Single-cell RNA sequencing profiles age-related transcriptional landscapes in human hair follicle cells Authors: Qian Zhao, Rui Ma, Kun Huang, Juan Wang, Donglin Zhang, Jingyuan Wang, Xiaofeng Ding, Feiyun Chen, Sijia Zhao, Na Ni, Xiaodie Zhang, Qian Du, Xiaojun Lin, Hua Wan, Jianglin Zhang, Xiaolei Ding, Shuang Yang, Fengping Xu, Yongxian Lai Published: 2025 ABSTRACT AbstractHair loss and graying, the earliest visible signs of skin aging, are driven by the functional decline of hair follicle stem cells and their niches. To elucidate the transcriptional mechanisms involved in scalp aging, we conducted a comprehensive analysis of human scalp samples using single-cell RNA sequencing and spatial transcriptomic technologies. Our study profiled the transcriptomes of 57,181 cells from scalp samples obtained from four young, six middle-aged, and one elderly individual. The integrated bioinformatic pipeline included cell clustering, spatial deconvolution, pseudotime trajectory, as well as cell-type specific gene expression, and intercellular communication analysis. An additional 92 volunteers were included, comprising 90 (37 young, 27 middle-aged, and 26 elderly) for trichoscopic examination, one young individual for senescence-associated β-galactosidase (SA-β-gal) staining, and one elderly individual for both MKI67 immunofluorescence and SA-β-gal staining. This approach led to several key findings: we identified three subtypes of mitotic keratinocytes that localized in the interfollicular epidermis (IFE), outer root sheath (ORS), and hair matrix, with pseudotime trajectory further confirming their transitional stage. Furthermore, in middle-aged scalps, we observed activated activator protein 1 (AP-1) transcription factor complex in keratinocytes, upregulated DCT gene in melanocytes, and decreased bone morphogenetic protein (BMP) and noncanonical wingless/integrated (ncWNT) signaling in dermal papilla (DP)–keratinocytes cross-talk. Due to the insufficient sample size and under-representation of elderly samples, transcriptional features associated with late aging, sex, and scalp regions were not completely captured. Nevertheless, our study provides valuable cell-resolved transcriptional insights into hair follicle aging and may support the development of future regenerative therapies.Keywords: hair follicle, aging, cell-type-specific gene expression, intercellular communication INTRODUCTION Skin, the largest organ of mammals, is our outermost physical barrier against external insults such as sun exposure, air pollution, and microorganism invasion [1,2]. The skin comprises three major layers—epidermis, dermis, and hypodermis—which consists of highly heterogeneous groups of cells including keratinocytes, fibroblasts, adipocytes, melanocytes, sebocytes, immune cells, endothelial cells, etc. Cellular composition varies significantly depending on its anatomical location. The skin replenishes itself every 4 weeks in human beings. Its appendage—hair follicles—undergo cycles of growth (anagen), regression (catagen), and resting (telogen), which are regulated by paracrine signals or direct cell‒cell interactions between the hair follicle stem cells (HFSCs) and their niche [3]. Due to its continuous periodic renewal and the impact of hemostatic disruptions on dermatological health, the skin and its hair follicles are intensively studied to advance our understanding of adult stem cell biology [[3], [4], [5]] and the aging process [6,7]. Single-cell RNA (scRNA) sequencing has been proven to be a valuable tool in identifying transcriptomic changes by cell type [8,9]. The recent advent of the spatial transcriptomic technologies [10] complements scRNA sequencing technologies with spatial information. The combination of these two powerful technologies has already provided new insights into the pathology and treatments of dermatological diseases [11]. While several studies have explored the single-cell molecular mechanisms of skin aging [7,12], relatively few have focused on the aging of hair follicles. Hair loss and graying are early visible signs of skin aging that often cause social anxiety and psychological stress. Although most skin diseases are not life-threatening, they can significantly reduce the quality of life of those affected. As a highly heterogeneous and dynamic organ, the skin, along with its appendages such as hair follicles and sebaceous glands, undergoes complex aging processes [[13], [14], [15]]. Understanding the molecular mechanisms underlying hair follicle aging holds significant scientific, clinical, and cosmetic value. Therefore, we studied the aging mechanism in human hair follicle cells using scRNA sequencing of human scalp samples combined with spatial transcriptomics and identified distinct transcriptional patterns associated with aging in the data. RESULTS Annotation of Human Scalp Cells A total of 11 volunteers aged between 20 and 59 years were recruited and donated their scalp samples for scRNA sequencing (Figure 1A). Four samples obtained from individuals aged 20–29 years (1 man and 3 women) were assigned to the young group, and 6 samples obtained from individuals aged 34–42 years (3 men and 3 women) were assigned to the middle-aged group. Additionally, one sample from a 59-year-old man was designated as the elderly group. A section from one scalp sample of the young group (from a 20-year-old male) was used for spatial transcriptomic sequencing. To evaluate whether hair phenotypes were associated with age, another 90 volunteers (49 women and 41 man) aged between 20 and 65 years were recruited for trichoscopy examination. The results showed that the hair density was associated with age (Figure 1B). Although terminal hair density, hair diameter, and vellus hair showed no significance between the young and middle-aged group, there was a significant difference between the elderly group and the other two groups (Figure 1B). The cellular senescence phenotype was evaluated in 2 samples (both men, one at 30 and another at 52) using the senescence-associated β-galactosidase (SA-β-gal) staining. We observed that the elderly sample showed a markedly higher percentage of β-galactosidase-positive area than the young sample, although the difference did not reach statistical significance (P = 0.05, Figure 1C–1D). Experimental overview (A) Experimental overview of single-cell RNA and spatial transcriptomic sequencing of human scalp samples at different ages. (B) Box plots of hair density, terminal hair density, hair diameter, and vellus hair in the young (n = 37), middle-aged (n = 27), and elderly group (n = 26) of the 90 volunteers. P values were calculated by one-way ANOVA test. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ns, not significant. (C) Comparison of SA-β-gal staining in sections from 1 young and 1 elderly sample. Scale bar: 25 μm. (D) The quantitative analysis of the proportion of positive area in the SA-β-gal staining experiment. n = 3 for each section. Data are presented as mean ± SD. P = 0.05 by Mann-Whitney test. Abbreviations: ANOVA, analysis of variance; SA-β-gal, senescence-associated β-galactosidase; SD, standard deviation. After scRNA sequencing, in total 57,181 high-quality cells without obvious batch effects remained in the subsequent analysis from 11 samples after stringent quality control and filtering (Figure S1A–S1B). Seventeen distinct cell clusters were identified following dimensionality reduction and clustering using the Seurat package, as visualized by Uniform Manifold Approximation and Projection (UMAP) (Figure 2A). The relative abundance of each cell type in each sample is shown in Figure 2B. These clusters were annotated and classified into specific cell types based on canonical marker genes (Figures 2C, S1C, and S2): interfollicular epidermis (IFE) basal cells (IB) (COL17A1, KRT14), IFE spinous cells (IS) (KRT10, KRT1), IFE granular cells (IG) (CALML5, DSC1, SPINK5), mitotic cells (Mi) (MKI67, TUBA1B, HMGB2), bulge cells (Bu) (CXCL14, KRT15), outer root sheath cells (ORS) (KRT16, GJB6, KRT17), inner root sheath cells (IRS) (KRT28, KRT71), matrix/cortex cells (MC) (KRT35, KRT85), endothelial cells (En) (PECAM1, SPARCL1), melanocytes (Me) (DCT, MLANA), sebaceous/eccrine cells (SE) (DCD, MUCL1), gland duct cells (GD) (KRT7), muscle cells (Mu) (TAGLN, TPM1, MYL9), fibroblast (Fib) (APOD, CFD, THY1, COL1A1), T cells (TC) (CD3D), dendritic/macrophages (DM) (CD74, CD1C, ITGAM, ITGAX, MRC1, CD86), and mast cells (Ma) (TPSAB1, KIT). The spatial transcriptomic data confirmed the annotation of cell types (Figures 2D and S3). The top 30 highly expressed genes identified in these clusters revealed unique transcriptional features of each cell type (Figure 2E), and the enriched pathways were associated with their distinct cell functions (Figure S1D). For example, bulge cells are mainly enriched with genes involved in extracellular structure organization and extracellular matrix organization. Single-cell and spatial transcriptomic data of human scalp samples (A) UMAP plot of cells from 11 human scalp samples after quality control. Cells are colored by type and annotated at right. The acronym of each cell type is defined in parentheses. (B) Proportion of each cell cluster in 11 samples with the same color code as in (A). (C) Dot plot showing the expression of marker genes for each cell type. (D) Nucleic acid dye staining of scalp tissue sections (left) and spatial distribution of cell types in frozen sections by computational devolution and projection from one scalp sample. Scale bar, 250 μm. (E) Heatmap showing the gene expression levels of each cell type. The color key from blue to red indicates low to high gene expression levels. The top bar indicates the cell type with the same color code as in (A). Abbreviations: UMAP, Uniform Manifold Approximation and Projection; IFE, interfollicular epidermis; ORS, outer root sheath cells; IRS, inner root sheath cells. Identification of Mitotic Cell Subpopulations and Their Transitional States via Spatial Transcriptomics and Trajectory Analysis Single-cell clustering identified a group of cells with high expression of cell cycle-related genes (Figure 2C). When projected onto the spatial transcriptomic data, these cells were distributed across the IFE, bulge-ORS, and the hair follicle matrix. Therefore, these cells were extracted and analyzed separately by dimensionality reduction and clustering. UMAP clustering revealed that these cells clearly segregated into three distinct clusters, each of which was characterized by high expression of known marker genes associated with epidermal cells (IS and IB) (KRT1, KRT10, COL17A1, KRT14), ORS (KRT17, KRT16, GJB6), and matrix cells (KRT35, KRT85) (Figures 3A–3C and S4A). These three subclusters were annotated as IFE mitotic cells (IM), ORS mitotic cells (OM), and matrix mitotic cells (MM), respectively. These annotations were confirmed in spatial enhanced-resolution omics sequencing (stereo-seq) data and the proliferation marker protein Ki-67 (MKI67) immunostaining (Figure 3D–3E). Heterogeneity of the mitotic cells (A) UMAP plot showing subpopulations of mitotic cells. Cells are colored by type and annotated at right. The acronyms of each cell type are defined in parentheses. (B) Dot plot showing the expression level of selected marker genes for each subpopulation. (C) Proportion of each cell subpopulation in each sample with the same color code as in (A). (D) Spatial mapping of each subpopulation from the spatial transcriptomic data. Scale bar, 250 μm. (E) Immunofluorescence staining of MKI67 in a section from a 52-year-old scalp tissue sample. The first panel (Mi) shows the whole tissue section at a scale bar of 250 μm, while the following three panels show the magnified specific regions (IM, OM, MM) at a scale bar of 50 μm. Red dots indicated MKI67-positive cells. (F) Heatmap showing the gene expression levels of each cell type. The color key from blue to red indicates low to high gene expression levels. The top bar indicates the cell type with the same color code in (A). (G) GO enrichment analysis of signature genes for each subpopulation. Abbreviations: UMAP, Uniform Manifold Approximation and Projection; IFE, interfollicular epidermis; ORS, outer root sheath cells; Mi, mitotic cells; GO, Gene Ontology. Gene Ontology (GO) analysis of the top 30 genes revealed the transcriptional features across the three subpopulations (Figure 3F–3G). In OM cells, highly expressed genes were enriched in biological functions such as actin filament organization, axon development, cell junction assembly, and wound healing. IM cells exhibit enrichment in pathways related to antigen processing and presentation as well as epidermis and skin development. MM cells showed strong enrichment in RNA metabolism and splicing-related pathways, including regulation of mRNA metabolic processes, RNA splicing, RNA splicing via transesterification reactions with bulged adenosine as nucleophile, and mRNA splicing via spliceosome. To investigate the differentiation trajectories of IFE cells and hair follicle keratinocyte cells, we performed pseudotime analysis using Monocle3 and Monocle2. We selected IB, IM, Bu, and MC cells as root populations to construct semi-supervised pseudotime trajectories. The self-renewal of basal cells in the IFE and the classic model of upward epidermal differentiation from basal to spinous and granular cells [16] were recapitulated (Figures 4A and S4B), consistent with previous findings [7,17,18]. Along the pseudotime axis, keratinocytes showed a loss of basal cell marker expression and an increase in suprabasal marker gene expression. The pseudotime trajectory curve and heatmap demonstrated that the marker gene COL17A1 for the IB cells was highly expressed at the early stage and gradually declined, while the marker genes KRT1 and CALML5 for the IS and IG cells markedly increased during the mid to late stages, respectively (Figure 4B–4C). Pseudotime trajectory analysis of keratinocytes (A, D, and G) UMAP feature plots showing pseudotime trajectories of IFE (A), hair follicle stem cells (D), and hair follicle keratinocytes at anagen (G). Cells are colored by cell type (top) and inferred pseudotime (bottom). Pseudotime is depicted from dark purple to light yellow. (B, E and H) Expression levels of selected marker genes along the pseudotime trajectory. Dots are colored by cell type. (C, F, and I) Heatmaps showing expression patterns of signature genes across IFE subpopulations along the pseudotime trajectory. Abbreviations: UMAP, Uniform Manifold Approximation and Projection; IFE, interfollicular epidermis; IM, IFE mitotic cells; IB, IFE basal cells; IS, IFE spinous cells; IG, IFE granular cells; Bu, bulge cells; ORS, outer root sheath cells; OM, ORS mitotic cells; MM, matrix mitotic cells; IRS, inner root sheath cells; MC, matrix/cortex cells. As shown in Figures 4D and S4C, Bu cells differentiated into OM cells, which subsequently developed into ORS cells. The pseudotime trajectory curve and heatmap showed that the Bu cell marker gene CXCL14 was highly expressed in the early stage and gradually declined over time, while marker genes—TUBA1B and MKI67—for the OM cells peaked at the mid-stage and the ORS cell marker gene KRT16 showed increased expression during the later stage (Figure 4E–4F). Collectively, these findings provide transcriptomic evidence that bulge descendants transition into transit-amplifying cells that retain partial stemness and migrate downward to populate the ORS layer [19]. Previous scRNA sequencing and assay for transposase-accessible chromatin using sequencing (ATAC-seq) data analysis of murine skin supported the differentiation trajectory of matrix cells into IRS and hair shaft (cuticle/cortex/medulla) [20,21]. We inferred the differentiation trajectory of these cells, including the MM subpopulation, in our dataset and validated that MC cells follow differentiation trajectories toward MM and IRS cells (Figures 4G–4I and S4D). Through the integration of pseudotime trajectory analysis and spatial transcriptomic mapping, we validated the transition state of the IB subpopulation, consistent with prior research [7,17,22]. Additionally, we elucidated the transition state of bulge-ORS and matrix cells. These insights substantially contributed to our knowledge of keratinocyte development and differentiation during the anagen phase of hair follicles. The Heterogeneity of Fibroblast Cells Fibroblasts in human skin are morphologically and functionally heterogeneous, playing key roles in skin development [23], the hair cycle [23], wound healing [23,24], and skin remodeling [24]. Single-cell transcriptional analysis has also revealed significant heterogeneity within these cells [12,25]. To investigate the heterogeneity in our data, dimensionality reduction and clustering on the Fib population were performed. The results showed that these cells were segregated into three major clusters (Figures 5A‒5B and S4E): dermal papilla cells (DP) (CRABP1, BMP7, RUNX3, ALX4, PTCH1, VCAN, WNT5A), dermal sheath cells (DS) (THY1, COL11A1), and reticular/papillary fibroblasts (fib) (APOD, CFD, APOE, COL1A1). The proportional distribution of each subpopulation within every sample (Figure 5C) revealed that DP cells were not detected in three middle-aged samples. The spatial localization of fib/DP/DS cells on the spatial transcriptomic chip further supported the accuracy of the annotation (Figure 5D). GO functional enrichment analysis of the highly expressed genes in each cell type (Figure 5E) revealed that genes highly expressed in DP cells were enriched in pathways closely related to hair follicle development and regeneration. Notably, these included various aspects of the WNT signaling pathway, such as regulation of WNT signaling, cell–cell signaling by WNT, and the WNT signaling pathway itself [26]. Heterogeneity of fibroblast cells (A) UMAP plot showing subpopulations of Fib cells are colored by type and annotated at right. The acronyms of each cell type were defined in parentheses. (B) Dot plot showing the expression level of representative marker genes for each subpopulation. (C) Proportion of each cell subpopulation in each sample. (D) Spatial mapping of each subpopulation from the spatial transcriptomic data. Scale bar, 250 μm. (E) GO enrichment analysis of the signature genes for each subpopulation. (F) Dot plots showing the expression level of representative marker genes for Fib subpopulations (DP, DS, and fib) in human and mouse scRNA sequencing datasets. “Human” refers to the data from this study and “mouse_5 w” refers to public mouse scRNA sequencing data from Joost et al.'s study ([21], GSE129218). Abbreviations: UMAP, Uniform Manifold Approximation and Projection; Fib, fibroblast; GO, Gene Ontology; scRNA, single-cell RNA. Compared with other cell types, fibroblasts might be the most heterogeneous cells in the skin, showing considerable variation across species and among different scRNA sequencing datasets [27]. Comparing our human dataset (group: human) with public mouse scRNA sequencing data from Joost et al.'s study ([21], GSE129218; group: mouse_5 w), the fib, DS, and DP subpopulations in human skin showed different marker genes (Figure 5F). Human DP cells showed higher expression of BMP7 and RUNX3, whereas mouse DP cells expressed higher levels of Grem2, Enpp2, Corin, Cntn1, Notum, and Nrg2. Human DS cells showed higher expression of COL11A1 and APCDD1, while mouse DS cells expressed high levels of Acta2 and Cavin2. The human fib subpopulation expressed higher levels of CFD, APOD, APOE, and THY1 compared to mouse fib subpopulation. Given the critical roles of DP and DS cells in regulating the hair cycle [28,29], species-specific differences in these cells can significantly impact the translational relevance of findings from mice to humans. Therefore, it is essential to account for the differences in hair follicle cell populations between mice and humans when using mouse models to study hair cycle-related disorders. Age-associated Transcriptional Patterns in Hair Follicle Cells Dynamic changes in gene expression are observed in a tissue- and cell-specific manner as age increases [[30], [31], [32]]. To investigate the mechanism of hair follicular aging, we divided the dataset into young and middle-aged groups and compared the cellular composition and cell-type-specific gene expression profiles between the two groups. Differential cell-type abundance analysis using Milo [33] revealed that ORS, OM, SE, and Bu cells were reduced in the middle-aged group (Figure 6A–6B). Gene set enrichment analysis (GSEA) of established gene sets for senescence, senescence-associated secretory phenotype (SASP), and apoptosis to score the cells revealed a significant difference between the two groups in all cells, including Bu, ORS, OM, and MM, respectively (Figure 6C). Age-associated gene expression patterns (A) Milo beeswarm plot showing differential abundance of cellular neighborhoods in scalp tissue across age groups, annotated by broad cell-type labels. Red and blue points represent neighborhoods significantly enriched in the young and middle-aged groups, respectively. Color intensity reflects the level of statistical significance. (B) Bar plot showing the proportional distribution of Mi subpopulations in young and middle-aged groups. (C) Left panel: Box plots of gene set enrichment scores for senescence, SASP, and apoptosis in young and middle-aged cells. Right panel: Gene set enrichment scores for these senescence, SASP, and apoptosis across different subpopulations in young and middle-aged groups. Statistical significance was evaluated using a two-sided Wilcoxon rank-sum test between the two groups within each subpopulation, with outliers removed based on the 1.5× IQR rule. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ns, not significant. (D) Volcano plot of DEGs between middle-aged and young groups (|avg_logFC| > 0.25, adjusted P value < 0.05). Red and blue points indicate genes upregulated and downregulated in the middle-aged group, respectively. (E–F) GO enrichment bubble plots of upregulated genes (E) and downregulated genes (F) in hair follicular keratinocytes between the middle-aged and young groups. Abbreviations: Ma, mast cells; DM, dendritic/macrophages; TC, T cells; GD, gland duct; Mu, muscle cells; SE, sebaceous/eccrine cells; Me, melanocytes; En, endothelial cells; MC, matrix/cortex cells; IRS, inner root sheath cells; ORS, outer root sheath cells; Bu, bulge cells; IG, interfollicular epidermis (IFE) granular cells; IS, IFE spinous cells; IB, IFE basal cells; DS, dermal sheath cells; DP, dermal papilla cells; fib, reticular/papillary fibroblasts; MM, matrix mitotic cells; OM, ORS mitotic cells; IM, IFE mitotic cells; SASP, senescence-associated secretory phenotype; IQR, interquartile range; DEG, differentially expressed gene; GO, Gene Ontology. We identified 2649 differentially expressed genes (DEGs) associated with the aging process (average logarithmic fold change: |avg_logFC| > 0.25 and adjusted P values: P_val_adj < 0.05), including 1590 upregulated DEGs and 1059 downregulated DEGs in the middle-aged group (Figure 6D and Table S1). The GO enrichment analysis revealed that upregulated genes in the middle-aged group (Figure 6E) were enriched in pathways related to stress response (such as response to steroid hormone, response to corticosterone, and integrated stress response signaling), immune and inflammatory responses (cellular response to glucocorticoid stimulus and mononuclear cell differentiation), and aging (regulation of telomere maintenance, positive regulation of cell aging, and DNA damage response). Downregulated genes in the middle-aged group (Figure 6F) were mainly associated with nervous system development (regulation of nervous system development and regulation of neurogenesis), intercellular communication (response to fibroblast growth factor), and stem cell development (stem cell fate commitment). Several genes reported to be associated with skin aging exhibited distinct expression patterns across different cell types. HES1 expression was decreased in the fibroblasts in eyelid skin aging [7]; however, its expression was upregulated in the IFE cells and hair follicle stem cells and significantly downregulated in sebaceous gland cells in the middle-aged group in our data (Figure 7A). KLF6 expression, which was downregulated in basal cells in eyelid aging, was found to be upregulated in the bulge, ORS, and OM cells (Figure 7B). ATF3 expression, which was decreased in basal cells of the back and arm skin during aging [34], was found to be upregulated in the bulge, ORS, and OM cells (Figure 7C). Significant genes and regulons between middle-aged and young individuals (A–D) Box plots showing the expression levels of HES1 (A), KLF6 (B), ATF3 (C), and DCT (D) in middle-aged (orange) and young (blue) groups across different cell types. Statistical significance is assessed using a two-sided Wilcoxon rank-sum test between the two groups within each cell type. Outliers were removed based on the 1.5× IQR rule. P values are not adjusted for multiple tests. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. (E) Rank of transcription factors in middle-aged (top) and young (bottom) groups based on RSS. Higher RSS values indicate stronger specificity of regulon activity in the group. The top 10 transcription factors with the highest RSS in each group are highlighted in red. (F–G) Box plots showing the expression levels of JUN (F) and FOS (G) genes in middle-aged (orange) and young (blue) groups. Statistical testing and outlier removal are performed as in panels (A–D). (H) Regulatory networks of DEGs associated with aging under the regulation of AP-1 complex. The left panel shows the networks in the Bu, ORS, and OM subpopulations, while the right panel shows those in the IB, IS, IG, and IM subpopulations. Node size is positively correlated with the number of edges (i.e., connectivity). Orange nodes represent components of the AP-1 complex (JUN, JUNB, JUND, FOS, and FOSB), and blue nodes represent DEGs associated with aging. Abbreviations: Ma, mast cells; DM, dendritic/macrophages; TC, T cells; GD, gland duct; Mu, muscle cells; SE, sebaceous/eccrine cells; Me, melanocytes; En, endothelial cells; MC, matrix/cortex cells; IRS, inner root sheath cells; ORS, outer root sheath cells; Bu, bulge cells; IG, interfollicular epidermis (IFE) granular cells; IS, IFE spinous cells; IB, IFE basal cells; DS, dermal sheath cells; DP, dermal papilla cells; fib, reticular/papillary fibroblasts; MM, matrix mitotic cells; OM, ORS mitotic cells; IM, IFE mitotic cells; IQR, interquartile range; RSS, regulon specificity scores; DEG, differentially expressed gene. The DCT expression in melanocytes was significantly elevated in middle-aged samples compared to young samples (Figure 7D). The DCT gene plays a key role in the melanin synthesis pathway and is involved in the melanogenesis process. We hypothesized that this elevated level of DCT expression may be the evidence of overactive melanin synthesis in middle-aged people caused by chronic low-grade inflammation associated with aging (i.e., inflammaging) [[35], [36], [37], [38]]. Inflammation induces the expression of several promelanin-synthesizing cytokines and enzymes, such as tumor necrosis factor-alpha (TNF-α) and interleukin-6 (IL-6) [39]. These inflammatory mediators enhance melanocyte activity and upregulate the expression of melanin synthesis-related genes, leading to excessive melanin accumulation in melanocytes [40]. Then, single-cell regulatory network inference and clustering (SCENIC) analysis was performed to assess the activity of key transcription factors between the two groups by calculating the regulon specificity score (RSS) (Figure 7E). The results showed that members of the activator protein 1 (AP-1) complex (JUND, JUN, FOS, FOSB, JUNB) as well as KLF4 were activated in the middle-aged group, while transcription factors associated with cell proliferation and differentiation, metabolism, autophagy, epidermal barrier function, and maintenance of stemness (such as ILF2, ENO1, RAB7A, GRHL1, and SOX9) were activated in the young group. Differential expression analysis also revealed significant upregulation of JUN, FOS, JUNB, and FOSB in hair follicle stem cells (Bu, ORS, and OM) and epidermal cells (IM, IB, IS, and IG) (Figure 7F–7G) in the middle-aged group. AP-1, a MAPK-responsive transcription factor complex composed of proteins from the JUN, FOS, ATF, and MAF families, orchestrates extensive chromatin remodeling during aging by activating late-life enhancers and decommissioning early-life, cell-identity regulatory elements, thereby reprogramming the transcriptome toward a senescence-associated gene expression profile [41,42]. Elevated AP-1 levels, especially members such as JUN and FOS, are a conserved signature of immune aging, contributing to age-related inflammation (inflammaging) [43,44]. To understand the regulatory mechanisms involving the AP-1 complex in aging skin, we constructed a regulatory network (Figure 7H, Tables S2‒S3) by focusing on the intersection of age-associated DEGs and AP-1 target genes. Here, we specifically demonstrated the interplay between AP-1 activation and the upregulation of MYC, DUSP1, NR4A1, NFKBIA, and SOCS3. AP-1 regulates MYC expression by binding to AP-1 response elements within its promoter, and the pattern of AP-1 activation and its subunit composition determines whether MYC is upregulated or repressed [45]. Overexpression of MYC drives stem cells from a quiescent or self-renewing state into active proliferation and differentiation, depleting the stem-cell pool and reducing long-term tissue regenerative capacity [46]. The SOCS3 gene, whose promoter contains a functional AP-1 site [47], encodes a protein involved in the negative regulation of cytokines that signal through the JAK/STAT pathway. In aged human muscle stem cells, elevated SOCS3 protein levels induced by IL-6 have been reported to block STAT3 mitogenic signaling and blunt regenerative capacity [48]. Multiple studies demonstrate that AP-1 activation is accompanied by the coordinated upregulation of immediate-early genes, such as DUSP1, EGR1, and NR4A1, forming a conserved gene module often coexpressed in various contexts, including cancer cells and senescence [49,50]. AP-1 can induce DUSP1 expression to forge a negative feedback loop that limits MAPK overactivation through the dephosphorylation of MAPK1/ERK2. The NR4A1 gene encodes a nuclear transcription factor and inhibits nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) signaling. The NFKBIA gene encodes NF-κB inhibitor alpha (IκBα), a member of the NF-κB inhibitor family. In mouse fibroblasts, AP-1 components c-Jun and JunD bind to the Nfkbia promoter exclusively in the presence of NF-κB p65 [51]. A study on mouse skin demonstrated that following irradiation, the expression of Nfkbia was upregulated during the rapid early phase of NF-κB activation, yet became repressed in the subsequent second phase associated with senescence [52]. AP-1 activation and the coregulated genes coordinate the activation of both proliferative and anti-inflammatory components, reflecting the cell's dynamic strategy to adapt to stress, maintain homeostasis, and mitigate potential damage. Age-associated Differences in Intercellular Communication Altered intercellular communication is a hallmark of aging [53]. To explore alterations in intercellular communication associated with aging, we analyzed the interactions among cell populations using CellChat version 2 [54] between the middle-aged and young groups. We found that intercellular communications in the middle-aged group were less frequent and attenuated compared to those in the young group (Figure 8A), a phenomenon that was also observed in a study on skin fibroblast cells [12]. The results showed that DP, DS, and fib subpopulations exhibited stronger outgoing interactions with other cell subpopulations in the young group, whereas DP cells lost their ability to communicate with other cells in the middle-aged group (Figure 8B). The primary receptor cells in both groups were hair follicle stem cells, including Bu, ORS, and OM (Figure 8B). Altered intercellular communication between middle-aged and young cells (A) Total number of interactions (left) and overall interaction strength (right) of intracellular and intercellular communication networks in middle-aged and young groups. (B) Overall incoming and outgoing communication strength based on receptor and ligand expression in middle-aged (left) and young (right) groups. Point size is proportional to the number of cells in each cluster. (C) Key signaling pathways ranked by differences in overall information flow between middle-aged and young groups. Pathways enriched in the middle-aged group are shown in orange; those enriched in the young group are shown in blue. (D, F, and H) Cell‒cell communication networks of the BMP (D), WNT (F), and ncWNT (H) signaling pathways in middle-aged (left) and young (right) groups. Nodes represent cell types; edges indicate interactions, with edge width proportional to interaction probability, reflecting signaling strength. (E, G, and I) Violin plots showing the expression of genes involved in the BMP (E), WNT (G), and ncWNT (I) signaling pathways in middle-aged (orange) and young (blue) groups. Plots were generated using CellChat to visualize age-related differences in signaling gene expression. Abbreviations: Ma, mast cells; DM, dendritic/macrophages; TC, T cells; GD, gland duct; Mu, muscle cells; SE, sebaceous/eccrine cells; Me, melanocytes; En, endothelial cells; MC, matrix/cortex cells; IRS, inner root sheath cells; ORS, outer root sheath cells; Bu, bulge cells; IG, interfollicular epidermis (IFE) granular cells; IS, IFE spinous cells; IB, IFE basal cells; DS, dermal sheath cells; DP, dermal papilla cells; fib, reticular/papillary fibroblasts; MM, matrix mitotic cells; OM, ORS mitotic cells; IM, IFE mitotic cells; ncWNT, non-canonical WNT. Information flow analysis by accumulating the potential cell‒cell communications of all receptor‒ligand pairs within the same cell type revealed that 108 signaling pathways were activated in both the middle-aged and young groups (Figure 8C, Table S4). The TNF and transforming growth factor-beta (TGF-β) signaling pathways, both of which were related to skin photoaging [[55], [56], [57]], were activated exclusively in the middle-aged scalp cells. The glutamate signaling pathway, which mediates keratinocyte and hair follicle proliferation as well as stem cell fate reversibility and maintenance of the HFSC regeneration [58], was activated exclusively in the young group. The BMP and WNT signaling pathways, as well as their dynamic interplay between DP and hair follicle cells, play pivotal roles in the hair follicle stem cell activity and hemostasis, hair cycling, and regeneration [3]. We observed that the BMP signaling in the young group was active but almost lost in the middle-aged group (Figure 8D–8E), and the WNT signaling was weak in both groups (Figure 8F–8G). Notably, the expression of ligand genes BMP4, BMP5, and BMP7 decreased in the middle-aged samples (Figure 8E). In addition, we observed that the non-canonical WNT (ncWNT) signaling was very weak in the middle-aged group, while it was intensive in the young group between DP and hair follicle cells, including ORS, Bu, IRS, MC, OM, and MM (Figure 8H). The expression levels of the ligand gene WNT5A and the receptor genes FZD6 and FZD7 decreased in the DP cells and hair follicle cells, respectively, in the middle-aged group (Figure 8I). Studies in mice have suggested that Wnt5a suppresses canonical WNT signaling and inhibits the telogen-to-anagen transition [59,60]. Further investigations into the dynamic interplay among BMP, WNT, and ncWNT signaling in the hair cycle will be beneficial for understanding their roles in the aging process. DISCUSSION We studied age-related transcriptional patterns in 11 scalp samples using scRNA sequencing combined with spatial transcriptomics of one sample. Since only one of the subjects was over the age of 50, the corresponding data were excluded from the age-related analysis. Analysis of eyelid skin revealed that the middle-aged group was transcriptionally close to the elderly group as shown by principal component analysis [7]. This suggests that the middle-aged group likely captures the early transcriptional changes associated with scalp aging. The integration of single-cell and spatial transcriptomic technologies facilitated mapping of hair follicle subpopulations in reference to the morphological structures of skin and its appendages. We identified three subpopulations of mitotic keratinocytes in the scRNA sequencing data and mapped them in the IFE, bulge-ORS, and hair matrix by computational deconvolution and projection on the spatial transcriptomic chip. The presence of mitotic cells has long been observed in human skin keratinocytes [7,18,22]. Our data provided transcriptional evidence of the existence of transient amplifying bulge descendants (OM cells) migrating downward in human hair follicles [19]. We also identified three subpopulations of Fib and classified them as DP/DS/fib, which was also confirmed by their locations on the spatial transcriptomic chip. In Shin et al.'s work [61], a group of progenitor cells that expressed cell cycle-related genes as well as DP and DS marker genes were found in mouse skin through scRNA sequencing of fluorescence-activated cell sorting (FACS)-sorted mesenchymal cells. However, we did not identify identical cell clusters in our human data. We analyzed the differences in transcriptional markers of DP/DS cells between humans and mice and emphasized that these differences should be taken into account when using mouse models to study hair disorders, given the critical role of DP/DS cells in regulating the hair cycle. In the age-associated analysis of single-cell transcriptomics, AP-1 activation emerged as a hallmark of middle-aged hair follicle and epidermal cells, consistent with its known role in chromatin remodeling and senescence-associated transcriptional reprogramming. The downstream targets of AP-1—such as MYC, SOCS3, DUSP1, NR4A1, and NFKBIA—form an intricate regulatory network that influences cell cycle progression, inflammatory responses, and stem cell depletion. This coordinated regulation reflects a dynamic cellular strategy in aging skin—balancing stem cell activation and stress adaptation while restraining excessive proliferative and inflammatory signaling to maintain tissue homeostasis. In addition, DCT was upregulated in melanocytes in the middle-age group, suggesting overactive melanin synthesis caused by inflammaging. Future studies leveraging in vivo and in vitro human hair follicle models are essential to elucidate the causal role of this AP-1-centered network and to evaluate whether targeting AP-1 or its downstream pathways could delay stem cell depletion and offer novel therapeutic avenues for age-related hair loss and graying. It has long been illustrated that the anatomic location of skin has different characteristics, such as the well-known androgen paradox [62], which means that androgens stimulate beard growth but suppress hair growth in androgenetic alopecia. In the current study, the expression patterns of several genes (HES1, KLF6, and ATF3) exhibit distinct trends across different cell types from scalp tissues, compared to findings from studies focusing on other anatomical sites such as the face, arm, and back. These observations highlight the need for caution when integrating single-cell transcriptomic data from different anatomical sites of the skin, as gene expression patterns may vary substantially. Intercellular communication analysis revealed reduced BMP and ncWNT signaling, as well as almost absent WNT signaling in the middle-aged group, suggesting that these changes may underlie aging-related disruption of hair cycles. It has been demonstrated that interaction between BMP and WNT signaling plays a pivotal role in regulating the hair cycle, particularly in the transition between the anagen and telogen phases [26,[63], [64], [65]]. The BMP signaling is associated with maintaining the telogen phase and keeping hair follicle stem cells quiescent, which is crucial for the cyclical nature of hair growth. In contrast, WNT signaling initiates the anagen phase by triggering the activation and proliferation of hair follicle stem cells. Due to their antagonistic relationship, the balance between BMP and WNT signaling is crucial for preventing premature activation of hair follicle stem cells, thereby ensuring that the hair cycle progresses in a controlled manner. Several reports have indicated that aging disrupted the balance, leading to abnormal stem cell activation and depletion. Aging white adipose tissue in mice exhibited elevated expression of genes involved in BMP signaling, alongside a reduction in WNT signaling activity [66]. Aging HFSCs presented decreased canonical WNT signaling and increased Wnt5a-driven ncWNT signaling, which antagonized canonical WNT signaling in mice, leading to delayed anagen onset and impaired hair regeneration [67]. Further investigation of their dynamic changes in human hair follicles during aging is both necessary and warranted. There are several limitations of the current study. Transcriptomic changes during aging are dynamic, while some genes show continuous up- or down-regulation with age, others exhibit nonlinear patterns that only become apparent upon entering the senescent state [68]. The absence of a dedicated elderly group limited our ability to construct a comprehensive transcriptional trajectory across the entire aging process, preventing the identification of key events occurring in late-stage aging. Additionally, the small sample size restricted the feasibility of stratified analyses based on sex and sampling sites. Hair loss manifests with distinct clinical patterns between women and men. Male-pattern hair loss (MPHL) typically involves bitemporal recession and vertex thinning but rarely affects the occipital scalp, whereas female-pattern hair loss (FPHL) typically presents as diffuse thinning over the crown with preserving the frontal hairline, known as the Ludwig pattern. The sex-linked gene XIST was found to be downregulated across all cell types in the current study (Figure 6D). However, given the limited and uneven gender distribution, our study cannot definitively ascertain how sex influences the aging process. Moreover, the study did not make full use of spatial transcriptomics. As a powerful technology, spatial transcriptomic sequencing offers a unique advantage to capture the spatial organization and physical interplay between cells in their native tissue environment. It allows for a more precise characterization of the stem cell niche, including DP cells and the infiltration of immune cells in aging follicles. By analyzing the signaling molecules and visualizing the physical relationships between these cell types during aging, research can gain a deeper understanding of how functional decline or fluctuations in the abundance of specific cell types impact the regenerative capacity of neighboring stem cells. Future investigations incorporating spatial transcriptomic sequencing with larger sample sizes, balanced sex representation, and inclusion of elderly cohorts—potentially achieved through collaborations and the integration of public datasets—will help overcome these limitations, improve statistical power, and yield valuable insights to inform the development of targeted therapies. In summary, the current study identifies the major cell types within the skin and its appendages, providing spatially resolved locations, and investigates cell-type-specific transcriptional patterns of hair follicle cells associated with the early aging process, without further stratification by sex or scalp regions. Our results provide new insights into understanding the aging mechanism of hair follicles, which could be applied in developing therapies or cosmetic solutions for conditions related to hair follicle aging. MATERIALS AND METHODS Ethical Statement and Study Design All relevant procedures involving human sample collection and experiments presented in this study are compliant with ethical regulations, the principles of the Declaration of Helsinki, and applicable local regulatory requirements and laws. All procedures were conducted under the approval of the Institutional Review Board of BGI (BGI-IRB 22047-T1) and Shanghai Skin Disease Hospital (2023–45 (ke) N2), and informed consent was obtained from all donors. Eligible subjects were described as adults aged 20–59 years with normal, healthy hair and skin. Exclusion criteria included the presence of skin disorders or infectious disease, self-reported pregnancy, and failure to provide informed consent. Eleven adults (four young [1 male and 3 females], six middle-aged [3 males and 3 females], and one elderly individual [1 male]) were included for single-cell and spatial transcriptomic sequencing from March 1 through April 30, 2023, with an average age of 36. Human scalp samples (10 mm × 10 mm) were dissected from volunteers undergoing dermatological surgeries under the guidance of the Department of Dermatologic Surgery, Shanghai Skin Disease Hospital (Shanghai, China). Each tissue was cut into two parts: one (6 mm × 10 mm) was immediately snap-frozen in liquid nitrogen pre-chilled-isopentane in Tissue-Tek OCT (Sakura, Nagano Prefecture, Japan; Cat#4583) and transferred to a −80 °C freezer for storage; the other part (4 mm × 10 mm) was sent to undergo tissue disassociation for single-cell RNA sequencing. In addition, 92 volunteers were also recruited based on the same criteria, including 90 volunteers (37 young [18 males and 19 females], 27 middle-aged [13 males and 14 females], and 26 elderly [10 males and 16 females]) for trichoscopic examination; one young male (30-year-old) for SA-β-gal staining; and one elderly male (52-year-old) for both MKI67 immunofluorescence and SA-β-gal staining. All the sample information is provided in Table S5. Trichoscopic Examination Quantitative trichoscopic parameters (hair density, follicular density, terminal hair ratio, and vellus hair ratio) were measured in 90 volunteers using the Bodystudio ATBM trichoscope (FotoFinder Systems GmbH, Bavaria, Germany). The examination focused on a designated scalp area, where standardized microimages were captured at 50× magnification to evaluate hair and follicular density. Stereo-seq Library Preparation and Sequencing Tissue processing Cryosections of a scalp tissue sample from the young group (the 20-year-old male), embedded in OCT, were cut at a thickness of 10 μm using a Leica CM1950 cryostat (Hessian, Germany). Then, the Stereo-seq library preparation and sequencing were performed as previously described [10]. Briefly, tissue sections were adhered to the stereo-seq chip surface (10 mm × 10 mm) and incubated at 37 °C for 3–5 min. The sections were fixed in methanol and incubated for 40 min at −20 °C before stereo-seq library preparation. Where indicated, the same sections were stained with nucleic acid dye (Thermo Fisher Scientific, Massachusetts, USA; Cat#Q10212), and imaging was performed with a Ti-7 Nikon Eclipse microscope (Tokyo, Japan) before in situ capture at the fluorescein isothiocyanate (FITC) channel. In situ reverse transcription After washed with 0.1× SSC buffer (Thermo Fisher Scientific, Massachusetts, USA; Cat#AM9770) supplemented with 0.05 U/mL RNase inhibitor (NEB, Massachusetts, USA; Cat#M0314L), tissue sections placed on the chip were permeabilized using 0.1% pepsin (Sigma, Missouri, USA; Cat#P7000) in 0.01 M HCl buffer, incubated at 37 °C for 12 min, and then washed again with 0.1× SSC buffer supplemented with 0.05 U/mL RNase inhibitor. The RNA released from the permeabilized tissue and captured by the DNA nanoball (DNB) was reverse transcribed overnight at 42 °C using SuperScript II (Invitrogen, California, USA; Cat#18064-014; 10 U/mL reverse transcriptase, 1 mM dNTPs, 1 M betaine solution PCR reagent, 7.5 mM MgCl2, 5 mM DTT, 2 U/mL RNase inhibitor, 2.5 mM stereo-seq-template-switching oligonucleotide [TSO], and 1× First-Strand buffer). After reverse transcription, tissue sections were washed twice with 0.1× SSC buffer and digested with tissue removal buffer (10 mM Tris-HCl, 25 mM EDTA, 100 mM NaCl, 0.5% SDS) at 37 °C for 30 min The cDNA-containing chips were then subjected to Exonuclease I (NEB, Massachusetts, USA; Cat#M0293L) treatment for 1 h at 37 °C and were finally washed once with 0.1× SSC buffer. Amplification The resulting cDNAs were amplified with KAPA HiFi Hotstart Ready Mix (Roche, Basel-City, Switzerland; Cat#KK2602) with 0.8 mM cDNA-PCR primer. The PCR reactions were conducted as follows: incubation at 95 °C for 5 min, 15 cycles at 98 °C for 20 s, 58 °C for 20 s, 72 °C for 3 min, and a final incubation at 72 °C for 5 min. Library construction and sequencing The concentrations of the resulting PCR products were quantified by QubitTM dsDNA Assay Kit (Thermo Fisher Scientific, Massachusetts, USA; Cat#Q32854). A total of 20 ng of DNA was then fragmented with in-house Tn5 transposase at 55 °C for 10 min, after which the reactions was stopped by the addition of 0.02% SDS and gently mixing at 37 °C for 5 min. After fragmentation, the products were amplified as follows: 25 mL of the fragmentation product, 1× KAPA HiFi Hotstart Ready Mix, and 0.3 mM Stereo-seq-Library-F primer, and 0.3 mM Stereo-seq-Library-R primer were combined in a total volume of 100 ml with the addition of nuclease-free H2O. The reaction was then run as follows: 1 cycle of 95 °C for 5 min, 13 cycles of 98 °C for 20 s, 58 °C for 20 s, 72 °C for 30 s, and 1 cycle at 72 °C for 5 min. The PCR products were purified using the AMPure XP Beads (0.6× and 0.15×), used for DNB generation, and finally sequenced on the MGI DNBSEQ-Tx sequencer (MGI, Guangdong, China) [10]. Droplet-based scRNA Sequencing Library Preparation and Sequencing Tissue dissociation The 11 scalp samples underwent removal of subcutaneous and adipose tissues simultaneously were cut into 2 mm × 2 mm pieces, and incubated in 0.25% trypsin-EDTA (Gibco, California, USA; Cat#25200-072) for 60 min at 37 °C. The resulting cell suspension was filtered through 70 μm and 40 μm strainers to remove hair and cornified layers for scRNA sequencing library preparation. Library preparation The scRNA sequencing libraries were prepared using the DNBelab C Series Single-Cell Library Preparation Kit (MGI, Guangdong, China; Cat#1000021082) and the MGIEasy Circularize Kit (MGI, Guangdong, China; Cat#1000005259). The scRNA sequencing libraries were generated through droplet encapsulation, emulsion breaking, mRNA capture bead collection, reverse transcription, cDNA amplification, and purification, transforming the single-cell suspension into barcoded scRNA-seq libraries. Index sequencing libraries were constructed according to the manufacturer's protocol. The concentration and quality of the sequencing libraries were quantified and assessed using the Qubit dsDNA Assay Kit (Thermo Fisher Scientific, Massachusetts, USA; Cat#Q23854) and the Qubit ssDNA Assay Kit (Thermo Fisher Scientific, Massachusetts, USA; Cat#Q10212). Sequencing and obtaining the expression matrix The scRNA sequencing libraries were sequenced on the DNBSEQ-T7 and MGISEQ-2000 sequencers (MGI, Guangdong, China). The reference genome used was GRCh38. Firstly, the database was constructed using DNBC4 tools (version 2.0.7); then the reads are aligned to the constructed reference genome using the default parameters of the DNBC4 tools software (version 2.0.7) to obtain the final expression matrix. Data Analysis Public mouse data The mouse single-cell transcriptomic expression matrix used to analyze cross-species fibroblast heterogeneity was obtained from the Gene Expression Omnibus (GEO) database (accession number GSE129218 [21]). Only three samples from 5-week-old mice at the anagen stage were included in the subsequent analyses. Stereo-seq raw data processing Raw data from the DNBSEQ-Tx sequencers were processed using Stereo-seq Analysis Workflow (SAW, https://github.com/STOmics/SAW). Firstly, coordinate ID (CID) sequences from the first reads were mapped to the coordinates of the chip obtained during the first round of sequencing, permitting a 1-base mismatch. Reads containing molecule identity (MID) with either N bases or more than 2 bases of low quality (e.g. score lower than 10) were excluded. The CIDs and MIDs were added to each read header. Reads were then mapped to the GRCh38 genome and annotated. Reads with a mapping quality (MAPQ) value < 10 were excluded [10]. Next, a custom R function to perform binning on the GEM file output from SAW (bin size = 20), dividing the original DNB units into 20 × 20 bin regions. The spatial location of each bin is assigned based on the coordinates (spatial_x, spatial_y) of the bottom-left DNB. The function simultaneously aggregated expression values, normalized spatial coordinates, and constructs a sparse matrix, ultimately returning a Seurat object. Quality control Using the R package SoupX [69] (version 1.6.2) in conjunction with the Seurat [70] (version 4.3.0.1) and sctransform [71] (version 0.3.5) packages, environmental RNA contamination was removed using default parameters. Initially, the autoEstCont function was employed to automatically calculate contamination scores, followed by utilizing the adjustCounts function to eliminate background contamination within the count matrix. Subsequently, the R package DoubletFinder [72] (version 2.0.3) was used for doublet detection, with parameters set as PCs = 1:30, pN = 0.25, and pK using the maximum PK value identified with find.pK, alongside nExp = nExp_poi, reuse.pANN = FALSE, sct = TRUE. Preprocessing quality control was conducted for cells and genes based on a minimum expression threshold of 200 genes and at least three cells expressing each gene (min.features = 200, min.cells = 3). For human data, additional quality control was performed to detect a minimum of 1000 and a maximum of 6000 genes per cell, with a maximum RNA molecule count of 50,000 per cell, and mitochondrial gene counts not exceeding 10% of the total gene count per cell. For the mouse data, additional quality control was performed based on detecting a minimum of 200 and a maximum of 7500 genes per cell, with a maximum RNA molecular count of 10,000 per cell, and mitochondrial gene counts not exceeding 25% of the total gene count per cell. Dimensionality reduction, clustering, and cell type annotation Further analysis was carried out using the R package Seurat. All mitochondrial genes were removed before dimensionality reduction and clustering. Normalization was performed for each sample using the “SCTransform” function. Highly variable genes were selected using the “SelectIntegrationFeatures” function. The integrated data were first subjected to linear dimensionality reduction using the “RunPCA function”, followed by batch effect removal using the “RunHarmony” function. Nonlinear dimensionality reduction was performed using the “RunUMAP” function from the Seurat package. Subsequently, cell similarities were computed, and a cell neighbor graph was constructed using the “FindNeighbors” function based on the reduced-dimensional data. Cell clustering was performed using the “FindClusters” function based on the constructed shared nearest neighbor (SNN) graph and the Louvain algorithm. The resolution was set to 2, with other parameters set to default values. When subclustering cell subpopulations, the optimal number of principal components for dimensionality reduction was determined based on the following criteria: the cumulative contribution of principal components exceeded 90%, the individual contribution of the second principal component to the variance was less than 5%, and the difference between consecutive principal components was less than 0.1%. Marker genes for each cluster were identified using the “FindAllMarkers” function and the Wilcoxon rank-sum test. Only differentially expressed genes meeting the criteria of |avg_logFC| > 0.25 and P_val_adj <0.05 were considered marker genes. Each cluster was annotated based on an extensive literature review and specific gene expression patterns identified through targeted gene expression analysis. To analyze cross-species heterogeneity of fibroblast cells, the DP, DS, and fib subpopulations from the human dataset and the public mouse dataset were merged using the Seurat “merge” function, and gene expression patterns were visualized using the “DotPlot” function. Annotating spatiotemporal chip cell types We used the R package spacexr [73] (version 2.2.1) to perform robust cell type decomposition (RCTD) deconvolution annotation. The RCTD learns cell type profiles from the scRNA sequencing dataset and uses these profiles to label the spatial transcriptomics pixels as cell types. First, the “Reference” function was used to create a single-cell reference based on annotated single-cell data. During this process, unnormalized raw data counts and unique molecular identifiers (UMIs) were utilized. Next, the “SpatialRNA” function was used to create a SpatialRNA object, where coordinates, counts, and nUMI were all raw, unnormalized data. When creating the RCTD object, the “doublet” model was selected. Gene set analysis We used the “AddModuleScore” function from the Seurat package to score all cells based on the average expression of genes within the corresponding gene sets of senescence (GSEA: M9143), SASP (GSEA: M48334), and apoptosis (KEGG: hsa04210) (full lists in Table S5) retrieved from the Molecular Signatures Database (MSigDB) and Kyoto Encyclopedia of Genes and Genomes (KEGG). Differential cell type abundance analysis We used the “miloR” R package (version 2.2.0) [33] to perform k-nearest neighbor graph-based differential cell type abundance testing between the middle-aged and young groups. To apply miloR to our integrated scRNA sequencing data, we first created a “SingleCellExperiment” R object from the count matrix of our Seurat project and then used the Seurat dimensionality reduction as the reduced.dim input for miloR in the “buildGraph” function. For the “buildGraph” function, we set k = 30 and d = 30 and used prop = 0.05 for the “makeNhoods” function. We then counted the cells in neighborhoods using the “countCells” function. Next, we used the “testNhoods” function to calculate spatial false discovery rate (FDR) correction with k-distance weighting, specifying the parameters model.contrasts = middle-young, fdr.weighting = graph-overlap, and norm.method = TMM. Finally, we visualized the differentially abundant cell neighborhoods with a SpatialFDR of < 0.9 using the “plotNhoodGraphDA” function. GO and KEGG enrichment analysis We used the “compareCluster” function in the clusterProfiler [74] package (version 4.6.2) to compute DEGs for each cell type, as well as GO terms and KEGG pathways associated with age-related DEGs within each cell type. We retained only items with a q value < 0.05. Visualization was performed using the ggplot2 [75] R package (version 3.4.2). Pseudotime trajectory analysis We utilized the R packages Monocle2 [76] (version 2.28.0) and Monocle3 [77] (version 1.3.1) to reconstruct the developmental trajectories of epidermal cells and hair follicle cells. Data normalization was completed using the sctransform package, followed by dimensionality reduction and clustering conducted using the Seurat package. After converting the Seurat object to a Monocle3 cell_data_set object, we followed the official Monocle3 tutorial to construct the trajectory. IB, IM, Bu, and MC cells were defined as the root state parameter and aligned using the “ordercells” function, respectively. Dimensionality reduction was then applied using “UMAP”, and the “plot_cell_trajectory” visualization function was used to plot each group along the same pseudotime trajectory. Transcription regulatory network analysis The PySCENIC [78] (version 0.12.1) workflow, integrating GRNBoost2 [79] and cisTarget [80], was applied to infer transcriptional regulatory networks based on transcription factor (TF) annotations from the hg38 genome. GRNBoost2 was used to construct coexpression networks and identify candidate TF-target interactions, followed by motif enrichment analysis using cisTarget to define high-confidence regulons. AUCell was applied to quantify regulon activity at the single-cell level, and regulon specificity scores (RSSs) were computed to assess differential TF activity across age groups. Finally, we intersected the SCENIC-predicted target genes with age-associated DEGs identified in the previous analysis to construct the transcriptional regulatory networks of AP-1 complex. The networks were visualized using Cytoscape [81] (version 3.10.3). Cell‒cell communication analysis To assess intercellular communication among different cell types, we utilized the R package Cellchat [54] (version 2.1.0) to infer cell‒cell communication networks from single-cell transcriptomic data. We employed the “identifyOverExpressedGenes” and “identifyOverExpressedInteractions” functions in the Cellchat package to calculate overexpressed genes and receptors, respectively. The probability of cell‒cell interactions was inferred based on the expression values of these genes. We utilized the “triMean” method to calculate the average gene expression for each cell. The minimum number of cells required in each cell group for cell‒cell communication was 10. MKI67 Immunofluorescence Staining Cryosections of scalp tissue from a 52-year-old male (Table S5) were fixed in 4% paraformaldehyde at 4 °C for 15 min, followed by washing with 1× washing buffer twice. Then 200 μL of PBS containing 1% fetal bovine serum (FBS) was added for blocking at 4 °C for 1 h. After that, 200 μL of working solution containing 0.1% Triton X-100 (Invitrogen, California, USA; Cat#HFH10) and 1% MKI67 antibody (Invitrogen, California, USA; Cat#MA5-14520) was added for overnight incubation at 4 °C. The sections were then washed three times with PBS (Sangon Biotech, Shanghai, China; Cat#E607008-0500), incubated with secondary antibody (Abcam, Cambridgeshire, UK; Cat#ab150075; 1:500 dilution) at room temperature for 1 h, and stained with 4′,6-diamidino-2-phenylindole (DAPI; Vector, California, USA; Cat#H-1800). The slides were mounted and then scanned on an inverted confocal microscope (IRX60; SUNNY GROUP, Zhejiang, China). Senescence-associated β-Galactosidase Staining According to the manufacturer's instructions, cellular senescence was detected in frozen sections of human scalp tissue (from both men, one at 30 and another at 52, Table S5) using a SA-β-gal Staining Kit (Beyotime, Shanghai, China). The sections were washed three times with phosphate-buffered saline (PBS), fixed with the kit-provided fixative at room temperature for 15 min, and incubated overnight at 37 °C with a working solution containing 0.05 mg/mL 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-gal). Images were taken on a Mica microhub (Leica, Hessian, Germany), and the percentage of positive regions was calculated and analyzed using ImageJ (version 1.54g; National Institutes of Health, Maryland, USA). Statistical Analyses Statistical analysis of trichoscopy data was performed using GraphPad Prism 9.0.0 (California, USA) with a one-way analysis of variance (ANOVA) test, where P < 0.05 was considered significant. The significance of positive regions in SA-β-gal staining experiments between the two groups was tested by the Mann-Whitney U test, with P < 0.05 considered significant. For gene set score analysis, statistical significance was evaluated using two-sided Wilcoxon rank-sum tests between the two groups within each subpopulation. Outliers were excluded based on the 1.5× interquartile range (IQR) rule. The P values less than 0.05 were considered statistically significant. To identify age-related DEGs, the “FindMarkers” function in the Seurat package was applied between middle-aged and young groups for each cell type. The logarithmic fold change (LogFC) and adjusted P values for each DEG were calculated using a non-parametric two-sided Wilcoxon rank-sum test. DEGs meeting the criteria of |avg_logFC| > 0.25 and P_val_adj <0.05 were considered as age-related DEGs. Author Contributions Qian Zhao: conceptualization, project administration, supervision, resources, formal analysis, writing – original draft, writing – review & editing. Rui Ma: software, formal analysis, visualization, writing – original draft, writing – review & editing. Kun Huang: project administration, supervision, methodology, writing – original draft. Juan Wang: supervision, methodology. Donglin Zhang: investigation, validation. Jingyuan Wang: software, data curation. Xiaofeng Ding: formal analysis, writing – original draft. Feiyun Chen: investigation. Sijia Zhao: investigation. Na Ni: investigation. Xiaodie Zhang: investigation. Qian Du: investigation. Xiaojun Lin: investigation. Hua Wan: supervision. Jianglin Zhang: resources. Xiaolei Ding: supervision, resources. Shuang Yang: resources. Fengping Xu: conceptualization, supervision, resources, writing – review & editing. Yongxian Lai: conceptualization, supervision, resources, writing – review & editing. Ethics Approval All relevant procedures involving human sample collection and experiments presented in this study are compliant with ethical regulations, principles of the Declaration of Helsinki, and applicable local regulatory requirements and laws. All procedures were conducted under the approval of the Institutional Review Board of BGI (BGI-IRB 22047-T1) and Shanghai Skin Disease Hospital (2023–45 (ke) N2), and informed consent was obtained from all donors. Data Availability The scRNA and spatial transcriptomic sequencing data reported in this paper can be accessed in the Open Archive for Miscellaneous Data (Omix) by accession number OMIX009698 (https://ngdc.cncb.ac.cn/omix/OMIX009698). They can also be accessed at the China National Science Data Center (CNSA; CNP0006374, https://db.cngb.org/cnsa/CNP0006374) and the Spatial Transcript Omics DataBase (STOmics DB; STT0000145, https://db.cngb.org/stomics/STT0000145), respectively. The clinical data are not publicly accessible to protect the privacy of the participants. Upon request, de-identified participant data will be made available from the corresponding author Y.L. Declaration of Competing Interests The authors declare that they have no competing financial interests or personal relationships that could influence the work reported in this paper. Acknowledgements We sincerely thank all participants for donating the samples. We are also grateful to Junpu Mei, Yuanwei Zhang, and Yufen Huang from BGI Research for their valuable advice and assistance with data analysis. We further acknowledge the support from the Shared Instrumentation Core facility at Hangzhou Institute of Medicine, Chinese Academy of Sciences. This research was supported by the Science, Technology, and Innovation Commission of Shenzhen Municipality (grant number JCYJ20220818102809021 to F.X.). Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.hlife.2025.10.003. Contributor Information Fengping Xu, Email: xufengping@genomics.cn. Yongxian Lai, Email: laiyx@shskin.com. Supplementary data The following are the Supplementary data to this article: Multimedia component 1 Multimedia component 2 Multimedia component 3 Multimedia component 4 Multimedia component 5 Multimedia component 6 Multimedia component 7 References Associated Data Supplementary Materials Multimedia component 1 Multimedia component 2 Multimedia component 3 Multimedia component 4 Multimedia component 5 Multimedia component 6 Multimedia component 7 Data Availability Statement The scRNA and spatial transcriptomic sequencing data reported in this paper can be accessed in the Open Archive for Miscellaneous Data (Omix) by accession number OMIX009698 (https://ngdc.cncb.ac.cn/omix/OMIX009698). They can also be accessed at the China National Science Data Center (CNSA; CNP0006374, https://db.cngb.org/cnsa/CNP0006374) and the Spatial Transcript Omics DataBase (STOmics DB; STT0000145, https://db.cngb.org/stomics/STT0000145), respectively. The clinical data are not publicly accessible to protect the privacy of the participants. Upon request, de-identified participant data will be made available from the corresponding author Y.L.