Spatial proteomics identifies JAKi as treatment for a lethal skin disease

Toxic epidermal necrolysis (TEN) is a rare but often fatal, drug-induced cutaneous adverse drug reaction (CADR) characterized by widespread keratinocyte death and epidermal detachment affecting more than 30% of body surface area. Despite multiple proposed mechanisms, the principal molecular drivers of keratinocyte cytotoxicity in TEN remain unclear and current management is largely supportive, with high mortality in severe cases. The broader CADR spectrum includes maculopapular rash (MPR), drug reaction with eosinophilia and systemic symptoms (DRESS), and Stevens–Johnson syndrome (SJS), with SJS–TEN overlap representing intermediate severity. To address the gap in mechanistic understanding and to identify druggable pathways, the study applies deep visual proteomics (DVP) to archived, formalin-fixed paraffin-embedded (FFPE) lesional biopsies, achieving cell-type-resolved, single-cell–based proteomic profiles of keratinocytes and infiltrating immune cells across CADR severities. The overarching goal is to map disease- and cell-type-specific molecular alterations that can inform targeted therapy for TEN.

Previous work has implicated various immune pathways and effector molecules in severe CADRs including TEN, such as HLA associations with drug hypersensitivity, T cell–mediated cytotoxicity, Fas/Fas ligand (CD95)–mediated apoptosis, TNF and IFN-γ involvement, nitric oxide synthase activation, and granulysin as a mediator of disseminated keratinocyte death. Clinical guidance has largely emphasized supportive care due to lack of proven disease-modifying therapies. Spatial and single-cell omics approaches have recently emerged to provide molecularly resolved views of intact tissues, enabling profiling of specific cell types within lesions. Viral reactivation has been linked to DRESS pathogenesis and proliferative signatures have been noted in immune infiltrates. However, a comprehensive, cell-type–resolved proteomic map across CADR subtypes to identify shared and divergent pathways, and to nominate therapeutic targets for TEN, has been lacking.

Study design and cohorts: Retrospective analysis of lesional FFPE skin biopsies from patients with CADRs of varying severity—MPR (n=5), DRESS (n=5), TEN (n=6)—and healthy controls (n=5), totaling n=21 individuals for initial DVP profiling. Additional expanded cohorts were used for targeted transcriptomics and validation assays. Deep Visual Proteomics (DVP): Tissue sections were stained for CD45 (immune cells) and pan-cytokeratin (keratinocytes). Machine learning–based cell segmentation identified cell-type contours. Laser microdissection collected contours per cell type for each individual, maintaining patient- and cell-type specificity. Proteins were extracted and analyzed by mass spectrometry, identifying approximately 5,000 unique proteins per cell type. Statistical analyses included unpaired two-sided t-tests for differential expression, one-way ANOVA with Tukey’s HSD for multi-group comparisons, and Benjamini–Hochberg FDR correction (<0.05). Principal component analysis (PCA), hierarchical clustering, overrepresentation analyses (Reactome), and gene set enrichment analyses (MSigDB Hallmark, GO BP) were performed. Immune cell subtype spatial proteomics: A multiplexed DIA (mDIA) workflow with a reference channel and the Astral mass analyzer enabled profiling of extremely sparse immune subsets—CD163+ macrophages, CD4+ T helper cells, and CD8+ cytotoxic T cells—from TEN/SJS-TEN skin. Approximately 20-cell equivalents per type and participant were analyzed, yielding a median of 2,104 proteins per sample (total 5,302). Differential expression between cell types and interferon-pathway protein intensities were assessed. Sample sizes: n=12 macrophage, 10 CD4+, 10 CD8+, 9 healthy samples from n=12 disease-affected and 7 healthy individuals; paired comparisons in n=10 individuals. Spatial keratinocyte proteomics (attached vs detached): Segmented keratinocytes were collected from detached (blister roof) and attached (adjacent) regions within the same TEN biopsy (about 20 cells per type/patient). PCA, ANOVA clustering, and identification of drivers distinguishing attached vs detached and vs healthy were performed (n=7 attached, 11 detached, 9 healthy samples; n=11 disease-affected and 7 healthy individuals). Targeted transcriptomics and histology: Targeted mRNA profiling of cytokines and JAK/STAT components was conducted in an expanded cohort (n=10 per group), assessing differential expression (t-tests with BH correction). Immunohistochemistry/immunofluorescence visualized STAT1 and phosphorylated STAT1 (pSTAT1) localization and levels across cohorts; nuclear localization and phosphorylation status were quantified. In vitro cytotoxicity model: An autologous co-culture assay was developed in which activated peripheral blood mononuclear cells (PBMCs) killed keratinocytes within 72 h. Live-cell imaging quantified keratinocyte cytotoxicity over time with or without JAK inhibition. The pan-JAK inhibitor tofacitinib was tested in dose-dependent fashion (representative condition: 50 nM), with n=6 biological replicates per condition and 4 fields of view per well. Mouse models of TEN and JAK inhibition: A smac-mimetic (IAP antagonist)–induced TEN-like mouse model reproduced cutaneous inflammation, epidermal detachment, and histologic features of TEN. Oral JAK inhibitors were administered either one day before induction (pre-treat model) or 3 h after induction (treatment model). Doses: tofacitinib 30 mg/kg, baricitinib 10 mg/kg, abrocitinib 20 mg/kg. Outcomes included clinical scores (days 1 and 3), histology (day 1), dermal thickness (15 measurements per mouse), lesion size, body weight change, pSTAT1 levels, keratinocyte death (TUNEL+ and cleaved caspase-3+), immune infiltration (CD45+), and keratinocyte proliferation (Ki67). Statistical tests included one-tailed unpaired Welch’s t-tests and unpaired two-sided t-tests as specified; two independent experiments; each data point represents one mouse. Humanized mouse model: Immunodeficient mice received intravenous PBMCs from a TEN survivor, followed by daily culprit-drug administration, inducing severe ocular conjunctivitis and conjunctival epithelial cell death by day 14. Oral baricitinib (10 mg/kg) vs vehicle was tested for therapeutic efficacy, assessing ocular reaction, disease occurrence, and percentage subepithelial cell death. Clinical application (compassionate use): Seven patients with TEN or SJS–TEN overlap received off-label oral JAK inhibitors (including abrocitinib; others as clinically indicated). Clinical course, extent of epidermal detachment, re-epithelialization, survival to day 30, adverse events, and cutaneous pSTAT1 levels pre/post treatment were documented.

- DVP revealed distinct, cell-type–resolved proteomic signatures across CADR severities, with keratinocyte proteomes clustering by phenotype and PC1 separating TEN from other CADRs and healthy controls. PC1 drivers included calcium homeostasis and stress-response proteins (HSPA5, TMCO1, ATP2A2) and a dominant interferon signature with STAT3. - Differential expression increased with disease severity; DRESS keratinocytes had >2-fold increases in MHC class I antigen presentation machinery (TAP1/2, TAPBP, HLA-A). TEN keratinocytes showed strong upregulation of antimicrobial and interferon-induced proteins (e.g., WARS1) and downregulation of SOD1 (approximately halved), implicating oxidative stress in keratinocyte apoptosis. - Hierarchical clustering identified a TEN-specific keratinocyte cluster enriched for complement activation and clotting; nearly all complement factors were uniquely upregulated in TEN keratinocytes. - Immune-cell proteomes also clustered by disease. DRESS displayed proliferative and viral pathway enrichment (E2F/MYC targets, DNA replication/ATR response), while TEN showed a dominant type I/II interferon signature with STAT1 as a key driver. DRESS had the largest number of DEPs vs healthy, followed by TEN and MPR. - Interferon pathway proteins were most abundant in CD163+ macrophages compared to CD4+ and CD8+ T cells; STAT1 levels were consistently higher in macrophages, along with myeloid markers (MNDA, MRC1) and interferon-driven proteins (STAT1, GBP5). - Spatial keratinocyte profiling (attached vs detached) showed complement/inflammation proteins (e.g., MX1, C3, KRT6, KRT16) elevated in both regions, indicating inflammatory activation already present adjacent to blisters; subtle but consistent differences separated attached from detached cells, with several proteins associated with early pre-detachment changes. - Six proteins (WARS1, STAT1, S100A9, LYZ, GBP1, APOL2) were upregulated ≥4-fold in both keratinocytes and immune cells in TEN (vs healthy), all within the interferon/JAK–STAT axis. Interferon-stimulated genes were strongly induced (e.g., GBP1 ~9-fold; GBP5 detectable only in TEN). - Targeted transcriptomics showed IFNG was the most potently upregulated JAK/STAT-signaling cytokine in TEN, exceeding TNF. JAKs and STATs were broadly upregulated in TEN and SJS–TEN overlap. STAT1 protein was increased and nuclear-localized in keratinocytes and immune cells; phosphorylated STAT1 was markedly elevated. - In vitro, activated PBMCs efficiently killed keratinocytes within 72 h; the pan-JAK inhibitor tofacitinib reduced cytotoxicity in a dose-dependent manner (e.g., 50 nM effective). - In a smac-mimetic TEN-like mouse model, oral tofacitinib and baricitinib significantly reduced clinical severity (days 1 and 3), decreased pSTAT1 signal, normalized dermal thickness, reduced keratinocyte death (TUNEL+ and cleaved caspase-3+), and lowered immune infiltration (CD45+). Epidermal recovery accelerated with increased basal-layer Ki67, suggesting no impairment of wound healing. - JAK1-selective inhibitors (abrocitinib, upadacitinib) also significantly reduced disease severity. Therapeutic administration after disease induction similarly decreased clinical severity, lesion size, and recovery time. - In a humanized mouse model (TEN survivor PBMCs + culprit drug), baricitinib reduced ocular conjunctivitis and conjunctival epithelial cell death compared to vehicle. - Compassionate-use clinical data: Seven patients with TEN or SJS–TEN overlap—some refractory to high-dose corticosteroids—received JAK inhibitors. All survived to day 30 without reported side effects, with rapid cessation of disease progression (e.g., within 48 h in an index case with SCORTEN 4), marked re-epithelialization (e.g., ~95% by day 16), and reduced cutaneous pSTAT1 post-treatment.

Cell-type–resolved spatial proteomics identified the JAK/STAT and interferon signaling pathways as central, shared drivers of inflammation and cytotoxicity in TEN, distinct from less severe CADRs. The convergence of strong interferon signatures and STAT1 activation in both keratinocytes and infiltrating immune cells suggests a self-amplifying inflammatory loop that propagates tissue damage. Macrophages emerged as prominent contributors, exhibiting the highest interferon-pathway activation among tissue immune cells, which may help explain the breadth and severity of epidermal injury. Spatial analyses demonstrated that keratinocytes adjacent to overt blisters already display robust complement and inflammatory activation, consistent with clinical observations (e.g., indirect Nikolsky sign) and supporting an active role for keratinocytes in perpetuating local cytotoxic responses. Translationally, JAK inhibition—pan-JAK or JAK1-selective—effectively attenuated keratinocyte killing in vitro, reduced disease severity and pathologic hallmarks in two distinct mouse models (including a humanized system), and was associated with rapid clinical improvement and re-epithelialization in seven patients without observed short-term adverse effects. These results indicate that targeting JAK/STAT can disrupt the pathogenic interferon-driven cascade and restore skin integrity in TEN. Findings in DRESS, including EZH2 detection and proliferative/viral-response signatures, underscore disease-specific pathways in other CADRs and highlight the value of spatial proteomics for discovering actionable mechanisms across inflammatory dermatoses. Overall, the work addresses the key question of TEN pathogenesis by pinpointing the JAK/STAT axis as the dominant, druggable pathway and provides multi-level evidence—molecular, cellular, animal models, and initial clinical observations—supporting JAK inhibitors as a promising therapeutic strategy.

This study delivers a comprehensive, cell-type–resolved spatial proteomic atlas across CADRs and identifies the JAK/STAT–interferon axis as a principal driver of TEN. Mechanistic insights include marked STAT1 activation in keratinocytes and macrophages, complement pathway upregulation, and broad ISG induction. Translationally, JAK inhibition—both pan-JAK and JAK1-selective—reduced disease severity and pathologic features in preclinical models and was associated with rapid re-epithelialization and survival without observed short-term adverse effects in seven patients with TEN/SJS–TEN overlap. These findings provide a strong rationale for controlled clinical trials of early JAK inhibitor therapy in TEN and suggest broader applicability of spatial proteomics to identify druggable pathways in complex inflammatory diseases. Future work should include randomized studies to confirm efficacy and safety of short-term JAK inhibition in TEN, deeper characterization of macrophage contributions and interferon sources, and exploration of related mechanisms (e.g., epigenetic regulators in DRESS) to tailor interventions across CADR subtypes.

The proteomic cohort size was modest (approximately five individuals per CADR group and healthy controls, six for TEN) and based on retrospective FFPE biopsies, which may limit generalizability. Some spatial and subtype analyses relied on extremely sparse cell populations, potentially constraining depth despite advanced mDIA methods. The clinical application comprised seven off-label treated patients without a randomized control group, precluding definitive conclusions on efficacy and safety. Mouse models, including smac-mimetic induction and humanized PBMC systems, recapitulate key features but may not capture all aspects of human TEN. Longer-term safety of JAK inhibition in TEN and potential interactions with oncologic therapies require evaluation in prospective trials.