Cross-species comparative spatial transcriptomics of hair follicle-T cell interactions identifies conserved drivers of cutaneous lupus erythematosus skin disease and associated hair loss

Abstract

Cutaneous lupus erythematosus (CLE) encompasses several autoimmune entities characterized by a T cell-rich infiltrate and alopecia. Here, we perform spatial transcriptomics across three mammalian species to characterize potential T cell-hair follicle communication pathways in the microanatomical niche. In mice, adoptive transfer of either autoreactive CD4 + /OT2 or CD8 + /OT1 T results in skin disease with 35% gene expression overlap. CD8 T cells specifically contribute to extracellular matrix reorganization. Digital spatial profiling of skin identifies CXCR3 ligands and IFN response genes in hair follicles, also conserved in pet dog and human Discoid lupus erythematosus (DLE) biopsies. Knock-out of CXCR3 on OT2 T cells alleviates skin disease in CLE mice. Moreover, B cell depletion in OT1 CLE recipient mice reduces skin disease scores as relevant to human B cell depletion therapies for lupus. Last, we identify CFD and S100A8/9 as conserved targets that could be further explored through future veterinary and human trials.

Cutaneous lupus erythematosus comprises several skin disease subsets, including primary cicatricial alopecia (PCA), characterized by skin inflammation, hair loss and collagen deposition in the follicles. How immune cells contribute to the pathogenesis is controversial. Here, the authors present a comparative spatial transcriptomic analysis of hair follicle-T cell interaction across a CLE/PCA mouse model, pet dog and human biopsies and report an enrichment of CXCR3 + T cells and of B cells in CLE lesional skin as well as changes in CFD and S100 expression.

Introduction

Cutaneous lupus erythematosus (CLE) encompasses several rare and common clinical entities that are all characterized by skin inflammation and wound-healing type-responses in the absence of obvious physical injury¹. These disorders can cause psychological stress and disfigurement². The initiating factors for distinct clinical entities are likely different and include age/sex hormones³ in addition to genetics and environmental triggers⁴ that lead to collapse of immune privilege in the hair follicle⁵. Since most patients present to the clinic after disease onset, stopping inflammation and reversing the fibrotic process are of utmost importance for improving prognosis and chances of disease reversal. Furthermore, studying shared pathological processes in different disorders and models will yield better insights into fibrotic processes and may accelerate treatments for diseases with similar features.

Clinically, the discoid subtype of CLE (DLE) is considered a subtype of lymphocytic primary cicatricial alopecia (PCA), or scarring alopecia. Other clinical subtypes of lymphocytic PCA include chronic cutaneous lupus erythematosus (CCLE)⁶, lichen planopilaris (LPP) and frontal fibrosing alopecia (FFA)⁷, among others⁷. The pathophysiology of these conditions shares similar features: inflammation, hair loss, and collagen deposition in the follicle. Different clinical subtypes of PCA have unique gene signatures: for example, CCLE has an interface dermatitis histological pattern in addition to hair follicle involvement, LPP appears to be more inflammatory (with higher numbers of CD8+ T cells⁵ and macrophages⁸), and FFA appears more fibrotic (with higher markers of epithelial to mesenchymal transition⁸). It is not clear whether CD8+ or CD4+ T cells contribute more to skin inflammation or alopecia in these conditions, and altered ratios of these T cell subsets have been reported in DLE and other forms of CLE^9–15. A paucity of animal models has hindered preclinical studies of potential treatments of CLE and DLE in particular due to challenges in treating fibrosis.

Companion animals spontaneously develop many of the same diseases as their human companions and may also serve as large animal models of disease. For example, pet dogs can develop CLE, DLE and other forms of alopecia⁹. Dogs share the same environment and lifestyle as their human companions, and dogs owned by SLE patients have a higher risk of developing lupus themselves¹⁶. Previous work from us and others has characterized the transcriptome of DLE and demonstrated shared gene expression in bulk microarray samples¹⁷.

Here, we sought to examine cellular neighborhoods and better define the microanatomical niche of T cells in and around hair follicles in mice, pet dogs and humans using spatial transcriptomics, with confirming mechanistic studies in mice. We demonstrate that CXCR3 ligands are expressed in the hair follicle niche, and that CXCR3 deficient CD4+/OT2 T cells are less able to induce skin disease in mice. We also demonstrate that B cells are enriched in human and canine DLE, and that B cell depletion prevents skin disease in CD8+/OT1 recipient mice. Importantly, we also identify overlap of S100 expression between pet dog and human, and complement expression in all 3 species, thereby providing new potential drug targets that are currently in development for other indications and could be tested in future veterinary and human clinical trials. More broadly, these findings may be applicable to other fibrotic skin conditions.

Results

Development of a mouse model of cutaneous lupus erythematosus that is dependent on the loss of TLR9 signaling

We developed a B6 model of cutaneous lupus erythematosus based on a similar model on the Balb/c background presented in Mande et al.¹⁸. We bred K5TGO rtTA mice to TLR9KO mice to generate CLE recipients (Fig. 1A). The TLR9KO background is lupus-prone, as it lacks immune regulatory signaling capacity¹⁹. When injected with activated OT2 T cells and administered doxycycline chow, these mice develop skin disease that resembles human CLE clinically with rashes and alopecia.

Fig. 1. Development of an inducible mouse model of cutaneous lupus erythematosus (CLE) that is dependent on loss of TLR9 signaling and autoreactive CD4+ T cell responses.

A BioRender schematic of the CLE mouse model induction protocol. Briefly, OT2 donor T cells are activated in vitro and transferred into sublethally irradiated recipients. Recipient mice were generated by breeding TLR9KO mice to K5-TGO-rtTA mice. Doxycycline chow is provided to the recipient mice to induce expression of ovalbumin in keratinocytes. (Created in BioRender. Richmond, J. (2026) https://BioRender.com/m31y149). B Example photographs demonstrating presence or absence of skin disease in recipient mice with the indicated genotypes. C Skin scores at week 3–4, maximal disease, from the genotypes as indicated (n = 34 TLR9KO rtTA+ (19 female, 15 male), 8 TLR9het rtTA+ (2 female, 6 male), 13 TLR9KO rtTA- (8 female, 5 male), 5 TLR9het rtTA- (4 female, 1 male) and 20 uninjected control mice (3 female, 17 male) pooled from 3 to 5 experiments; one-way ANOVA with Tukey’s posttests significant as indicated). D Spleen weights in mice of the indicated genotypes (one-way ANOVA with Tukey’s posttests significant as indicated). E Example ANA staining (scale bar 250 µm). F ANA grading performed by blinded observers (n = 4 TLR9KO rtTA+, 3 TLR9het rtTA+, 4 rtTA- littermates and one uninjected control representative sera from mice as in panel (C); PL2-3 is used as a positive control and buffer as a negative control for assays; one-way ANOVA with Tukey’s posttests significant as indicated). G Microarray gene expression analysis in B6 CLE mice versus littermate controls (n = 4 CLE (2 male, 2 female) and 4 littermates (3 male, 1 female). H Microarray gene expression analysis in Balb/c CLE mice versus littermate controls (n = 4 per group; queried from Mande et al. JCI 2018). I Comparison of DEGs from the B6 and Balb/c CLE mouse models (Created with BioVenn). Error bars represent the standard error of the mean (SEM).

TLR9 deficiency is generally considered to be a lupus-prone genetic background. To ascertain that our mice require this, we also evaluated TLR9 heterozygous mice as well as transgenic negative littermates (Fig. 1B). We observed an intermediate skin phenotype in TLR9het mice (Fig. 1C), similar to Nickerson et al.²⁰. Only TLR9KO mice had splenomegaly significantly higher than uninjected control mice (Fig. 1D). Anti-nuclear antibody (ANA) staining revealed that TLR9KO and TLR9het mice, but not uninjected controls, make ANAs (Fig. 1E, F). Littermate controls make lower titers of ANAs that typically have a nuclear pattern rather than signet ring pattern (Fig. 1E, F). To characterize the skin inflammation in these mice, we used a microarray to compare TLR9KO to littermate control mice (Fig. 1G). We also compared gene expression to the historical Balb/c model (Fig. 1H). Both models shared overlapping differentially expressed genes (DEGs), but there were also unique DEGs depending upon the genetic background of the mice (Fig. 1I). Taken together, these data indicate that the B6 genetic background can also serve as a model for CLE, facilitating evaluation of genetically targeted T cell donor and/or recipient mice given the wealth of genetically modified mice available on this genetic background.

Transfer of autoreactive CD8+ T cells also induces skin disease in CLE recipient mice

One benefit of the B6 genetic background is that we can also perform adoptive transfer of ovalbumin-specific CD8+ T cells (OT1) into recipient mice expressing ovalbumin in Keratin 5+ cells, which includes hair follicle keratinocytes (Fig. 2A). This model is similar to the Katz/Okiyama lab model of GVHD²¹, but differs in that our recipients express the K5-ovalbumin under a tetracycline response element so that we can induce flares²². The resulting disease in our mice resembles DLE, including alopecia, fibrosis and mucin deposition (Fig. 2B, C; Supplementary Fig. 1). Though this model is similar to GVHD models in skin disease score (Fig. 2C), mice do not significantly lose weight compared to littermate controls (Fig. 2D). The rtTA+ K5TGO+TLR9−/− OT1 recipients develop anti-nuclear antibodies (ANA, Fig. 2E) and splenomegaly (Fig. 2F).

Fig. 2. Autoreactive CD8+ T cells also induce skin disease in CLE recipient mice and attack both hair follicle and interfollicular keratinocytes.

A Example photographs from OT1 CLE recipients demonstrating hair loss and skin disease. B Example H&E, trichrome, mucin and Verhoff Van Gieson staining of skin sections from mice (scale bar 200 µm). C Skin scores over time. D Mouse weights over time. E Example ANA staining (scale bar 20 µm). F Spleen weights demonstrating splenomegaly in recipient mice (panels (A)–(F) n = 8 littermate (4 female, 4 male) and 11 OT1 CLE mice (6 female, 5 male) compiled from 3 to 4 separate experiments; Student’s unpaired two-tailed t test significant as indicated). G Microarray gene expression analysis in OT1 CLE recipient mice versus littermate controls (n = 4 CLE (3 male, 1 female) and 4 littermates (3 male, 1 female)). H Gene set enrichment analysis of microarray data from mice. I Schematic of photoconvertible mouse model for in vivo ear imaging (Created in BioRender. Yildiz altay, U. (2026) https://BioRender.com/s5ho7ru). J Stills from 2 photon in vivo ear imaging videos (scale bar 100 µm). K Quantification of time spent in frame by the T cells. L Quantification of hair follicle and interfollicular epithelial cell types in littermates versus CLE mice using flow cytometry (n = 14 littermates (3 female 11 male), and 14 OT1 CLE mice (9 female 5 male) per group pooled from 4 experiments; multiple T tests with Bonferroni correction q values significant as indicated). Error bars represent the standard error of the mean (SEM).

We sought to characterize gene expression in lesional skin of CLE mice versus littermate controls using a microarray (Fig. 2G). We noted significant increases in chemokines/cytokines, complement family members and CD markers (p < 0.05, log2fold change +/−1.5). Gene set enrichment analysis revealed multiple inflammation-related pathways are upregulated in OT1 CLE recipient mice versus littermate controls (Fig. 2H). We compared our mouse model DEG list to published human datasets (GSE186075²³ and GSE95474⁶). We find overlapping gene expression in our model as compared to human lymphocytic PCAs including DLE (Supplementary Fig. 2), which makes sense given the model is dependent upon adoptive transfer of T cells.

Cytotoxic T cells mediate destruction of multiple keratinocyte and hair follicle cell populations

To observe T cell behavior in the follicles, we bred OT1 mice to Kikume photoconvertible mice and used cells from these donors to induce CLE²⁴ (Fig. 2I). We performed in vivo imaging of the ear skin²⁵ in CLE mice and littermate controls 24 h after photoconversion with a 405 nm violet laser. We observed GFP+ cells moving quickly in littermate controls, likely in the vasculature (Fig. 2J, Supplementary movie 1). In contrast, we found RFP+ cells arrested near hair follicles, surrounded by clouds of GFP+ cells (Fig. 2K, Supplementary movie 2). Next, we asked what target cells the T cells were attacking in the OT1 CLE mouse model. We performed flow cytometry on lesional skin from mice using a panel for flow cytometry of hair follicles²⁶ (Supplementary Fig. 3). We noted significant decreases in CD34+ bulge cells, which are typically attacked in scarring alopecia, from hair follicles and interfollicular epithelial cells in OT1 CLE mice compared to littermate controls (Fig. 2L). However, T cells also attacked other hair follicle and interfollicular keratinocyte populations, indicating that the K5-driven autoantigen expression occurs throughout the skin, matching its known distribution. Taken together, these data characterize OT1 transfers in lupus-prone mice and demonstrate that the disease can model aspects of human CLE and DLE in particular.

Spatial transcriptomics of OT1 CLE mouse skin compared to OT2 CLE mouse skin reveals different pathways allowing for interactions between CD8 versus CD4 T cells and stromal cells

We wanted to further characterize potential binding interactions between T cells and hair follicle cells in our mouse model. We hypothesized that autoreactive CD8+ T cells use discreet recognition pathways for interacting with hair follicle stromal cells as compared to CD4+ T cells. Therefore, we performed transcriptomics on skin from CD8+ recipients (OT1 recipient mice), CD4+ recipients (OT2 recipient mice) and littermate controls (Fig. 3A). First, we analyzed total gene expression in skin tissue using a bulk microarray (Fig. 3B). We noted unique and overlapping genes depending upon whether the rtTA+ K5TGO+TLR9−/− recipients received OT2 (CD4+) or OT1 (CD8+) T cells (Fig. 3C). We used Digital Spatial Profiling with CD3, CD8, and CD45 morphology markers to identify T cells and hair follicles, which we defined as CD45- from morphology marker selection in regions that had hair follicle morphology (Fig. 3D). We used cell segmentation approaches to pull out CD3+CD8+, CD3+CD8− and other CD45+ cells from regions of interest (ROIs), and used geometric ROIs to pull out hair follicles and keratinocytes for comparators (Fig. 3E, Supplementary Fig. 4). Comparing gene expression in the hair follicle from OT1 recipients vs healthy mice (Fig. 3F) and OT2 recipients vs healthy mice (Fig. 3G) revealed many differentially expressed genes. Examples of genes that were upregulated in the hair follicle from both mouse models include ATF4, a transcription factor that orchestrates the integrated stress response in hair follicles²⁷ (Fig. 3H), HMGB1, a danger signal that induces hair regrowth following trauma²⁸ but serves as a key danger-associated molecular pattern in lupus²⁹ (Fig. 3I), and CXCL16, a chemokine that is important for generation of Trm in the context of melanoma^30,31 (Fig. 3J). CD200R1, a classical immune privilege signal in the hair follicle³², was reduced in both models (Fig. 3K).

Fig. 3. Spatial transcriptomics of CLE mouse model OT1/CD8 recipient mice compared to OT2/CD4 recipient mice reveals conserved and differential gene expression.

A Schematic demonstrating CD4 (OT2) versus CD8 (OT1) recipient mouse models (Created in BioRender. Richmond, J. (2026) https://BioRender.com/pxkmk9k). B Bulk microarray data from OT2 versus OT1 mouse skin (n = 4 OT2 (2 male, 2 female) and 4 OT1 (3 male, 1 female)). C Biovenn analysis of unique and shared DEGs from the skin of recipient mice (Created with BioVenn). D Example images for ROI selection (scale bar 500 µm). E Schematic of involved versus uninvolved/healthy hair follicle ROI comparison (Created in BioRender. Richmond, J. (2026) https://BioRender.com/mpx60td). F OT1 versus healthy and G. OT2 versus healthy hair follicle gene expression. H ATF4, I HMGB1, and J CXCL16 were upregulated in both OT1 and OT2 hair follicles compared to healthy (one-way ANOVAs with Holm–Šídák’s multiple comparisons tests significant as indicated). K CD200R1 expression was lost in hair follicles in both models (one-way ANOVAs with Dunnett’s multiple comparisons test significant as indicated). L Comparing OT1 versus OT2 recipient hair follicle gene expression. M Pathway analysis of OT1 versus OT2 recipient hair follicle gene expression. N DBI and O ACE2 were upregulated in OT1 hair follicles compared to OT2. P WIF1 expression was lower in OT1 hair follicles compared to OT2 (one-way ANOVA with Tukey’s posttests significant as indicated). (n = 4 healthy, 4 OT2 and 4 OT1 tissues sectioned; ROIs analyzed that passed QC include 3 healthy, 3 OT2 and 4 OT1 hair follicles). Error bars represent the standard error of the mean (SEM).

We compared hair follicles from OT2 vs OT1 recipient mice, and found many DEGs (Fig. 3L). Pathway analysis of these genes demonstrated terms including “degradation of the extracellular matrix (ECM)”, “non-integrin membrane-ECM interactions”, and “platelet activation, signaling and aggregation” specifically in OT1 recipient mouse hair follicles (Fig. 3M). Notably, DBI (Fig. 3N), which is involved in lipid metabolism, and ACE2 (Fig. 3O), which is the major receptor associated with Sars-CoV-2 cell entry and has been implicated in androgenic alopecia³³ and post-COVID-19 disease³⁴ or vaccination-associated alopecia³⁵ were significantly upregulated in OT1 recipient hair follicles. In contrast, WIF1 (Fig. 3P), the loss of which is associated with baldness³⁶, was downregulated compared to OT2 recipient mice. Taken together, these data demonstrate unique and overlapping hair follicle danger signals in the B6 mouse models, and show that CD4 versus CD8 autoreactivity may result in slightly different responses in hair follicles, with CD8 specifically contributing to ECM reorganization and potentially hair follicle fibrosis.

Examination of spatial gene expression in spontaneous discoid lupus erythematosus and mucocutaneous lupus in client-owned dogs reveals pathways conserved with mouse and human DLE

Dogs also spontaneously develop DLE⁹. Our lab previously characterized bulk gene expression in lesional skin from companion dogs that spontaneously developed DLE using a microarray¹⁷. Here, we compared healthy leg margin skin excisions from amputations to DLE and mucocutaneous lupus erythematosus (MLE) biopsies³⁷ (Fig. 4A). First, we performed bulk RNA sequencing on curls from FFPE blocks from DLE cases versus healthy margins (Fig. 4B). We noted a loss of CD34 in DLE compared to healthy margins, and upregulation of chemokines (CCL5), keratins (KRT3, 4, 13, 76, 222), interferon response genes (MX1, PSMB8), and S100 family genes (S100A8).

Fig. 4. Spatial transcriptomics of spontaneous discoid (DLE) and mucocutaneous lupus (MLE) in client-owned dogs demonstrates increased S100 expression and loss of PPARG as pertinent to human discoid lupus.

A Left: example clinical photos of canine facial DLE featuring scarring, dyspigmentation, erosions/ulcerations and loss of nasal planum architecture. Right: example clinical photos of periorbital (top) and genital (bottom) MCLE in pet dogs. B Bulk RNAseq from DLE and healthy margin FFPE curls demonstrates upregulation of chemokines, keratins and S100 genes. C Example ROI selection from canine sections (scale bar 600 µm). D Schematic for involved versus uninvolved hair follicle ROI comparisons (Created in BioRender. Richmond, J. (2026) https://BioRender.com/mpx60td). E DLE versus healthy and F MLE versus healthy hair follicle comparisons. G Example S100A8 staining compared to isotype staining (scale bar 400 µm). H Quantification of S100A8 staining in different microanatomical compartments determined by ImageJ (one-way ANOVA with Tukey’s posttests significant as indicated). I Comparison of DLE versus MLE hair follicle gene expression. J Pathway analysis revealed ‘innate immune system’, ‘CLEC7A signaling’ and ‘neutrophil degranulation’ were enriched in canine DLE hair follicles. K IRF1 and L MMP9 were upregulated specifically in DLE hair follicles. M DLE hair follicles revealed a significant loss of PPARG compared to MLE and healthy (one-way ANOVA with Tukey’s posttests significant as indicated). N Example human DLE biopsy (GSE182825, scale bar 600 µm). O Heatmap of S100 family gene expression divided by ROI. S100A8/9 were among the top upregulated genes in keratinocytes. Error bars represent the standard error of the mean (SEM).

We performed Digital Spatial Profiling with the Canine Cancer Atlas (CCA) probe set using CD3, CD8 and CD45 morphology markers. We employed a similar ROI selection as in the mouse WTA, selecting hair follicles as well as T cell and immune cell ROIs (Fig. 4C), and examined involved versus healthy hair follicles from our spatial data (Fig. 4D). Comparing gene expression in the hair follicle from DLE versus healthy skin (Fig. 4E) and MLE vs healthy skin (Fig. 4F) revealed many differentially expressed genes. Some of the highest upregulated genes were S100 family genes. We confirmed expression of S100A8 at the protein level in keratinocytes using immunohistochemistry (IHC) (Fig. 4G), and found that staining was enriched in the epidermis and hair follicles of DLE compared to the dermis and lymphocyte-rich regions and to healthy skin (Fig. 4H).

We compared hair follicles from DLE vs MLE dogs and found many DEGs (Fig. 4I). “Signaling by VEGF” pathway was enriched in MLE hair follicles, whereas “Innate Immune System”, “CLEC7A (Dectin-1) signaling” and “Neutrophil degranulation” were enriched in DLE hair follicles (Fig. 4J). IRF1, which is an interferon response gene previously reported to be upregulated in plucked human hair follicles from chronic DLE patients³⁸, and MMP9, a matrix metalloproteinase previously reported to be elevated in human CCCA³⁹, were significantly upregulated in DLE dog hair follicles (Fig. 4K, L). There was also a significant loss of PPARG, which when deleted from hair follicle stem cells causes scarring alopecia in mice⁴⁰ (Fig. 4M).

We queried keratinocyte ROIs from our previously deposited human healthy, DLE and SCLE WTA dataset GSE182825⁴¹ to determine if S100 family genes are conserved across canine and human (Fig. 4N). Comparing CD3+CD45+ and CD45− keratinocyte geometric ROIs demonstrated highest counts of S100A7, A8 and A9 in keratinocytes (Fig. 4O). We quantified S100A8 and A9 (Supplementary Fig. 5) given that these were also elevated in the canine dataset and found that indeed they are upregulated in keratinocyte ROIs compared to CD3+ and CD45+ ROIs. Taken together, our comparative spatial transcriptomics of canine DLE demonstrates conservation of S100 family proteins in keratinocytes, as well as IFN responses and loss of protective factors in hair follicles including PPARG, which is also downregulated in human LPP⁴².

We compared perifollicular versus interfollicular CD8+ T cells in our mouse model and spontaneous DLE in pet dogs. Several DEGs overlapped, including CXCL12 (Supplementary Fig. 6). Pathway analysis using the GeoMX pathways tool identified “hemostasis” and “platelet activation” terms as conserved. This is interesting, given that platelet rich plasma (PRP) is used as a treatment for PCA in people⁴³.

Human DLE biopsies exhibit markers of T cell memory, chemokine ligands and fibrotic responses in and around hair follicles

DLE disproportionately affects skin of color patients⁴⁴. Therefore, we sought to perform spatial transcriptomics on DLE biopsies from archival blocks from skin of color patients to better understand the microanatomical niche in this patient population (Fig. 5A). First, we performed bulk RNA analysis using a microarray (Fig. 5B). We noted significant upregulation of markers identified in our mouse model (CD69, SELL, IL2RB/CD122), chemokines (CCL3, CXCL9, CXCL11), cytokines (TNF), IFN response genes (IFIT1, MX1, MX2, IRF7, ISG15, STAT1), as well as markers previously associated with lupus (MYD88, FASLG).

Fig. 5. Spatial transcriptomics of DLE biopsies from skin of color patients reveals FOS and STING1 expression in affected hair follicles.

A Example clinical photo of DLE. B Bulk microarray data from matched biopsies reveals upregulation of CD69, SELL, chemokines and IFN response genes. (n = 6 healthy and 12 biopsies from 10 DLE patients). C Schematic of affected vs unaffected hair follicle comparison (Created in BioRender. Richmond, J. (2026) https://BioRender.com/mpx60td). D Example ROI selections from biopsies (scale bar 500 µm). E Gene expression in DLE affected versus unaffected hair follicles. F Pathway analysis reveals ‘interferon signaling’, ‘neutrophil degranulation’, ‘IFN alpha/beta signaling’, ‘regulation of IGF’, ‘intrinsic pathway of fibrin clot formation’ and ‘ISG15 antiviral mechanism’ as top pathways enriched in DLE hair follicles. G FOS and H STING1 were significantly upregulated in affected hair follicles (Kruskal–Wallis test with Dunn’s multiple comparisons tests significant as indicated). (ROIs from n = 9 acute DLE, 10 DLE, 13 SLE, 3 SCLE and 12 tumid LE). Error bars represent the standard error of the mean (SEM).

We used the Whole Transcriptome Atlas (WTA) to compare DLE blocks to other types of non-scarring CLE, including SCLE and SLE (Fig. 5C, Supplementary Table 3). We examined affected hair follicles from DLE and acute DLE patients compared to unaffected hair follicles from all CLE biopsies (Fig. 5D). Affected follicles specifically upregulated CXCL11, ROCK2 and CASP9 among other genes (Fig. 5E). Pathway analysis revealed “interferon signaling”, “neutrophil degranulation”, “interferon alpha/beta signaling”, “regulation of IGF transport and uptake”, “intrinsic pathway of fibrin clot formation” and “ISG15 antiviral mechanism” terms were enriched in DLE hair follicles (Fig. 5F). Interestingly, FOS, which is involved in wound healing⁴⁵, and STING1, which is involved in hair follicle stem cell replicative stress⁴⁶, were significantly upregulated in affected DLE hair follicles (Fig. 5G, H). Taken together, our data confirm previously reported lupus-associated gene expression and identify novel pathways that have not been studied in human DLE.

T cell-hair follicle gene ontology terms “metabolic process”, “localization” and “response to stimulus” are conserved across mammalian species

We compared T cells in hair follicles to interfollicular T cells to better understand how they interact with T cells during primary cicatricial alopecia. We pooled all T cell ROIs and analyzed perifollicular vs interfollicular ROIs (Supplementary Fig. 7A). In OT1 recipient mice, CXCL12, CLIP, ZFP146, MET, OPN5, SCL35D2 were enriched in perifollicular T cells, whereas SOCS3, SERPINB13, and IRX3 were enriched in interfollicular T cells (Supplementary Fig. 7B). We exported the DEG lists and, using p < 0.05 and log2FC cutoff of 1.0, we used PantherDB to analyze Gene Ontology (GO) pathways as Biological processes. The top signatures excluding non-annotated genes were “cellular process”, “biological regulation”, “metabolic process”, “localization” and “response to stimulus” (Supplementary Fig. 7C). In DLE dogs, CXCL12 was conserved in perifollicular T cells (Supplementary Fig. 7D). We also noted IFNG, which has previously been described as a “node” in DLE⁴⁷, was upregulated in perifollicular T cells whereas FOS was upregulated in interfollicular T cells. GO pathways were similar, with “cellular process”, “biological regulation”, “metabolic process”, and “response to stimulus” being enriched in perifollicular T cells (Supplementary Fig. 7E). In human DLE samples, we noted CD69 and SELL were enriched in perifollicular T cells, along with IFNL3, CCL3, CCL8 and BACH1 genes (Supplementary Fig. 7F). Interfollicular T cells were enriched for expression of MAP2K5, SFRP2, CCL17 and GATA5 (Supplementary Fig. 7F). GO pathways were again similar, with “cellular process”, “biological regulation”, “metabolic process”, “localization” and “response to stimulus” as being the most enriched upregulated terms (Supplementary Fig. 7G). Taken together, these data suggest that T cell metabolism and migration are key biological processes conserved across species in DLE.

We then wanted to compare hair follicles themselves across species (Supplementary Fig. 8A). We again exported the DEG lists and, using p < 0.05 and log2FC cutoff of 1.0, we used PantherDB to analyze GO pathways as Biological processes. The top signatures in OT1 recipient mice excluding non-annotated genes were “cellular process”, “biological regulation”, “metabolic process”, “localization” and “response to stimulus” (Supplementary Fig. 8B). We also examined IFN response genes. IRF1 was upregulated in both OT2 and OT1 mouse models (Supplementary Fig. 8C). Last, we examined spatial CFD expression, which was a top DEG in bulk microarray data. CFD was highest in deep dermal ROIs (Supplementary Fig. 8D). We were able to confirm CFD expression at the protein level in mouse skin using IHC (Supplementary Fig. 8E). In pet dog affected hair follicles, the top GO terms were “cellular process”, “biological regulation”, “response to stimulus”, “metabolic process” as well as “developmental process” and “multicellular organismal process” (Supplementary Fig. 8F). IRF1 was conserved as a key IFN response gene (Supplementary Fig. 8G). CFD, however, was not enriched in affected hair follicles, but was significantly higher in perifollicular T cells (Supplementary Fig. 8H). In human DLE affected hair follicles, “cellular process” “biological regulation”, “response to stimulus” “metabolic process” as well as “developmental process” and “multicellular organismal process” were the key GO terms (Supplementary Fig. 8I). IRF1 was not significantly higher in DLE affected hair follicles; rather IFIT1 was significantly upregulated in hair follicles (Supplementary Fig. 8J). CFD was enriched in CD45+ ROIs (Supplementary Fig. 8K). Taken together, these data suggest that for proper understanding of biological processes across species, it is necessary to also have knowledge of the functional gene systems to know which genes can compensate for one another in a given biological system. Specifically, similar gene expression networks might be expressed by different cell types, as exemplified by CFD. Our data also recapture the IFN pathways as key biological processes that are conserved across mammalian species during CLE.

CXCR3 ligands are expressed in the hair follicle niche, and CXCR3 deficiency in autoreactive CD4+ T cells results in reduced skin disease scores in mice

Based on our previous individual species analyses, we noted CXCR3 ligands were upregulated across species. We sought to examine hair follicles as a potential source of these chemokines as danger signals (Fig. 6A). CXCL9 was significantly upregulated in OT1 mouse hair follicles (Fig. 6B). In dog, CXCL10 was the key CXCR3 ligand upregulated in affected hair follicles (Fig. 6C). CXCL11 was the key CXCR3 ligand upregulated in affected hair follicles (Fig. 6D). We confirmed the ligands were elevated in bulk microarray data from both CD4+/OT2 and CD8+/OT1 recipient mouse models (Fig. 6E). We also confirmed expression at the protein level using REX3 reporter mice, which we had bred onto CLE recipient background (Fig. 6F; Supplementary Figs. 9, 10).

Fig. 6. CXCR3 ligands are expressed in CLE mice, dogs and humans, and CXCR3 deficient OT2 cells are less able to induce skin disease in recipient mice.

A Schematic of hair follicle ROI analysis (Created in BioRender. Richmond, J. (2026) https://BioRender.com/mpx60td). B CXCR3 ligand expression in affected hair follicles of mice, C dogs and D humans (one-way Kruskal–Wallis tests with Dunn’s multiple comparisons significant as indicated). E CXCL9/10 expression from Microarray data from mouse skin (n = 4 OT1 (3 male, 1 female), 4 OT2 (2 male, 2 female), 8 littermate (6 male, 2 female) and 4 uninjected control mice (males); two-way ANOVA with Šídák’s multiple comparisons tests significant as indicated). F Quantification of REX3 reporter expression in skin from OT1 and OT2 recipients versus littermate controls (n = 5 littermates (4 male, 1 female), 3 OT1 recipients (male) and 3 OT2 recipients (2 female, 1 male) pooled from 2 to 3 experiments; two-way ANOVA with Tukey’s posttests significant as indicated). G Median fluorescence intensity (MFI) of CXCR3 on the indicated immune cells from healthy or lupus donor blood (n = 6 lupus and 4 healthy donors; two-way ANOVA with Sidak’s posttests significant as indicated). H Example flow cytometry staining for CXCR3 and OT2 TCR Vα2 chain on WT and CXCR3KO OT2 donor T cells. I Example photographs of WT OT2, CXCR3KO OT2 CLE recipients and uninjected controls. J Quantification of skin disease scores (n = 6 WT recipients, 6 CXCR3KO recipients, 6 littermates and 3 uninjected controls, all males, pooled from 2 separate experiments; one-way ANOVA with Tukey’s posttests significant as indicated). K Spleen weights from mice (one-way ANOVA with Tukey’s posttests significant as indicated). L Example ANA staining (scale bar 20 µm). M ImageJ quantification (CTCF) of ANA staining intensity from sera from representative mice (n = 3 WT, 4 CXCR3KO, 1 uninjected control and example PL2-3 positive control; one-way ANOVA with Tukey’s posttests significant as indicated). Error bars represent the standard error of the mean (SEM).

Next, we examined expression of the CXCR3 receptor. We were unable to detect increases in the microarray data, as CXCR3 is likely regulated at the protein level due to GPCR recycling. Therefore, we queried our previously deposited human lupus and healthy peripheral blood flow cytometry datasets (Supplementary Fig. 11, FlowRepository accessions FR-FCM-Z4PL, FR-FCM-Z4PM, FR-FCM-Z4PN, FR-FCM-Z4PQ, FR-FCM-Z4PX, FR-FCM-Z6UN, FR-FCM-Z7ZP, FR-FCM-Z7ZQ). We found that CXCR3 median fluorescence intensity (MFI) was higher on multiple cell types in lupus blood compared to healthy blood (Fig. 6G). We also confirmed expression of CXCR3 in all 3 species using immunohistochemistry (Supplementary Fig. 12). Next, we bred CXCR3KO mice to OT2 donor mice and used them as donors in the CLE model (Fig. 6H, I). Mice that received CXCR3KO OT2 cells had lower skin disease scores than WT recipients (Fig. 6J). Spleen weights and ANAs were also reduced (Fig. 6K–M). Taken together, these data support the role of CXCR3 expression on T cells in driving CLE.

B cells are enriched in CLE skin, and B cell depletion is efficacious for skin disease and alopecia in OT1 recipient mice

Next, we assessed what immune cell subsets were infiltrating the skin during DLE across mammalian species. We used the cell type enrichment analysis module in nSolver software on bulk RNA microarray data, and predictive scores are reported for cell types that could be accurately predicted according to a p cutoff of 0.05 in the analysis software. The mouse models demonstrated significant increases in neutrophils and macrophages (Fig. 7A). To ascertain whether B cells were present, we also analyzed individual genes in case the phenotype of B cells was changed in response to the inflammatory millieu. Fcgr2b expression demonstrated increases in OT1 and OT2 recipients compared to uninjected controls and littermates (Fig. 7B). Cell type enrichment analysis in dog and human revealed significant increases in cytotoxic cells, B cells, neutrophils and T cells compared to healthy margin skin (Fig. 7C, D). Clinically, B cell depletion is used as a therapy for lupus. Therefore, we tested whether our mouse model recapitulates this aspect of disease. We also tested IL7R/CD127 blockade, which is typically expressed on memory T cells, given that this gene was enriched in recipient mice and that total T cell depletion has not worked for lupus treatments. B cells were depleted using the protocol in Keren et al. using a combination of anti-CD19/B220/CD22/Kappa light chain antibodies and were then entered into the OT1 model. CD127 blockade was performed thrice weekly beginning the day after disease induction. B cell depletion but not CD127 blockade prevented skin disease in mice (Fig. 7E, F). Interestingly, CD127 blockade but not B cell depletion reduced spleen weights to littermate control levels (Fig. 7G). We examined T and B cell numbers in skin by flow cytometry and confirmed that B cell depletion reduced B cells and T cells, though not statistically significantly so (Fig. 7H, I, Supplementary Fig. 13). CD127 blockade reduced skin T cells but not B cells. We also examined the antigen specific T cells using TCR antibody and found that CD127 blockade, but not B cell depletion, reduced antigen-specific T cells in skin (Fig. 7J, Supplementary Fig. 14). Paradoxically, B cell depletion seemed to increase antigen-specific T cell numbers in skin, which may be due to increased space in the skin. We also assessed ANAs from these mice, which demonstrated that B cell depletion significantly reduced ANA titers compared to isotype treated mice (Fig. 7K, L). Taken together, these data indicate that the OT1 CLE model is dependent on B cells, and that complex interplay exists between skin immune cell pools. Further studies are warranted to examine the recirculation of different lymphocyte populations and how they might be selectively targeted for CLE therapy.

Fig. 7. B cells are enriched in CLE skin, and B cell depletion, but not CD127 blockade, is efficacious for treatment of skin disease in OT1 recipient mice.

A Cell type score quantification from CLE mice. B FCGR2B normalized counts from CLE mice may better predict B cells in skin (n = 4 OT2 CLE (2 male, 2 female), 4 OT1 CLE (3 male, 1 female), 4 OT2 littermates (3 male, 1 female), 4 OT1 littermates (3 male, 1 female) and 4 uninjected mice (males); two-way ANOVA and one-way ANOVA with posttests significant as indicated). C Dog bulk microarray data cell type scores (n = 8 CCLE and 5 healthy dog skin samples; two-way RM ANOVA with Fisher’s LSD posttests significant as indicated). D Human bulk microarray data cell type scores demonstrated increases in cytotoxic cells, B cells, neutrophils and T cells compared to healthy margin skin (n = 6 healthy and 12 biopsies from 10 human DLE patients; multiple unpaired t tests with Holm–Šídák method correction significant as indicated). E Example photos of OT1 recipient mice treated with isotype antibody, B cell depleting antibodies (anti-CD19/20/kappa light chain) or CD127/IL7R antibody. F Skin scores from treated mice (one-way ANOVA with Tukey’s posttests significant as indicated). G Spleen weights from treated mice (one-way ANOVA with Tukey’s posttests significant as indicated). H Quantification of CD19 + B cells and I CD3 + T cells in skin from treated mice. J Quantification of OT1 cells in the skin. K Example ANAs from treated mice (scale bar 20 µm). L ANA grades from blinded reviewers. (unpaired t test isotype vs B cell depletion significant as indicated; n = 6 littermates (5 female, 1 male), 4 isotype treated (2 female, 2 male), 3 B cell depleted (male) and 4 CD127 blockade treated (female) CLE mice pooled from 2 separate experiments). Error bars represent the standard error of the mean (SEM).

Discussion

Here, we present a comparative spatial transcriptomics study of hair follicles and T cells in the context of CLE. Our data recapitulate previous studies identifying key pathways and processes in CLE, such as IFN⁴⁷, sterol biosynthesis⁴⁸ and fibrosis²³. We also identify potential novel targets. CFD is a highly upregulated DEG in our novel OT1 CLE mouse model, which is expressed in mouse hair follicles, dog T cells and human immune cells. CFD plays an essential role in the formation of C3 convertase, leading to heightened complement activation⁴⁹. While the cellular context of this signal was slightly different, the complement pathway terms were identified in several datasets presented here, and C3 immunofluorescence staining occurs in hair follicles in CCCA biopsies⁵⁰. Recent studies have demonstrated a link between CD8+ T cells and complement system activation, showing that GZMK-expressing CD8+ T cells activate complements and promote inflammation by cleaving complements such as C2, C3, C4 and C5⁵¹. This evidence supports our findings that complement pathways are relevant in CLE, further emphasizing the potential target of complement pathways. The CFD inhibitor Danicopan is currently FDA-approved for extravascular haemolysis (EVH) in adults with paroxysmal nocturnal haemoglobinuria (PNH)⁵². Danicopan is currently being tested in clinical trials for Geographic atrophy (GA), a severe form of age-related macular degeneration (AMD)⁵³. It would be interesting to test drugs in the complement inhibitor class such as C1, C3, C5 and Factor B or D inhibitors for topical, intralesional and, if necessary, oral administration preclinically as well as in veterinary and human trials for CLE, especially given that they eye and hair follicle are both immune privileged organs.

Using comparative spatial transcriptomics, we demonstrate that CXCR3 ligands are produced in hair follicles and that CXCR3KO OT2 T cells are less capable of inducing skin disease in our mice. We also noted that CXCL12 is produced by hair follicles. SDF-1/CXCL12 was previously noted in hair follicle diseases to inhibit hair regrowth⁵⁴. It is possible that, given the lack of 100% protection in CXCR3KO OT2 recipient mice, CXCL12 can serve as a redundant signal to recruit autoreactive T cells into hair follicles.

Similar to human DLE, canine DLE is responsive to hydroxychloroquine⁵⁵, tacrolimus⁵⁶ and was recently shown to be responsive to the JAK inhibitor oclacitinib^57,58. Here we show that S100A8/9 were highly conserved between canine and human in DLE keratinocytes. S100A8/A9 were conserved across pet dog and human keratinocytes in DLE lesions, but not in uninvolved skin. While functional studies are still needed, their concentration in follicular and epidermal regions may point to a specific role in DLE. Small molecule inhibitors of S100A8/9 are in preclinical development. Most studies have used these inhibitors to treat myocardial dysfunction⁵⁹, with some recent work demonstrating S100A8/9 inhibition prevents COVID-19 induced lung damage⁶⁰. Additionally, these proteins have been associated with liver fibrosis associated with cirrhosis⁶¹. It would be interesting to test drugs in this class for topical, intralesional and, if necessary, oral administration in veterinary and human trials.

Adachi et al. demonstrated that hair follicle-derived IL-7 and IL-15 mediate skin-resident memory T cells at homeostasis and during cutaneous T cell lymphoma (CTCL), specifically demonstrating dependency of CD8+ Trm on IL-15 and IL-7, and CD4+ Trm on IL-7⁶². Christiano lab demonstrated that either IL7 or IL15 blockade prevents alopecia areata in C3H mice^63,64. Here we found that blockade of IL7 signaling pathways was not sufficient to prevent disease in OT1 CLE mice. While these pathways are not unique disease markers, their pathogenic and protective roles vary by context, emphasizing the importance of tailored preclinical testing. Further investigation is required to understand the context of IL7 signaling in scarring alopecia.

Limitations of our study include small sample size, targeted panels and limited morphology markers. We also only examined the TLR9KO background of lupus, yet there are several genetically driven mouse models that recapitulate features of lupus organ involvement. Despite this, higher TLR9 expression in cutaneous lupus is associated with positive response to hydroxychloroquine⁶⁵, and given that our mouse model still recapitulates gene expression in human DLE, we may be able to use this for exploration of novel CLE treatments for recalcitrant patients. DBI total body knockout mice exhibit alopecia⁶⁶. However, in our mouse model, we noted an increase in DBI expression in OT1 CLE hair follicles. Further studies would need to be conducted to better understand the context of DBI activity in the TLR9KO lupus-prone background. We also note the inability to micro-dissect hair follicle sub-anatomical niches as a weakness. Last, our human biopsies were aged, resulting in lower counts, though samples did pass quality control. In the future, we plan to prospectively collect samples for flash frozen RNA spatial analyses, and to employ spatial sequencing tools in development such as the 10X Genomics canine panel.

Methods

Mouse model

Mouse studies were conducted at UMass Chan on an IACUC-approved protocol at an AAALAC approved barrier facility. Experimental and control animals were co-housed for all experiments. Animals were euthanized by CO₂ asphyxiation and secondary cervical dislocation per AVMA guidelines. Mice were on the B6 genetic background, and both male and female animals of 8 weeks of age or greater were used in experiments. Numbers of animals in each group and numbers of experimental repeats are reported in each figure legend.

We generated a B6 version of the previously published Balb/c cutaneous lupus erythematosus (CLE) model¹⁸ by breeding K5-TGO mice²² to TLR9^−/− mice⁶⁷ to generate lupus-prone recipients that express the OVA model autoantigen in keratin 5-expressing cells. We abbreviate these recipient mice, whose genotypes are K5-TGO+ rtTA+ TLR9^−/−, as CLE mice. Littermates encompass the genotypes K5-TGO−, rtTA−, TLR9+/+, TLR9+/− and combinations thereof.

To model CLE, we used OT2 T cells (JAX # 004194) instead of DO.11T cells to induce disease in these mice by transferring 10 million Th2 skewed T cells as previously described⁶⁸. Mice received doxycycline chow to turn on OVA expression in K5+ cells, and developed CLE-like skin lesions in 3–4 weeks. Disease scores were measured on a scale of 0–4 with 0 being no evidence of skin disease, score of 1 representing up to 25% body surface involvement (BSI), score 2 up to 50% BSI, score 3 up to 75% BSI and score 4 up to 100% BSI. Our BSI score is inclusive of both alopecia and rash.

Alternatively, we transferred 1 million freshly isolated OT1 T cells (JAX # 003831) to induce disease. Mice received doxycycline chow to turn on OVA expression in K5+ cells. This model is similar to the Okiyama and Katz OT1 transfer model of GVHD in WT recipients that includes inflammation and fibrosis of the skin and hair²¹; however our mice are TLR9KO which makes them lupus-prone. We therefore adopted a similar scoring system from this study, scoring five criteria of (1) rash, (2) alopecia, (3) mucosal involvement, (4) hunched appearance, and (5) weight loss on a scale of 0-2 with 0 representing no involvement, 1 mild and 2 severe involvement. Weight loss was scored <5%-0, 5-15%-1, >15%-2, and the maximum score an individual mouse could receive was 9 points. To specifically assess skin disease, we also separately compiled the criteria of (1) rash, (2) alopecia, (3) mucosal involvement, which could have a maximum score of 6. Mice developed peak skin disease at week 2-3.

All T cells were isolated from donor spleens using MojoSort kits (Biolegend, CD4 kits for OT2 and CD8 kits for OT1) per the manufacturer’s instructions. All recipients in both models were sublethally irradiated with 400R to facilitate engraftment of the transferred T cells.

Mouse model treatments

To deplete B cells in vivo, we followed the protocol in Keren et al.⁶⁹. Briefly, mice were injected intraperitoneally with the following mixture of monoclonal antibodies at 150 μg/mouse each: rat anti–mouse CD19 (clone 1D3), rat anti–mouse B220 (clone RA36B2), and mouse anti–mouse CD22 (clone CY34). After 48 h, the mice were injected with a secondary antibody mouse anti–rat κ (clone TIB216) at 150 μg/mouse. Alternatively, mice received isotype control antibodies. To block IL7R/CD127 signaling, we followed the protocol in Gratz et al.⁷⁰. Briefly, mice were injected intraperitoneally with anti- IL-7Rα antibody clone A7R34 (500 µg per mouse) 2×/week for a total of 3 weeks. All antibodies and sterile PBS vehicle were purchased from BioXCell as InVivo Mab grade (Lebanon, NH).

Canine biopsies

Biopsies from client-owned companion dogs were selected from the Tufts Cummings School of Veterinary Medicine biorepository. Samples were deposited with written owner consent on an IACUC-approved protocol. Cases were re-reviewed by a board-certified veterinary dermatologist (RMA) and were sectioned onto Leica bond plus slides after floating in RNAse free water for use in the Canine Cancer Atlas (CCA) Digital Spatial Profiling assay (NanoString).

Canine FFPE bulk RNA sequencing

RNA was isolated from 30 μM curls using the Qiagen RNEasy FFPE kit per the manufacturer’s instructions. Samples were shipped to GeneWiz for processing and sequencing, using the NEBNext Ultra II RNA library prep kit with sequencing on Illumina HiSeq 4000 with paired-end 150 bp read configuration. FASTQ files were aligned to canfam 1 for further processing.

Human biopsies

Human DLE scalp biopsies were selected from the Howard University biorepository on an archival tissue IRB protocols that were approved by the UMass Chan IRB protocol #H00020503 and the Howard University IRB protocol # IRB-21-MED-13, with a Memorandum of Understanding (MOU) and Material Transfer Agreement (MTA) for cross-institutional studies. The approving committees have provided a HIPAA waiver for use of archival material. Cases were re-reviewed by a board-certified pathologist (BS) and dermatologist (CF). Blocks were sectioned onto Leica bond plus slides in the UMass Chan morphology core and were prepped for spatial analyses in the UMass Chan SCOPE core.

Spatial transcriptomics

Biopsies from affected hair-bearing skin from mice, dogs or humans were sectioned onto Leica Bond Plus slides and were stored at 4 °C in slide boxes with desiccant until use (3 weeks maximum). NanoString GeoMX Digital Spatial Profiling platform was used for cross-species comparisons. Mouse and Human Whole Transcriptome Atlas (WTA) assays and Canine Cancer Transcriptome Atlas (CCA) were used to analyze samples. CD3, CD8 and CD45 were used as morphology markers with SYTO13 nuclear stain (antibodies in Supplementary Table 1). GeoMX sample preparation was performed in the UMass Chan SCOPE core (RRID:SCR_022721). Region of interest (ROI) selection was based on hair follicle morphology, of which we included stromal cells defined as CD45- from morphology marker selection, and cell segmentation using the same ROI selection methods for mice, dogs and humans for consistency. Collection plates were sequenced on an Illumina HiSeq in the UMass Chan High Throughput Sequencing Core. Data were analyzed in the GeoMX data analysis suite including biological probe QC, and Q3 normalization. For T cell ROIs, housekeeping gene normalization was also performed using the NanoString housekeeping gene list from the human TCR microarray panel threshholded to 1%.

Comparison to publicly available datasets

We performed gene expression analysis comparison of skin from our mice to GEO datasets GSE186075²³ and GSE95474⁶ from human PCA and DLE, respectively. We also compared the OT1 CLE model to human GVHD GSE216645⁷¹, as the original model was described as a GVHD model. Top tables were generated using Geo2R software and were exported into Excel. The human gene lists were truncated to match the common denominator genes in the mouse microarray panel as previously described using the VLOOKUP IS ERROR function¹⁷. Tables were sorted based on p-value and normalized expression, and differentially expressed genes (DEG) were compared using BioVenn software⁷².

Histology & immunohistochemistry

H&E staining, trichrome staining, Verhoeff Van Gieson and IHC staining were performed in the UMass Chan Morphology Core. Complement Factor D polyclonal antibody (Thermo cat # PA579034) or isotype control were used at 1:100 dilution using Leica Bond Plus autostainer.

Immunohistochemical studies for CXCR3 were performed on 5-μm sections from formalin-fixed, paraffin-embedded skin samples in the Comparative Pathology & Shared Genomics Resource (CPGSR) at Tufts (RRID:SCR_028706). After deparaffinization and rehydration, the antigen retrieval was performed with EDTA buffer (pH 8.5) for human and mouse skin, or citrate buffer (pH 6.0) for canine skin. The sections were permeabilized with IntelliPATH FLX Buffer (TBS w/tween), and then treated with 3% hydrogen peroxide. The sections were incubated with Novus Biological Rabbit CXCR3 (0.05–1.0 mg/mL, LS-B10183/WO44003) with a 1:1500 dilution for human/mouse or 1:1000 dilution for dog for 30 min at room temperature, followed by MACH 2 Rabbit polymer (Biocare RHRP520L). The staining was visualized with Biocare Betazoid DAB Chromogen Kit (BDB2004H). All the sections were counterstained with hematoxylin, and images were taken using a 3DHistech Pannoramic Midi Digital Slide Scanner.

QuPath immunohistochemistry quantification

IHC whole-slide images were analyzed using QuPath v0.6.0. First, the DAB (3,3′-diaminobenzidine) stain vector was added to the project to enable accurate color deconvolution. For each slide, a closed polygon annotation was manually drawn to encompass the entire tissue section. Cell detection was then performed using Analyze → Cell Detection, which identified all nuclei and cell boundaries within the annotated region. To classify DAB-positive cells, we used Classify → Object Classifier → Create Single Measurement Classifier, which was trained on representative regions and subsequently applied uniformly to all samples. Quantification was calculated as (number of DAB-positive cells/tissue area in µm²) ×10⁶ to normalize counts across samples. Statistical comparisons between groups were performed using an unpaired t-test in GraphPad Prism 10.6.0.

Flow cytometry

Healthy mice skin and CLE mice skin were harvested, minced into small pieces, and digested in the skin digesting buffer with 2.0 mg/ml collagenase XI from Clostridium histolyticum (Sigma-Aldrich), 0.5 mg/ml hyaluronidase from bovine testes (Sigma-Aldrich), and 0.1 mg/ml DNAse (Sigma-Aldrich) for 40 min at 37 °C. The single cell suspensions were washed with RPMI 1640 (Corning, #20623011), filtered through a 40 μm cell strainer (Fisher Scientific, #22363547), and stained with fluorescent antibodies (Supplementary Table 2) for flow cytometry. The stained samples were analyzed using a Cytek Aurora cytometer (Cytek) and data analysis was performed with FlowJo 10.10.0 software. Examples of mouse skin flow cytometry gating strategy for B cells and host and antigen-specific T cells are available in Supplementary Figs. 13, 14, respectively.

2 photon in vivo microscopy

We performed in vivo imaging of ear skin in mice that received Kikume OT1 T cells as previously described²⁵. Briefly, mice were anesthetized with ketamine/xylazine solution and were placed in a 3D printed stage. Images were captured with a Bliq Photonics Upright multiphoton video-rate microscopy system with a Spectra Physics Insight DeepSee laser, Olympus 20× XLUMPlanFL 0.95 NA water immersion objective, 520/40 nm filter for the green channel, a 630/92 nm filter for the red channel and Axicon volumetric imaging. The excitation wavelength was 920 nm and 100 nm averaging was used. Images were compiled in Fiji and total duration of interactions of T cells with skin is reported.

Data analysis & statistics

Two-way comparisons for differentially expressed genes (DEGs) and pathway analyses were performed in GeoMX software to calculate volcano plots, and individual hypotheses of gene families were tested by querying specific gene sets within the Q3 normalized count file. Data were graphed with GraphPad Prism software, and ANOVAs with post-tests were performed for groups of 3 or more, and t tests for comparing 2 groups, using p < 0.05 as a cutoff. Data underwent normality testing to determine if parametric or non-parametric tests should be used, and tests are reported in the corresponding figure legends.

Author contributions

Conceptualization – J.M.R. Methodology - C.B., M.A. Software - N/A. Validation - R.A., W.A., L.Y., E.S., and H.B. Formal analysis - U.Y.A., A.R., and J.M.R. Investigation - U.Y.A., N.S., R.L., H.R., Y.Z., T.C., K.M., S.S., M.D., Q.T., M.O., F.B., and J.M.R. Resources A.M.R., M.D.R., B.S., A.S.B., C.F., C.B., A.M., A.K., R.M.A., and J.M.R. Data Curation - U.Y.A., R.L., and J.M.R. Writing - Original Draft - U.Y.A., J.M.R. Writing - Review & Editing - all authors. Visualization - U.Y.A., J.M.R. Supervision - J.M.R., C.F., and C.B. Project administration – J.M.R. Funding acquisition - J.M.R., C.B., W.A., and Y.Z.

Peer review

Peer review information

Nature Communications thanks Raphael Clynes, Seon-Pil Jin and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Supported by a Target Identification in Lupus Award from the Lupus Research Alliance, a Mechanisms & Targets Award from the Lupus Research Alliance, and startup funds from UMass Chan (JMR). WA and YZ were supported by Diversity Research Supplement Awards from the Dermatology Foundation.

Data availability

The Mouse microarray data have been deposited in the Gene Expression Omnibus (GEO) Database under accession GSE285892. The Canine bulk FFPE RNA-seq data have been deposited in the Sequence Read Archive (SRA) Database under accession PRJNA1119020. The canine, mouse, and human spatial transcriptomics datasets have been deposited in the SRA Database under accession codes PRJNA1156903, PRJNA1170259, and PRJNA1478613, respectively. The following publicly available datasets, deposited on the GEO Database, were analyzed in the study: GSE182825, GSE186075, GSE95474, and GSE216645, The Human blood flow cytometry data have been deposited in FlowRepository under accession #s FR-FCM-Z4PL, FR-FCM-Z4PM, FR-FCM-Z4PN, FR-FCM-Z4PQFR-FCM-Z4PX, FR-FCM-Z6UN, FR-FCM-Z7ZP, FR-FCM-Z7ZQ. The mouse flow cytometry data have been deposited in FigShare under [10.6084/m9.figshare.32193621], All data are included in the Supplementary Information or available from the authors, as are unique reagents used in this Article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided with this paper.

Competing interests

J.M.R. and R.M.A. are inventors on a patent application for the diagnosis of skin conditions in veterinary and human patients (#63/478,900). J.M.R. is an inventor on patents for targeting CXCR3 (0#15/851,651) and IL15 (# 62489191) for the treatment of vitiligo. A.S.B. is the inaugural recipient of the Skin of Color Society Career Development Award as well as the Society for Investigative Dermatology Freinkel Diversity Fellowship Award, and a recipient of the Robert A. Winn Diversity in Clinical Trials Career Development Award (Winn CDA) funded by Bristol Myers Squibb Foundation (BMSF); she is a consultant for Senté, Inc. and Sonoma Biotherapeutics. F.B. is an employee of Bliq Photonics. All other authors declare no financial conflicts of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-76048-8.