Myeloid infiltration and epidermal dysregulation characterize cutaneous photosensitivity in systemic lupus erythematosus

Abstract

For patients with systemic lupus erythematosus (SLE), sun avoidance and sunscreen use are the only preventive therapies for photosensitivity, a process by which exposure to ultraviolet (UV) light triggers cutaneous and systemic flares of disease. Photosensitivity is exceedingly common amongst patients with SLE who develop rashes, affecting up to 93% of this population. Treatment of skin and systemic disease triggered by the sun can be challenging, and thus there exists a critical need for better therapeutic and preventative strategies. Innovation of targeted therapies for photosensitive skin diseases requires a precise dissection of the cellular players and signaling pathways that contribute to aberrant inflammation following UV exposure. Here, we used cutaneous UVB photoexposure of healthy control and patients with SLE to provide an atlas of the cutaneous inflammatory response 24 hours after UV exposure in single-cell resolution. We demonstrate that the epidermis is dysregulated in lupus, where epidermal-dermal crosstalk in SLE is dominated by inflammatory keratinocytes following UV exposure in contrast to the healthy control epidermis. In both healthy control and SLE, the downstream consequence of UV exposure is the recruitment of a population of monocyte-derived dendritic cells (moDCs). Although moDCs are present in relatively equal proportions in SLE and healthy control UV-treated skin, they appear to be derived from distinct circulating monocyte populations. SLE UV moDCs are characterized by inflammatory and antigen presentation pathways and may represent a precursor to lesional cutaneous lupus CD16⁺ DCs, identifying a potential link between cutaneous UV responses and cutaneous lupus flare.

One Sentence Summary:

In photosensitive lupus, dysregulated cutaneous UV responses recruit and activate monocyte-derived dendritic cells that may drive lesion development.

INTRODUCTION

Exposure to ultraviolet (UV) light normally results in anti-inflammatory responses and is used to treat various types of inflammatory skin diseases (1). However, patients with systemic lupus erythematosus (SLE) can experience flares of cutaneous and systemic disease following even moderate doses of UV light in a phenomenon termed photosensitivity (2). SLE is a debilitating chronic autoimmune disorder that commonly affects the skin and is characterized by overproduction of type I interferons (IFNs). Up to 93% of patients with SLE who experience skin lesions, called cutaneous lupus erythematosus (CLE), are photosensitive (3). Despite advances in mechanistic understanding of photosensitive responses in SLE and CLE, sun avoidance and sunscreen usage remain the only preventative therapies for UV-induced flares, and compliance rates are low (4). Further, flares are difficult to treat, contributing to poor quality of life (5), and many medications have severe side effects or infection risks. With no FDA-approved therapies specific for CLE lesions, there exists a critical need for a precise dissection of mechanisms that drive aberrant inflammation in lupus skin exposed to UV light.

UV light exists in UVA, UVB, and UVC wavelengths (2). UVB, which is primarily absorbed by the basal layer of the epidermis (6), is the best described of the three to trigger photosensitive skin rashes in lupus (7). Chronic, elevated cutaneous cytokines, primarily type I IFNs (8–10), are thought to drive cutaneous photosensitive responses following UVB exposure through priming for inflammatory responses, such as increased interleukin 6 (IL-6) production and increased keratinocyte apoptosis (8, 11). Activation of early IFN responses following cutaneous UV exposure may also contribute to cutaneous myeloid cell recruitment and activation (8, 12–14). Although some studies have explored differences between lupus and health control (HC) skin following UV exposure (15–17), a summative analysis of the transcriptional differences between the HC and SLE skin response to UV at a single-cell level has not yet been explored.

Here, we used single-cell RNA-sequencing (scRNA-seq) to delineate the early cutaneous immune response to UVB irradiation in nonlesional, sun-protected skin from patients with SLE and HC participants. In doing so, we have identified monocyte-derived dendritic cell (moDC) infiltration as a common response in both HC and SLE skin. However, the SLE UV response is characterized by dysregulation of cutaneous epidermal, stromal, and immune cell responses and skewing of the moDC population towards inflammatory activation. Comparison of SLE UV-treated skin myeloid cells with lupus rash-associated myeloid cells shows an overlap between SLE-derived UV moDCs and inflammatory CD16⁺ DC populations that have been identified in CLE lesions (9). Thus, targeting signals that contribute to moDC activation in lupus skin may help to prevent photosensitive immune activation in SLE.

RESULTS

Single-cell RNA-sequencing of SLE and HC participants reveals distinct UV-induced skin and immune cell populations

The skin of patients with SLE produces increased inflammatory mediators after UV exposure, but the changes that occur on a cellular level are not well-understood. To thus map the cellular photoresponses of healthy control and SLE skin, we performed single-cell RNA sequencing (scRNA-seq) on UVB-treated and -untreated skin of seven healthy controls and eight patients with SLE (Table 1) with the goal of elucidating key cutaneous cellular populations and their transcriptional differences following exposure to UVB light (Fig. 1A). Patients underwent UVB phototesting with doses ranging from 10 to 80 mJ/cm². Biopsies were collected from untreated, sun-protected skin sites near the upper buttock and at the minimal erythema dose (MED) or 80 mJ/cm² if MED was not in the tested range. Biopsies were subsequently processed for scRNA-seq. Using uniform manifold approximation and projection (UMAP) visualization, we identified 24 clusters from the skin biopsies that corresponded to 10 major skin-associated cell populations and immune cells, comprising keratinocytes (KCs), melanocytes (MLNCs), fibroblasts (FBs), pericytes (PRCs), endothelial cells (ECs), lymphatic endothelial cells (LECs), Schwann cells, T cells, myeloid cells, and mast cells (Fig. 1, B to E). Neutrophils were not captured by our methodology. Similar percentages of each cell population were captured across disease and treatment conditions (Fig. 1, C and D). As type I IFNs are well-described to drive SLE pathogenesis and are up-regulated after UV exposure (18, 19), we assessed type I IFN responses in UV-exposed SLE versus healthy control skin by the expression of the IFN-stimulated gene IFI6, which was highly modified by UV treatment (Fig. 1F), and composite IFN signatures [derived from (20)] for cutaneous cell populations after UV treatment (Fig. 1G). At 24 hours after UV treatment, healthy control melanocytes, fibroblasts, and endothelial cells exhibited increases in IFI6 and IFN gene scores compared to baseline. By contrast, SLE biopsies demonstrated robust IFN-regulated gene expression across the majority of cell types at baseline with further increases seen after UVB (Fig. 1, F and G). Thus, we find our data replicated the known IFN skewing of SLE skin following UV exposure, validating scRNA-seq as a tool to study photoresponses in SLE skin.

Table 1.

Demographics of participants enrolled for phototesting and biopsy.

Participant	Age	Sex	Race	Ethnicity	Fitzpatrick score	SLEDAI	CLASI	CLE subtype	MED biopsy site mJ/cm2	Lupus Medications at time of biopsy
HC	33	F	Black	non-H	V	N/A	N/A	N/A	70	N/A
HC	63	F	White	non-H	III	N/A	N/A	N/A	60	N/A
HC	76	F	White	non-H	III	N/A	N/A	N/A	40	N/A
HC	19	M	Asian	non-H	III	N/A	N/A	N/A	40	N/A
HC	19	F	Black/White	non-H	III	N/A	N/A	N/A	40	N/A
HC	23	M	Black	non-H	IV	N/A	N/A	N/A	80	N/A
HC	27	F	Asian	non-H	III	N/A	N/A	N/A	40	N/A
SLE	54	F	White	non-H	II	8	2	ACLE	80	HCQ, hydrocortisone 20 mg BID
SLE	42	F	White	non-H	II	0	6	DLE	40	none
SLE	43	F	Black	non-H	V	4	5	SCLE	80	Belimumab, Myfortic, HCQ
SLE	42	F	White	non-H	II	2	1	DLE and lupus panniculitis	60	HCQ, Cellcept, prednisone 5 mg daily
SLE	44	F	Black	non-H	VI	8	4	DLE	80	Rituximab
SLE	33	F	Black	non-H	V	2	1	DLE	80	HCQ, belimumab, prednisone 5 mg daily
SLE	63	F	Black	non-H	V	0	0	ACLE	80	none
SLE	74	F	White	non-H	II	2	4	SCLE	30	HCQ

All data reflect disease state at time of UV exposure and biopsy. ACLE, acute cutaneous lupus erythematosus; DLE, discoid lupus erythematosus; F, female; HC, healthy control; HCQ, hydroxychloroquine; M, male; MED, minimal erythema dose; non-H, non-Hispanic; SCLE, subacute cutaneous lupus erythematosus; SLE, systemic lupus erythematosus.

Fig. 1. Single-cell RNA-sequencing of SLE and HC participants reveals distinct UV-induced skin and immune cell populations.

(A) Study overview. Eight patients with systemic lupus erythematosus (SLE) and seven healthy control (HC) participants were exposed to varying doses (0 to 80 mJ/cm²) of UVB light on non–sun-exposed skin, which were biopsied and processed for scRNA-seq 24 hours later. Graphics created with BioRender (89). (B and C) Uniform manifold approximation projection (UMAP) demonstrating (B) major cell types captured by single-cell RNA sequencing and (C) major cell types by disease and treatment. (D) Proportion of each major cell type by disease and treatment condition. (E) Marker genes for each major cell type. (F and G) Violin plots showing expression of (F) IFI6 or (G) interferon (IFN) gene score in SLE and HC skin samples with or without UVB exposure. [(F) and (G)] Data were analyzed by Welch’s t test with comparisons shown between SLE-UV and HC-UV. ***P ≤ 0.001 and ****P ≤ 0.0001.

Inflammatory keratinocytes exhibit hyperinflammatory responses in SLE skin

The basal epidermis is primarily composed of basal KCs and absorbs the majority of UVB (6). We therefore examined whether KC subpopulations exhibited differential responses to UVB light in SLE versus HC subjects. We subclustered KCs (Fig. 2, A and B, and fig. S1 A and B) and identified two subclusters in both SLE and HC samples that were primarily present after UV exposure (Fig. 2C). These clusters were transcriptionally similar to basal and spinous KCs but were characterized by expression of stress keratin and S100 protein transcripts and were thus termed basal inflammatory and spinous inflammatory KCs, respectively (Fig. 2, B and C, and fig. S1B). Consistent with our previous report (9), SLE skin exhibited higher proportions of basal and spinous inflammatory KCs relative to HC skin at baseline; however, both SLE and HC exhibited an increase in these populations after UVB exposure (Fig. 2C and fig. S1, C and D).

Fig. 2. Inflammatory keratinocyte subsets are expanded following UVB exposure.

(A and B) UMAPs depicting (A) disease and treatment source of each subcluster and (B) cell subsets corresponding to the subclusters. (C) Barplot revealing proportions of each keratinocyte subset within each disease and treatment condition. (D) Representative immunofluorescent images of SLE (n = 3) and HC (n = 4) skin treated with or without UVB stained for keratin 14 (red) to denote basal keratinocytes (KC), keratin 10 (purple) to denote more superficial KC populations, and IFI27 (green) as an IFN-regulated gene. (Right) Quantification of IFI27 in the basal epidermis (as defined by keratin 14 staining). Scale bars, 50 μm. (E) Volcano plot depicting HC or SLE skewed gene expression in basal inflammatory KCs after UVB treatment as calculated by the difference in gene expression of UV compared to no UV in (right side) SLE versus HC or (left side) HC versus SLE. (F) Upstream regulators (URs) as defined by Ingenuity Pathway Analysis (IPA) for SLE UV versus HC UV Basal Inflammatory KCs. (G) IFNB gene expression in SLE KCs (n = 6 patients, treated in triplicate) cultured in vitro with DMSO (vehicle control) or rapamycin (to inhibit mTOR) pretreatment prior to UVB. [(D) and (G)] Data were analyzed by one-way Anova corrected for multiple comparisons. *P ≤ 0.05, ***P ≤ 0.001, and ****P ≤ 0.0001. n=8 SLE and 7 HC for all analyses except where indicated.

We then asked how these cell populations were differentially skewed by UV light in SLE versus HC skin. SLE basal inflammatory and spinous inflammatory KCs primarily upregulated inflammatory genes including chemokines like CXCL14 and IFN-stimulated genes like IFI27 and IFI6. This increased expression of IFI27 was confirmed by immunostaining (Fig. 2, D and E). HC basal inflammatory and spinous inflammatory KCs primarily upregulated stress response genes like S100A8 and S100A9 and cell cycle components like RAN and TUBA1C, ultimately suggesting that SLE inflammatory KCs may be more active in promoting cutaneous inflammation following UVB (Fig. 2, D and E, fig. S1E, and tables S1 and S2).

To assess the factors upstream of these transcriptional differences, the predicted upstream regulators (URs) that drive these differential KC responses between SLE and HC were compared using Ingenuity Pathway Analysis (IPA) (Fig. 2F and fig. S1F). Cytokines such as IL-9 and IL-27 and drivers of IFN-related responses such as TLR7, IFN-λ, and RIG-I were identified as significant (Z score > 2.5) URs in SLE basal KCs. RICTOR, a critical component of mTOR complex 2 (mTORC2), was the top predicted upstream regulator in SLE UV basal inflammatory KCs. mTORC2 and RICTOR are critical for the transcription of IFN stimulated genes (21), highlighting a potential novel regulator in UV-induced IFN responses in SLE (fig. 2F). In contrast to SLE, HC basal inflammatory KC responses were predicted to be driven by RB1 and STK11, which have important roles in the DNA damage response (22, 23) (fig. 2F). Additionally, HC basal inflammatory KC responses were predicted to be driven downstream of mediators like NFE2L2, which suppresses production of the cytosolic DNA sensor STING and limits production of pro-inflammatory cytokines (24, 25). We next tested the role of mTORC2 signaling in type I IFN production following UV exposure in SLE versus HC KCs using rapamycin, which inhibits mTORC1 with short incubations (1 hour) and mTORC2 with longer exposures (24 hours) (26). In SLE KCs, rapamycin repressed UV-induced IFNB production with long, but not short, incubation, confirming that mTORC2 complex regulation may be an important mediator of downstream UV responses in SLE KCs. Similar effects were not found in HC KCs (fig. S2G and fig. S1G). SLE spinous inflammatory KC responses did not exhibit URs that drive IFN or other inflammatory responses, highlighting the hyperinflammatory nature of basal inflammatory KCs even relative to other lupus KC populations (fig. S1E) (8, 27). Additionally, HC spinous inflammatory KC responses were predicted to be driven by KLF4 and SOCS3, which function to maintain the skin barrier and suppress cytokine production, particularly IL-6 (28–30) (fig. S1E) (31). Taken together, these data indicate a hyperinflammatory response to UVB in the SLE basal epidermis in contrast to HC basal inflammatory KCs, which exhibit a more regulatory, anti-inflammatory phenotype.

Given that UVB-treated SLE basal inflammatory KCs demonstrated this skewing, we questioned what unique factors contributed to the formation of this population. We used pseudotime trajectory analysis to explore the relationship between UVB-treated and -untreated basal KC and basal inflammatory KCs in SLE and HC. SLE and HC KCs were clustered in separate UMAPs and pseudotime trajectories were generated for HC and SLE samples (fig. S2, A and B). In both HC and SLE, UV basal inflammatory KCs populated the end of the trajectory, indicating that these cells represent an inflammatory state derived from basal KC precursors (fig. S2, C and D). Analysis of the top 80 upregulated genes across each of the SLE and HC trajectories revealed differences among these groups (fig. S2, C and D). Up-regulation of S100 genes was common across both trajectories, whereas SLE KCs upregulated IFN-stimulated genes (ISGs) (IFI27), genes related to oxidative stress (TXNDC17, SRM, and GSTP1) and genes related to metabolism (GAPDH, NDUFA4, and SLC25A5) as they develop into UV basal inflammatory KCs. By contrast, the HC end-trajectory states demonstrated upregulation of genes related to maintenance of skin hydration and integrity (AQP3 and KRT6C), confirming a more balanced inflammatory and protective response to UVB exposure in the HC epidermis.

Finally, we sought to understand the unique soluble signals that contribute to formation of SLE UV basal inflammatory KCs compared with healthy controls. Differentially expressed genes were analyzed between the beginning- and end-trajectory states in SLE and HC, and predicted upstream regulators (URs) were identified using IPA. Predicted cytokine URs that were unique to the SLE trajectory were plotted (fig. S2E) and included inflammatory mediators such as CSF1, IFNL1, CXCL12, and TNFSF10. The SLE-specific UR most predictive (by Z score) of the SLE basal KC to UV basal inflammatory KC differentiation was IL-27, a pleiotropic cytokine produced by dermal myeloid cells following skin injury (32) that can activate STAT1- and IRF3-dependent production of ISGs by keratinocytes (33). Additionally, CSF1 emerged as an upstream regulator. Neutrophils, which we did not capture by scRNA-seq, are known to infiltrated skin shortly after UV exposure (34) and produce CSF1 (35); However, whether CSF1 directly acts on KCs is unclear as KCs typically do not express the appropriate receptor for CSF1. Thus, basal inflammatory KCs are expanded in both SLE and HC following UV exposure, however the signals these cells receive differ and determine downstream UVB-induced inflammatory responses in SLE skin.

Basal inflammatory keratinocytes are the primary epidermal cell transmitting UV-mediated signals to dermal cells following UV exposure in SLE skin

As UV light is primarily absorbed by the epidermis, any perturbations in dermal inflammation likely occur as a result of crosstalk between the epidermis and the dermis. Consequently, we sought to understand the primary dermal cell populations receiving UV-mediated signals from the epidermis and to examine which epidermal populations are primarily transmitting these signals in SLE versus healthy control skin. Using CellPhoneDB, we mapped predicted cell-cell communication via ligands originating from epidermal cell populations, including KCs, Langerhans cells (LCs) and melanocytes, transmitted to receptors expressed by dermal cell populations encompassing ECs, FBs, LECs, mast cells, myeloid cells, Schwann cells, PRCs, and T cells. In total, after UVB treatment, the SLE epidermis engaged in more predicted interactions with dermal cells than the HC epidermis, suggesting that the lupus epidermis is more active in the cutaneous response to UVB than HC (table S3). In both HC and SLE skin, ECs, FBs, myeloid cells, and PRCs were the major dermal cell types receiving the most epidermal signal following UVB exposure (fig. 3A). We next assessed which subsets of epidermal cells were primarily communicating with the dermis. In healthy control skin, LCs and a subset of melanocytes dominated the UV-induced epidermal-dermal crosstalk. By contrast, SLE epidermal-dermal crosstalk was driven primarily by basal inflammatory KCs, suggesting that transmission of UVB-induced signals was guided by a dysregulated epidermal inflammatory response (Fig. 3B and table S3). Analysis of dermal subsets receiving epidermal signal revealed similarities between the SLE and HC dermis UV signal recipients (Fig. 3, C to F). In both SLE and HC, capillary ECs, IL6⁺ FBs, IFN FBs, monocyte-derived dendritic cells (moDCs), NG2⁺ PRCs, and NG2⁻ PRCs received the most epidermal-derived UV signal (Fig. 3, C to F, and fig. S3, A to F). However, differences between SLE and HC were noted in skin-resident myeloid cells; in HC, perivascular macrophages (PVMs) received more epidermal signal, whereas a subpopulation of myeloid cells called lipid-associated macrophages (LAMs) received more epidermal signal in SLE. The dermal recipients were relatively constant across disease states and supports the hypothesis that the aberrant response to UVB in lupus were guided by a dysregulated epidermis.

Fig. 3. Basal inflammatory KCs drive the epidermal-dermal crosstalk in SLE UV-exposed skin.

(A to F) CellPhoneDB was used to assess predicted crosstalk interactions in which ligands were produced by epidermal cells and receptors were produced by dermal cell populations. Number of crosstalk interactions in UV > no UV for HC and SLE was plotted as a heatmap. (A) Total epidermal to dermal crosstalk (B) Total ligand production by subsets of epidermal cells. Epidermal-dermal crosstalk between subsets of epidermal cells and (C) subsets of endothelial cells, (D) subsets of fibroblasts, (E) subsets of myeloid cells, (F) subsets of pericytes. (G and H) CellphoneDB was used to map the total communication connectome in (G) HC UV > no UV and (H) SLE UV > no UV. Nodes represent cell types. Node size is proportional to the total number of predicted crosstalk interactions the node is engaged in when as source. Lines between nodes represent predicted bidirectional crosstalk. Line color is proportional to the number of crosstalk interactions between the two nodes the line connects, with red color representing higher relative interactions. n=8 SLE and 7 HC for all analyses.

Finally, using CellPhoneDB, we assessed the total bidirectional cellular crosstalk landscape following UV exposure in SLE versus HC skin (Fig. 3, G and H). In both SLE and HC UV-treated skin, venule ECs were engaged in a substantial amount of crosstalk, both with epidermal cells and with dermal cells, suggesting they may represent an important intermediate cell type transmitting epidermal UV signal to downstream dermal cells. In HC, venule endothelial cells were engaged in the most crosstalk with skin-resident populations, including KC subsets, LECs, IL6⁺ and IFN fibroblasts, and PVMs (Fig. 3G). Unlike HC, the total crosstalk landscape in SLE was considerably more complex, including a diversity of both stromal and immune cell types (Fig. 3H). SLE LECs were engaged in more dynamic crosstalk with the skin system than their HC counterparts, consistent with recent reports of lymphatic dysfunction contributing to photosensitivity in lupus (36). In both HC and SLE skin, venule ECs engaged in crosstalk with structural cells – primarily basal and spinous inflammatory KCs, IFN and IL6⁺ fibroblasts, and LAMs. In SLE, however, capillary ECs were also engaged and the overall crosstalk was more complex.

SLE skin exhibited more robust crosstalk with a population of myeloid antigen-presenting cells (APCs) defined as moDCs. Although the presence of moDCs was not specific to SLE skin, SLE moDCs were engaged in stronger predicted interactions than their HC counterparts, suggesting that this population may be one of interest in the propagation of inflammation in photosensitive skin disease (Fig. 3H). To better understand how moDCs may communicate differently in UVB-treated SLE versus HC skin, we next analyzed the ligand-receptor pairs involving moDCs. Pairs noted to be unique in either HC or SLE revealed that SLE moDCs uniquely expressed CXCL1 and VEGF (fig. S4, A and B). LAMs were strong receivers of moDC chemokines and cytokines in SLE UVB-treated skin. MoDCs in both HC and SLE skin also received TNF signals, but activating signals from SLE ECs to moDCs included SELE (E-selectin), a molecule upregulated on activated endothelium. To understand whether moDCs from SLE exhibited more robust ligand production for any ligand made by moDCs, the ligand-receptor pairs from moDCs to other cell types were analyzed and the top 20 secreted ligands produced by moDCs were compared between SLE and HC skin after UV exposure. Both HC and SLE UV moDCs produced IL-1β and CCL3; however, HC UV moDCs signaled more via CCL3L1 and CCL4L2, encoding two chemokines that bind to CCR5, a receptor that facilitates recruitment of myeloid-derived suppressor cells to sites of inflammation where they are stimulated to produce IL-10 to activate T_reg cells (37). SLE UV moDCs expressed higher CXCL8, a neutrophil chemoattractant, and TNFSF13B, which generates the lupus-associated cytokine BAFF. This suggests that recruitment of moDCs into lupus skin results in a chemokine environment that facilitates neutrophil recruitment and promotes B cell activation and maturation (fig. S4C). Our data identify a contrasting response to UVB in HC and SLE skin. The HC epidermis exhibits more anti-inflammatory responses and a less diverse activation of inflammatory crosstalk in the dermis. By contrast, the SLE epidermis is more activated and exhibits signals driving ISG activation following UVB exposure, contributing to increased propagation of communication signals likely through venule ECs as an intermediate, leading to heightened crosstalk with immune cells, particularly moDCs.

SLE endothelial cells are skewed by excess type I IFNs

We next evaluated stromal cell populations participating in the cutaneous immune response to UVB exposure. As ECs receive a plurality of the epidermal signal following UV exposure in both HC and SLE skin, we next questioned whether ECs were differentially skewed by the HC and SLE skin environment. We therefore subclustered ECs and identified 10 subclusters corresponding to arteriolar, capillary, and venule ECs (Fig. 4, A to D and fig. S5A). All three EC cell types were represented in equal proportion across disease state and treatment (Fig. 4C). We next queried the cytokine signatures of UV-exposed EC cell types between SLE and HC, identifying increased IL-1b, TNFa, and IFNa signatures in SLE arteriole, venule, and capillary ECs compared with HC (fig. S5, B to D). As both capillary and venule ECs were major recipients of epidermal signals, we evaluated the differentially expressed genes (DEGs) in SLE versus HC UV-exposed capillary and venule ECs. This comparison identified an increase in interferon-stimulated genes like IFI27 and IFI6 and MHC class I molecule transcripts in both subtypes of SLE UV-exposed ECs, indicating that these cells experience type I IFN signaling after UVB treatment (Fig. 4E and table S4). To assess the soluble mediators predicted to drive these transcriptional differences between SLE and HC capillary ECs, we performed upstream regulator analysis using IPA. HC capillary EC responses exhibited more of a T_H2 cell phenotype with predicted upstream regulators such as IL-4 and IL-5, whereas the predicted regulators contributing to SLE capillary EC responses were dominated by IFN signaling (Fig. 4F). HC ECs were predicted to be driven by transcription factors like MYC that have been shown to limit EC-mediated inflammation (38, 39) (fig. S5E). Thus, SLE ECs exhibit inflammatory signaling after UVB exposure primarily dominated by upregulation of interferon-regulated genes.

Fig. 4. SLE UV capillary endothelial cells are skewed by type I, II, and III IFNs.

(A and B) UMAP of SLE and HC endothelial cells by (A) disease and treatment and (B) endothelial subtypes. (C) Bar plot of proportions of each EC subset by disease and treatment. (D) Dot plot of marker genes for each cell type. (E) Volcano plot of DEGs induced by UV treatment in (right) SLE and (left) HC in (top) capillary and (bottom) venule ECs. (F) Upstream cytokine regulators of DEGs between SLE and HC UV (top) capillary and (bottom) venule ECs. n=8 SLE and 7 HC for all analyses.

SLE fibroblasts exhibit diminished anti-fibrotic responses following UVB exposure

Given evidence for connections between EC activation and fibrosis (40, 41), fibroblast (FB) involvement in cutaneous immune responses (9, 42, 43), and their prominence in UV skin crosstalk (Fig. 3, G and H), we next investigated the changes in FB populations after UV exposure. We subclustered fibroblasts and identified seven total subclusters: SFRP2⁺ FBs, IFN FBs, IL6⁺ FBs, APOE⁺ FBs, FMO2⁺ FBs, TNN⁺ FBs, and CLDN1⁺ FBs (fig. S6, A to E). Of these, IFN FBs and IL6⁺ FBs populations were increased in both SLE and HC skin following UV irradiation (Fig. S6D) (P = 0.001 for HC IFN FB, P = 0.025 for SLE IFN FB, P = 0.0004 for HC IL6⁺ FB, and P = 0.27 for SLE IL6⁺ FB via one-sided paired Student’s t-test). IFN FBs upregulated IFN-stimulated genes, indicating that they experienced IFN stimulation after UV exposure. IL6⁺ FBs expressed many of the same IFN signature genes but additionally exhibited upregulation of IL6 transcripts. Upstream regulator analysis of HC and SLE FBs confirmed that IL6⁺ and IFN FBs received dramatically different predicted upstream signals following UV exposure depending on disease state (fig. S6, F and G). In HC skin, IFN FBs exhibit negative regulation of nucleic acid signaling with RNASEH2B and TREX1 as top upstream regulators, whereas SLE IFN FBs received signals from IFNs and reactive oxygen species mediators (fig. S6F). In IL-6⁺ FB, HC skin exhibited an anti-inflammatory program with RUNX1, a transcription factor known to negatively regulate inflammatory genes (44), and IRGM, a negative regulator of the NLRP3 inflammasome (36), identified as top predicted upstream regulators. Anti-fibrotic signals including SMAD7, miR-16-5p, and miR-29b-3p (45–48) were also noted in HC but not SLE IL6⁺ FB. By contrast, in SLE IL6⁺ FB, inflammatory mediators like IFN⍺, TNF⍺, IFNγ, IFNλ, IL-1β, and IL-6 were predicted to drive the transcriptional program after UV exposure (fig. S6G). Thus, the absence of these anti-fibrotic and anti-inflammatory signals upstream of SLE IL6⁺ and IFN FB transcriptional responses to UV suggest a dysregulated fibrotic and inflammatory response in specific lupus FB subsets secondary to a hyperinflammatory cutaneous milieu.

Cutaneous regulatory T cells exhibit exhausted phenotypes in SLE UV-exposed skin

We next sought to examine immune cells in UV-treated SLE and HC skin. Analysis of the T cell compartment revealed five subsets comprising CD4 T cells, regulatory T (T_reg) cells, CD8 T effector memory (T_em) cells, resident CD8⁺ memory T cells (CD8⁺ T_rm), innate lymphoid cells (ILC), and natural killer (NK) cells (Fig. 5, A to D). None of the populations exhibited a substantial shift in proportion based on disease or UV treatment (Fig. 5C). We next probed activation, cytotoxicity, interferon-stimulated gene (ISG), and exhaustion signature scores as described (49) (Fig. 5, E to H). Compared with HC, SLE CD8⁺ effector memory T cells demonstrated increased ISG and activation scores but similar exhaustion and cytotoxicity scores. By contrast, SLE NK cells were noted to maintain a prominent cytotoxicity signature after UV. T_regs in UV-treated SLE skin exhibited the highest ISG signatures among the T cell compartment (Fig. 5E). SLE UV T_regs also exhibited higher cytotoxicity (Fig. 5G) and exhaustion (Fig. 5H) signatures compared with HC UV T_regs, suggesting dysfunction of post-UV cutaneous regulatory mechanisms in SLE. Type I IFNs have been reported to contribute to T_reg dysfunction in lupus (18, 50) and may specifically promote T_reg exhaustion (51). Thus, these data suggest that shifts in cytotoxicity after UVB are not an apparent feature in SLE skin but rather SLE UVB responses are dominated by IFN and exhaustion signatures in T_regs.

Fig. 5. SLE T_reg cells exhibit higher ISG and exhaustion signatures than HC T_reg cells following UVB exposure.

(A and B) UMAPs depicting T cells by (A) disease and treatment and (B) T cell subsets. (C) Barplot of proportions of each T cell subset by disease and treatment. (D) Marker genes for each T cell subset (E to H) Gene signatures for each T cell subset across disease and treatment conditions for (E) interferon-stimulated gene (ISG) signature, (F) activation signature, (G) cytotoxicity signature, and (H) exhaustion signature. Data [(E) to (H)] were analyzed by Welch’s t test with comparisons shown between SLE UV and HC UV. *P ≤ 0.05, ***P ≤ 0.001, and ****P ≤ 0.0001. ns, not significant. n=8 SLE and 7 HC for all analyses.

UV-recruited moDCs are more inflammatory in SLE skin

Our crosstalk data (Fig. 3) identified increased communication in the SLE epidermis and dermis with moDCs. Myeloid lineage cells have recently been implicated in propagation of cutaneous inflammation in lupus (9, 13, 14, 52–54). Thus, we next examined the myeloid cell compartment in our dataset. Subclustering revealed the presence of eight myeloid populations including LCs, cDC1s, cDC2s, moDCs, mature regulatory DCs (mregDCs), perivascular macrophages (PVMs), lipid-associated macrophages (LAM), and a proliferating population that could not be further classified (Fig. 6, A to C, and fig. S7, A and B). Consistent with prior reports, LCs were relatively depleted in UVB-treated SLE skin compared to HC (55) (Fig. 6C). Plasmacytoid dendritic cells (pDCs) were not identified in our myeloid clusters. Given reports of UV-induced pDC recruitment to SLE skin (56), we conducted immunohistochemistry for the pDC associated marker BDCA-2. HC and SLE skin exhibited an increase in dermal staining for the pDC marker BDCA-2 after UV treatment (fig. S7C), suggesting that cells expressing this pDC marker were present 24 hours after UV exposure but were not captured by scRNA-seq methodologies. LAMs were almost exclusively found in SLE skin, whereas healthy control skin had so few that analysis to compare LAMs across disease state was not possible, which may suggest the importance of this population in propagation of cutaneous inflammation in SLE skin (Fig. 6C). moDCs were annotated as such given their expression of markers like THBS1 and EREG which have been identified in myeloid DC populations in systemic sclerosis skin (57) and in inflammatory-monocyte derived dendritic cells. (58, 59, 60).

Fig. 6. moDCs are recruited to SLE and HC UV-exposed skin.

(A and B) UMAPs of skin myeloid cells by (A) disease and treatment conditions and (B) cell subtype. n=8 SLE and 7 HC. (C) Barplot of proportions of each myeloid subset by disease and treatment. (D) Representative immunofluorescence microscopy of HC and SLE (n=3 per group) UV-treated skin depicting CD14⁺ (red) EREG⁺ (green) moDCs. Dotted line in indicates 100 μm from DEJ; white arrows indicate colocalized cells. Scale bars, 50 μm. (E) Percent of total CD14⁺ EREG⁺ cells within 100 μm of DEJ in SLE and HC. Quantification performed using QuPath software. Each point represents a unique sample. (F) Nearest neighbor analysis of Xenium spatial profiling obtained on SLE nonlesional skin before and after UVB exposure. Analysis examines likelihood of moDC proximity to denoted KC subtypes for 3 SLE patients. (G) SLE nonlesional violin plots of SLE and HC UV moDC keratinocyte-specific gene scores. (H) Percent of UV moDC cells with positive and negative KC gene scores in SLE (blue) and HC (yellow). (I) Donut plot of top ten IPA predicted upstream regulators filtered for |z| ≥ 2 and soluble mediators (cytokines and chemokines). Proportion of each mediator made by each cellular compartment is denoted by the donut plot for each cytokine. Data (G) were analyzed by Welch’s one-sided t test. *P ≤ 0.05.

Of the myeloid cells detected, only moDCs increased in proportion in the skin following UV exposure at the 24-hour timepoint (Fig. 6C).Both HC and SLE skin exhibited recruitment of this cell population. We next asked where these cells were localized in SLE and HC UV-treated skin. Immunofluorescence microscopy was performed for CD14 as well as the moDC marker EREG (Fig. 6D and fig. S7A). Though the total number of dermal cells and CD14⁺ EREG⁺ moDCs was similar between the two groups (fig. S7, D and E), quantitation of the localization of moDCs in UV-treated HC and SLE skin demonstrated that a greater percentage of moDCs localized to the dermoepidermal junction (DEJ) in SLE versus HC (Fig. 6E). Using Xenium to spatially profile expression of 5100 genes of four UV-treated and four untreated SLE biopsies, we conducted nearest neighbor analysis to evaluate which KCs were in close proximity to the moDC population. For each moDC in the sample, the 10 nearest neighbors were recorded and the percentage of each cell population near the moDC was ploted. This revealed that UV treatment of SLE skin resulted in closer association of moDCs with basal inflammatory KCs versus other KC populations in untreated skin (Fig. 6F). This positioning may promote miseducation of moDCs by cytokines secreted by basal inflammatory KCs. Additionally, this suggested that moDCs may be spatially positioned to phagocytose apoptotic KCs, which may provide a source of autoantigens. Thus, we asked whether there was evidence of KC phagocytosis in SLE versus HC moDCs. We assessed this by generating a KC module score made up of KC-specific genes including KRT10 and KRT14 as a read-out of potential apoptotic KC uptake by moDCs (table S6). The presence of RNA from KC-specific genes was found in both SLE and HC UV-induced moDCs in our scRNA-seq dataset, suggesting they may be engulfing apoptotic KCs (Fig. 6G). In line with this hypothesis, a greater percentage of SLE moDCs had a positive KC score compared with their HC counterparts following UV exposure (Fig. 6H). Given that SLE KCs undergo increased apoptosis after UV exposure (8, 61), this may reflect more substrate to engulf or increased phagocytic capacity in SLE moDCs, as suggested in murine models of SLE (14).

We next examined whether there were differences in the downstream inflammatory responses present in SLE versus HC moDCs. Pathway analysis revealed hyperinflammatory responses in UVB-treated SLE moDCs particularly related to inflammatory cytokines and type I IFNs, identifying a more inflammatory phenotype in SLE moDCs (fig. S7F and table S7). By contrast, upregulated pathways in HC moDCs following UV treatment included those related to homeostatic functions, as well as FXR/RXR activation, which attenuates myeloid cell production of pro-inflammatory cytokines including type I IFNs (62, 63). These data suggest that moDCs are recruited to UVB-treated skin in both HC and SLE subjects, but SLE moDCs exhibit enhanced trafficking to the DEJ, potential uptake of apoptotic KCs, and a pro-inflammatory phenotype. By contrast, HC moDCs mount homeostatic and anti-inflammatory responses after UV treatment.

Finally, we questioned what signals upstream of these pathways were driving the inflammatory responses in SLE UV moDCs and which cellular compartments were contributing to the production of those signals. Thus, we filtered predicted URs for SLE UV moDCs for soluble mediators like cytokines, identifying that SLE UV moDCs are skewed by a host of inflammatory cytokines including IL-1β and TNF (table S8). Then, for each predicted cytokine UR, we calculated the percentage of the total transcript made by each cellular compartment in the skin (Fig. 6I). Thus, KCs, myeloid cells, T cells, and FBs as the primary cellular source of these inflammatory cytokine URs, highlighting the importance of epidermal signals but also the collective skin system in driving aberrant inflammatory responses in immune cells recruited following UVB.

UV-recruited moDCs are derived from distinct populations of circulating monocytes in SLE and HC

We then asked whether UV-recruited moDC cells were derived from the same circulating pool in HC and SLE subjects. Myeloid cells from skin samples as well as paired peripheral blood myeloid cells isolated from the same HC and patients with SLE were clustered together via UMAP (Fig. 7, A and B). Using pseudotime trajectory analysis, we then explored the relationship between circulating classical and non-classical monocytes and skin moDCs. The combined trajectory using SLE and HC cells showed two branches with one dominated by classical monocytes and a second that showed a close placement of moDCs and non-classical monocytes (fig. S8). However, when the trajectory was separated by disease state, a relationship between classical monocytes and moDCs was mapped in the HC group (Fig. 7, C and D), whereas in SLE, non-classical monocytes and moDCs clustered together at the end of the pseudotime trajectory (Fig. 7, E and F), suggesting the origin of moDCs may differ between HC and SLE. Indeed, finer analysis of the top 80 genes across the HC and SLE pseudotime trajectories revealed upregulation of distinct genes across the differentiation trajectory from blood monocyte to skin moDC. As HC monocytes entered the skin, there was primarily upregulation of ribosomal protein genes, genes known to limit autoimmunity like PTX3 (64, 65), and genes that are important for promoting tolerogenic phenotypes in DCs following efferocytosis like PTGS2 (66) (Fig. 7G). By contrast, as SLE monocytes entered UV-treated lupus skin and differentiated into moDCs, they upregulated genes related to antigen presentation and processing like CTSD, CTSS, HLA-A, and HLA-DRA (Fig. 7H). Genes upregulated in the final differentiation state in the SLE trajectory were primarily HLA-related genes and genes related to immune recruitment and activation like CXCL3, CXCL8, IL1B, CCL3, and CCL4, reflecting the pro-inflammatory nature of SLE moDCs in contrast to HC moDCs. Thus, cutaneous moDCs in HC are derived from classical monocytes and enact a homeostatic, regulatory program following UV exposure, whereas SLE moDCs are derived from nonclassical monocytes and express inflammatory and antigen presentation signatures.

Fig. 7. SLE and HC UV moDCs are recruited from different populations of circulating monocytes.

(A and B) UMAPs of SLE (n=8) and HC (n=5) PBMC monocytes and skin myeloid cells (n=8 SLE and 7 HC) by (A) tissue source and (B) cell subsets. (C to E) HC monocyte to moDC pseudotime trajectories by (C) pseudotime, (D) cell subset, (E) HC pseudotime heatmap showing top 80 genes expressed across the pseudotime. (F to H) SLE monocyte to moDC pseudotime trajectories by (F) pseudotime and (G) cell subsets and (H) SLE pseudotime heatmap showing top 80 genes expressed across the pseudotime.

UV-recruited moDCs represent a potential precursor to lesional CD16⁺ dendritic cells in lupus skin

Previous work from our research group identified a subpopulation of myeloid-derived DCs, CD16⁺ DCs, as potentially important in lesional inflammation in lupus skin (9). As our published data support a potential derivation of SLE skin moDCs from circulating nonclassical monocytes, we sought to understand the relationship between UV-recruited moDCs and lupus skin-associated CD16⁺ DCs. We thus clustered myeloid cells derived from non-lesional and lesional scRNA seq data from patients with SLE with active CLE lesions (and one CLE-only patient) from our previous work (9) and combined them with myeloid populations derived from UV-treated SLE skin in this study. The overlay of our myeloid clusters was consistent between both studies (Fig. 8, A and B). We noted overlap between moDCs and CD16⁺ DCs populations from UV-treated skin (Fig. 8, A and B), suggesting a potential relationship between UV moDCs and lesional-derived CD16⁺ DCs (Fig. 8B). Consistent with the idea that UV moDCs are transcriptionally related to lesional CD16⁺ DCs, marker gene analysis revealed shared marker gene expression between lesional CD16⁺ DCs and moDCs (Fig. 8C). To further probe the relationship between these populations, we performed pseudotime trajectory analysis which provided further support of UV moDCs as precursors of CD16⁺ DCs (Fig. 8, D to F). Assessment of the pseudotime trajectory by cell source indicated that both lesional and nonlesional CD16⁺ DCs clustered at the end of the trajectory; however, only lesional CD16⁺ DCs populated the center of the trajectory alongside UV moDCs, potentially suggesting a population of actively recruited moDCs to lesional skin which progressively mature over time as they are exposed to the hyperinflammatory lesional skin environment (Fig. 8F). Nonlesional skin CD16⁺ DC may mature differently given there is less overall inflammation in non-lesional skin. The main DEGs across the pseudotime suggest that CD16⁺ DCs are differentiated by additional upregulation of antigen presentation machinery and CD81, a tetraspanin important for cell migration (67–69), and downregulation of genes such as EREG and CCL20 (Fig. 8G). Thus, UV-mediated moDC recruitment in SLE skin could be a potential source for CD16⁺ DCs.

Fig. 8. UV-recruited moDCs represent a likely precursor to lesional CD16⁺ dendritic cells.

(A and B) Myeloid cells from UV-treated SLE skin (n=8) were clustered with lesional CLE and nonlesional myeloid cells from seven SLE paired skin samples. (A) UMAP showing myeloid cells by subtype. (B) UMAP of myeloid cells by source. Red circles show location of CD16⁺ and UV induced mo-DCs (C) Dotplot of marker genes for UV moDCs (UV), lesional CD16⁺ DCs (L), and Nonlesional CD16⁺ DCs (N). (D to F) Pseudotime trajectories from UV moDCs to CD16⁺ DCs by (D) pseudotime, (E) cell type, and (F) disease state. (G) Pseudotime heatmap showing top 80 expressed genes across the pseudotime, (H) UV moDC verus lesional CD16⁺ DC upstream regulators.

Finally, as we are ultimately interested in understanding the links between UV-induced skin injury and lesion development, we assessed the mediators that may predict transition from moDC phenotype to lesional CD16⁺ DC phenotype following UV (Fig. 8H). ETV3 and ETV6, transcription factors that drive monocyte differentiation into moDCs instead of macrophages (70), were the top URs in the UV moDC phenotype, likely reflecting the recency with which these cells had differentiated at the 24 hour timepoint at which they were harvested. URs of lesional CD16⁺ DCs were primarily inflammatory cytokines, including types I, II, and III IFNs and IL-27, and factors like NONO, which has been described to promote activation of the cGAS-STING pathway (71). Thus, these data ultimately suggest that the moDC to CD16⁺ DC trajectory may be initiated by UV light exposure and is highly influenced by the inflammatory context of SLE skin. Thus, UV-induced moDCs could serve as a contributor to photosensitive inflammatory responses and lesion development (fig. S9) and could potentially serve as an important target for prevention of photosensitive flares of CLE.

DISCUSSION

Using scRNA-seq, we have elucidated the drivers of UV-mediated inflammation in healthy and lupus skin at the single-cell level. In doing so, we have identified substantial differences in the SLE UV response that are driven by hyperinflammatory and skewed epidermal responses leading to inflammatory moDC recruitment to the DEJ in SLE skin.

Following UV exposure, the major cellular actors in the cutaneous response to UVB in SLE and HC are largely the same; however, the difference in cutaneous inflammatory microenvironment between SLE and HC drives differential responses, likely secondary to the overproduction of type I IFNs and other inflammatory cytokines in SLE skin. The genesis of this dysregulated UVB response appears to begin with dysregulation at the level of the epidermis. In the HC epidermis, basal inflammatory KCs are expanded by UVB but relatively inert in cell-cell crosstalk. By contrast, SLE basal inflammatory KCs drive downstream inflammatory responses through robust cell-cell crosstalk following UVB treatment. Ultimately, this reinforces the importance of the tissue context in driving immune responses in SLE. Indeed, recruitment of moDC to UVB-treated skin is a shared feature of the cutaneous UVB response in SLE and HC; however, SLE moDCs traffic closer to the DEJ, where more of these cells appear to uptake KC antigens. Although uptake of apoptotic material is typically a tolerogenic event, SLE UV moDCs paradoxically exhibit upregulation of inflammatory pathways that lead to activation and upregulation of antigen presentation pathways. This may facilitate downstream adaptive immune cell activation, which is also influenced by the inflammatory milieu in SLE skin. Indeed, HC cell populations seem to be upregulating anti-inflammatory signaling mechanisms after UV exposure, and this is repressed in SLE. Further highlighting the importance of tissue in driving inflammatory responses, this inflammatory milieu is amplified in lesional skin such that CD16⁺ DCs exhibit the highest inflammatory signal. This suggests that moDCs recruited to UVB-injured SLE skin are educated by the chronic inflammatory, type I IFN-rich cutaneous milieu that can lead to unchecked immune activation.

The mechanisms underlying a primed hyperinflammatory immune response in SLE skin are under active investigation but remain incompletely understood. SLE keratinocytes exhibit chronic type I IFN signaling which drives increased apoptosis and heightened cytokine response to UV exposure (8, 72). This persistent type I IFN production may trigger further inflammatory responses through upregulation of Z-DNA binding protein 1 (ZBP1) and the capture of cytosolic oxidized mitochondrial DNA in the Z-conformation (73). In addition, methylation changes of SLE KCs may skew Hippo signaling pathways, increasing propensity for apoptosis following UV light (10). Metabolic dysfunction and type I IFN responses may also skew T_reg function in response to UV exposure (18, 74). Indeed, rapamycin may restore T_regs in SLE (75), and targeted treatment of T_regs may improve cutaneous lupus lesions (76). Previous studies of UV-provoked human lupus skin demonstrated increased expression of antigen presentation genes in lupus skin compared with control (77) and identified monocytes as early infiltrators following UV exposure (78). More recent data demonstrate the presence of a myeloid population which is characterized by CD16 expression and high IFN signatures and is present in both nonlesional and lesional lupus skin (9). Together, these findings suggest the importance of skin-activated myeloid DCs in initiation and propagation of lesional inflammation.

Our study is not without limitations. First, our data are limited by the capture of a single time point following UVB irradiation. Despite this limitation, we were able to detect notable differences in the inflammatory education of numerous cell populations after UVB exposure. Secondly, the methodology of scRNA-seq did not allow for capture of neutrophils, which are important responders to UV exposure, or pDCs, which may have some SLE-specific pathogenic roles. Additional methods such as capture of cells in fixed tissue will be helpful in the future to address the role of these cell populations in SLE skin. Third, because of the low expression level of type I IFN genes, we are unable to use our scRNA-seq data to answer the question of which cells are critical producers of type I IFN in the skin after UVB treatment. Based on previous data, the etiology of cutaneous type I IFNs likely includes epidermal and myeloid sources (8, 11, 54). Critically, however, this structural cell production of IFNs and inflammatory cytokines is important for myeloid cell activation after UV exposure and, these limitations notwithstanding, substantial and informative differences were identified.

In summary, we posit that altered UV responses in patients with SLE-related photosensitivity lead to inflammatory epidermal and stromal responses that skew the transcriptional program from homeostatic to one characterized by strong IFN responses, inflammatory cytokine production, and robust recruitment and activation of moDCs in the skin, identifying important targets for prevention and treatment of skin and systemic flares in lupus.

MATERIALS AND METHODS

Study Design

Our objectives in this study were to understand the differential effects of ultraviolet B (UVB) exposure on nonlesional lupus and healthy control skin, to elucidate the circulating precursors to inflammatory cells infiltrating irradiated skin, and to investigate the relationship between UV-irradiated and lesional lupus skin. Seven healthy control (defined as no history of autoimmune disease or skin inflammation) and eight lupus patients (Table 1) underwent informed, written consent according to study procedures approved by the University of Michigan IRBMED HUM00171985. All patients were treated according to the declaration of Helsinki. The patients with SLE had no or mild active skin disease at the time of study. Following signing of written informed consent, subjects were exposed to varying doses of UVB light via a Kodacel lamp ranging from 10 mJ/cm² to 80 mJ/cm². At 24 hours after UV exposure, 6 mm punch skin biopsies were taken from UV-unexposed skin as well as the portion of skin corresponding with the minimal erythema dose (MED). Patients who did not achieve MED were biopsied at the 80 mJ/cm² site. All samples from all patients enrolled were included in the analysis. No outliers were excluded. Exclusion criteria for the study were active pregnancy, history of bleeding disorder, use of NSAIDs in the past 72 hours, Cutaneous Lupus Erythematosus Disease Area and Severity Index (CLASI) > 10 or Systemic Lupus Erythematosus Disease Activity Index (SLEDAI) > 8 (reflecting high disease activity that could be exacerbated by UV exposure), ongoing use of prednisone > 10 mg daily, active Cytoxan treatment or exposure within the past 6 weeks, absence of non-lesional skin on buttocks, or sunburn in the past month. Peripheral blood mononuclear cells (PBMC) samples for single-cell RNA sequencing (scRNA-seq) were drawn the day of skin biopsy (24 hours after UV treatment).

Library preparation, sequencing, and alignment for single-cell RNA sequencing

Skin biopsies were transported to the laboratory on ice in Hanks Balanced Salt Solution (HBSS) buffer (14025092, Gibco) and processed immediately. Biopsies were incubated in 0.4% dispase (D4693–1G, Sigma) in Hank’s Balanced Saline Solution (HBSS) (Gibco) overnight at 4°C to separate epidermis from dermis. Digestion of the epidermis was performed in 0.25% trypsin-EDTA (25200056, Gibco) with 10U/ml of DNase I (4536282001, Sigma) for 1 hour at 37°C. Simultaneously, dermal samples were dissociated mechanically and then digested in 0.2% Collagenase II (C6885–500MG, Sigma) and 0.2% collagenase V (C9263–25MG, Sigma) with 10U/mL DNaseI in plain medium for 1.5 hours at 37°C. Following their respective incubation times, epidermal and dermal samples were strained through a 40-μm mesh filter and centrifuged. Cells were subsequently mixed at a 1:1 ratio before being submitted for scRNA-seq. The 10x Chromium system with chemistry v3 was used by the University of Michigan Advanced Genomics Core to generate libraries, which were subsequently sequenced on the Illumina NovaSeq 6000 platform, generating 150-base pair paired-end reads. The 10x Cellranger software was used to process data including gene quantification, quality control, and read alignment (hg38). Please see the Supplementary Materials for additional methods on cell clustering and annotation.

Module score analysis

Module scores for all skin cells (IFN score), ECs (IFN-α score, IL-1β score, and TNF-α score), T cells (ISG score, activation score, cytotoxicity score, and exhaustion score), and UV moDCs (Keratinocyte score) were calculated using the AddModuleScore function on previously published gene signatures. For the skin score, the gene signatures are available within table S1 in (20); for ECs, the gene signatures are available within data file S2 in (9); for T cells, the ISG, activation, cytotoxicity, and exhaustion signatures are available within table S3 in (49). The UV moDC keratinocyte scores were calculated using 19 signature genes we curated, which are available in table S6.

Ligand-receptor interaction analysis

Ligand-Receptor analysis was performed using CellPhoneDB (v3.0.0). Please see the Supplementary Materials for details.

EC cytokine signatures

We used RNA-seq–constructed cytokine response signatures for the following cytokines: IFN-α (5 ng/ml), TNF (10 ng/ml), and IL-1β (10 ng/ml) as previously described by our group (9, 79, 80). To summarize, KCs from 50 distinct donors were treated with the above cytokines and after 8 hours in culture, cells were harvested, and RNA was isolated. Simultaneously, unstimulated KCs were cultured and harvested for RNA. Bulk RNA-seq was performed using the Illumina NovaSeq 6000 with assistance from the University of Michigan Advanced Genomics Core. For each condition, DEG analysis was performed between the stimulated cells versus unstimulated controls using DESeq2 (81). Response signatures were constructed using DEGs (twofold increase; false discovery rate < 0.05) as published (9).

Pseudotime trajectory construction

Pseudotime trajectories were constructed using the R package “monocle” (v2.28.0) (82). In the first analysis, trajectories were constructed using skin moDCs, together with cMono and ncMono from PBMC. Trajectories in HC and SLE were plotted. In the second analysis, trajectories were constructed using UV moDCs within SLE together with CD16⁺ DCs within the 7 SLE (including both lesional and nonlesional) patients from previously published work from our laboratory (9).

In both cases, the gene expressions of cells in raw counts were taken from the Seuratobjects. Following the protocol of using monocle2, estimateSizeFactors and estimateDispersions were ran, using the default parameters, to handle normalization across cells and variance across genes, respectively. Only genes present in greater than 10 cells were retained for further analyses. The differentialGeneTest function was used to determine variable genes using a model against Seurat subcluster identities. Genes with an adjusted P value (or q value) less than 0.01 were used for trajectory inference, which is achieved by setting ordering_genes within the set OrderingFilter function, and the reduceDimension function was used to construct the trajectory with default reduction method of DDRTree. The starting points (roots) of the trajectories were manually chosen or adjusted based on biological interpretation, not automatically. The expression changes of the top 80 genes across pseudotime were plotted by the plot_pseudotime_heatmap function.

Spatial Profiling

Data from NCT05048238 was obtained for further analysis. Here four patients with SLE and a history of cutaneous lupus underwent UV phototesting as reported in this study and Xenium 5000+100 custom gene panel spatial profiling was conducted. Using the cell type annotated scRNA-seq data as reference, we projected the cells from the Xenium data to the reference space (83), and nearest neighbor was used for cell type inference. Spatial network was constructed using the createSpatialDelaunayNetwork function from the R Giotto package with default settings. Numbers of neighboring cell types in the spatial network were counted for each cell type and condition. Counts were the normalized by dividing by the number of corresponding cells in the condition to remove the cell number bias.

Immunohistochemistry and Immunofluorescence

Paraffin-embedded tissue sections from punch biopsies taken at 1 dose higher than the MED of patients recruited to the study (Table 1) were heated for 30 min at 60°C, deparaffinized, and rehydrated. Slides were then placed in antigen retrieval buffer at the appropriate pH (9 for IHC, 6 for IF) and heated to 90°C for 20 min. Once cooled, slides were either stained for immunohistochemistry (IHC) or immunofluorescence (IF) microscopy. For IHC, slides were treated with 3% H₂O₂ for 5 min and subsequently blocked with 10% goat serum (005000121, Jackson Immuno Research Labs) for 30 min. Slides were stained with rabbit anti-human BDCA2 (1:20, Sigma, HPA029432) overnight at 4°C. The following day, slides were washed, treated with anti-rabbit HRP-conjugated secondary antibodies (SC2357 Santa Cruz Biotechnology), peroxidase, and diaminobenzidine substrate and subsequently imaged. For IF, slides were washed, blocked with 10% goat serum, and stained with mouse anti-human CD14 (1:200, Proteintech, 501736350) and rabbit anti-human epiregulin (1:200, Invitrogen, PA5–24727) at 4°C overnight. The following day, slides were washed, stained with goat anti-mouse conjugated to Alexa Fluor Plus 488 (1:500, Invitrogen, A-11001), and goat anti-rabbit conjugated Alexa Fluor Plus 647 (1:500, Invitrogen, A-21244), mounted with 4′,6-diamidino-2-phenylindole (DAPI) (P36931, Invitrogen), and imaged. Quantification was performed using QuPath (v0.5.1).

For IF staining of Keratin10, Keratin14, and IFI27, FFPE sections were heated at 60°C for 1 hour, deparaffinized, rehydrated, and antigen retrieval was performed using a pressure cooker and antigen retrieval buffer of pH 6 (550524, BD Biosciences). Slides were washed and blocked with 10% normal goat serum (005000121, Jackson Immuno Research Labs). Next, slides were incubated overnight at 4°C with primary antibodies against keratin 14 (1:500, chicken polyclonal, BioLegend, 906004), IFI27 (1:200, mouse monoclonal, Novus Biologicals, NBP3–26876), and keratin 10 (1:200, rabbit polyclonal, Proteintech, 18343–1-AP). All slides were then washed three times and incubated for 30 min with secondary the antibodies goat anti-chicken conjugated to Alexa Fluor Plus 555 (1:500, Invitrogen, A-21437), goat anti-mouse conjugated to Alexa Fluor Plus 488 (1:500, Invitrogen, A-11001), and goat anti-rabbit conjugated Alexa Fluor Plus 647 (1:500, Invitrogen, A-21244). Nuclei were counterstained with DAPI (1:200). After washing, slides were washed and were mounted using ProLong Glass Antifade Mountant (Invitrogen, Cat. No. P36980). Images were taken the next day with 40x magnification using a Nikon Yokogawa Spinning disk microscope. For each sample, three images were taken based on DAPI staining to identify skin structures and to reduce operator bias. Quantification of immunofluorescence signals was performed using Fiji (ImageJ 1.54f). Quantification of epidermal mean fluorescence intensity of IFN27 was done in three independent regions of interest per biological sample based on K14 and DAPI-staining.

Keratinocyte culture and MTORC testing

Primary human keratinocytes from healthy controls (HC) (n = 4) and patients with SLE (n = 6) were seeded at 30,000 cells per well in 12-well plates and cultured in EpiLife medium (MEPI500CA, Gibco) supplemented with human keratinocyte growth supplement (HKGS) (S0015, Gibco) until reaching approximately 70% confluency. Cells were then pretreated with 100 nM rapamycin (S1039, Selleck Chemicals) for either 1 hour (to inhibit mTORC1) or 24 hours (to inhibit mTORC2), with parallel control groups receiving dimethyl sulfoxide (DMSO) for the same durations. Untreated control and UVB-only groups were also included. Following pretreatment, cells were exposed to 50 mJ/cm² UVB in PBS, washed, and incubated in fresh medium for an additional 4 hours. IFNB gene expression was analyzed by RT-qPCR using SYBR Green (A25742, Applied Biosystems), with assays performed with the assistance of the University of Michigan Advanced Genomics Core, using the following primers: IFNB Forward: 5′-ACGCCGCATTGACCATCTAT-3′; IFNB Reverse: 5′-GTCTCATTACAGCCAGTGCT-3′; RPLP0 Forward: 5′-GGCAGCATCTACAACCCTGA-3′; RPLP0 Reverse: 5′-CTGCAGACAGACACTGGCAA-3′.

Statistical analysis

The FindMarkers function from the Seurat package was used to analyze differential gene expression between disease and treatment conditions using |Log₂FC| ≥ 0.25. Welch’s t-test was performed to compare the gene expressions and module scores between two conditions for the scRNA-seq data. Predicted upstream regulators (URs) were identified using Ingenuity Pathway Analysis (IPA) v.2024. In all analyses, URs were filtered for |z| ≥ 2. Unless otherwise specified, graphed molecules were filtered for the following: cytokine, enzyme, fusion gene/product, G-protein coupled receptor, group, growth factor, ion channel, kinase, ligand-dependent nuclear receptor, mature microRNA, microRNA, other, peptidase, phosphatase, transcription regulator, translation regulator, transmembrane receptor, transporter. For in vitro assays, Welch’s t-test or one way ANOVA were used for comparisons as indicated in the figure legends with P < 0.05 considered statistically significant. Raw data for these experiments are provided in data file S1.

Supplementary Material

Supplementary Materials and Methods, includes references 86–90

Figs. S1 to S9

Tables S1 to S9

References (##–##)

Data files S1

MDAR Reproducibility checklist

Data, code, and materials availability:

All data associated with this study are in the paper or supplementary materials. scRNA-data from the main analysis has been deposited with GEO under accession number 289389. Any patient-level metadata beyond what are supplied in GEO will require a data transfer agreement for release. No new code was written for analysis of data in this paper. N/TERT keratinocytes require an MTA with the originator, Dr. J. Rheinwald (Brigham and Women’s Hospital). All other materials used or generated in this study are commercially available.