Spatial transcriptomics reveals organized and distinct immune activation in cutaneous granulomatous disorders

Abstract

Background:

Non-infectious (inflammatory) cutaneous granulomatous disorders include cutaneous sarcoidosis (CS), granuloma annulare (GA), necrobiosis lipoidica (NL), and necrobiotic xanthogranuloma (NXG). These disorders share macrophage predominant inflammation histologically, but the inflammatory architecture and the pattern of extracellular matrix alteration varies. The underlying molecular explanations for these differences remain unclear.

Objective:

To understand spatial gene expression characteristics in these disorders.

Methods:

We performed spatial transcriptomics in cases of CS, GA, NL, and NXG to compare patterns of immune activation and other molecular features in a spatially resolved fashion.

Results:

CS is characterized by a polarized, spatially organized T helper (Th) 1 predominant response with classical macrophage activation. GA is characterized by a mixed, but spatially organized pattern of Th1 and Th2 polarization with both classical and alternative macrophage activation. NL showed concomitant activation of Th1, Th2, and Th17 immunity with a mixed pattern of macrophage activation. Activation of type 1 immunity was shared among, CS, GA, and NL and included upregulation of IL-32. NXG showed upregulation of CXCR4-CXCL12/14 chemokine signaling and exaggerated alternative macrophage polarization. Histologic alteration of extracellular matrix correlated with hypoxia and glycolysis programs and type 2 immune activation.

Conclusions:

Inflammatory cutaneous granulomatous disorders show distinct and spatially organized immune activation that correlate with hallmark histologic changes.

Clinical Implication:

The molecular profiles of these disorders imply multiple avenues for novel therapeutic intervention and begin to shed light on the immunologic differences amongst disorders.

Graphical Abstract

Capsule summary

There are no FDA approved therapies for inflammatory cutaneous granulomatous disorders. Spatial transcriptomics identifies distinct immune programs in each disorder that may guide future therapeutic development.

Introduction

Macrophages play central roles in many functions including antimicrobial responses, tissue repair including wound healing, and tissue homeostasis.^1,2 During infection, particularly with mycobacterial organisms, aggregates of activated macrophages called granulomas may form and are thought to help defend against spread of microbes.³ Macrophages also serve critical roles in tissue repair such as in wound healing and transiently accumulate in tissue during the repair process.⁴

Monocytes and macrophages can also become inappropriately activated and accumulate in tissue where they contribute to inflammation and damage during autoimmunity or autoinflammation.⁵ In some such non-infectious inflammatory disorders, macrophage activation and accumulation in tissue is particularly prominent and granuloma formation may be seen.⁶ In dermatology this group of disorders includes cutaneous sarcoidosis (CS), granuloma annulare (GA), and necrobiosis lipoidica (NL) which are unified by clinical lesions which are most commonly dermally-based and sometimes (but not always) clinically annular (ovoid lesions with prominent raised edges). Macrophage-rich (often referred to as granulomatous) inflammation is seen histologically.

Despite the overlapping features of these disorders, there are also distinct clinical and histological features that distinguish them from one another. In sarcoidosis, there is a proclivity for cutaneous lesions to involve the head and neck (including the face), scars, and tattoos.⁷ Histologically, well-organized tuberculoid granulomas composed of tightly-packed epithelioid macrophages with relatively sparse lymphocytic inflammation are seen. Prominent visual alteration of the dermal extracellular matrix is typically not observed. Extracutaneous involvement, especially of the lungs and lymph nodes is very common, and often occurs without skin involvement.⁷

GA lesions are common on the extremities (especially distal) and trunk. Histologically, two main patterns of inflammation are seen: palisaded and interstitial.^8,9 In palisaded GA, macrophages surround relatively pauci-cellular areas, often with visibly altered (degraded) extracellular matrix (ECM) (collagen and elastin) and deposition of mucinous material and glycogen.^10–12 In the interstitial pattern, extracellular matrix alteration is more subtle, and macrophages are present in between collagen bundles without well-developed palisading. Perivascular lymphocytes are a reproducible finding.

NL almost always involves the anterior lower legs and is often but not always found in patients with diabetes.¹³ Histologically, there is palisaded granulomatous inflammation which is typically pan-dermal (as opposed to the patchy dermal involvement in GA) and the palisades may be organized into horizonal tiers. Necrobiosis (pauci-cellular areas of altered ECM with collagen and elastin degeneration, typically without prominent mucin) is a hallmark. Fibrosis may be marked (in contrast to GA) resulting in a characteristic squared biopsy shape histologically.¹⁴ Plasma cells are more commonly encountered histologically in NL, compared to GA.¹⁴

Necrobiotic xanthogranuloma (NXG) is a non-Langerhans cell histiocytosis typically attributable to paraproteiemia, plasma-cell dyscrasias (including multiple myeloma), or other lymphoproliferative disorders.¹⁵ Peri-orbital involvement by granulomatous plaques or nodules is common. Histologically palisaded granulomatous inflammation with necrobiosis is observed and multi-nucleate cells are often prominent and may include very large and bizarre forms. Macrophages often have a lipidized appearance and cholesterol clefts may be seen. Thus, despite its unique ontogeny, there is clinical and histologic overlap with inflammatory granulomatous disorders, particularly NL.

The underlying immunologic programs co-opted by these overlapping yet distinct inflammatory disorders have not been fully elucidated. New molecularly targeted immunologic therapeutics are often directed against key modulators of conserved immune programs, in particular T cell derived cytokines. For example, IL-4/13 inhibition is used to treat type 2 predominant disorders and IL-23-IL-17 axis inhibition is used to treat type 3 (Th17) predominant disorders. In terms of the macrophages themselves, a commonly used (albeit overly simplified) classification system is M1 versus M2. M1 macrophages are “classically activated” and represent effector macrophage polarization seen during cell-mediated type 1 immune responses. The activity of IFN-γ and TNF (and/or lipopolysaccharide) are thought to drive this pro-inflammatory phenotype.¹⁶ In contrast, “alternatively activated” (or M2) macrophages are thought to play an important role in tissue repair and may have anti-inflammatory / homeostatic properties. Type 2 cytokines including IL-4 and IL-13 are thought to be key drivers of this macrophage phenotype.¹⁶ Better deciphering the immunologic cues driving cutaneous granulomatous disorders has great potential to inform our understanding of macrophage biology, cutaneous inflammation, ECM regulation, and to identify novel treatment strategies for these disorders; for which there are collectively no FDA approved therapies (except prednisone and corticotrophin repository for pulmonary sarcoidosis).

Spatial transcriptomics allows relatively unbiased transcriptomic evaluation of tissue in a spatially resolved manner. Given the characteristic and spatially organized nature of the histologic inflammation in CS, GA, NL and NXG, we use spatial transcriptomics to better delineate and compare the underlying immunologic features of these disorders.

Materials and Methods

Methods Summary

Cases were selected using careful clinicopathologic correlation. Additional details are available in the Online Repository and Tables E1 and E2. For samples subjected to spatial transcriptomics, formalin-fixed paraffin embedded (FFPE) blocks were cut and slides were prepared in accordance with the 10X Genomics’ Visium FFPE pipeline. Details on sequencing and the analytic pipeline are described in detail in the Online Repository. RNA in situ hybridization was performed using the RNAscope platform (Advanced Cell Diagnostics) and immunohistochemistry was performed using standard methods; additional details can be found in the Online Repository.

Results

Distinct transcriptomic regions are present in cutaneous granulomatous disorders

Cases of GA, CS, NL, and NXG with classic clinical and histologic features were selected for inclusion in the study (Figure 1A and Table E1). Normal skin from patients without these disorders was included as a control. We performed spatial transcriptomics using 10X Visium to quantify RNA transcripts within a 55 μM spot size. Cases with sufficient spots passing quality control cutoffs (250 unique features present in > 20 spots) were used in the analysis. Based on these criteria, four samples were excluded (1 GA and 3 CS). Thus, the final analysis cohort included 5 cases of NL, 4 cases of GA, 2 cases of CS, 1 case of NXG, and 3 cases of normal control skin (Table E1). The analysis cohort included 12,665 spots with 3,940 spots from GA samples, 1,511 spots from CS samples, 4,853 spots from NL samples, and 1,252 spots from the NXG sample. In non-control cases, there was an average of 963 spots with an average of 1856 unique genes and 3445 total reads. Normal control sections had a mean of 370 spots per section with 721 unique features and 1065 total counts per spot on average.

Figure 1. GA, CS, NL, and NXG are characterized by spatially distinct patterns of gene expression.

A. Cartoon summarizing the typical histologic features of these disorders. B. H&E (left), spatial feature plot showing transcriptionally distinct regions (middle), and UMAP plot (right panel) of a representative GA sample. C. H&E (left), spatial feature plot showing transcriptionally distinct regions (middle), and UMAP plot (right panel) of a representative CS sample. D. H&E (left), spatial feature plot showing transcriptionally distinct regions (middle), and UMAP plot (right panel) of a representative NL sample. E. H&E (left), spatial feature plot showing transcriptionally distinct regions (middle), and UMAP plot (right panel) of NXG. F. Spatial features plots showing expression to myeloid (CD68), T cell (CD3E), and fibroblast (COL1A1) markers in representative cases of GA, NL, CS, and NXG. G. Heatmap showing cell type deconvolution results among distinct spot types in each disorder.

Biologic replicates from each condition were informatically merged for dimensionality reduction analysis and compared to merged data sets for other conditions. Unsupervised clustering of each condition revealed multiple transcriptionally distinct regions. The clusters were then compared across conditions and similarities in transcriptional profiles corresponding to distinct histologic areas allowed classification into distinct spot types: 1: macrophage rich (M), 2: dermis with lymphocytic rich inflammation (DL), 3: pauci-inflammatory dermis with minimal histologic change (DU), 4: necrobiotic areas (N), 5: epidermis predominant (E), and 6: uninflamed control dermis from healthy skin (C) (Figure 1B–E and Figure E1). Similar regions were observed among replicates within each condition (Figure E2–E4). Given contamination from keratinocyte genes and the dermal predominant inflammatory pattern in these diseases, epidermal predominant (cluster 5) spots were excluded from subsequent analysis. Spots in control samples were filtered to only include those in dermis regions based on H&E staining.

The macrophage rich spots (M) corresponded to areas of granulomatous inflammation histologically. The lymphocyte rich spots (DL) corresponded to perivascular-predominant dermal inflammation which was often nearby macrophage rich areas, while pauci-inflammatory spots (DU) corresponded to relatively uninflamed areas of dermis histologically (Figure 1F). Spot-based cell type deconvolution analysis was performed on each sample using SpatialDWLS (Giotto package).²¹ Spot gene signatures were mapped to a previously published scRNAseq reference data set from patients with GA, CS, and normal skin from healthy controls.^22,23 This analysis showed that indeed myeloid gene signatures were enriched in the M spots and T cell genes were enriched in DL compared to DU. Fibroblast signatures also corresponding to DL and DU spots (Figure 1G).

Each disorder shows a distinct immune polarization program

Apart from CS (type 1 predominant), the overarching immune and macrophage polarization characteristics in these cutaneous granulomatous disorders is not firmly established. To compare immune polarization among disorders, data from biologic replicates for each condition were informatically merged and canonical immune polarization markers were compared across conditions and spot types. We found that CS showed high expression of type 1 immune markers supporting a type 1 predominant response (TBX21, IL12A, IL12B, IFNG, IL15, CXCL9, and CXCL10) with relatively little expression of Th2 and Th17 markers (Figure 2A) consistent with prior reports.^23,27 These signals were prevalent within the macrophage rich spots, consistent with intermixing of activated lymphocytes among the macrophages. CS also showed strong expression of markers associated with M1 macrophage polarization (NOS2, IRF3, IRF5, STAT1, CD274, and IL18BP) with relatively low expression of M2 markers (Figure 2A).

Figure 2. Differential, spatially-organized immune polarization is observed in each disorder.

A. Heatmap showing expression of key immune polarization genes across conditions and spot types. B. UMAP projection of macrophage-rich (M) spots among different conditions. C. Volcano plot showing differentially expressed genes among disorders. D. Violin plots showing expression of selected macrophage polarization makers across conditions (within the macrophage-rich (M) spots). E. Pathway analysis showing enrichment of selected transcriptional programs across conditions (within the macrophage-rich (M) spots).

The other conditions showed mixed patterns of immune polarization and relatively lower expression of T cell markers within the macrophage rich areas. NL and GA showed relatively increased expression of Th2 markers (IL4, IL5, and IL13). Th17 markers were relatively increased in NL compared to other conditions.

Thus, we next focused specifically on macrophage-rich (M) spots for further analysis and compared the macrophage-rich (M) spots across conditions. Macrophages showed a mixed M1 and M2 polarization pattern in GA, NL, and NXG. Compared to CS these conditions showed more prominent expression of M2 markers including CX3CR1, MERTK and CD163 (Figure 2A). NL also showed prominent M1 marker expression, including high expression of effector molecules (e.g. IL6 and IL1B) that were comparable to or even exceeded the expression level seen in CS. UMAP projection of the macrophage rich spots indeed showed distinct clusters in each condition, with the most heterogeneity observed in GA (Figure 2B).

Differential gene expression (DEG) analysis was performed in the macrophage rich (M) spots using Seurat and compared among conditions, showing distinct transcripts preferentially upregulated in each condition (Figure 2C and 2D). In CS, markers of Th1/M1 polarization including CXCL9, CHIT1, SOD, WARS, CHI3L1, STAT1, and GBP5 were among the most differentially expressed transcripts. CCL19 (macrophage inflammatory protein 3b) and PTGDS (prostaglandin D2 synthase) were also uniquely and markedly upregulated in CS.

GA showed prominent upregulation of fibroblast and extracellular matrix genes (COL1A1, COL1A2, COL6A1, COL6A2, FBN1, and TNXB) consistent with the ECM alteration observed histologically in GA. GA also showed marked upregulation of specific chemokines including CCL18 (macrophage inflammatory protein 4), CCL13 (monocyte chemotactic protein 4) and CCL17 (TARC).

NL showed prominent upregulation of calprotectin genes (S100A8 and S100A9), methallothionein genes (MT2A, MT1G, and MT1H), and SLC11A1 (divalent metal ion transporter). CXCL8 was also upregulated in NL compared to the other conditions and is likely consistent with an element of type 3 polarization.

NXG showed the most prominent upregulation of tissue resident macrophage markers (which somewhat overlap with M2 markers) including CD163, MARCO, VMO1, CXCR4, RNASE1, and VSIG4 (Figure 2C and 2D).²⁸ Lipid metabolism genes were also upregulated in NXG including ME1, PLTP, PLD3 (Figure 2C) and are consistent with the often-lipidized appearance (reflecting high lipid content) of macrophages histologically in NXG.

To better understand the biologic origins of these differing transcriptional patterns in macrophage rich (M) spots, gene set enrichment analysis (GSEA) was performed on the most differentially regulated genes in each condition using EnrichR.²⁴ In CS, IFN-γ response programs and allograft rejection were the most strongly upregulated programs, consistent with the observed type 1 / M1 immune polarization (Figure 2E). In GA, epithelial mesenchymal transition was by far the most upregulated pathway likely reflecting relatively increased transcription of mesenchymal (ECM) genes in the macrophage-rich spots in this disorder. This observation may unify the histologic ECM alteration, observed M2 macrophage polarization, and physiologic role of M2 macrophages in tissue repair responses.

NL showed a mix of transcriptional programs including IFN-γ and TNF responses. Interestingly there was also a prominent hypoxia signature and upregulation of “mineral absorption” transcriptional programs in NL. The mineral absorption program appeared to in part reflect upregulation of metallothonein genes (metallothioneins act as binders of heavy metals and also play a role in oxidative stress responses). NXG showed a mixture of transcriptional programs with phagocytosis (phagosome and lysosome programs) and cholesterol metabolism being relatively upregulated compared to the other disorders.

CS shows polarized Th1 inflammation with classical (M1) macrophage activation

We next sought to understand spatial characteristics of the variable immune programs observed in each condition. To better understand patterns of gene expression as they relate to established transcriptional programs, we used single-parameter BinSpect (Giotto package) to generate a list of 1000 spatially variable genes in each condition,¹⁸ which were then used to generate modules of spatially co-expressed gene sets. Analysis of CS showed 5 distinct transcriptional modules, a representative case is shown in Figure 3A. Module 1 and 2 were similar in their spatial organization and showed highest expression within macrophage rich (M) spots. These spots corresponded to areas of increased myeloid gene expression and areas of well-organized granulomas on H&E (Figure 3B). Module 4 was highest in between the granulomas in the histologically relatively uninflamed dermis (corresponding to DU spots). Module 4 showed predominantly fibroblast and ECM markers.

Figure 3. CS shows spatially organized activation of Type 1 immunity.

A. Heatmap showing expression of spot-type markers in each CS transcriptional module. B. Spatial feature plots showing expression of myeloid and lymphocyte spot markers (box) and each module in a representative case. H&E section and high-power inset (scale bar 50 μm) of a sarcoidal granuloma. C. Selected results of pathway analysis in representative modules, nGenes: number of genes in the module. D. Spatial feature plots showing expression of IFNG, CD3E (T cells) and CD68 (macrophages). E-F. Spatial feature plots for selected transcripts. G. Violin plots showing expression of CXCL9 and CCL19 among all replicates of each condition. H. Quantification of YKL40 (CHI3L1) IHC in areas of granulomatous inflammation in cases of GA (n=13), CS (n=10), NL (n=10), NXG (n=7) and CNTRL normal skin (n=10), error bars represent standard error of the mean (SEM). I. Representative examples of YKL40 (CHI3L1) IHC (brown chromogen), scale bar: 100 μm.

Pathway analysis was performed using the genes that made up the modules and showed strong activation of type 1 immunity in Module 1, with IFN-γ response being the most upregulated transcriptional program (Figure 3C). IFNG high spots tended to correspond to spots with relatively increased CD3E expression and were enriched around the periphery of granulomas (Figure 3D). Markers of response to IFN-γ, including M1 polarization genes such as IRF1, CHIT1, CHI3L1, FBP1, STAT1, IFI30, CXCL9, and NFKBIA were concentrated within the granulomas. Expression of CCL19 was upregulated relative to other conditions and tended to be enriched within and at the edge of granulomas (Figure 3E–G).

CHI3L1 (YKL40), a hydrolase secreted by activated M1 macrophages, is upregulated in response to IFN-γ²⁹ and has previously been identified as being upregulated in pulmonary sarcoidosis.³⁰ We performed immunohistochemistry (IHC) for CHI3L1/YKL40 on additional cases of GA (n=13), CS (n=10), NL (n=10), NXG (n=7), and healthy control skin (n=10) and found marked upregulation in CS in areas of granulomatous inflammation compared to the other disorders (Figure 3H). In CS, diffuse strong labeling of granuloma macrophages was noted, while in GA, NL, and NXG only a minority of individual cells stained positively (Figure 3I). NOS2 encodes iNOS and is also a useful marker of M1 macrophage polarization.³¹ We also performed chromogenic RNA in situ hybridization (RISH) staining for NOS2 on these additional cases. The most significant upregulation of NOS2 was observed in CS, with only rare staining of individual and small clusters of cells observed in GA, NL, and NXG (Figure E5).

Trajectory inference analysis can be used to infer changes in cell type and/or differentiation and may be applied to spatial transcriptomic data. We utilized SpatialPCA to perform spatially-aware clustering,²⁵ and then used slingshot to construct transcriptional trajectories.²⁶ The trajectories were instructed to originate within macrophage rich spots and point outward as transcriptional changes occurred. This analysis suggested that signals leading to inflammatory macrophage polarization were the most intense within the center of the granuloma and dissipated moving outwards towards the edge of the granuloma and are ultimately diminished in the dermis outside the granulomas (Figure E6).

Taken together, these observations suggest that compared to other inflammatory cutaneous granulomatous disorders, CS is characterized by a relatively pure type 1 immune activation including M1 macrophage polarization. The magnitude of the inflammation appears to be highest in the center of the granuloma.

GA show a hierarchical structure with overlapping but distinct type 1 and type 2 inflammation

Analysis of spatially co-expressed gene modules was also performed in GA and revealed 8 distinct modules, a representative case is shown in Figure 4A. The module landscape was more complex than in CS, consistent with the more complex histology of GA which in addition to macrophage rich inflammation also shows perivascular lymphocytes and ECM alteration. Modules 2, 3 and 6 showed the strongest signal in macrophage rich (M) spots. Although there was spatial overlap in these modules, module 2 genes were most intense along the inner edge of the macrophage palisade extending through areas of ECM alteration (Figure 4B). Modules 3 and 6 were most intense within the macrophage palisade, with module 6 being more diffuse and module 3 being more localized. Modules 1, 5, 7, and 8 were enriched in areas of dermis outside the palisaded inflammation (both DL and DU).

Figure 4. GA shows spatially organized activation of Type 1 and Type 2 immunity.

A. Heatmap showing expression of spot-type markers in each GA transcriptional module. B. Spatial feature plots showing expression of myeloid, fibroblast, and lymphocyte spot markers (box) and each module in a representative case. H&E with high power inset (scale bar 100 μm) of macrophage-rich interstitial inflammation. C. Selected results of pathway analysis in representative modules, nGenes: number of genes in the module. D-F. Spatial feature plots for selected transcripts. G. Violin plots showing expression of CCL18, CCL17, and CCL13 among all replicates of each condition. H. Quantification of SPP1 (osteopontin) RISH staining in cases of GA (n=13), CS (n=10), NL (n=10), NXG (n=7) and CNTRL normal skin (n=10), error bars represent standard error of the mean (SEM). I. Representative example of SPP1 (osteopontin) RISH staining (red chromogen) in GA, scale bar: 100 μm. Dashed lines show the paucicellular areas typical of GA. J. Representative example of SPP1 (osteopontin) RISH staining (red chromogen) in GA, CS, and NL.

GSEA was performed on the genes composing the modules. Modules 3 and 6 (M spots) showed overlap in immune activation pathways (Figure 4C). The most significantly upregulated transcriptional program in module 6 was asthma, consistent with type 2 immunity; though type 1 immune programs (e.g. allograft rejection and graft versus host disease) were also upregulated. CCL13 (monocyte chemoattractant protein 4), CCL17 (TARC), and CCL18 (macrophage inflammatory protein 4) are chemokines known to contribute to type 2 inflammation and have also been implicated in M2 macrophage polarization and tissue repair responses.^32–37 In looking at the spatial expression patterns of these chemokines, we saw that CCL18 more so than CCL13 expression was enriched within spots encompassing the macrophage palisade (module 6) (Figure 4D). CCL17 was also expressed at highest levels within the macrophage palisade and overlapped more so with module 3. These chemokines were much higher in GA than the other conditions, across biologic replicates (Figure 4G).

Module 3 showed more prominent activation of type 1 immune programs with top programs being allograft rejection, IFN-γ response and NF-κB signaling. Markers of Th1 immunity including CXCL9, GBP5, IFI30 overlapped most closely with Module 3 consistent with the results of pathway analysis (Figure 4E). Together, these findings suggest overlapping, but spatially distinct activation of type 1 and type 2 immunity in the macrophage palisade of GA; and are consistent with prior bulk RNA expression analyses in GA.³⁸

We next focused on module 2 which corresponded to areas of altered ECM within the macrophage palisade histologically. By far, the strongest transcriptional programs identified by pathway analysis in this module were hypoxia and glycolytic programs (Figure 4C). HK2 and VEGFA (reflecting these gene signatures) were upregulated in this area (Figure 4F). TGFBI (TGF-β) was also prominently upregulated in this area. Although the significance of these constellation of observations is not fully clear, they are interesting because TGF-β is a well-known regulator of extracellular matrix biology, thought to be generally anti-inflammatory, and has recently been implicated in promoting glycolytic metabolism in macrophages.³⁹ SPP1 (osteopontin) was also highly upregulated in module 3. Osteopontin is a cytokine-like protein that is known to regulate ECM turnover.⁴⁰

We were intrigued by the hypoxia and tissue remodeling signature in the areas of altered ECM. To follow-up on these observations, RISH staining was first performed for SPP1 in the additional cases of GA, CS, NL, NXG, and healthy control skin. We found prominent upregulation of SPP1 at the interface between the areas of granulomatous inflammation and the areas of altered collagen in GA, NL, and NXG (Figure 4H–J). While focal SPP1 labeling was observed in some CS cases, it tended to be uncommon, and when present, expression was most prominent in central areas of the granuloma (Figure 4J). RISH was also performed for VEFGA on these cases. We found that VEGFA was much more prominently expressed in GA, NL, and NXG than CS (Figure E7).

Trajectory inference analysis was also performed in GA and compared to CS was more complex. The macrophage palisade particularly, the area corresponding to module 3 (type 1 immune predominant) showed the strongest activation at early inferred time signatures and then transitioning towards areas of unaffected tissue (Figure E6). Taken together, these findings suggest that a type 1 immune response, perhaps directed against a fibroblast antigen, could be the proximal immune event in GA and that type 2 / M2 signatures may be secondary; perhaps being activated in an attempt to repair damaged dermal tissue.

NL shows a highly inflammatory state with multiple overlapping immune programs

Analysis of spatially co-expressed gene modules was also performed in NL, a representative example is shown in (Figure 5A). Modules 2 and 4 were the predominant modules corresponding to areas of granulomatous inflammation histologically and to macrophage rich (M) spots (Figure 5B). Module 4 was more closely associated with the macrophage palisade whereas module 2 enrichment was more diffuse and involved both the macrophage palisade and adjacent areas of lymphocytic inflammation (DL). Module 3 was enriched along the inner aspect of the macrophage palisade extending through the necrobiotic areas. Module 1 corresponded to areas of inflamed dermis (DL). Module 6 was highest in deeper parts of the tissue and corresponded to aggregates of lymphocytes with admixed plasma cells (Figure 5B).

Figure 5. NL shows spatially organized activation of Type 1, 2, and 3 immunity.

A. Heatmap showing expression of spot-type markers in each NL transcriptional module. B. Spatial feature plots showing expression of myeloid, fibroblast, and lymphocyte spot markers (box) and each module in a representative case. H&E with high power inset of a lymphoid aggregate, scale bar: 50 μm. C. Selected results of pathway analysis in representative modules, nGenes: number of genes in the module. D-I. Spatial feature plots for selected transcripts with (F) violin plots showing expression of CXCL8 and CXCL2 across conditions and biologic replicates.

GSEA was performed using the genes composing each module to better understand the biologic programs underlying these transcriptional modules. Module 4, which was enriched in the macrophage palisade, showed activation antigen presentation, allograft rejection, and IFN-γ response, suggesting type 1 predominant immunity in these areas, and reflected by expression patterns of CXCL9, IFI30, and CD274 (PD-L1), and STAT1 (Figure 5D). Type 1 immune polarization signals were more prominent in magnitude in NL relative to GA, being more comparable to that observed in CS. Module 4 also showed weak enrichment for asthma transcriptional programs suggesting a lesser component of type 2 immunity in these areas; type 2 chemokines CCL13, CCL17, and CCL18 were much less prominent than in GA (Figure 4G and 5E). Module 2 which was also upregulated in the macrophage rich spots (as well as adjacent lymphocyte rich areas) also showed upregulation of allograft rejection and IFN-γ response, though to a lesser degree than module 4. Activation of type 3 immunity was also observed in module 2 (Figure 5C). CXCL2 and CXCL8 are chemokine biomarkers of type 3 immunity,⁴¹ and were elevated in NL compared to the other conditions (Figure 5F). Thus, macrophage rich spots in NL were enriched for a mixed, but spatially distinct pattern with type 1 immunity activation to a greater degree than type 2 and type 3 immunity; notably significant type 3 immune activation was only evident in NL and not detected in the other conditions.

GSEA of module 3, which was enriched along the inner aspect of the macrophage palisade extending through the necrobiotic areas, showed that hypoxia and glycolytic transcriptional signatures were prominent in these spots (Figure 5C) as reflected by expression of VEGFA and HK2 (Figure 5G). This pattern was similar to the inner edge of GA palisades extending into areas of altered ECM. As in GA, SPP1 (osteopntin) expression was also highest in these areas. In contrast to GA, TGFBI expression in NL was not as prominent in these areas. Pathway analysis also identified mineral absorption as an upregulated pathway in NL (Figure 5H) and likely corresponded in part to expression of metallothionein genes (eg. MT2A and MT1H), which are involved in regulation of oxidative stress.

Module 6 showed upregulation of B cell receptor signaling and corresponded to markers of B and plasma cells (MS4A1, PAX5, and XBP1) as well as to areas of lymphoid aggregates with plasma cells on H&E (Figure 5B and 5I).

Trajectory inference analysis showed the earliest inferred time points corresponding to areas of well-developed macrophage predominant gene expression and corresponded most closely to module 4 with this signal dissipating in favor of other transcriptional programs with progression towards inflamed dermis and areas of necrobiosis (Figure E6). In sum, NL showed marked upregulation of type 1 immunity and to a lesser degree, activation of types 2 and 3 immunity. Necrobiotic areas showed an oxidative stress and hypoxia signature.

NXG shows prominent expression of M2 markers and upregulation of the CXCL12/CXCL14-CXCR4 axis

Analysis of spatially co-expressed gene modules was also performed in NXG revealing 8 gene expression modules (Figure 6A). Modules 3 and 7 were the predominant modules corresponding to areas of granulomatous inflammation histologically, were highly spatially overlapping, and corresponded to macrophage rich (M) spots (Figure 6B). Module 6 was highest in deeper parts of the tissue (near the deep dermal, subcutaneous junction) and corresponded to aggregates of plasma cells histologically (Figure 6B).

Figure 6. NXG shows an exaggerated M2 phenotype and upregulated CXCR4 signaling components.

A. Heatmap showing expression of spot-type markers in each NXG transcriptional module. B. Spatial feature plots showing expression of myeloid, fibroblast, and lymphocyte spot markers (box) and each module in a representative case. H&E with high power inset of plasma cell rich area, scale bar: 25 μm. C. Selected results of pathway analysis in representative modules, nGenes: number of genes in the module. D-F. Spatial feature plots for selected transcripts. G. Violin plots showing expression of selected markers across conditions including all biologic replicates. H. Quantification of CD163 IHC in cases of GA (n=13), CS (n=10), NL (n=10), NXG (n=7) and CNTRL normal skin (n=10), error bars represent standard error of the mean (SEM). I. Representative example of CD163 IHC (brown chromogen) in NXG, scale bar: 100 μm. J. Representative examples CD163 (brown chromogen) staining in GA, CS, NL, and CNTRL, scale bar: 100 μm.

GSEA showed that in module 7 asthma (reflecting activation of type 2 immunity) was the most highly upregulated transcriptional program (Figure 6C). Module 3 which was also prominent in macrophage rich (M) spots and extended into necrobiotic zones showed signatures for hypoxia and glycolysis; similar to areas of necrobiosis in NL. Other signatures including mTORC1 signaling and cholesterol homeostasis were also upregulated and could correspond to the known lipid-rich (“foamy”) appearance of NXG macrophages histologically.

We also noted that markers of tissue resident macrophages including VSIG4, RNASE1, MARCO, VMO1, and CXCR4 were among the most prominent genes expressed in modules 3 and 7 (Figure 6D and 6G),²⁸ consistent with the prominent upregulation of these genes in macrophage rich (M) spots in NXG compared to other sample types (Figure 2C). These spots also showed high upregulation of the M2 marker CD163 (Figure 6G).

We performed IHC staining for CD163 in second cohort of GA, CS, NL, NXG, and healthy control cases. We found that CD163 labeling the majority of macrophages in most cases of GA, NL, and NXG (Figure 6H–J). The intensity of the labeling was highest in NXG with prominent labeling of the large, multinucleate macrophages typical of NXG (Figure 6I). In CS, while some CD163 labeling was seen, the labeling was limited to mononuclear cells at the periphery of the granuloma; the epithelioid macrophages composing most of the granuloma in CS did not show significant staining (Figure 6J). IHC staining for MARCO was also performed in these cases. MARCO expression was strongest in GA, NL, and NXG (Figure E8). As with CD163 staining, in CS, the most prominent staining was in the non-epithelioid mononuclear cells as the periphery of the granulomas. Faint MARCO staining was also present in the epithelioid cells in CS, the significance of which is not fully clear; for the purposes of quantification these cells were judged as negative (Figure E8).

NXG most commonly occurs in the setting of monoclonal paraproteinemia with a subset of patients also meeting diagnostic criteria for multiple myeloma (a plasma cell malignancy). The NXG sample utilized for spatial transcriptomics in this study was derived from a patient with IgGκ multiple myeloma; many of the NXG samples from patients in the second cohort also have IgGκ paraproteinemia (Table E2). Thus, we next shifted our attention to module 6, which corresponded closely with aggregates of plasma cells histologically (Figure 6B). Indeed, plasma cell markers including SDC1, CD79, IRF4 were expressed at the highest levels in these spots (Figure 6E). GSEA showed activation of adaptative immune programs in this module. B cell receptor signaling programs were noted in this region as well as cytokine-chemokine receptor interaction. Given this observation, we looked for cytokines and chemokines that were highly expressed in module 6 and we noted high expression of CXCL12 and CXCL14. CXCL12 and CXCL14 are chemokine ligands for CXCR4, which was also highly upregulated on the macrophage rich spots in NXG (Figure 6G). Interestingly, CXCL12-CXCR4 signaling has been previously shown to regulate M2 polarization in multiple myeloma in bone marrow.⁴² CXCR4 was also expressed to some extent in NL; albeit it at lower levels (significant expression was not detected in CS and GA). CXCL12 and CXCL14 ligands were produced across conditions (Figure 6G). Taken together, we hypothesize the CXCL12/CXCL14-CCR4 axis may be of particular importance in NXG, and to a lesser degree also active in NL.

Trajectory inference analysis showed numerous discrete areas with an early pseudo time corresponding to macrophage rich spots and consistent with diffuse tissue infiltration of activated macrophages (Figure E6). In sum, analysis of NXG showed a distinct phenotype of macrophages which show a tissue resident-like phenotype with exaggerated M2 polarization and patterns compatible with a role for CXCL12/CXCL14-CCR4 signaling.

Increased IL-32 expression is increased in cutaneous granulomatous disorders

Throughout our analyses, we noted that IL32 expression was often a prominent component of myeloid rich spots in CS, GA, NL, and NXG (Figure 7A). The magnitude of expression in the macrophage rich spots was highest in CS (Figure 7B). In contrast to the other disorders, in NXG IL32 expression was highest in the plasma cell rich spots (Figure 7A), consistent with the prior observation that constitutive IL-32 upregulation may occur in multiple myeloma plasma cells.⁴³

Figure 7. IL-32 overexpression is a feature of inflammatory cutaneous granulomatous disorders.

A. Spatial feature plots showing expression of IL32 in a representative case for each condition. B. Violin plot showing expression of IL32 across all biologic replicates. C. Quantification of IL32 RISH staining in cases of GA (n=13), CS (n=10), NL (n=10), NXG (n=7) and CNTRL normal skin (n=10), error bars represent SEM. D. Representative examples of RISH staining (red chromogen) in CS and CNTRL, scale bar: 200 μm.

RISH staining was also performed for IL32 in the second cohort of cases. CS showed the strongest staining, but prominent staining was also observed in GA, NL, and NXG (Figure 7C and 7D). There was some expression of IL32 in control skin as well, mostly in the around adnexal structures (Figure 7D). IL-32 is activated downstream of multiple pro-inflammatory cytokines including TNF and IFN-γ and is known to play a particularly important role in anti-mycobacterial immune responses,⁴⁴ suggesting this may be molecular signal in inflammatory cutaneous granulomatous disorders worthy or further investigation.

Discussion

We analyzed tissue sections from multiple non-infectious cutaneous granulomatous disorders using spatial transcriptomics and found activation of spatially organized immunologic programs that have both overlapping and distinct features and appear to correlate with the varied histologic features in each of these diseases.

CS was characterized by relatively pure type 1 immune activation that was most concentrated in the center of granulomas and dissipated outwards. Type 1 immune polarization has been previously described in CS using both scRNAseq and spatial transcriptomics,^23,27 and is consistent with the observation that inhibition of type 1 immunity with TNF inhibitors and Janus kinase (JAK) inhibitors can be effective in treatment of CS.⁴⁵ Another relatively unique signal observed in CS relative to other samples included upregulation of CCL19, which has been described previously in pulmonary and cutaneous sarcoidosis.^22,46 CCL19 is known to bind to CCR7 and in doing so play a role in the maturation of dendritic cells and trafficking of lymphocytes and DCs towards secondary lymphoid tissue. This data supports the previously proposed concept that highly organized granuloma formation in sarcoidosis may involve co-option of signals involved in formation of normal lymphoid organs.²³

In contrast to CS, type 1 immune activation, while present in GA was relatively muted in its magnitude. Unlike CS, detectable type 2 immune activation was also present and is consistent with a prior report.³⁸ Thus, GA shows an overlapping, but spatially distinct pattern of type 1 and type 2 immune activation, corresponding to the mixed M1 and M2 macrophage polarization phenotypes. CCL13, CCL17, and CCL18 were prominently upregulated in GA compared to other conditions. These chemokines have previously been described as type 2 immune biomarkers in diseases such as atopic dermatitis and asthma.^32,47–49 These chemokines have also been linked to M2 macrophage polarization, increased collagen production, and tissue repair; all features that appear to be present in GA. Interestingly, these chemokines may even act as eosinophil chemoattractants and is consistent with the observation that eosinophils are found histologically in ~40% of GA cases, but are very uncommon in CS.^50,51

While inhibition of type 1 immunity with TNF inhibitors or inhibition of type 1 and type 2 immunity with JAK inhibitors appear to be effective in GA, the role of type 2 specific inhibition, for example with dupilumab, in the treatment of GA is not fully clear with reports of both improvement, but also of development of GA in the setting of dupilumab.⁴⁵ Further work examining the functional role of type 2 immunity (e.g. primary driver vs secondary phenomenon) and its role in GA will require additional work. JAK inhibitors generally cover both type 1 and type 2 immunity and may explain the anecdotal apparent increased efficacy of JAK inhibition over dupilumab, but further clinical study is required to evaluate this hypothesis.

NL was the only condition that exhibited activation of all three major immune programs. While type 1 immunity was the predominant signature (more so than in GA and rivaling that seen in CS), activation of type 2 and 3 immune activation were also evident. Response of NL to TNF inhibition (type 1 and type 3) supports a role for these immune programs as a therapeutic target NL.⁴⁵ Successful use of JAK inhibitors such as tofacitinib (JAK1/3>2) and ruxolitinib (JAK2) have also been reported in NL.^52,53 Conventional JAK inhibitors better inhibit type 1 and type 2 immunity relative to type 3 immunity. In this context, it is interesting that while improvement in NL with JAK inhibitors has indeed been reported, complete resolution of lesions has not been reported to occur, and could involve residual type 3 immunity which was not fully controlled. In terms of the specifically targeting type 3 immunity in NL, a recent small open-label trial showed that secukinumab (an IL-17A inhibitor) led to stable disease to modest improvement in patients with NL.⁵⁴ These data suggest that activation of type 3 immunity, may not be a primary driver of NL, but further work will be needed to understand the relative importance of type 3 immunity in NL.

Metallotheioneins are cysteine-rich metal binding proteins that play a role in metal ion metabolism and are critical detoxifiers of reactive oxygen species.⁵⁵ Metallothionein 1 and 2 expression is constitutive in the skin but may be increased under conditions of oxidative stress in which free radical scavenging is imperative to prevent untoward tissue damage. Metallothionein expression can also be induced by pro-inflammatory signals and facilitates upregulation of anti-microbial reactive oxygen species.⁵⁶ The upregulation of metallothionein 1 and 2 genes in NL is interesting and appears to occur to a magnitude not seen in other cutaneous inflammatory granulomatous disorders. Reactive oxygen specifies can be genotoxic over time which is interesting given the proclivity of long-standing NL lesions to occasionally develop superimposed squamous cell carcinoma⁵⁷; however, other factors including chronic ulceration may also play a role. Concomitant development of overlying carcinoma is generally not seen in other cutaneous granulomatous disorders. Ultimately, understanding the etiology and implications of increased metallothioneins in NL and its potential relationship to squamous cell carcinoma will require further study.

In NXG, we also found upregulation of this pathway in skin along with CXCL14 which can also signal via CXCR4. Prior work in multiple myeloma identified the CXCR4-CXCL12 axis as important in the development of a M2 macrophage phenotype and a tumor supportive microenvironment in the bone marrow.⁴² Compared to other conditions NXG showed enhanced expression of M2 markers, some of which are also makers for tissue resident memory cells in the liver (Kuppfer cells) and are highlighted by markers including VSIG4, VMO1, and CD163. Taken together we interpret these data as suggesting that this chemokine axis is particularly important in NXG and may provide a link whereby malignant plasma cells promote macrophage accumulation and activation in skin in the setting of paraproteinemia by producing CXCL12 and/or CXCL14. Whether or not there is a contribution of skin resident macrophages to NXG or whether these markers reflect extreme M2 like polarization of circulating monocyte derived cells will require future investigation.

Plerixafor is an approved CXCR4 antagonist and could be worth evaluating in patients with multiple myeloma with skin manifestations, or paraproteinemia with granulomatous dermatologic disease manifestations warranting systemic therapy. Other investigational CXCR4 inhibitors are being evaluated as cancer immunotherapy adjuvants.⁵⁸

Necrobosis is a histologic hallmark of NL and NXG. In GA, areas of altered ECM are also characteristic of cases showing palisaded histology, and in some cases of GA, necrobiotic-like zones may develop. The pathomechanism of necrobiosis is not known. Electron microscopy studies in GA and NL have shown degenerative changes in both collagen and fibroblasts in the pauciceullar and necrobiotic areas.^10–12,59 Interestingly, immunohistochemical studies have shown loss of fibroblasts (using CD34+ staining) in lesional GA and NL.^60,61 Taken together, along with our molecular data, we propose that this phenomenon could reflect cell-directed immunity (type 1) against fibroblasts resulting in their depletion and then resulting in a tissue-repair response (type 2) and possibly explaining the mixed patterns of inflammation in these disorders characterized by ECM alteration.

The strongest molecular signature in the necrobiotic areas in this study was hypoxia and glycolytic metabolism. The reason for this remains unclear. Prior work showed that there was an increased (not decreased) microvessel density in GA and NL compared to normal skin, suggesting that blood supply, per se, is not the reason for this.⁶² Expression of osteopontin and TGF-β, two known regulators of ECM turnover that may even cooperate during tissue repair,^63–65 were also highly expressed in the necrobiotic areas. Interestingly, TGF-β and has recently been implicated in promoting glycolytic metabolism in macrophages,³⁹ possibly linking these observations, but further work will be required to better understand how these observations relate and what the proximal event is.

We found that IL-32 expression was increased across all the granulomatous disorders examined in this study. IL-32 is a proinflammatory cytokine with multiple isoforms. Relative to other cytokines, the function of IL-32 is still poorly understood; both intracellular and extracellular (secreted) functions have been described and the effect may be context dependent (pro- and rarely, anti-inflammatory action has been reported). Further, the cognate cell surface receptor is not known. IL-32 is upregulated during infection and has been shown to be important in defense against mycobacterial organisms,⁴⁴ where granulomatous inflammation is typical of the host response. Further work will be required to assess the exact role of IL-32 in these disorders and whether or not its inhibition could be a viable treatment strategy.

Limitations of the study include the relatively small sample size. Further, due to the retrospective study design, non-lesional skin from affected individuals with these disorders was not analyzed. We cannot entirely exclude that some of the variation observed was influenced by anatomic location (for example, all NL specimens were from the legs; a typical site for this disease). Another limitation was relative lack of racial/ethnic diversity in some of the groups; for example, for GA and NL, it was difficult to find cases in our tissue archive from black individuals; which may to some extent reflect the epidemiology of these diseases in the U.S.⁶⁶

In sum, we delineate distinct molecular changes in inflammatory cutaneous granulomatous disorders including major immune polarization programs. These data will be useful as the next generation of targeted therapies for these disorders is developed and evaluated.

Abbreviations:

CS

cutaneous sarcoidosis

FDA

U.S. Food and Drug Administration

ECM

extracellular matrix

GA

granuloma annulare

GSEA

gene set enrichment analysis

H&E

hematoxylin and eosin

IHC

immunohistochemistry

IL

interleukin

JAK

Janus kinase

NL

necrobiosis lipoidica

NXG

necrobiotoic xanthogranuloma

RISH

RNA in situ hybridization

SEM

standard error of the mean

Spot type abbreviations:

C

uninflamed control dermis from healthy skin

DL

dermis with lymphocyte rich inflammation

DU

pauci-inflammatory dermis with minimal histologic change

E

epidermis predominant spots

M

macrophage rich spots

N

necrobiotic areas