Organism-wide cellular dynamics and epigenomic remodeling in mammalian aging

Abstract

Aging leads to functional decline across tissues, often accompanied by profound changes in cellular composition and cell-intrinsic molecular states. However, a comprehensive catalog of how the population of individual cell types change with age and the associated epigenomic dynamics is lacking. Here, we constructed a single-cell chromatin accessibility atlas consisting of ~7 million cells from 21 tissue types spanning three age groups in both sexes. This dataset revealed 536 main cell types and 1,828 finer-grained subtypes, defined by unique chromatin accessibility landscapes at ~1.3 million cis-regulatory elements. We observed widespread remodeling of immune lineages, with increases in plasma cells and macrophages, and depletion of T and B cell progenitors. Additionally, non-immune cell populations, including kidney podocytes, ovary granulosa cells, muscle tenocytes and lung aerocytes, showed marked reductions with age. Meanwhile, many subtypes changed synchronously across multiple organs, underscoring the potential influence of systemic inflammatory signals or hormonal cues. At the molecular level, aging was marked by thousands of differentially accessible regions, with the most concordant changes shared across cell types linked to genes related to inflammation or development. Putative upstream factors, such as intrinsic shifts in transcription factor usages and extrinsic cytokine signatures, were identified. Notably, around 40% of aging-associated main cell types and subtypes showed sex-dependent differences, with tens of thousands of chromatin accessibility peaks altered exclusively in one sex. Together, these findings present a comprehensive framework of how aging reshapes the chromatin landscape and cellular composition across diverse tissues, offering a comprehensive resource for understanding the molecular and cellular programs underlying aging and supporting the exploration of targeted therapeutic strategies to address age-related dysfunction.

Introduction:

Aging is the leading risk factor for many diseases, including neurodegenerative disorders, cardiovascular diseases, metabolic conditions, and cancer(1). This association underscores the potential of therapies targeting the aging process itself to delay or prevent various age-related diseases, rather than focusing on specific diseases or organs. However, this approach is challenging due to the complexity of aging, which involves widespread changes across various systems and organs (2, 3). Each organ consists of hundreds of unique cellular states with specialized functions, each undergoing varied aging-associated changes. Therefore, a systematic characterization of aging-associated cell states—including their genetic and epigenetic alterations underlying population dynamics—is essential for identifying cellular and molecular targets that could guide the development of targeted therapeutic interventions.

Recent advancements in single-cell genomics have significantly enhanced our ability to map aging-associated cell states. For example, single-cell transcriptomics analyses have enabled us to profile alterations in thousands of genes across millions of cells as they age (4–7). However, these analyses typically focus on protein-coding genes and overlook the impact of non-coding genetic elements in regulating cell-type-specific dynamics in aging. Moreover, current single-cell atlas studies often focus on a limited set of organs (e.g., liver, kidney), largely because some other tissues, like the pancreas, are difficult to profile due to high RNA degradation (8, 9). This limitation impedes our ability to assess aging across the entire organism. As a result, we still lack a comprehensive analysis of cell population dynamics and chromatin landscape changes throughout the mammalian body for understanding how non-coding regions, particularly cis-regulatory DNA elements, affect cell states and population dynamics as they age.

Single-cell ATAC-seq(10), which analyzes genome-wide chromatin accessibility at the single-cell level, has enabled the mapping of the cell-type-specific chromatin landscape in a range of mammalian tissues(11–14), aiding in the identification of disease-associated non-coding regions and their corresponding cell types. Among varied scATAC-seq approaches, single-cell ATAC-seq by combinatorial indexing (sci-ATAC-seq), allows for high-throughput analysis of tens of thousands of cells from highly heterogeneous tissues in a single experiment(11). With a further optimized version of sci-ATAC-seq (EasySci-ATAC), we recently mapped the cell-type-specific chromatin landscape and aging-associated cell population changes in the mammalian brain(15). Building on this work, we now extend our analysis beyond a single organ to explore aging-associated changes in cell populations across the entire organism, identifying changes in cellular chromatin states, population dynamics, and associated genetic elements—including cis-regulatory elements and transcription factor motifs—across different ages and sexes.

Results

An organismal, single-cell chromatin accessibility atlas of aging

To map aging-associated cell population dynamics and the cell-type-specific chromatin landscape changes at the organismal scale, we developed an optimized EasySci-ATAC(15) protocol that enables high-throughput analysis of mouse tissues across the entire organism (Methods). Next we applied this approach to profile 552 samples collected from thirty-two individual mice representing three age groups (1-month, 5-month, and 21-month)(Table S1). Each group contained 8–12 sex-balanced biological replicates, except for the female ovary and uterus samples. Twenty-one different tissue types spanning major biological systems were profiled, including the immune system (i.e., bone marrow, spleen, thymus), the cardiovascular system (i.e., heart), the respiratory system (i.e., lung), the digestive system (i.e., liver, pancreas, esophagus, stomach, intestine, cecum, colon), the urinary system (i.e., kidney), the sensory system (i.e., eye), the reproductive system (i.e., ovary/uterus), the musculoskeletal system (i.e., muscle), the integumentary system (i.e., skin), and the adipose tissues, which includes brown adipose tissue (BAT) and white adipose tissues (WAT) in various anatomical locations (i.e., gonadal, inguinal, and mesenteric) (Fig. 1A). Of note, the brain was not included here, as it has been extensively analyzed in our previous study(15). Also, to enable the investigation of newborn cell dynamics in a companion study, we labeled mice with 5-Ethynyl-2-deoxyuridine (EdU) before tissue collection, similar to our previous work (16). This allowed single-cell chromatin accessibility profiling of both sorted DAPI singlets (‘all’ cells, this study) and EdU+ nuclei (‘newborn’ cells, analyzed in the companion study) from the same set of samples.

Figure 1. An organismal, single-cell chromatin accessibility atlas of aging.

(A) Experimental scheme to construct an organismal cell atlas of chromatin accessibility across different ages and both sexes.

(B) Barplot showing the number of cells profiled in each tissue, colored by the sources. DAPI singlets represent the global cell population investigated in this study.

(C) UMAP visualization of cells from each tissue, colored by 144 unique cell types collapsed from all tissues.

The extracted nuclei from each tissue sample were sorted and combinatorial barcoded through indexed tagmentation, indexed ligation, and indexed PCR (Fig. 1A). The resulting libraries were sequenced on the Illumina NovaSeq 6000 system, yielding a total of 95 billion paired-end raw reads. After filtering out low-quality cells and doublets similar to our previous study (15) (Fig. S1A), we obtained chromatin accessibility profiles for a total of 10,956,311 single cells, including 6,839,086 DAPI singlets and 4,117,225 cells gated on EdU positivity (Fig. 1B and Fig. S1B). The median number of cells recovered per tissue range from 115,060 in the esophagus to 1,173,739 in the kidney (Fig. 1B). With a low sequencing duplication rate of only 20.1%, we detected a median of 3,031 unique ATAC-seq fragments per nucleus. On average, 29.8% of reads mapped to promoter regions (within ±1 kb of the transcription start site, TSS), aligning with performances from previous single-cell ATAC-seq studies (11, 15) (Fig. S1C and S1D). Also, the pseudo bulk ATAC-seq profiles are clustered by tissue type rather than by individual mice, indicating the low technical batch effect across individuals (Fig. S1E).

Next, we identified and annotated heterogeneous cell types within each tissue. Applying SnapATAC2(17), we performed dimensionality reduction and Leiden clustering, and manually curated the annotation based on the gene accessibility of known markers (Table S2, Fig. S2). Of note, we utilized the full dataset containing 10,956,311 cells for clustering and annotations. Chromatin accessibility alone enabled the identification of diverse cell types, including tissues that are challenging to profile by scRNA-seq due to high RNase levels, such as ductal cells (marked by Ccn2), beta-cells (marked by Gcg), alpha-cells (marked by Isn1) and delta-cells (marked by Sst) in the pancreas (18, 19) (Fig. S2A). In total, we annotated 536 organ-level main cell types, which were consolidated into 144 unique categories, with a median of 25 cell types per tissue and 4,860 nuclei per cell type (Fig. 1C). While most cell types were restricted to one organ (n = 102), we identified eighteen broadly distributed cell types shared by more than ten tissues, including immune cells (e.g., T cells, B cells), stromal cells (e.g., fibroblasts, smooth muscle cells) and endothelial cells (e.g., vascular and lymphatic endothelial cells).

To globally characterize the cell-type-specific chromatin accessibility landscape, we combined reads from cells of the same type and same tissue, performed peak calling, and iteratively merged the resulting peaks to a universal set of 1,341,077 open chromatin regions (±250 base pairs from peak summit; Methods). The genomic distribution of these peaks was largely consistent with previous sn-ATAC-seq profiling (14) (Fig. S3A): they predominantly located within introns (n=648,244; 48.3%) and intergenic regions (n=404,499; 30.1%), and to a lesser extent at TSS-promoter regions (n=34,487; 2.57%). These regions occupy 12.3% of the genome and intersect with 89.1% of the 339,815 CREs identified by the ENCODE consortium (20) (Fig. S3B). Using this universal peak set, cells of the same lineage clustered together across tissues, such as immune cells (e.g., T/NK cells, B cells, plasma cells, macrophages) and stromal cells (e.g., adipocytes, pericytes), further validating the accuracy of cell annotations (Fig. S3C).

Using an entropy-based method(12), we identified 320,304 cis-regulatory elements (CREs) that are cell-type-specific (Fig. S3D, Table S3; Methods). We next applied chromVar (21) to pinpoint TF motifs specific to each cell type (Fig. S3E). The results were consistent with established findings, highlighting the enrichment of Runx3 in cytotoxic T cells and NK cells (22), Sfpi1 in general myeloid cells (23), Irf4 in plasma cells (24), and Ebf1 in B cells and pericytes (25, 26) (Fig. S3E). The enrichment of these motifs was further validated by gene activity and cross-validated across different organs (Fig. S3F). Interestingly, we also observed TFs with a negative correlation between motif accessibility and gene activity, such as Gfi1b in erythroid cells (Fig. S3F), consistent with its role as a transcription repressor during erythroid development in humans(11, 27).

Next, we sought to use cell-type-specific accessible peaks to interpret genetic variants associated with complex traits and diseases. Specifically, we first mapped human SNPs to orthologous coordinates in the mouse genome and then applied linkage disequilibrium score regression (LDSC) to assess GWAS SNP enrichment within cell-type-specific accessible peaks, similar to previous studies(28) (Methods). Given that many GWAS SNPs reside in non-coding regions, this analysis provides insights into their functional relevance by prioritizing cell types of interest (Fig. S3G). For instance, Alzheimer’s disease SNPs were significantly enriched in microglia-specific CREs, consistent with prior studies(29). Meanwhile, SNPs associated with multiple sclerosis were enriched in B cells, T cells, and plasma cells(30, 31), while LDL-associated SNPs were linked to hepatocytes(32) and height-associated SNPs were linked to tenocytes. SNPs for type II diabetes showed the strongest association with pancreatic beta cells(33), and those for hypertension with juxtaglomerular cells in the kidney(34) (Fig. S3G).

Age- and sex- dependent cell population change at the main cluster level

By quantifying cell type abundances in tissues from 5-month-old mice, we observed significant variability across organs, with both highly abundant populations (e.g., hepatocytes, cecal epithelial cells, brown adipocytes) and rare, organ-specific cell types detected (Fig. 2A). Notably, 43 organ-specific cell types constituted less than 1% of their tissue’s cellular composition, such as Ascl1+ neuroendocrine cells in the lung (0.039%) and Gja8+ lens epithelial cells in the eye (0.051%), underscoring the high sensitivity of scATAC-seq to identify low abundant but critical cell populations.

Figure 2. Age- and sex-dependent cell population changes at the main cell type level.

(A) Histogram showing the distribution of cell type specific fractions within tissue. Only cells from adult mice (5-month-old) were used in this quantification.

(B) K-means classifiers were trained to distinguish male and female cells for each main cell type using spectral embedding calculated by SnapATAC2 (17). Scatterplot comparing the area under the curve (AUC) values of models between adult and aged cells. Cell types with high sex dimorphism in both age groups are labeled.

(C) Left: UMAP plots of all kidney cells, colored by sex (right), highlighting distinct chromatin states between sexes of proximal tubule cells (PT). Right: Heatmap displaying accessibility of autosomal peaks with sex- and cell-type-specific patterns, conserved across ages.

(D) Venn diagram showing the number of main cell types whose relative proportions in corresponding tissues are significantly associated with age or age-sex interactions.

(E) Left: Dotplot showing all significantly altered main cell types from (D) across tissues, colored by Pearson correlation between their relative proportion and age. Cell types with significant age or age-sex interactions are marked. Right: Barplot summarizing the number of tissues with significant changes for each cell type, colored by direction (up/down) or interaction.

(F) Scatterplot illustrating examples of main cell types with significant age-related proportional changes (left), alongside validation using published snRNA-seq data (7) for the same cell type and tissue (right).

(G) Heatmap showing relative proportions of main cell types with significant age-sex interactions, including female-biased (left) and male-biased (right) cell types.

(H) Scatterplot highlighting cell types that change in the same direction with aging but exhibit sex-dependent magnitude.

(I-J) Scatterplots of two kidney proximal tubule cell types with opposite sex-specific proportional trends (I) and snRNA-seq validation results (J).

We next examined sex dimorphism for each main cell type. While most cell types exhibited similar population sizes between sexes (Pearson correlation r = 0.99, 0.98, 0.95 for young, adult and aged mice, respectively) (Fig. S4A), we identified a subset of cell types displaying strong sex-specific differences in chromatin states. Specifically, we trained k-nearest neighbor (KNN) classifiers to distinguish female and male cells of the same age for each cell type based on chromatin embeddings (Methods). We then used the area under the curve (AUC) metric to quantify classification accuracy, and cell types with AUC > 0.9 were considered to exhibit high sex-associated disparity (Fig. 2B). This approach revealed high sex-associated chromatin states in hepatocytes, proximal tubule cells, type IIB myonuclei, and multiple cell types from the gonadal white adipose tissue (e.g., adipocytes, adipose stem and progenitor cells, mesothelial cells, and smooth muscle cells)(Fig. 2B), mirroring previous reports based on gene expressions (7, 35, 36). For example, proximal tubule (PT) cells in the kidney, including the general states spanning spatially connected segments (PTS1, PTS2 and PTS3; marked by Slc34a1) and a distinct state of segment 3 (PTS3T2, marked by Slc22a7) (37) was separated into clusters determined by sexes with unique marker gene accessibility (Fig. 2C and Fig. S4B). Furthermore, extensive changes underlying the sex difference were found on autosomal chromosomes and exhibited high cell-type-specificity (Fig. 2C, right).

We further investigated how aging influences the cell population dynamics across various tissues. Using linear regression, which included an age-sex interaction term, we pinpointed specific cell types that undergo significant changes with age within each organ (Methods). This analysis revealed that 146 organ-specific cell types are significantly altered upon aging (FDR of 0.05, R² > 0.4), with 55 of these showing significant differences between males and females in aging (Fig. 2D–E; Table S4). Although aging-associated changes were noted in all profiled tissues, certain organs (e.g., ovary/uterus, liver, thymus) exhibited a higher proportion of cell types with significant age-related changes (Fig. S5A), suggesting higher susceptibility to the aging effects in these organs.

A total of 62 main cell types were significantly expanded in aging regardless of sex (Fig. 2E, Fig. S5B top). Notably, the majority (68%) of these aging-expanded cell types are immune cells distributed across various organs (Fig. S5C). Plasma cells and macrophages were among the most significantly expanded, showing an increased relative proportion in 14 and 12 tissues, respectively, indicating a widespread increase across the organism (Fig. 2E). Several less characterized cell population changes were detected across multiple organs, such as the expansion of dendritic cells in the BAT and stomach, and ILC3s in the liver, colon, and cecum (Fig. S5D). Additionally, we observed the expansion of other immune cell types with more tissue-specific distributions, such as Kupffer cells in the liver and gamma-delta (γδ) T cells in the iWAT (Fig. S5D), consistent with previous studies(38–42). In addition to immune cells, we detected a notable expansion of various non-immune cell types during aging. For example, specialized myonuclei associated with neuromuscular junctions (NMJs) and myotendinous junctions showed increased abundance (Fig. 2E and 2F left), potentially acting as a compensatory mechanism to maintain muscle integrity and contractile function under age-related stress(43). Furthermore, we observed a rise in enteroendocrine cells within the stomach (Fig. 2E and 2F left), which could contribute to dysregulated digestive processes in aged individuals (44). Notably, these expansions were consistent across different sexes and were further validated by re-analysis of an independent single-cell transcriptome atlas of aging(7) (Fig. 2F, right).

The 29 aging-depleted cell types are predominantly tissue-specific (Fig. 2E, Fig. S5B bottom). This includes a notable reduction in immune cell progenitors such as pro-/pre-B cells in the bone marrow, which contributes to the age-related decline in immune function (45). Similarly, in the thymus, there is a decrease in immature T cells during the double-negative (DN) and double-positive (DP3) stages, indicative of diminished T cell production in the aging thymus (46). Other aging-depleted immune cell types, like NK cells and pDC cells in the spleen, were also detected, aligning with previous findings (47, 48) (Fig. 2E, Fig. S5B, S5E). Meanwhile, multiple non-immune cell types are significantly depleted across organs, especially in the ovary/uterus, skin, and lung (Fig. S6A). In the ovary/uterus, for example, granulosa cells and theca cells that collaborate in sex hormones production decline with age, together with basal epithelial cells and vascular endothelial cells. (Fig. 2E, Fig. S6B). This aligns with the decrease in ovarian follicles that characterizes reproductive aging (49, 50). In the skin, aging is associated with fewer melanocytes—a trend consistent with a 10–20% reduction per decade in individuals over 25–30 years of age (51)—as well as lower proportions of dermal papilla cells, outer root sheath cells, and lymphatic endothelial cells (Fig. 2E, Fig. S6B), the latter consistent with reduced skin lymphatic density observed in aging(52). Similarly, the aged lung exhibits declines in stromal cell types associated with blood vessels, including pericytes, smooth muscle cells (Fig. 2E, Fig. S6B), and a specialized endothelial cell type known as aerocytes(53) (Fig. 2F). These cells are important for maintaining proper vascular structure and function, suggesting that their depletion may contribute to compromised blood flow regulation in the elderly lung(54).

Moreover, we observed a decline in other functional cell types, such as kidney podocytes that filter primary urine from plasma (Fig. 2E–F), aligning with previous studies(55). The depleted cell populations also include a group of less characterized Ncam1+ parietal epithelial cells (Fig. 2E, Fig. S6B), which serve as progenitor cells capable of proliferation and differentiation into podocytes (56). The depletion of both mature podocytes and their precursor cells could contribute to the impaired capacity of the kidney to regenerate and maintain proper filtration over time. Additionally, aging-depleted cell types include liver hepatocytes (57), muscle tenocytes(58), muscle satellite cells (59), stomach enteric neurons, and adipocytes from multiple organs (e.g., lung, gWAT, and ovary/uterus) (Fig. 2E–F and S5B), validated by re-analysis of external single-cell transcriptome dataset(7) (Fig. 2F). Interestingly, some of these aging-decreased cell types (e.g., muscle satellite cells and tenocytes, bone marrow pro-/pre-B cells) exhibited reductions of more than 50% at relatively early stages (between 1 and 5 months) (Fig. 2F, S5B bottom, S6B), highlighting an early onset of aging impacts on the population of these cell types.

We further analyzed 55 cell types displaying significant age-sex interactions. Most sex-specific cell types (35 out of 55) are immune cells, including 19 that preferentially expanded in aging females and 15 in aging males (Fig. 2G). For instance, aged male mice showed significant increases in neutrophils (in the lung and intestine) and eosinophils in the muscle (Fig. 2G), consistent with male-specific expansions of these cells observed in prior human studies(60, 61). Conversely, aged females demonstrated increased populations of basophils (in the liver) and NK cells (in the pancreas and bone marrow) (Fig. 2G). Additionally, some cell types exhibited similar trends in both sexes but differed in the extent of change in different organs. For example, ILC2 cells in the bone marrow and hepatic stellate cells in the liver expanded more significantly in males, whereas the increase of interstitial macrophages in the lung and depletion of naive/memory B cells in the spleen was more pronounced in the female (Fig. 2G–H). Interestingly, certain cell types showed opposite trends between the sexes. For instance, the aforementioned specialized proximal tubule (PT) cell type (PT S3T2, characterized by Slc22a7) increased in females but decreased in males, whereas the other PT cells (PT S1S2S3, marked by Slc34a1) exhibited the opposite pattern (Fig. 2I). The findings were further validated by re-analysis of independent single-cell transcriptome atlas profiling both adult and aged tissues (7) (Fig. 2J).

Age- and sex- dependent cell population change at the subtype level

We further explored cell population dynamics at the subtype level. First, we re-clustered each main cell type using SnapATAC2(17), and manually curated cellular subtypes based on cluster-specific gene accessibility. Our analysis revealed significant heterogeneity within main cell types. For instance, amacrine cells in the eye, known for their extensive diversity(62), were clustered into thirty-four distinct subtypes (Fig. 3A). These subtypes were characterized by unique gene activity, peak accessibility, and TF motif enrichments (Fig. 3A, Fig. S7A–B), aligning with previous findings based on single-cell transcriptomic analyses (62). In total, our clustering analysis identified 1,828 subtypes from 386 main cell types across the entire organism, with a median of 1,610 cells per subtype and four subtypes per main cell type (Fig. S7C). Most subtypes (93.4%) were consistently detected across more than ten samples (Fig. S7C). Notably, we identified many rare cell subtypes that have not been extensively explored in prior chromatin accessibility studies. Notable examples include diverse enteroendocrine cells in the small intestine, including L cells (marked by Gcg, 0.092% of the total population), K cells (marked by Gip, 0.070%), D cells (marked by Sst, 0.028%), enterochromaffin cells (marked by Tph1, 0.084%), and enteroendocrine progenitor cells (marked by Neurog3, 0.081%) (Fig. S7D–S7E).

Figure 3. Age- and sex-dependent cell population change at the subtype level.

(A) Left: schematic of the sub-clustering workflow. Middle: example UMAP plots showing sub-clustering analysis of eye amacrine cells, colored by subcluster ID. Right: dotplot displaying marker gene accessibility for amacrine cell subtypes.

(B) Venn diagram showing the number of cell subtypes with significant changes associated with age or age-sex interactions.

(C) Top: barplot showing the number of aging-associated subtypes for each main cell type. Center: dotplot summarizing the number aging-associated subtypes across main cell types and tissues. Right: barplot showing the number of aging-associated subtypes per tissue.

(D) Scatterplot comparing aging-associated cellular fraction changes between aging-associated subtypes and their corresponding main cell types. Subtypes with distinct trends compared to their parent main cell types are labeled.

(E) UMAP plots showing sub-clustering of bone marrow HSPCs to identify different subtypes, colored by subcluster ID (top middle), age group (top right), and marker gene accessibilities for lymphoid progenitor cells (bottom left) and age-associated megakaryocyte progenitors (bottom right).

(F) Scatter plot showing cell proportion changes for lymphoid progenitor cells (HSPC-3) and age-associated megakaryocyte progenitors (HSPC-4), with a linear regression fit. Each dot represents one animal.

(G) UMAP plot showing sub-clustering results of kidney vascular endothelial cells to identify different subtypes, colored by subcluster ID (bottom).

(H) UMAP plots of kidney vascular endothelial cells, colored by the accessibility of genes that mark vascular cells across different spatial locations.

(I) Scatter plots and boxplots showing the fractions of all kidney endothelial cells (top), vein endothelial cells (Lhx6+ VEC-4, center), and medulla endothelial cells (Igf1+ Cyp1b1+ VEC-5, bottom) across three age groups.

(J) UMAP plots of vascular endothelial cells from a published snRNA-seq study (7), colored by the expression of genes marking medulla endothelial cells (as in H).

(K) Scatterplot showing the aging-associated decline in the kidney medulla endothelial cells identified from a snRNA-seq atlas of aging(7).

(L) UMAP plots showing all kidney cells (Left) and sub-clustering analysis (middle and right) for thick ascending limb cells (TAL, top), distal convoluted tubule cells (DCT, center), and connecting tubule cells (CNT, bottom), colored by main cell type specific markers (left), subtype ID (middle) and age group (right). Squares indicate the aging-associated cell subtypes.

(M) Scatter plots showing the relative fractions of three reactive subtypes identified in (K) across ages, with separate linear regression fits for each sex.

(N) Dotplots showing the gene marker accessibility marking the reactive states (TAL-5, DCT-4 and CNT-5) in each renal epithelial cell type.

(O) Violinplots showing the motif accessibilities of transcription factors marking the reactive states in each renal epithelial cell type.

(P) UMAP plots showing sub-clustering results for TAL, DCT, and CNT cells from published snRNA-seq data(7), colored by the expression of genes that mark reactive states (as in N).

(Q) Scatterplots showing the relative fractions of reactive subtypes identified in (P) across aging, with separate linear regression fits for each sex.

We next investigated cell population changes at the subtype level. Using the same analytical pipeline for identifying aging-associated main cell types, we identified 499 subtypes (out of 228 organ-specific main cell types) with significant cell population dynamics associated with aging (FDR of 0.05, R² > 0.4) (Fig. 3B, Table S5). Of these, 193 subtypes exhibited significant sex-specific aging differences (FDR of 0.05) (Fig. 3C). A median of 24 significantly altered subtypes was identified per tissue, ranging from 51 in the kidney to 3 in the esophagus (Fig. 3C). While the dynamics of aging-associated subtypes were generally consistent with their main cell type changes (Pearson r: 0.72, p-value < 2.2 * 10⁻²⁶), distinct trends emerged in certain subtypes compared to their parent main cell types (Fig. 3D). For instance, specific aging-associated subtypes such as Lepr1+ Abcg1+ fibroblasts and Ccr3+ C5ar1+ C5ar2+ mast cells expanded in the aged intestine, while their main cell types remained relatively stable with age (Fig. S7F–G). In the liver and lung, while the overall main T cell population increased in aged tissues, subtypes corresponding to naive T cells were depleted (Fig. S7F–G). In contrast, we observed an expansion of a subtype corresponding to Il10+ Foxp3+ regulatory T cell in the iWAT, despite the overall CD4 T cell proportion decreasing with aging (Fig. S7F–G). We also observed opposing aging-associated dynamics of subtypes from the same cell type. For example, in the bone marrow hematopoietic stem and progenitor cells (HSPCs), there is a marked decrease in lymphoid progenitors (HSPC-3, marked by Il7r (63)), accompanied by a significant expansion of aging-associated megakaryocyte progenitors (HSPC-4, marked by Gata2, Fhl1, and Hoxb5) (Fig. 3E–F), consistent with findings from a recent study(64).

Several aging-associated subtypes correspond to the same cell type in distinct spatial locations, uncovering region-specific vulnerabilities during aging. For example, renal vascular endothelial cells (ECs) can be clustered into heterogeneous subtypes representing ECs from different anatomical locations within the nephron(65), including the artery (Sox6+), cortex (Npr3+, Igfbp3+), vein (Lhx6+), medulla (Igf1+, Cyp1b1+), vasa recta (Aqp1+), and glomeruli (Gata5+) (Fig. 3G–H). Interestingly, while the total abundance of ECs remains relatively unchanged at the main cell type level, we observed a pronounced depletion of two EC subtypes—Lhx6+ vein ECs and Igf1+, Cyp1b1+ medulla ECs (Fig. 3I)—indicating that endothelial cells in these specific regions are more vulnerable to aging. The findings were further validated through a published snRNA-seq dataset(7): we observed a consistent decline of Igf1+ Cyp1b1+ medulla ECs across aging in both sexes, despite a higher baseline proportion in females (Fig. 3J–K).

Meanwhile, certain aging-associated cell subtypes across different spatial locations share similar molecular features. For example, sub-clustering analysis of three specialized renal epithelial cell types—the thick ascending limb of the Loop of Henle (TAL), distal convoluted tubule (DCT), and connecting tubule (CNT)—revealed three subtypes (i.e., TAL-5, DCT-4, and CNT-5) that exhibit significant expansion in aging (Fig. 3L–M), particularly in females. All three subtypes converged into a reactive aging-associated epithelial cell state that is characterized by the activation of transcription factors linked to inflammation and stress responses (e.g., Creb5, Bnc2, Runx1, Bach2, Fosl2, and Klf7), confirmed through gene body accessibility (Fig. 3N) and TF motif activity (Fig. 3O). We further validated the existence of these states (Fig. 3P) and female-biased expansions (Fig. 3Q) by re-clustering the same cell type from a single-cell transcriptomic atlas of aging(7). The emergence of this reactive cellular state underscores the potential influence of inflammatory signaling on epithelial cells specifically in the aged female kidney. In summary, our subtype-level analysis revealed both region-specific and shared aging-associated changes.

Aging-associated population dynamics of broadly distributed immune subtypes

We next examined whether broadly distributed cells displayed similar dynamics across multiple organs in aging. We first analyzed immune cells, including T cells (n = 1,264,322 cells), B cells (n = 1,079,641 cells), plasma cells (n = 157,545 cells), innate lymphoid cells (NK cells, ILC1, ILC2 and ILC3) (n = 87,550 cells), monocytes and macrophages (n = 581,877 cells), and dendritic cells (n = 306,562 cells) isolated from various organs.

Clustering analysis for each cell type revealed a total of sixty-seven highly heterogeneous immune cell subtypes, each defined by distinct chromatin accessibility patterns at key marker genes (Fig. 4A–B). For instance, the combined clustering of T cells yielded sixteen subtypes spanning a differentiation trajectory from T progenitor/double-negative (DN) cells, through double-positive (DP) cells, and onward to mature T cell states (Fig. 4A–B, top left), including CD4 T cells (Cd4+), CD8 T cells (Cd8b1+), γδ T cells (Tcrg-C4+), and innate-like T cells (Zbtb16+) (Fig. S8A). Similarly, we identified eight B cell clusters forming a trajectory from pro-B and pre-B cells (Erg+, Rag1+, Il7r+) through naive/memory B cells (Cd55+, Fcer2a+) and finally transitioned into aging-associated states marked by either Tbx21 or Pstpip2 (Fig. 4A–B, top right; Fig. S8B). These aging-associated B cell states closely mirror recent single-cell transcriptomic findings(66). Within the plasma cell population, we detected IgA, IgM, and IgG plasma cells, each comprising several distinct subtypes (Fig. 4A–B, middle left). For innate lymphoid cells, ten clusters were identified, including multiple subtypes within NK cells, ILC2, and ILC3 (Fig. 4A–B, middle right; Fig. S8C). Monocytes and macrophages were resolved into twelve subtypes (Fig. 4A–B, bottom left), capturing the transition from Ace+ monocytes to Lyve1+ tissue-resident macrophages, and various specialized macrophages across different tissues. Dendritic cells (DC) were categorized into fourteen subtypes (Fig. 4A–B, bottom right), including Xcr1+ type I DCs, Sirpa+ type II DCs, Ccr7+ migratory DCs, and Siglech+ plasmacytoid DCs, with diverse subsets observed within each major type.

Figure 4. Aging-associated dynamics of broadly distributed immune subtypes.

(A) UMAP plots showing combined clustering results for T cells, B cells, plasma cells, innate lymphoid cells, macrophages and dendritic cells across tissues, colored by subtype id.

(B) Dotplots showing accessibilities of genes used for annotating immune cell subtypes.

(C) Stacked bar plots showing the tissue composition for each immune subtype.

(D) Dotplots showing the proportional changes of each immune subtype in each tissue across aging, colored by Pearson correlation between their relative proportion (within each tissue) and age. Subtypes with significant age or age-sex interactions are marked.

(E) UMAP plots highlighting aging-expanded subtypes of plasma cell (left) and macrophage (right) in gastrointestinal tissues, colored by marker gene accessibility (Gpr15 for plasma cell-3 and Tlr12 for macrophage-7).

(F) Scatterplot showing the relative fractions of Gpr15+ Igha+ plasma cells and Tlr12+ macrophages along aging.

(G) Scatterplots showing the relative fractions of Gzmk+ CD8 T cells (left) and Il23r+ Zbtb16+ type 17 innate T cells (right) along aging, with separate linear regression fits for each sex.

(H) Scatterplot showing the relative fractions of Tbx21+ aging-associated B cells along aging, with separate linear regression fits for each sex.

(I) UMAP plots showing subtypes exhibiting female-biased expansion along aging, colored by accessibilities of marker genes. (Gpr55 for NK-3, Ifnlr1 for macrophage-11, Dtx1 and Sirbp1a for dendritic cells-7).

(J) Scatterplot showing the relative fractions of Gpr55+ NK cells (left), Ifnlr1+ macrophages (middle) and Dtx1+ Sirbp1a+ type 2 dendritic cells (right) along aging, with separate linear regression fits for each sex.

We then quantified the tissue composition of each immune subtype. Although many subtypes were detected globally, certain subtypes exhibited high organ specificity (Fig. 4C). For example, immature T cells (T cells −1, T cells - 2, T cells −3, and T cells - 4) were predominantly in the thymus, while Gzma+ γδ T cells (T cells - 13) were mainly localized to the intestine, cecum, and colon(Fig. 4C). Similarly, B cell progenitors (B cells - 1 and B cells - 2) were primarily found in the bone marrow, and IgA+ plasma cells (plasma cells - 1, plasma cells −2, and plasma cells - 3) were enriched in gastrointestinal organs (e.g., intestine, cecum, and colon)(Fig. 4C). For macrophages (MAC), Cxcl16+ Tlr12+ macrophages (MAC-6 and MAC-7) were constrained to the small and large intestine, while Spic+ macrophages were unique to bone marrow (MAC-9) and spleen (MAC-10)(Fig. 4C). The latter aligns with previous reports about the role of Spic in the development of iron-recycling macrophages in these two organs(67). Certain dendritic cells also demonstrated tissue-specific characteristics, such as Apol7c+ type I DCs (DC-3), Il22ra2+, and C1qtnf3+ type II DCs (DC-9 and DC-10), which were exclusively identified in the gut(Fig. 4C).

We next assessed how these immune subtypes changed with aging (Fig. 4D). Six immune subtypes were found significantly depleted across multiple tissues, including naive CD4+ T cells in ten tissues, naive CD8+ T cells in five tissues, and immature B cells in the bone marrow, spleen, and mWAT(Fig. 4D). We also observed a significant depletion of organ-specific immune progenitor cells, such as immature T cells in the thymus and pro-B/pre-B cells in the bone marrow, which subsequently led to a reduction of naive/memory B cells in tissues like the spleen and mWAT(Fig. 4D). In addition, we uncovered less-characterized immune cell dynamics validated across multiple organs. For instance, Zfp683+ tissue-resident memory T cells were consistently depleted in the intestine, iWAT, and mWAT (T cells - 12, Fig. 4D), and Gzma+ resting NK cells (68) were depleted in both the spleen and bone marrow (NK - 1; Fig. 4D). Notably, the depletion of this resting NK cell subtype occurred concurrently with an expansion of a Gpr55+ Ifi211+ proinflammatory NK cell subtype (69) (NK - 3) in the same organs (Fig. 4D).

In contrast to these depletions, the majority of immune subtypes (42 out of 67) significantly expanded during aging in at least one tissue (Fig. 4D). These expansions included nine T cell subtypes, eight NK and ILC subtypes, five B cell subtypes, seven dendritic cell subtypes, and nearly all plasma cell and monocytes/macrophages (‘MoMa’) subtypes (Fig. 4D). Notable examples of broadly expanding populations included Zbtb16+ Il23r+ type 17 innate T cells (T cells - 15), Gzmk+ CD8 T cells (T cells - 10), Il21+ follicular helper T cells (T cells - 9), Foxp3+ regulatory T cells (T cells - 7), Tbx21+ aging-associated B cells (B cells - 6), Chst1+ Wipf3+ IgM plasma cells (plasma cells - 5), Clec4d+ monocytes (MoMa - 4), and F13a1+ Lyve1+ macrophages (MoMa - 5), each of which expanded in multiple organs (Fig. 4D). Beyond these widespread patterns, some aging-associated expansions were confined to specific organs. For example, several immune subtypes expanded preferentially in the gastrointestinal tract, such as Gpr15+ Stc2+ IgA plasma cells (plasma cells - 3; Fig. 4E–F, left), Tlr12+ macrophages (macrophages - 7; Fig. 4E–F, right), Ccr6+ Trappc3l+ ILC3 cells (NK/ILC - 7), and Apol7c+ type I DCs (dendritic cells - 3) (Fig. 4D). We also identified tissue-specific expansions of Cd163I+ Sox13+ ILC3 cells (NK/ILC - 8) in the eye, kidney, and liver, as well as Pcsk1+ ILC2s (NK/ILC - 9) in the ovary/uterus and Cybb+ ILC2s (NK/ILC - 10) in the bone marrow (Fig. 4D). Interestingly, some immune subtypes exhibited opposite trends depending on the organ. For instance, Klrk1+ Gas7+ CD8 effector T cells (T cells - 8) expanded in the liver and lung but were depleted in gastrointestinal tissues (e.g., colon and cecum) (Fig. 4D), potentially due to distinct local microenvironmental cues that shape tissue-specific aging dynamics.

Aging-associated immune cell dynamics also differed notably between males and females. For example, Gzmk+ CD8 exhausted T cells exhibited greater expansion in females (Fig. 4G, left), notably in the lung, pancreas, and mWAT, while type 17 innate T cells showed a stronger increase in males, especially in the stomach, gWAT, and mWAT (Fig. 4G, right). Overall, however, we observed that females tend to have a greater expansion of immune cell subtypes than males during aging (53 female-biased subtypes compared to 31 male-biased subtypes, Fig. S8D), including Tbx21+ aging-associated B cells (in mWAT, pancreas, thymus; Fig. 4H), Gpr55+ Ifi211+ NK cells (in bone marrow, pancreas, iWAT, spleen; Fig. 4I–J, left), Sox5+ Ifnlr1+ Zfp362+ macrophages (in thymus, gWAT; Fig. 4I–J, middle) and Dtx1+ Sripb1a+ type II dendritic cells (Fig. 4I–J, right). These subtypes potentially contribute to the higher prevalence of autoimmune diseases in women(70). Interestingly, we observed an enrichment of male-biased subtypes within gastrointestinal tissues, with 15 out of 31 originating from the esophagus, stomach, intestine, cecum, and colon, compared to only 2 out of 53 in female-biased ones (Fig. S8D). Noteworthy examples include the expansion of Ccr9+ ILC3s (NK/ILC-5) and Cyp26b1+ ILC3s (NK/ILC-6) in the intestine (Fig. S8D), reflecting more pronounced changes in the innate immunity during male intestinal aging (71).

Aging-associated population dynamics of broadly distributed non-immune subtypes

Next, we expanded our analysis to aging-associated dynamics in non-immune cell populations, focusing on vascular endothelial cells (VECs) (n = 426,160 cells, Fig. 5A–B, top), fibroblasts (n = 377,181 cells, Fig. 5A–B, middle), and smooth muscle cells (n = 101,284 cells, Fig. 5A–B, bottom), which are broadly distributed across organs. Unlike immune cells that share molecular states across tissues, these non-immune cell populations displayed remarkable organ specificity, with most (34 out of 61) clusters dominated by a single tissue (>90% of cells from one tissue). For example, we identified VECs specific to the liver (marked by Gata4) and the lung (marked by Foxf1) (Fig. 5C). The tissue specificity was further confirmed through motif footprinting analyses of marker TFs (Fig. 5D), and consistent with previous study (72). Similarly, tissue-specific TFs were observed in different fibroblast populations: Six1 in bone marrow fibroblasts, Foxf1 in gastrointestinal fibroblasts, Tbx20 in cardiac fibroblasts, Hoxc9 in kidney fibroblasts, and Tbx5 in lung fibroblasts (Fig. 5E). We also identified subtypes exhibiting sex specificity. For instance, we identified subtypes of smooth muscle cells unique to female gonadal organs (e.g., ovary, uterus, and gWAT). These cells were marked by the transcription factors Wt1 and Esr1, supported by both gene accessibility and motif activity (Fig. 5F).

Figure 5. Aging-associated dynamics of vascular endothelial cells, fibroblasts, and smooth muscle cells across organs.

(A) UMAP visualization showing the combined clustering for vascular endothelial cells (VECs, top), fibroblasts (FBs, middle) and smooth muscle cells (SMCs, bottom) across tissues, colored by tissue origin (Left) and cluster ID (Right).

(B) Dot plots showing the accessibility of marker genes for cell clusters of VEC (top), FB (middle) and SMC (bottom), including a general marker for each cell type (Flt1 for VEC, Pdgfra for FBs, Myh11 for SMCs)

(C) UMAP plot of VECs colored by accessibilities of the liver VEC marker Gata4 and lung VEC marker Foxf1

(D) Motif footprinting analysis showing enrichment of GATA4 motifs in liver VECs and FOXF1 motifs in lung VECs compared with other tissues.

(E) Heatmaps showing the motif enrichments and gene body accessibility of tissue-specific transcription factors. Motif enrichment scores (−log10 q-value) were computed with HOMER (73) using tissue-specific peaks and scaled to the top enriched tissue.

(F) UMAP plots of SMCs colored by gene accessibility (Top) and chromVar motif deviation scores (Bottom) for gonadal SMC markers Wt1 and Esr1.

(G) The stacking bar plots show tissue composition per cluster for VECs (Left), FBs (Middle), and SMCs (Right). The dot plots show aging-associated proportional changes of clusters across tissues, colored by Pearson correlation between cluster proportion (normalized per tissue) and age. Clusters with significant age/age-sex effects are annotated (p-adjusted < 0.05).

(H) UMAP of VECs colored by the accessibility of VEC-5 markers (Meox2, Lhx6, Slc7a14).

(I) Scatterplots showing the age-related decline in VEC-5 proportions in muscle (quadriceps), gonadal white adipose tissue (gWAT), and ovary/uterus.

(J) UMAP of FBs colored by the accessibility of FB-18 markers (Ifi207, Ear1).

(K) Scatterplots showing the age-associated increase in FB-5 proportions in mesenteric WAT (mWAT), stomach, cecum, and esophagus.

(L) UMAP plot showing gWAT VEC cells isolated from snRNA-seq data (7), colored by Lhx6 expression (VEC-5 marker).

(M) Scatterplot showing age-dependent reduction in Lhx6+ gWAT endothelial cells.

We further examined how these broadly distributed non-immune subtypes changed with aging (Fig. 5G). Compared to immune cells that overall expand across organs, these broadly distributed non-immune cell types exhibited strong tissue-specific changes during aging (Fig. 5G). For example, we observed an age-associated depletion of various organ-specific endothelial subtypes, such as cells in the kidney (VEC-14, Igf1+ kidney medulla endothelial), lung (VEC-7, Tbx2+ lung aerocytes), and muscle (VEC-1, Aqp7+) (Fig. 5G and S9A). By contrast, organ-specific endothelial cells in the liver (VEC-9, Clec4g+ sinusoidal endothelial) and pancreas (VEC-4, Lamb1+ VECs) expanded significantly during aging (Fig. 5G and S9B). The expansion of liver sinusoidal endothelial cells aligns with previous findings from Tabula Muris Senis(57). Fibroblasts (FB) likewise displayed tissue-specific aging trajectories, showing expansion in aged bone marrow (Btla+ FB-13), colon (Edil3+ FB-21), cecum (Prkcq+ FB-19), stomach (Csf2rb+ FB-3), and mWAT (Nkx6–1+ FB-17) (Fig. 5G and S9C). In contrast, two fibroblast subtypes (ovary Tgfa+ FB-9 and stomach Sox6+ FB-23) significantly decreased in aging (Fig. 5G and S9D). For smooth muscle cell (SMC) subtypes, there is a strong depletion of several subtypes across organs, such as colon (Hoxd3+ SMC-15), lung (Tbx5+ SMC-3), and ovary/uterus (Gata2+ SMC-2) (Fig. 5G and S9F). In the cecum, however, there was an increased population of SMCs marked by inflammatory gene markers such as Il7 and Serpina1c (SMC-5, Fig. 5G and S9G).

Importantly, we identified several cell subtypes that shared similar aging dynamics across multiple tissues. For example, a Slc7a14+ Lhx6+ Meox2+ vein endothelial cell subtype (VEC-5) was significantly depleted in several aged tissues (e.g., muscle, gWAT, and ovary/uterus) (Fig. 5H–II), which could contribute to the endothelium dysfunction in aging. Likewise, an aging-associated fibroblast subtype characterized by elevated inflammatory markers (e.g., Ifi207, Ifi205, Ear1) expanded in several aged organs (e.g., mWAT, stomach, cecum, esophagus) in a male-specific manner (Fig. 5J–K). Of note, these aging-associated subtype dynamics were validated through external datasets (7), such as the depletion of Lhx6+ endothelial cells in gWAT (Fig. 5L–M) and the expansion of Csf2rb+ fibroblasts in the stomach (Fig. S9E).

Age- and sex-dependent epigenetic signatures of aging

We next investigated the organism-wide epigenetic aging program by examining 349 organ-specific main cell types. For each major cell type in each tissue, we performed differential accessibility (DA) analyses in females and males separately (Methods). From these analyses, we identified both shared and sex-specific age-associated alterations in chromatin accessibility. We then carried out linkage analysis to map downstream genetic targets of these chromatin sites and examined upstream regulators by assessing the activity of intrinsic transcription factors and extrinsic cytokine signaling pathways that drive epigenetic reprogramming during mammalian aging (Fig. 6A).

Figure 6. Aging-associated changes of chromatin accessibility.

(A) Scheme for computational analyses to identify differentially accessible (DA) peaks along aging, their putative gene targets, and upstream regulators.

(B) Scatterplot comparing the number of DA peaks between males and females for each main cell type. Cell types with a high number of DA peaks are labeled.

(C) Venn plot of cell-type-specific DA peaks shared or unique to males and females.

(D) Correlation of log transformed fold change (logFC, aging-associated accessibility changes per month, calculated via edgeR (77)) for DA peaks shared between sexes.

(E) Scatterplot summarizing DA peaks with concordant changes with aging in both sexes. Axes show counts of cell types with increased (x-axis) or decreased (y-axis) accessibility; color denotes peak density.

(F) Dot plots showing log transformed fold change (logFC) for DA peaks with consistent increases (promoters of Nr3c2 and Il7), decreases (promoters of Sox4), or divergent trends (intronic peak of Epha3) across cell types.

(G) Genomic tracks visualizing the accessibilities of DA peaks in (F) across three age groups.

(H) Dot plots of sex-specific DA peaks for females (Left) and males (Right). Y-axis: number of cell types with significant increases; x-axis: −log10(q-value) for the most significant cell type. Color denotes the number of tissues with significant increases.

(I) Genomic tracks of female-specific DA peak in liver hepatocytes (Left) and male-specific DA peak kidney proximal tubule cells S3T2 (Right).

(J) Barplot showing validation of sex-specific chromatin accessibility changes using gene expression data (7) of the same genes in the same cell type as in (I). Dots represent gene expression of target genes across independent animals, quantified by transcripts per million (TPM).

(K) Barplot showing the number of cell-type-specific, aging-associated gene-promoter-CRE linkages (Left) and the number of unique genes within those linkages (Right) for each tissue.

(L) Dot plot of genes with aging-upregulated linkages in both sexes. X-axis: cell-type count; y-axis: CRE count per gene. Color denotes the number of tissues containing cell types with significant increases.

(M) Genome browser plot showing links between putative cis-regulatory elements and promoters for Nr3c2.

(N) Heatmaps showing coordinated increases in gene expression, promoter accessibility, and linked distal CRE accessibility of Nr3c2 across multiple immune cells with aging.

In total, we identified 1,318,007 and 959,361 cell-type-specific and aging-associated DA peaks in females and males, respectively (Fig. 6B, Fig. S10A, Table S6–S8). While the number of DA peaks generally correlated with cell numbers, a few cell types exhibited notably strong aging-related chromatin changes, standing out as outliers with high regression residuals—such as B cells from brown adipose tissue or kidney, tenocytes, and muscle satellite cells (Fig. S10B). Among the identified DA peaks, a total of 540,386 cell-type-specific DA peaks were present in both sexes, most (96.9%) exhibiting consistent changes in the same cell type across both males and females (Spearman correlation of log transformed fold change = 0.86, p-value < 2.2 * 10⁻¹⁶, Fig. 6C–D).

While many DA peaks were restricted to only a few cell types (91.4% found in no more than 3 cell types), we also uncovered a subset of peaks (n = 803) that were consistently altered across many cell types (n >= 10) and tissues (Fig. 6E). This subset highlights coordinated chromatin dynamics across diverse cell types and regions during mammalian aging. Notable examples include increased promoter accessibility of the mineralocorticoid receptor Nr3c2 (in 63 cell types, Fig. 6F–G, left) and the proinflammatory cytokine Il7 (in 26 cell types, Fig. 6F–G, middle left), predominantly within immune cells. Increased Nr3c2 activity also marked the aging-associated B cell states (Fig. 4B). Conversely, we identified DA peaks with decreased accessibility in aging, such as promoters of Sox4 (in 25 cell types) and Sox11 (in 19 cell types)(Fig. 6E). Specifically, Sox11, a transcription factor involved in angiogenesis (74), showed reduced accessibility in endothelial cells across multiple organs (Fig. 6F–G, middle right), which could contribute to the global functional decline of the vascular system.

Aging-associated chromatin alterations were not always uniform across cell types. For example, 131 peaks showed opposite changes in different cell types (at least five cell types in each direction). One example is a peak in the intron of Epha3, which displayed increased accessibility in T cells and macrophages across 12 tissues but decreased accessibility in non-immune cell types such as vascular cells, neurons, and epithelial cells (Fig. 6F–G, right, Fig. S10C). These cell-type-specific patterns were confirmed across both sexes (Fig. 6F), suggesting that aging-associated chromatin alterations are shaped by unique cellular contexts and lineage-specific regulatory networks.

These epigenetic changes in aging also vary by sex. Considering peaks with significantly increased accessibility in aging, we found 195,893 sex-specific peaks in females and 140,770 in males (Fig. 6H). Top female-specific DA peaks validated across multiple cell types from different tissues included intronic peaks at the senescence-associated gene Cdkn2b (in 18 cell types) and promoter peaks of the interferon-inducible gene Ligp1 (in 14 cell types) and Ifi211 (in 16 cell types) (Fig. 6H). Similarly, we observed male-specific DA peaks shared across multiple cell types, such as the increased promoter accessibility of Tsbp1 across 18 cell types (Fig. 6H). We also identified cell-type-specific chromatin accessibility changes that were highly sex-dependent. For example, aged hepatocytes in females showed up-regulation of peaks overlapping with Bnc2 (Fig. 6I, left), a gene involved in extracellular matrix production and degradation (75). In contrast, aged hepatocytes in males showed up-regulation of peaks overlapping with Cidea (Fig. S6D, left), a gene involved in lipid particle organizations. Similarly, in a subtype of proximal tubule cells segment 3 (PT S3T2), peaks overlapping Mill2 were more accessible in aged females (Fig. S6D, right), whereas intronic peaks in Tifa—a gene associated with programmed cell death (76)—were more accessible in aged males (Fig. 6I, right). Consistent with these findings, this cellular state specifically declined in aged males (Fig. 2H). Notably, these sex- and cell-type-specific peaks were further validated by corresponding changes in promoter accessibility and gene expression during aging (Fig. 6I–J and S6E).

To infer putative target genes of aging-associated chromatin alterations in cis-regulatory elements (CRE), we linked non-promoter DA peaks to nearby promoters based on their co-accessibility across samples for each cell type, thus generating promoter-CRE linkages. We then integrated the data with a published snRNA-seq atlas of aging (7) to identify genes exhibiting consistent changes between promoters and expressions with age (Methods). This analysis identified 21,360 and 14,564 cell-type-specific gene-promoter-CRE linkages in males and females, respectively, with 12,862 linkages shared by both sexes (Table S9). In total, this revealed 15,090 CREs linked to aging-associated changes in 2,001 differentially expressed (DE) genes across 86 organ-specific cell types (Fig. 6K). Collapsing sex-shared, up-regulation linkages across cell types revealed concordant changes supported in three molecular evidences (i.e, genes, promoters and putative enhancers) (Fig. 6L). One notable example is Nr3c2, linked to 25 CREs, most located in intronic regions (Fig. 6M). These CREs showed strongly increased accessibility during aging across multiple immune cell types and tissues, in line with changes in promoter accessibility and gene expression from the same cell type (Fig. 6N).

Intrinsic and extrinsic regulators of epigenetic reprogramming in mammalian aging

We then explored upstream regulators of these aging-associated chromatin changes by performing motif enrichment analyses of DA peaks in each cell type. Aging-upregulated peaks were enriched for inflammatory motifs that are confirmed across multiple cell types, including IFN-stimulated response elements (ISRE) in 80 cell types, IRF1 in 77 cell types, IRF2 in 85 cell types, NF-κB in 57 cell types, and AP-1 in 71 cell types (Fig. 7A–B). In contrast, downregulated peaks were enriched for TF families linked to stem cell maintenance and development, such as SOX15 ( in 18 cell types) and E2A (in 29 cell types) (Fig. 7A). Sex-specific patterns also emerged. For example, female-specific upregulated peaks were enriched for POU motifs in B cells and male-specific upregulated peaks were enriched for inflammatory TF motifs (IRFs and AP-1) in intestinal epithelial cells (Fig. 7C).

Figure 7. Identifying upstream regulators of aging-associated chromatin changes.

(A) Sex-shared aging-associated up-regulated and down-regulated differentially accessible (DA) peaks for each main cell type were analyzed for transcription factor (TF) motif enrichment using HOMER (73). The scatter plot summarizes the number of cell types exhibiting significant motif enrichments (p.adj < 0.05) in either up-regulated (red) or down-regulated (blue) peak sets. TF motifs enriched across multiple cell types are labeled.

(B) Dot plot showing the enrichment of interferon-stimulated response element (ISRE) motifs in aging-up DA peaks across cell types from diverse tissues. Dot size represents the fold change in ISRE-containing peaks in DA sets compared to a background peak set (Methods). Each dot represents one cell type from a specific tissue, colored by the tissue of origin.

(C) Comparison of TF motif enrichment significance between female-specific and male-specific aging-up peak sets for each main cell type. Cell types with strong sex-specific motif enrichments are labeled.

(D) Scatter plot showing enriched TF motifs in sex-shared, aging-up (red), and aging-down (blue) peaks from lung B cells. The x-axis denotes the difference in motif frequency (%) between DA and background peaks; the y-axis represents statistical significance.

(E) Barplots showing motif enrichment for exemplary TFs: Ebf1 and Ets1 (down-regulated in both sexes), Irf2 (up-regulated in both sexes), and Pou2f2 (female-biased up-regulation) in lung B cells.

(F) Scatterplot and barplot showing age-related expression changes of TFs highlighted in (E) in lung B cells, quantified using transcripts per million (TPM) for each animal.

(G) Enrichment significance of EBF and IRF motifs for DA peaks from B cells across tissues, demonstrating conserved aging-associated regulatory patterns.

(H) Scheme for computational analyses to identify cytokines potentially drive aging-associated chromatin remodeling in lung B cells.

(I) Circle plot showing enrichments (−log10(p-value), hypergeometric test) of cytokine-specific signatures (derived from Immune Dictionary dataset (82)) in aging-associated genes identified in lung B cells from this study.

(J) Dot plot displaying aggregated gene expressions of 350 genes shared between aging and cytokine treatments. Top conditions recapitulating aging-associated expression profiles were labeled.

(K) Left: dot plot showing the enrichment score (−log10(p-value), hypergeometric test) of cytokine signatures in the male and female lung B cells as in (I). Right: Heatmap indicating the aging-associated changes in the accessibility of cytokine secretion or receptor genes.

(L) Genomic tracks showing aging-associated accessibility increases at Ifng and Il21 in lung T cells and Il4 and Il2rb in B cells.

(N) Scatterplot showing increased gene expressions of Il21 in T cells and Il2rb in B cells with age, together with a linear regression line.

(M) Left, dot plot summarizing the enriched cytokine signatures using aging-up peak sets of B cells from each tissue. Only signatures that can be supported by increased accessibilities of secretion or receptor genes are shown. Right: barplot summarizing the number of tissues with activated signatures for each cytokine, splitting males and females.

These TF activity changes were further supported by corresponding gene expression dynamics observed in single-cell RNA-seq data. For example, in the aged B cells in the lung, we detected reduced motif accessibility for Ebf1 and Ets1 (Fig. 7D–E), transcription factors involved in B cell development (25) and quiescent state maintenance (78), aligning with their decreased expression in both sexes (Fig. 7F). Meanwhile, aged B cells exhibited decreased expression and increased motif accessibility for Irf2 (Fig. 7D–E), reflecting its role as a transcriptional repressor of inflammatory signaling (79, 80). Pou2f2, a TF involved in B proliferation and plasma cell differentiation(81), showed increased motif accessibility exclusively in female-upregulated peak sets, corroborated by female-specific increases in Pou2f2 expression(Fig. 7D–E). This pattern is consistent with the preferential expansion of aging-associated B cells in female tissues (Fig. 4D and 4H). Moreover, the aging-associated changes in EBF and IRF motifs were detected in B cells body-wide, highlighting a global chromatin transition in B cells during aging (Fig. 7G).

Recognizing the strong inflammatory signature underlying these changes, we next explored whether cytokines known to drive immune responses could contribute to global epigenetic remodeling. Using the Immune Dictionary dataset (82), which encompasses scRNA-seq profiles of various immune cell types responding to 86 cytokines, we compared aging-related DA peaks and associated gene changes with those induced by different cytokines (Fig. 7H, Methods). This revealed that aging signatures, especially in female B cells from the lung, resembled those induced by specific cytokines such as interferons (IFNs), IL4, IL12, IL15, IL18, and IL21 (Fig. 7I). Cross-referencing aging-upregulated gene expression changes with cytokine-responsive transcripts further validated these links (Fig. 7J).

To reinforce this connection, we examined whether cytokine signatures were supported by changes in the accessibility of either secretion or receptor genes (Fig. 7K–L). For example, for secretion factors, we observed an increased accessibility of Infg and Il21 in T cells, as well as Il4 and Il15 in B cells (Fig. 7L), aligning with activation of corresponding signatures, particularly in females (Fig. 7I). On the other hand, the upregulated response to IL2 and IL15 matched with increased accessibility of the receptor gene, Il2rb, in the aged B cells (Fig. 7L). A consistent trend was observed by analysis of recent snRNA-seq data (7), such as the increased expression of Il21 in T cells, Il2rb and Il15 in B cells (Fig. 7N and Fig. S10F).

Extending this analysis to all B cells across tissues uncovered the widespread emergence of cytokine signatures associated with aging, especially IL2, IL15, IFNγ, IL4, and IL21(Fig. 7M). Notably, the upregulated IL21 signatures were consistent with the global expansion of Il21+ T cells from in our previous analysis (T cells - 9, Fig. 4B and 4D) and were further confirmed at the plasma protein level in humans (3)(Fig. S10G). Interestingly, a previous study has described the function of IL21 in inducing the expression of Tbx21 (a marker of aging-associated B cells, Fig. 4B) in B cells and promoting their differentiation into plasma cells in systemic lupus erythematosus (83), suggesting a potential similar role of Il21 in the epigenome reprogramming of B cells during aging. Similarly, we analyzed cytokine signatures in macrophages by leveraging expression changes following cytokine treatments of the same cell type from Immune Dictionary. This revealed activation of a distinct set of cytokines, including IL15, IL7, IL1α, IL18, and TNFα (Fig. S10H), that may underlie age-related molecular shifts in macrophages.

In summary, our integrative multi-level framework reveals a global, sex-dependent epigenetic reprogramming linked to inflammatory signaling and cytokine-driven pathways. These findings provide a roadmap for understanding the regulatory underpinnings of aging across diverse cell types and tissues—from intrinsic TF regulators to extrinsic cytokine signaling—and highlight candidate upstream factors that could be targeted to mitigate age-related epigenetic dysregulation.

Discussion

Aging involves global deterioration in cellular and organismal function. While advancements in genomics techniques, especially single-cell RNA-seq and spatial transcriptomics, have been used to explore age-associated changes in cell-type- and region-specific gene expression (15, 57, 84, 85), few studies have systematically investigated cell population shifts, chromatin dynamics, as well as age-sex interaction on the scale of the entire organism. Here, by profiling around seven million cells from 21 mouse tissues at three ages and in both sexes, we constructed a single-cell chromatin accessibility atlas of aging with unprecedented depth and breadth. We reported the aging-associated dynamics of 536 tissue-level main cell types and 1,828 finer-grained subtypes, each defined by the accessibility at approximately 1.3 million cis-regulatory elements. Our findings reveal widespread cell population shifts, lineage-specific epigenomic reprogramming, and striking sex-dependent patterns, laying a foundation for understanding the regulatory logic of mammalian aging.

A key insight of our study is the extensive cell population dynamics accompanying age. Notably, 146 organ-specific cell types (approximately 25% of all organ-specific cell types) changed significantly with age. Many of these are immune cells. For example, we observed a widespread increase in plasma cells and macrophages across various tissues, aligning with previous findings of immunoglobulin accumulation (85) and monocyte/macrophage propagation (86) in aging. Tissue-specific population changes, such as the expansion of neutrophils in the liver and lung, and depletion of lymphoid progenitors in primary immune organs (e.g., pre-/pro-B cells in the bone marrow and immature T cells in the thymus) were identified. Beyond immune-related changes, our study also reveals additional vulnerabilities in non-immune cell lineages. For example, we observed profound depletion of specialized functional cell types across multiple organs, such as kidney podocytes, muscle tenocytes and satellite cells, ovary granulosa cells, and lung aerocytes. These depletions likely contribute to age-related loss of tissue homeostasis and functions.

Beyond shifts in broad cell types, our high-resolution analysis identified 1,828 subtypes—nearly one-third of which show significant aging dynamics. We identified subtypes exhibiting cell state changes that diverge from their parent population. For example, naive T cell subtypes in the lung and liver sharply decline, although total T cell proportions rise with age. Likewise, we detected region-specific vulnerability of endothelial cells in the kidney medulla and vein, distinct expansions in fibroblast and mast cell subpopulations in the intestine, and the emergence of a shared reactive state (Creb5+ Bnc2+ Runx1+) across different epithelial cell types in the kidney. These findings suggest that aging promotes a variety of cell-state transitions, including those localized to particular anatomical niches or widespread across distinct parent cell types potentially triggered by common circulating factors.

Coordinated expansion or depletion of the same subtype in multiple organs underscores a broader organism-level program. Immune subtypes, in particular, display strikingly synchronous changes, such as the global expansion of Gzmk+ CD8 T cells, regulatory T cells, Il21+ CD4 T cells, Il23r+ innate T cells and Tbx21+ B cells. Certain changes were restricted to a few organs, including the depletion of Zfp683+ tissue-resident memory T cells in adipose and gut tissues, or the decline of Gzma+ resting NK cells in spleen and bone marrow. However, this synchronization extends beyond mobile immune lineages: certain endothelial and fibroblast populations, also show parallel shifts across distinct tissues, such as the depletion of Slc7a14+ Lhx6+ Meox2+ vein endothelial cell and the expansion of an Ifi207+ Ear1+ aging-associated fibroblast subtype. This points to systemic inflammatory signals or hormonal cues that modulate cell fates body-wide.

At the molecular level, we identified extensive chromatin reprogramming associated with aging. Our differential accessibility analyses uncovered 279,401 peaks changing with age, including a subset consistently altered across many cell types (e.g., promoter peaks of Nr3c2 and Il7 that frequently gain accessibility, and Sox4 or Sox11 peaks that frequently lose accessibility). Linking chromatin changes to RNA-seq data validated the effects from accessibilities to expressions and revealed putative enhancers regulating aging-associated genes. Motif analysis further supports a shift from TFs involved in tissue maintenance or stemness (e.g., SOX, E2A) to those mediating inflammatory and stress responses (e.g., NF-κB, IRFs, AP-1). Specifically, as an example to showcase the potential of our data to reveal anti-aging targets, we focused on cytokines and identified the activation of IL15, IL21, IL4, and IFNγ signatures in aged B cells, indicating these extrinsic signals may reinforce or drive the cell-state transitions observed in aging.

Crucially, we found extensive sexual dimorphism. Around forty percent of aging-associated main cell types (55 out of 146) and subtypes (193 out of 499) exhibited sex differences. Notable examples include the male-biased expansion of hepatic stellate cells in the liver and the female-biased expansion of interstitial macrophages in the lung. Overall, female mice appeared to mount a more pronounced immune activation in aging at both proportional and molecular levels, consistent with the higher incidence of autoimmune diseases in women (87, 88). This included female-specific upregulation of TFs like Pou2f2 in B cells, aligning with female-biased expansions of aging-associated B cell subtypes. We also observed subtypes that shift in opposite directions by sex, such as Slc22a7+ proximal tubule cells (PT S3T2) increasing in aged females but decreasing in males—trends replicated by single-cell transcriptomic data. At the epigenetic level, we detected tens of thousands of cell-type-specific peaks that changed exclusively in one sex, highlighting sex as a major axis of aging heterogeneity.

In summary, this organism-level single-cell chromatin accessibility atlas illuminates how aging remodels the cellular composition and regulatory landscape of multiple tissues. Although many changes are highly tissue-specific, we uncovered “coordinated” cellular and molecular dynamics that are shared across different organs, including immune remodeling, broad depletion of functional cell types, the emergence of inflammation-related states, and sex-dependent trajectories. By cataloging these changes, we offer a resource for understanding the molecular logic of aging and for guiding therapeutic strategies aimed at preserving or restoring youthful tissue states. This work also highlights molecular candidates to mitigate aging-related dysfunction by targeting broad-spectral cell types across tissues, including intrinsic transcription factors, extrinsic cytokines, and individual receptors (Nr3c2 for immune cells, Epha3 for T cells, Il2rb for B cells). Looking forward, we anticipate that this atlas will serve as a critical reference for evaluating anti-aging interventions and refining our understanding of how chromatin remodeling, cell state transitions, and tissue physiology intersect in the aging mammalian body.

Data availability:

Raw FASTQ files, processed count matrices, cell metadata, and peak metadata can be downloaded from NCBI GEO under accession number GSE288730 (reviewer’s token: exitcckgbfmzpkl). An interactive website facilitating the visualization of our data across cell type, age, and sex is available at https://mouseagingatacatlas.org.