All authors conceived and designed the experiments. JMZ, SP, DKI, AL, AC, ATW, DDT, KMM, EGL, KRB, XX, MPM, GF, MSHK, DS, JF, SKK, WHW, ABZ, and KGB performed the experiments. JMZ, ABO, MG, and SKK analyzed the data. SP, ABO, DKI, AL, AC, ATW, DDT, MG, KMM, EGL, KRB, XX, MPM, GF, MSHK, DS, JF, SKK, WHW, ABZ, SKK, and KGB contributed reagents/materials/analysis tools. JMZ, SP, ABO, SKK, and KGB wrote the paper.
The authors have declared that no competing interests exist.
We present the AGEMAP (Atlas of Gene Expression in Mouse Aging Project) gene expression database, which is a resource that catalogs changes in gene expression as a function of age in mice. The AGEMAP database includes expression changes for 8,932 genes in 16 tissues as a function of age. We found great heterogeneity in the amount of transcriptional changes with age in different tissues. Some tissues displayed large transcriptional differences in old mice, suggesting that these tissues may contribute strongly to organismal decline. Other tissues showed few or no changes in expression with age, indicating strong levels of homeostasis throughout life. Based on the pattern of age-related transcriptional changes, we found that tissues could be classified into one of three aging processes: (1) a pattern common to neural tissues, (2) a pattern for vascular tissues, and (3) a pattern for steroid-responsive tissues. We observed that different tissues age in a coordinated fashion in individual mice, such that certain mice exhibit rapid aging, whereas others exhibit slow aging for multiple tissues. Finally, we compared the transcriptional profiles for aging in mice to those from humans, flies, and worms. We found that genes involved in the electron transport chain show common age regulation in all four species, indicating that these genes may be exceptionally good markers of aging. However, we saw no overall correlation of age regulation between mice and humans, suggesting that aging processes in mice and humans may be fundamentally different.
This work studies the aging process in mice using DNA microarrays to identify genes that change expression during aging. The entire set of age-regulated genes constitutes a transcriptional profile that can be used to measure ages of different individuals. Furthermore, the aging expression profile highlights genetic and metabolic pathways that change with age, providing key insights about possible molecular changes that may contribute to cell senescence and physiological decline. This aging study is massive in scope, involving gene expression measurements from 16 different tissues at four different ages. Expression data for genes of interest can be queried over the web by researchers interested in aging. Different tissues were found to have strikingly different levels of age-related change, and could be divided into three groups based on their patterns of aging: a neural group, a vascular group, and a steroid-responsive group. The electron transport chain pathway stands out because it is the only genetic pathway that shows a similar pattern of age-related change in mice, humans, worms, and flies. However, there is little overall similarity between changes in gene expression during aging of humans and mice, consistent with evolutionary theories suggesting that aging lies outside the force of natural selection.
Aging is characterized by the progressive functional decline of multiple organs and tissues, eventually culminating in death. A long sought after goal has been to identify biomarkers of aging to characterize aging. Biomarkers of aging may indeed be tightly linked to aging processes, and hence may provide insight regarding mechanisms underlying aging. Changes in cellular morphology or cell type number are quantifiable traits that can be correlated with physiological age. For example, age-related changes in specific tissues include hair whitening, muscle atrophy, wrinkling of skin, thymic involution, and glomerulosclerosis of the kidney. However, these age-related tissue changes do not pinpoint specific genes or molecular pathways involved in distinct aging processes. Further, it is difficult to determine whether age-related changes in different tissues are due to common or distinct molecular causes. At the molecular level, an attractive approach is to use changes in gene expression as biomarkers of aging. DNA arrays can be used to scan a large fraction of the genome for genes that change expression with age. Identification of age-regulated genes provides key insights regarding aging.
In mice, at least 19 studies have profiled changes in gene expression with age in nine separate mouse tissues. Many of these studies find that similar classes of genes are regulated with age in different tissues. For instance, genes involved in the inflammatory response and heat shock factors have been found to increase expression levels with age in mouse brain, muscle, and heart [
These DNA microarray studies were performed using different experimental protocols, with different types of gene arrays (e.g., Affymetrix GeneChips or DNA microarrays), and by different labs. Because of these differences in methodology, most studies in mouse aging have not compared their results to each other. In this study, we present AGEMAP (Atlas of Gene Expression in Mouse Aging Project), which is a highly standardized study of gene expression changes as a function of age in mice. AGEMAP includes expression data for 8,932 genes from 16 tissues taken from the same set of C57BL/6 mice using an identical gene array platform and experimental protocol. Because the gene array experiments were analyzed at the same time and in the same way, we can compare gene expression profiles for aging between different tissues in order to identify common aging biomarkers.
Previous studies have shown that the overall pattern of expression from age-regulated genes (a gene expression profile) can be used as a biomarker of aging. For instance, a transcriptional profile of aging in human kidneys correlated strongly with a histological measure of kidney function, and thus could be used to predict physiological age [
Expression profiles can be used to compare aging in different tissues. In humans, there is extensive overlap in gene expression profiles for aging between different parts of the same tissue. Specifically, similar aging transcriptional profiles were found between the cortex and medulla of the kidney and between different sections of the prefrontal cortex of the brain [
Expression profiles of aging for one species can be compared to that of another species, thereby showing which age-related changes in expression are unique to one species and which are shared by different animals [
In this work, we present an analysis of data from AGEMAP, a database of changes in gene expression with respect to age in mice, which can be used as a resource for gerontological research. We generated mRNA transcript profiles of aging involving 8,932 genes and 16 mouse tissues at four times during aging. The raw data can be downloaded and reanalyzed by researchers interested in specific tissues or genes. We show that the entire gene expression profile of aging can be analyzed as a molecular phenotype of aging. First, we used the number of age-regulated genes in different tissues to estimate the magnitude of age-related decline among tissues and found that there were great differences in the levels of age-related change in different tissues. Some tissues showed little or no changes in expression (e.g., liver and striatum), whereas other tissues showed a great deal of transcriptional change with aging (e.g., thymus). Second, the aging expression profile for different tissues showed three main patterns for murine aging (neural, vascular, and steroid-responsive). Finally, we compared the expression profiles for mouse aging to those from human aging. Although there are a small number of genetic pathways that age similarly between the two organisms, we found no overall correlation in age-related transcriptional changes between mouse and human.
We profiled the effects of aging on gene expression in different tissues dissected from C57BL/6 mice. Mice were of ages 1, 6, 16, and 24 mo, with ten mice per age cohort (five mice of each sex). We dissected 16 tissues from each mouse: the cerebellum, cerebrum, striatum, hippocampus, spinal cord, adrenal glands, heart, lung, liver, kidney, muscle, spleen, thymus, bone marrow, eye, and gonads. These particular tissues were surgically accessible and relatively easy to harvest in the limited amount of time available before the onset of tissue ischemia. We isolated mRNA from each tissue sample and generated radiolabeled cDNA, which was then hybridized to two filter membranes containing a total of 16,896 cDNA clones corresponding to 8,932 genes. Data from 23 arrays were eliminated for quality-control reasons. In total, we generated expression data from 617 gene arrays including 16 tissues and four ages. The full set of expression data from these experiments can be found in
First, we wanted to identify genes that showed age-related changes in each tissue. We employed a multiple regression model taking age and sex into account in order to calculate the slope of expression with age for each of the 8,932 genes on the gene array (
Age-Regulated Genes
One of the tissues (gonads) consists of 20 ovary samples and 20 testes samples. When the ovary and testes samples were analyzed separately, we did not observe a significant number of age-regulated genes, possibly because the sample size is smaller than the other tissues. However, surprisingly, when we grouped the ovaries and testes together, we found a relatively high number of age-regulated genes (45 genes,
We also explored the impact of aging on the overall expression levels of gene sets and molecular pathways within mouse tissues. We used a modified gene set enrichment analysis [
Only nine of the 16 tissues showed significant levels of age-related gene expression changes. To test whether different levels of age regulation reflect different quality of expression data from the gene filters, we calculated the average Pearson correlation for all genes between biological repeats. We found no association between experimental variability and the number of age-regulated genes found in different tissues (
We were interested in finding which genes and gene sets are similarly age-regulated in multiple mouse tissues, which could potentially identify core cell biological pathways involved in common aging mechanisms. We used empirical meta-analysis to find genes, as well as gene sets that are age regulated across multiple tissues. Empirical meta-analysis combines single
We also investigated the false discovery distribution by permuting ages for all samples 1,000 times and recalculating empirical meta-analysis
We examined 401 gene sets defined by the Gene Ontology Consortium [
To determine which tissues were aging similarly, we used the commonly age-regulated genes and gene sets to group tissues by unsupervised hierarchical clustering (
(A) Rows indicate individual genes, arranged from the gene that exhibits the greatest increase in expression with age across multiple tissues at top, to the gene that shows the greatest decrease in expression with age in multiple tissues at bottom. Each column corresponds to an individual mouse tissue. Scale corresponds to the slope of the change in log2 expression with age (
(B) Rows indicate individual gene sets, arranged from the gene set increasing expression most on average with age in multiple tissues at top, to the most decreasing with age. Each column corresponds to an individual mouse tissue. Scale is the van der Waerden
To characterize the differences between the three different modes of tissue aging, we searched for gene sets that were strongly age regulated within one tissue group but not others. For each of the tissue groups, we first selected gene sets that showed strong age regulation in that tissue group (statistical significance at
Columns refer to tissues, and rows correspond to gene sets. Vertical red bars divide tissues into neural, vascular, and steroid-responsive tissue groups from left to right. Tissue names are colored to indicate affiliation with either neural (red), vascular (blue), or steroid-responsive (yellow) tissue groups. Horizontal red bars divide gene sets into those chosen for being particularly age-regulated in steroid-responsive, neural, or vascular tissues (
Genes in the cytosolic ribosome set stand out as the strongest age-related difference in expression between tissues. This gene set shows an increasing trend in expression in the neural group, but decreasing trends in the vascular and steroid-responsive groups. Increasing expression of ribosomal genes in neural tissues is curious, as rates of protein synthesis are known to decrease with age [
The steroid-responsive aging group shows decreased expression of genes in the protein targeting gene set. Genes in this gene set play an important role in protein secretion, suggesting a role for changes in hormone and growth factor expression in aging of steroid-responsive tissues.
Besides clustering tissues based on the expression profile of commonly age-regulated pathways, we also compared the degree of overlap in age regulation between each tissue with every other tissue. For each of the nine age-regulated mouse tissues, we selected genes that decrease expression with age, and genes that increase expression with age (
Similarities of Age Regulation for Nine Mouse Tissues
Individuals age at different rates, even when they have nearly identical genetic backgrounds and are grown under similar conditions. Different rates of aging can be seen by measuring the lifespan of individuals, the morphology and physiology of different tissues, or expression profiles of age-related genes. To what extent is the aging of one tissue coordinated with other tissues in the same individual? For instance, one might expect different organs in the cardiovascular system (heart and lung) to age in step with one another, but aging of the brain may not be tightly linked to aging of the heart. To test for coordination of aging among different tissues, we first developed a score for apparent age of each tissue in mice from the same age cohort based on the expression profile for that tissue. We then determined whether the apparent age of one tissue was correlated with the apparent age of another tissue in the same mouse.
To calculate an aging score for a specific tissue, we first use
For each age cohort, we were interested in seeing whether there was coordinate aging among tissues. If aging is coordinated among tissues, then a mouse may contain many tissues with high apparent ages (represented by high fractional scores) and thus generate a very high overall aging score. Conversely, another mouse may contain many young tissues and generate a very small overall aging score. If aging is not coordinated among tissues, all mice should show a mix of old and young tissues, and thus have an overall aging score close to the average for all mice. We summed the fractional scores of all tissue for a given mouse within an age cohort, resulting in ten overall scores. Coordinate aging among tissues within an age cohort should generate a set of ten overall scores that shows a high variance, whereas independent aging should show a variance similar to those generated by random permutation. For each age cohort, we compared the variance of the ten overall scores to variances generated by permutation to calculate a
At 6, 16, and 24 mo of age, we found that there is coordinate aging of different tissues (
Columns refer to tissues. The rightmost column refers to the sum of fractional scores for each mouse. Rows refer to individual mice. Identification of individual mice indicates sex (M or F), age (A1 to A4), and biological repeat (a to e); e.g., FA1a is female mouse, age 1 mo, biological repeat a. Scale of tissue columns indicates the fractional score of that individual mouse with its tissue and age cohort; a low fractional score corresponds to a mouse that expresses age-regulated genes as a younger mouse, while a high fractional score expresses age-regulated genes as an older mouse. All fractional scores are summed for each mouse; the result is the overall score shown in the overall column. A low overall score corresponds to a mouse that expresses age-regulated genes as a younger mouse, while a high overall score denotes a mouse expressing age-regulated genes as an older mouse. Each
We wanted to find genes and gene sets that show similar age regulation in different species:
This search yielded 22 genes and 17 gene sets that are commonly age regulated in mice and humans (
(A) Comparison of age-related expression changes for 2,578 orthologous genes in humans, mice, flies, and worms. Rows correspond to genes, arranged from the genes showing the greatest decrease in expression with age in both human and mouse at top to the genes showing the greatest increase in expression with age in both human and mouse at bottom. Columns correspond to tissues; tissue names are colored to denote their species of origin, either human (red) or mouse (black). Scale corresponds to the slope of the change in log2 expression with age (
(B) Comparison of age-related expression changes for 280 gene sets in humans, mice, flies, and worms. First row is the mitochondrial electron transport chain gene set, which was found to be age regulated in flies and worms in addition to mammals. Second row is the lysosome gene set, which was found to be age regulated in flies in addition to mammals. Other rows correspond to gene groups that are commonly age related in humans and mice alone. Scale refers to a Van der Waerden
One other gene set (the lysosomal gene set) showed a common increasing trend in expression with age in humans, mice, and flies, but not worms. The lysosome is involved in degradation of extracellular proteins, and increased expression of lysosomal genes may reflect increased turnover of damaged cell surface proteins in old age.
Fifteen gene pathways show common age regulation in humans and mice but not flies or worms. The peroxisome gene set shows a decreasing trend in expression with age. The peroxisome is important for detoxification of foreign compounds and also produces hydrogen peroxide, a known precursor of free radicals. Decreasing expression of genes in the peroxisome pathway might lead to decreased tolerance toward toxins and decreased levels of oxidative damage in old age. The inflammatory response and cytokine activity gene sets show an increasing trend in expression with age, suggesting that inflammation becomes more active with age. Finally, genes involved in cell cycle arrest tend to increase expression with age, implying a tendency for cells to lose the ability to proliferate in the elderly.
Next, we determined whether there is an overall correlation between age-related expression in mice and in humans. That is, among the entire set of genes that increase expression with age in mice, is there also a tendency for these genes to increase expression with age in humans, and vice versa? As a measure of age regulation across nine mouse tissues and three human tissues, we used the
(A) Comparison of age-related expression changes for 2,578 orthologous genes in mouse and human. The
(B) Comparison of age-related expression changes for 280 gene sets in mouse and human. The
We have provided an expression database on mouse aging to be used as a resource for the research community. This dataset is unprecedented in its magnitude and experimental uniformity; it contains a total of 5,519,976 expression values for 8,932 genes in 16 tissues at four times during aging with ten biological repeats at each time. The tissue samples were isolated from the same set of 40 mice, and were hybridized to similar gene arrays using identical protocols in the same lab to maximize uniformity. Out of practical consideration, the experiments were limited to the C57BL/6 strain, and future experiments will be needed to find age-related differences between C57BL/6 and other mouse strains. The full dataset can be downloaded (
In addition to revealing expression of one gene at a time during aging, the entire set of age-regulated genes can be used as a molecular phenotype of aging and may be used to assess the influence of antiaging interventions on the aging process. The expression levels from all of the age-regulated genes can be compiled into one score revealing the relative age of a tissue sample. One can view the age of a tissue at the molecular level using such a transcriptional profile of aging, similar to viewing the cellular age of tissues with a microscope or the age of individuals with a camera. We used the aging expression phenotype to discern whether aging has a great or small effect on specific tissues, to categorize different tissues into one of three distinct aging groups, and to compare molecular phenotypes of aging among different tissues and among different species.
Biomarkers of aging have been sought for some time for their potential utility in aging research [
In this study, we identify gene expression profiles for nine different tissues in mice. Our results show a substantial heterogeneity in the amount of age regulation in diverse tissues. Nine tissues show some level of age-related expression changes, ranging from 17 to 346 age-regulated genes. Since the great majority of physiological changes in a tissue will also result in changes in expression (e.g., oxidative damage induces the oxidative damage stress response), we expect there to be a high correlation between those tissues that age the most and those that show the greatest changes in gene expression. Tissues with a large number of age-regulated genes (such as the thymus, eye, lung, and spinal cord) may be exceptionally prone to age-related changes, and thus may contribute a disproportionate amount to aging in the mouse overall. Age-related expression changes may be due either to changes in gene expression or to changes in cell heterogeneity within a tissue. For example, the thymus is known to involute in old age, as thymocytes are replaced by fat cells and the overall size of the thymus is dramatically reduced. The expression profile for aging in the thymus contains genes with high expression in fat cells that show increased abundance in the thymus in old age, and genes with high expression in thymocytes that show decreased abundance. Both changes in cell-intrinsic gene expression as well as changes in heterogeneous cell populations could functionally impact aging tissues, and we do not distinguish between the two in this work. Seven tissues showed little or no significant changes in expression with respect to age, despite having similar levels of technical variability as the rest of the tissues. The lack of transcriptional change with age indicates that these tissues may be exceptionally resilient to the aging process. However, aging may affect these tissues in ways that would not be detected by the gene arrays, such as changes in protein levels or amounts of oxidative damage.
Using the aging transcriptional profiles, we were able to cluster the tissues into three groups corresponding to three distinct modes of aging: neural, vascular, and steroid responsive. The formation of distinct aging clusters suggests that different tissues age via different pathways. The three distinct aging clusters might be affected to different extents in different mice, suggesting that individual mice may be composed of a mosaic of tissues with different physiological ages. However, these three age-related modes are defined by changes in gene expression and not by physiological measurements, and it will be important to determine whether these age-related transcriptional patterns accurately depict functional decline.
Age regulation of 14 gene sets distinguishes the different aging patterns of the three tissue groups. The strongest signature differentiating the three tissue groups involves genes that encode ribosomal subunits. Genes in this pathway decrease overall expression with age in the steroid-responsive and vascular tissues, but increase expression with age in neural tissues. This finding suggests a strong role for protein synthesis and ribosomal regulation in determining how different tissues age. Although the steroid-responsive and vascular tissues showed fairly similar patterns of age regulation, a chief difference between the two tissue groups was in the age-dependent regulation of oxidoreductase genes.
We have shown that many genes and several genetic pathways are age regulated in a similar fashion between human and mouse, despite the large disparity in lifespan between the two species. We have previously shown that the mitochondrial electron transport chain gene set shows an overall decreasing trend in expression with age in humans, mice, and flies [
However, we found no global, overall correlation between age-related expression changes in mice with those from humans; we found similarity only in a few isolated, specific gene sets. Thus, an aging pathway that showed increased expression with age in mice is as likely to decrease with age in humans as it is to increase. Similarly, for pathways that decrease expression in mice. We performed this analysis by comparing the overall change in expression averaged from all nine mouse tissues and all three human tissues. We repeated the analysis by comparing age-related expression changes between mice and humans in single tissues (kidney and muscle). We found that none of these two tissues showed a significant correlation in age-related changes in expression between mice and humans (
The results from this paper indicate that age-related changes in a gene or pathway in mice do not reliably predict age-related changes in humans. Future experiments using a larger sample size, improved technology, or different analytical techniques may show an overall correlation between aging in mice and humans. Unless an overall correlation can be found, in order to understand human aging, it will be necessary to perform human experiments to find out which aspects of aging are shared between mice and humans. Once identified, focused experimental attention on these public age-related pathways in mice has the greatest potential to reveal mechanisms relevant to human aging.
Male and female C57BL/6 mice were obtained from the National Institute on Aging (NIA) colony at 3 wk and 5, 15, and 23 mo of age. Mice were euthanized by cervical dislocation at 1, 6, 16, and 24 mo of age. Each mouse was dissected in a span of time ranging from 10–12 min. After cervical dislocation, mice were decapitated and an incision was made in the ventral side. Following the incision, the skin was degloved and harvested, exposing the abdominal tissues, which were harvested in order of pancreas, gonads, and liver. Following abdominal dissection, thoracic tissues were harvested in order of heart, lung, thymus, and kidney. Simultaneously with abdominal and thoracic dissection, the head was dissected and brain tissues (cerebellum, cortex, hippocampus, striatum, and spinal cord) collected. Finally, skeletal muscle and bone marrow were harvested last. All tissues were collected in RNAlater (Ambion) on ice and stored at −20 °C. The animal study protocol was reviewed and approved by the gerontology research center's animal care and use committee.
All tissue samples were homogenized using a minibeadbeater-8 with 1.0mm zirconia/silica beads (Biospec Products). Total RNA was isolated from each sample homogenate using a RNeasy Mini Kit (Qiagen) in accordance with the manufacturer's protocol. RNA quality was assessed using an Agilent 2100 Bioanalyzer.
All hybridizations were to NIA mouse 17K cDNA arrays. These arrays are composed of cDNA spots taken from both the NIA mouse 15K cDNA and NIA mouse 7.5K cDNA clone sets [
We calculated log2(expression level), and then used the
We used a multiple regression model to measure changes in expression with age for 8,932 genes in each of 16 individual tissues. This model assumed that differences in levels of gene expression may exist because of age or sex:
We wanted to determine whether the observed trends in expression of these genes were driven by developmental changes between 1-mo-old mice and older mice. We excluded the 1-mo-old cohort from the data, and then recalculated the slopes with respect to age for all of the genes that were originally found to be age related in
At
We used a modified version of gene set enrichment analysis [
For gene
Fisher's method is very powerful, but we needed to modify it. The statistic
The statistic
There were 346 genes with
The gene list is potentially too small because the factor of 2 may be too conservative and potentially too large because of dependencies among tissues. We validated the list empirically using the 8,932 observed genes, through a permutation analysis. In each resampling, the ages of the 20 male mice were randomly permuted, as were those of the 20 female mice. Male and female mouse ages were permuted independently. The newly generated ages were used for all tissues from the same animal. Among 1,000 such permutations, only three of them had more than the 314 significant genes found in the data.
Fisher's meta-analysis is perhaps the best-known method for combining multiple test statistics on a single hypothesis, but other approaches are available. Birnbaum [
For each age-regulated tissue, we first identified the 50 genes most increasing expression with age (as determined by slope with age) as well as the 50 genes with the greatest decrease in expression with age. For each tissue and mouse pair, we then calculated the expression:
We considered the possibility that the results could reflect a technical artifact in which the RNA from all of the tissues of an individual mouse was degraded, due to mishandling, disease, or some other factor. If so, then all of the transcripts from that mouse would show weak correlations with the other mice, not just the age-regulated genes. In order to address this possibility, we calculated the Pearson coefficients between individual mice within each tissue for all 8,932 genes. We did not observe any mice exhibiting an unusually low Pearson correlation with other mice (unpublished data), showing that the tissue coordination effect is seen for age-regulated genes.
(A) Comparison of aging coefficients for 106 genes found to be age regulated in human kidney in both human and mouse. The
(642 KB TIF)
(7.9 MB XLS)
(361 KB XLS)
(50 KB XLS)
(27 KB XLS)
(197 KB XLS)
(40 KB XLS)
(258 KB XLS)
(38 KB XLS)
(18 KB XLS)
(19 KB XLS)
We gratefully acknowledge Dan L. Longo for his contributions to this manuscript, for originating the notion for the study, providing resources for collecting the data, and assisting us in interpreting the results. We are indebted to the members of Stuart K. Kim's laboratory and the Research Resources Branch of the National Institute on Aging for insight and discussion in the course of generating this work. SKK and KCB contributed equally to this study.
Atlas of Gene Expression in Mouse Aging Project