Multitissue, multi–time point transcriptomic atlas of aging in mice and rats Authors: Tea Shavlakadze, Kun Xiong, Romain Donne, Corissa McEwen, Abhilash Gadi, Matthew Wakai, Hunter Salmon, Jacob Treinish, Elias Pavlopoulos, Larry Proctor, Shuo Li, Nicole Negron, Min Ni, Yi Wei, Yu Bai, David J. Glass Published: 2026 ABSTRACT To determine the genes and pathways that are up- or down-regulated in a consistent manner throughout the rodent lifespan, we generated a high N age-related gene expression atlas in mice and rats, by profiling 28 tissues in male and female C57BL/6J mice and 32 tissues in male Sprague Dawley rats (>5000 samples) over multiple time points. We identified age-related genes and pathways that change either early in life, at mid-age, late in life, or linearly throughout the animals’ lifespan. Linear genes dominated many but not all tissues, and certain tissues were relatively spared from age-related changes. We explored common and different features of aging between tissues, sexes, and species. Given the expanse of our transcriptomic dataset, we believe that this study will serve as a useful resource for understanding the timing, tissue specificity, sex-specificity, and species specificity of age-related gene and pathway changes in mice and rats. INTRODUCTION To study age-related changes in molecular pathways, transcriptomic studies have been conducted in various tissues in mice, rats (1–3), and humans (4–6). Also, recent studies have focused on profiling large human cohorts for metabolomic (7, 8) and proteomic (9–11) markers of aging. In addition, determination of DNA methylation patterns that correlate with biological age has been another strategy to develop molecular signatures of aging (12–14). Perturbations in age-related signaling pathways may create a permissive environment for the emergence of morbidities and comorbidities and, eventually, the emergence of chronic conditions with age (15). Thus, for many diseases, including cancer, heart disease, kidney disease, dementia, Alzheimer’s disease, osteoarthritis, sarcopenia, and frailty, aging is the biggest risk factor (16–19). An extensive body of work aimed at deciphering molecular signatures of aging has been conducted using murine models since the mouse is a widely used species in aging and biomedical research. A transcriptomic atlas of multitissue aging in mice has been published using bulk RNA sequencing (RNA-seq) (20), single-cell RNA-seq (scRNA-seq) (2), and single-nuclear RNA-seq (snRNA-seq) (21). The prior bulk RNA-seq atlas for mouse aging was constructed from 17 organs from males aged 1 to 27 months and females aged 1 to 21 months. Tissues were sampled at 3-month intervals, N = 2 to 4 per group (20). The single-cell transcriptomic atlas across the lifespan of the mouse included 24 organs/tissues and was constructed using males aged 1, 3, 18, 21, 24, and 30 months and females aged 3, 18, and 21 months, with N = 2 to 4 mice per group, except for N = 7 males aged 3 months (2). A recent study used single-nucleus transcriptomic profiling and characterized age-related transcriptomic changes in 14 organs in male and female mice aged 3, 6, 12, 16, and 23 months, with N = 2 to 4 mice per group (21). Overall, the field is moving toward prioritizing single-cell or single-nucleus transcriptomic databases as these enable discovery of age-related transcriptomic alterations in specific cell types and, to some degree, offer a glimpse in cell population changes in aging organs. However, relatively limited transcript coverage remains a disadvantage for single-cell transcriptomic analyses (22), and the Ns in these studies were low, which is a concern (23). Therefore, transcriptomic data generated by whole-tissue deep sequencing would provide a more complete picture of the transcriptomic landscape. In addition, bulk RNA-seq can benefit from a much higher sample size, which helps to reduce false-positive or false-negative results (23). We have conducted a focused effort to generate comprehensive, high N, transcriptomic signatures of aging in mice and rats because we wanted to know whether examination of multiple tissues at various ages could lead to previously unidentified insights into the potential global and tissue-specific mechanisms of aging that might be causal for age-related pathologies. We are also interested in determining whether one rodent species is superior to another for aging studies focused on particular organ systems. Our approach is to use deep sequencing of multiple tissues in rodents across their lifespan—using high sample numbers per group and, where possible, including both males and females. To this end, we have previously published two studies that analyzed aging transcriptomic signatures in four tissues (liver, kidney, hippocampus, and gastrocnemius muscle) in male rats (1) and in multiple skeletal muscles in male rats and male and female mice (3). Here, we expanded our effort to develop multitissue and multi–time point gene expression signatures in male and female C57Bl6/J mice and male Sprague-Dawley rats aged 6, 9 (rats only), 12, 18, 21, 24, and 27 months. We profiled 28 tissues in male and female mice and 32 tissues in male rats by performing whole-tissue RNA-seq at 100 million reads per sample. We did not include female rats because of the high prevalence of mammary gland tumors in females in this strain, which we felt would highly complicate our study (24). We found genes that change in a coordinate and uniform direction in every tissue; genes that are regulated in early logistic, mid-logistic, late-logistic, and linear fashion in particular tissues; and pathways that are regulated in multiple tissues, giving some indication of common mechanisms of aging. These findings can also serve as robust markers of aging in particular tissues—a gene expression “molecular clock.” RESULTS Age-related changes in gene expression differ by tissues, sex, and species We collected samples from female and male mice across six time points (6, 12, 18, 21, 24, and 27 months and from male rats across seven time points (6, 9, 12, 18, 21, 24, and 27 months) (Fig. 1A). Life expectancy for these species is similar: the Jackson Laboratory (Mouse Phenome Database) reports the median lifespan of male and female C57BL/6J mice to be ~29.9 months (25), and the median lifespan for male Sprague-Dawley rats is reported to be 29 to 30 months (26). Others report ~50% survival at 26 months for C57BL/6 mice (27). The actual lifespan of laboratory rodents differs by research facility and is influenced by factors such as housing density and housing conditions. We see more than 50% death at 26 to 27 months in our rodent cohorts. Age-related genes in rodent tissues. (A) The experimental design: Numbers of analyzed tissues are in table S1. Animal icons were created by BioRender [Zhou, A. (2025) https://BioRender.com/1tbw9v6, Zhou, A. (2025) https://BioRender.com/qx25yis]. m, months. (B) Flowchart for age-related gene identification. The expression trend of each DEG was fit by linear, logistic, and null trends. Genes that were best fit by a linear or a logistic trend were considered age related. (C) The schematic representation of linear-, early-, mid-, and late-logistic genes from rats. Each gray curve is the expression trend of an age-related gene in a tissue. A red curve shows lowess-smoothened trend of a gene expression averaged across all tissues. Examples of age-related genes in mouse tissues are in fig. S2A. (D) The heatmap representing numbers of DEGs in each tissue from each age group compared with 6 months. Top panel shows shared tissues, which were analyzed in both species. The middle and the bottom panels, respectively, show tissues analyzed in mice only or in rats only. The heatmap represents the number of down-regulated (blue to white) and the number of up-regulated (red to white) genes. Gray boxes indicate lack of samples. (E) The proportion of age-related genes in all DEGs in each tissue for each species. Linear age–related genes are shown in teal, and logistic genes are shown in red. See fig. S2B for proportions of early-, mid-, and late-logistic genes in all analyzed tissues. (F) Two most ubiquitous age-related genes: Gpnmb and Asxl3, with the expression trends plotted similarly as in (C). We performed bulk RNA-seq on 28 tissues collected for each sex and age from mice (n = 6 to 10 for each sex and each age), and 32 tissues from rats (N = 10 to 12 for each age for most tissues) (Fig. 1A and table S1). The same 24 tissues were profiled in both female and male mice and in male rats (Fig. 1E). Samples were excluded from analyses if their RNA integrity was too low to construct the appropriate sequencing library, or if they failed to pass additional quality inspection after sequencing (see Materials and Methods). As a result, the rat samples lack the following groups of tissues: 9-month cortex, 18-month thymus, 9-month, and 24-month kidney. In addition, the 12-month and 24-month rat cortex contains only two and three samples, respectively. Because there are adequate samples from rat cortex at each of the other five time points, our analysis, which focused on the overall trend of expression across seven time points, should remain reliable. The sample numbers for all age groups are listed in table S1. To validate the RNA-seq analyses, we have previously performed RT-qPCR amplification of the top five up- and down-regulated genes in all muscle groups (tibialis anterior, gastrocnemius, soleus, and diaphragm) from the same animals as used in the present study (3). Principal components analysis (PCA) on gene expression confirmed that the samples clustered by their tissues of origin (fig. S1, A to C), indicating that most of the variance observed among samples reflects intertissue differences. We calculated the proportion of variance related to age in each tissue and found that thymus showed the highest age-related variance: 36% for female mice, 43% for male mice, and 43% for male rats (fig. S1D), indicating that the thymus is strongly affected by aging. We also found that for certain tissues (e.g. thymus, bone marrow, cornea, etc.), samples from older groups have a larger variance compared to younger groups (fig. S1E). This sort of increase in variance in older animals has been reported previously (1, 28–30). However, we also noted that not all tissues in our study showed an age-related increase in gene expression variance and that certain tissues (e.g., pituitary gland and thyroid) showed lower variance in older animals (fig. S1E), suggesting that an age-related increase in gene expression variance may be tissue dependent. We next identified age-related genes in each tissue using a previously established method (Fig. 1B) (1). Briefly, we first identified differentially expressed genes (DEGs) between 6 months and each of the latter ages in each tissue. We next focused on DEGs whose expression consistently increased or decreased with age: We refer to these genes as age related because they are either up- or down-regulated in a single direction throughout life. For example, we did not include genes that might be increased from 6 months to 9 months but then decreased from 9 months to 12 months, etc. The age-related genes were classified either as linear, which show constant rates of expression change across ages, or logistic, which show a large increase or decrease during a specific age period but remain relatively consistent at younger and/or older ages (Fig. 1C and fig. S2A). On the basis of when the specific change in expression takes place, the logistic genes were further classified into early, mid, and late subtypes (Fig. 1C and fig. S2A). In general, the numbers of DEGs increase with age across tissues, and there are more up-regulated genes than down-regulated genes (Fig. 1D). On average, a rat tissue shows 1323 genes up-regulated and 789 genes down-regulated in at least one age group versus the 6-month group. The respective numbers are 1098 and 570 for male mice and 1147 and 516 for female mice. Among all tissues, thymus consistently shows the largest numbers of DEGs, totaling 10,020 up-regulated or down-regulated in male rats, 6343 in male mice, and 4136 in female mice (Fig. 1D). This result is consistent with the finding that the thymus shows the highest proportion of age-related variance across samples (fig. S1E). In many tissues, most of the DEGs are age related, and most age-related genes are linear (Fig. 1E). In mouse tissues, the proportions of age-related genes among total DEGs are generally consistent between males (52% on average across tissues) and females (53%), with the highest proportions in the thymus (93% for males and 88% for females) and the central nervous system tissues (80% and 90 for the hippocampus, 87 and 79% for the hypothalamus, 92.2 and 73% for the cortex, and 82 and 79% for the cerebellum, for males and females, respectively), and the lowest proportions in the gut tissues (55 and 48% for the duodenum, 35 and 30% for the ileum, 45 and 58% for the jejunum, and 46 and 65% for the colon, for males and females, respectively), the pituitary gland (35 and 51%, for males and females, respectively), and the skin (40 and 36%, for males and females, respectively) (Fig. 1E). Some tissues show noticeable differences between male and female mice. For example, the thymus and the liver show larger proportions of logistic age-related genes in male mice (66% for the thymus and 34% for the liver) compared with females (12% for the thymus and 6% for the liver) (Fig. 1E). In addition, the logistic age-related genes in the male thymus are all late-life, while female thymus has early-, mid-, and late-life age-related genes (fig. S2B). In the male liver, late-life logistic genes make up 91% of all logistic genes, while only 9% of logistic genes in the female liver are late-life (fig. S2B). When comparing between species, tissues of male rats generally show larger proportions of logistic genes compared with the same tissues in male and female mice (Fig. 1E). To verify, if this difference reflects a true interspecies difference, we randomly picked eight rat samples, per age, per tissue, to match the sample size of male mouse tissues. We also removed the 9-month group from the rat samples because we did not have 9-month-old mice. This downsampled rat dataset mainly caused a reduction in the numbers of early-up, early-down, and late-up logistic genes across tissues (fig. S2C). Consequently, the downsampled rat dataset displayed a higher (although not substantially) ratio of late-logistic genes to early-logistic genes, compared with the male mouse dataset (fig. S2D). However, the downsampled rat dataset still displayed a higher ratio of logistic genes to linear genes, compared with the male mouse dataset (fig. S2E). Therefore, the proportionally more logistic age-related genes in rat tissues were likely real and not an artifact caused by the larger sample size or the additional age group of rat samples. We checked the age-related genes that appeared most frequently among all tissues and identified glycoprotein nonmetastatic melanoma protein B (Gpnmb) and Asxl transcription regulator 3 (Asxl3) as the top age-related genes that were up-regulated and down-regulated, respectively, in many tissues from all groups (female and male mice and male rats) (Fig. 1F). Gpnmb was up-regulated in 20 tissues of female mice, 22 tissues of male mice, and 21 tissues of rats. Asxl3 was down-regulated in 13 tissues of female mice, 7 tissues of male mice, and 11 tissues of rats (Fig. 1F). A recent study also found that Gpnmb was the top gene that universally tracked tissue aging in mice and rats (31). The full list of age-related genes in all tissues is available in table S2. Immune response pathways are universally up-regulated To get an overview of the biological functions perturbed by aging, we performed pathway enrichment analysis on the age-related genes in each tissue (Table S3). The enrichment analysis was performed on the up-regulated (i.e. linear-up and logistic-up genes were merged to one group) and on the down-regulated age-related genes. As shown in Fig. 2A, many Reactome pathways that were frequently enriched in tissues as a result of aging, hereafter denoted as age-related pathways, are related to immune response. In particular, eight pathways, “neutrophil degranulation,” “interactions between lymphocytes and other cells,” “signaling by interleukins,” “GPCR ligand binding,” “complement cascade,” “interferon signaling,” “response to elevated platelet cytosolic Ca2+,” and “DAP12 interactions,” are enriched in at least half of the tissues available for mice and/or rats: Therefore, we consider these as “common age-related pathways.” Age-related up-regulation of inflammatory and immune response is common across tissues, sex, and species. (A) Top Reactome pathways enriched in age-related genes. x axis, the numbers of tissues that enrich for a given pathway. We show only pathways that are enriched in at least eight tissues in mice or rats. (B) Age-related gene pathways that are most frequently shared among tissues (≥14 for mouse and ≥16 for rat). Heatmaps show the adjusted P value of the enrichment in each tissue; Padj. ≥ 0.01 are shown in gray. Bar plots under each heatmap show the average fold change of the up-regulated age-related genes that are related to immune response. From left to right, the bars show the log2 fold change for 12, 18, 21, 24, and 27 months versus 6 months for mice and for 9, 12, 18, 21, 24, and 27 months versus 6 months for rats. In both (A) and (B), pathways related to immune and inflammatory response are in bold. (C) Proportions of up-regulated age-related genes that are related to immune response. Each dot represents the proportion of these genes in a tissue. To ensure fair comparison between mice and rats, species-specific genes were excluded. (D) The average log2 fold-change between 27 and 6 months for genes shown in (C). Each dot represents a tissue. (E) Ratio of age-related genes to age-unrelated DEGs in immune-related and immune-unrelated genes (all but the immune-related genes). In (C) to (E), two-way paired Wilcox rank tests were performed between the indicated groups. **P < 0.01 and ***P < 0.001. n.s., not significant. Figure 2B shows the specific tissues that enrich with the common pathways defined above, and the average fold changes of the age-related genes in these pathways. Across sex and species, tissues from brain and heart frequently show enrichment with common age-related pathways, while tissues from the gut (except for duodenum in male mice) do not. We did not perfuse animals before collecting tissues. Since aging can change immune cell signatures (2, 32, 33), the residual blood in tissues could have contributed to the increased expression of immune-related genes in old animals. To estimate the potential impact of residual blood, we identified genes that were up-regulated with age in the white blood cells in Tabula Muris Senis (2) and in the peripheral blood mononuclear cells (PBMCs) in Teo et al. (32). Next, we compared these up-regulated genes with our data and found that on average, about 2% of the up-regulated immune-related genes in any mouse tissue in our study was also up-regulated in old white blood cells in Tabula Muris Senis (fig. S3A) and in old PBMCs in Teo et al. (fig. S3B) (32). We ran additional analysis on the Tabula Muris Sensis data. Tabula Muris Sensis perfused the animals when collecting samples, and there were 11 tissues in Tabula Muris Senis that correspond to 17 tissues in our data: The brain and heart in Tabular Muris Senis were each matched to four subregions of the same organs in our study. We identified genes up-regulated by age in each of these 11 tissues in Tabula Muris Senis (see Materials and Methods) and compared them with the up-regulated immune-related genes in the corresponding tissues in our data. We found that the four cardiac compartments, the tibialis anterior and skin from our data showed small overlap in genes with the corresponding tissues of Tabula Muris Senis (fig. S3C). However, for the remaining 11 tissues from our study, on average, about 36% of the up-regulated immune-related genes in a tissue were also up-regulated in the corresponding tissue in Tabula Muris Sensis (fig. S3C). On the basis of these results, we think that the residual blood in our samples has limited contribution to the age-related up-regulation of immune-related genes. However, since we did not perfuse tissues, the contribution of residual blood cells to age-related perturbations in immune and inflammatory pathways found in rodent tissues in our study cannot be entirely excluded. In addition, we used the mouse snRNA-seq atlas (PanSci) (21) as a reference to estimate changes in various cell populations during aging in our study. The PanSci dataset includes 10 tissues that correspond to 18 tissues in our study (heart, brain, and skeletal muscles in PanSci each correspond to multiple tissues in ours). We deconvoluted our bulk RNA-seq data for these 18 mouse tissues (males and females) by using the snRNA-seq data from the matching tissues and sex (see Materials and Methods for details). We observed an age-related increase in the immune cell populations, mostly myeloid cells, in the four regions of the brain, gonadal white adipose tissue (gWAT), tibialis anterior, kidney, and the four heart compartments, for both sexes (fig. S4 and table S4). Thus, it is possible that the age-related increase of immune cell populations might contribute to the up-regulated expression of immune-related genes locally in these tissues (Fig. 2B). It is possible that these immune cells are inherently tissue resident, or they derive from blood monocytes. To explore more age-related changes in the immune system, we defined immune-related genes as genes under the Reactome top annotation “Immune System,” which covered 1876 mouse genes and 1870 rat genes in our annotations. Overall, the relative numbers and fold changes in the expression of up-regulated immune-related genes were similar between male and female mouse tissues but were lower in rat tissues (Fig. 2, C and D). We next compared the age-related changes in the immune-related genes and in the immune-unrelated genes, which we defined as all genes minus the immune-related genes. We found that both the immune-related genes and the immune-unrelated genes have a higher tendency to be up-regulated with age and that the proportion of up-regulated genes is larger for immune-regulated genes (fig. S5, A and B). We also found that the ratio of age-related genes to age-unrelated DEGs (i.e., DEGs that do not show monotonic gene expression with age) is higher for the immune-related genes compared to the immune-unrelated genes (Fig. 2E and fig. S5C). On the other hand, these two groups of genes overall did not show significant differences in the ratio of linear genes to logistic genes (fig. S5D). However, at more granular level (fig. S5E), several pathways had a higher ratio of logistic genes: Examples of such pathways are “digestion and absorption,” “circadian clock,” etc. (fig. S5E). These findings suggests that aging differentially modulates expression of genes that belong to different functions. We also performed pathways enrichment analysis separately on the early, mid, and late subtypes of age-related genes and linear genes. As shown in fig. S6, linear genes enriched for pathways that are most frequently shared by multiple tissues in both species, and the most common pathways, are related to immune function, G protein–coupled receptor (GPCR) signaling, and extracellular matrix organization (ECM). The late-logistic genes were the second most frequent subtype: These genes enriched for multiple pathways shared by three or more tissues in either species (fig. S6). Immune pathways were again featured among the most prevalent late-occurring transcriptomic changes (fig. S6). Late–down-regulated pathways were mostly found in rats, and these were linked with down-regulation of mitochondrial metabolism (pyruvate metabolism and the tricarboxylic cycle) and fatty acid, glycogen, and glucose metabolism (fig. S6). We searched for transcription factors (TFs) that could be responsible for the up-regulation of immune-related genes. We compiled known TF-target interactions from public databases and then checked TFs that showed high correlation in gene expression with the up-regulated immune pathways in each tissue (see Materials and Methods for details). We found that Ikzf1, Runx1, Pou2f2, Runx3, and Irf8 frequently showed high positive correlation with the immune pathways in tissues, in both sex and both species (fig. S7). These TFs are known to regulate the proliferation and development of myeloid and lymphoid cells (34–37), suggesting that the increase in expression of genes related to immune function in rodent tissues could be a result of an increased presence of immune cell in older tissues (fig. S7), as has been reported (2), or tissues responding to perturbations such as the senescence-induced secretory program (38). A full list of TFs that were associated with all age-related pathways in each tissue is listed in table S5. Differences between female and male mice To characterize biological processes that show differences between sexes during aging (“intersex” differences), we compared the numbers of the age-related genes (fig. S8A) and the age-related pathways (Fig. 3) for each tissue between female and male mice. Because immune-related pathways were frequently enriched in the tissues from both sexes (Fig. 2, A and B), we excluded these pathways from the intersex comparisons in Fig. 3 to determine sex-dependent enrichment for other pathways. Results of more comprehensive comparison including immune-related pathways were shown in figs. S8C and S9 to S14. Age-related gene pathways in mice that show sex-dependent changes. (A) Numbers of pathways enriched by age-related genes in each mouse tissue. Reactome pathways under Immune System were excluded from the analysis. Red and blue bars respectively denote numbers of pathway that are enriched in females only or males only. Yellow bars denote pathways that are enriched in both sexes. Asterisks indicate whether the numbers of male pathways versus female pathways are significantly different from each other in a tissue, analyzed by two-way G test, *P < 0.05, **P < 0.01, ***P < 0.001. (B) Age-related pathways that show sex-dependent enrichment in mice. We omitted immune-related pathways and pathways that showed differences between sexes in less than three tissues. The bar plot on the left shows numbers of tissues enriching for the given pathways. Yellow, blue, and red show numbers of tissues that enrich for a pathway in both sexes, in males only, and in females only, respectively. We considered that a pathway is male enriched if yellow + blue > red. Female-enriched pathways are defined similarly. The remaining pathways were considered as sex divergent (fig. S8B). Asterisks indicate whether the numbers of male tissues versus female tissues are significantly different from each other for a pathway, analyzed by two-way G test *P < 0.05, **P < 0.01. In heatmaps, red and blue show adjusted P values of pathways that are significantly enriched with up-regulated and down-regulated genes, respectively. Gray marks nonsignificant enrichment (Padj ≥ 0.01). Pathways enriched in the up-regulated genes and down-regulated genes are also marked by the initial “U:” and “D:,” respectively. Pathways are grouped on the basis of their top-level function in Reactome. We found that the pituitary gland (fig. S12) and the hypothalamus (fig. S8C) enrich for more pathways in females, compared with males (Fig. 3A and fig. S8A). The colon, lymph node, cerebellum, cortex, hippocampus, left atrium, and soleus from females also enrich for more pathways than their male counterparts, although the differences are not significant (Fig. 3A and figs. S8C, S11, and S13). These observations are consistent with recent findings that brain tissues of female mice age faster than those in males (39). On the other hand, the thymus, femur, gWAT, liver, and duodenum had significantly more age-related pathways in males compared with females (Fig. 3A and fig. S8A, S10, S12, and S14). We next characterized pathways that showed sex-dependent enrichment (sex-enriched pathways). We considered a pathway sex-enriched to the female if 50% or more of the tissues that enriched for the pathway were only in the females. Male-enriched pathways follow a similar definition. We note that this arbitrary classification of pathways was meant to group the pathways for better visualization. Under this classification, many of the female-enriched pathways are related to the ECM reorganization and are down-regulated with aging (Fig. 3B). Age-related pathways that appear only in male tissues include pathways related to cell proliferation, and these are down-regulated in the thymus, skin, and ileum (Fig. 3B). We found that down-regulation of these male-enriched pathways was correlated with down-regulation of TFs involved in cell proliferation [e.g., Foxm1 proliferation–associated TF (40) and E2f8—a transcriptional regulator of cell cycle–dependent gene expression (41)] and DNA damage repair (e.g., Brca1—a tumor suppressor gene) (fig. S7). We also characterized pathways that overlapped highly, but not completely, between male and female mice (fig. S8B), indicating limited intersex divergence in these pathways. For example, while down-regulation of pathways related to collagen turnover and assembly were found in the diaphragm, gastrocnemius, soleus, and skin in both sexes, these pathways were additionally found in the pituitary gland and femur of the female and in the tibialis anterior of the male (fig. S8B). Last, we present a more granular comparison of the neuronal tissues in male and female mice, including the cerebellum, hypothalamus, hippocampus, and cortex (fig.S8C). The most pronounced intersex difference here is that significantly down-regulated pathways are found only in females, with GPCR ligands being linearly down-regulated in the female cerebellum (fig. S8C). Overall, although immune pathways enrich to brain tissues in both sexes, up-regulation of these pathways is more exaggerated in females (fig. S8C). Similar comparisons for all other tissues are presented in figs. S9 to S14. Inter-rodent species differences in aging—Comparing mice and rats To evaluate differences between species, we compared pathways enriched by age-related genes between tissues of male mice and male rats. The quality of the gene and pathway annotations differ between mouse and rat, which can bias the comparison. Therefore, when performing pathway enrichment analysis, we removed species-specific genes from the age-related gene and pathway annotations. Because fewer genes were used in the pathway enrichment analysis, we relaxed the threshold for the adjusted P value of enriched pathways from 0.01 to 0.05. Since we already showed that the immune response is overall stronger and more prevalent in mouse tissues compared with rat tissues (Fig. 2B), here, we excluded pathways related to the immune response to focus on additional interspecies differences. We found that mouse duodenum, right ventricle, right atrium, gWAT, and liver enriched for significantly more age-related pathways, compared with the counterpart tissues from rats (Fig. 4A). On the other hand, the thymus, gastrocnemius, colon, and jejunum enriched for significantly more pathways in rats compared with mice (Fig. 4A). The inter-species differences in these tissues held up when we compared the numbers of age-related genes between male mice and male rats (fig. S15A). We also compared the downsampled rat data with the data from the male mouse. We did this to control for potential artifacts caused by differences in samples sizes between rats and mice. Most tissues that showed more age-related genes in the full rat dataset versus the male mouse dataset held up with down-sampling (fig. S15B). Thus, difference between male rats and male mice are likely to reflect true biological differences rather than artifacts due to sample size variations. Age-related gene pathways that show species-dependent changes. (A) Numbers of pathways enriched in age-related genes in male mouse and rat. Reactome pathways under Immune System were excluded from the analysis. Yellow and blue bars, respectively, denote pathway numbers that are enriched only in rats or only in mice. Red bars denote pathways that are enriched in both species. Asterisks indicate whether the numbers of pathways in mice are significantly different from rats when analyzed by two-way G test *P < 0.05, **P < 0.01, ***P < 0.001. To avoid bias in gene and pathway annotations, only the common genes and pathways were used in the enrichment analysis. (B) Pathways that show species-dependent enrichment. We omitted Reactome pathways under Immune System and only analyzed pathways that were different between species in at least three tissues. A bar plot on the left shows numbers of tissues enriching for the given pathways. Red, blue, and yellow show numbers of tissues that enrich for a pathway in both species, in male mice only, and in male rats only, respectively. We consider the pathway mouse enriched, if red + blue > yellow. Rat-enriched pathways are defined similarly. The remaining pathways were species divergent (fig. S15D). Asterisks indicate whether the numbers of mouse tissues versus rat tissues are significantly different from each other for a pathway, analyzed by two-way G test *P < 0.05. In heatmaps, red and blue show adjusted P values of pathways that are significantly enriched with up-regulated and down-regulated genes, respectively. Gray marks nonsignificant enrichment (Padj ≥ 0.05). Pathways enriched in the up-regulated genes and down-regulated genes are also marked by the initial U: and D:, respectively. We next characterized pathways that showed species-dependent enrichment (species-enriched pathways). We defined rat-enriched pathways and mouse-enriched pathways in the similar ways as was done for the sex-enriched pathways described earlier. For many pathways that showed species-dependent enrichment, the rat or mouse tissues that contributed to these differences have diverse functions or anatomical origins (Fig. 4B), making it difficult to infer biological implications. However, we observed a few pathways where the species-dependent differences involve functionally related tissues, perhaps suggesting a true interspecies difference in aging in these tissues. For example, the lower-limb muscles (soleus, gastrocnemius, and tibialis anterior) showed down-regulation of pathways related to glycogen metabolism and respiratory electron transport in male rats but not in male mice (Fig. 4B). We previously reported this difference in the skeletal muscle data subset and associated these transcriptomic changes with muscle age-related atrophy (3). Another example is the liver and gWAT: These tissues showed down-regulation of pathways related to amino acid metabolism in male mice but not in male rats (Fig. 4B), suggesting a larger reduction in nutrient metabolism in old mice, compared with rats. To ensure that the species-dependent pathways were not confounded by different sample sizes for mice versus rats, we repeated the analysis by using the downsampled rat data. As shown in fig. S15C, the two above examples were still valid. Many of the pathways that showed less species-dependent differences were related to collagen synthesis, assembly, and degradation (fig. S15D). While these pathways were enriched in the skeletal muscle, thymus, kidney, and bone marrow of both species, they were up-regulated in the liver and spleen of male mice but were down-regulated in the same tissues of rats (fig. S15D), which suggests subtle differences in the ECM remodeling during the aging of the two species. Next, we present a detailed comparison among the neuronal tissues between male mice and male rats (fig. S16). For both species, we analyzed four brain tissues (cerebellum, hippocampus, hypothalamus, and cortex). Overall, most pathways in these brain tissues of both species were enriched in the linear-up genes and were related to immune and inflammation pathways (fig. S16). For rats, in addition to brain tissues, we also included the lumbar region of the spinal cord sampled at the L3-L6 level; this includes the L4-L5 segments where the sciatic nerve initiates in Sprague-Dawley rats (42). The lumbar region of the rat spinal cord showed many mid-up pathways (in addition to linearly up-regulated pathways), which were rare in the brain tissues in either species (fig. S16). Most of the mid-up pathways in rat spinal cord were related to extracellular matrix turnover, immune, and inflammation (fig. S16). Interspecies comparisons for other tissues are presented in figs. S17 to S22. Correlated aging among tissues Last, we asked whether age-related transcriptomic changes correlated among tissues. We focused on changes in the expression of immune genes due to their prevalent up-regulation in rodent tissues. For each tissue, we asked whether the average expression of immune genes in an individual sample is higher or lower compared with the age-group average. We then calculated a Spearman correlation in the departure-from-average between samples from two tissues (see Materials and Methods for more details). We observed significant correlation among tissues that were from the same organ or had related functions. For example, the cardiac compartments showed significant correlation among themselves and with liver in female mice, male mice, and male rats, and with kidney in the female and male mice (Fig. 5, A to C). While we did not observe correlation between the cardiac compartments and kidneys in the rat (Fig. 5C), this may have been because we were missing kidney samples from two age groups in rats. In both female and male mice, the up-regulation of immune and inflammatory response appeared earlier in the kidney compared with the liver or the heart (Fig. 5, D and E), suggesting that age-related biological changes in kidneys may precede those in the liver and the heart during the mouse aging. Tissue correlation based on their “age gaps.” (A) Correlation in the age gaps between tissues in female mice. We calculated an apparent age for each sample based on the expression levels of up-regulated immune genes and defined the age gap as the difference between the apparent age of the sample and the average apparent age of all samples that were from the same age group and tissue as the given sample (see Materials and Methods for details). Heatmap shows the Spearman correlation coefficients between the age gaps of samples of two tissues. For clarity in visualization, we colored insignificant (Padj ≥ 0.05) correlations to white, but the tissues were clustered with the raw correlation coefficient. (B and C) Correlations among tissues in male mice and male rats, respectively. (D to F) Transition ages for immune gene up-regulation in the tissues of female mice, male mice, and male rats, respectively. The transition age is the extrapolated age at which the expression level for a given gene increases by 50% relative to 6 months. In male rats, we observed correlations among the lower-limb muscles, gastrocnemius, tibialis anterior, and soleus (Fig. 5C), and there was a correlation between the gastrocnemius muscle and the lumbar region of the spinal cord (Fig. 5C). Because the sciatic nerve that innervates lower limbs initiates at the lumbar region, the correlation between the spinal cord and the gastrocnemius muscle likely reflects coordinated aging of the neuromuscular system. The timing for up-regulation of immune response is similar in the spinal cord, tibialis anterior, and gastrocnemius (Fig. 5F), while in the soleus muscle, the increase in immune response occurs later (Fig. 5F). We also measured the correlation between two tissues by using the number of up-regulated age-related genes related to immune functions shared between the two tissues. Under this metric, highly correlated tissues are mainly those that had the same anatomical origin (fig. S23A). The result was similar when we measured the correlation between two tissues by the numbers of all age-related genes shared between the two tissues (fig. S23B). DISCUSSION In this study, we generated a comprehensive, high N, multi–time-point transcriptomic atlas of aging in rodents by profiling 28 tissues in male and female C57BL/6J mice and 32 tissues in male Sprague-Dawley rats. Mice and rats are the most commonly used species in preclinical research, including aging studies. By characterizing age-related genes, i.e., genes that show monotonic changes in expression during aging in each tissue, we compared the aging signatures across tissues, sex, and species. Although mice (Mus musculus) and rats (Rattus norvegicus) belong to the same rodent family Muridae (43), with similar lifespans (25, 26), we found many differences in the molecular profile of aging between equivalent tissues from the two species. We also analyzed similarities and differences between male and female mice (we did not include female rats because of the high prevalence of breast tumors seen in female Sprague-Dawley). Side-by-side comparison of the age-related transcriptional landscape between species may help to choose a more appropriate model for preclinical testing; depending on the tissue of interest, a mouse may be more appropriate in certain instances, while the rat may be the better model in other settings—given the data in this study. For instance, as we have previously reported, lower-limb muscles of rats show more age-related genes that are linked with sarcopenia compared with age-matched mice; also, only the rats demonstrated a decline in mitochondrial gene expression, which had been reported in humans, and, hence, rats may be more useful for modeling human sarcopenia (3). Compared with similar studies of multitissue aging signatures in mice (2, 20, 21), we profiled more tissues [28 mouse tissues in our study versus 17 tissues in Schaum et al. (20) and 14 tissues in Zhang et al. (21)], included rats, and used more biological replicates. The large sample size increased the power to detect changes in gene expression; however, because we profiled 5297 samples, we had to rely on bulk RNA-seq, but this allowed us to be more comprehensive in our analysis. While the field is moving toward characterizing aging signatures by using scRNA-seq (2) and snRNA-seq (21), bulk RNA-seq performed at high number of reads (100 million reads for the present study), provides coverage of a more comprehensive number of transcripts and is more sensitive in detecting lowly expressed genes. Therefore, for example, our dataset is also suitable for exploring the age-related regulation of transposons, as well as noncoding RNAs: This adds value given increasing interest in the role of noncoding RNAs (44, 45) and transposon activation (46, 47) in age-associated diseases. However, we are aware of the limitations of bulk RNA-seq, such as its lack of resolution at the single-cell level, inability to distinguish heterogeneity within a cell population, and the potential to miss rare cell types or states. To this end, combining our data with the scRNA-seq atlas generated by the Tabula Muris Consortium (2) and the recently developed single-cell nuclei (PanSci) atlas (21) will greatly enhance existing resources for aging biology. While scRNA-seq and snRNA-seq provide distinct cellular profiles that are missing in our datasets, our study provides more comprehensive survey of gene expression in each tissue while avoiding perturbations that may be induced because of cell stress and viability caused by tissue digestion. As a result of having analyzed multiple age groups in both species (6, 12, 18, 21, 24, and 27 for male and female C57Bl/6J mice, plus an additional 9 months for rats), we were able to more granularly track the behavior of age-related genes. To this end, we identified genes whose expression changes in a linear way from young adulthood (6 months) to very old age (27 months), as well as genes whose expression experiences a distinct change (“inflection point”) at a set time, during early-life, mid-life, or late-life. Behavior of the “linear” age-related genes is reminiscent of biological clocks as these track (up or down) the age of the animal. Similar use of transcriptomic profiling to generate an aging clock has been reported in the case of human blood cells (4) and human dermal fibroblasts (48). The age-related perturbations in logistic genes echo with several nonlinear behaviors reported before, e.g., the expansion of certain immune cells in mice toward late-life (21) and dysregulation of the transcriptome and proteome in middle-age and old human samples (8). Gene signatures that experience a distinct change (inflection point) at a set time (early, mid, or late-life) raise follow-up questions as to which set of molecular signatures are causative to age-associated decline in tissue functions. Identification of “causative” factors will require extensive mechanistic studies. For our skeletal muscle datasets, we have previously performed correlative analyses between gene signatures and decline in muscle phenotype. To this end, we identified late-logistic genes and signatures that coincide with sarcopenia, age-associated decline in muscle mass and function (3, 49). In rat lower-limb muscles, which suffer from severe sarcopenia, analysis of gene pathways that enrich with late response genes revealed down-regulation of metabolic and mitochondrial pathways (fig. S20). In addition, perturbations of gene markers of myofiber denervation and atrophy coincided with the time course of muscle wasting, with the onset at 18 months (3). Skeletal muscles in mice, which display milder sarcopenia compared with rats, were spared from such dramatic perturbations of late-life transcriptomic signatures (3). Similar correlative analyses between age-related decline in organ function and transcriptomic signatures can be performed for other tissues, provided that organ function data are available. In this study, we found that the up-regulation of immune and inflammatory pathways was one unifying theme of aging in many rodent tissues. Expression of genes that relate to immune and inflammatory pathways tended to increase with age, and this trend was common to many tissues in both species. These findings suggest that during aging, the immune and inflammatory environment tends to start changing relatively early in life, and these changes progress at a constant pace and direction, while the rate and/or direction of changes in other biological process varies by tissue. Age-related up-regulation of immune and inflammatory pathways is a common finding in aging studies in humans (50–52) and rodents (1, 2, 20) and is thought to be a major contributor of failing proteostasis observed in resident cells (53). Here, we did not conduct scRNA-seq to identify cells that contribute to the increased inflammatory milieu in old rodent tissues. However, other reports suggest that increased inflammation may be due to age-related changes in the immune system, related to dysregulation of macrophages with an increasing proinflammatory phenotype (53), infiltration of certain subpopulations of T and B cells (20, 21) including an increase in Gzmk+ T cells associated with increased inflammation (33), and accumulation of immunoglobulin G–expressing cells (54). In addition to reinforcing prior findings that an increase in the inflammatory milieu is a characteristic of aging, we enrich this prior knowledge by providing side-by-side comparison between sexes (male and female mice) and species (male mice and male rats) and specificity as to the particular inflammatory pathways that we found to be up-regulated. The consistent finding of an increase in interferon signaling was especially notable. It has been suggested that this may be due to a loss of heterochromatin, resulting in de novo transcription of particular transposons, and thus increases in double-stranded RNA, due to their sequence structure. Thus, the cells seem to be responding in part as if they are experiencing a viral infection (55). In addition to the interferon pathway, we saw evidence of an increase in cytokine signaling, given the up-regulation of the signaling by interleukins pathway. Overall, the degree of increase in the inflammatory and immune responses was comparable in female and male mice. However, when making interspecies comparisons, increase in immune and inflammatory pathways was overall lower in rats. Since we did not perfuse animals before collecting samples, we could not exclude the possibility that immune cells from the residual blood could have contributed to age-related perturbations in immune-related gene expression in the tissues. To address this limitation, we compared our datasets with the publicly available data (2, 21, 32) and showed that actual contribution of residual blood cells to immune gene expression in our tissues was relatively small. First, the up-regulated immune-related genes in our aging mouse tissues had only a small overlap with genes up-regulated in PBMCs (32) or in the white blood cells from old mice (2). However, we found a larger overlap with genes up-regulated with age in the perfused mouse tissues from Tabula Muris Senis (2). Second, when we used the data from PanSci snRNA atlas (21) as the reference to deconvolute our mouse datasets, we found that many of our mouse tissues showed age-related expansion of the immune cell populations: The age-related expansion of microglia in mouse brain tissues is one notable example. These two pieces of evidence, albeit indirect, suggest that the age-related up-regulation of immune-related genes in many tissues in our study could be a result of the expansion of tissue-resident and/or tissue-infiltrating immune cells. Nevertheless, we cannot completely exclude the impact of residual blood in the rodent tissues collected in our study since the animals were not perfused. We found that the “burden” of aging, measured by the numbers of age-related genes, differ among tissues. In female and male mice, intestinal tissues (duodenum and ilium) and skin had the lowest number of age-regulated genes, and most of these genes were linear. In rats, the duodenum also had the lowest number of age-related genes compared to other rat tissues. Moreover, intestine tissues from both mice and rats seem to be refractory to age-related inflammatory changes and did not enrich for inflammatory pathways (except for duodenum in male mice). Earlier studies have reported that mouse intestinal cells show low numbers of age-related genes (20, 56, 57). A recent study also reported that mouse intestine tissues showed minimum expansion of CD8+ Gzmk+ T cells during aging (21). These findings led to the question as to why different tissues seem to experience aging differently. Several studies proposed that quiescent or slow-proliferating tissues, like the liver, brain, etc., could accumulate more DNA damage and therefore show stronger signs of senescence, whereas fast proliferating tissues, such as the gut and skin, replace damaged cells more efficiently (58, 59). Our mouse data seem to agree with this proposal. Many tissues showed sex-dependent and/or species-dependent aging. A comparison of males to females revealed the tissues that are divergent between the sexes; we characterized tissues with the fewest overlapping pathways and the tissues that are more similar, which had the most overlapping pathways. Once we removed immune and inflammatory pathways (since these are major shared aging pathways), the pituitary gland and hypothalamus were two tissues that stood out, where female mice had many more pathways that were age-regulated compared with males (female enrichment). The femur, gWAT, liver, and duodenum showed male enrichment. Prominent sex effects on fat and liver aging were also observed in a prior study (20). Comparison between species (Fig. 4A) also revealed tissues that are the most different in age-related gene expression when comparing male mice and male rats. For instance, brain tissues—cerebellum and hippocampus—had more mouse-enriched pathways, while gut tissues, colon and jejunum had more rat-enriched pathways. In addition, we have previously reported differences in the lower limb muscle aging between mice and rats (3). Exploring the sources of the sex-related and species-related differences in the aging of these tissues may inform certain mechanisms of aging. It might be somewhat of a surprise to note significant differences in aging-related gene expression changes between mice and rats since each of these species are often studied in the hope that they represent the human. Learning that there are significant age-related differences even between these often studied rodent species might increase caution when assuming translation, forcing us to validate specific age-related settings for their translatability. This study should serve as a useful source of comparison to the human for this purpose. Aging has a systematic impact on the whole body; therefore, different organs and tissues could display correlated changes (60, 61). Organs and tissues could interact with each other to facilitate age-related changes, for example, through secreted proteins and through cell-to-cell contact (e.g., the neuromuscular junction provides a way for motor neurons to communicate with skeletal muscle), and there are reports of potential causal interaction among senescence between tissues since senescent cells secrete a host of proteins labeled as the “senescence-associated secretory phenotype” or SASP (62). We found correlations in the aging status of tissues with closely linked functions, e.g., the heart and kidney (63), or tissues that were from the same organs, e.g., the four cardiac compartments. While further studies are required to verify these intertissue correlations, we think that certain correlations could reflect potential interactions during the aging of two tissues. Below, we explore this possibility by using the observed correlation between the lumbar region of the spinal cord and the lower-limb muscles in rats as an example. In the current study, which illustrates the value of having a comprehensive dataset, we can now contextualize our prior work in skeletal muscle by comparing it to the new study of the spinal cord. In the case of rats, because we included the lumbar region of the spinal cord, we were able to analyze temporary transcriptomic changes in this tissue, in relation to muscles that are innervated by the sciatic nerve, which initiates from the lumbar region. One of the direst consequences of aging is deterioration of neuromuscular junctions (NMJs), resulting in the loss of effective innervation and diminished muscle contractility (64–66). Age-related functional denervation of myofibers has been described in humans (67, 68), mice (69, 70), and rats (71), with fast (glycolytic) muscles being more affected than slow (oxidative) muscles. However, despite extensive evidence of neural dysfunction as a hallmark of human sarcopenia, it remains unclear whether myofiber denervation is initiated and driven primarily by changes in the central nervous system (motoneurons), or by myofibers, or both (72). Thus, it is believed that NMJ integrity is maintained by bidirectional signaling: a retrograde signaling from the muscle to the motoneuron body and anterograde signaling from the motoneuron body to the muscle (73). Our approach tracks a cross-sectional time course of age-related transcriptomic changes in this neuromuscular unit (muscle and spinal cord) across seven ages along the lifespan of rats. In rat gastrocnemius and tibialis anterior muscles (fast glycolytic muscles), markers of NMJ destabilization (e.g., Musk, Chrnd, and Ncam1) are significantly up-regulated at 18 months, compared to all analyzed younger ages, and expression of these genes progressively increases by 27 months (3). We compared the chronology of perturbations of age-related pathways in these muscles and the spinal cord (fig. S22). Both the spinal cord and the lower-limb muscles have many linear age–related pathways, among which increase of immune and inflammatory pathways is prominent. In addition to the linear pathways, which start to change early in life, the spinal cord shows many mid-life pathways that enrich with age-related genes with the point of inflection between 12 and 18 months. At this point, induction of markers of activated microglia (e.g., Cd68) (74) and disease-associated microglia (e.g., Apoe, Ccl3, Atf3, and Lgals3) (75) was seen in the rat spinal cord, pointing that microglia in rat spinal cords gradually assume neurodegenerative phenotype. In conclusion, the spinal cord changes in the rat come before the dominant changes seen in the rat skeletal muscle, i.e., the lumbar region of the spinal cord shows many mid-logistic genes, whereas many late-life pathways are seen in lower-limb skeletal muscles. These findings suggest the possibility that age-related motor neuron changes might be causative of muscle changes—or at least need to be considered earlier than muscle changes. The most consistently up-regulated gene in our study is Gpnmb. The most consistently down-regulated is Asxl3. Gpnmb was up-regulated in 20 tissues of female mice, 22 tissues of male mice, and 21 tissues of rats. Asxl3 was down-regulated in 13 tissues of female mice, 7 tissues of male mice, and 11 tissues of rats. It is tempting to look to these as reliable markers of aging and to follow up on these findings as to the age-related function of these genes in particular. Given the expanse of our transcriptomic dataset and the evidence of both its consistency and further enhancement of prior data, we hope this study will serve as a useful resource for understanding the timing, the tissue specificity, the gender specificity, and the species specificity of age-related gene and pathway changes in mice and rats. The timing, in particular, may point to the earliest changes as potential “triggers,” inducing some of the changes seen later on. Inclusion of both sexes helps to mitigate overreliance of preclinical experiments on males since more historical data are available for male rodents. The differences between rats and mice might further help the researcher make informed decisions when it comes to selecting preclinical animal models to study age-related disorders. Also, the high N used in this study should help the researcher to define the N required for future age-related gene expression studies. From these perspectives, these data should be helpful in choosing the most appropriate rodent, the most appropriate tissue, and the most appropriate N to model distinct aspects of human aging. MATERIALS AND METHODS Animal maintenance and tissue collection All procedures involving animals were approved by the Institutional Animal Care and Use Committee of Regeneron (NYS AWP-ID# A-125) and New York Medical College (NYS AWP ID# A-122), NY, USA. Male and female C57BL/6J mice were purchased from the Jackson Laboratory (USA) at various ages and maintained at Regeneron animal holding facilities for ~12 weeks before tissue collection under specific pathogen–free (SPF) conditions. The mice were housed in groups of four to five per cage with controlled temperature and light (22°C, 12-hour light/12-hour dark cycle: lights on at 0600 hours/lights off at 1800 hours) and with ad libitum access to food (PicoLab Rodent Diet 20, Lab Supply) and water. Male Sprague-Dawley rats were purchased at 3 to 4 weeks of age from Envigo (Indianapolis, USA) and aged at Envigo under SPF conditions. Rats were imported to New York Medical College animal holding facilities 8 to 12 weeks before tissue collection. Once imported, the rats were housed two per cage at the SPF facility with controlled temperature and light (22°C, 12-hour light/12-hour dark cycle: lights on at 0600 hours/lights off at 1800 hours) and with ad libitum access to food (2014 Teklad Global 14% Protein diet, Envigo) and water. Tissue collection occurred over several months, and tissues from the same age group (n = 8 to 12 per group) were collected on the same day between 9 a.m. and 1 p.m. Before tissue collection, the mice were euthanized by CO2 asphyxiation. The rats were anesthetized with 3.5% isoflurane and euthanized by exsanguination and thoracotomy. Organs were removed, and the tissues were dissected on ice, preserved in RNAlater solution (Thermo Fisher Scientific), and stored at −80°C until further processing. The list of analyzed tissues for female and male mice and male rats is shown in table S1. RNA sequencing RNA was extracted using MagMax-96 total RNA isolation kit (Thermo Fisher Scientific). The RNA concentration was quantified using NanoDrop Spectrophotometer (NanoDrop Technologies), and the integrity was assessed using Fragment Analyzer (Agilent). Strand-specific total RNA-seq libraries were prepared using Kapa RNA HyperPrep kit (Roche) with Fastselect rRNA HMR ribosome RNA removal kit (Qiagen). RNA-seq libraries were sequenced at 100 million reads using Illumina Novo-Seq 6000 platforms in paired-end mode to a length of 2 × 76 base pairs. Sequencing reads were processed with ArrayStudio (11.7.4) RNA-seq pipeline to quantify gene expression levels. For the mouse, we used mm10 genome reference and Refseq-based inhouse gene model. For the rat, we used Rn7.0 genome reference and Refseq-based inhouse gene model. Given large number of samples collected for sequencing, rodent tissues were consistently collected within the same time frame each day, between approximately 9:00 a.m. and 1:00 p.m., to minimize potential circadian or metabolic effects. For each tissue, RNA extraction was performed for all samples in the same run, and sequencing was conducted simultaneously for all samples to ensure consistency. Sample quality control We applied a consistent standard to remove outliers in each tissue. Samples of each tissue, sex, and species were analyzed separately. We previously reported results of the diaphragm, soleus, tibialis anterior, and gastrocnemius from mice and rats (3). Here, we reanalyzed the same samples so that results from all tissues could be compared consistently. We normalized the raw counts of gene expression using the limma package (3.58.1) (76) and then used isotree (0.6.1-1) (77) to score the level of abnormality of each sample. Specifically, we pooled samples of all time points and ran isotree with ndim = 2, nTrees = 5000, and scoring_metric = “depth” while using the default values for all other parameters. The scores of samples were then converted to z scores. Samples were considered outlier if (i) their z score > 3.3, (ii) their z score > 2.33 and their RNA quality was relatively low [DV200 < 70%, RQN (RNA quality number) < 5, 28 s/18 s <0.5, or concentration < 50 ng/μl], or (iii) their z score > 2.33 and their proportions of mappable reads were below 0.8. After the initial inspection, we observed unusually large numbers of DEGs in some cerebellum samples of rats when comparing 18 to 6 months (fig. S24). We visually inspected the PCA plot for these samples and identified six 18-month samples that could have been outliers: These six samples were excluded from the downstream analysis (fig. S24). We also removed all 18-month rat thymus samples. While these samples passed the initial quality check, they expressed little to no Prss16, Psmb11, and Rag1, which are known to be expressed in the thymus. See table S1 for the numbers of samples before and after quality check. After the initial quality check, we noticed a batch effect in the samples of the four rat heart compartments. In each heart compartment of rat, the proportions of reads mapped to exons among all mapped reads were lower in samples from 6-, 12-, and 21-month groups (fig. S25A). Under this batch effect, there were much more DEGs (versus 6 months) at 9 and 18 months than at 12 and 21 months in each heart compartment (fig. S25B). However, we found no abnormality in the RNA quality and other RNA-seq metrics of these samples (fig. S25C). We also excluded the possibility of sample swapping between time points or heart compartments (fig. S26). Because samples from all the other tissues of rat showed good quality, indicating that the animals were healthy, and because that the aging status of heart matters, we decided to keep samples of rat heart that had passed the initial quality check. When identifying age-related genes from the heart of rats, we applied additional steps to filter out genes that were affected by the batch effect. Details are described in section Identification of age-related genes. Downsampling We randomly downsampled the rat data such that each age group of a tissue had at most eight samples. This was done to match mouse numbers used per age group and to make sure that analyses are not confounded by varying numbers of animals per group between species. Identification of age-related genes All data analyses were performed with R (4.3.3). We identified age-related genes from RNA-seq data by using the earlier described method (1), with minor modifications. We processed the data from each species, sex, and tissue separately. Some genes had the same gene names in the gene model we used; therefore, we added the raw counts of genes that had the same gene names. We used the R package EdgeR (4.0.16) (78) to preprocess the counts, filtering genes with low expression using the function filterByExpr and normalizing out library size using calcNormFactors. We then used the limma package (3.58.1) (76) to correct for mean-variance relation and convert the corrected gene expression levels to log2CPM (counts per minute). We used the ages of samples as the grouping variate in preprocessing and mean-variance relation correction. Because we filtered out low-expression genes, we denoted this normalized log2CPM as the “filtered log2CPM” and used it for identifying DEGs and age-related genes (see below). The filtered log2CPM was not convenient for analysis involving multiple tissues because the low-expression genes could be different from tissue to tissues. Therefore, we generated an “unfiltered log2CPM” by applying all but the filtering steps described above. We used this unfiltered log2CPM for analyzing variance among samples and calculating correlations among samples. We identified DEGs between 6 months and any other age. Genes were considered age related if their expression was up-regulated or down-regulated by ≥1.5-fold at any age compared to 6 months, and the Benjamini-Hochberg–adjusted P value <0.05. Next, for each age-related gene, we fitted a null model (constant expression), a linear model, and a four-parameter logistic model to the filtered log2(CPM) over time. We used the R package drc (3.0.1) (https://doi.org/10.1371/journal.pone.0146021) to fit the four-parameter logistic model. The four-parameter logistic model is given asLog2(CPM)=c+d−c1+(AGEe)bwhere b (Hill coefficient), c, d, and e (inflection point) are the four parameters to be estimated. We used the Bayesian Information Content (BIC) of each model to identify the best model. Following our previous work (3), we filtered out genes that had null_BIC smaller than linear_BIC and logistic_BIC by 6 or more. The remaining genes were classified as linear, unless the logistic_BIC was smaller than the linear_BIC by more than 2. Genes that were classified as logistic were further classified as early-logistic if the inflection point occurred earlier than 12 months, mid-logistic if the inflection point occurred between 12 and 21 months, and late-logistic if the inflection point occurred after 21 months. Linear genes with a positive slope and logistics genes for which b * (d − c) < 0 are classified as increasing with age. Linear genes with a negative slope and logistics genes for which b * (d − c) > 0 are classified as decreasing with age. Compared with the Akaike Information Content, another commonly used metric in model selection, BIC is more suitable for identifying the true model among candidate models and favors a simpler model over a more complex one (79). Our choice of using BIC was not motivated by finding the true relation between expression of a gene and the animal age but by avoiding overfitting. Our method therefore can be regarded as a conservative approach as it may underestimate the numbers of age-related genes and favor linear genes over logistic genes. We applied additional steps when identifying age-related genes in the four heart compartments of rats due to a batch effect in these samples (see section Sample quality control). Under the batch effect, the expression levels of certain genes fluctuated across the time points in each heart compartment of rats. To identify genes that showed strong fluctuations in a heart compartment, we called DEGs between each two consecutive time points using a fold change cutoff of 1.25 and P value of 0.05. We considered a gene fluctuating if it was up-regulated (down-regulated) then down-regulated (up-regulated) in any three consecutive time points. We then removed the fluctuating genes from the age-related genes, as well as from the DEGs between 6 months and later ages, in each rat heart compartment. Variance analysis We used principal variance component analysis (80) to calculate the proportion of variance in gene expression that were related to age in each tissue. This method uses PCA to reduce the dimension of gene expression matrix, and then each principal component (PC) was regressed on age to estimate the amount of age-associated variance and that of the remaining variance. Using the eigenvalues of the PC as weight, the age-associated variance and the remaining variance were weighted averaged, respectively. The proportion of age-associated variance was calculated using the two weighted averages. We analyzed the tissues separately using the R package pvca (1.42.0) on the unfiltered log2CPM of genes. We set the minimal proportion of variance that needed to be captured by the PCs to be 0.8. For each of the four cardiac compartments of rat, we excluded the fluctuating genes before calculating the age-associated variance. We also used the PCs to calculate variance in each age groups of a tissue. Specifically, we ran PCA on the log2CPM of genes in each tissue and computed PCs that accounted for at least 80% of the total variance. For each age group, we determined the centroid in the PCA space by averaging the coordinates of individual samples in the age group. We then calculated the Euclidian distance between individual samples and the centroid, weighting the distance on each PC by the eigenvalue. The average distance between samples and the centroid was taken as the variance among samples in the age group. For each of the four cardiac compartments of rat, we excluded the fluctuating genes before calculating the variance in each age group. Pathway enrichment analysis We obtained species-dependent REACTOME pathways (81) and their associated genes from the R package Msigdbr (7.5.1). Genes that were not present in our gene annotations were removed. To focus on general biological processes, we only used the third-level annotation of Reactome. For example, the top annotation set Immune System contains a subset “Innate Immune System,” which contains a subset Complement cascade. We kept Complement cascade and removed all annotations above and below. We performed several pathways enrichment analyses that differ in the genes and pathways (see below) but used the same criteria to determine significant enrichment. We discarded pathways that overlap with less than five genes without performing Fisher’s exact test. The P values were adjusted with a Benjamini-Hochberg correction for multiple hypothesis testing, and an adjusted P value smaller than 0.01 (unless stated overwise) was considered to be a significant overlap between pathway and age-related genes. The main variation in the pathway enrichment analyses was the input genes. In the main figures, the input genes were all the up-regulated age-related genes (linear plus logistic) and all the down-regulated age-related genes. We ran enrichment analyses on the eight categories of age-regulated separately in some of the extended figures and supplementary figures. There was another variation when we compared mice with rats side by side. We generated a common pathway annotation that excluded species-specific genes to remove potential bias due to differences in the quality of gene/pathway annotations between mice and rats. To identify genes shared between mouse and rat, we downloaded a list of orthologous genes in mouse and rat from the Mouse Genome Informatics (www.informatics.jax.org). We removed ortholog genes that had multiple orthologs in the other species and those with ambiguous gene names due to synonyms. We ended finding 17,125 pairs of genes in our data that had reliable one-to-one orthologous relation between mouse and rat. Genes that were not in this list were excluded from the pathway annotation of each species and from the age-related genes. Calculate correlation among tissues We transformed the unfiltered log2CPM of a gene among samples to z scores to make sure that the range of gene expression levels were the same. The z scores of genes were calculated for each tissue separately. To calculate an aging status of a sample, we selected all the up-regulated age-related genes in a tissue that were related to immune response. We used the average z score of these genes as a metric of the aging status of the sample. The aging status of two tissues can show trivial correlation due to systematic differences among the age groups. To remove this trivial correlation, we subtracted from the aging status of a sample the average aging status of the tissue in the sample’s age group. This is similar to the age gap proposed in several studies (60, 61). We calculated the Spearman correlation between two tissues using this aging status, and considered the correlations significant if the Benjamini-Hochberg–adjusted P value <0.05. Identifying TFs associated with age-related pathways We compiled TF-gene interactions in human and mouse from ChEA3 (82), TRRUST version 2 (83), and hTFtarget (84). We mapped the genes in this dataset to the rat genes and mouse genes in our data, excluding genes that had ambiguous mapping between species, using the same method as described in the previous section. The interactions were filtered separately for rat and mouse by requiring that the TFs were from the gene ontology (GO) terms GO0003700, GO0003712, and GO0140297 (85, 86) of the given species. After excluding duplicates and self-regulatory interactions, the final lists contain 781 TFs and 0.9-m interactions in rat and 794 TFs and 1-m interactions in mouse. To associate a TF with a pathway, we checked whether the expression levels of the TF and its target genes in the pathway were highly correlated. We analyzed the tissues separately, filtering out TFs and target genes that were not age related in the tissue from the master TF-target list. To calculate the Spearman correlation in expression between TFs and their target genes, we used the z scores obtained from analyzing tissue-tissue correlations (see above). To avoid trivial correlation, we subtracted from a gene’s z score in a sample the mean z score of that gene in the sample’s age group before calculating correlation. We ranked all TF-target pairs by their correlation coefficients and used fgsea (1.28.0) (87) to test whether the rankings of a TF and its target genes in a pathway were extreme. We ran fgsea on all TF-pathway pairs where the TF had at least three targets in the pathway. We considered the association between a TF and a pathway significant if the Benjamini-Hochberg–adjusted P value <0.05 and the normalized enrichment score had the same sign as the average correlation coefficient between the TF and its target genes in the pathway. When presenting the results, we kept TF-pathway pairs that were positively correlated in at least three tissues or negatively correlated in at least three tissues. This also filtered out TF-pathways that were positively correlated in some tissues but negatively in others. Comparison with additional dataset We compared the up-regulated immune-related genes in our dataset with those from two datasets (2, 32). The first dataset was the bulk RNA-seq of Tabula Muris Senis (2). We downloaded the raw fastqs of the tibialis anterior, brain, heart, duodenum, spleen, kidney, gonadal adipose tissue, skin, bone, bone marrow, liver, and white blood cells of 6- and 27-month animals from GSE (GSE132040). We mapped the fastqs to the same genome assembly and gene model as our mice data. We identified DEGs in each tissue between 6 and 27 months by using the same method as our mice data. The only difference was that we merged the male samples and female samples in the analysis because the sample size of Tabula Muris Senis was relatively small. We included sex as a covariate when calling the functions FilterbyExpr and voom. We called significant DEGs by a fold change ≥1.5 and adjusted P value <0.05. The second dataset was an scRNA-seq data of PBMCs from young and old mice from Teo et al. (32). We downloaded the raw fastqs of this study from GEO (GSE1202505), and used Cell Ranger (6.1.2, 10X Genomics) to demultiplex and align the fastqs to the same genome assembly and gene model used in our mice data. We used Seurat to analyze the processed data, and kept the same cells that had been analyzed by the original authors. To identify genes differentially expressed between the old and young PBMCs, we used the Seurat (4.3.0.1) (88) function FindMarkers and set min.pct to 0.005. We used avg_log2FC ≥ log2(1.5) and p_val_adj < 0.05 to call genes significantly up-regulated in the PBMCs of old mice. Deconvolution of bulk RNA-seq We used the snRNA data of PanSci (21) as reference to deconvolute our bulk RNA-seq. We downloaded the author-processed scRNA data of the brain, duodenum, jejunum, ileum, gWAT, colon, heart, muscle, kidney, and liver from GEO (GSE247719). For each tissue, we kept the wild-type cells of 6, 12, and 23 months and removed genes not present in our data. We split the cells of each tissue into male and female so that we could deconvolute our data by using cells of the same tissue and sex as the reference. We used Adroit (0.2.2) (89) for deconvolution. Specifically, for each tissue and sex, we removed cell types that had less than 10 cells and then used the Adroit function get.hvfs to identify 2000 genes highly variable among cells. We set get.hvfs to balance the sizes of cell types when calculating variance of genes. We built a deconvolution reference for the given tissue and sex by using the highly variable genes and the function ref.build with the default parameters. We then used the function AdRoit.est. with default parameters to estimate the proportion of each cell type in a sample of our data that has the matching tissue and sex. We used the Jonckheere-Terpstra test to identify cell types that expanded or shrank with age in a tissue. Specifically, we used the function JonckheereTerpstraTest from the R package DesTools (0.99.60), setting nperm to 5000 and alternative to “two side” i.e., a two-side test.