Spatial characterization of skin lesions in discoid and systemic lupus erythematosus

Abstract

Lupus erythematosus (LE) skin lesions are associated with significant dysregulation of the local immune microenvironment. However, the spatial distribution and interactions between stromal and immune cells within affected tissues remain poorly characterized. In this study, we employed Stereo-seq to construct a single-cell resolution spatial transcriptomic atlas of lesional skin from patients with discoid lupus erythematosus (DLE) and systemic lupus erythematosus (SLE). Our analysis revealed that keratinocyte subpopulations in distinct differentiation states presented cell type-specific profiles of inflammatory mediator expression. Notably, a subset of stress keratinocytes located predominantly at the epidermis-dermis interface mediated the recruitment of T cells and plasma cells, potentially through an IFN-γ-JAK-STAT axis driving CXCL9/10/11-CXCR3 signaling, and these effects were more pronounced in active SLE lesions compared to chronic DLE lesions. Spatial profiling revealed immune cell niches resembling tertiary lymphoid structures (TLS), which were particularly prominent in chronic DLE and SLE lesions. These lupus-associated TLS structures were characterized by the coordinated infiltration of multiple cell subsets, including age-associated B cells, precursors of germinal center B cells, naïve B cells, regulatory T cells, T peripheral helper and/or T follicular helper cells, memory T cells, and CCL14⁺ vascular endothelial cells, which presented the spatial characteristics of the lymphoid follicle-like structures in chronic LE skin lesions. Collectively, our findings provide a comprehensive spatial characterization of the cellular components and molecular features within LE skin lesions.

Skin lesions are common manifestations of systemic lupus erythematosus (SLE). Here, the authors present a spatial transcriptomic atlas of lesional skin from patients with lupus and identify an acute inflammatory signature in SLE, contrasting with chronic damage in tertiary lymphoid structures in discoid lupus.

Introduction

Lupus erythematosus (LE) is a complicated autoimmune disease with a wide spectrum of clinical manifestations, ranging from cutaneous changes to life-threatening systemic symptoms¹. Here, lupus limited to the skin is referred to as cutaneous LE (CLE) and is distinguished from LE with systemic involvement, which is termed systemic LE (SLE). Discoid LE (DLE), accounting for up to 80% of CLE cases, represents the most prevalent subtype of CLE². Notably, the skin is the second most frequently affected organ in patients with SLE, with cutaneous manifestations developing in 70-85% of patients during disease progression³. Consequently, cutaneous manifestations represent both a common clinical feature and a crucial pathogenic component in LE.

Insights into the pathogenesis of LE lesions are implicated in immune homeostasis dysregulation, particularly with the unwanted stimulation of innate immune responses and the activation of adaptive immunity⁴. Strong evidence suggests that in individuals with genetic susceptibility to LE, exposure to environmental factors such as ultraviolet (UV) radiation triggers keratinocyte stress responses, leading to apoptosis and the release of autoantigens⁵. These autoantigens initiate antigen presentation, thereby activating a complex pathogenic immune response mediated by T and B cells in lupus lesions⁵. However, this process may not explain the development of skin lesions in patients with CLE who lack autoantibodies. Moreover, there are no Food and Drug Administration (FDA)-approved targeted therapies for CLE, and empirical and “off-label” use of topical and systemic medications is common in the treatment of CLE⁶. Hence, a deeper understanding of the molecular mechanisms in the lesional skin of patients with DLE and SLE, particularly with respect to their cellular compositions and molecular characteristics, will facilitate the development of personalized treatment options for CLE.

The development of single-cell RNA sequencing (scRNA-seq) has provided insight into affected tissue samples from patients with lupus to define the cellular composition and identify underlying inflammatory mediators that contribute to the pathogenesis of lupus. Profiling the single-cell landscape of patients with lupus nephritis (LN) revealed a strong interferon (IFN) response and fibrotic signature in tubular cells from renal biopsies⁷. Recent findings have also demonstrated that the IFN-rich signature is present in the stromal and inflammatory cells of both lesional and non-lesional lupus skin^8–10. However, tissue dissociation in these studies removes the spatial context of interacting cell types, and the casual mechanism of complicated cellular interactions in the local microenvironments of lupus lesions is not entirely understood. Recently, spatial transcriptomics at single-cell resolution has provided a readout of how cellular identity, transcriptional activity, and potential function are influenced by specific microenvironmental interactions within tissues. This approach is particularly valuable for elucidating the spatial distribution of pathogenic cells and identifying cell-cell interactions during lupus progression for potential therapeutic intervention.

In this study, we characterize the spatial cellular composition, location, and interactions of stromal and immune cells within lesional skin at the single-cell level in lupus patients using spatial transcriptomics sequencing, providing insights into the mechanisms underlying disease initiation and persistence.

Results

Spatial transcriptomic atlas of lesional skin in patients with DLE and SLE

We performed spatial transcriptomics analysis with Stereo-seq on freshly frozen skin lesions from patients with DLE and SLE to determine the spatial cellular composition and distribution of lesional skin at the single-cell level (Fig. 1a). The whole tissue section was spatially resolved in high-resolution chips with a diameter of 220 nm/spot and an effective area of 100 mm². We generated Stereo-seq datasets of skin biopsy samples obtained from patients who had been pathologically diagnosed with LE (patients with DLE, n = 5; patients with SLE, n = 6), as well as healthy controls (HC, n = 4) (Supplementary Fig. 1a; Supplementary Data 1). We applied Cellbin, a cell segmentation method that is used to accurately identify cells with nucleic acid staining, to analyze the skin tissue slice matrix and achieve single-cell resolution¹¹. After filtering out the low-capture cells, we obtained 311,815 segmented cells with an average of 1153 UMIs and 730 genes per cell (Supplementary Fig. 2a-d; Supplementary Data 1). We integrated our previously published scRNA-seq data (GEO: GSE179633) of 58,446 cells from 10 epidermal samples (HC, n = 2; DLE, n = 4; SLE, n = 4) (Supplementary Data 2) and 81,207 cells from 10 dermal samples (HC, n = 2; DLE, n = 4; SLE, n = 4) (Supplementary Data 2) to establish an unbiased reference expression fingerprint of cell types to define the spatial mapping of distinct cell populations in Stereo-seq slides (details in Methods). Among the participants included in our study, patients with DLE presented with higher CLASI damage scores, whereas patients with SLE exhibited higher CLASI activity scores (Supplementary Fig. 1b).

Fig. 1. Spatiotemporal transcriptomic landscape of skin lesions in patients with DLE and SLE.

a Schematic representation of the overall study design. Skin lesions from 4 HCs, 5 DLE patients, and 6 SLE patients were included in this study and Stereo-seq analysis was performed. scRNA-seq from our published datasets were utilized for data analysis. In vivo experiments, in vitro experiments, and mIHC were utilized for results validation. This image was created by Biorender.com. b Representative H&E staining images of tissue sections from HC, DLE, and SLE patients used for Stereo-seq. Main images: Full-thickness skin sections showing the overall tissue architecture. Boxed areas: Magnified views highlighting characteristic features such as epidermal thickening or immune cell infiltration in LE patients. The images shown represent the results of 3 independent experiments. Scale bars, 1 mm for main images and 50 μm, 100 μm, 200 μm, 150 μm, and 200 μm for magnified images. c Box plots showing epidermal thickness across disease conditions in Stereo-seq data (HC, n = 4; DLE, n = 5; SLE, n = 6). d Uniform manifold approximation and projection (UMAP) plot showing major cell clusters from 139,653 cells in scRNA-seq data. e Spatial visualization of cell type maps in representative skin samples from HCs, DLE, and SLE patients in Stereo-seq data. Scale bars, 1 mm. f Heatmap showing the expression of marker genes for major cell types in Stereo-seq data. g Spatial visualization of cell types and marker genes in representative skin samples from HCs, DLE, and SLE patients in Stereo-seq data. Scale bars, 1 mm. h Box plots showing the overall cell density in HCs (n = 4), DLE (n = 5), and SLE (n = 6) based on Stereo-seq data, separated by epidermis and dermis. Epi, epidermis; Der, dermis. i Box plots showing the density of major cell types in whole skin sections from HCs (n = 4), DLE (n = 5), and SLE (n = 6) based on Stereo-seq data. j Violin plots showing the Gene Set Variation Analysis (GSVA) scores of pathways in the 6 regions based on Stereo-seq data. HC_Epi, epidermis of HC; DLE_Epi, epidermis of DLE; SLE_Epi, epidermis of SLE; HC_Der, dermis of HC; DLE_Der, dermis of DLE; SLE_Der, dermis of SLE. Data are presented as boxplots with the median (horizontal line), the 25th and 75th percentiles (bounds of box), and the maximum and minimum values (whiskers) (c, h, i), and P values were determined by two-sided Kruskal–Wallis test followed by Dunn’s multiple comparison test (c, h, i).

We used Seurat to cluster 139,653 qualified cells from scRNA-seq data (details in Methods)¹², which integrated epidermal and dermal cells into 13 major cell types, including keratinocytes (KRT10 and KRT1), melanocytes (MLANA and DCT), Schwann cells (CDH19 and MPZ), sweat gland cells (DCD and SCGB2A2), fibroblasts (DCN and COL1A1), endothelial cells (PECAM1 and CLDN5), smooth muscle cells (TAGLN and MYL9), T cells (CD3D and CD3E), NK cells (NKG7 and GNLY), B cells (MS4A1 and CD79A), plasma cells (JCHAIN and MZB1), macrophages/DCs (LYZ and AIF1), and mast cells (TPSAB1 and TPSB2) (Fig. 1d; Supplementary Fig. 2e). We assigned each segmented cell to a specific cell type with the highest probability ratio (details in Methods), and most cell types were highly correlated in both the scRNA-seq and Stereo-seq data (Supplementary Fig. 3a, c). Spatial visualization of the major cell clusters in the Stereo-seq slides revealed a well-organized distribution pattern of distinct cell types, consistent with the histopathological features indicated by H&E staining (Fig. 1b, e; Supplementary Fig. 1a, 3b). Expression of the canonical cell type marker genes in the scRNA-seq dataset validated the cell type annotations assigned to the segmented cells in the Stereo-seq slides (Fig. 1f, g).

Stereo-seq analysis identified substantial tissue remodeling in lesional skin. Compared to HCs, DLE and SLE lesions exhibited significantly increased epidermal thickness and overall cellular density (Fig. 1c; Supplementary Fig. 3d), primarily driven by a marked expansion of dermal cells (Fig. 1h). At the subset level, the densities of endothelial cells, and major immune populations (T cells, B cells, NK cells, and macrophages/DCs) were significantly elevated in both patient groups compared to HCs (Fig. 1i, g). While no significant differences in cellular densities were detected between DLE and SLE lesions, a trend toward a higher proportion of T cells and B cells was observed in DLE samples compared with the SLE group (Supplementary Fig. 3e).

To further investigate the transcriptional signatures of different regions in the lesional skin of lupus patients, we performed epidermis and dermis segmentation by manually identifying the basement membrane (BM) location on ssDNA images using H&E-stained images as a reference (details in Methods). Compared with HCs, functional enrichment analysis of local bulk RNA profiles in different regions revealed that DNA repair, oxidative phosphorylation, the P53 pathway, and the PI3K/Akt/mTOR signaling pathway were highly enriched in the epidermis of patients with DLE and SLE (Fig. 1j). Additionally, greater enrichment of pathways involved in interferon responses was observed in both the epidermal and dermal areas of patients with DLE and SLE than in HCs (Fig. 1j).

Overall, these data establish a comprehensive spatial cellular atlas of lesional skin in DLE and SLE patients, revealing characteristic regional distribution patterns.

Inflammatory mediators exhibit a distinctive distribution across keratinocytes at different differentiation states in the lupus epidermis

To further characterize epidermal functional alterations in DLE and SLE lesions, we categorized 38,027 keratinocytes into the following 13 clusters: granular KCs (FLG and LCE1F), spinous KCs 1-4 (KRT1 and KRT2), basal KCs (KRT15 and KRT14), channel KCs (GJB6 and ATP1B1), stress KCs (KRT6A and S100A8), cycling KCs (MKI67 and TOP2A), inner root sheaths (IRSs, KRT17 and KRT75), outer root sheaths (ORSs, POSTN and LGR5), outer bulges (OBs, FST and PTHLH), and hair follicle-sebaceous glands (HF-SGs, KRT79 and MGST1) (Fig. 2a, b). Spatial visualization of keratinocyte subclusters revealed distinct stratification patterns corresponding to differentiation states across epidermal layers. Granular KCs were predominantly localized to the upper epidermal layers, whereas basal KCs and cycling KCs were primarily confined to the basal layers adjacent to the BM (Fig. 2c, e; Supplementary Fig. 4a, b). Upon phenotypic enrichment of these keratinocyte clusters, granular KCs, OBs, stress KCs, cycling KCs, and ORSs were preferentially enriched in both DLE and SLE lesions (Fig. 2d).

Fig. 2. Spatiotemporal heterogeneity of keratinocyte subtypes and characteristics across regions of lesional skin in patients with DLE and SLE.

a UMAP plot showing keratinocyte subtypes from 38,027 cells in scRNA-seq data. Granular KC, granular keratinocyte; Spinous KC, spinous keratinocyte; Basal KC, basal keratinocyte; IRS, inner root sheath; ORS, outer root sheath; OB, outer bulge; HF-SG, hair follicle-sebaceous gland; Channel KC, channel keratinocyte; Stress KC, stress keratinocyte; Cycling KC, cycling keratinocyte. b Dot plot showing the expression of marker genes for keratinocyte subtypes in scRNA-seq data. c Spatial visualization of keratinocyte subtypes in representative LE sample from Stereo-seq data. Scale bars, 1 mm for the main image, and 50 μm and 100 μm for magnified images. d Composition of keratinocyte subtypes across disease conditions based on Stereo-seq data. e Line graphs showing the cell proportions of granular KCs, ORSs, OBs, stress KCs, and cycling KCs at different epidermal depths in lesional skin of HCs (n = 4), DLE (n = 5), and SLE (n = 6). Epidermis was divided into 10 equal layers, with depth 0 indicating the superficial side and depth 1 representing the BM side. BM, basement membrane. f Heatmap showing enriched pathways in granular KCs, ORSs, OBs, stress KCs, and cycling KCs of DLE and SLE patients in scRNA-seq data. g Spatial visualization and scoring of signature gene sets of keratinocytes across disease conditions in Stereo-seq data (HC, n = 4; DLE, n = 5; SLE, n = 6). Scale bars, 1 mm for main images and 100 μm for magnified images. h Heatmap showing the expression of interferon (IFN)-related genes, cytokines, and chemokines in keratinocytes across disease conditions based on scRNA-seq data. i Left: Line graphs showing the expression of IL36G, CXCL10, and IFIH1 at different epidermal depths in lesional skin of HCs (n = 4), DLE (n = 5), and SLE (n = 6) based on Stereo-seq data. Right: Spatial visualization of inflammatory mediators on the left side of representative skin samples from HCs, DLE, and SLE patients. Scale bars, 1 mm for main images and 100 μm for magnified images. Data are presented as mean ± SEM (e, i) and boxplots with the median (horizontal line), the 25th and 75th percentiles (bounds of box), and the maximum and minimum values (whiskers) (g), and P values were determined by two-sided Wilcoxon rank-sum test (e, i) and two-sided Kruskal–Wallis test followed by Dunn’s multiple comparison test (g). *P < 0.05, **P < 0.01 (e, i).

We subsequently investigated the functional properties of lupus-enriched keratinocyte subpopulations. Granular KCs exhibited unique enrichment in keratinization and epidermal development, whereas stress KCs and cycling KCs were enriched in oxidative phosphorylation (Fig. 2f; Supplementary Data 5). Notably, OBs and stress KCs were enriched in the innate immune response, the cellular response to cytokine stimulus, the type II interferon response, and antigen presentation pathways, with more pronounced activation in the SLE group than in the DLE group (Fig. 2f; Supplementary Data 5). Spatially, keratinization was predominant in the upper epidermis, whereas interferon (IFN)-α and IFN-γ responses were predominant in the lower epidermis (Fig. 2g; Supplementary Data 4).

Recently, keratinocytes have been described as “cytokinocytes”, which contribute to the lupus lesions by producing IFNs and IFN-regulated cytokines and chemokines^13,14. To systematically identify inflammatory mediator sources in the local microenvironment, we investigated the expression of key cytokines and chemokines at spatial resolution. We initially evaluated the expression levels of established inflammatory mediators implicated in lupus pathogenesis in the scRNA-seq data split by disease conditions. Interestingly, most interferon-stimulated genes (ISGs), such as IFI27, IFITM1, IRF1, and BST2, were significantly enriched in both the DLE and SLE groups, with the highest expression levels detected in the SLE group (Fig. 2h). Other inflammatory mediators enriched in lupus lesions included IL20, IL34, and CXCL3, which were predominant in DLE patients, and IL36G, CCL20, CXCL9, CXCL10, and CXCL11, which predominated in the SLE group (Fig. 2h). Based on the Stereo-seq data, we then segmented the epidermis areas into 10 equal layers along the direction of the BM on each slide (details in Methods). Layer-specific analysis revealed significant enrichment of ISGs, including IFIH1, IFI44L, and IFITM1, throughout the whole epidermis in the DLE and SLE groups compared with the HC group (Fig. 2i; Supplementary Fig. 5). Intriguingly, cytokine gradients (IL36G, IL36RN, IL18, and IL1RN) in lupus lesions exhibited progressive attenuation from superficial epidermal layers to the BM, with gene expression levels decreasing from 0.40-0.75 to nearly 0 (Fig. 2i; Supplementary Fig. 5). Conversely, the expression of chemokines (CXCL9, CXCL10, and CXCL11) gradually increased from the superficial layers toward the BM (Fig. 2i; Supplementary Fig. 5).

Taken together, these findings reveal disease-specific inflammatory mediator signatures in the epidermal lesions of patients with DLE and SLE, characterized by stratified spatial distribution patterns.

Enrichment of stress keratinocytes at the epidermis-dermis interface in lupus lesions and their interactions with immune cells

A unique keratinocyte subset, stress KCs, was markedly increased in the lesional skin of patients with DLE and SLE (Fig. 2d). The transcriptional profile of these cells was defined by the marked enrichment of genes atypical for the interfollicular epidermis, including markers of UV damage (KRT6A), chemokines (CXCL10), and alarmins (S100A8) (Fig. 2b). Notably, alarmins, which are released during cellular stress, are early and sensitive biomarkers for monitoring inflammatory conditions¹⁵. Spatial mapping revealed that stress KCs were enriched proximal to the epidermis-dermis interface and perifollicular regions (Fig. 2c; Supplementary Fig. 4a, 6a). Intriguingly, a distinct keratinocyte subset, OBs, co-expressing CXCL10 but lacking stress keratins, was found to spatially surround the dermal stress KCs (Fig. 2b; Supplementary Fig. 6b).

To investigate the differentiation features of stress KCs, we performed pseudotime trajectory inference on keratinocyte subsets from all samples, except for pilosebaceous-related keratinocytes (details in Methods). Employing the Monocle2 tool¹⁶, we found that the stress KCs appeared to represent a transitional state in keratinocyte differentiation, originating from basal state to the terminally differentiated granular state (Supplementary Fig. 6c, d). Furthermore, known proliferation and differentiation dynamics of keratinocytes were observed from basal layers to granular layers (Supplementary Fig. 6c, d). These data indicated that stress KCs exhibited a transition dynamics characteristic, distinct from the normal keratinocyte differentiation trajectory.

Interestingly, while the overall proportion of stress KCs was comparable between SLE and DLE, their spatial distribution exhibited a higher cell density in SLE lesions, particularly within the epidermis (Supplementary Fig. 6e). Further underscoring their clinical relevance, we found a significantly positive correlation between the density of stress KCs in the epidermis of SLE lesions and the CLASI activity score (R = 0.89, P = 0.033) (Supplementary Fig. 6f). These findings suggested that stress KCs may play a more active role in mediating acute inflammation in SLE.

To further explore how stress KCs mediate immune infiltration, we employed CellChat to deconvolute intercellular networks (details in Methods)¹⁷. Lesional skin from both DLE and SLE patients exhibited a broad upregulation of inflammatory pathways compared to HCs (Fig. 3a). Critically, communication analysis revealed a significant enhancement of the CXCL9/10/11-CXCR3 signaling axis from stress KCs to immune cells in SLE compared to DLE (Fig. 3b; Supplementary Fig. 6g, h). Among the predicted interactions, the CXCL10-CXCR3 interaction was the most prominent, with its activity primarily involving T cells and plasma cells (Fig. 3b, c). This underscored the central role of CXCL10 in the immune cell recruitment driven by stress KCs. Notably, this signaling axis was also contributed by OBs, with stronger activity observed in SLE lesions (Fig. 3c). Spatial distance analysis further confirmed the proximity between stress KCs and T cells (Supplementary Fig. 6i).

Fig. 3. Stress keratinocytes enriched in lupus lesions mediate immune infiltration through IFN-γ-JAK-STAT-mediated production of CXCL9, CXCL10, and CXCL11.

a The relative information flow of all significant signaling pathways within the inferred networks in HCs, DLE, and SLE patients. b Bubble plots showing CXCL9/10/11-CXCR3 ligand-receptor pairs between stress KCs and T cells, NK cells, and plasma cells based on scRNA-seq data. Dot size indicates P value, colored by communication probability. P values were computed by one-sided permutation test. c Predicted cell-to-cell interactions of CXCL10-CXCR3 axis in DLE and SLE patients, respectively. The edge width is proportional to the inferred communication probabilities. d qPCR was performed to quantify the mRNA levels of chemokines (CXCL9, CXCL10, CXCL11), antimicrobial peptide (S100A8), UV damage (KRT6A), and JAK-STAT pathway (JAK2, STAT1, STAT2)-related genes in human primary keratinocytes stimulated with IFN-γ and/or irradiated with UVB in culture (n = 4 samples for each group). e Left: Schematic diagram of the T cell and B cell transwell migration assay using conditioned medium from treated keratinocytes. This image was created by Biorender.com. Right: Quantification of chemotactic effects of keratinocyte-conditioned media on activated T cells and B cells (n = 3 samples for each group). f Top: Schematic diagram of the experimental timeline of SLE mouse model. This image was created by Biorender.com. Bottom: Representative  flow cytometry plots and quantification of the frequency of immune cells within skin lesions of SLE mouse model following intradermal injection of CXCL10 protein (n = 6 samples for each group). All results are representative of at least three independent experiments. Data are presented as mean ± SEM (d, e, f), and P values were determined by one-way ANOVA followed by Holm–Sidak’s multiple comparisons test (d) and two-tailed unpaired Student’s t test (e, f).

The pivotal role of IFN-γ in driving the CXCL9/10/11-CXCR3 axis has been well established from studies on skin inflammation^18–20. However, the specific stimuli that drive transcriptional changes of stress KCs, including the upregulation of stress keratins and alarmins, remain unclear. In vitro treatment of human and mouse primary keratinocytes, we revealed that the transcriptional signatures of stress KCs (e.g., KRT6A, S100A8, CXCL10, and CXCL11) and interferon pathway genes (STAT1, STAT2, JAK2) were significantly upregulated upon co-stimulation with IFN-γ and UV (Fig. 3d; Supplementary Fig. 6l). Since UV radiation can trigger DNA damage cascade inducing apoptosis, we computed the signature score of apoptosis in keratinocytes (Supplementary Data 4). As expected, higher scores were observed in stress KCs and basal epidermis where stress KCs were mainly located (Supplementary Fig. 6j, k).

To figure out the specific role of key chemokine CXCL10 produced by stress KCs, we performed transwell migration assays, which confirmed that chemokines secreted by stress KCs promoted the dose-dependent migration of activated CD4⁺ T cells and B cells (Fig. 3e). Next, we administered intradermal CXCL10 injections in a spontaneous SLE mouse model. This treatment resulted in significant recruitment of CD45⁺ leukocytes, specifically enriching for helper T 1 (Th1) cells, tissue-resident memory T (Trm) cells, follicular helper T (Tfh) cells, plasma cells, and NK cells within lesional skin, whereas Th17 cell proportions were markedly reduced (Fig. 3f; Supplementary Fig. 7).

To validate our findings in the basal epidermal layer on other interface dermatitis diseases, we analyzed publicly available spatial transcriptomics data from lichen planus (LP) (Supplementary Data 3; details in Methods)²¹. As expected, LP lesions demonstrated a significant upregulation of ISGs in the epidermis, concomitant with increased expression of key inflammatory mediators, including IL18, IL36G, CXCL9, CXCL10, and CXCL11 (Supplementary Fig. 8a). Consistent with our data, depth-dependent changes in key chemokines CXCL9 and CXCL10 were observed in the epidermis of LP lesions (Supplementary Fig. 8d). Spatial cell deconvolution revealed that stress KCs with high expression of CXCL10 were enriched in the lower epidermis, while T cells and B cells highly expressing the CXCR3 receptor were concentrated in the upper dermis adjacent to the BM (Supplementary Fig. 8b, e). Noteworthy, we observed that T_IFN cells and B_IFN cells were the dominant immune cell subsets in interface dermatitis of LP, distinct from the enrichment of Trm cells, Tfh cells, and plasma cells observed in CLE (Supplementary Fig. 8c).

Collectively, these data indicate that the enrichment of stress KCs adjacent to the epidermis-dermis interface may represent a shared hallmark of interface dermatitis. However, the subsequent recruitment of pathogenic immune cells mediated by stress KCs is differential. In lupus, particularly in SLE lesions, stress KCs contribute more actively to acute inflammation, which may coordinate the recruitment of adaptive immune cells through the IFN-γ-JAK-STAT-CXCL9/10/11-CXCR3 axis.

TLSs are present in the dermis of chronic lupus lesions

Spatially, dense immune cell infiltration was observed surrounding follicular and vascular structures within lupus lesional skin (Fig. 4a). To profile the spatial organization of the immune landscape, we analyzed the fractions and features of immune cells in dermal regions along the vertical direction of the BM, with each layer in an area 250 µm wide parallel to the BM (details in Methods). We observed significant immune cell infiltration in the superficial dermis compared with the deep dermal side (Fig. 4b, c). Specifically, macrophages/DCs were the most abundant cell types throughout the lesional dermis, exhibiting maximal diversity in the BM-proximal layer (0-250 µm) (Fig. 4b, c). The diversity of plasma cell abundance progressively decreased with increasing distance from the BM, whereas T cells and B cells showed preferential localization within the 250-500 µm stratum (Fig. 4b, c). Notably, dermal densities of T and B cells demonstrated a strong positive correlation with lesion duration (R = 0.84, P = 0.0011) (Supplementary Fig. 9a). Although a similar positive trend was observed between total dermal immune cell density and lesion duration, this association did not reach statistical significance (R = 0.56, P = 0.075) (Supplementary Fig. 9a). These findings revealed a disease-associated reorganization of both immune cell density and spatial distribution patterns in lupus lesions.

Fig. 4. Spatial heterogeneity of immune microenvironment in skin lesions of patients with DLE and SLE.

a Top: Representative H&E staining of dermal immune microenvironment in SLE. Bottom: Representative spatial visualization of immune niche, HF niche, and endothelial cells in dermal region of skin lesion of SLE. The H&E staining images shown represent the results of 3 independent experiments. Scale bars, 1 mm for main images and 200 μm for magnified images. b Line graphs showing the fractions of immune cell subtypes at different distances to the BM as determined using Stereo-seq data (n = 11 samples of LE patients). Regions greater than 1500 µm from the BM were merged into one area. c Spatial distribution of major immune cell types relative to the BM. Scale bar, 200 μm. d Barplots showing the percentages of T cells, B cells, NK cells, plasma cells, and macrophages/DCs within each cluster of immune cell niche. e Bar plots showing the percentage of each immune cell niche cluster across disease conditions. f Heatmap showing the cell type compositions and fractions of the 8 immune niche clusters in each skin sample. g Top: Spatial visualization of immune niche clusters in representative skin lesions of HC, DLE, and SLE patients. Bottom: Magnification images showing spatial distribution of major immune cell types in regions of cluster 6 and 7 in patients with DLE and SLE, respectively. Scale bars, 1 mm for main images and 200 μm for magnified images. h Violin plots showing TLS signatures scores across immune cell niche clusters under different disease conditions. TLS, tertiary lymphoid structure. i Spatial visualization of TLS score in representative lesional skin of HC, DLE, and SLE patients. Scale bars, 1 mm. j Multiplexed IHC staining of CD3 (red) and CD20 (green) in paraffin section of skin from patients with DLE and SLE. The images shown represent the results of 3 independent experiments. Scale bars, 200 μm for main images, 75 μm for DLE5 magnified image, and 50 μm for SLE5 magnified image. k Correlation analysis between the numbers of TLS and duration of lesions in LE patients with TLS structures (n = 9). l Volcano plot showing differentially expressed genes between the TLS region and the region 200 μm from TLS boundary. Genes with adjusted P value < 0.05 and |log₂(fold change)| > 0.25 were color coded. m Bar plot showing enriched pathways for TLS⁺ and TLS^- regions. Data are presented as mean ± SEM (b), and P values were determined by two-sided Kruskal–Wallis test followed by Dunn’s multiple comparison test (b) and Spearman’s rank correlation (k). *P < 0.05, **P < 0.01, ***P < 0.001,  ****P < 0.0001 (b).

To further characterize the immune microenvironment architecture in lupus lesions, we computed immune cell frequencies within a 150 µm radius surrounding all dermal immune cells in the sample (details in Methods). Leiden clustering of the cell-neighbor matrix initially revealed 17 distinct clusters (Supplementary Fig. 9b). Following the manual merging of compositionally similar clusters, we finally defined 8 immune microenvironment subregions with unique cellular compositions and spatial distributions (Fig. 4d). Interestingly, the healthy skin samples consisted of only 4 subregions, including clusters 1, 3, 4, and 5, whereas clusters 6 and 7 were specifically enriched in skin lesions from both DLE and SLE, with a higher proportion in DLE (Fig. 4e, f).

TLSs are increasingly described in settings of chronic inflammation and autoimmunity, forming ectopic lymphoid-like structures that are composed of well-organized T cells and B cells^22,23. A hallmark feature of TLSs is the ectopic expression of genes encoding lymphotoxin and lymphoid chemokines, which are critically involved in secondary lymphoid organ development²⁴. We found that the cellular composition and appearance of clusters 6 and 7 presented characteristics of immune aggregates resembling TLSs (Fig. 4d, g). Specifically, clusters 6 and 7 displayed an organized lymphoid architecture with B-cell cores surrounded by T-cell zones and peripherally distributed plasma cells (Fig. 4g). To provide supportive evidence for the presence of TLS-like structures in lupus lesions, we analyzed published TLS-related gene signatures on spatial slides (Supplementary Data 4)²⁵. Subregions with high TLS activity scores were found in the spatial distribution of clusters 6 and 7 (Fig. 4h, i). In addition, clusters 2 and 8 also presented high TLS activity, while both clusters were confined to a single sample, with greater infiltration of macrophages/DCs and lower B-cell density, which differed from that of canonical TLSs (Supplementary Fig. 9c). Multiplex immunohistochemistry (mIHC) further validated the presence of TLSs in lesional skin by labeling CD3⁺ T cells and CD20⁺ B cells in FFPE samples (Fig. 4j). In summary, we identified the presence of TLSs in lupus lesions and proposed dividing the slides into TLS⁺ and TLS^- regions for further analysis. TLS⁺ regions were identified in 9 of the 11 patients, revealing a significant positive correlation between the duration of lesions and the number of TLSs per unit area (R = 0.61, P = 0.049), whereas no significant association was detected with the TLS area (R = 0.58, P = 0.1) (Fig. 4k; Supplementary Fig. 9d). Notably, 2 patients with SLE tested negative for TLSs, with lesions persisting for 0.25 and 7 months (Supplementary Data 1).

To explore the mechanism of TLS formation and function in lupus lesions, we designated areas containing clusters 6 and 7 as TLS⁺ regions and 200-µm-wide areas away from the border of the TLS zone as TLS^- regions to obtain local bulk RNA profiles (Supplementary Fig. 9e; details in Methods). DEG analysis of immune cells in the TLS⁺ regions and TLS^- regions revealed that genes related to lymphocyte activation, B-cell activation, the antigen receptor-mediated signaling pathway, and hematopoietic or lymphoid organ development were enriched in the TLS⁺ regions (Fig. 4l, m; Supplementary Data 6, 7). Moreover, genes associated with chemotaxis- and migration-related pathways, including positive regulation of cell migration, positive regulation of locomotion and chemotaxis, and regulation of cell-cell adhesion, were upregulated in the TLS-regions (Fig. 4l, m; Supplementary Data 6, 7).

Together, these findings confirm the prevalence of structured TLSs in chronic lesions of patients with DLE and SLE.

Spatial localization of ABCs, pre-GC B cells, Bn, Tph/Tfh cells, Treg cells, and Tm cells within TLS structures of lupus lesions

We further subdivided 39,845 T cells, 6459 B/plasma cells, and 6208 macrophages/DCs into 25 subsets, including 9 T-cell subsets, 6 B/plasma cell subsets, and 10 macrophage/DC subsets to delineate the cellular and molecular characteristics of TLS structures in lupus patients (Fig. 5a; Supplementary Fig. 10a). We first investigated the spatial distribution of immune cell subpopulations across disease conditions and observed significant lymphocyte enrichment in patients with DLE and SLE compared with HCs, as well as a preferential distribution of T-cell and B-cell subsets in DLE lesions than in SLE lesions (Fig. 5b). Next, we explored the heterogeneity of immune cell subsets in the TLS and non-TLS regions of lupus skin. Compared with those in non-TLS regions, we detected increased proportions of memory T (Tm) cells, regulatory T cells (Tregs), T peripheral helper and/or T follicular helper (Tph/Tfh) cells, IFN T (T_IFN) cells, cytotoxic T (Tc) cells, γδT cells, naïve B cells (Bn), age-associated B cells (B_ABCs), and IFN B (B_IFN) cells in TLS regions (Fig. 5c; Supplementary Fig. 10b). We also detected a decreasing trend in monocyte-like macrophages (Mac_Mono), C1Q-related macrophages (Mac_C1Q), resident tissue macrophages (Mac_RT), and IFN macrophages (Mac_IFN) in TLS regions compared with non-TLS regions, but no significant differences in the distributions of plasma cells or other macrophage/DC subsets were found (Supplementary Fig. 10b, c).

Fig. 5. Spatial cellular composition and molecular characteristics of TLS structures in lesional skin of patients with DLE and SLE.

a UMAP plot showing subtypes of 39,845 T cells, 6459 B/plasma cells, and 6208 macrophages/DCs in scRNA-seq data. Tn, naïve T cell; Tm, memory T cell; Trm, tissue-resident memory T cell; Treg, regulatory T cell; Tph/Tfh, T peripheral helper and/or T follicular helper cell; T_IFN, IFN T cell; T_STR, stressed T cell; Bn, naïve B cell; B_ABC, age-associated B cell; B_IFN, IFN B cell; B_STR, stressed B cell; B_preGC, precursor of germinal center B cell; Mac_Mono, monocyte-like macrophage; Mac_C1Q, C1Q-related macrophage; Mac_RT, resident tissue macrophage; Mac_IFN, IFN macrophage. b Heatmap showing the phenotype preferences of subtypes by Ro/e based on Stereo-seq data. c Box plots showing the percentage of immune cell subtypes between TLS (n = 9) and non-TLS regions (n = 11). d Heatmap showing the correlation of cell density between pairs of subtypes of T cells and B/plasma cells in TLS regions. Others represent all macrophages/DCs. e Heatmap showing the median distance observed between pairs of neighboring cells in TLS regions. Median distances < 30 μm were labeled with asterisk symbol. f Spatial distribution of immune cell subtypes of interest in representative TLS region. Scale bars, 150 μm for the main image and 25 μm for the magnified image. g Line graphs showing the mean cell density of Tm, Treg, Tph/Tfh, and B_preGC at different distances from B_ABC. h Multiplexed IHC staining of CD20 (red) and T-bet (yellow) in paraffin section of skin from patients with LE. The images shown represent the results of 3 independent experiments. Scale bars, 200 μm for the main image and 50 μm for the magnified image. i Heatmap showing the regulon activity of top 10 transcription factors in B_ABC across B cell subsets based on scRNA-seq data. j Spatial visualization of TF activity of TFEB and BATF in representative TLS region. TF, transcription factor. Scale bars, 75 μm. k Biological functional classification and network analysis of targeted genes regulated by TFEB and BATF. l Bubble plots showing ligand-receptor pairs between TLS feature subsets based on scRNA-seq data. Dot size indicates P value, colored by communication probability. P values were computed by one-sided permutation test. m Spatial visualization of the expression of CD86-CD28, CD40LG-CD40, CXCL13-CXCR5 in T and B cells within representative TLS region. Scale bars, 75 μm. n Multiplexed IHC staining of CD4 (green), FOXP3 (red), and CXCL13 (yellow) in paraffin section of skin from patients with LE. The images shown represent the results of 3 independent experiments. Scale bars, 200 μm for the main image and 50 μm for the magnified image. Data are presented as boxplots with the median (horizontal line), the 25th and 75th percentiles (bounds of box), and the maximum and minimum values (whiskers) (c) and mean (g), and P values were determined by two-sided Wilcoxon rank-sum test (c) and Spearman’s rank correlation (d). *P < 0.05,  **P < 0.01, ***P < 0.001 (d).

To characterize the potential spatial co-occurrence patterns of pathogenic immune cell types in TLSs, we conducted a regional cell density correlation analysis on lymphocyte subsets localized in TLS regions of lupus lesions (details in Methods). The most significant spatial correlations of cell density were observed between precursors of germinal center B cells (B_pre-GCs) and B_ABCs and between Tph/Tfh cells and Tregs (Fig. 5d; Supplementary Data 8a). Other positive correlations were detected between the cell density of Bn cells and B_pre-GCs, B_IFN cells, B_STR cells, and B_ABCs; between B_STR cells and Tn cells, Trm cells, B_ABCs, B_IFN cells, and B_pre-GCs; and between Tm cells and naïve T (Tn) cells, cytotoxic T (Tc) cells, γδT cells, B_IFN cells, and B_pre-GCs (Fig. 5d; Supplementary Data 8a). We then analyzed the spatial distances among major immune cell subsets within the TLS structures of lupus lesional skin (details in Methods). Minimal distances were revealed between pairs of lymphocytes (median of 20 µm), whereas maximal distances were observed between B_STR cells and plasma cells (median of 80 µm) (Fig. 5e; Supplementary Data 8b). We found that the median distances between B_ABCs and Tm cells, Tregs, and Bn cells; between B_pre-GCs and Tm cells, Tregs, Bn cells, B_ABCs, and B_IFN cells; and between Tregs and Tph/Tfh cells, T_IFN cells, B_ABCs, B_IFN cells, and B_pre-GCs were physically proximate (Fig. 5e; Supplementary Data 8b). These observations suggested that TLSs in lupus skin lesions were characterized by enrichment of signature populations, including B_ABCs, B_pre-GCs, Bn cells, Tph/Tfh cells, Tregs, and Tm cells.

To further confirm the cellular composition of TLSs in lupus lesions, we leveraged an independent CLE cohort profiled by NanoString Digital Spatial Profiling (DSP) (Supplementary Data 3; details in Methods)²⁶. We observed robust TLS signatures in CD3⁺ and CD45⁺CD3^- dermal immune cell regions of CLE lesions, especially in DLE (Supplementary Fig. 11b). Cellular deconvolution analysis further identified the expansion of Tn cells, Tregs, Trm cells, Tc cells, Bn cells, B_ABCs, and plasma cells in DLE lesions (Supplementary Fig. 11c). Although the predefined regions of interest (ROIs) did not capture enrichment for Tph/Tfh cells, Tm cells, and B_pre-GCs subsets, the high consistency of most features underscores the generalizability of our main findings.

Spatial visualization of the dominant infiltrating immune cell types in the TLS structures of lupus lesions revealed a well-organized distribution pattern, with B_pre-GCs, Bn cells, and B_ABCs predominantly enriched in the center of TLSs, whereas Tregs, Tph/Tfh cells, and Tm cells were scattered throughout (Fig. 5f). Immunofluorescence costaining of the B-cell marker CD20 and the representative transcription factor T-bet verified the abundance of B_ABCs in lupus skin (Fig. 5h). We also calculated the density gradients of Tregs, Tm cells, Tph/Tfh cells, Bn cells, and B_pre-GCs radiating from B_ABCs in TLS regions and found that the densities of the detected cell components gradually decreased with increasing distance from B_ABCs (Fig. 5g). In accordance with our previous findings, we observed a greater density of B_pre-GCs adjacent to B_ABCs than of Tregs, Tph/Tfh cells, Tm cells, and Bn cells (Fig. 5g). Moreover, the cell density of B_pre-GCs decreased 20-30 µm away from B_ABCs, where Tregs were more abundant in these TLS regions (Fig. 5g, n).

Given the critical role of B-cell development in extrafollicular responses in inflamed tissues²⁷, we used single-cell regulatory network inference and clustering (SCENIC) analysis to analyze TF regulon activity across B-cell subsets in the scRNA-seq data (details in Methods)²⁸. As B_ABCs are a hallmark of the extrafollicular pathway, we focused on the top 10 most activated TFs in these cells. Notably, BATF and TFEB, which regulate class-switch recombination and B-cell antigen receptor activation in B cells, were particularly enriched in B_ABCs within TLS regions of lupus lesions (Fig. 5i, j; Supplementary Fig. 10d). Moreover, these TFs were predicted to collectively regulate a network of target genes involved in key processes such as the cell cycle, autophagy, immune response activation, and lymphocyte activation (Fig. 5k; Supplementary Data 9). Our findings further supported the established characteristics of B_ABCs, including their long-term survival capacity and hyperactivated state^29,30.

Additionally, we explored the potential cellular crosstalk within lupus TLS structures via CellChat analysis (details in Methods)¹⁷. Among the ligand-receptor pairs pertaining to B_ABCs and B_pre-GCs, CD86-CD28 and CD86-CTLA4 were significantly enriched in Tregs and Tph‒Tfh cells, highlighting the role of B_ABCs in activating Tregs and Tph/Tfh cells (Fig. 5l, m). In addition, interactions associated with cell activation were also observed between Tph/Tfh cells and Bn cells, B_ABCs, and B_pre-GCs through the CD40LG‒CD40 axis (Fig. 5l, m; Supplementary Fig. 10e). Interestingly, CXCL13 could bind to CXCR5 expressed by B_ABCs and B_pre-GCs, which was secreted by Tph/Tfh cells, thereby facilitating the immune cells recruitment to TLS niches (Fig. 5l, m; Supplementary Fig. 10e).

Collectively, these results delineate the cellular compositions and molecular signatures of TLS-associated immune cell subsets in chronic cutaneous lesions of DLE and SLE patients.

Spatial localization of stromal cells within TLS structures of lupus lesions

Previous studies have highlighted the involvement of specialized vascular structures, including high endothelial venules (HEVs) and lymphatic vessels (LVs), in facilitating cellular homing and migration within secondary lymphoid organs such as lymph nodes (LNs)^31,32. mIHC revealed the existence of blood vessels in the center and at the edges of organized lymphoid structures (Fig. 6m). Hence, we hypothesized that there could be communication pathways between immune cells and vascular endothelial cells (VECs), potentially instigating immune infiltration and subsequent formation of TLSs. To validate this hypothesis, we regrouped the stromal cells of 26,541 fibroblasts (FBs) and 4583 endothelial cells (ECs) into 10 and 4 cell subpopulations, respectively (Fig. 6a; Supplementary Fig. 12a). Expansion of VEC_CCL14⁺, FB_Cycling, and FB_Inflamed and decreased abundance of FB_COL11A1⁺ and FB_SFRP2⁺ were observed in the TLS regions compared with the non-TLS regions (Fig. 6b). To characterize the spatial distribution of stromal cells, we quantified the cell density within 100 µm of the inner and outer TLS boundaries (details in Methods). VEC_CCL14⁺ predominantly localized to the center of the TLS and sharply decreased beyond 20 µm outside the TLS border (Fig. 6c, d; Supplementary Fig. 12c). We also noted that FB_IFN was markedly reduced at the TLS boundary, although this difference was not statistically significant (Supplementary Fig. 12c).

Fig. 6. Spatial characteristics of stromal cells within TLS region in patients with DLE and SLE.

a UMAP plot showing subtypes of 26,541 fibroblasts and 4583 endothelial cells in scRNA-seq data. FB, fibroblast; Inflam, inflamed; VEC, vascular endothelial cell; LEC, lymphatic endothelial cell. b Barplots showing the composition of subtypes of fibroblast and endothelial cell in TLS and non-TLS region, respectively. c Line graph showing the density of VEC_CCL14⁺ at different distances from the boundary of TLS (n = 8 samples of TLS regions containing VEC_CCL14⁺ cells). d Representative spatial visualization of the distribution of fibroblast and endothelial cell subtypes and TLS structures. Scale bars, 500 μm for the main image and 125 μm for the magnified image. e Line graph showing the fraction of FB_IFN subset at different distances to the BM as determined using Stereo-seq data (n = 11 samples of LE patients). f Representative spatial distribution of FB_IFN subset and TLS boundaries in skin lesion of LE patients. Scale bar, 500 μm. g Violin plot showing the expression of adhesion and chemotaxis signatures across endothelial cell and fibroblast subtypes, respectively, based on scRNA-seq data. h Bubble plots showing ligand-receptor pairs between TLS feature subsets and endothelial cell subtypes based on Stereo-seq data. Dot size indicates P values, colored by communication probability. P values were computed by one-sided permutation test. i Predicted interactions of CXCL12-CXCR4 axis between major immune cell types and fibroblast subtypes in HC and LE patients, respectively. The edge width is proportional to the inferred communication probabilities. j Box plot showing the distance of FB_IFN subset to major immune cell types (n = 11 samples of LE patients). k Left: Correlation analysis between the density of VEC_CCL14⁺ subset and the density of Tn subset in LE patients. Middle: Correlation analysis between the density of VEC_CCL14⁺ subset and the density of  Tph/Tfh subset in LE patients. Right: Correlation analysis between the density of FB_IFN subset and the density of macrophages/DCs in LE patients. l Left: Spatial visualization of the expression of SELPLG-SELE in T cells and VEC_CCL14⁺ subset within  representative TLS region. Right: Spatial visualization of the expression of CXCL12-CXCR4 in fibroblasts and major immune cell types in representative skin lesion of LE patients. Scale bars, 1 mm for main images, and 100 μm and 20 μm for magnified images. m Multiplexed IHC staining of CD3 (red), CD20 (green), and CD31 (orange) in paraffin section of skin from patients with LE. The images shown represent the results of 3 independent experiments. Scale bars, 200 μm for the main image and 50 μm for the magnified image. Data are presented as mean ± SEM (c, e) and boxplots with the median (horizontal line), the 25th and 75th percentiles (bounds of box), and the maximum and minimum values (whiskers) (j), and P values were determined by two-sided Wilcoxon rank-sum test (other distances to TLS boundary vs −20/20 μm to TLS boundary) (c), two-sided Kruskal–Wallis test followed by Dunn’s multiple comparison test (e) and Spearman’s rank correlation (k).

To illustrate the role of TLS-enriched stromal cells in TLS development, we calculated gene set scores for the adhesion and chemotaxis modules across all EC and fibroblast clusters, respectively (Supplementary Data 4). VEC_CCL14⁺ presented the highest adhesion scores, whereas elevated chemotaxis scores were observed in the FB_cycling, FB_IFN, and FB_COL23A1⁺ populations (Fig. 6g). Next, we conducted spatial cell-cell communication analysis between VEC_CCL14⁺ and TLS-associated immune cells using COMMOT (details in Methods)³³. In terms of signaling from EC subsets to immune cells, the SELE-CD44, ICAM1-SPN, and ICAM1-ITGAL pairs were notably enriched specifically in VEC_CCL14⁺ in comparison with other EC clusters (Fig. 6h). Additionally, for EC clusters that emerged as signaling targets, increased communication from Tn cells, Tm cells, Tregs, Tph/Tfh cells, Bn cells, B_ABCs, and B_pre-GCs to VEC_CCL14⁺ via the SELPLG-SELE pair was observed (Fig. 6h, l). Notably, VEC_CCL14⁺ cell density was significantly correlated with Tn cells (R = 0.67, P = 0.039), while its association with Tph/Tfh cells did not reach statistical significance (R = 0.55, P = 0.087) (Fig. 6k). These findings imply that VEC_CCL14⁺ might play major roles in TLS formation and activation by promoting immune cell infiltration.

Furthermore, FB_IFN was specifically enriched in lupus lesions, constituting a predominant stromal cell population with a widespread distribution throughout the superficial dermis (Fig. 6b, f; Supplementary Fig. 12b). As expected, significant enrichment of FB_IFN was observed within the first layer (0-250 µm) closest to the BM boundary, where the spatial distribution predominantly coincided with dermal infiltration of macrophages/DCs (Fig. 6e-j). Cellular interactions between fibroblasts and immune cells according to the scRNA-seq data revealed that intercellular communication, especially that involving the CXCL12-CXCR4 and CCL19-CCR7 axes, was more complex in lupus lesions than in HCs (Fig. 6i; Supplementary Fig. 12e). Subsequently, we examined the spatial expression of the interactive signals observed above and found that, compared with those in HCs, the ligand-related genes CXCL12 and CCL19 were highly expressed in fibroblasts from LE patients, especially in FB_IFN, and the receptor-related genes CXCR4 and CCR7 were enriched in T cells, NK cells, B cells, and macrophages/DCs (Fig. 6l; Supplementary Fig. 12f, g). Notably, significant positive interactions were observed between the density of FB_IFN and macrophages/DCs (R = 0.65, P = 0.037) and between the density of FB_IFN and the density of NK cells on the basis of the Stereo-seq data (R = 0.76, P = 0.0092) (Fig. 6k; Supplementary Fig. 12d). These findings highlighted the ability of FB_IFN to mediate diffuse immune cell infiltration, particularly of myeloid cells, within a reticular network structure, in contrast with the TLS-enriched distribution of VEC_CCL14⁺.

Taken together, our findings imply that VEC_CCL14⁺ and FB_IFN cells formed specialized stromal niches that coordinated TLS development in lupus skin lesions.

Discussion

Our study revealed the spatial, cellular and transcriptional signatures of the skin ecosystem in lupus patients and identified a network of cell-cell interactions involved in the pathogenesis of lupus, highlighting the dominant contributions of keratinocyte dysregulation and immune cell infiltration. By combining scRNA-seq and Stereo-seq, we demonstrated that keratinocytes in different differentiation states at lesional sites presented cell-specific expression patterns of inflammatory mediators. Notably, we identified a specific subset of keratinocytes, the stress KCs, which mediated the infiltration of Th1 cells, Trm cells, Tfh cells, and plasma cells at the epidermal-dermal interface, thereby driving acute inflammation in lesional skin of lupus, especially in active SLE lesions. Furthermore, our data shed light on the role of TLS structures in the persistence of localized chronic lesions in lupus patients and identified the spatial preferences of pathogenic TLS components, including B_ABCs, B_pre-GCs, Bn cells, Tph/Tfh cells, Tregs, Tm cells, and VEC_CCL14⁺ cells, which may be the potential pathogenesis of chronic refractory lupus lesions.

Lupus lesions were initially considered to be caused by aberrant apoptosis of keratinocytes exposed to UV radiation or other damaging triggers, followed by debris generation and chemokine production, which has been shown to activate the immune system^4,34,35. Here, we found that keratinocytes in different differentiation states strikingly expressed diverse transcripts of inflammatory mediators. Spatially, the strongest cytokine responses (e.g., IL36G, IL36RN, IL18, and IL1RN) in lupus lesions were in the granular layer of the upper epidermis, while chemokines (e.g., CXCL9, CXCL10, and CXCL11) were most prominently expressed in the basal layer of the lupus epidermis. Published studies have shown that the overexpression of disease-promoting cytokines is involved in the pathogenesis of CLE, whereas these cytokines are mostly derived from immune cells^36–38. Although the expression of cytokine transcripts was limited in skin lesions, we successfully detected them in spatial-specific patterns in lupus keratinocytes. Interestingly, cytokines enriched in the upper epidermis of lupus lesions all belonged to the IL-1 superfamily. The role of IL-1 family cytokines in proinflammatory responses is well established³⁹. These cytokines are excellent sensors of environmental threats, especially considering their distribution in the outer layers of the granular epidermis. IL-36G and IL-18 remain inactive under steady-state conditions but are upregulated in the context of contact with invading pathogens following microbial infection or mechanical damage, thereby generating a strong inflammatory response^40,41. Notably, IL-18 may upregulate the production of CXCL9, CXCL10, and CXCL11 in keratinocytes by activating the PI3K/Akt signaling pathway⁴².

Intriguingly, we identified a unique subpopulation of keratinocytes, stress KCs, which were located adjacent to the epidermal-dermal interface and communicated with T cells, NK cells, and plasma cells with the CXCL10-CXCR3 axis. Established studies have revealed the significant role of the IFN-γ-induced chemokines CXCL9/10/11 in the recruitment and maintenance of T-cell infiltrates in the tissue microenvironments of inflammatory skin diseases^43–46. However, our study provided a critical conceptual advance by resolving this biology at the single-cell level. First, we demonstrated that this pathogenic chemokine production was not ubiquitous across the epidermis but was highly restricted to a discrete, transcriptionally defined subpopulation of epidermal stress KCs. Mechanistically, we delineated the specific environmental-immune synergy governing this phenotype. While stress KCs expressing alarmins have been identified as sensitive biomarkers of cellular stress, we experimentally demonstrated that the combination of UVB irradiation and IFN-γ stimulation significantly upregulated the expression of signature markers of stress KCs. Clinically, the positive correlation between epidermal stress KC density and CLASI activity scores in SLE patients implied that this cellular state may be more involved in the acute inflammatory infiltration of SLE. While the CLASI scores reflect the overall severity of skin lesions, we believed that the observed cellular changes in the representative SLE lesions supported the notion that stress KCs may contribute to the acute inflammatory response of active SLE lesions characterized by erythema and edema. Finally, we also compared our data to a recently published study focusing on LP, which was recognized as the common interface dermatitis diseases²¹. Our results revealed that the enrichment of stress KCs at the epidermis-dermis interface may represent a shared hallmark of interface dermatitis. However, our data highlighted that the subsequent recruitment of pathogenic immune cells driven by stress KCs in interface dermatitis was differential. The pronounced expansion of T_IFN cells and B_IFN cells was observed highly specific to the superficial dermis of LP, while increased infiltration of Trm cells, Tfh cells, and plasma cells were uniquely observed in CLE lesions. These results underlay the distinct immunopathogenic mechanisms mediated by stress KCs in CLE and LP.

TLSs manifest as ectopic lymphoid aggregates that are formed in nonlymphoid organs under chronic inflammatory conditions, such as infection, cancer, and autoimmune disease^22,23,47. However, it is crucial to distinguish their different functional implications under the context of various diseases. In the tumor microenvironment, TLSs are frequently associated with a favorable prognosis, serving as local hubs for generating robust anti-tumor immunity⁴⁸. In contrast, in autoimmune diseases, TLSs serve as pathogenic hubs that drive tissue destruction by facilitating the local activation of autoreactive lymphocytes and autoantibody production, consistently correlating with a poor prognosis⁴⁹. However, the phenotypes, functions, and clinical significance of TLSs in the skin tissues of patients with lupus, as well as their pathogenic contributions to disease progression, remain incompletely characterized. Our analysis confirmed the presence of TLS structures in chronic lupus lesions, where TLS density showed a direct correlation with lesion duration, consistent with the notion that TLS development is a gradual, chronic process.

At the cellular level, our spatial analysis unveiled a pathogenic architecture within lupus TLSs driven by specific T- and B-cell subsets. We identified a spatially organized network where B_pre-GCs, B_ABCs, and Bn cells accumulated in B-cell follicles at the center of TLSs, while Tregs, Tph/Tfh cells, and Tm cells were dispersed in surrounding T-cell-rich areas. Notably, ABCs are signatures of extrafollicular B-cell responses and are commonly described in autoimmune diseases and chronic infections, with the ability to differentiate into plasma cells to be produced upon antigen reencounter or innate stimulation³⁰. In the present study, we discovered that the local activation of B_ABCs in lupus lesions was enriched in the immune response, lymphocyte activation, autophagy, and the cell cycle. These features were consistent with the existing published characterization of ABCs, including hyperactivation, long-term survival, and migration to inflamed tissues^29,50,51. Furthermore, we observed a critical interaction between B_ABCs and Tph/Tfh cells via the CXCL13-CXCR5 axis, indicating the pathogenic contribution of CXCL13 in TLSs. Several studies have reported that the administration of recombinant CXCL13 induces TLS progression in cancer models^52–54. Pathogenic expansion of CXC13⁺CD4⁺ T cells in TLSs has been identified in joint tissues from rheumatoid arthritis patients and blisters from pemphigus patients, and these cells were observed to interact with Tregs or B cells through CD153-CD30 signaling^55,56

Stromal cells, encompassing fibroblasts, blood and lymphatic endothelial cells, have been recognized as vital coordinators of immune responses and contributors to disease persistence, extending far beyond their traditional architectural function⁵⁷. Here, we identified a distinct vascular endothelial cell subset, the VEC_CCL14⁺ cells, which was characterized by the expression of key selectins and adhesion molecules and was spatially enclosed within organized lymphocyte aggregates in the TLSs of lupus lesions. This anatomical positioning strongly suggested that VEC_CCL14⁺ cells were not passive bystanders but active participants in inflammation, potentially by regulating lymphocytes recruitment into tissues along with TLS formation. Supporting this notion, peripheral lymphocytes, including Tn cells, Tregs, Tph/Tfh cells, B_ABCs, and B_pre-GCs, were observed to be recruited into TLSs in a manner dependent on SELE-CD44, ICAM1-SPN, and ICAM1-ITGAL. Previous studies have found that CCL21 is highly expressed in endothelial cells, particularly in lymphatic endothelial cells, which may contribute to the migration of CCR7⁺ infiltrating T cells in autoimmune skin diseases^58,59. However, recent research has increasingly recognized that vascular endothelial cells, rather than lymphatic endothelial cells, participated in an unexpected role in the initiation of TLSs, which may involve in the loss of Notch signaling or aberrant activation of the cGAS-STING pathway^60,61. Further investigation is warranted to elucidate the detailed molecular mechanisms and therapeutic potential of vascular endothelial cells, especially VEC_CCL14⁺ cells, underlying in the TLS formation of lupus.

We acknowledge several limitations inherent in our study. First, our spatial analyses were based on a small number of lesion samples from patients with DLE and SLE. To address this directly, we validated our findings on an independent public dataset of CLE lesions. Future longitudinal studies on serial biopsies of lesional and nonlesional skin at different disease stages will help address this caveat. Second, despite the high nanoscale resolution of Stereo-seq, the performance of the cell segmentation was limited, especially due to difficulties in identifying cell boundaries in sections with dense lymphocyte aggregates. We employed strict quality control standards, filtering out Cellbins with an area greater than 2500 to mitigate potential artifacts and ensuring the reliability of our data. Notably, an emerging method called TopACT has been developed to automatically identify cell types without pre-defined cell boundaries, thus bypassing the segmentation step⁶². This tool is promising for directly addressing this issue in future studies. Third, while we have detailed the cellular and molecular landscape of TLSs in lupus lesions, the absence of direct functional validation for our mechanistic interpretation regarding TLSs in lupus lesions, which temper definitive conclusions. Future mechanistic investigations will help to understand the upstream triggers of TLS formation and establish their causal role in disease progression.

In conclusion, the combination of spatial transcriptomics and single-cell transcriptomics analysis enabled a comprehensive spatial profile of lesional skin in patients with DLE and SLE. We demonstrated unique spatial molecular signatures in different layers of epidermal lesions and identified the pathogenic role of TLS structures in chronic inflammatory lupus lesions, with TLS formation mediated by cross-talk between immune cells and endothelial cells. Considering lupus as a spectrum disease, we speculate that the distinct clinical manifestations of lesional skin in lupus may be mediated by different dominant pathogenic mechanisms. Specifically, the acute active inflammation in SLE is more associated with stress KCs, whereas the chronic damaging inflammation in DLE is more related to the presence and persistence of TLSs. Based on these features, our findings can be translated into potential therapeutic strategies, such as the stress KCs and VEC_CCL14⁺ subsets, which can be selectively modulated to enable personalized interventions for the lupus spectrum.

Methods

Human participants

This study was approved by the Medical Ethics Committee of the Second Xiangya Hospital of Central South University (No. 2019-30-044), the Xiangya Hospital of Central South University, and the Institute of Dermatology (No. 201212074), Chinese Academy of Medical Sciences (No. 2021-KY-054). All the samples included in this study were obtained from patients and healthy controls who granted their consent for the utilization of their tissue for research purposes. Representative skin biopsy specimens were collected from patients diagnosed with DLE and SLE based on clinical characteristics, laboratory examinations, and histopathological tests. Specifically, all enrolled SLE patients presented with active CLE lesions. Among them, 3 patients exhibited the classic malar rash, while the remaining 3 patients displayed active erythematous rash characterized by distinct edema and infiltration. In contrast, the lesions from DLE patients presented as localized discoid erythematous plaques with distinct dyspigmentation, scarring, and atrophy. Matched HCs were obtained from healthy skin samples from sun-exposed areas adjacent to the nevus of subjects who underwent nevus removal surgery. The detailed demographic and clinical information for the Stereo-seq and scRNA-seq samples is provided in Supplementary Data 1 and 2, respectively. The CLE disease area and severity index (CLASI) scores were evaluated in patients with DLE and SLE by three dermatologists⁶³.

scRNA-seq data processing

We utilized our published scRNA-seq dataset of cutaneous lesions in LE (GSE179633). Further quality control and cell annotation analysis were performed using Seurat (v4.3.0)¹². First, we performed cell filtering, retaining cells with 400 to 4000 detected genes, UMI counts greater than 1000, and mitochondrial percentage less than 20%. Then, the top 2000 highly variable genes were selected for principal component analysis. The gene expression matrix was normalized and the batch effects among samples were corrected using Harmony (v 0.1.1)⁶⁴. Next, the top 40 principal components were used to construct a neighborhood graph, followed by Leiden clustering (resolution = 1.0). Finally, dimensionality reduction visualization was performed using uniform manifold approximation projection (UMAP). Subsequently, we compared cluster-specific genes with known classical marker genes to determine cell types. We performed sub-cluster analysis using the same steps as above. To eliminate interference, we removed cells introduced by sequencing bias before performing sub-cluster analysis.

Stereo-seq tissue processing

OCT-embedded skin tissues were sectioned at a thickness of 400 µm and total RNA was extracted using the RNeasy Mini Kit (Qiagen, 74104). RNA integrity was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies), and only samples with RNA Integrity Number (RIN) ≥ 6.5 were used for subsequent experiments.

Frozen skin tissues in OCT were sectioned at a thickness of 10 µm using the Stereo-seq Permeabilization Set for Chip-on-a-slide V1.0 (MGI, 101SP118) and permeabilization was tested for 6, 12, 18, and 24 min, with free RNA as a positive control. The optimal permeabilization time was determined as 12 min for HC1, DLE1, DLE2, DLE4, SLE4, and SLE5 samples; and 18 min for HC2, HC3, HC4, DLE3, DLE5, SLE1, SLE2, SLE3, and SLE6 samples to balance tissue fluorescence intensity and minimal RNA diffusion.

Stereo-seq in situ reverse transcription and amplification

Stereo-seq assay was performed according to the manufacturer’s protocol using Stereo-seq Transcriptomics Set for Chip-on-a-slide V1.2 (MGI, 101ST114). Briefly, 10 μm frozen tissue sections were transferred to the surface of the Stereo-seq chip, incubated at 37 °C for 5 min to allow adherence, and then fixed in −20 °C methanol for 30 min. After complete methanol evaporation, the chip was stained with nucleic acid dye for single-stranded DNA (ssDNA) visualization, while the adjacent section adhered to the slide was stained with Hematoxylin and Eosin (H&E) staining. Both staining results were imaged using a Motic Custom PA53 FS6 microscope.

Uniformly add 150 μL 1× permeabilization buffer on chip and incubate at 37 °C for 12 or 18 min. Wash the tissue with wash buffer and add 200 μL/chip Reverse Transcription Mix for reverse transcription, and incubated at 45 °C for 2 h. After reverse transcription, wash the chip with 0.1× SSC and release cDNA with 180 μL/chip cDNA Release Mix, and incubated at 55 °C for 10 min. The released cDNA was transferred to a 1.5 mL centrifuge tube, neutralized with 23 μL of Neutralizer, made up to 198 μL with nuclease-free water, and denatured at 95 °C for 5 min.

cDNA was amplified using the reagents provided in the kit, and the PCR conditions were as follows: 95 °C for 5 min; 98 °C for 20 s, 58 °C for 20 s, and 72 °C for 3 min, 13 cycles; and a final extension at 72 °C for 5 min. PCR products were purified using VAHTSTM DNA Clean Beads (Vazyme, N411-03) and quantified using the Qubit dsDNA HS Assay Kit (Invitrogen, Q32854).

Stereo-seq library construction and sequencing

The sequencing library was constructed according to the manufacturer’s instructions of the Stereo-seq 16 Barcode Library Preparation Kit (MGI, 101KB016). Briefly, 100 ng of cDNA was mixed with 10 μL of KMB and nuclease-free water to a final volume of 45 μL, followed by incubation at 95 °C for 5 min and 40 °C for 3 min. Then, 5 μL of KME was added, and the mixture was incubated at 37 °C for 10 min for cDNA multiple displacement amplification. The ssDNA was purified using VAHTSTM DNA Clean Beads (Vazyme). Subsequently, 100 ng of the purified product was amplified using the PCR mix provided in the kit under the following conditions: 95 °C for 5 min; 13 cycles of 98 °C for 20 s, 58 °C for 20 s, and 72 °C for 30 s; and a final extension at 72 °C for 5 min. The PCR product was purified with VAHTSTM DNA Clean Beads (Vazyme, N411-03) for DNA nanoball (DNB) generation. Finally, sequencing was performed on the MGI DNBSEQ-Tx sequencer (MGI) with read lengths of 50 bp for read 1 and 100 bp for read 2.

Stereo-seq raw data processing

Fastq files were processed using SAW software⁶⁵. Read 1 is consisted of 25 bp CID sequences (coordinate identity) and 10 bp MID sequences (molecular identifiers). CID sequences were matched with the designed coordinates of the in situ captured chip to determine the spatial location of each read on the chip, only allowing 1 base mismatch. Reads with MID sequences containing either N bases or more than 2 bases with quality scores below 10 were filtered out. The remaining reads were aligned to the human reference genome (GRCh38) using STAR (v2.7.2b)⁶⁶. Finally, we generated an expression profile matrix for each transcript at each spatial spot on the chip for subsequent analysis.

Cell segmentation

The Stereo-seq chip contains pre-etched track lines for image registration. Using StereoCell (v0.2.0) software, we aligned the track lines from ssDNA images with those from mRNA expression images. Nuclei segmentation masks were then identified from ssDNA images using QUPATH (v0.4.3) for image analysis⁶⁷. To capture cytoplasmic mRNA, we expanded the nuclear masks by 8 pixels outward to generate spatial single-cell masks. Finally, an in-house pipeline was applied to generate gene expression matrix for each spatial single cell (Cellbin) for downstream analysis.

Quality control of segmented cells

Low quality Cellbins with an area greater than 2500 or less than 100, fewer than 200 detected genes, and mitochondrial percentage ≥ 10% were filtered out. All query genes were guaranteed to be expressed in at least three cells prior to be retained. Our filtering strategy, primarily based on gene counts, removed 34.99% of the cells. Of these, 96.40% of the cells had ≤ 200 genes detected, 16.46% of the cells had ≥ 10% mitochondrial percentage, and 7.45% of the cells had an area ≤ 100 or ≥ 2500.

Cell type annotation of segmented cells

To validate spatial distribution of the identified cell types in our skin sections, we employed Cell2location (v0.1.3) to map cell types from scRNA-seq onto spatial locations from Stereo-seq⁶⁸. First, scRNA-seq data were filtered using the software’s default parameters. Subsequently, the negative binomial regression model with default settings was applied to estimate gene expression signatures for each scRNA-seq cell type. Finally, these scRNA-seq expression signatures were mapped onto Stereo-seq data to obtain the abundance of each cell type at every spatial location. During this process, the hyperparameter N_cells_per_location, which influences cell abundance estimation in Stereo-seq data, was set to 1 (based on cell area considerations). The cell type with the highest q05_cell_abundance_w_sf cell abundance at each spatial spot was assigned as the definitive cell type for that Cellbin. To evaluate the reliability of the Stereo-seq annotations, we identified the top 5000 highly variable genes from the scRNA-seq data using Scanpy (flavor = “cell_ranger”). Subsequently, Spearman correlation coefficients were calculated between the average expression profiles of the scRNA-seq and Stereo-seq datasets at the major cell type level.

Define regions of epidermis and dermis

The location of the basement membrane (BM) was manually identified on ssDNA images based on H&E staining images, enabling classification of each pixel into either the epidermal or dermal region. By matching the coordinates of ssDNA pixels with the x-axis and y-axis positions of Stereo-seq Cellbins, each Cellbin was subsequently classified as belonging to either the epidermal or dermal layer. The Euclidean distance from each Cellbin to the BM was calculated for further analyses.

Spatial distribution of keratinocyte subtypes and inflammatory mediators in epidermis

To analyze the spatial gradient distribution of keratinocyte subtype proportions and inflammatory mediator expression levels in the epidermis, we calculated the relative position of each Cellbin within the epidermis to the top of epidermis and the BM. The relative positions were divided into 10 equal intervals, ranging from 0 to 1. The value of 0 indicates the top of epidermis, and 1 indicates the BM. The proportions of keratinocyte cell types and the expression levels of inflammatory mediators at each epidermis depth were subsequently calculated.

Spatial distribution of immune cells in dermis

To assess the spatial gradient distribution of immune cells in the dermis, we calculated the Euclidean distance of dermal immune cells to the BM in disease samples. The distances were stratified into 6 intervals (0-1500 μm), with distances exceeding 1500 μm grouped into a single interval. The proportions of different immune cell types within each distance range were subsequently quantified.

Signature gene set score

Signature gene set scores were calculated using Scanpy (v1.10.1) with the “scanpy.tl.score_genes” function (parameter setting: ctrl_size = 100) to evaluate expression levels in individual cells or Cellbins⁶⁹. The gene sets were derived from the HALLMARK and Gene Ontology (GO) biological processes  in the Molecular Signatures Database (MSigDB, v2024.1.Hs)⁷⁰. All gene sets were listed in Supplementary Data 4.

Single-sample gene set enrichment analysis

Single-sample gene set enrichment analysis (ssGSEA) was performed in Stereo-seq data using the R package scGSVA (v0.0.22) with parameter settings (method = “ssgsea”) to compare pathway enrichment differences among various groups and tissues. Gene sets were obtained from the HALLMARK database.

Pseudotime trajectory analysis

To infer potential differentiation trajectory of keratinocyte subtypes, we performed pseudotemporal analysis on the scRNA-seq data using Monocle2 (v2.24.1)¹⁶. The significantly differentially expressed genes identified by the Seurat “FindAllMarkers” function (adjusted P value < 0.05 and |log₂(fold change)| > 0.5) were used as the ordering gene set¹². A CellDataSet object was built based on the raw UMI counts and preprocessed by estimating size factors and gene dispersion. Subsequently, the DDRTree algorithm was applied for dimensionality reduction and trajectory inference, and cells were ordered along the pseudotime trajectory using the orderCells function. The inferred cell trajectory was visualized using the Monocle2 built-in function plot_cell_trajectory.

Cell-cell interaction analysis

CellChat (v2.1.1) was performed to analyze cell-cell interactions between interested cell types across disease conditions in scRNA-seq data¹⁷. The algorithm integrates a manually curated ligand-receptor interaction database (CellChatDB) with single-cell expression profiles, quantifying intercellular communication probabilities of ligand-receptor pairs through probabilistic modeling. Specifically, it calculates interaction strength by evaluating co-expression levels of ligands in sender cells and their cognate receptors in receiver cells, weighted by pathway-specific confidence scores. To ensure analytical robustness, CellChat employs permutation tests (100 permutations) for statistical validation of communication signals, retaining only interactions with corrected P value < 0.05.

Spatial cell-cell communication analysis

Due to the insufficient number of VEC_CCL14⁺ cells in the disease state in the scRNA-seq data for CellChat analysis, we employed COMMOT (v0.0.3) to analyze cellular cross-talk between endothelial cells and immune cells³³. The software incorporates spatial constraints to prevent implausible interactions between distantly located cells, and we considered interactions reliable only between cells within a spatial distance of 100 μm. Similar to CellChat, COMMOT utilizes permutation tests (100 permutations) for statistical validation of communication signals, retaining only interactions with corrected P < 0.05.

Identification of immune cell niches

For the identification of immune cell niches, we performed the following steps: First, we calculated the number of other immune cells within a 150 μm radius for each immune cell in each sample, generating a matrix of neighborhood cell counts. Second, we performed Leiden clustering using Scanpy with default parameters for scanpy.pp.neighbors and resolution = 0.08 for scanpy.tl.leiden⁶⁹. This process initially generated 17 clusters, which were then manually merged based on similar immune cell proportions, resulting in 8 distinct clusters with significant compositional differences. Third, we applied DBSCAN clustering from sklearn (v1.3.0) to remove distant and low-density cells with eps = 150 and min_samples = 20. Fourth, we determined the contours of each cell cluster using the alphashape from Python with alpha = 0.01. Fifth, we identified potential tertiary lymphoid structures (TLS) among the clusters based on the ratio of T cells and B cells and the TLS gene set score (Supplementary Data 4)⁷¹.

Spatial characteristic analysis of immune subsets within TLS

To assess the spatial characteristics of immune cells within TLSs, we performed cell density correlation and intercellular distance analyses on immune subsets for each TLS. First, the density of each immune cell subtype was calculated. Spearman correlation analysis was performed pairwise for all cell subtypes, and the Benjamini-Hochberg adjusted P value < 0.05 were considered as significant correlation in cell density. Second, the Euclidean distance between each immune cell and other different subtypes within each TLS was calculated. The median distance between two subtypes within TLSs was calculated and visualized using a heatmap.

Spatial distribution of stromal cell subsets on both sides of TLS boundary

To analyze the changes in density of endothelial cells and fibroblasts at the TLS boundary with distance, we calculated the shortest Euclidean distance from each subtype of endothelial cells and fibroblasts to the TLS boundary and statistically analyzed the cell density at different distances, comparing it with the density at the closest distance (20 μm or −20 μm from the TLS boundary).

Differential gene expression and functional enrichment analysis

Differential gene expression analysis between different groups was performed using the “FindMarkers” function in Seurat¹². For scRNA-seq data, genes with min.pct > 0.1, |log₂(fold change)| > 0.5, and Bonferroni correction adjusted P value < 0.05 were considered significantly differentially expressed and used for further analyses. For Stereo-seq data, we compared gene expression profiles of immune cells within TLS regions with those located in adjacent areas (200 μm outside the TLS boundary). Genes with min.pct > 0.05, |log₂(fold change)| > 0.25, and Bonferroni correction adjusted P value < 0.05 were considered significantly differentially expressed and used for further analyses.

Gene Ontology enrichment analysis was conducted using Metascape with the human dataset as ref. ⁷². The top significant terms (adjusted P value < 0.05) were selected for visualization.

Transcription factor analysis

pySCENIC (v0.12.1) was performed to evaluate transcription factor (TF) activity across B cell subtypes in scRNA-seq data²⁸. The top 10 TFs with highest regulon specificity scores in B_ABC cells was visualized using ComplexHeatmap (v2.22.0). GO enrichment analysis of target genes regulated by BATF and TFEB was performed using Metascape, and the target gene-function relationships were visualized with Cytoscape (v3.10.1)⁷³. Based on the TF regulatory networks identified in scRNA-seq data, we predicted TF activity for each Cellbin in Stereo-seq data using AUCell.

Publicly available lichen planus spatial transcriptomics data analysis

Publicly available spatial transcriptomics data of lichen planus using the Visium technology of 10X Genomics (GSE206391) was downloaded to validate our findings on interface dermatitis²¹. The dataset included profiles from both lesional and non-lesional skin of 4 individuals (Supplementary Data 3). Following the same analytical pipeline applied to our primary data, we performed epidermal-dermal segmentation based on H&E images and assigned each spatial spot to its respective compartment. The relative distances of each spot to the top of epidermis, the bottom of dermis, and the BM were calculated. Cellular composition for each spot was estimated using Cell2location (v0.1.3)⁶⁸. Bivariate Moran’s I analysis was performed using the Moran_Local_BV function from the Python esda package to reveal the spatial co-localization of specific gene expression and target cell populations in skin lesions.

Publicly available CLE spatial transcriptomics data analysis

Publicly available NanoString Digital Spatial Profiling (DSP) Whole Transcriptome Atlas (WTA) dataset (GSE182825) was downloaded and performed further analysis²⁶. This dataset comprises spatial transcriptomic profiles across multiple tissue slides encompassing HC, DLE, and SLE conditions. Due to the lack of explicit patient-to-ROI mapping in the public metadata, our analysis was conducted at the region of interest (ROI) level, categorized by clinical diagnosis and cellular morphology. A total of 26 ROIs were analyzed, including CD45⁺CD3⁺ T-cell areas, CD45⁺CD3^- non-T immune cell areas, and keratinocyte segments (Supplementary Data 3). The raw count data, which had undergone prior quality control, filtering, and Quantile 3 normalization, were used for all downstream analyses. We employed Gene Set Variation Analysis (GSVA, v1.52.3) in R to calculate TLS scores within the CD45⁺CD3⁺ and CD45⁺CD3^- ROIs⁷⁴. Furthermore, we evaluated the relative abundance of keratinocyte and immune cell subtypes using SpatialDecon (v1.6.0)⁷⁵. Given that the extensive expression of interferons led to a disproportionate abundance of T_IFN cells and B_IFN cells, which obscured other subtypes, we excluded these two groups prior to analysis and visualized the results using the ComplexHeatmap R package⁷⁶.

Multiplex immunohistochemistry

The Opal 7-color immunohistochemistry (IHC) detection kit (PerkinElmer, NEL811001KT) was conducted to multiplex IHC staining of human FFPE skin tissues according to the manufacturer’s instructions by applying primary antibodies against targets including CD20 (MXB, MAB-0669), CD3 (MXB, MAB-0740), CD4 (MXB, RMA-1086), FOXP3 (Abcam, ab215206, dilution 1:500), CXCL10 (Abcam, ab318282, dilution 1:1000), CXCL13 (Abcam, ab246518, dilution 1:1000), T-bet (Abcam, ab154200, dilution 1:1000), CD31 (MXB, MAB-0720), and KRT6B (Proteintech, 17391-1-AP). They were followed by incubation with Opal anti-rabbit/mouse horseradish peroxidase (HRP) secondary antibody (PerkinElmer, ARH1001EA) and tyramide signal amplification. The slides were then high-pressure antigen retrieval after each TSA procedure. Nuclei were stained with DAPI (Abcam, ab104139) after all the human antigens had been labeled. All images were captured by the PerkinElmer Vectra multispectral imaging system (PerkinElmer), and analyzed by inForm software (v2.4.11).

Primary keratinocytes isolation and treatment

Primary human keratinocytes were purchased from Lifeline Cell Technology (FC-0007). For isolation of primary mouse keratinocytes, skin was collected from C57BL/6 J newborn mice and digested overnight with 4 mg/mL dispase (Sigma-Aldrich, D4693) at 4 °C. After enzymatic digestion, manually separated epidermis and dermis and incubated the skin in trypsin solution at room temperature on a horizontal shaker with gentle agitation for 20 min. Primary keratinocytes were cultured in basal medium (Lifeline Cell Technology, LM-0004) supplemented with life factors (Lifeline Cell Technology, LS-1030). When primary keratinocytes were grown to 80% confluence, cells were treated with 10 ng/mL recombinant human IFN-γ (PeproTech, 300-02) or recombinant murine IFN-γ (PeproTech, 315-05) for 3 h and/or irradiated with 40 mJ/cm² UVB (310 nm) and collected for RNA isolation and qPCR.

Transwell migration assay

Transwell migration of lymphocytes was performed with activated T cells or B cells and concentrated keratinocyte-conditioned medium. Briefly, CD4⁺ T cells and CD19⁺ B cells from mouse spleens were isolated using CD4⁺ T Cell Isolation Kit (Stemcell, 19852) and Pan-B Cell Isolation Kit (Stemcell, 19844). Isolated T cells and B cells were stimulated with 5 μg/mL plate-bound anti-CD3 antibody (eBioscience, 16-0031-85) and 2 μg/mL anti-CD28 antibody (eBioscience, 16-0281-85) for 3 days and 10 μg/mL anti-IgM polyclonal antibody (Jackson Laboratories, 109-006-129) and 10 ng/mL IL-4 (PeproTech, 214-14) for 2 days, respectively. To obtain keratinocyte-conditioned medium, primary mouse keratinocytes were cultured in basal medium (Lifeline Cell Technology, LM-0004) supplemented with life factors (Lifeline Cell Technology, LS-1030) with or without the addition of 10 ng/mL recombinant murine IFN-γ (PeproTech, 315-05) for 3 h and/or irradiated with 40 mJ/cm² UVB (310 nm). The medium was then concentrated (1×, 5×, or 10×) for chemo-attractants. In the transwell cell migration assay, 24-well 5-μm inserts from Corning were used. Migration medium (500 μL) containing either keratinocyte-conditioned medium with the indicated concentrations or control medium were loaded into the lower chamber. 5 × 10⁵ activated T cells or B cells were suspended in 200 μL medium and placed in the upper chamber. At 3 h after incubation at 37 °C and 5% CO₂, the migrated and unmigrated immune cells were collected from the lower and upper chambers, respectively. To normalize samples, beads of known concentration were added to all samples before counting by FACS. The migration ratio was determined by dividing the number of migrated cells by the number of migrated plus unmigrated cells.

RNA isolation and RT-qPCR

Total RNA from keratinocytes was extracted using the Super FastPure Cell RNA Isolation Kit (Vazyme, RC102). For cDNA synthesis, equal amounts of RNA were reverse-transcribed using HiScript IV RT SuperMix (Vazyme, R423). qPCR was conducted using LightCycler 96 (Roche) with ChamQ Universal SYBR qPCR Master Mix (Vazyme, Q711-02). The relative expression levels of genes were calculated by the 2^−ΔCt method, which normalized to the reference gene ACTB or  Rplp0. The primers used in this study are listed in Supplementary Data 10.

Construction of SLE murine model and CXCL10 intradermal injection

The C57BL/6 J, Pparg^fl/^fl (NM-CKO-190071), and Krt5^creERT2/+ (NM-KI-190016) mice were purchased from the Shanghai Biomodel Organism Science & Technology Development Co. Ltd. Pparg^fl/fl mice were bred with Krt5^creERT2/+ mice to generate Pparg^fl/fl; Krt5^creERT2/+ keratinocyte conditional Pparg knockout mice. Eight-week-old mice were used for all experiments. Mice were housed under specific pathogen-free conditions (20-25 °C, 40-60% humidity) with a 12 h light/dark cycle. All experiments were repeated three times with 5-6 mice per group. All mouse experiments were performed on both female and male mice, and there is no sex difference in the above mouse experiments. Mice were randomly assigned to experimental and control groups. Animal welfare was monitored, euthanasia was conducted according to the guidelines, and the ethical committee of the Institute of Dermatology, Chinese Academy of Medical Sciences approved all animal procedures.

To induce spontaneous SLE-like phenotype, 4-hydroxytamoxifen (5 mg/mL; Sigma-Aldrich, H6278) suspended in 10% DMSO (Sigma-Aldrich, D2650) and 90% corn oil (Beyotime, ST1177) was topically applied to both ears of each Pparg^fl/fl; Krt5^creERT2/+ mice for 5 consecutive days to activate Cre recombinant protein⁷⁷. After the spontaneous SLE-like phenotype appeared, 500 ng CXCL10 protein was intradermally injected into the back of mice. Skin lesions were harvested 6 h after injection for flow cytometric analysis.

Flow cytometric analysis

Full-thickness skin lesions (1 cm × 1 cm) were excised from the back of mice and washed three times with cold Hank’s Balanced Salt Solution (Gibco, 14170112). The tissue was gently cut into small pieces using curved scissors and placed in 5 mL of dissociation buffer containing Collagenase I with a final concentration of 2 mg/mL with DMEM high glucose (Gibco, 11965092), and incubated at 37 °C and 5% CO₂ for 1 h. Then, the cell suspension was filtered through a 40 μm cell strainer and washed with stain buffer (BD Biosciences, 554656) to obtain a single-cell suspension.

For cell surface staining, the following antibodies were used: Zombie Aqua™ Fixable Viability Kit (Biolegend, 423102), anti-mouse CD45-AF700 antibody (eBioscience, 56-0451-82, dilution 1:500), anti-mouse CD4-FITC antibody (Biolegend, 116004, dilution 1:500), anti-mouse CD25-BB515 antibody (BD Biosciences, 564424, dilution 1:500), anti-mouse CXCR5-APC antibody (Biolegend, 145506, dilution 1:500), anti-mouse PD-1-BB700 antibody (BD Biosciences, 748242, dilution 1:500), anti-mouse CD69-BV605 antibody (Biolegend, 104530, dilution 1:500), anti-mouse CD103-BV421 antibody (Biolegend, 156915, dilution 1:500), anti-mouse CD19-BV711 antibody (Biolegend, 115555, dilution 1:500), anti-mouse CD138-PE-Cy7 antibody (Biolegend, 142514, dilution 1:500), anti-mouse NK1.1-BV650 antibody (Biolegend, 156547, dilution 1:500), and anti-mouse CD11c-APC-Cy7 antibody (Biolegend, 117324, dilution 1:500). For cell nuclear staining, nuclear membranes were permeabilized and stained according to the manufacturer’s protocol of Foxp3 Staining Buffer Set (eBioscience, 00-5523-00) with anti-mouse FOXP3-PE antibody (Biolegend, 126404, dilution 1:100). For cell intracellular staining, cells were stimulated with leukocyte activation cocktail (BD Biosciences, 550583) at 37 °C and 5% CO₂ for 6 h and then stained with anti-mouse IFN-γ-APC antibody (Biolegend, 505810, dilution 1:100), anti-mouse IL-4-BV421 antibody (Biolegend, 504120, dilution 1:200), and anti-mouse IL-17A-PE-Cy7 antibody (Biolegend, 506922, dilution 1:200). Flow cytometry data were acquired on NL-3000 flow cytometer (CYTEK Biosciences) and analyzed via FlowJo software (v10.8.1).

Statistical analysis

Statistical analyses for Stereo-seq data were conducted using R 4.2.1. For comparisons between two independent groups, the two-sided Wilcoxon rank-sum test was performed. For comparisons involving three or more groups, the two-sided Kruskal–Wallis test followed by Dunn’s multiple comparison test was employed. Correlations analyses were performed using Spearman’s rank correlation. Statistical analyses of experimental data were conducted using GraphPad Prism (v10.3.1). For two-group comparisons, two-tailed unpaired Student’s t test was used. For three or more groups, the one-way analysis of variance (ANOVA) followed by Holm–Sidak’s multiple comparisons test was performed. Data are presented as the mean ± SEM, with sample numbers indicated. No statistical methods were employed to predetermine the sample size. Mice in this study were randomly allocated to different groups.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Q.L., M.Z., H.W. and J.M. conceived and designed this study. W.Z. and Y.H. analyzed and interpreted the data and wrote the manuscript. W.Z. and Y.L. performed all experiments except for mIHC staining. J.T. provided the SLE murine model adopted in the study. H.L., W.S. and H.Z. recruited patients. H.Y. and Q.L. contributed to sample preparation for Stereo-seq tissue. M.Z. and Z.H. assisted in subclusters analysis of scRNA-seq data. W.Z. and B.Z. conducted the mIHC staining of skin tissues. Z.Z. contributed to the analysis of immune cell niches. J.H. was involved in generating the spatial distribution map of keratinocytes. Y.Y. was responsible for implementing the code for the identification of epidermal and dermal regions. R.C. and X.F. provided the help for Stereo-seq experiment. All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The Stereo-seq data generated in this study have been deposited in the National Genomics Data Center (NGDC) under accession code OMIX012568 and the China National GeneBank DataBase (CNGBdb) under accession code STT0000149. Due to the administrative policies regarding human genetic resources in China, these data are available under controlled access. Access can be obtained by application through the respective repository’s managed access system and will be released upon approval by the corresponding authors. The scRNA-seq data of cutaneous LE lesions used in this study were available in the Gene Expression Omnibus (GEO) database with accession code GSE179633. The Visium spatial transcriptomics data of lichen planus were obtained from GEO database under accession code GSE206391. The NanoString Digital Spatial Profiling (DSP) data of CLE were available in GEO under accession code GSE182825. All relevant approvals were obtained from the Ministry of Science and Technology (MOST) of China regarding the management and external provision of human genetic resources relevant to this work. Any other details supporting the findings of the present study are available from the corresponding authors upon request. Source data are provided with this paper.

Code availability

All the codes used for processing and analyzing the data in this study have been deposited in GitHub (https://github.com/hyf−2021/Skin_LE_script).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72720-1.