Integrated RNA-Seq Profiling Identifies NRXN1+ Cell-Fibroblast Associated MAPK Signaling in Prurigo Nodularis

Abstract

Purpose

Prurigo nodularis (PN) is chronic pruritic dermatosis with hyperkeratotic nodules and persistent itch, occurring alone or with atopic dermatitis (AD). The molecular basis of nodular architecture and sustained pruritus remain poorly defined, particularly the role of fibroblasts and their interactions with neuronal components. This study aimed to identify PN-specific pathogenic profiles by comparing lesional transcriptomes from AD, atopic prurigo (AP), and PN.

Methods

Bulk RNA sequencing was performed on lesional skin from AD (n = 6), AP (n = 7), PN (n = 7), and healthy controls (HC; n = 4). Publicly available single-cell RNA-sequencing datasets were reanalyzed for cell-type signatures. Integrated analyses included differentially expressed genes, pathway enrichment, cell-type deconvolution, ligand–receptor inference, and immune cell profiling.

Results

PN lesions showed MAPK/ERK pathway enrichment in fibroblasts compared with AD and AP. NRXN1+ cells were inferred as major interacting partners of fibroblasts, with ligand–receptor interactions consistent with enhanced fibroblast MAPK/ERK activity. Immune pathway analyses showed weaker Th2 pathway activity in PN compared with AD/AP, providing context for neuronal-stromal interactions as a putative distinguishing feature.

Conclusions

PN shows enhanced MAPK/ERK pathway enrichment in fibroblasts and inferred NRXN1+ cell-fibroblast interactions. This candidate axis may contribute to nodular lesion architecture and persistent pruritus, suggesting potential therapeutic strategies.

INTRODUCTION

Prurigo nodularis (PN), also referred to as chronic nodular prurigo, is highly burdensome pruritic inflammatory dermatosis with multiple hyperkeratotic nodules and a persistent itch-scratch cycle. PN affects individuals across all age groups, although it is more common in older ages, and epidemiological studies indicate that it occurs worldwide.¹ Patients with PN frequently experience profound impairment in quality of life, including sleep disturbance, reduced daily functioning, and psychological distress; effective long-term disease control remains challenging.²

Accumulating evidence suggests that PN represents a neuroimmune disorder rather than merely a secondary consequence of scratching.³,⁴ Lesional skin demonstrates altered cutaneous innervation, including increased dermal nerve density⁵ and reduced epidermal nerve fibers,⁶ along with the enhanced expression of pruritogenic mediators.⁵ These findings support the dysregulation of the neurocutaneous unit, in which sensory neurons, immune cells, keratinocytes, and stromal cells engage in bidirectional communication that sustains chronic pruritus and lesion persistence.³ Despite these advances, the cell-type-specific molecular profiles and intracellular signaling networks that characterized PN lesions and contribute to chronic itch remain incompletely defined.

A defining histopathological feature of PN is the presence of firm dermal nodules accompanied by stromal expansion and extracellular matrix (ECM) remodeling.⁷ Fibroblasts are central regulators of dermal architecture and are capable of responding to inflammatory and neural cues through production of cytokines, growth factors, and neuroactive mediators.²,⁷,⁸ Given their anatomical proximity to cutaneous nerve endings and their established role in tissue remodeling, fibroblasts may serve as critical intermediaries linking neuronal dysregulation to the structural changes observed in PN nodules. However, while emerging evidence supports fibroblast-neuronal interactions in PN, a comprehensive, integrated transcriptomic characterization of PN fibroblast remains lacking.

Clinically, PN may develop in individuals with or without an atopic background. A subset of patients with atopic dermatitis (AD) develops prurigo-like nodules, often referred to as atopic prurigo (AP).⁹ Although these conditions share type 2 inflammation features,³,¹⁰,¹¹ they differ in dominant morphology and histopathological architecture, suggesting the presence of disease-specific molecular drivers. The nosological position of AP remains incompletely defined, and direct cell-type resolved comparisons among AD, AP, and PN are limited.

To address these gaps, we sought to delineate the transcriptional signatures and intercellular communication networks that distinguish PN from related pruritic dermatoses. We integrated bulk RNA sequencing from the well-characterized lesional biopsies of AD, AP, and PN with publicly available single-cell RNA sequencing (scRNA-seq) datasets to identify cell-type-specific signatures and candidate pathways underlying nodular remodeling and persistent pruritus.

MATERIALS AND METHODS

Study cohort and sample collection

Four-millimeter punch biopsies were obtained from patients with AD (n = 6), AP (n = 7), and PN (n = 7), as well as in healthy volunteers (n = 4), after obtaining written informed consent. AD was diagnosed according to the Hanifin and Rajka criteria. AP was defined as the development of PN-like nodules in patients with a history of AD, whereas PN was defined as PN without a prior history of AD. Biopsy specimens were taken from representative lichenified lesions in AD and nodular lesions in AP/PN at the initial visit, prior to any pharmacological treatment. Samples were preserved in RNAlater solution (ThermoFisher Scientific, Waltham, MA, USA) and stored at −20°C until further analysis.

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The study was approved by the Institutional Review Board at National Medical Center (approval No. NMC 2022-07-089). The procedures used in this study adhere to the tenets of the Declaration of Helsinki. The patients in this study have given written informed consent to the publication of their case details.

RNA extraction and analysis

Samples were processed with a grinder, and RNA was extracted using TRizol (ThermoFisher Scientific) according to the manufacturer’s instructions. Quality control regarding the concentration and integrity of the isolated RNA was performed with an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Then cDNA was reverse transcribed and amplified with Quant-seq with dual indices (Lexogen, Wien, Austria), and sequencing was carried out with Illumina Hiseq 2500 platform as paired-end reads. Following UMI extraction with umi-tools,¹² trimming was performed with BBTools function bbduk. After adapter trimming, sequence reads were aligned to the human genome (GRCh37) using STAR.¹³ Count matrix was made with Htseq.¹⁴ Differential expression analysis was carried out using DESeq2,¹⁵ with differentially expressed genes (DEGs) defined as |log2fold-change (FC)| ≥ 1 and Benjamini–Hochberg adjusted P-value < 0.05. Over-representation and enrichment analyses were conducted using the clusterProfiler package,¹⁶ utilizing the enrichGO function. Gene set variation analysis (GSVA) was carried out using the GSVA package.¹⁷ Cell-type deconvolution was conducted using the MuSiC,¹⁸ with scRNA-seq data as a reference. Overall analysis was performed using R (v4.2; R Foundation for Statistical Computing, Vienna, Austria).

Single-cell RNA sequencing

The scRNA-seq data were acquired from Gene Expression Omnibus (GEO) as in accession number GSE222840 and GSE173205. Quality control thresholds were applied to filter out low-quality cells, including min_genes = 200, min_cells = 3, less than 10%, or 15% mitochondrial reads using Scanpy,¹⁹ followed by doublet filter using Scrublet.²⁰ Dataset was log-normalized and 5,000 highly variable genes were selected to cluster cells, and integration was performed using scVI-tools.²¹ Clustering and sub-clustering was performed with the Leiden graph-based method (scanpy.tl.leiden) and visualized using Uniform Manifold Approximation and Projection (UMAP) by the scanpy.tl.umap function. Differential abundance analysis of cellular neighborhoods was performed with milopy.²² The pyDESeq2²³ was used to identify DEGs in each sample types with |log2FC| ≥ 1 and Benjamini–Hochberg adjusted P value < 0.05. Over-representation and enrichment analyses were conducted by gseapy package²⁴ using databases from the gseapy library. Gene expression imputation was carried out using the Markov Affinity-based Graph Imputation of Cells²⁵ algorithm. Receptor-ligand analysis and transcription factor network visualization were performed with Liana+²⁶ and decoupler.²⁷ Overall analysis was performed using Python and the computing server of the Genomic Medicine Institute Research Service Center.

Statistical analysis

The Wilcoxon rank-sum test was used to compare non-normally distributed continuous variables between independent groups. Statistical significance was defined as P < 0.05.

Data availability

Bulk RNA-seq datasets generated during the current study are available from the corresponding author on reasonable request, and scRNA-seq datasets are publicly available via GEO under GSE222840 and GSE173205, respectively, for disease and healthy control (HC) samples.

RESULTS

Bulk transcriptomic gene expression patterns in AD, AP, and PN

The demographic and clinical characteristics of the study participants are summarized in Supplementary Table S1. Age was older in PN (mean age 47.3 ± 18.1 years, 57.1% male) compared with AD (mean age 27.2 ± 8.5 years, 50% male), AP (mean age 35.3 ± 17.6 years, 42.9% male) and HC (mean age 26.5 ± 4.4 years, 50% male).

To identify disease-specific transcriptomic differences among AD, AP, and PN (using HC as reference), we performed bulk RNA-seq analysis and utilized publicly available scRNA-seq datasets from the GEO repository (Fig. 1). Unsupervised principal component analysis demonstrated clear segregation of disease groups from HC, with partial overlap observed among the disease groups (Supplementary Fig. S1A). To better interpret this partial overlap, we next applied GSVA using curated gene sets (Supplementary Table S2). This analysis revealed a distinct transcriptomic signature in AD, including enhanced responses to bacterial molecules and signatures consistent with impaired epithelial barrier integrity compared with AP and PN (Supplementary Fig. S1B). DEGs further underscored divergence between AD and AP or PN (Supplementary Fig. S1C). Notably, a neuropeptide precursor proenkephalin (PENK) and acute-phase inflammatory mediator orosomucoid 1 (ORM1) were upregulated in AP and PN compared with AD (Supplementary Fig. S1D). Gene Ontology (GO) analysis comparing each disease group to HC revealed significant enrichment of ERK1/2 cascade-related pathways in AP and PN compared with AD (Supplementary Fig. S1E). Overall, PN showed a distinct transcriptional profile from AD, with shared signaling mechanisms with AP, underlying their transcriptomic similarity.

Fig. 1. Overview of the study design. Skin biopsies were processed for bulk RNA-seq in this study, including RNA extraction, library preparation, sequencing, and downstream data processing. The scRNA-seq data were obtained from publicly available datasets (GEO: GSE222840, GSE173205) and reanalyzed, including quality control, dimensionality reduction/clustering, and cell-type annotation. Bulk transcriptomes were further deconvoluted using MuSiC with cell-type gene expression signatures derived from scRNA-seq reference. Figure created with Biorender.com.

RNA-seq, RNA-sequencing; scRNA-seq, single cell RNA-sequencing; GEO, Gene Expression Omnibus.

Enhanced MAPK signaling in fibroblast of PN

To validate and further characterize molecular features of diseases at single-cell resolution, we leveraged scRNA-seq datasets including biopsies from AD (n = 5), AP (n = 1), PN (n = 7) and HC (n = 4). After quality control and integration, 15 distinct cell clusters were identified (Supplementary Fig. S2A and B) and annotated based on canonical marker genes (Supplementary Fig. S2C).

Comparative analysis of cell-type proportions with HC as a reference showed an elevated proportion of keratinocytes across all three disease groups, with the greatest increase observed in AD (Supplementary Fig. S2D). In addition, there was a trend of an increase in smooth muscle cells and mast cells in PN, while AP showed an increased proportion of B cells, and plasma cells. Keratinocytes remained the predominant expanded cell type in AD. At a subcluster level, using Milo, we detected enriched fibroblast subsets and endothelial cells and a reduced keratinocyte population in PN compared with AD (false discovery rate < 0.1) (Supplementary Fig. S2E).

In addition, pseudo-bulk analysis derived from aggregating single-cell transcriptomes identified disease-specific pathway differences between PN and AD (Fig. 2A). PN was enriched for ECM organization, angiogenesis-related pathways, and MAPK/ERK signaling, along with neuronal pathways such as axon guidance and synapse organization. In contrast, AD showed enrichment of mitochondrial-related pathways, including mitochondrial gene expression, transcription, and oxidative phosphorylation.

Fig. 2. Increased activation of the MAPK pathway in the fibroblasts of PN.

(A) GO enrichment analysis based on DEGs identified from pseudo-bulk comparisons between the AD and PN groups. Enriched GO terms are categorized by biological functions, with AD-related terms shown in yellow and PN-related terms in pink. ERK1/2 cascade-related pathways are highlighted in bold. (B) Heatmap showing GO enrichment analysis of DEGs from fibroblast pseudo-bulk across HC and disease groups. The left panel represents combined scores, and the right panel shows −log10 (adjusted P values). Key terms with significant adjusted P-values are shown, and darker colors indicate stronger enrichment or higher significance. (C) GSEA analysis of selected signaling pathways, with comparison of fibroblasts in PN versus HC (top) and PN versus AD (bottom). Bolded group labels indicate the condition with upregulated pathways relative to the other group. (D) Bar plot illustrating an increased activity of key TFs in PN fibroblasts compared with AD fibroblasts. Red bars indicate increased activity, while blue bars indicate decreased activity. Bolded labels indicate the condition with upregulated TFs relative to the other group. (E) Network plot of key genes regulated by TFs. TFs are colored based on enrichment scores, and targets are colored by gene expression.

PN, prurigo nodularis; GO, Gene Ontology; DEG, differentially expressed gene; AD, atopic dermatitis; GSEA, Gene Set Enrichment Analysis; HC, healthy controls; TF, transcription factors.

Because PN showed both an increased proportion of fibroblasts and strong pseudo-bulk enrichment of ECM remodeling and MAPK/ERK signaling, we next performed cell-type-resolved analyses to pinpoint the cellular sources of these pathways, with a particular focus on fibroblasts. Fibroblast-associated pathways were analyzed by comparing each disease group with HC to evaluate the enrichment scores (combined scores) and adjusted P values (Fig. 2B). Fibroblasts from PN exhibited significant enrichment of ERK1/2 cascade, angiogenesis, MAPK signaling, extracellular matrix organization and epithelial-to-mesenchymal transition (EMT) pathways, while fibroblasts from AD predominantly featured type II interferon signaling pathways. Gene Set Enrichment Analysis supported the enrichment of the MAPK signaling pathway in fibroblasts of PN relative to those from HC (normalized enrichment score [NES], 1.77; P = 0.03) and AD (NES, 2.81; P = 0.00) respectively (Fig. 2C).

Potential transcription factors regulating these processes in PN fibroblasts were systematically analyzed, revealing higher activity of the SMAD family (SMAD1, SMAD2, SMAD3, STAT3, MEF2A, JUN) in PN fibroblasts compared with AD (Fig. 2D). These transcription factors predominantly targeted inflammation-related (IL11, IL10, NFKBIA, A2M, SOCS1) and EMT-related (TWIST2, BMP2, RUNX2, COL1A1, and COL1A2) genes (Fig. 2E), pointing to a coordinated transcriptional network that promotes inflammatory signaling and EMT/ECM remodeling. The concurrent upregulation of collagen- and matrix metalloproteinase-related genes further supports an activated matrix remodeling state, consistent with dermal remodeling underlying the nodular phenotype (Supplementary Fig. S2F).

NRXN1+ cells are associated with fibroblast MAPK/ERK activation in PN

To examine neurocutaneous signatures in PN, we assessed the expression of neuronal- and neuropeptide-related genes across disease groups. Genes involved in axon guidance and synapse organization (SEMA5A, NRCAM, and EPHB2), together with neuroactive molecules associated with synaptic transmission and neuroendocrine regulation (GRIA3, NMB, and POMC), were upregulated in AP and PN compared with AD. Notably, their expression patterns were concordant between bulk RNA-seq and the fibroblast cluster in the scRNA-seq data (Fig. 3A). To investigate whether these neurogenic signatures are accompanied by altered intercellular signaling, we next refined cluster annotations with a particular focus on the neuron-labeled cluster in the scRNA-seq data. Because neuronal populations are challenging to fully capture by scRNA-seq, we re-examined this cluster using established neuronal markers (SCN7A and NRXN1). Cells with highest expression levels were designated as NRXN1+ cells (Supplementary Fig. S3A). Schwann cell markers (SOX10, GLDN, and TJP1) were detected only in a small subset of cells (Supplementary Fig. S3B). We therefore classified this population as NRXN1+ cells to better reflect its neuronal-like transcriptional identity. This cluster showed enrichment in nervous system development, signal release from synapses, and anterograde trans-synaptic signaling (Supplementary Table S3).

Fig. 3. NRXN1+ cell–fibroblast interactions and fibroblast MAPK/ERK signatures in PN.

(A) Boxplot showing the expression of various neural activity-related genes and neuropeptides across disease groups in fibroblasts from bulk RNA-seq data and scRNA-seq. The dotted line represents the median expression for each gene. The scRNA-seq gene expression was imputed using MAGIC. (B) Dot plot illustrating interactions between NRXN1+ cells and fibroblasts in PN versus AD across cell clusters. Dotted box highlights interactions between NRXN1+ cells as a source and fibroblasts as target. Dot size represents significance (ligand_P-value), and color indicates interaction strength (interaction_statistic). (C) Predicted downstream regulatory pathways in PN fibroblasts (left) and keratinocytes (right) associated with NRXN1+ cell interaction. Red indicates upregulation, and blue indicates downregulation.

PN, prurigo nodularis; RNA-seq, RNA-sequencing; scRNA-seq, single cell RNA-sequencing; MAGIC, Markov Affinity-based Graph Imputation of Cells; AD, atopic dermatitis; AP, atopic prurigo; HC, healthy controls.

Building on these observations, we next asked whether these neurogenic features could be shaped by intercellular communication with neuron-like populations. To address this, we performed ligand–receptor interaction analysis focusing on interactions between NRXN1+ cells (neuron-like) and stromal cells, in particular fibroblasts. This analysis highlighted stronger predicted communication from NRXN1+ cells to fibroblasts and keratinocytes in PN (Fig. 3B). NRXN1+ cells were predicted to provide ligands, including neuroactive mediators (PLAT and PSAP) that interact with LRP1, and ECM-associated/secreted ligand (MFGE8, COL18A1, COL1A1, SPARC, and LAMB1), that engage fibroblast cell-surface receptors (PDGFRB, ITGB5/ITGA5, FGFR1, and ITGA1) related to MAPK signaling. Subsequent regulatory network analysis implicated the MAPK signaling cascades as a downstream pathway associated with communication between NRXN1+ cells and fibroblasts (Fig. 3C). In fibroblasts, MAPK signaling genes were upregulated, consistent with increased MAPK/ERK-associated activity. In contrast, keratinocytes showed a reduced expression of MAPK signaling genes, driven from decreased ERBB2/CDK1/MAPK3K8/MAPK pathway signaling. These findings highlight cell-type-specific differences in MAPK/ERK signaling outcomes.

Th2 pathway activity differs between AD and PN

Given that AD and PN share type 2 inflammation yet differ in clinical and tissue phenotypes, we profiled the immune cell composition and T-cell pathway activity. First, immune cells from scRNA-seq data were sub-clustered, identifying 13 distinct immune cell subsets, including T cells, B cells and plasma cells, NK cells, and innate lymphoid cells, as well as myeloid-derived cells such as dendritic cells, monocytes, macrophages, mast cells and langerhans cells (Fig. 4A and B). Clusters were effectively distinguished by canonical marker gene expression, displaying unique clustering patterns across the disease groups that reflect differences in immune cell composition (Supplementary Fig. S4A and B). PN showed increased mast cells, AP showed increased B cells and plasma cells; AD showed increased proliferating T cells. Using MuSiC, we deconvoluted bulk RNA-seq data and observed increased macrophage and monocyte clusters in PN compared with HC, with no significant differences in other immune cell types across disease groups (Fig. 4D and E, Supplementary Fig. S4C, and Supplementary Table S4).

Fig. 4. Immune cell composition and T-cell pathway profiling across AD, AP, and PN.

(A) UMAP visualization of immune cell clusters in scRNA-seq across disease groups. (B) UMAP embedding of immune cells (n = 24,896) identified from scRNA-seq after sub-clustering. Clusters are colored according to annotated immune cell-types. (C) Cell-type proportions across disease groups in scRNA-seq data. (D) Boxplots showing the estimated proportions of various immune and non-immune cell types across disease groups derived from deconvolution analysis (MuSiC) of bulk RNA-seq data. Each subplot represents a specific cell-type. The center line indicates the mean proportion. (E) Relative proportions of MP and Mo across disease groups from bulk RNA-seq deconvolution data. (F) Boxplots of GSVA scores for Th1, Th2, and Th17/22 immune pathways calculated from bulk RNA-seq data across disease groups. Significant differences are marked with asterisks.

AD, atopic dermatitis; AP, atopic prurigo; PN, prurigo nodularis; UMAP, Uniform Manifold Approximation and Projection; scRNA-seq, single cell RNA-sequencing; RNA-seq, RNA-sequencing; GSVA, gene set variation analysis; Th, helper T cell; HC, healthy controls; B, B cell; DC, dendritic cell; EC, endothelial cell; F, fibroblast; ILC, innate lymphoid cell; KC, keratinocyte; LC, Langerhans cell; MC, mast cell; MEL, melanocyte; Mo, monocyte; MP, macrophage; NK, natural killer cell; PC, plasma cell; ProT, proliferating T cell; SGC, sweat gland cell; SMC, smooth muscle cell; Tc, cytotoxic T cell; Treg, regulatory T cell; Mo, monocyte; ns, non-significant.

^*P < 0.05, ^**P < 0.01.

To further characterize T-cell-mediated immune responses, we analyzed bulk RNA-seq data by comparing log2FC of genes associated with Th1, Th2, and Th17/22 pathways in each disease group relative to HC. Although these values do not represent direct statistical comparisons between disease groups, the relative magnitude of HC-normalized changes allow for an indirect assessment of pathway activity. Based on this comparison, Th2- associated genes showed higher or statistically significant log2FC in AD and AP than in PN when referenced to HC (Supplementary Fig. S4D). Consistent with this observation, GSVA analysis performed on bulk RNA-seq data confirmed relatively significant Th2 pathway activation in AD than in PN and AP (Fig. 4F).

DISCUSSION

PN is increasingly recognized as a neuroimmune disorder characterized by chronic itch, repeated scratching, and progressive dermal remodeling that culminates in hyperkeratotic nodules. By integrating bulk RNA-seq with the reanalysis of publicly available scRNA-seq datasets, our study highlights MAPK/ERK-associated fibroblast activation and stromal remodeling as key features. NRXN1+ cells emerged as a neuron-like population showing prominent predicted ligand–receptor interactions with fibroblasts, consistent with the enrichment of MAPK/ERK-associated signatures in fibroblasts.

Previous studies have reported elevated neuropeptide expression²⁸,²⁹ along with the infiltration of immune cells and inflammatory mediators in PN supporting a neuropathic component. One notable transcriptomic difference between PN versus AD was the upregulation of PENK, which encodes proenkephalin—a neuropeptide precursor involved in pain modulation and stress response.³⁰ Although PENK upregulation in lesional skin does not establish causality, its enrichment in PN relative to AD supports the concept that PN lesions harbor a distinct neurosensory milieu. Moreover, we observed the enrichment of axon signaling and synapse organization programs, particularly in fibroblasts, reinforcing the coupling between neuronal dysregulation and fibroblast activity more prominently in PN than in AD. Notably, fibroblasts in PN also showed an increased expression of multiple neuropeptides and neuroactive molecules, further supporting a neurogenic fibroblast state in PN lesions.

Importantly, NRXN1+ cells were also predicted to communicate with keratinocytes. However, the inferred downstream outcome differed by cell-type. Whereas fibroblasts showed the relative upregulation of MAPK/ERK-associated genes, keratinocytes exhibited a reduced expression of MAPK signaling genes, suggesting cell-type-specific differences in signaling directionality and pathway engagement.

Among inferred ligand-receptor pairs, PLAT (tPA) and PSAP were predicted to engage LRP1 on fibroblasts. Prior work has shown that tPA-LRP1 signaling can promote fibroblast activation³¹ and trigger ERK1/2-associated signaling.³² LRP1 has also been implicated in fibroblast ECM remodeling,³³ supporting biological plausibility. However, because tPA/PSAP are not cell-type restricted and because our interactions are inference based, whether NRXN1+ cells provide these ligands to activate fibroblast LRP1 in PN lesions requires further validation.

Because PN is inflammatory dermatosis, PN fibroblasts also exhibit the enrichment of gene profiles related to inflammation, fibrosis, ECM remodeling, and EMT. TF analysis indicated an increased activity of SMAD family proteins, which are known to play key roles in EMT and fibrosis through transforming the growth factor (TGF)-β signaling pathway.³⁴,³⁵ The co-activation of SMAD1 and SP7 indicates the concurrent engagement of the BMP-TGF-β signaling axis,³⁶ providing a mechanistic basis for the enhanced ECM remodeling and fibrotic responses³⁷ observed in PN fibroblasts. SP7 (Osterix), albeit classically associated with osteoblast differentiation, has been shown to promote the expression of key ECM components such as COL1A1 and COL1A2.³⁸,³⁹ This SP7-driven ECM-related activity further supports heightened fibroblast activation and extensive ECM remodeling in PN.

This increased expression of neuropeptides and neuron-associated transcripts in fibroblasts may further amplify the pain-itch cycle by increasing the sensitivity of cutaneous sensory neurons, leading to exaggerated response to both painful stimuli and itch triggers. Such neuro-stromal dysregulation may occur alongside the local tissue microenvironment, including vascular changes that can facilitate neuroimmune interactions. In line with this, numerous studies that have described the role of vascularization in prurigo,⁴⁰ particularly its association with epidermal remodeling as evidenced by the presence of sprouting angiogenesis and increased epidermal thickness has been observed across all patients.⁴¹

In addition to stromal and neuronal dysregulation, PN exhibits an immune cell landscape characterized by an increased infiltration of monocytes, macrophages, and mast cells. These innate immune cells are well known to sustain chronic inflammation through the persistent secretion of pro-inflammatory cytokines and pruritogenic mediators. Monocytes recruited from the circulation differentiate into tissue macrophages that contribute to inflammatory amplification via tumor necrosis factor-α, interleukin (IL)-1B, and chemokine production. In the murine models of PN, M2-polarized macrophages have been identified as a primary source of IL-31, and their depletion markedly reduced both IL-31 expression and scratching behavior, suggesting a direct link to chronic pruritus.⁴² Mast cells, in close proximity to sensory nerve endings, further intensify neurogenic inflammation by releasing histamine and IL-31, thereby contributing to an itch-inflammation amplification loop.⁴³

Similar to PN, AP showed the intermediate activation of MAPK and EMT associated pathway activation between AD and PN, consistent with its clinical view. AP also showed an increased expression of neuropeptide genes. We sought to unravel the interaction between NRXN1+ cells and fibroblast clusters in AP, which also exhibited neuronal activity along with PN. However, this analysis was not feasible due to the limited number of cells and the small number of AP samples.

Previous studies have demonstrated that oxidative stress plays a critical role in AD pathogenesis, with an increased expression of genes encoding oxidative stress response proteins in the skin of patients with non-lesional atopic dermatitis, highlighting the role of mitochondria as the key regulators of reactive oxygen species (ROS) production and oxidative stress.⁴⁴ ROS not only contribute to skin inflammation but also drive keratinocyte differentiation,⁴⁵ implicating mitochondrial dysfunction in AD-related skin barrier defects.

As widely reported,⁴⁶,⁴⁷ AD showed a Th2-mediated response, with elevated levels of cytokines such as IL-4, IL-5, and IL-13. However, Th22 pathways, marked by IL-22 production, and variable contributions from Th1 and Th17 axes also play roles, depending on the AD subtype.⁴⁸ Our data showed an increase in Th2 responses in AD. Cytokines, such as IL4, IL17A, and IFNG, could not be reliably detected in our dataset, as they did not pass the quality control threshold (average count > 0). The absence or low detection of cytokine transcripts reported in previous studies is likely due to their transient expression and rapid secretion and degradation.⁴⁹ Accordingly, we interpreted T-cell-related activity primarily at the pathway level and observed that Th2 pathway activity was more pronounced in AD than in AP or PN. This immune context motivated our subsequent analyses, in which neuronal-stromal interactions emerged as a major axis distinguishing PN, beyond differences in Th2 immune pathway activation alone.

Taken together, our findings suggest a novel interaction between NRXN1+ cells and the fibroblasts of PN as well as a unique signaling pathway to describe potential PN pathogenesis. These findings advance our current understanding of PN, providing insights into novel therapeutic targets related to fibroblast and neuronal activity.

However, this study has several limitations that should be acknowledged. First, only one AP sample was available in the scRNA-seq dataset, which limits the strength and generalizability of AP-specific findings at single-cell resolution. Although the overall AP transcriptional pattern was partially consistent with our bulk RNA-seq results, direct comparisons between PN without AD and PN with AD should be interpreted with caution. Consequently, AP-related observations in this study should be regarded as preliminary and require validation in larger, independent cohorts with more balanced sample representation. Second, the neuronal-fibroblast interactions identified in PN were inferred from ligand–receptor analyses and transcriptional correlations rather than direct spatial or functional evidence. As a result, the proposed interaction between NRXN1+ cells and fibroblasts should be regarded as putative and require validation through spatial transcriptomics, multiplex immunostaining, or in vitro functional assays. Third, rare neuronal populations, such as NRXN1+ cells, are difficult to capture consistently in skin single-cell datasets, and their low abundance may have limited the resolution of neuronal subtype classification. Finally, although integrated bulk and single-cell analyses provide complementary insights, the cross-platform integration may introduce technical variability inherent to differences in sample preparation, sequencing depth, and data preprocessing.

Although our key findings were consistent across multiple analytical layers, we acknowledge that the PN cohort was older than the HC, which may partially contribute to group-level differences in skin transcriptomes. Therefore, our results should be interpreted with consideration of age as a potential confounder and validation in age-matched cohorts will be important for future work. Further studies with larger cohorts and spatially resolved, multimodal approaches will be essential for substantiating and extending these findings.

SUPPLEMENTARY MATERIALS

Supplementary Table S1

Summary of the demographic and clinical characteristics of the study cohorts

Supplementary Table S2

Input gene lists for GSVA

Supplementary Table S3

Gene ontology of NRXN1+ cluster

Supplementary Table S4

Cell-type proportions derived from bulk RNA-seq deconvolution using scRNA-seq data

Supplementary Fig. S1

Bulk RNA-seq analysis of pruritic skin disorders.

Supplementary Fig. S2

scRNA-seq cell clusters and characteristics of PN.

Supplementary Fig. S3

NRXN1 and SCN7A expression defines neuron-associated cells.

Supplementary Fig. S4

Immune cell profiles from scRNA-seq.