IgG autoantibodies in bullous pemphigoid induce a pathogenic MyD88-dependent pro-inflammatory response in keratinocytes

Abstract

Autoantibodies in bullous pemphigoid (BP) are known to activate the innate immune response. Nevertheless, the direct effect of autoantibodies on keratinocytes and the contribution of keratinocyte responses to the pathology of BP are largely unknown. Here, by performing multiplex immunoassays and RNA-seq on primary keratinocytes treated with IgG derived from BP patients, we identify a MyD88-dependent pro-inflammatory and proteolytic response characterized by the release of several cytokines (IL-6, IL-24, TGF-β1), chemokines (CXCL16, MIP-3β, RANTES), C1s, DPP4, and MMP-9. The activation of this MyD88-dependent response is further validated using spatial transcriptomics and scRNA-seq of diseased skin. Blistering of the skin appears to significantly impact this inflammatory response, with attached BP skin and spongiotic dermatitis revealing indistinguishable transcriptomes. In a preclinical mouse model of BP, Krt14-specific Myd88 knockout significantly decreases disease severity and reduces serum levels of IL-4 and IL-9, indicating a contributory role of keratinocyte-derived skin inflammation in the systemic response. Thus, our work highlights key contributions of keratinocytes in response to autoantibodies in BP.

The role of autoantibodies in bullous pemphigoid (BP) and their impact on keratinocytes and the response to BP pathology remains underexplored. By leveraging transcriptomics analysis and large-scale protein assays, here the authors identify keratinocyte MyD88 as a regulator of the pro-inflammatory response in BP, uncovering the role of keratinocytes in this disease pathology.

Introduction

Bullous pemphigoid (BP) is an autoimmune blistering disease caused by autoantibodies targeting the hemidesmosomal proteins BP180 (Type 17 collagen, COL17) and, to a lesser extent, BP230^1,2. Autoantibodies can be internalized in a BP-IgG/BP180 complex via macropinocytosis^3,4, as well as interact with Fc receptors (FcR) and complement^5–7. Clinically, BP presents with numerous morphologies including urticarial plaques, eczematous patches, and/or bullae, with accompanying pruritus1. Histologically, BP most commonly presents with abundant tissue eosinophilia, particularly with eosinophils lining the dermo-epidermal junction in early (urticarial) phase and/or entering the inflamed epidermis (eosinophilic spongiosis)^8,9. Both complement-fixing and non-fixing IgG isotypes are seen amongst autoantibodies^5,6,10,11. Aside from IgG autoantibodies, IgE autoantibodies are detected in up to 70% of patients, driving basophil degranulation^5,8,12–14. Both IgG and IgE from patients with BP have been shown to cause the release of IL-6 and IL-8 directly from cultured keratinocytes, indicating an FcR and complement-independent pathologic mechanism^15–17. This suggests that autoantibody-triggered manipulation of BP180 on keratinocytes may regulate inflammatory pathways and directly initiate and/or amplify the local inflammatory response in BP. IL-6 and IL-8 upregulation alone would, however, not sufficiently explain the lesional Th2 polarization^18,19 and eosinophil influx observed in BP. Thus, we hypothesize that IgG from BP patients activates alternative pathologic pathways. For example, we have recently discovered direct inflammatory effects and pauci-inflammatory blistering mechanisms of autoantibodies in anti-laminin 332 pemphigoid^20,21.

Despite the known roles of BP180 in the basement membrane zone (BMZ), as an autoantigen in BP, and as a prognostic marker in certain cancers²², its role in the regulation of skin inflammation and BMZ homeostasis is not well described BP²³. Several animal mutant models of Col17a1^24,25, as well as case reports of epidermolysis bullosa with COL17A1 mutations^26–28 strongly suggest a direct regulatory role of COL17 on tissue granulocyte infiltration. COL17 additionally undergoes physiological ectodomain shedding²⁹, which has been associated with carcinogenesis³⁰, but has an unclear role in skin inflammation^{22,23,31–34}. Notably, spontaneous inflammation occurs in two animal models with a mutation placed in the humanized COL17 NC16a domain²⁴, or the corresponding mouse NC14a (ΔNC14A) domain²⁵. In the first model, mice developed spontaneous scratching behavior even when crossed with Rag2 ^−/− mice²⁴, indicating a lack of adaptive immune involvement. Cultured keratinocytes from these ΔNC14A mice demonstrated elevated TSLP levels relative to controls. The study also noted a significant increase in TSLP levels in BP skin relative to normal, implicating TSLP as an epidermal contributor towards pathogenesis. We, however, noted epithelial TSLP elevation to be a non-specific finding of multiple inflammatory skin diseases³⁵. In the second disease model, introducing a mutation in NC14a (ΔNC14A)²⁵ also resulted in a BP-like phenotype with dermal eosinophil infiltrations, peripheral eosinophilia, elevated serum IgE, and pruritus²⁵. Oddly, despite the removal of the major BP epitope region, some ΔNC14A mice developed IgG and IgA autoantibodies with subepidermal reactivity. These IgG autoantibodies recognized a 180-kDa keratinocyte protein that was sensitive to collagenase digestion, consistent with COL17. Thus, it is unclear to what extent the symptomatology is due to direct COL17 mutations or the generation of anti-skin antibodies in this model.

In light of these studies, we hypothesize that disturbance of BP180 homeostasis by autoantibodies in BP drives an inflammatory response from keratinocytes, extending beyond IL-6 and IL-8. In this study, we demonstrate that IgG from BP patients induces an inflammatory keratinocyte response, with much of this inflammatory response depending on the Myeloid Differentiation primary response 88 (MyD88) protein. Further, we demonstrate the in vivo relevance of this response using a murine model of BP, revealing that keratinocyte-specific knockout of MyD88 is sufficient to mitigate disease severity and reduce circulating type 2 inflammatory cytokines such as IL-4 and IL-9. Collectively, these findings demonstrate that autoantibodies in BP induce a pathologically relevant inflammatory response, with divergence in bullous and non-bullous lesions.

Results

Pro-inflammatory Cytokines and Chemokines are Released from BP-IgGtreated primary human keratinocytes

To investigate the direct effect of BP-IgG on keratinocytes, we cultured primary human keratinocytes (PHK) overnight with affinity-purified single donor BP-IgG or control-IgG. The demographics and serotypes of patients are summarized in (Supplementary Table 1). We first analyzed protein levels of 118 cytokines, chemokines, and proteases from keratinocyte supernatants from two independent experiments (Fig. 1A). Analytes falling below the linear detection range were excluded from further analysis (Supplementary Dataset 1). Setting a threshold with correction for false discovery rates (FDR) of P_adj < 0.05, revealed upregulation of several cytokines (IL-6, IL-24, TGF-β1), chemokines (CXCL16, CTACK, MIP-3β, RANTES), complement components (C1s), DPP4, and proteases (MMP-9) (Fig. 1B). To better understand co-expression of protein components, we next generated a correlation matrix which highlighted a strong correlation between CTACK, IL-8, IL-6, I-TAC, IL-1α, EN-78 and MIP-3β expression (Fig. 1C). Subanalysis of PHK treated with IgG from patients with only anti-BP180 or anti-BP230 IgG, but not both, revealed comparable findings, other than reduced G-CSF, IL-8, MCP-1, and TNF in anti-BP230 IgG treated keratinocytes relative to anti-BP180 treated keratinocytes (Fig. 1D). Given the elevated MMP-9 production and known role of anti-BP180 IgG in inducing BP180 internalization, we sought to confirm the ability of BP-IgG to induce blistering in the absence of immune cells using human 3D skin equivalents. This revealed that BP-IgG alone was sufficient to induce blistering in the absence of immune cells (Fig. 1E, F). Thus, our findings suggest that IgG from BP patients induces the release of numerous proinflammatory and proteolytic molecules from keratinocytes.

Fig. 1. BP-IgG induces expression of numerous pro-inflammatory proteins relative to control-IgG treated primary human keratinocytes.

A Heat map demonstrating supernatant protein expression of BP-IgG vs. control-IgG-treated primary human keratinocytes. B Volcano plot of significant up- and down-regulated proteins with P_adj <0.05. (Mann-Whitney U-test with False Discovery Rate (FDR) correction with Benjamini, Krieger, and Yekutieli method). C Correlation matrix of supernatant protein levels with hierarchical clustering. A–C Data shown are representative of (n = 16) BP-IgG and control-IgG (n = 9) treated samples pooled from two independent experiments. Box and whiskers plot shown as mean, minimum, and maximum with box showing 25th to 75th percentile. D Proteins decreased in BP patients with antibodies detectable only against BP230 (n = 3) vs. those with antibodies only detectable against BP180 (n = 9) normalized to control-IgG-treated keratinocytes. (Welch’s t-test). E BP-IgG induces histologic blistering in 3D HSE in the absence of immune cells. (Mann-Whitney U-test). F Direct immunofluorescence demonstrates deposition of anti-basement membrane IgG in BP-IgG but not control-IgG-treated 3D HSE. E, F data shown are representative of n = 9/cohort pooled from two independent experiments. Data shown is mean ± SEM. NS not significant. Scale bar = 100 μm.

BP-IgG induces a Pro-inflammatory transcriptional response in primary human keratinocytes

To validate these findings and to better understand the global effects of BP-IgG on keratinocytes, we performed bulk RNA-seq on PHK treated with IgG from BP patients or controls. BP-IgG treated PHK demonstrated significant gene dysregulation relative to control-IgG treated PHK (Fig. 2A, Supplementary Dataset 2) with notable dysregulation of numerous cytokines and chemokines (Fig. 2B). Significant upregulation of CXCL16, IL24, and TGFB1 corresponded with our protein findings, though CXCL8 was notably downregulated at the RNA level in contrast to prior studies and protein levels¹⁵. We additionally identified significant dysregulation of complement components including upregulation of C1S and C1R, as well as S100 proteins, toll-like receptors, extracellular matrix (ECM) components, matrix metalloproteinases including MMP9, and skin barrier components (Fig. 2C). Likewise, there was dysregulation of cytokeratins, including downregulation of KRT10 and upregulation of KRT17, a marker of keratinocyte inflammation³⁶. Several cluster of differentiation (CD) markers were significantly upregulated including DPP4 (CD26), CD14, and CD68³⁷ (Supplementary Fig. 1). Gene ontology analysis revealed enrichment in skin development, as well as chemotaxis, and T-cell modulation in response to BP-IgG (Supplementary Fig. 2). To further validate our RNA-seq findings, we confirmed MMP9 upregulation by RT-PCR using a separate cohort of BP patients and PHKs (Fig. 2D). Transcriptomics therefore aligned with our identification of proinflammatory and protease release from BP-IgG-stimulated keratinocytes, while also demonstrating significant impact on keratinocyte function with upregulation of numerous desmosomal genes.

Fig. 2. Whole transcriptome analysis of BP-IgG versus control-IgG-treated keratinocytes reveals dysregulation of numerous inflammatory and structural genes.

A Heat map of bulk RNA-seq from BP-IgG vs. control-IgG-treated primary human keratinocytes. B Violin plots demonstrating normalized expression of significantly differentially expressed cytokines/chemokines, C Complement components, S100 family, toll-like receptors, extracellular matrix (ECM) components, matrix metalloproteinases, and skin barrier genes. The data shown is the mean Log2(TPM + 1) representative of n = 4 per group. Data is representative of two independent experiments. All data shown are significant at P_adj <0.05 (EdgeR). D Validation of MMP9 upregulation by qPCR from one of two independent experiments in a unique patient cohort (n = 4 IgG donors/cohort) is shown as mean, minimum, and maximum (unpaired t-test).

Blistered epithelium in BP mirrors BP-IgG Keratinocyte culture treatments

In light of these prior findings, we next aimed to determine how our BP-IgG-induced in vitro findings related to BP patient skin. First, we generated tissue microarrays from FFPE sections of patients with confirmed BP (Fig. 3A). FFPE blocks were obtained from 20 patients with confirmed BP, as well as 20 samples of age-matched normal skin (NL), and spongiotic dermatitis with eosinophils (SD) as a control for eosinophil-rich inflammation. Patient demographics are summarized in Supplementary Table 2. Age did not significantly differ between BP and normal samples (i.e. mean age73.6 vs. 75.4, P = 0.52), while patients with spongiotic dermatitis were significantly younger than those with BP (i.e., mean age 54.5 vs. 73.6, P < 0.01). Tissue microarrays were generated, sectioned, and placed on slides for spatial transcriptomics using GeoMx. To minimize the chance of including intraepithelial eosinophils or neutrophils, slides were stained for eosinophil peroxidase (EPX) or neutrophil elastase, respectively. Regions of interest (ROIs) were manually drawn onto the epithelium (Fig. 3B). To define whether the epithelium was from intact versus blistering skin, we scored blistering as 0 (attached skin), 1 (fulcrum of a blister), or 2 (distant blister). Following quality control and normalization, spatial deconvolution was performed to confirm alterations in immune cell infiltration. No significant differences in cell abundance were observed, except for fewer melanocytes in the control groups (SD + NL) relative to BP (Supplementary Fig. 3).

Fig. 3. Spatial transcriptomics demonstrates a divergent transcriptome between blistered and attached skin in BP.

A Representative tissue microarray (of 3) and distinction of attached vs. detached skin stained with H&E. B Representative region of interest selection strategy (eosinophil peroxidase, EPX) and (neutrophil elastase, NE) is shown in green and magenta respectively; DAPI is shown in blue. C Two-dimensional UMAP plot demonstrating clustering of attached and detached epithelium in BP. D Volcano plot of differentially expressed genes in detached vs. attached epithelia in BP. The top 10 differentially expressed genes sorted by P_adj are shown P_adj < 0.05 (EdgeR). E Heat map comparing attached and detached gene expression from selected genes from BP-IgG-treated human keratinocyte experiments. F Venn diagram of up- and down-regulated genes between BP-IgG vs. control-IgG-treated primary human keratinocytes and detached vs. attached epithelium in BP. G Dot plots of gene ontology (GO) and WikiPathways for detached vs. attached skin in BP (FDR-corrected). H Differential gene expression of COL17A1 and DST in detached vs. attached BP skin. I Gene expression trajectory of all BP patients in aggregate, and (J) a single patient for whom all levels of blistering were available. The data shown is representative of 26 unique cores selected from BP patients (n = 20). NS not significant, ROI region of interest. Scale bar = 200 μm.

We first questioned how detached (blistered) epithelia differed from attached epithelia in BP patient skin. Clustering and dimensionality reduction by two-dimensional UMAP revealed a clear distinction between the transcriptomes of attached and detached epithelia (Fig. 3C). Differential gene expression analysis revealed 1140 upregulated and 1434 downregulated genes, defined as P_adj < 0.05 (Fig. 3D, Supplementary Dataset 3). Numerous differentially expressed genes from BP-IgG treated keratinocytes were also noted in the detached BP epithelium, including small proline-rich proteins (SPRRs), kallikreins, IVL, S100A7, CXCL16, CD68, MMP9, and JUP (Fig. 3E). We additionally identified a decrease in C1R expression in blistered skin without a significant difference between BP attached and normal skin. While several genes were similarly upregulated or downregulated in spatial transcriptomics and primary cell experiments, others demonstrated opposite responses (Fig. 3F, Supplementary Dataset 4). For example, CXCL8 was significantly upregulated in patient skin, despite being downregulated in PHK experiments. Notably, spatial transcriptomics is less sensitive in identifying cytokine gene dysregulation, as described in previous studies on inflammatory skin disease, which may account for the failure to identify differential expression of several cytokines and chemokines detected in 2D culture experiments, such as IL-6 and IL-24³⁸.

Gene ontology analyses further supported enrichment of keratinocyte differentiation pathways as well as VEGF and hypoxia-inducible factor-1 signaling (HIF-1a) pathways which presumably correspond with separation from the BMZ (Fig. 3G). Upregulation of cornified envelope is also notable, as we have previously identified this as a response to dipeptidyl peptidase 4 (DPP4) inhibitors, a known trigger of BP³⁹. Interestingly, COL17A1 and DST were both significantly downregulated in detached skin, demonstrating a lack of a reparative mechanism and keratinocyte differentiation (Fig. 3H).

We next sought to characterize the trajectory of gene expression as a function of blistering (Fig. 3I). We aggregated patient transcriptomes from each level of blistering, scored 0-2, to identify the top differentially expressed genes of each component. We plotted these as a heatmap, revealing a pseudo-trajectory from attached to blistered BP epithelium. This revealed increasing expression of keratinocyte differentiation markers such as SPRRs, kallikreins, and cathepsin L2, as well as late cornified envelope components. We then assessed this in a single patient in whom all ROIs of all three levels of blistering were present (Fig. 3J), demonstrating a similar trajectory and differentially expressed genes of each component as a function of blistering.

Attached BP epithelium demonstrates a pro-inflammatory transcriptome shared with spongiotic dermatitis

Next, we questioned how BP epithelia differed from control skin to identify unique epithelial responses to BP-IgG. Based on our observations regarding the substantive differences between blistered and attached skin, we only considered attached skin to minimize potential bias from the blistering process for differential gene expression. Comparing BP epithelia to healthy control and spongiotic dermatitis as an inflammatory and eosinophil-rich control in aggregate revealed minimal transcriptional alterations (Fig. 4A), mostly driven by a lack of significant differences between BP epithelia and spongiotic dermatitis (Fig. 4B). Clustering by two-dimensional UMAP demonstrated that spongiotic and BP attached epithelium mostly clustered together, with polarization between normal skin and detached BP epithelia (Fig. 4C). Thus, in our comparative bioinformatic analyses, non-blistered epithelia in BP appears essentially indistinguishable from that of spongiotic dermatitis. This is surprising given the presence of autoantibodies in the former.

Fig. 4. Spatial transcriptomics of attached BP skin and spongiotic dermatitis share numerous features distinct from normal skin.

A Volcano plot comparing BP epithelia with normal skin (NL) and spongiotic dermatitis (SD), revealing few differentially expressed genes along with (B) BP vs. spongiotic dermatitis alone. C Two-dimensional UMAP plot demonstrating significant overlap between BP (attached) and spongiotic dermatitis (SD) epithelial transcriptomes with polarization of BP (detached) and normal skin. D Volcano plot demonstrating significantly differentially expressed genes between BP (attached) and normal (NL) skin defined as Log2F.C. > 0.5 and P_adj < 0.05 (EdgeR). E Gene ontology (GO) and WikiPathways bubble plots of BP (attached) vs. normal skin (NL). F Upregulation of STAT3 (P_adj = 0.01) and MYD88 (P_adj = 0.12) expression in BP vs. normal skin (EdgeR). G Venn diagram showing overlapping genes between BP (attached) vs. normal skin (NL) and BP-IgG vs. control-IgG-treated primary human keratinocytes. The data shown are representative of n = 20 patients per cohort. NS not significant.

When compared to normal epithelia only, attached BP epithelia demonstrated numerous differentially expressed genes (Fig. 4D). Upregulation of several genes including DSG3, KRT6A/KRT16, as well as S100A8/9, IL4R, VEGFA, and ADAM8 were noted, as well as downregulation of filaggrin genes (Supplementary Dataset 5). Significantly enriched pathways in BP skin relative to normal skin include cytoplasmic vesicle lumen, consistent with expected macropinocytosis induced by anti-BP180-IgG in vitro, as well as VEGFA signaling (Fig. 4E). Interestingly, we noted significant upregulation of STAT3 in attached BP epithelia versus normal skin, with a trend towards increased MYD88 (Fig. 4F). We then investigated the overlap between upregulated genes in BP epithelia relative to normal, and those differentially expressed in PHK experiments (Fig. 4G, Supplementary Dataset 6). Notably, VEGFA and ADAM8 were downregulated in BP-IgG-treated keratinocytes yet upregulated in BP epithelium, suggesting indirect mechanisms of upregulation.

Comparison of keratinocyte transcriptomes by pseudo-bulk scRNA-seq

We next utilized the recently published database by Liu et al. which included single cell transcriptomes from the lesion skin of 8 BP patients and 5 healthy controls⁴⁰. In total we assessed 27,591 keratinocyte transcriptomes (Fig. 5A, B). Keratinocytes were divided as basal, suprabasal, and granular based on expression of canonical markers, including KRT14, KRT10, and IVL respectively, with a comparable proportion between samples (Supplementary Fig. 4). We next performed pseudo-bulk differential gene expression of keratinocytes in aggregate (Fig. 5C, Supplementary Dataset 7), which similarly reflected differential gene expression of basal keratinocytes between BP and controls (Supplementary Dataset 8). These findings mirrored our spatial transcriptomics experiments, with upregulation of several inflammatory and differentiation markers, including ADAM8, C1S, DPP4, KRT6A/KRT16, IL4R, MYD88, TGFBI, and VEGFA (Fig. 5D).

Fig. 5. scRNA-seq of BP versus normal keratinocytes.

A Two-dimensional t-SNE feature plot demonstrating bioinformatically gated keratinocyte transcriptomes from BP (n = 5) vs. healthy control (n = 8, HC) patient skin. B Heat map of BP vs. HC skin. C Volcano plot with the top 10 differentially expressed genes between all keratinocytes in BP patients vs. HC skin cohort. D Violin plots of key differentially expressed genes between BP keratinocytes and HC keratinocytes using pseudobulk analysis of scRNA-seq reveals significant dysregulation of several inflammatory, differentiation, and protease markers (DeSeq2). NS not significant.

Keratinocyte MyD88 regulates much of the inflammatory response to BP-IgG

Given the upregulation of numerous MyD88 and NF-κB regulated cytokines in PHK experiments^41–45, the upregulation of STAT3⁴⁶ in BP epithelia, upregulation of MYD88 in scRNA-seq data, upregulation of MYD88 in detached BP skin relative to control (Log₂fold 0.79, P_adj = 0.003) as well as slight upregulation in BP-IgG treated keratinocytes (Supplementary Fig. 5A, B), we sought to investigate MyD88 regulation of the keratinocyte further. We utilized a MyD88 CRISPR/Cas9-mediated knockout (MyD88 KO) nTERT cells along with transfection control nTERT⁴⁴ cells, repeating treatment with BP-IgG or control-IgG. Successful knockout of MyD88 was confirmed by Sanger sequencing and western blot (Supplementary Fig. 5C).

First, we performed bulk-RNA-seq of BP-IgG-treated nTERT and MyD88 KO cells, as well as control-IgG-treated MyD88 KO cells (Fig. 6A, Supplementary Dataset 9). Principal component analysis (PCA) demonstrated distinct clustering of each respective cohort (Fig. 6B). Bulk RNA-seq revealed a diminished upregulation of numerous inflammatory markers in response to BP-IgG in MyD88 KO relative to BP-IgG-treated control cells, including IL1B, IL24, IL36G, TSLP, CXCL8, MMP1, CSF3, and S100A7/8/9/P (Fig. 6D). Barrier alarmin molecules were additionally downregulated in MyD88 KO in response to BP-IgG, including CLDN1/7/14/23, KRT6b, and KRT16 (Supplementary Fig. 6). Bulk RNA-seq of BP-IgG vs control-IgG in MyD88 KO still demonstrated several findings from PHK (Supplementary Dataset 10), including upregulation of MMP9 and IL24, indicating that MyD88 partially regulates the inflammatory molecules and proteases released by keratinocytes upon BP-IgG treatment.

Fig. 6. MyD88 regulates a number of responses to BP-IgG in keratinocytes.

A Heat map of gene expression from BP-IgG-treated control (CBP, n = 5) or MyD88-deficient nTERT cells (88BP, n = 5) and control-IgG-treated MyD88-deficient nTERT cells (88 C, n = 4). B Two-dimensional PCA plot demonstrates clustering of samples, supporting significant transcriptional changes due to MyD88 deficiency. C Heat map of supernatant proteins from BP-IgG- vs. control-IgG-treated MyD88 KO nTERT (MyD88BP, n = 4; MyD88C, n = 5) and control nTERT cells (TERTBP, n = 4; TERTC, n = 5). D Violin plots demonstrating selected genes induced by BP-IgG that are affected by MyD88 deficiency. All data shown are significant at P_adj < 0.05 (EdgeR). E MyD88 knockout blunts protein expression of IL-8, IL-24, MMP-9, TGFα, GROα, and TIMP-1. F BP-IgG induced protein expression of TGF-β1, PDGF-AB/BB, and RANTES occurs regardless of MyD88 deficiency. G MyD88 knockout results in a relative increase in MMP-1, MMP-3, and IL-20 responses. Data shown are from a single experiment as mean±SEM. E–G One way ANOVA with Tukey’s test. NS not significant. *P < 0.05, **P < 0.01, ***P < 0.001, and ***P < 0.0001.

Analysis of supernatant cytokines and chemokines demonstrated similar clustering based on supernatant protein expression (Fig. 6C, Supplementary Dataset 11). As suspected, IL-8, IL-24, MMP9, TGFα, GROα, and TIMP-1 were significantly blunted in MyD88 KOs (Fig. 6E). In contrast, TGF-β1, PDGF-AB/BB, and RANTES were upregulated by BP-IgG treatment independent of MyD88 status (Fig. 6F), while BP-IgG induced a more robust MMP-1, MMP-3, and IL-20 response in MyD88 KO than control cells (Fig. 6G). Notably, G-CSF was undetectable, while IL-1α, IP-10, GROα, M-CSF, and TNF were significantly reduced in MyD88 KO cells regardless of IgG treatments (Supplementary Fig. 6).

Keratinocyte dependent MyD88 knockout decreases disease severity in an experimental model of Bullous pemphigoid

In light of these findings, we questioned whether keratinocyte-specific MyD88 knockout could limit pathology in an in vivo model of BP induced by anti-COL17^NC14-1 IgG. To test this, we generated Krt14-dependent MyD88 knockouts (Krt14-Cre⁺;Myd88 ^fl/fl) as previously described⁴⁷, using Krt14-Cre^-;Myd88 ^fl/fl mice as controls. Krt14-Cre and Myd88 ^fl/fl have each previously undergone extensive characterization supporting epidermal and gene specificity, respectively^48,49. Knockout efficiency was confirmed by western blot (Fig. 7A). Rabbit anti-mouse COL17^NC14-1 was injected into age- and sex- matched mice over a period 2 weeks as previously described⁵⁰. Krt14-Cre⁺;Myd88 ^fl/fl mice demonstrated a significantly decreased affected body surface area relative to controls (Fig. 7B–D), without appreciable differences by sex (Fig. 7E). After excluding samples with poor RNA quality, we performed RT-PCR from lesional skin to assess Il4, Il9, Il6, Il24, Tgfb1, Cl5, C1s, and Mmp9 which demonstrated a trend towards decreased expression of Cl5 and C1s (Supplementary Fig. 7). While our original intent was to compare neutrophil infiltration, step sections demonstrated significant heterogeneity with histologic skip lesions. This heterogeneity was supported by 10³ orders of magnitude variability in Cxcl2 expression within conditions (Supplementary Fig. 8).

Fig. 7. Krt14-dependent Myd88 deficiency decreases disease severity in BP, reducing inflammatory cytokines.

A Western blot from one of each genotype demonstrates knockout of MyD88 in skin from Krt14-Cre+ mice, as well as Krt14 limited expression of Cre. B, C Representative phenotype, H&E, and direct immunofluorescence of Krt14-Cre- or Krt14-Cre + ;Myd88^fl/fl crosses treated with passive antibody transfer against Col17a1^NC1-14. D Decreased affected body surface area (ABSA) score is seen at day 14 in Krt14-Cre + ;Myd88^fl/fl (unpaired t-test), (E) without significant sex-based differences appreciated (one-way ANOVA). F Volcano plot demonstrating decreased serum levels of IL-1β, IL-4, MIP-2, and IL-9 in keratinocyte-dependent MyD88 knockouts relative to controls. B–E Data shown as mean ± SEM representative of n = 7/condition pooled from two independent experiments. NS not significant. Scale bar = 50 μm.

Keratinocyte-dependent MyD88 deficiency decreases key circulating type 2 cytokines and chemokines

As we demonstrated an overall decrease in disease severity as a result of MyD88 knockout in murine keratinocytes, we hypothesized that this decreased keratinocyte response would translate to a decreased systemic inflammatory response in BP. As such, we performed a 32-plex cytokine/chemokine multiplex immunoassay of serum from Krt14-Cre⁺;Myd88 ^fl/fl or controls. Notably, serum from keratinocyte-dependent MyD88 knockout mice demonstrated a significant decrease in serum IL-1β, IL-4, IL-9, and MIP-2 levels relative to controls, but not in other cytokines and chemokines assessed (Fig. 7F).

Discussion

Our data reveal a role for keratinocytes as an orchestrator of the inflammatory and proteolytic response seen in BP. Both transcriptomic and large-scale protein assays demonstrated a proinflammatory response in primary keratinocytes treated with BP-IgG that was similar to gene expression in blistered epithelium of BP patients, providing evidence that these autoantibody-dependent effects may contribute to disease pathology. Non-blistered epithelium in BP, by contrast, is not significantly transcriptionally different from the epithelium in spongiotic dermatitis. This is a surprising finding, as autoantibodies are, by definition, present in non-bullous BP as identified by direct immunofluorescence. This builds on prior evidence that bullous and non-bullous phenotypes demonstrate different inflammatory responses, and points to two distinct pathologic phases of disease^51,52 This also suggests that an initial keratinocyte response to a Th2 microenvironment is initially shared, as can be noted by upregulation of IL4RA in both non-bullous BP and spongiotic dermatitis. Many of the genes upregulated in BP-IgG-treated primary keratinocytes are known to be regulated by MyD88 and/or NFκB, and we found that knockout of keratinocyte MyD88 in vitro and in vivo results in blunting of many of these factors, decreasing clinical disease severity in a murine model of BP, as well as serum cytokines including IL-1β, IL-4, and IL-9.

Whether keratinocytes are a bystander in BP, or the driver of systemic response is not clear^5,6. We posit that keratinocytes are active drivers of disease in response to BP-IgG. Prior studies demonstrating keratinocyte upregulation of IL-6 and IL-8 in response to anti-BP180 antibodies suggest more than a bystander role in the inflammatory response. The occurrence of this with both IgG and IgE autoantibodies additionally highlights independence from the immune cell Fc receptor¹³. As keratinocytes release numerous pro-inflammatory markers, proteases, and complement components, this provides further support of a keratinocyte-derived systemic response. The translation of these 2D culture findings to in vivo pathogenicity remains unclear. The efficacy of whole-body topical corticosteroid treatment relative to oral corticosteroids does provide some insight into skin as an inflammatory organ⁵³. While this may be a result of direct action on skin-infiltrating leukocytes, topical corticosteroids may also directly inhibit inflammatory responses of keratinocytes. The upregulation of C1 complex components seen in BP-IgG treated keratinocytes also raises the question as to whether the keratinocyte pathologically contributes towards a complement-dependent response⁵⁴. Notably, a C1s antibody was shown to completely block complement pathway activation using serum from BP patients in vitro⁵⁵. Thus, the observation of release of C1s from keratinocytes induced by BP-IgG in serum free 2D conditions, could play a major role in the complement-dependent response in BP.

The blunting of systemic IL-4 and IL-9 in Krt14-specific Myd88 knockout mice may particularly be clinically relevant. The anti-IL4Rα antibody dupilumab is currently under investigation for the treatment of BP, with numerous retrospective studies demonstrating its efficacy^56–59. Likewise, IL-9 has recently been shown to correlate with clinical responses in BP60. Th9 cells have been identified in skin samples of BP and are able to drive eosinophil-rich and type 2 inflammatory responses in numerous other diseases^60–66. The expression of IL-1β and TGF-β1 seen in BP-IgG-treated keratinocytes is also noteworthy, as when combined with IL-4, these can drive the differentiation of naïve T cells to Th9 cells^61,67. As neither IL-4 nor IL-9 is expressed by keratinocytes, our finding of decreased serum levels in Krt14-specific Myd88 knockout mice highlights the interaction between keratinocyte-mediated inflammation as a response to BP-IgG, and development of Th2/Th9 responses thereafter. The role of other T-cell stimulatory factors identified in BP-IgG-treated keratinocytes such as CTACK⁶⁸, IL-24⁶⁹, IL-6⁷⁰, CXCR16⁷¹, and RANTES^72,73 warrants further investigation.

VEGFA was also highly upregulated in BP epithelium in our spatial transcriptomics analysis, along with NRP1 which is required for VEGFA endothelial function⁷⁴. While numerous factors can induce VEGFA, previous work has demonstrated that keratinocytes treated with IL-9 upregulated VEGFA. Circulating VEGFA has also been associated with an increased risk of venous thromboembolism^75,76, a known complication of BP thought to occur due to local coagulation factors^77–79. This finding aligns with previous studies identifying elevated VEGFA protein in BP blister fluid⁸⁰, serum⁸¹, and skin⁸². As VEGF-A was not directly induced by BP-IgG on keratinocytes, this would indicate an intermediary stimulus to induce upregulation of certain factors, such as IL-9.

Keratinocyte-derived proteases should also be considered in the context of granulocyte chemotaxis. While MMP-9 is capable of cleaving BP180, it also appears to regulate eosinophil chemotaxis to the skin in BP. A recent study utilizing an IgE model of BP demonstrated decreased eosinophil infiltration in Mmp9 knockout mice⁸³. We have also identified ADAM8 in BP epithelium which can also act as an eosinophil chemoattractant in other disease models⁸⁴. Notably, we did not detect ADAM8 in BP-IgG treated PHK supernatants, again likely indicating indirect stimulation on human skin. ADAM8 has been shown to be essential in the development of experimental asthma⁸⁵ and its inhibition has been shown to decrease eosinophil and Th2 lymphocyte infiltration in the lungs⁸⁶. It is thought to function through cleavage of CD23 into a soluble form⁸⁷. Interestingly, ADAM8 was found to correlate with airway allergen induced pneumonitis, but not with blood (i.e., drug-induced) pneumonitis, thus further implicating epithelial-immune interactions⁸⁷. There is only limited data pertaining to ADAM8 in the skin, with a study demonstrating that transgenic animals exhibit more severe oxazolone-induced cutaneous hypersensitivity⁸⁸. Further analysis in the context of BP is needed, especially since high levels of sCD23 were found in BP blister fluid compared to suction blisters⁸⁹.

Despite the histological abundance of eosinophils in BP skin, there was a notable lack of major eosinophil chemotactic factors induced by BP-IgG, aside from MMP-9 and RANTES⁹⁰. While eotaxin-1 is expressed in BP skin as well as in spongiotic dermatitis^35,91, basal expression on keratinocytes is decreased when treated with BP-IgG. While RANTES also acts as a chemotactic factor for eosinophils, its expression on BP has also been questioned, as it was not detected in an immunohistochemistry study of BP skin versus normal skin⁹² or in BP serum⁹³. Thus, we hypothesize that keratinocytes release factors that indirectly drive eosinophil chemotaxis. However, this warrants further investigation.

We additionally identified upregulation of DPP4 in both BP-IgG-treated keratinocytes and BP skin. As DPP4 inhibitors lead to a significantly increased risk of developing BP⁹⁴, this finding appears counterintuitive. DPP4 is an exopeptidase that can act on numerous secreted factors. DPP4 can drive Th1/Th17 proliferation⁹⁴, while increased number of Th17 cells have been described in early BP skin lesions⁹⁵. Soluble DPP4 levels have been noted to negatively correlate with systemic inflammation which contrasts with our findings of upregulated DPP4 in the setting of release of inflammatory molecules⁹⁶. DPP4 can, however, also cleave RANTES which may account for prior findings of a lack of detectable protein on human skin⁹⁷. DPP4 can also cleave eotaxin⁹⁸. Inhibition of DPP4 on keratinocytes leads to upregulation of the late cornified envelope and cytoskeletal remodeling, but an absence of inflammatory or proteolytic findings^39,99. Likewise, DPP4 expression did not differ in MyD88-deficient keratinocytes. Thus, whether oral DPP4 inhibitors pathologically act on keratinocyte-derived DPP4 induced by BP-IgG, or through inhibition of upstream T-cell responses, remains unanswered.

BP antibodies appear to also exert a significant effect on keratinocyte differentiation. We identified significant downregulation of numerous genes linked to cell-matrix adhesion on detached BP epithelium including COL17A1, DST, ITGB4, ITGA3, and the recently characterized LAMB4 which serves as the autoantigen in p200 pemphigoid^100,101. Decreased KRT5 and KRT14 along with increases in granular layer markers such as IVL, TGM1, FLG, and LCE3D further point towards keratinocyte differentiation. We have previously demonstrated that blockade of the laminin-332 and integrin α6β4 interaction was sufficient to induce blistering and result in keratinocyte differentiation via protein kinase C and NOTCH21. Thus, BP appears to demonstrate a similar phenomenon, whereby autoantibody-induced blistering is perpetuated by a loss of differentiation with loss of hemidesmosomal gene expression.

Our study has several limitations. Total IgG rather than BP180- or BP230-specific IgG was extracted from serum. Thus, the impact of alternative autoantigens or differences in the dose of pathogenic autoantibodies cannot be ruled out. We did not, however, identify differences amongst most markers assessed in patients with BP180 versus BP230 antibodies. Future studies with epitope-limited (i.e. NC16a vs. C-terminus) autoantibodies will provide further insight into BP180 site-specific regulation of inflammation. We also lacked a sufficient sample size of patients receiving DPP4 inhibitors, which are described to target different epitopes of BP180 and have a less inflammatory response¹⁰², to perform a subset analysis. Likewise, due to differences in diagnostic tests and batch effects between experiments, we were unable to pool samples to perform an adequate correlation analysis of indirect immunofluorescence titers or anti-BP180/BP230 antibody levels with protein or gene expression.

Spatial transcriptomics provides a pseudo-bulk-transcriptomic view that lacks single-cell resolution and prevents the identification of single epithelial and immune cells and their respective interactions, thus limiting the study of regional differences in keratinocytes. This was partially supplemented by employing and analyzing a scRNA-seq dataset of BP patients and controls, though these were from different patients. This dataset likewise revealed skin lesions with different cellular composition from prior studies identifying lymphocyte and eosinophil-predominant inflammation¹⁰³, though this was presumably due to technical challenges of scRNA-seq in capturing granulocyte transcriptomes¹⁰⁴. Nonetheless, by clustering blistered versus attached BP skin, we were able to gain significant insight into transcriptional differences and perform a pseudo-trajectory analysis. This demonstrated a critical role of keratinocyte differentiation as a function of skin separation. Addtionally, as 19 of 20 patients demonstrated C3 deposition on direct immunofluorescence, we were likewise unable to perform a subanalysis based on the presence or absence of complement deposition. As blood and skin samples were also not matched, we could not compare individual paired keratinocyte responses and skin expression of key inflammatory markers.

Lastly, given we noted significantly decreased IL-1α, G-CSF, IP-10, M-CSF and TNF in MyD88-deficient keratinocytes regardless of BP-IgG treatment, we cannot rule out that these were the critical factors resulting in decreased disease severity in vivo and type 2 response rather than blunted IL-24 or MMP9 expression. Likewise, as CXCL8 is not expressed in mice, it is unclear to what extent cytokines and chemokines are blocked in murine MyD88-deficient keratinocytes relative to human keratinocytes. Still, given a decrease in disease severity in vivo, these observations nonetheless implicate keratinocytes in the coordination of the systemic response to BP-IgG.

In conclusion, we demonstrate that keratinocytes play an active role in the response to autoantibodies in BP. IgG autoantibodies induce an inflammatory response, leading to the release of proteases. These antibodies are sufficient to induce blistering independent of granulocytes, leading to keratinocyte differentiation and loss of expression of hemidesmosomal genes. This keratinocyte response is regulated in part through MyD88 signaling, resulting in decreased disease severity and a reduction of several cytokines and proteases. Thus, our study highlights the important contributory role of keratinocytes to the pathogenesis of BP.

Materials and Methods

Ethics approvals

Studies utilizing human samples were approved by the Institutional Review Board at Rush University Medical Center (IRB #: 20121406) and conducted according to the principle of the Declaration of Helsinki. Animal studies were approved by the Institutional Animal Care and Use Committee at Rush University Medical Center (IACUC #: 20-079, 22-024) in accordance with NIH guidelines.

Patients

Peripheral blood was collected from patients with active BP following written informed consent. A diagnosis of BP was confirmed using previously reported standards¹⁰⁵ including clinical suspicion of BP, positive direct immunofluorescence, and a positive serologic test either for BP180 and/or BP230 autoantibodies by ELISA, and/or deposition on the roof of a salt split skin blister.

Mice

B6N.Cg-Tg (Krt14-Cre)1Amc/J mice were crossed for at least 6 generations with C57B6/J mice (referred to in the text as Krt14-Cre). Myd88 ^fl/fl (B6.129P2(SJL)-Myd88tm1Defr/J) mice were generated as previously described⁴⁹. This strain contains a loxP site flanking exon 3 of Myd88 which retains MyD88 function in the absence of Cre recombinase. When homozygous mice are bred to Cre-expressing mice, they demonstrate deletion of exon 3 in Myd88. All strains (JAX Stock No.018964, 000664 and 008888, respectively) were purchased from Jackson Laboratories (Bar Harbor, Maine, USA). Genotyping was performed using gDNA collected from the ear by TransnetYX. Animals were housed under specific pathogen-free conditions with a 12 h light–dark cycle 70–75⁰ at 40–60% humidity and fed standard chow ad libitum.

Multiplex immunoassay and ELISA

Multiplex measurements of supernatants were performed using the Luminex 200 System (Luminex, Austin, Texas, USA) by Eve Technologies Corp. (Calgary, Alberta, Canada). TGF-β 3-Plex Discovery Assay® (MilliporeSigma, Burlington, Massachusetts, USA), Eve Technologies’ Human MMP/TIMP 13-Plex Discovery Assay, and Eve Technologies’ Human Cytokine 96-Plex Discovery Assay were used to measure supernatant cytokine, chemokine, and metalloprotease expression according to the manufacturer’s instructions. Additional specific markers including ADAM8 (Biorbyt, Durham, North Carolina, USA), IL-36G (Thermo Fisher Scientific, Waltham, Massachusetts, USA), C1s (RayBiotech, Peachtree, Corners, Georgia, USA), and CD26 (Thermo Fisher Scientific, Waltham, Massachusetts, USA) were measured by ELISA following the manufacturer’s instructions. Mouse cytokines were measured using the 32-plex cytokine assay (MilliporeSigma, Burlington, Massachusetts, USA). Standard curves were generated for each analyte to define the concentrations of each analyte. All samples were run in duplicate, with the average of the duplicates used for further analysis. A complete list of analytes assessed is shown in Supplementary Table 3.

2D cell culture and treatments

Serum underwent affinity purification using ToxOut (BioVision, Milpitas, California, USA; Cat# K2505) or Nab columns (Thermo Fisher Scientific Waltham, Massachusetts, USA; Cat# 89957), followed by buffer exchange and concentration using Amicon Ultra-15 centrifugal filter units with a 50 kDa filter in PBS (Millipore Sigma, Burlington, Massachusetts, USA; Cat# UFC905008) with magnetic azide removal (Nanopartz, Loveland, Colarado, USA; Cat# PPZ-KIT-10). Multiple purchased pooled human control IgG and individual normal serum controls were utilized to increase biological variability of control IgG. Pooled IgG was purchased from MP Biomedicals (Santa Ana, California, USA; Cat# 0855908), Cell Sciences (Newburyport, Massachusetts, USA; Cat# CSI11983A), and Sigma (Saint Louis, Missouri, USA; Cat# I4506). Individual donor sera were purchased from Innovative Research (Novi, Michigan, USA; Cat# ISERS2) and underwent the same purification as BP samples. Primary adult human keratinocytes (PHKs) were cultured in EpiLife Medium with Human Keratinocyte Growth Supplement (all from Thermo Fisher Scientific Waltham, Massachusetts, USA; Cat# C0055C, MEPICF500, s0015) in a humidified atmosphere of 5% CO₂ at 37 °C. MyD88 knockout N/TERT cells or CRISPR/Cas9 transfection controls were generated as previously described⁴⁴ with permission of Dr. James G. Rheinwald¹⁰⁶ for original N/TERT cells. N/TERT cells were grown in KC-SFM medium supplemented with 30 μg/ml bovine pituitary extract, 0.2 ng/ml epidermal growth factor, and 0.3 mM calcium chloride (Thermo Fisher Scientific Waltham, Massachusetts, USA; Cat# 17005042, 13028014, PHG0314). When confluency reached approximately 70%, cells were treated with patient or control-IgG at a final concentration of 4.0 μg/μl overnight. Each sample represents a distinct culture. Supernatants were removed and stored at −80 °C until protein measurement. Cells pellets were washed twice with PBS and stored at −80 °C until RNA extraction. Independent experiments were performed on different passages.

3D human skin equivalents

3D human skin equivalents (HSE) were generated and propagated as previously described using primary neonatal human keratinocytes¹⁰⁷. Three unique keratinocyte donors were used for 3D HSE. After 9 days of growth at the air-liquid interface, 3D HSE were either treated with 2 mg/mL of pooled BP-IgG from 4 donors with BP180 reactivity or pooled control-IgG for 72 h. All cultures were harvested on day 12. Sections were either fixed in formalin or placed in TissueTek OCT compound (Sakura Finetek, Torrance, CA, USA; Cat#4583) with routine histology, or direct immunofluorescence, respectively, performed by standard procedures.

RNA extraction

RNA was extracted from cultured primary human keratinocytes (PHKs) or N/TERT keratinocytes using a miRNeasy Mini Kit according to the manufacturer’s instructions (Qiagen Inc., Germantown, Maryland, USA; Cat# 217084). On-Column DNase I Digestion was used to prevent genomic DNA contamination. For RT-PCR of lesional murine tissue, 5 μM cryosections of OCT-embedded tissue were processed using the Promega SV Total RNA Isolation System (Promega, Fitchburg, Wisconsin, USA; Cat# Z3105) per manufacturer instructions. Samples were run as duplicates (skin) or triplicates (keratinocytes).

RT-PCR

RT-PCR validation was performed as previously described in ref.²⁰. Individual primers are shown in Supplementary Table 4. Primers were synthesized by IDT (Coralville, Iowa, USA). Data for qPCR were generated using a separate cohort of BP-IgG patients and PHKs.

Bulk RNA-seq

RNA quantity was determined using a Nanodrop 1000 spectrophotometer (Nanodrop, Wilmington, Delaware, USA). RNA integrity number (RIN) and concentration were determined using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, California, USA). A minimum RIN of 7 was required for further processing. Poly (A) mRNA was subsequently extracted and enriched from total RNA utilizing oligo (dT)-attached magnetic beads according to the manufacturer’s instructions (Illumina, San Diego, California, USA). Enriched and purified mRNA was fragmented into approximately 200 nucleotide short mRNAs, followed by first-strand cDNA synthesis using random hexamers as primers. Second-strand cDNA was constructed in a buffer containing dNTPs, DNA polymerase I, and RNase H. Appropriate fragments were isolated and enriched by PCR amplification. cDNA libraries were then pair-end sequenced using an Illumina NovaSeq sequencing platform (Illumina, San Diego, California, USA) with a read length of 2 × 150 and approximately 20 million reads per sample.

Tissue microarray generation and spatial transcriptomics

We utilized the Rush University Medical Center Department of Pathology’s search feature to identify retrospective formalin fixed paraffin embedded (FFPE) tissue sections consistent with BP, spongiotic dermatitis with eosinophils (SD), or age matched normal skin (NL). BP samples were further confirmed by a positive serologic test. Individual sections from each specimen were stained with H&E per standard protocol to confirm the initial impression and suitability for the generation of a tissue microarray.

Regions of interest (ROIs) were pre-selected by the principal investigator (K.T.A.) and reviewed by a pathologist (V.M.) for histological verification. The array layout (4 columns x 7 rows) was defined following specific requirements to arrange donor cores in the center of a standard recipient block in an area no larger than 35 mm × 14 mm per GeoMx size requirements. An automated microarrayer, the TMA Master (3DHISTECH, Ltd., Budapest, Hungary) was used to remove and transfer cylindrical tissue cores from each donor block (cases & controls) and re-embed them into a recipient TMA paraffin block. The donor tissue core diameter was 2 mm and the distance between donor tissue cores was 0.7 mm. At least one core was chosen per patient, with additional cores taken for distinct regions such as the presence or absence of blistering. Three TMA blocks were built with samples randomized across the three blocks. Once construction was completed, the finalized TMA blocks were tempered overnight in a CO₂ incubator at 37 °C, facing down onto a clean glass slide to allow closure of any gap between the donor tissue cores and the surrounding paraffin wax. After tempering, the TMA blocks were left overnight to cool at room temperature before the glass slides were removed, and the blocks were ready for further use.

Representation of at least the epithelium of all samples was confirmed by cutting serial sections and performing H&E staining, followed by slide scanning utilizing the Aperio AT2 Slide scanner (Leica Biosystems, Wetzlar, Germany). Samples for transcriptomic analysis were cut to a thickness of 5 μm. Samples were stained with antibodies targeting EPX¹⁰⁸ (kindly provided by Dr. Elizabeth Jacob) directly conjugated to Alexa Fluor 594 (Lightning-Link, Abcam, Cambridge, UK; Cat# ab269822), and neutrophil elastase conjugated to Alexa Fluor 647 (Novus, Centenial, Colorado, USA; Cat# MAB9167)) with DAPI counter stain to identify potential intraepidermal granulocytes. We then utilized the Nanostring GeoMx system (Nanostring, Seattle, Washington, USA) following the manufacturer’s recommend protocol to generate ROI and libraries. ROIs were manually assigned based on identification of epithelial morphology, excluding intraepithelial granulocytes. For BP samples, ROIs were defined as either blistered or adherent skin. Of 96 epithelial ROIs, 7 were removed due to low gene detection rate ( <10%) or low alignment) (n = 5 BP, n = 1 SD, n = 1 NL). Background signal was detected using negative control probes. Normalization was performed by the Third Quartile approach. Data and figure generation were subsequently performed in R (4.1.0). Spatial deconvolution¹⁰⁹ was performed to estimate the relative abundance of cell types in each ROI. Due to a lack of granulocytes in normal skin, we utilized two scRNA-seq data sets: Solé-Boldo et al.¹¹⁰ and Danaher et al.¹⁰⁹ for further downstream query and analyses.

Generation of MyD88 KO keratinocytes (KCs) in N/TERT-2G

CRISPR KO KCs were generated as previously described¹¹¹. In brief, single-guide RNA (sgRNA) target sequence was developed (MYD88 sgRNA: CTGCTCTCAACATGCGAGTG for MyD88 KO using a web interface for CRISPR design (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). Synthetic sgRNA target sequences were inserted into a cloning backbone, pSpCas9 (BB)-2A-GFP (PX458) (Addgene plasmid Stock No. 48138), and then cloned into One Shot Stbl3 chemically competent E. coli cells (Thermo Fisher Scientific, Rockford, Illinois, USA; Cat# C737303). Proper insertion was validated by Sanger sequencing. The plasmid with proper insertion was then transfected into an immortalized KC line (N/TERT-2G) using the TransfeX transfection kit (ATCC, Manassas, Virginia, USA; Cat# ACS4005) in the presence of JAK1/JAK2 inhibitor, baricitinib (10 mg/ml, AchemBlock, Hayward, California, USA; Cat# G-5743). GFP-positive single cells were plated and then expanded. Cells were then genotyped and analyzed by Sanger sequencing. MyD88 KO keratinocytes were validated by western blot (Anti-MyD88 1:1000, Abcam, Cambridge, UK; Cat# ab2064); (anti-beta actin, 1:5000, Millipore-Sigma, Milwaukee, WI, USA; Cat# A5441).

Validation of Krt14-specific MyD88 KO

To validate skin-specific knockout, fresh ear tissues of Krt14-Cre⁺- or Krt14-Cre^--MyD88 ^fl/fl crosses were lysed with RIPA lysis buffer. Proteins in the supernatants were quantified using the BCA method. 25 µg of total protein was loaded and separated by 12% SDS-polyacrylamide gel electrophoresis and transferred to PVDF membranes. After blocking, the membranes were incubated with the appropriate primary antibodies against MyD88 (clone 4D6, 1:500, Invitrogen, Carlsbad, California, USA; Cat# MA5-16231), Cre recombinase (clone D7L7L, 1:500, Cell Signaling, Boston, Massachusetts, USA; Cat# 15036S) or GAPDH polyclonal antibody (1:500, Thermo Fisher Scientific, Rockford, Illinois, USA; Cat# MA5-15738) at 4 °C overnight. After washing, the blots were subsequently incubated with horseradish peroxidase- (HRP) conjugated secondary antibodies (1:3000). Protein signals were developed using SuperSignal West Pico Plus Chemiluminescent Substrate (Thermo Fisher Scientific, Rockford, Illinois, USA; Cat# 34580) and imaged using the LI-COR Odyssey XF.

Experimental pemphigoid model

The experimental BP murine model was generated as a previously described in ref. ⁵⁰. Briefly, recombinant NC14-1 cDNA was generated using gene synthesis (Eurofins MWG Operon, Ebersberg, Germany). After codon optimization for E. coli, cDNA was cloned into the pET24d-N expression vector where the NC14-1 protein was expressed with a terminal-His tag. Following expression in E. coli the recombinant protein was purified by affinity chromatography using TALON metal affinity resin (TaKaRa, San Jose, California, USA; Cat# 635501). Anti-NC14-1 IgG was generated by immunization of New Zealand white rabbits (Kaneka Eurogentec S.A. Seraing, Belgium) with recombinant NC14-1. Total IgG from immunized rabbit sera was affinity purified using protein G Sepharose (Genscript, Piscataway, New Jersey, USA; Cat# L00209). Age- and sex-matched Krt14-Cre⁺;Myd88 ^fl/fl or Krt14-Cre^-;Myd88 ^fl/fl co-housed littermates aged 8-12 weeks were injected intraperitoneally (i.p.) thrice weekly for 2 weeks to induce a cutaneous phenotype. Non-mendelian sex distribution was noted with fewer females per litter. As such, 2 female and 5 female mice were treated per cohort. The percentage of affected body surface area (ABSA) was measured by an investigator blinded to the experimental conditions on days 0, 4, 9, and 14. Disease scoring was modified from a previous mouse model of epidermolysis bullosa acquisita¹¹². The percentage of ABSA was multiplied by 0.5 for alopecia, or 1.0 for erythema, crusting or erosions. Initial power analysis was performed to determine sizing using α = 0.05 and β = 0.8, with an anticipated mean disease surface area of 5 ( ± 1) and an expected observed severity of 3. Hematoxylin and eosin stains were performed using standard protocols. Mouse direct immunofluorescence was performed as previously described¹¹², using anti-rabbit secondary antibody to confirm successful passive transfer. Perilesional tissue was placed in OCT with 5μm sections and stained with AlexFluor 488 conjugated goat anti-Rabbit IgG (1:1000) and DAPI counterstain were purchased from Thermo Fisher Scientific (Rockford, IL, USA; Cat# A-11008). All slides were immediately photographed following staining using an Evos FL microscope (Thermo Fisher Scientific, Rockford, Illinois, USA).

Bulk RNA-seq analyses

Paired-end reads were aligned to the human genome (Hg38/gencode.v42) with bowtie (Version 1.2.3) and quantified using the RNA-seq by Expectation-Maximization algorithm (RSEM) (Version 1.3.3) using standard parameters¹¹³. Samples were filtered and only protein-coding and lncRNA-coding genes expressed at minimum 1 TPM in at least one sample in all biological replicates were considered for downstream analyses. A pseudocount of 0.1 was added and samples were Log-transformed. Pair-wise comparisons were performed using edgeR as per developer’s suggestions with correction for FDR¹¹⁴. Gene ontology pathway analyses were performed with (Enrichr)^115–117. Downstream visualization including heatmaps (complexheatmap)^118,119, principal component analyses, and volcano plots (enhancedvolcano), were performed as suggested by developer with minor modifications.

Spatial transcriptomics

GeoMx-NGS RNA expression analysis was performed with Geomx tools as previously described^120,121. Briefly, GeoMx raw count files, and metadata (DCC, PKC, and annotation files) were loaded in R Studio (Version 2024.04.2 + 764). Quality control, filtering and normalization was then performed as suggested by the developer with minor modifications. Q3 normalized data was used for differential gene expression analyses and performed as previously described^120,121. Downstream visualizations and statistical analysis including dimensionality reduction with UMAP and t-SNE, heatmaps, volcano plots, pathway analyses with Enrichr^115–117, and differential expression analyses with linear mixed effect models were performed as before and as suggested by developer. Raw sequencing files have been uploaded to the GEO server.

Pseudo-bulk scRNA-seq analyses

We utilized the recently published scRNA-seq dataset described by Liu et al.⁴⁰. Briefly, Cell Ranger output files of 8 healthy controls (HC) and 5 BP patients were downloaded from the Zenodo database (Accession code: 10924853; https://zenodo.org/records/10924853). Count matrices were pre-processed and doublets/multiplets were removed using Single-Cell Remover of Doublets (Scrublet)¹²² (Version 0.2.1). The resultant AnnData object was exported as compressed AnnData H5AD file and were further read in Seurat for downstream low-quality cell pruning¹²³. Seurat objects were merged, normalized, and highly variable genes (features) and scaling were performed using SCTransform¹²⁴ as described before¹⁰⁴. Basal, suprabasal, and granular keratinocytes were identified based on their bona fide expression of KRT14, KRT10, and IVL expression, respectively. Differential gene expression in all keratinocytes between HC and BP patients was determined by DESeq2 with minor modifications¹²⁵. Downstream visualization including heatmaps (complexheatmap), volcano plots (enhancedvolcano), and two-dimensional t-distributed Stochastic Neighbor Embedding (tSNE) were performed as suggested by the developer with minor modifications.

Statistical analyses

For non-transcriptomic data, statistical analysis was performed in Prism (Version 10.5.0, GraphPad Inc, La Jolla, CA, USA). All tests were two sided. Significance was defined as . <0.05. Correction for multiple corrections were applied as indicated in the legends. EdgeR and DeSeq2 analysis corrected for multiple comparisons by FDR.

Reporting summary

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

Author contributions

Conceptualization: M.A.M., C.M., C.D.N., K.T.A.; Investigation: L.B., C.F.G.-J., J.L., M.P., S.E., Y.L., A.P.M., Y.W., B.E.P.W., N.C.N., C.C., M.K.S., V.M., M.S.; Supervision: S.G., K.K., J.E.G., E.S., K.T.A.; Resources: M.P., S.E., A.A., N.I., T.H., S.G., K.K., M.K.S., J.E.G.; Formal analysis: K.T.A., C.F.G.-J., Y.L., K.T.A.; Writing original draft: C.F.G.-J., K.T.A.; Visualization: C.F.G.-J., Y.L., K.T.A.; Funding acquisition: K.T.A.; Reviewed final manuscript: L.B., C.F.G.-J., J.L., M.P., S.E., Y.L., A.P.M., Y.F.W., A.A., N.I., T.H., B.E.P.W., S.G., K.K., N.C.K., C.C., M.K.S., J.E.G., V.M., M.S., M.A.M., C.M., C.D.N., E.S., K.T.A.

Peer review

Peer review information

Nature Communications thanks Hideyuki Ujiie and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The transcriptomic data generated in this study have been deposited in the Gene Expression Omnibus database under accession code GSE280619 and GSE280620, (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE280620). Prior scRNA-seq data sets were accessed from the Genome Sequence Archive (GSA) database under accession code HRA003993, HRA000145, and HRA000471. The other data generated in this study 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

This work was supported by Astra Zeneca. A.M., C.M. and C.N. are past or present employees of AstraZeneca and may hold stock and/or stock options or interests in the company. E.S. has a scientific cooperation with Astra Zeneca and received consulting fees from Astra Zeneca (both not related to the present project). The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-62495-2.