Aging-associated transcriptional programs in T cells signify constituents of TGF-β signaling for immunosenescence Authors: Renke He, Chanchan Xiao, Wen Lei, Guobing Chen, Yangqiu Li, Oscar Junhong Luo Published: 2025 ABSTRACT BackgroundImmunosenescence, particularly the altered ratio of naïve and memory T cells, contributes to a diminished immune reserve and impaired adaptive immunity in aging and frail populations. The role of TGF-β signaling pathway—a critical hallmark of organismal senescence and T-cell exhaustion—in terminally differentiated effector memory T (Temra) cells remains elusive. We devised single-cell and bulk-cell RNA sequencing (RNA-seq) datasets to identify age-group-specific transcriptional regulatory networks in T cells and elucidate the roles of TGF-β signaling constituents associated with immunosenescence in Temra.ResultsAnalysis of scRNA-seq data from peripheral T cells across healthy human age groups revealed young-specific regulons controlled by FOXP1, TCF7, LEF1, and IKZF1 and old-specific regulons governed by EOMES, TBX21, RUNX3, and NFATC2. Transcription factor (TF)-binding-motif enrichment analysis implicated TGF-β signaling pathway components ZEB2 and TGFBR3 as pivotal target genes coregulated by multiple TFs, potentially facilitating T-cell terminal differentiation and exhaustion. Pseudotime analysis and bulk-cell RNA-seq further corroborated these regulons, validating their association with T-cell self-renewal capacity (young-specific) or effector/terminal differentiation (old-specific). In terms of aging, multiple TGF-β signaling activation components, including TGFB1, TGFBR1, SMAD3, ZEB2, and TGFBR3, were significantly upregulated in CD8 + Temra cells relative to CD8 + naïve T cells.ConclusionsOur study used systematic approaches for delineating age-dependent transcriptional networks for T-cell-associated immunosenescence. We identified multiple components of the TGF-β signaling pathway as potential biomarkers of Temra, which are strongly associated with senescence features including impaired differentiation plasticity, high cytotoxicity, and inflammatory chemotaxis capacity.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12915-025-02484-5. Background The aging of the immune system, characterized by a decrease in the absolute number of T cells and naïve T-cell reservoirs, an increase in the proportion of effector and terminally differentiated T cells, and a decrease in TCR diversity, leads to increased susceptibility to infections and failure of immune surveillance in the old and frail populations [1]. Temra (terminally differentiated effector memory T cells re-expressing CD45RA) cells represent the final differentiation stage of effector memory T cells, which are characterized by high cytotoxicity, inflammatory chemotaxis, and proliferation defects [2]. Temra is significantly more prevalent in older individuals than in young adults because of thymic degeneration and persistent stimulation by chronic antigens (e.g., cytomegalovirus), which leads to dysregulation of adaptive immunity, exacerbation of “inflammaging,” and a decline in tumor immunosurveillance [3, 4]. Thus, the increase in Temra-associated functionality serves as a well-recognized biomarker of immunosenescence in the elderly population [5]. The activation of transforming growth factor-β (TGF-β) signaling pathway has been well established as a driver of aging and frailty in humans [6]. For example, a significant increase in TGF-β in the blood induces the upregulation of the cell cycle inhibitory factor p21 in the kidneys, brain, and lungs, accompanied by decreases in multiple organ function indices, and blockade of the TGF-β signaling pathway alleviates aging in these multiple organs [7]. The classical TGF-β signaling pathway is regulated in T cells by transforming growth factor beta 1 (TGFB1), transforming growth factor beta receptor complex (TGFBR1/2/3), SMAD family signaling proteins (SMAD2/3/4/7), and the zinc finger E-box-binding homeobox transcription factors (ZEB1 and ZEB2). Upon TGFB1 binding, TGFBR2 recruits and phosphorylates TGFBR1, enabling intracellular signal transduction that promotes the nuclear translocation of SMAD2/3 complexes, thereby activating the canonical TGF-β pathway. Conversely, SMAD7 disrupts TGFBR1/2 heterodimerization, facilitates receptor ubiquitination-mediated degradation, and inhibits SMAD2/3 signaling, serving as a negative feedback regulator. TGFBR3 stabilizes the receptor complex by fine-tuning the signal intensity and duration to modulate pathway output [8–10]. TGFBR3 synergistically enhances pathway activation with TGFBR1 and TGFBR2 [11]. In Alzheimer’s disease, TGFBR3 was identified as a senescence marker for CD8 + Tem cells with IL7RAlow in the peripheral blood [12]. TGFBR3 is highly expressed in the peripheral blood T cells of healthy old donors and drove CD4 + T-cell differentiation toward Th9 cells, which causes chronic inflammation [13]. ZEB2 mediates epithelial–mesenchymal transition (EMT) in the tumor microenvironment in conjunction with TGFBR3 [14, 15]. A mouse lymphocytic choriomeningitis virus (LCMV) model demonstrated that ZEB2 acted independently or synergistically with TBX21 to drive cytotoxic T lymphocytes (CTLs) to lose their memory phenotype and acquire a terminally differentiated state. Additionally, ZEB2 deficiency promoted the upregulation of memory/precursor T-cell-related genes [16, 17]. SMAD7 inhibits TGF-β-mediated chimeric antigen receptor (CAR) T-cell exhaustion and enhances T-cell infiltration and cytotoxic potency in tumors [18–20], and it is upregulated in T cells in autoimmune diseases, thereby contributing to inflammation-related tissue damage [21]. Notably, existing studies have not elucidated the roles of ZEB2, TGFBR3, and SMAD7, along with other TGF-β pathway constituents, in regulating immunosenescence and Temra phenotypes within healthy human populations. Although our team [22] and others [23–25] have delineated the aging of peripheral blood immune cells via single-cell RNA sequencing (scRNA-seq), the transcriptional regulatory network governing the Temra phenotype under physiological conditions, along with its key transcription factors and target genes, remains elusive. Here, via a systematic analytical framework, we identified young-specific regulons, primarily targeted by FOXP1 (forkhead box P1), TCF7 (transcription factor 7), LEF1 (lymphoid enhancer binding factor 1), and IKZF1 (IKAROS family zinc finger 1), which characterize T-cell homeostasis and self-renewal capacity. We also identified old-specific regulons, specifically those targeted by EOMES (eomesodermin), TBX21 (T-box transcription factor 21), RUNX3 (RUNX family transcription factor 3), and NFATC2 (nuclear factor of activated T cells 2), which characterize T-cell terminal differentiation. Subsequently, we explored the dynamic shift in expression patterns from young-specific to old-specific regulons in T cells during the transition from high stemness to terminal differentiation. We identified ZEB2 and TGFBR3, two key components of the TGF-β signaling pathway, as common target genes of EOMES, TBX21, and NFATC2. Notably, multiple constituents involved in TGF-β signaling pathway activation—including TGFB1, TGFBR1, TGFBR3, SMAD3, and ZEB2—along with the negative feedback regulator SMAD7, were significantly upregulated in Temra cells. These genes exhibited synergistic overexpression in Temra cells and serve as molecular markers for characterizing immunosenescence. Results Delineation of T-cell subtypes across age groups We devised a systematic approach for investigating aging-related regulatory programs in T cells. First, we sought to identify the functional cell types that account for the phenotypic differences in immune homeostasis and immunosenescence across age groups. We reanalyzed the publicly available scRNA-seq datasets of human peripheral blood mononuclear cells (PBMCs) from donors of different age groups, including neonates (cord blood), young individuals, and healthy and frail old individuals. The age information of each donor was taken from the original publication [22] and summarized in Table S1 (Additional file 2). After data quality control (Additional file 1: Fig. S1a and b), dimensionality reduction, and clustering, we identified the canonical cell types in PBMCs (Additional file 1: Fig. S1c, “Methods”). Compared with the young group, the old and frail groups exhibited a decrease in the proportion of Tn (CCR7hi, SELLhi) cells and an increase in Tem (effector memory T, CCR7−, SELL−), mNK (mature NK, FCGR3Ahi, KLRD1hi), and cMono (classical monocyte, CD14hi, S100A8hi, S100A9hi) cells (Additional file 1: Fig. S1d and e, Additional file 3: Table S2). We further confirmed the validity of the above cellular annotations via pathway enrichment analysis. For example, the differentially expressed genes (DEGs) of CD4 + Tn, CD4 + Tcm, and CD8 + Tn cells were enriched in “T-cell differentiation” and “T-cell selection,” CD8 + Tem and mNK cells were enriched in “cell killing,” and cMono cells were enriched in “granulocyte chemotaxis” and “leukocyte migration” (Additional file 1: Fig. S1f, Additional file 4: Table S3). Adaptive immunodeficiency due to imbalance and dysfunction of T-cell naïve/memory subpopulations is one of the pivotal features of immunosenescence under normal chronological aging. Therefore, we hereafter focused on T cells by extracting CD4 + Tn, CD4 + Tcm, Treg, CD8 + Tn, and CD8 + Tem cells from young and old groups and further clustered them into more functionally refined subtypes. The frail donors were excluded from the subsequent analyses, as they had more chronic conditions and poorer physical fitness [22], which is beyond the scope of this study. In addition to the canonical T-cell subtypes, the other subtypes included CD8 + Tcm_KLRB1 (IL7Rhi, KLRB1hi), CD8 + Tem_ISGs (ISGs: interferon-stimulated genes, CCR7−, SELL−, OASLhi, IFIT2hi, HERC5hi), γδT (TRGV9hi, TRDV2hi), and Tex (exhausted T, MALAT1hi, NEAT1hi, SYNE2hi) (Fig. 1a and b, Additional file 1: Fig. S2a and b, Additional file 3: Table S2). In this process, we randomly sampled the T cells of the old group so that the number of cells in the young and old groups was similar, thereby eliminating the potential effect of quantity variance on downstream analysis (Fig. 1a). We observed that the proportions of CD8 + Tcm and CD8 + Tem cells were significantly greater in the old group than in the young group, whereas the proportion of CD8 + Tn cells was significantly lower. Although statistical significance was not reached, the populations of CD8 + Tem_ISGs and CD4 + Tcm and Tex cells were greater in the old group compared to the young group, whereas the abundances of CD4 + Tn and γδT cells were lower in the old group (Fig. 1c and d). Through these analyses, we systematically characterized the distinct subtype distributions of T cells between the young and old groups at the single-cell level. These findings are consistent with the established results regarding shifts in T-cell immunosenescence [26, 27].Fig. 1Identification and comparison of T-cell subtypes between healthy young and old individuals via scRNA-seq. a Uniform manifold approximation and projection (UMAP) visualization and annotation of T-cell subtypes from the young and old groups, respectively. ISGs, interferon-stimulated genes; ex, exhausted. b Dot plot of the marker genes used for all cell subtype annotations. The color scale represents the normalized expression levels of every gene, and the size of the dot represents the expression percentage of each gene in each cell type. The same applies to the dot plots of gene expression in this paper. c Stacked bar plot showing the proportions of each cell type for each age group. The colors of the cell types are the same as those in a. d Box plot for comparing the abundance of each T-cell subtype between the young and old groups. Each dot represents a donor. P-values were derived from the Wilcoxon test Identification and comparison of T-cell subtypes between healthy young and old individuals via scRNA-seq. a Uniform manifold approximation and projection (UMAP) visualization and annotation of T-cell subtypes from the young and old groups, respectively. ISGs, interferon-stimulated genes; ex, exhausted. b Dot plot of the marker genes used for all cell subtype annotations. The color scale represents the normalized expression levels of every gene, and the size of the dot represents the expression percentage of each gene in each cell type. The same applies to the dot plots of gene expression in this paper. c Stacked bar plot showing the proportions of each cell type for each age group. The colors of the cell types are the same as those in a. d Box plot for comparing the abundance of each T-cell subtype between the young and old groups. Each dot represents a donor. P-values were derived from the Wilcoxon test Aging-related eigengene modules To minimize the analytical burden of downstream analyses and increase the biological interpretability of the results, we excluded CD8 + Tem_ISGs and γδT cells, which were low in abundance and did not show significant intergroup differences. We performed high-dimensional weighted gene co-expression network analysis (hdWGCNA) on the remaining T-cell subtypes. As a result, five eigengene modules (MEs) were identified as specific to particular cell subtype(s) and age group: ME1 (turquoise), ME2 (green), ME3 (blue), ME4 (yellow), and ME5 (brown) (Fig. 2a). At the T-cell subset level, genes in ME1 presented high expression in Tn cells; ME2 presented moderate expression in all subtypes except Tex cells; ME4 was expressed in CD8 + Tem, CD8 + Tcm, and CD8 + Tcm_KLRB1 cells, with the highest expression in CD8 + Tem cells; and ME3 and ME5 both presented the highest expression in Tex cells. At the age group level, genes in ME1 and ME2 presented specifically high expression in the young group, whereas those in ME3 to ME5 presented stronger expression in the old group (Fig. 2b and c). Across all the analyzed T-cell subtypes, the gene expression signature scores of ME1 and ME2 were significantly greater in the young group, even in Tex cells, whereas the signature scores of ME3, ME4, and ME5 were significantly greater in the T cells of the old group including Tn cells but except ME5 in Tex cells (Fig. 2d).Fig. 2Identification of age-specific MEs via hdWGCNA. a Dendrogram showing the eigengene modules identified by hdWGCNA. Each leaf on the dendrogram represents a single gene, and the color at the bottom indicates the co-expression ME assignment. Genes denoted in gray were not classified in any of the co-expression ME. b Dot plot of ME expression in distinct T-cell types or age groups. The top legend is for the left panel, and the right legend is for the right panel. The color scale represents the normalized average expression levels of genes in each ME, and the size of the dot represents the expression percentage of each ME in each cell type or age group. c UMAP plots with projection of ME1 to ME5 scores calculated by UCell. UMAP with cell type color coding is also shown to aid interpretation. Each color scale represents the expression activity level of the corresponding ME. The colors used for the MEs are the same as those in a.d Half-violin plots for comparing the activity difference of each ME between the young and old groups. The red dot for each half-violin represents the median of activity. *P < 0.01, **P < 0.01, ***P < 0.001, ****P < 0.0001. P-values were calculated by Wilcoxon’s test. e Dot plot shows the top 5 enriched GO functions for each ME Identification of age-specific MEs via hdWGCNA. a Dendrogram showing the eigengene modules identified by hdWGCNA. Each leaf on the dendrogram represents a single gene, and the color at the bottom indicates the co-expression ME assignment. Genes denoted in gray were not classified in any of the co-expression ME. b Dot plot of ME expression in distinct T-cell types or age groups. The top legend is for the left panel, and the right legend is for the right panel. The color scale represents the normalized average expression levels of genes in each ME, and the size of the dot represents the expression percentage of each ME in each cell type or age group. c UMAP plots with projection of ME1 to ME5 scores calculated by UCell. UMAP with cell type color coding is also shown to aid interpretation. Each color scale represents the expression activity level of the corresponding ME. The colors used for the MEs are the same as those in a.d Half-violin plots for comparing the activity difference of each ME between the young and old groups. The red dot for each half-violin represents the median of activity. *P < 0.01, **P < 0.01, ***P < 0.001, ****P < 0.0001. P-values were calculated by Wilcoxon’s test. e Dot plot shows the top 5 enriched GO functions for each ME In terms of functional enrichment, pathways enriched in ME1 and ME2 were associated with cellular respiration, metabolism, and robust energy production; ME3 was enriched in epigenetic modification-related pathways, particularly histone deacetylation, which not only inhibits the transcriptional process but also promotes the aging of hematopoietic stem cells and myeloid differentiation bias [28, 29]; ME4 was enriched in pathways involved in MHC molecule assembly and CD8 + T-cell activation; and ME5 was enriched in IL-9 signaling pathways, which can mediate chronic inflammatory states associated with the senescence-associated secretory phenotype (SASP) [13], and regulatory pathways of necrotic programmed cell death (Fig. 2e). These results suggest that ME1 and ME2 could reflect the immune homeostasis of T cells in young adults and represent immune homeostasis characterized by the active biosynthetic capacity and high self-renewal potential of Tn cells. Conversely, ME3 to ME5 indicate an immunosenescent state in the older individuals, manifested by an imbalance between naïve and effector T cells alongside cytotoxic and proinflammatory functions. Age-group-specific regulatory programs We next sought to characterize the genes in the MEs via SCENIC analysis to examine potential transcriptional regulatory connections within each module, which could be specific to distinct T-cell subtypes of different age groups (Fig. 3a, “Methods”). By focusing on the top 150 genes from each ME for regulatory network analysis, we identified a list of regulons with T-cell subtype specificity. Regulons governed by TCF7 [30], FOXP1 [30], IKZF1 [31], and YBX1 [32] exhibited high activity in Tn and Treg cells and were enriched for “T-cell differentiation,” “Positive regulation of leukocyte cell adhesion,” and “Negative regulation of leukocyte apoptotic process” (Fig. 3b, Additional file 1: Fig. S3a, Additional file 4: Table S3). Conversely, RUNX3 [33], EOMES [33], TBX21 [34], and NFATC2 [35] exhibited high regulatory activity in CD8 + Tcm, CD8 + Tcm_KLRB1, and CD8 + Tem (Fig. 3b) on genes enriched in “Cell killing,” “Negative regulation of lymphocyte differentiation,” “Negative regulation of lymphocyte activation,” and “Peptide antigen assembly with MHC protein complex” (Additional file 1: Fig. S3b, Additional file 4: Table S3). Additionally, regulons targeted by POLR2A (RNA polymerase II subunit A) [36] and EP300 (E1A binding protein P300) [37], which regulate transcriptional processes, were highly active in the CD8 + Tcm, CD8 + Tcm_KLRB1, CD8 + Tem, and Tex cells (Fig. 3b, Additional file 4: Table S3). The presence of the binding motif provides potential evidence for direct DNA binding in the TF-target regulatory relationship [38]. The TF-motif enrichment results suggested the presence of binding motifs for FOXP1 or TCF7 in the promoter regions of multiple naïve or precursor T-cell characteristic genes (e.g., CCR7, IL7R, and LEF1) (Additional file 1: Fig. S4a and b). The promoters of ZEB2 and TGFBR3 shared the same binding motif of EOMES, TBX21, and NFATC2 (Additional file 1: Fig. S4c, d, and e).Fig. 3Identification and characterization of age-group-specific regulons. a Process of screening MEs for SCENIC analysis. The numbers in parentheses represent the gene count at each step. Total genes, genes from original scRNA-seq data. Common genes in regulons and MEs, the intersection of the genes from regulons generated by SCENIC and the top 150 genes from each ME. Age-group-specific regulons, genes with differential expression between the young and old groups. Young-specific regulons, the intersection of young-specific genes and genes in ME1 and ME2. Old-specific regulons, the intersection of old-specific genes and genes in ME3 to ME5. b Heatmap visualization of the average expression activity (Z-score transformed) of regulons in distinct T-cell subtypes. The gene symbols on the right of the heatmap denote the TFs governing the corresponding regulons. The number of genes included in the regulons is shown in parentheses. c Dot plots showing TF regulon specificity scores (RSSs, y-axis) vs. regulon rank (x-axis) in young (left) and old (right) groups. The selected regulons with relatively high RSS specificity are highlighted in color: green represents relatively high values in the young group, and purple represents relatively high values in the old group. d Violin plots with boxplot inserts for comparing UCell scores for the genes from the age group-specific regulons between the young and old groups. The red dashed line represents the median of the UCell scores, and the same applies to e and Fig. S5. ****P < 0.0001, Wilcoxon test. e Violin plots with boxplot inserts for comparing UCell scores between cross-age groups for the genes from the age group-specific regulons among distinct T-cell subtypes. ****P < 0.0001, Kruskal–Wallis test. f UMAP plots with expression level projections of selected TFs regulating age-group-specific regulons, split by age groups. The color scale represents the normalized expression. UMAP with cell type color coding is also shown to aid interpretation Identification and characterization of age-group-specific regulons. a Process of screening MEs for SCENIC analysis. The numbers in parentheses represent the gene count at each step. Total genes, genes from original scRNA-seq data. Common genes in regulons and MEs, the intersection of the genes from regulons generated by SCENIC and the top 150 genes from each ME. Age-group-specific regulons, genes with differential expression between the young and old groups. Young-specific regulons, the intersection of young-specific genes and genes in ME1 and ME2. Old-specific regulons, the intersection of old-specific genes and genes in ME3 to ME5. b Heatmap visualization of the average expression activity (Z-score transformed) of regulons in distinct T-cell subtypes. The gene symbols on the right of the heatmap denote the TFs governing the corresponding regulons. The number of genes included in the regulons is shown in parentheses. c Dot plots showing TF regulon specificity scores (RSSs, y-axis) vs. regulon rank (x-axis) in young (left) and old (right) groups. The selected regulons with relatively high RSS specificity are highlighted in color: green represents relatively high values in the young group, and purple represents relatively high values in the old group. d Violin plots with boxplot inserts for comparing UCell scores for the genes from the age group-specific regulons between the young and old groups. The red dashed line represents the median of the UCell scores, and the same applies to e and Fig. S5. ****P < 0.0001, Wilcoxon test. e Violin plots with boxplot inserts for comparing UCell scores between cross-age groups for the genes from the age group-specific regulons among distinct T-cell subtypes. ****P < 0.0001, Kruskal–Wallis test. f UMAP plots with expression level projections of selected TFs regulating age-group-specific regulons, split by age groups. The color scale represents the normalized expression. UMAP with cell type color coding is also shown to aid interpretation In terms of T-cell subtype enrichment per age group, regulons targeted by TCF7, FOXP1, IKZF1, and YBX1 had greater regulon specificity scores (RSSs) in the young group, whereas regulons targeted by RUNX3, EOMES, TBX21, YBX1, POLR2A, and EP300 had greater RSSs in the old group (Fig. 3c, Additional file 5: Table S4). We further examined the overall expression of genes (UCell score) in the young-specific and old-specific regulons, respectively, validating that young-specific regulons were significantly upregulated in the young group and vice versa (P < 0.05) (Fig. 3d). At the cell subtype level, we observed consistent trends with the aforementioned intergroup differences. Moreover, the elevated expression of genes in the young-specific regulons was more profound in the Tn and Treg cells; conversely, the old-specific regulons were more significantly upregulated in the Tcm, Tem, and Tex cells (P < 0.05) (Fig. 3e). The activated TGF-β signaling pathway can drive T-cell terminal differentiation and exhaustion [8]. The enrichment analysis of DEGs between young and old groups suggested that the signature score of the TGF-β signaling pathway-related gene set was significantly greater in the old group than in the young group, especially for CD8 + Tcm, CD8 + Tcm_KLRB1, CD8 + Tem, and Tex cells (P < 0.05). More importantly, ZEB2 and TGFBR3 were both critical constituents of the TGF-β signaling-associated pathways (Additional file 1: Fig. S5, Additional file 4: Table S3). Furthermore, gene expression projection on the UMAP plot revealed that FOXP1, TCF7, LEF1, and IKZF1 were highly expressed in CD8 + Tn, CD4 + Tn, CD4 + Tcm, and Treg cells. Conversely, EOMES, TBX21, NFATC2, and ZEB2 were highly expressed in CD8 + Tcm, CD8 + Tcm_KLRB1, CD8 + Tem, and Tex cells (Fig. 3f). Kernel density plots reflected the specificity of TFs and selected target genes from young-specific or old-specific regulons across the aforementioned T-cell subtypes (Additional file 1: Fig. S6a). Genes from age-group-specific regulons, including ZEB2 and TGFBR3, demonstrated significant differences in normalized expression between the young and old groups (Additional file 1: Fig. S6b). Overall, the age-group-specific regulons effectively distinguished T-cell expression patterns between young and old individuals. ZEB2 and TGFBR3, as potential target genes of old-specific regulons and constituents of the TGF-β pathway, may promote T-cell differentiation toward effector or terminal states, thereby plausibly contributing to differences in T-cell lineages between elderly and young populations. Dynamics of age-group-specific regulons in T-cell terminal differentiation To further characterize the role of age-group-specific regulons in T-cell phenotypic transitions, we employed pseudotime analysis and mapped the expression dynamics of these regulons throughout T-cell development from naïve to terminal differentiation states (“Methods”). As expected, pseudotime analysis revealed that CD8 + Tn and CD4 + Tn cells were predominantly localized at the origin of the developmental trajectory, which coincided with the region of the highest signature scores for the young-specific regulons and the highest differentiation potential signature scores (Fig. 4a, b, c, d, e). Compared with Tn cells, CD4 + Tcm, Treg, and CD8 + Tcm_KLRB1 cells were positioned in the middle segment of the trajectory, exhibiting overlapping expression characteristics of both young-specific and old-specific regulons, with lower differentiation potential. CD8 + Tcm, CD8 + Tem, and Tex cells were positioned at the termini of the trajectory, highly expressed genes from the old-specific regulons, and presented significantly lower differentiation potential than the other T-cell subtypes did (Fig. 4a, b, c, d, e).Fig. 4Significance of age-specific regulons in functional shifts in T cells during immunosenescence. a DDR (discriminative dimensionality reduction) visualization of T cells by mapping pseudotime (left) and T-cell subtype information (right). The black dashed arrows represent the direction of differentiation, and the same applies to the other DDR plots in this paper. b DDR visualization with UCell score mapping of genes in the young-specific regulons (left) and the old-specific regulons (right). c Cell density plot showing the changes in the abundance (y-axis) of distinct T-cell subtypes along the predicted pseudotime (x-axis). d DDR visualization with projection of the predicted differentiation order. The higher the predicted order value is, the greater the degree of cell stemness and the lower the degree of differentiation. e Box plots for comparing the predicted differentiation order of cells in each subtype. f Waterfall diagram showing the correlations between gene expression and the predicted differentiation order in the analyzed T cells. Selected genes from the young-specific regulons and old-specific regulons are highlighted in green and purple, respectively. g Left: expression heatmap of the top DRGs that had the most correlated (or anti-correlated) expression profile (Z-score transformed and smoothed) with the T-cell trajectory pseudotime. Right: Dot plot showing the top 5 enriched GO BP functions for each DRG cluster. GO function names are followed by enriched genes in parentheses. The numbers in parentheses indicate the number of enriched genes in each cluster (x-axis) Significance of age-specific regulons in functional shifts in T cells during immunosenescence. a DDR (discriminative dimensionality reduction) visualization of T cells by mapping pseudotime (left) and T-cell subtype information (right). The black dashed arrows represent the direction of differentiation, and the same applies to the other DDR plots in this paper. b DDR visualization with UCell score mapping of genes in the young-specific regulons (left) and the old-specific regulons (right). c Cell density plot showing the changes in the abundance (y-axis) of distinct T-cell subtypes along the predicted pseudotime (x-axis). d DDR visualization with projection of the predicted differentiation order. The higher the predicted order value is, the greater the degree of cell stemness and the lower the degree of differentiation. e Box plots for comparing the predicted differentiation order of cells in each subtype. f Waterfall diagram showing the correlations between gene expression and the predicted differentiation order in the analyzed T cells. Selected genes from the young-specific regulons and old-specific regulons are highlighted in green and purple, respectively. g Left: expression heatmap of the top DRGs that had the most correlated (or anti-correlated) expression profile (Z-score transformed and smoothed) with the T-cell trajectory pseudotime. Right: Dot plot showing the top 5 enriched GO BP functions for each DRG cluster. GO function names are followed by enriched genes in parentheses. The numbers in parentheses indicate the number of enriched genes in each cluster (x-axis) Through genes or pathways associated with specific functions, we were able to elucidate T-cell phenotypic shifts more intuitively. Several genes positively correlated with differentiation potential were derived from young-specific regulons, including SELL (selectin L), IL7R (interleukin-7 receptor), TCF7, LEF1, NOSIP (nitric oxide synthase-interacting protein) [39], LTB (lymphotoxin beta) [40], which are established markers of naïve and precursor T cells, and PABPC1 (poly(A)-binding protein cytoplasmic 1) and NPM1 (nucleophosmin 1), which have been demonstrated to promote the antiaging capacity of hematopoietic stem cells (HSCs) by increasing their resistance to stress [41]. In contrast, a series of genes negatively correlated with differentiation potential were extracted from the old-specific regulons, including proinflammatory chemokines such as CCL4 (C–C motif chemokine ligand 4) and CCL5 [42], cytotoxicity and cytolysis-related genes (e.g., GZMB, CTSW, GNLY, PRF1), and TFs of the old-specific regulons: TBX21, multiple MHC molecules, and TGF-β signaling components, such as TGFB1, TGFBR3, and ZEB2 (Fig. 4f, Additional file 1: Fig. S3a and b). We next clustered the genes with correlated (or anticorrelated) expression trends with the developmental pseudotime for functional enrichment analysis, thereby characterizing functional transitions during T-cell differentiation. Among the defined gene clusters, cluster 3 was enriched in pathways characterizing T-cell stemness, transcription initiation, and antiapoptosis, whereas cluster 2 was enriched in signaling pathways such as the NF-κB pathway and leukocyte adhesion, which promote T-cell activation and maintain T-cell immune memory [43]. Cluster 4 was enriched in oxidative stress and regulates IL-2 production, which is associated with promoting cellular damage and senescence [44]. Clusters 1 and 5 were enriched in pathways related to cell killing, cytotoxicity, the assembly of MHC molecules, and the regulation of T-cell toxicity and activation, which were related to the clearance of pathogens and proinflammatory functions (Fig. 4g, Additional file 4: Table S3, Additional file 6: Table S5). Thus, the dynamic expression of genes in age-group-specific regulons along the pseudotime trajectory intuitively and reasonably reflected the phenotypic transition of naïve T cells to terminally differentiated T cells. Characteristics of age-group-specific regulons and TGF-β signaling in Tn, Tcm/Tem, and Temra cells To further test the validity of the age-group-specific regulons in characterizing T-cell immune homeostasis and immunosenescence, and to investigate the associations between components of the TGF-β signaling pathway (including ZEB2 and TGFBR3) and the T-cell terminal differentiation state, we utilized a public bulk-cell RNA-seq dataset of Tn, Tcm/Tem, and Temra cells from 28 healthy old individuals for independent analysis [45] (Additional file 7: Table S6, “Methods”). We found that the gene expression signature scores of the young-specific regulons progressively decreased from Tn to Tcm/Tem and further decreased to Temra cells in the analyzed individuals, whereas those of the old-specific regulons progressively increased (P < 0.05, Fig. 5a). Furthermore, the young-specific and old-specific regulons presented clear differential expression patterns in these distinct T-cell subtypes, respectively, validating the specificity of these age-group-specific regulons and their association with T-cell stemness and senescence (Fig. 5b).Fig. 5Characterization of genes associated with age-group-specific regulons in Tn, Tcm/Tem, and Terma cells. a Violin plots with boxplot inserts visualizing the signature scores for the genes in the young-specific regulons or old-specific regulons among Tn, Tcm/Tem, and Temra cells. The red dashed line represents the median; each point corresponds to a sample. *P < 0.01, **P < 0.01, ***P < 0.001, ****P < 0.0001. P-values were calculated by the Wilcoxon test. b Heatmap visualizing the signature score (Z-score transformed and smoothed) for the young-specific and old-specific regulons in T-cell subtypes. c Bar plot showing the functional enrichment of genes in the young-specific and old-specific regulons. Genes in red were also identified as DRGs in Fig. 4g. d Violin plots with boxplot inserts visualizing the expression differences of genes from age-group-specific regulons among Tn, Tcm/Tem, and Temra. The black dashed line separates the genes from the young-specific regulons (top row) and from the old-specific regulons (bottom row). The gene symbol of the governing TF is shown at the top of each panel. e Scatter plots with fitted lines demonstrating the correlation between the expression of pairs of TFs. The black dashed line separates the TFs from the young-specific regulons (left) and the old-specific regulons (right). The correlation coefficient (r) calculated via Pearson’s correlation and the associated P-value are shown at the top of each panel. Each point corresponds to a sample; the x-axis and y-axis represent the normalized expression of the corresponding TFs. The same applies to Fig. S9 Characterization of genes associated with age-group-specific regulons in Tn, Tcm/Tem, and Terma cells. a Violin plots with boxplot inserts visualizing the signature scores for the genes in the young-specific regulons or old-specific regulons among Tn, Tcm/Tem, and Temra cells. The red dashed line represents the median; each point corresponds to a sample. *P < 0.01, **P < 0.01, ***P < 0.001, ****P < 0.0001. P-values were calculated by the Wilcoxon test. b Heatmap visualizing the signature score (Z-score transformed and smoothed) for the young-specific and old-specific regulons in T-cell subtypes. c Bar plot showing the functional enrichment of genes in the young-specific and old-specific regulons. Genes in red were also identified as DRGs in Fig. 4g. d Violin plots with boxplot inserts visualizing the expression differences of genes from age-group-specific regulons among Tn, Tcm/Tem, and Temra. The black dashed line separates the genes from the young-specific regulons (top row) and from the old-specific regulons (bottom row). The gene symbol of the governing TF is shown at the top of each panel. e Scatter plots with fitted lines demonstrating the correlation between the expression of pairs of TFs. The black dashed line separates the TFs from the young-specific regulons (left) and the old-specific regulons (right). The correlation coefficient (r) calculated via Pearson’s correlation and the associated P-value are shown at the top of each panel. Each point corresponds to a sample; the x-axis and y-axis represent the normalized expression of the corresponding TFs. The same applies to Fig. S9 At the functional level, genes from young-specific regulons were enriched in pathways regulating T-cell differentiation, anti-apoptosis, and the promotion of transcription and biosynthesis. In contrast, genes from old-specific regulons were enriched in pathways related to cytotoxicity, cytolysis, antigen presentation, and immune response regulation (Fig. 5c, Additional file 4: Table S3). These findings aligned with the functional enrichment observed for age-group-specific regulons from the scRNA-seq analyses (Additional file 1: Fig. S3a and b). The TFs and their targets from the age-group-specific regulons were differentially expressed in the three analyzed T-cell subtypes. Compared to Temra, FOXP1, TCF7, LEF1, LTB, PABPC1, NPM1, genes related to adhesion and migration molecules (SELL and CCR7), and co-stimulation molecules (CD27 and CD28) were significantly higher expressed in Tn and Tcm/Tem cells (P < 0.05) (Fig. 5d, Additional file 1: Fig. S7a). Conversely, EOMES, TBX21, NFATC2, ZEB2, TGFBR3, and genes related to proinflammation (XCL1, CCL3, CCL5), cytotoxicity (CTSW, FGFBP2, FCGR3A, NCAM1, CMC1, GZMB, PRF1, NKG7, GNLY), immunosuppression (TIGIT, TOX, LAG3), terminal differentiation [5, 46] (KLRD1, KLRG1, B3GAT1/CD57, CX3CR1), and cell cycle arrest [47] (CDKN2A/p16) were significantly more highly expressed in Temra (P < 0.05) (Fig. 5d, Additional file 1: Fig. S7a). These results reflect a progressive shift from Tn to Tcm/Tem, and then to Temra, characterized by a decrease in self-renewal capacity and differentiation potential, along with an increase in proinflammatory and cytolytic capacities, as well as exhaustion (Additional file 1: Fig. S7b). In addition, FOXP1, TCF7, LEF1, and IKZF1 presented moderate synergistic expression (P < 0.05, r > 0.3), whereas EOMES, TBX21, ZEB2, and NFATC2 presented significantly stronger co-expression (P < 0.05, r > 0.75, Fig. 5e). These findings revealed the transcriptional regulatory profiles and expression characteristics of young-specific regulons involved in maintaining T-cell immune homeostasis, as well as old-specific regulons involved in mediating T-cell-related immunosenescence. To further characterize the role of the TGF-β signaling pathway in T-cell senescence, we compared the expression profiles of additional components of the TGF-β pathway other than TGFBR3 and ZEB2 across these three T-cell subtypes (Additional file 1: Fig. S8). Compared with Tn cells, Temra cells presented significant upregulation of the expression of the TGF-β pathway activators: TGFB1, TGFBR1, and SMAD3, accompanied by downregulation of TGFBR2 (P < 0.05). Concurrently, the negative feedback regulator SMAD7 was also upregulated in Temra (P < 0.05). The upregulation of ZEB2, TGFBR3, and TGFB1 in Tcm/Tem and Temra cells, along with the upregulation of ZEB1 and TGFBR2 in Tn cells, corroborated the findings from the scRNA-seq analysis (Additional file 1: Fig. S6a). Considering ZEB2 is one of the pivotal transcription factors after TGF-β signaling activation and is defined as a driver of T-cell exhaustion and terminal differentiation [16, 17, 48–50], we further investigated the expression correlation between ZEB2 and T-cell functionally important genes. Through Tn to Tcm/Tem and to Temra, ZEB2 was positively correlated with proinflammation (e.g., XCL1 and CCL5), cytotoxicity (e.g., NKG7 and GNLY), NK-specific markers (e.g., FCGR3A and KLRD1), biomarkers of T-cell terminal differentiation (e.g., KLRG1 and CX3CR1), components of the TGF-β signaling pathway (e.g., TGFBR3 and SMAD7), and cellular senescence (CDKN2A) (r ≥ 0.5, P < 0.05, Fig. S9). It was negatively correlated with co-stimulation (CD27), antistress (PABPC1, NPM1), stemness biomarkers (e.g., FOXP1 and TCF7), and components of the TGF-β signaling pathway (TGFBR2, ZEB1) (r ≤ − 0.4, P < 0.05, Additional file 1: Fig. S9). Such expression correlation between these two clusters of genes signifies T-cell terminal differentiation and differentiation plasticity, respectively (Additional file 1: Fig. S10). Moreover, the activity of the TGF-β pathway was greater in Temra cells than in Tn and Tcm/Tem cells, which was consistent with the results from the scRNA-seq analysis (Additional file 1: Fig. S5, Additional file 1: Fig. S11a and b). Overall, these results collectively suggest that age-group-specific regulons effectively delineate the functional states of Tn, Tcm/Tem, and Temra. Meanwhile, ZEB2, TGFBR3, and SMAD7 are putative regulators involved in the generation of Temra and positively correlated with T-cell immunosenescence (Fig. 6).Fig. 6Age-group-specific regulons characterize the phenotypic shift of T cells from Tn to Temra through aging. Schematic of transcriptional regulatory network dynamics in T cells during immunosenescence. Through aging, naïve T-cell (Tn) population declines, while terminally differentiated effector memory T-cell (Temra) abundance increases. Concomitantly, transcriptional regulation shifts from a FOXP1-TCF7-LEF1-IKZF1 network sustaining self-renewal capacity in Tn cells to an EOME Age-group-specific regulons characterize the phenotypic shift of T cells from Tn to Temra through aging. Schematic of transcriptional regulatory network dynamics in T cells during immunosenescence. Through aging, naïve T-cell (Tn) population declines, while terminally differentiated effector memory T-cell (Temra) abundance increases. Concomitantly, transcriptional regulation shifts from a FOXP1-TCF7-LEF1-IKZF1 network sustaining self-renewal capacity in Tn cells to an EOME Discussion In this study, we devised scRNA-seq datasets and a systematic approach for identifying aging-related transcriptional regulatory programs in T cells. In particular, the old-specific regulons were targeted by transcription factors such as EOMES, TBX21, and NFATC2, which promote T-cell terminal differentiation and senescence. The target genes of these regulons included multiple components of the TGF-β signaling pathway, such as TGFBR3 and ZEB2. We subsequently demonstrated that the TGF-β signaling pathway was more active in T cells from the older populations than in those from young populations, as evidenced by the upregulation of TGF-β signaling activation members (e.g., TGFB1, TGFBR1, TGFBR3, SMAD3, and ZEB2) and the negative feedback factor SMAD7 in Temra subpopulations. These findings thus characterized Temra as both senescent and proinflammatory. Young-specific regulons at the functional gene level were targeted by FOXP1, TCF7, LEF1, and IKZF1. TCF7 and LEF1 have been confirmed to maintain the self-renewal capacity of T cells [51]. Defects or downregulated FOXP1 expression leads to aging in multiple tissues and organs, including mesenchymal stem cells [52–54]. IKZF1 defects lead to neurodegenerative pathologies and dysfunction in the differentiation and function of types of multiple lymphocytes [55]. The young-specific regulons exhibited T-cell immune homeostasis and stem cell-like characteristics, which are consistent with those characterized by pan-age single-cell transcriptome profiles of naïve T cells [24]. In addition, we identified TFs that drive T-cell differentiation toward effector or terminally exhausted phenotypes from old-specific regulons, which are primarily targeted by EOMES, TBX21, RUNX3, and NFATC2. EOMES, TBX21, and RUNX3 can promote T-cell differentiation toward effector/memory phenotypes in both physiological and pathological contexts and have been established as drivers of T-cell senescence in multiple studies [56–60]. NFATC2, a member of the nuclear factor of activated T cells (NFAT) transcription factor family, regulates T-cell activity [35] and drives the generation of exhausted T-cell phenotypes in both physiological immunosenescence and the tumor microenvironment [25, 61]. The old-specific regulons represented a transcriptional regulatory imprint of T cells preferentially differentiating toward effector/exhaustion phenotypes, particularly in Temra. These factors contribute to the reduced T-cell reserves and diminished activity-maintaining capacity in elderly individuals, as well as the organismic state of immune dysregulation with chronic inflammation [1]. In summary, we propose that age-group-specific regulons—identified on the basis of T-cell subtype-specific expression profiles and age-group expression differences (Fig. 3a, “Methods”)—could distinguish the overall differences in the T-cell functional state between young and old individuals (Fig. 3d) and characterize the variation in gene expression across distinct T-cell differentiation stages (Figs. 3e, and 4b, Additional file 1: Fig. 5a and b). Indeed, further experimental validation and multi-omics analyses are needed to dissect the complex regulatory mechanisms of age-group-specific regulons and identify genes that play pivotal regulatory roles. Although the activation of the TGF-β signaling pathway is a hallmark of senescence [6] and an important component of immunosuppressive cross-talk networks under chronic viral infection or oncogenesis, which suppresses the activation and differentiation of CD8 + T cells and facilitates the exhaustion of activated CD8 + T cells [8, 62, 63], it remains unclear which constituents of this pathway are potential Temra-related targets. In recent years, ZEB2 has been demonstrated to drive the generation of age-associated B cells [48] and CD4 + T-helper cells [49] with SASP and promote the terminal differentiation of CD8 + T cells in chronic viral infection and cancer [64, 65], and TGFBR3 has been shown to facilitate CD4 + T-cell differentiation toward Th9 in a proinflammatory phenotype and the generation of memory CD8 + T cells with a senescent phenotype in Alzheimer’s disease [12, 13]. Our analysis revealed that ZEB2 and TGFBR3 were co-enriched in multiple pathways regulating TGF-β signaling; were highly expressed in Tcm, Tem, and Tex cells; and were significantly more highly expressed in the old group than in the young group. Furthermore, the promoter regions of ZEB2 and TGFBR3 contain binding motifs for key transcription factors that drive effector/terminally exhausted T-cell phenotypes, such as EOMES, TBX21, and NFATC2. These findings led us to further investigate the significance of TGF-β signaling in T-cell senescence, especially in Temra, via bulk RNA-seq. Temra cells represent a paradoxical phenotype integrating exhaustion/senescence programs with active cytotoxic and inflammatory chemotactic functions, whereby the TGF-β signaling pathway constitutes a critical regulatory module within the complex network governing their phenotypic heterogeneity. Guan et al. demonstrated in a murine model that ZEB1 and ZEB2 differentially regulate T-cell memory and exhaustion phenotypes, respectively. In vitro experiments revealed that TGFB1 stimulation upregulated ZEB1 while downregulating ZEB2 in P14 + CD8 + T cells, whereas TGFBR2 knockdown enhanced ZEB2 expression and promoted an exhausted phenotype. However, this study did not address the effects of these experimental conditions on SMAD2/3 or SMAD7 expression [50]. While our findings on the role of TGF-β signaling activation in CD8 + T-cell terminal differentiation contrast with the abovementioned study, the observations that ZEB1 preserves T-cell self-renewal capacity and is more highly expressed in Tn and Tcm/Tem cells, whereas ZEB2 promotes T-cell terminal differentiation and is significantly upregulated in Temra cells, are corroborated by established literature [16, 50, 66]. On the basis of prior evidence, we propose that during physiological aging, TGF-β pathway activation drives T-cell senescence and Temra development, as evidenced by the coordinated upregulation of TGFBR1/3 receptors, the SMAD3 signaling molecule, ZEB2 transcription factor, and the ligand TGFB1. Concurrently, high SMAD7 expression mediates antagonistic regulation of TGF-β signaling, thereby enhancing the cytotoxic function of Temra [18–20]. Collectively, via bulk RNA-seq analysis, we validated the utility of age-specific regulons (derived from scRNA-seq) in distinguishing the expression profiles of purified T-cell subsets. This approach mitigates the sparsity of expression data for signaling proteins (e.g., SMAD2, SMAD3, SMAD7) in scRNA-seq, thereby advancing our understanding of age-specific transcriptional regulatory networks, as well as the functional associations between TGF-β signaling pathway components and T-cell subsets at distinct differentiation stages—particularly regarding Temra function. Given the complexity of the TGF-β signaling pathway, we further postulate that multiple mechanisms may contribute to Temra phenotypic complexity: (1) ZEB2-mediated feedback regulation of TGF-β pathway upstream components [67], (2) cross-species heterogeneity in TGF-β signaling, (3) disparities between in vitro and in vivo immune microenvironments, (4) epigenetic regulation and post-translational modifications [62], and (5) crosstalk with other signaling pathways (e.g., the Wnt, NF-κB, MAPK, and JAK-STAT pathways) [68]. The above speculative extension regarding the regulation of the Temra phenotype by TGF-β signaling pathway may provide a basis for future investigations into the role of this pathway in T-cell immune senescence. Conclusions We utilized a progressive analytical approach to systematically identify aging-related regulons with synergistic transcriptional regulatory effects and biological background consistency. Using this framework, we more precisely delineated the patterns of T-cell transcriptional regulation between young and old populations, ultimately identifying TGFB1, TGFBR1, TGFBR3, SMAD3, ZEB2, and SMAD7 as potential biomarkers for senescence and proinflammatory phenotypes in Temra. These findings may provide a foundation for in-depth exploration of TGF-β-mediated aging-related diseases and the development of immunotherapeutic strategies. Methods Dataset acquisition and preprocessing The scRNA-seq data were downloaded from the NCBI GEO database (GSE157007). For specifics on sample processing, genome alignment, and library construction, please refer to the original publication [22]. For upstream quality control (QC), thresholds were set according to the original article (nFeature_RNA = 200–3000, percent_mt < 10%), and doublets were removed via scDblFinder [69] (version 1.8.0). We performed canonical correlation analysis (CCA) via Seurat [70] (version 4.3.0) with the built-in FindIntegrationAnchors function (features = 3000, vars.to.regress = c("percent_mt", "nCount_RNA")). We then delineated the landscape of the immune cells in PBMCs by referencing the original marker panel [22]. Subsequently, we extracted T cells and performed secondary dimensionality reduction, clustering, and batch effect removal with Harmony [71] (version 1.2.0) (features = 2000, vars.to.regress = c("percent_mt", "nCount_RNA")). Following T-cell subtype annotation, we focused on the T cells from the young and old groups. We randomly down-sampled the T cells from the old group (n = 33,722), yielding 23,920 T cells (young group: 12,679, old group: 11,241) for further analysis. This approach eliminates the influence of disparities in cell number on the results of the downstream analyses. With uniform manifold approximation and projection (UMAP), we visualized the distribution of cell types across age groups and calculated the difference in the proportion of each T-cell subpopulation between the young and old groups. The bulk-cell RNA-seq data were derived from peripheral blood CD8 + T-cell samples obtained via flow sorting from a cohort of 28 healthy elderly donors (90 ≥ aged ≥ 65 years, NCBI GEO database: GSE241027). The sorted Tn (CCR7hi, CD45RAhi), Tcm_Tem (CCR7hi/lo, CD45RAlo), and Temra (CCR7lo, CD45RAhi) strains were subjected to total RNA sequencing as described in the original study. For detailed data processing workflows, please refer to the original publication [45]. Generation of MEs via hdWGCNA We conducted hdWGCNA [72] (version 0.4.05) analysis on the T-cell subtypes excluding CD8 + Tem_ISGs and γδT cells.Step 1: Construct hdWGCNA files with gene_select = "fraction" and fraction = 0.05 and then generate metacells using parameters k = 25, max_shared = 15, group.by = c("Celltype"), and min_cells = 100. The metacells were subsequently integrated for batch-effect removal.Step 2: Calculate the soft threshold and build a weighted gene co-expression network with soft_power = 3. MEs were retained, excluding the gray module.Step 3: Assess the differential expression of MEs across T-cell subtypes or age groups, map ME activity to UMAP plots, and compute expression signature scores for genes in each ME for every cell.Step 4: Functional enrichment analysis was performed for each ME, with the top 6 enriched pathways displayed for the respective modules. Step 1: Construct hdWGCNA files with gene_select = "fraction" and fraction = 0.05 and then generate metacells using parameters k = 25, max_shared = 15, group.by = c("Celltype"), and min_cells = 100. The metacells were subsequently integrated for batch-effect removal. Step 1: Construct hdWGCNA files with gene_select = "fraction" and fraction = 0.05 and then generate metacells using parameters k = 25, max_shared = 15, group.by = c("Celltype"), and min_cells = 100. The metacells were subsequently integrated for batch-effect removal. Step 2: Calculate the soft threshold and build a weighted gene co-expression network with soft_power = 3. MEs were retained, excluding the gray module. Step 2: Calculate the soft threshold and build a weighted gene co-expression network with soft_power = 3. MEs were retained, excluding the gray module. Step 3: Assess the differential expression of MEs across T-cell subtypes or age groups, map ME activity to UMAP plots, and compute expression signature scores for genes in each ME for every cell. Step 3: Assess the differential expression of MEs across T-cell subtypes or age groups, map ME activity to UMAP plots, and compute expression signature scores for genes in each ME for every cell. Step 4: Functional enrichment analysis was performed for each ME, with the top 6 enriched pathways displayed for the respective modules. Step 4: Functional enrichment analysis was performed for each ME, with the top 6 enriched pathways displayed for the respective modules. Identification of age-group-specific regulons through SCENIC analysis We utilized SCENIC [73] (version 1.3.1) to construct a gene regulatory network (GRN) based on the identified MEs.Step 1: Save the file containing the grouping conditions, initialize the analysis environment, and load the CisTarget [74] (version 1.14.0) genomic references.Step 2: Filter genes on the basis of low expression levels and frequencies via the following thresholds: minCountsPerGene = 3 × 0.03 × ncol(exprMat) and minSamples = ncol(exprMat) × 0.01. The intersection of the remaining genes with the defined ME was used for GRN construction.Step 3: Build regulatory modules (regulons) for transcription factors (TFs) and calculate the regulon activity per cell via AUCell.Step 4: Compute regulon specificity scores (RSSs) under grouping conditions and generate RSS plots.Step 5: GRN genes with age-group differences were screened via Wilcoxon’s test (P < 0.0001) and then intersected with ME1 and ME2 to generate young-specific regulons and with ME3 to ME5 to generate old-specific regulons, respectively.Step 6: Calculate AUCell-based gene expression signature scores for young-specific and old-specific regulons in each cell.Step 7: Regulatory networks of young-specific and old-specific regulators were mapped separately to demonstrate the regulatory relationships between TFs and their targetsStep 8: Visualize the TF-target motif logos with JASPAR (version 0.99.10) as a reference. Step 1: Save the file containing the grouping conditions, initialize the analysis environment, and load the CisTarget [74] (version 1.14.0) genomic references. Step 1: Save the file containing the grouping conditions, initialize the analysis environment, and load the CisTarget [74] (version 1.14.0) genomic references. Step 2: Filter genes on the basis of low expression levels and frequencies via the following thresholds: minCountsPerGene = 3 × 0.03 × ncol(exprMat) and minSamples = ncol(exprMat) × 0.01. The intersection of the remaining genes with the defined ME was used for GRN construction. Step 2: Filter genes on the basis of low expression levels and frequencies via the following thresholds: minCountsPerGene = 3 × 0.03 × ncol(exprMat) and minSamples = ncol(exprMat) × 0.01. The intersection of the remaining genes with the defined ME was used for GRN construction. Step 3: Build regulatory modules (regulons) for transcription factors (TFs) and calculate the regulon activity per cell via AUCell. Step 3: Build regulatory modules (regulons) for transcription factors (TFs) and calculate the regulon activity per cell via AUCell. Step 4: Compute regulon specificity scores (RSSs) under grouping conditions and generate RSS plots. Step 4: Compute regulon specificity scores (RSSs) under grouping conditions and generate RSS plots. Step 5: GRN genes with age-group differences were screened via Wilcoxon’s test (P < 0.0001) and then intersected with ME1 and ME2 to generate young-specific regulons and with ME3 to ME5 to generate old-specific regulons, respectively. Step 5: GRN genes with age-group differences were screened via Wilcoxon’s test (P < 0.0001) and then intersected with ME1 and ME2 to generate young-specific regulons and with ME3 to ME5 to generate old-specific regulons, respectively. Step 6: Calculate AUCell-based gene expression signature scores for young-specific and old-specific regulons in each cell. Step 6: Calculate AUCell-based gene expression signature scores for young-specific and old-specific regulons in each cell. Step 7: Regulatory networks of young-specific and old-specific regulators were mapped separately to demonstrate the regulatory relationships between TFs and their targets Step 7: Regulatory networks of young-specific and old-specific regulators were mapped separately to demonstrate the regulatory relationships between TFs and their targets Step 8: Visualize the TF-target motif logos with JASPAR (version 0.99.10) as a reference. Step 8: Visualize the TF-target motif logos with JASPAR (version 0.99.10) as a reference. Pseudotime trajectory analysis We evaluated the dynamic expression of genes from age-group-specific regulons following the differentiation process with monocle [75] (version 2.22.0).Step 1: Create a new CellDateSet object on the basis of the number of seurat.obj.Step 2: Calculate the size factor for the scRNA-seq data and estimate dispersions and then set the global expression detection threshold to be used with this CellDataSet (min_expr = 1.0).Step 3: Generation of differentiation-related genes (DRGs) via differentialGeneTest.Step 4: Construction of discriminative dimensionality reduction (DDRT) via reduceDimension (max_components = 2, reduction.method = 'DDRTree').Step 5: Calculate where each cell falls within the trajectory by orderCells.Step 6: Mapping of “Pseudotime,” “Celltype,” and “Old-specific regulons” and “Young-specific regulons” signature scores to the DDRT for visualization.Step 7: Clustering of DRGs according to expression patterns via plot_pseudotime_heatmap. We also analyzed the association of the differentiation potential of T-cell subtypes with age-group-specific regulons during differentiation via CytoTRACE [76] (version 0.3.3). Step 1: Create a new CellDateSet object on the basis of the number of seurat.obj. Step 1: Create a new CellDateSet object on the basis of the number of seurat.obj. Step 2: Calculate the size factor for the scRNA-seq data and estimate dispersions and then set the global expression detection threshold to be used with this CellDataSet (min_expr = 1.0). Step 2: Calculate the size factor for the scRNA-seq data and estimate dispersions and then set the global expression detection threshold to be used with this CellDataSet (min_expr = 1.0). Step 3: Generation of differentiation-related genes (DRGs) via differentialGeneTest. Step 3: Generation of differentiation-related genes (DRGs) via differentialGeneTest. Step 4: Construction of discriminative dimensionality reduction (DDRT) via reduceDimension (max_components = 2, reduction.method = 'DDRTree'). Step 4: Construction of discriminative dimensionality reduction (DDRT) via reduceDimension (max_components = 2, reduction.method = 'DDRTree'). Step 5: Calculate where each cell falls within the trajectory by orderCells. Step 5: Calculate where each cell falls within the trajectory by orderCells. Step 6: Mapping of “Pseudotime,” “Celltype,” and “Old-specific regulons” and “Young-specific regulons” signature scores to the DDRT for visualization. Step 6: Mapping of “Pseudotime,” “Celltype,” and “Old-specific regulons” and “Young-specific regulons” signature scores to the DDRT for visualization. Step 7: Clustering of DRGs according to expression patterns via plot_pseudotime_heatmap. We also analyzed the association of the differentiation potential of T-cell subtypes with age-group-specific regulons during differentiation via CytoTRACE [76] (version 0.3.3). Step 7: Clustering of DRGs according to expression patterns via plot_pseudotime_heatmap. We also analyzed the association of the differentiation potential of T-cell subtypes with age-group-specific regulons during differentiation via CytoTRACE [76] (version 0.3.3). Functional enrichment and calculation of signature scores of gene sets We performed functional enrichment analysis on specific gene sets or genes in regulons via clusterProfiler [77] (version 4.2.2). We obtained biological process (BP) annotation from the C5 collection via msigdbr [78] (version 7.5.1, https://www.gsea-msigdb.org/gsea/msigdb) for gene set scoring and performed functional enrichment of MEs derived from hdWGCNA via the “GO_Biological_Process_2023” database in enrichR [79] (version 3.2, https://maayanlab.cloud/Enrichr/). Differential functional pathways between the young and old groups were initially screened via irGSEA [80] (version 3.2.5). In addition, we calculated signature scores for MEs, age-group-specific regulons, and TGF-β signaling pathways via UCell [81] (version 2.0.0). The signature score of age-group-specific regulons was calculated via the gsva (method = "ssgsea") function in GSVA (version 1.42.0). Additionally, we calculated the activity scores for multiple classical pathways on the basis of the DEGs from comparisons of Temra versus Tn, or Temra versus Tcm/Tem, using the built-in multiple linear regression function run_mlm from decoupleR [82] (version 2.0.1). Data format conversion and statistical analysis We organized and generated files for plotting or statistical analyses via dplyr (version 1.1.4) and conducted statistical analyses via the ggpubr (version 0.6.0) and stats (version 4.1.0) R packages. Specifically, two-group p-values were calculated via t-test or Wilcoxon’s test, multiple-group comparisons were performed via the Kruskal–Wallis test, and correlation analysis was performed via Pearson’s correlation test. Visualization We visualized our data via the following R packages: ggplot2 (version 3.5.0), ComplexHeatmap (version 2.10.0), ggalluvial (version 0.12.3), ggrepel (version 0.9.5), and igraph (version 1.6.0). The schematic in Fig. 6 was drawn via BioRender (https://www.biorender.com/). Supplementary Information Additional file 1. Supplementary Figs. 1–11. Fig. S1 A overview of the PBMC scRNA-seq data used in this study. Fig. S2 Landscape of T cells from the young and old groups. Fig. S3 Network view of young-specific and old-specific regulons. Fig. S4 TF binding motifs discovered in the promoter regions of target genes. Fig. S5 Comparison of TGF-β signaling-related pathway activities in T-cell subtypes across distinct age groups. Fig. S6 Differential expression of genes from the young-specific and old-specific regulons. Fig. S7 Expression characterization of function-related genes in distinct T-cell subtypes (Tn, Tcm/Tem, and Temra). Fig. S8 Expression characterization of key components of the TGF-β signaling pathway in distinct T-cell subtypes (Tn, Tcm/Tem, and Temra). Fig. S9 Expression correlation between ZEB2 and other T-cell subtype-specific genes or genes in the TGF-β signaling pathway through T-cell differentiation. Fig. S10 Heatmap showing the pairwise expression correlations between genes from the age-group-specific regulons and the TGF-β signaling pathway. Fig. S11 Signaling pathways with different activities between T-cell subtypes. Additional file 1. Supplementary Figs. 1–11. Fig. S1 A overview of the PBMC scRNA-seq data used in this study. Fig. S2 Landscape of T cells from the young and old groups. Fig. S3 Network view of young-specific and old-specific regulons. Fig. S4 TF binding motifs discovered in the promoter regions of target genes. Fig. S5 Comparison of TGF-β signaling-related pathway activities in T-cell subtypes across distinct age groups. Fig. S6 Differential expression of genes from the young-specific and old-specific regulons. Fig. S7 Expression characterization of function-related genes in distinct T-cell subtypes (Tn, Tcm/Tem, and Temra). Fig. S8 Expression characterization of key components of the TGF-β signaling pathway in distinct T-cell subtypes (Tn, Tcm/Tem, and Temra). Fig. S9 Expression correlation between ZEB2 and other T-cell subtype-specific genes or genes in the TGF-β signaling pathway through T-cell differentiation. Fig. S10 Heatmap showing the pairwise expression correlations between genes from the age-group-specific regulons and the TGF-β signaling pathway. Fig. S11 Signaling pathways with different activities between T-cell subtypes. Additional file 2. Table S1 Summary of peripheral blood donors and number of cells in the scRNA-seq data per sample after quality control. Additional file 2. Table S1 Summary of peripheral blood donors and number of cells in the scRNA-seq data per sample after quality control. Additional file 3. Table S2 Differentially expressed genes (DEGs) of all cell types of PBMCs across age groups and T cell types of the old and young groups. Additional file 3. Table S2 Differentially expressed genes (DEGs) of all cell types of PBMCs across age groups and T cell types of the old and young groups. Additional file 4. Table S3 Functional enrichment analysis of all cell types across age groups, differentiation-related gene (DRG) clusters or age-group-specific regulons. Additional file 4. Table S3 Functional enrichment analysis of all cell types across age groups, differentiation-related gene (DRG) clusters or age-group-specific regulons. Additional file 5. Table S4 Age-group-specific regulons and the regulon specificity score (RSS) of the young and old groups. Additional file 5. Table S4 Age-group-specific regulons and the regulon specificity score (RSS) of the young and old groups. Additional file 6. Table S5 DRGs of T cell subtypes from young and old groups by monocle. Additional file 6. Table S5 DRGs of T cell subtypes from young and old groups by monocle. Additional file 7. Table S6 Expression matrix and metadata of GSE241027 dataset. Additional file 7. Table S6 Expression matrix and metadata of GSE241027 dataset.