably, ILC-KO mice developed significantly fewer tissue granulomas compared with Rag2–/– control mice (n = 3 and n = 7, respectively) (Figure 6, C and D). Finally, we confirmed the absence of ILCs in ILC-KO mice by flow cytometry (Figure 6E and Supplemental Figure 10E). Notably, we did not see increased recruitment of NK cells to QDOT-induced skin granulomas (Figure 6E). We conclude that ILCs are necessary for noninfectious granuloma formation.

CXCR4/CXCL12 signaling is upregulated in sarcoidosis. Multiple immune cell types in sarcoidosis-affected skin exhibited increased expression of TLS-specific pathways, including CXCL12/CXCR4 signaling. Compared with gene expression profiles of other inflammatory skin diseases, sarcoidosis induced the strongest global expression of CXCR4 (Figure 7A) (32, 47, 48). Consistently, spatial transcriptomics localized CXCR4 expression within sarcoid granulomas, and expression was minimal or not seen in patients with psoriasis or healthy unaffected skin (Figure 7B and Supplemental Figure 11) (49). Moreover, protein immunofluorescence confirmed strong CXCL12 expression within the center of sarcoid granulomas (Figure 7C).

CXCR4 is necessary for ILC1 migration and mouse noninfectious granuloma formation. (A) Dot plot comparing cytokine ligand and receptor gene expression in sarcoidosis and skin inflammatory diseases. Dot size reflects percentage cells expressing the gene, and color illustrates level of gene expression. MC, molluscum contagiosum; BP, bullous pemphigoid; AE, acrodermatitis enteropathica; AD, atopic dermatitis. (B) Representative spatial transcriptomics depicting CXCR4 and CXCL12 expression in affected sarcoidosis (n = 2), unaffected sarcoidosis (n = 2), psoriasis (n = 3), and healthy volunteer skin (n = 3). (C) Representative immunohistochemistry (n = 3) of CXCL12 depicting expression within sarcoid granulomas. Scale bars: 50 μm. (D) Fold change in CXCL12-mediated migration of CD45+ immune cells from healthy volunteers (HV, n = 4) and sarcoidosis blood (n = 4) with and without CXCR4 inhibitor (plerixafor). Mean ± SEM. Significance was calculated by 2-tailed, paired Student’s t test. (E) H&E staining highlighting pulmonary granuloma formation in lung tissue from cadmium nanoparticle–induced mice treated with PBS or plerixafor. Scale bars: 100 μm. Quantification of lung granuloma formation (n = 5 in each group). Data represented as scatter plots show mean ± SEM. *P < 0.05; ***P < 0.001 by 2-tailed, unpaired Student’s t test.

In addition to promoting TLS formation, CXCL12/CXCR4 signaling has been shown to regulate immune cell migration (50). To test whether CXCR4 regulates sarcoidosis-activated immune cell function, we performed a migration assay with PBMCs isolated from healthy volunteers and patients with sarcoidosis. In response to its cognate ligand CXCL12, CD45+ circulating immune cells from sarcoidosis patients exhibited approximately 2-fold greater cell migration compared with CD45+ immune cells from healthy volunteers (Figure 7D). When CD45+ circulating immune cells were treated with a well-established pharmacologic CXCR4 inhibitor, plerixafor (also known as AMD3100), they no longer exhibited improved migration compared to immune cells from healthy volunteers (Figure 7D). We conclude that the increased expression of CXCR4 on sarcoidosis-activated immune cells improved cell migration toward CXCL12.

We returned to the QDOT granuloma model to test whether CXCR4 signaling is necessary for noninfectious granuloma formation. Mice were administered plerixafor or vehicle control by osmotic pumps. Compared with control-treated mice, plerixafor-treated WT mice developed fewer QDOT-induced lung granulomas (n = 5 in control group; n = 4 in plerixafor-treated group) (Figure 7E). Taken together, these results show that CXCR4 signaling is necessary for formation of noninfectious tissue granulomas.