A whole lifespan mouse multi-tissue DNA methylation clock

  1. Margarita V Meer
  2. Dmitriy I Podolskiy
  3. Alexander Tyshkovskiy
  4. Vadim N Gladyshev  Is a corresponding author
  1. Brigham and Women’s Hospital and Harvard Medical School, United States
  2. Skolkovo Institute of Science and Technology, Russia
5 figures, 1 table and 5 additional files


Figure 1 with 4 supplements
Construction of the WLMT clock.

(A) Datasets used for the clock construction and initial validation. Radii of pie charts correspond to the number of samples in a group. The range of ages is shown above (in months). Performance of the whole-lifespan multi-tissue clock on training (B) and test (C) sets. MAE on training and test sets was 28.6 and 72.7 days, respectively. Tissues used are indicated on the right according to the color scheme.

Figure 1—figure supplement 1
Density plot for the number of CpG sites covered in all samples.

(A) and in samples with covered 2÷7106 sites, which were used to train the WLMT clock (B).

Figure 1—figure supplement 2
Principle component analysis of samples sequenced in the study.

Library ‘M20’ was not used in the preparation of the clock.

Figure 1—figure supplement 3
Performance of the WLMT clock on male (blue) and female (pink) samples.
Figure 1—figure supplement 4
Number of reads obtained for the sequenced samples.
Figure 2 with 6 supplements
Overlap in CpG sites among mouse DNAm clocks.

(A) overlap in CpG sites. (B) overlap in genes containing the clock sites. * - significant with p<104(two-tailed Chi-square test with Yates correction).

Figure 2—figure supplement 1
Distribution of WLMT clock CpG sites along the mouse genome.

Sites with positive weights (increasing methylation with age) are shown in green, and those with negative weights (decreasing methylation with age) in red.

Figure 2—figure supplement 2
Location of CpG sites forming the four epigenetic clocks in the mouse genome: WLMT sites are shown in blue, YOMT sites in green, blood clock sites in red and liver clock sites in orange.
Figure 2—figure supplement 3
Frequency of the occurrence of WLMT clock sites in 100 clocks generated based on the robustness study.

100 clocks are trained on different subsets of the set of samples used in WLMT construction. (A) Density plot showing the overall distribution of frequencies of WLMT clock sites as found in the generated 100 clocks. Carpet plot shows the exact values of frequencies. (B) Relation of the absolute values of weight of sites in the WLMT clock and frequency of these sites in the 100 alternative clocks.

Figure 2—figure supplement 4
Age-related changes in DNA methylation of the ten WLMT clock sites with highest positive weights.

Tissues used in construction of the clock are shown.

Figure 2—figure supplement 5
Age-related changes in DNA methylation for the ten WLMT clock sites with lowest negative weights.
Figure 2—figure supplement 6
Density plots showing distribution of weights of CpG clock sites in the four mouse clocks.
Figure 3 with 2 supplements
Absolute errors of four mouse DNAm clocks applied to untreated wild-type samples.

(A) In this case, WLMT clock was applied to its test set, whereas for the other age predictors this set also included their training set samples. (B) Blood samples which were not used in any of the clocks. (C) Only liver samples. (D) Samples of all tissues, excluding those used in the YOMT clock. (E) Samples of all non-blood tissues. (F) All tissues, excluding liver.

Figure 3—figure supplement 1
Performance of the WLMT, blood, YOMT and liver clocks on the liver samples not included into the WLMT and liver training sets.

∆Age shows difference between DNAm age and chronological age. Chronological age of the included samples is 180 days. MAE is 244 days for the WLMT and 65 days for the liver clock.

Figure 3—figure supplement 2
Performance of the WLMT, blood, YOMT and liver clocks.

WLMT: only test samples are included. YOMT: test and training sets are shown. Blood samples have not been used in development of this clock. Blood clock: all blood samples belong to the training set, while all other tissues to the test set. Liver clock: only test samples are shown.

DNAm ages estimated for an independent dataset.

An additional dataset (Reizel et al., 2018) was tested with the four clocks. Solid line represents a perfect correspondence between DNAm age and chronological age. Dashed lines show slopes of linear regressions made for each DNAm clock based on the age values estimated with it. Slopes: 0.14 (YOMT), 0.16 (liver), 1.24 (WLMT), 0.45 (blood). p-values: p=1.41039 (YOMT), p=2.81029 (liver), p=0.12 (WLMT), p=6.4104 (blood).

Difference in DNAm age estimated by four age predictors and chronological age based on WLMT, blood, YOMT and liver clocks.

Untreated and wild-type samples are shown in green, and intervention samples in orange. (A) Effect of caloric restriction in mouse strain C57BL/6, DNA from blood. (B) Effect of caloric restriction, mouse strain B6D2F1, DNA from blood. (C) Conversion of fibroblasts to iPSCs, DNA from cultured cells. (D) Growth hormone receptor knockout, DNA from blood. (E) Snell dwarf mice, DNA from blood. (F) Ames dwarf mice, DNA from liver. ΔAge was defined as the difference between the DNAm age and the chronological age. *- p<5102, **- p<1102, ***- p<103.



Table 1
Summary of available mouse DNAm age predictors.
precision, R2
Age range of
training set,
Number of CpG
sites in the model
Blood>0.90083.–3590(Petkovich et al., 2017)
Liver0.910.2–26148*(Wang et al., 2017)
Multi-tissue0.70.2–9.5329(Stubbs et al., 2017)
Multi-tissue0.890.2–35435This paper
  1. *Unlike other DNAm clocks, the liver clock is based on a combination of methylation levels on both strands and thus it corresponds to 296 genomic positions.

Additional files

Supplementary file 1

Sample information.

Supplementary file 2

Information on samples sequenced by RRBS in the study and included in the analysis.

Supplementary file 3

Mouse DNAm clocks.

Supplementary file 4

Genes overlapping in pairs of clocks.

Supplementary file 5

Distribution of samples across libraries.


Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Margarita V Meer
  2. Dmitriy I Podolskiy
  3. Alexander Tyshkovskiy
  4. Vadim N Gladyshev
A whole lifespan mouse multi-tissue DNA methylation clock
eLife 7:e40675.