https://orcid.org/0000-0002-7280-4310
Figures & data
Figure 1. Primary human fibroblasts undergo global DNA methylation changes across the cellular lifespan.
(a) Overview of study design, which includes 12 timepoints for a longitudinal assessment of DNAm across the cellular lifespan, and comparison with longitudinal data from human blood [Citation6]. (b) Cumulative population doublings for cultured primary dermal human fibroblasts. (c) Fibroblasts across different stages of the lifespan imaged by phase contrast microscopy after 5 days of growth at each respective passage. (d) Principal component analysis (PCA) of all CpG sites across the cellular lifespan; the first two components are plotted and each timepoint is indicated by the number of days in culture. (e) Correlation between the first and (f) second principal components and chronological age. (g) Frequency distribution of all CpG sites ordered by their rate of methylation change per month across the lifespan. Dotted lines indicate an arbitrary methylation rate threshold of 5%. Inset: Proportion of sites >5% per month that undergo age-related decrease (hypomethylation) or increase (hypermethylation) in DNAm.
![Figure 1. Primary human fibroblasts undergo global DNA methylation changes across the cellular lifespan.(a) Overview of study design, which includes 12 timepoints for a longitudinal assessment of DNAm across the cellular lifespan, and comparison with longitudinal data from human blood [Citation6]. (b) Cumulative population doublings for cultured primary dermal human fibroblasts. (c) Fibroblasts across different stages of the lifespan imaged by phase contrast microscopy after 5 days of growth at each respective passage. (d) Principal component analysis (PCA) of all CpG sites across the cellular lifespan; the first two components are plotted and each timepoint is indicated by the number of days in culture. (e) Correlation between the first and (f) second principal components and chronological age. (g) Frequency distribution of all CpG sites ordered by their rate of methylation change per month across the lifespan. Dotted lines indicate an arbitrary methylation rate threshold of 5%. Inset: Proportion of sites >5% per month that undergo age-related decrease (hypomethylation) or increase (hypermethylation) in DNAm.](/cms/asset/5141052b-d426-4d42-bb6e-2ea465bc27c5/kepi_a_1626651_f0001_c.jpg)
Figure 2. DNAm clocks track aging across the cellular lifespan, are sensitive to glucose levels, and reveal accelerated aging in vitro.
(a) Linear regression of chronological age (e.g. days in culture) and predicted DNAmAge in primary human fibroblasts using the Pan tissue [Citation10] and (b) Skin & Blood [Citation21] clocks. Cells were cultured either in normal glucose (n = 12 timepoints, 5.5mM) and high ‘diabetic’ levels of glucose (n = 14 timepoints, 25mM). The dotted line indicates the estimated point at which division rate substantially decreases (i.e. replicative senescence). Note that the Skin & Blood clock remains linear throughout the lifespan. (c-d) Box plots of the residuals from regressions in A and B, comparing normal and high glucose cells across lifespan. Each datapoint reflects the residual score for each timepoint assesses; non-parametric unpaired Mann-Whitney test. (e) The rate of epigenetic aging measured as the slope of DNAmAge and chronological time in years over the linear portion of the regressions in A and B. For the regression of the Pan Tissue Clock, only timepoints in early- and middle-life are used. (f) Regression of MiAge estimated cell divisions to actual population doublings calculated from cell counts performed at each passage. Divisions are shown relative to youngest 35 day timepoint.
![Figure 2. DNAm clocks track aging across the cellular lifespan, are sensitive to glucose levels, and reveal accelerated aging in vitro.(a) Linear regression of chronological age (e.g. days in culture) and predicted DNAmAge in primary human fibroblasts using the Pan tissue [Citation10] and (b) Skin & Blood [Citation21] clocks. Cells were cultured either in normal glucose (n = 12 timepoints, 5.5mM) and high ‘diabetic’ levels of glucose (n = 14 timepoints, 25mM). The dotted line indicates the estimated point at which division rate substantially decreases (i.e. replicative senescence). Note that the Skin & Blood clock remains linear throughout the lifespan. (c-d) Box plots of the residuals from regressions in A and B, comparing normal and high glucose cells across lifespan. Each datapoint reflects the residual score for each timepoint assesses; non-parametric unpaired Mann-Whitney test. (e) The rate of epigenetic aging measured as the slope of DNAmAge and chronological time in years over the linear portion of the regressions in A and B. For the regression of the Pan Tissue Clock, only timepoints in early- and middle-life are used. (f) Regression of MiAge estimated cell divisions to actual population doublings calculated from cell counts performed at each passage. Divisions are shown relative to youngest 35 day timepoint.](/cms/asset/f7ed8baa-9c1e-45f8-b9e8-323aca559c1f/kepi_a_1626651_f0002_c.jpg)
Figure 3. ELOVL2 DNAm topology and age-related hypermethylation and expression are conserved between human blood and cultured fibroblasts.
(a) Overview of the ELOVL2 gene with 8 exons, 7 introns, and 1 CpG Island (green box) located in the promoter region. All EPIC array CpG sites are mapped as vertical light-blue lines. CpGs with >75% (red) or <25% (blue) methylation levels are color coded, and CpGs exhibiting significant DNAm changes in human blood across the cellular lifespan are indicated by an arrow with their % change in DNAm level. (b) DNAm topology graph of ELOVL2 CpG sites within the promoter, first exon, and gene body present in both the EPIC (in vitro, fibroblasts) and 450k (in vivo, blood) arrays. Methylation levels from whole blood [top,, Citation6] and cultured human fibroblasts (bottom) are juxtaposed, highlighting their similar topology. In both datasets, each line represents a different individual (in vivo) or timepoint of the same individual’s cells (in vitro), color-coded by age. (c) Sigmoidal fit line of the DNAm levels for cg1686657 in vivo (top) and in vitro (bottom) across the lifespan. (d) ELOVL2 transcript levels quantified by RNA sequencing across the early- and mid-life portions of the cellular lifespan (linear regression, n = 3 timepoints, r2 = 0.98, p value = 0.012). (e) Analysis of topological similarity, quantified as the correlation across all CpGs that map to a given gene between in vivo and in vitro systems. Each datapoint is the average methylation values across all ages/passages, plotted for each of the top 29 age-associated genes reported in Wang et al. [Citation6]. Significant correlations (p < 0.05) are shown as thick regression lines, with 95% confidence interval (shaded area). Inset: proportion of in vivo – in vitro correlations that are significant correlations (p < 0.01). (f) Same as E but for 46 single CpGs whose methylation levels are positively correlated (p < 0.05) with age [Citation6]. Each graph is for a single CpG and each datapoint (n = 12) reflects time in culture (x axis) and corresponding human ages (y axis). Full size correlation graphs can be found in Supplemental Figure S4.
![Figure 3. ELOVL2 DNAm topology and age-related hypermethylation and expression are conserved between human blood and cultured fibroblasts.(a) Overview of the ELOVL2 gene with 8 exons, 7 introns, and 1 CpG Island (green box) located in the promoter region. All EPIC array CpG sites are mapped as vertical light-blue lines. CpGs with >75% (red) or <25% (blue) methylation levels are color coded, and CpGs exhibiting significant DNAm changes in human blood across the cellular lifespan are indicated by an arrow with their % change in DNAm level. (b) DNAm topology graph of ELOVL2 CpG sites within the promoter, first exon, and gene body present in both the EPIC (in vitro, fibroblasts) and 450k (in vivo, blood) arrays. Methylation levels from whole blood [top,, Citation6] and cultured human fibroblasts (bottom) are juxtaposed, highlighting their similar topology. In both datasets, each line represents a different individual (in vivo) or timepoint of the same individual’s cells (in vitro), color-coded by age. (c) Sigmoidal fit line of the DNAm levels for cg1686657 in vivo (top) and in vitro (bottom) across the lifespan. (d) ELOVL2 transcript levels quantified by RNA sequencing across the early- and mid-life portions of the cellular lifespan (linear regression, n = 3 timepoints, r2 = 0.98, p value = 0.012). (e) Analysis of topological similarity, quantified as the correlation across all CpGs that map to a given gene between in vivo and in vitro systems. Each datapoint is the average methylation values across all ages/passages, plotted for each of the top 29 age-associated genes reported in Wang et al. [Citation6]. Significant correlations (p < 0.05) are shown as thick regression lines, with 95% confidence interval (shaded area). Inset: proportion of in vivo – in vitro correlations that are significant correlations (p < 0.01). (f) Same as E but for 46 single CpGs whose methylation levels are positively correlated (p < 0.05) with age [Citation6]. Each graph is for a single CpG and each datapoint (n = 12) reflects time in culture (x axis) and corresponding human ages (y axis). Full size correlation graphs can be found in Supplemental Figure S4.](/cms/asset/1ae03a09-4e3a-483b-9dbf-bca712539274/kepi_a_1626651_f0003_c.jpg)
Figure 4. Lifespan trajectories of single CpGs reveal rapid linear and non-linear age-related changes in DNA methylation.
(a) Heatmap of the top 1,000 age-related CpGs with the lowest P values from the generalized additive models (GAM) analysis across the lifespan (see methods for details). Hierarchical clustering using complete linkage and Euclidean distance [sqrt(sum((xi – yi)^2)))]. (b) P values for all EPIC CpG sites arranged by degree of freedom (DoF). Coloring scheme indicates datapoint density (log scale). (c) Proportion of the top 1,000 age-related CpGs that decrease (Hypo) or increase (Hyper) in DNAm levels with age, organized by DoF. (d) Example of lifespan trajectories (fitted models) using the top 20 most significant CpG sites undergoing hypermethylation (top) and hypomethylation (bottom), for DoF 1–4. Bolded lines represent the average of all sites with similar trajectories. (e) Distribution of the top 1,000 sites by gene regions (f), relative to CpG islands, and (g) promoter region by DoF categories. The label ‘All sites’ corresponds to all sites included in the EPIC array and is used as reference to evaluate the enrichment in specific genomic locations.
![Figure 4. Lifespan trajectories of single CpGs reveal rapid linear and non-linear age-related changes in DNA methylation.(a) Heatmap of the top 1,000 age-related CpGs with the lowest P values from the generalized additive models (GAM) analysis across the lifespan (see methods for details). Hierarchical clustering using complete linkage and Euclidean distance [sqrt(sum((xi – yi)^2)))]. (b) P values for all EPIC CpG sites arranged by degree of freedom (DoF). Coloring scheme indicates datapoint density (log scale). (c) Proportion of the top 1,000 age-related CpGs that decrease (Hypo) or increase (Hyper) in DNAm levels with age, organized by DoF. (d) Example of lifespan trajectories (fitted models) using the top 20 most significant CpG sites undergoing hypermethylation (top) and hypomethylation (bottom), for DoF 1–4. Bolded lines represent the average of all sites with similar trajectories. (e) Distribution of the top 1,000 sites by gene regions (f), relative to CpG islands, and (g) promoter region by DoF categories. The label ‘All sites’ corresponds to all sites included in the EPIC array and is used as reference to evaluate the enrichment in specific genomic locations.](/cms/asset/4908b9a1-243c-4eee-8963-1723a0b2668d/kepi_a_1626651_f0004_c.jpg)
