An algorithm for calculation of inflammation profiles based on microarray data was used to establish a gene expression-based mapping of inflammation events associated with aging in human skin and other tissue types (see Methods section and Figure 2 from Swindell et al. [31]). Colors reflect age-associated gene expression patterns among “signature transcripts” that exhibit elevated expression in different cell populations (listed in left margin). Dark red colors indicate cases in which signature transcripts for a given cell population increase with age (i.e., the ratio of age-increased to age-decreased signature transcripts is greater than one). Filled black circles indicate significant bias towards age-increased expression among signature transcripts (FDR-adjusted P<0.05; Fisher's exact test), suggesting increased abundance or infiltration of that cell population with aging. The opposite pattern, in which signature transcripts are disproportionately decreased with age, is indicated by green squares within the chart (indicative of decreased cell type abundance with age). For most cell population categories (rows), data from multiple cell populations was available, which permitted calculation of several signature transcript ratios (r₁, r₂…r_n). In these cases, the maximum ratio is displayed in the chart if the median value of r₁, r₂… r_n is greater than 0.50 (since, in this case, most evidence suggests bias towards age-increased expression among the n replicate sets of signature transcripts). The minimum ratio is displayed if the median value of r₁, r₂… r_n is less than 0.50 (since, in this case, most evidence suggests bias towards age-decreased expression among the n replicate sets of signature transcripts).

We identified 500 probe sets (from the Affymetrix Human Genome U133 Plus 2.0 array) representing signature transcripts that exhibit high expression in CD4+ T-cells isolated from peripheral blood (based on data obtained under GEO series accession GSE14278). Scatterplots (A) and (C) display expression of these transcripts in skin samples from young and old human subjects. Red symbols represent probe sets with higher expression in old subjects, and green symbols represent probe sets with higher expression in young subjects. The relative proportion of these two probe set groups is indicated by the pie chart shown in each figure. In (A) (females), young subjects were between 18 and 22 years of age (n = 5) and old subjects were between 57 and 75 years of age (n = 5). In (C) (males), young subjects were between 18 and 27 years of age (n = 5) and old subjects were between 57 and 60 years of age (n = 5). Among all T-cell-associated probe sets, the average increase in old subjects was 11% in females, but in males, T-cell-associated probe sets decreased by 5.3% on average (P<0.001; two-sample t-test). Across all probe sets, moreover, fold-change differences between old and young mice were negatively correlated between the sexes (r = −0.13, P<0.001). In parts (B) and (D), subjects were divided into different age groups (n = 2–3 subjects per group). 40-year fold-change estimates (old/young) were then averaged across subjects belonging to the same age group for each of the 500 probe sets associated with the CD4+ T-cell signature. Grey boxes denote the median and inter-quartile range for the 500 fold-change estimates in each age group. The red background region denotes an average fold-change larger than one (i.e., signature transcripts are age-increased on average), while the green background region denotes an average fold-change less than one (i.e., signature transcripts are age-decreased on average).

We generated gene expression profiles for aging in human and mouse skin and compared these with 25 aging profiles from human tissues and 28 aging profiles from mouse tissues. All human profiles (black labels) were generated based on data from the Affymetrix Human Genome U133 Plus 2.0 array platform and all mouse profiles (red labels) were generated based on data from the Affymetrix Mouse Genome 430 2.0 array platform (Tables S1 and S2). The cluster analysis is based upon a total of 3785 human-mouse transcript pairs, which collectively, were associated with 3646 unique human genes and 3580 unique mouse genes. All human-mouse transcript pairs were significantly altered by aging with respect to at least three separate human profiles or at least three separate mouse profiles (FDR-adjusted P<0.05). Heat map colors denote an adjusted fold-change ratio (old/young) in units of standard deviations to permit visual and quantitative comparison across tissues and species. In humans, adjusted ratios were obtained by calculating the estimated 40-year fold change (old/young) for each transcript and dividing this ratio by the standard deviation of all 3785 ratios calculated in a given profile (i.e., each row of the heatmap). Similarly, in mice, adjusted ratios were obtained by calculating the 2-year fold change (old/young) for each transcript and dividing this ratio by the standard deviation of all 3785 ratios calculated in a given profile (i.e., each row of the heatmap). Further details on the experimental datasets used to generate human and mouse aging profiles is provided in the Methods section, Table S1 and Table S2.

The expression fold-change (old/young) over 40 years (human) or 2 years (mouse) was calculated for 15,203 transcripts associated with orthologous human and mouse genes. The scatterplot shows the human-mouse association between fold change estimates in (A) female human subjects versus female mice and (B) male human subjects versus male mice. The dashed blue line represents the least squares regression estimate, and the estimated Pearson correlation (r) is shown in the lower right quadrant (blue font). The yellow region in (A) and (B) outlines the central 50% of data points closest to the bivariate median (based upon Mahalanobis distance). The grey region outlines the central 80% of data points closest to the bivariate median (based upon Mahalanobis distance). A bivariate association, if present, would be indicated by an oblong (non-circular) shape of the yellow or grey region. In (A), the estimated correlation coefficient is significantly negative (r = −0.096; P<0.001). In (B), the estimated correlation coefficient is non-significant (r = −0.002; P = 0.77). Absence of significant correspondence between human and mouse aging effects was further supported based upon three other statistical methods for comparing gene expression profiles (see Figures S7, S8 and S9).

Gene expression profiles associated with aging in mouse tail skin (first two columns in chart) were aligned with 28 other aging profiles from a panel of mouse tissues. All profiles were generated from datasets that included young and old mice, where the same Affymetrix Mouse Genome 430 2.0 array platform was used to generate each dataset (see Methods and Table S2 for further detail). The chart lists in rank-order the top 50 genes increased by aging across all profiles (based upon the total number of significant (age-increased) results among all profiles for a given gene). Colors within the chart denote the estimated 2-year fold change (old/young) for a given gene (row) and profile (column) (see color scale), while up and down-triangles denote significant aging effects (comparison-wise P<0.05). Several alternative versions of this chart with different gene-ranking schemes are shown in Figure S10.

RT-PCR was used to evaluate the expression of Clec7a in mice from three age groups (5 months, 17 months and 30 months) and both sexes (n = 5–6 mice for each sex/age combination). The expression of 18S ribosomal RNA (Rn18s) was used as an internal control gene for calculation of relative expression in each sample. Plots (A)–(F) show mean expression levels in each group (± one standard error), which have been normalized to the average expression level of the young (5 month) female group. In each plot, and for each sex, the average expression of middle-aged (17 month) and old groups (30 months) was compared to that of the young group (5 months). A blue star is used to indicate a significant difference based upon a two-sample non-parametric statistical test (P<0.05; either Wilcoxon or Kruskal-Wallis rank sum test).

An algorithm for calculation of inflammation profiles based on microarray data was used to establish a gene expression-based mapping of inflammation events associated with aging in mouse tail skin (CB6F1 strain) and other tissue types [31] (see Methods). This figure can be interpreted as described in the Figure 1 legend and is based upon the influence of aging on the expression of “signature transcripts” associated with different cell populations (listed in left margin). Red colors in the chart denote evidence for an age-related expansion or infiltration of the cell population type listed in each row (sees color scale and Figure 1 legend). The opposite trend is denoted by green colors (see color scale and Figure 1 legend).

The chart lists the top-scoring motifs associated with genes (A) increased by aging of tail skin in female mice, (B) decreased by aging of tail skin in female mice, (C) increased by aging of tail skin in males, and (D) decreased by aging of tail skin in males. For each motif (left margin), gene sets associated with a varying number of motif occurrences (in the region 2000 BP upstream/200 BP downstream of the transcription start site) were derived. For any one motif, gene sets with fewer motif occurrences were always larger than those sets containing genes with more motif occurrences (see legend in right margin). Thus, larger symbols correspond to larger gene sets with few motif occurrences while smaller symbols correspond to smaller gene sets with a larger number of motif occurrences (see legend). For each set, the proportion of age-increased to age-decreased genes was calculated (open symbols and horizontal axis; see Figure S3 and S4 legends). A motif-expression association is indicated by deviation of this proportion from 0.50, with blue symbols corresponding to gene sets for which the estimated proportion differs significantly from 0.50 (comparison-wise P<0.05; Fisher's exact test), and yellow symbols corresponding to sets significant at a more stringent threshold (FDR-adjusted P<0.05; Bejamini-Hochberg correction). We noted two motif-aging associations that, for the same sex, were consistent in direction with those identified in humans (blue font; compare with Figures S3 and S4). However, we noted nine associations that, for the same sex, were contrasting in direction compared to those identified in humans (magenta font; compare with Figures S3 and S4). For each motif listed, candidate target genes are listed in the right margin (red labels for age-increased genes; green labels for age-decreased genes; see Figure S3 and S4 legends).

Plots (A) and (B) display the expression of human clusterin (CLU) relative to age, where each open circle indicates the microarray-based expression estimate in sun-protected skin from an individual female or male. Red lines correspond to the least squares estimate for the regression of expression versus subject age, and blue lines represent the same regression but obtained using the Loess procedure. Panel (C) displays mean microarray-based expression levels of mouse Clu (±1 standard error) in tail skin from young (5 month) and old (30 months) CB6F1 mice (n = 5 per group). Panels (D)–(G) display average expression of mouse Clu in mice from three age groups (5 months, 17 months and 30 months) and from both sexes (n = 5–6 mice for each sex/age combination; RT-PCR with 18S ribosomal RNA used as an internal control gene for calculation of relative expression). Each panel (D)–(G) displays the mean expression levels in each group (± one standard error) normalized to the average expression level of the young (5 month) female group. For each sex, the average expression of middle-aged (17 month) and old groups (30 months) was compared to that of the young group (5 months). A red star indicates a significant difference relative to the youngest group of the same sex (P<0.05; two-sided two-sample t-test). A blue star is used to indicate a significant difference based upon a two-sample non-parametric statistical test (P<0.05; either Wilcoxon or Kruskal-Wallis rank sum test). Panels (H) and (I) display average expression of mouse clusterin (Clu) in long-lived Pit1(dw/dw) dwarf mice as compared to littermate controls with normal phenotypes (n = 6 per genotype; males; six months of age; Actb was used as an internal reference for calculation of relative expression).