Skip to main content
PMC home page
eLife logoLink to eLife
. 2018 Aug 24;7:e37462. doi: 10.7554/eLife.37462

Epigenetic age-predictor for mice based on three CpG sites

Yang Han 1,2, Monika Eipel 1,2, Julia Franzen 1,2, Vadim Sakk 3, Bertien Dethmers-Ausema 4, Laura Yndriago 5, Ander Izeta 5,6, Gerald de Haan 4, Hartmut Geiger 3,7, Wolfgang Wagner 1,2,✉
Editors: Vadim N Gladyshev8, Jessica K Tyler9
PMCID: PMC6156076  PMID: 30142075

Abstract

Epigenetic clocks for mice were generated based on deep-sequencing analysis of the methylome. Here, we demonstrate that site-specific analysis of DNA methylation levels by pyrosequencing at only three CG dinucleotides (CpGs) in the genes Prima1, Hsf4, and Kcns1 facilitates precise estimation of chronological age in murine blood samples, too. DBA/2 mice revealed accelerated epigenetic aging as compared to C57BL6 mice, which is in line with their shorter life-expectancy. The three-CpG-predictor provides a simple and cost-effective biomarker to determine biological age in large intervention studies with mice.

Research organism: Mouse

eLife digest

Epigenetic marks are chemical modifications found throughout the genome – the DNA within cells. By influencing the activity of nearby genes, the marks govern developmental processes and help cells to adapt to changes in their surroundings. Some epigenetic marks can be gained or lost with age. A lot of aging research focuses on one type of mark, called “DNA methylation”. By measuring the presence or absence of specific methyl groups, scientists can estimate biological age – which may differ from calendar age.

Recent studies have developed computer models called epigenetic aging clocks to predict the biological age of mouse cells. These clocks use epigenetic data collected from the entire genomes of mice, and are useful for understanding how the aging process is affected by genetic parameters, diet, or other environmental factors. Yet, the genome sequencing methods used to construct most existing epigenetic clocks are expensive, labor-intensive, and cannot be easily applied to large groups of mice.

Han et al. have developed a new way to predict biological aging in mice that needs methylation information from just three particular sections of the genome. Even though this approach is much faster and less expensive than other epigenetic approaches to measuring aging, it has a similar level of accuracy to existing models. Han et al. use the new method to show that cells from different strains of laboratory mice age at different rates. Furthermore, in a strain that has a shorter life expectancy, aging seems to be accelerated.

The new approach developed by Han et al. will make it easier to study how aging in mice is affected by different interventions. Further studies will also be needed to better understand how epigenetic marks relate to biological aging.

Introduction

Age-associated DNA methylation (DNAm) was first described for humans after Illumina Bead Chip microarray data became available to enable cross comparison of thousands of CpG loci (Bocklandt et al., 2011; Koch and Wagner, 2011). Many of these age-associated CpGs were then integrated into epigenetic age-predictors (Hannum et al., 2013; Horvath, 2013; Weidner et al., 2014). However, site-specific DNAm analysis at individual CpGs can also provide robust biomarkers for aging. For example, we have described that DNAm analysis at only three CpGs enables age-predictions for human blood samples with a mean absolute deviation (MAD) from chronological age of less than five years (Weidner et al., 2014). Such simplistic age-predictors for human specimen are widely used because they enable fast and cost-effective analysis in large cohorts.

Recently, epigenetic clocks were also published for mice by using either reduced representation bisulfite sequencing (RRBS) or whole genome bisulfite sequencing (WGBS) (Petkovich et al., 2017; Stubbs et al., 2017; Wang et al., 2017). For example, Petkovich et al. described a 90 CpG model for blood (Petkovich et al., 2017), and Stubbs and coworkers a 329 CpG model for various different tissues (Stubbs et al., 2017). Nutrition and genetic background seem to affect the epigenetic age of mice – and thereby possibly aging of the organism (Cole et al., 2017; Hahn et al., 2017; Maegawa et al., 2017). In analogy, epigenetic aging of humans is associated with life expectancy, indicating that it rather reflects biological age than chronological age (Lin et al., 2016; Marioni et al., 2015). However, DNAm profiling by deep sequencing technology is technically still challenging, relatively expensive, and not every sequencing-run covers all relevant CpG sites with enough reading depth.

Results

Therefore, we established pyrosequencing assays for nine genomic regions of previously published predictors (Petkovich et al., 2017; Stubbs et al., 2017). These regions were preselected to have multiple age-associated CpGs in close vicinity. DNAm was then analyzed in 24 blood samples of female C57BL/6 mice that covered a broad range of 12 different age groups (11 to 117 weeks old). The nine amplicons covered a total of 71 CpG sites (Supplementary file 1) and we used machine learning to identify the best fitted model for epigenetic age-predictions using cross-fold validation on the training set. The best results were observed for 15 CpGs from five different amplicons that provided an extremely high correlation with chronological age in the training set (R2 = 0.99; mean absolute deviation [MAD]=2.76 weeks; Supplementary file 2), albeit the training set might be too small for this approach. To make the method more easily applicable and more cost-effective, we wanted to focus on less CpGs. When we varied the regularization parameters for models with less CpGs, the precision declined significantly. For example the best model with three CpGs comprised the three CpGs of Hsf4 (CpGs# 3,4,5) that also revealed the overall highest Pearson correlations with chronological age (R2 = 0.95; MAD = 5.24 weeks). However, combination of different hypo- and hypermethylated amplicons might be advantageous to facilitate better assessment of plausibility of the results. Therefore, we alternatively selected those three CpGs that revealed the highest Pearson correlation with chronological age in different amplicons. These three CpGs were associated with the genes Proline rich membrane anchor 1 (Prima1: chr12:103214639; R2 = 0.71), Heat shock transcription factor 4 (Hsf4: chr8:105271000; R2 = 0.95) and Potassium voltage-gated channel modifier subfamily S member 1 (Kcns1: chr2:164168110; R2 = 0.83; Figure 1A–C; Figure 1—figure supplement 1). Notably, all three CpGs were derived from the epigenetic age-predictor for blood samples (Petkovich et al., 2017). A multivariable model for age-predictions was established for DNAm at the CpGs in Prima 1 (α), Hsf4 (β), and Kcns1 (γ):

Figure 1. Three CpG epigenetic age-predictor for mice.

(a–c) DNA methylation (DNAm) of three CpGs in the genes Prima1, Hsf4 and Kcns1 was analyzed by pyrosequencing in 24 C57BL/6 mice (training set). Coefficient of determination (R2) of DNAm versus chronological age is indicated. (d) Based on these age-associated DNAm changes a multivariable model for age prediction was calculated. (e–g) Subsequently, two independent validation sets were analyzed: 21 C57BL/6 mice from the University of Ulm and 19 C57BL/6 mice from the University of Groningen (validation sets 1 and 2, respectively). (h) Age predictions with the three-CpG-model revealed a high correlation with chronological age in the independent validation sets (MAD = mean absolute deviation; MAE = median absolute error).

Figure 1.

Figure 1—figure supplement 1. Target sequences of pyrosequencing assays.

Figure 1—figure supplement 1.

Sequences for the three genomic regions are depicted and CpG sites (red) are numbered by the dispensation order. The relevant CpGs with highest age-correlation are highlighted in bold.
Open in a new tab

Predicted ageC57BL/6 (in weeks) = −58.076 + 0.25788 α + 3.06845 β + 1.00879 γ

Age-predictions correlated very well with the chronological age of C57BL/6 mice in the training set (R2 = 0.96; MAD = 4.86 weeks; Figure 1D).

Our three CpG age-predictor was subsequently validated in a blinded manner for 21 C57BL/6J mice (7 to 104 weeks old) from the University of Ulm (validation set 1) and 19 C57BL/6J mice (14 to 109 weeks old) from the University of Groningen (validation set 2). The results of both validation sets revealed high correlations with chronological age (R2 = 0.95 and 0.91, respectively; Figure 1E–H) with relatively small MADs (6.9 and 7.1 weeks) and median absolute errors (MAE; 5.0 and 5.9 weeks). Thus, our age-predictions seem to have similar precision as previously described for multi-CpG predictors based on RRBS or WGBS data (Petkovich et al., 2017; Stubbs et al., 2017; Wang et al., 2017).

Gender did not have significant impact on our epigenetic age-predictions for mice (Figure 2), as described before (Maegawa et al., 2017; Petkovich et al., 2017; Stubbs et al., 2017). In contrast, the human epigenetic clock is clearly accelerated in male donors (Hannum et al., 2013; Horvath, 2013; Weidner et al., 2014). This coincides with shorter life expectancy in men than woman, whereas in mice there are no consistent sex differences in longevity (Goodrick, 1975).

Figure 2. Gender does not affect epigenetic age predictions in mice.

Figure 2.

Open in a new tab

The deviations of predicted age by our three-CpG predictor versus chronological age did not reveal significant differences between female and male C57BL/6 mice (Mann–Whitney U test p=0.6).

To address the question if our three CpG signature was also applicable for other tissues than blood we analyzed the DNAm in skin, kidney, intestine, lung, liver, heart, brain, testis, and pancreas of 3 young (9.6 weeks old) and three old mice (56.9 weeks old). In all tissues tested the samples of old mice were predicted to be older using our three CpG signature. However, the different DNAm levels clearly demonstrate that the model needs to be retrained to be applied for these tissues (Figure 3).

Figure 3. Age-associated DNA methylation at the three CpG sites in different tissues.

Figure 3.

Open in a new tab

Different tissues were isolated of three young (9.6 weeks) and three old mice (56.9 weeks) and DNAm was analyzed at the three relevant CpGs in (a) Prima1, (b) Hsf4, and (c) Kcns1. Epigenetic age-predictions using the 3 CpG model for blood demonstrated also significant differences between young and old mice in skin, intestine, brain, and testis (mean ± standard deviation; Student t-tests: *p<0.05; **p<0.01; ***p<0.001).

Subsequently, we analyzed epigenetic aging of DBA/2 mice that have a shorter life expectancy than C57BL/6 mice (Goodrick, 1975) (33 mice from Ulm and Groningen; 6 to 109 weeks old). The three CpGs in Prima1, Hsf4 and Kcns1 revealed high correlation with chronological age (R2 = 0.91, 0.88 and 0.83, respectively), albeit the offset in DNAm between DBA/2 and C57BL/6 mice indicated that the signature needs to be retrained for different mouse strains (Figure 4a–c). Notably, the slopes were higher in DBA/2 mice, particularly for the CpG in Prima1. Furthermore, DNAm of Hsf4 increased at a higher rate in young DBA/2 mice, indicating that it is more accurately modelled as a function of logarithmic age. This has also been described in human for many age-associated CpGs in pediatric cohorts (Alisch et al., 2012). In fact, epigenetic age-predictions in DBA/2 mice seemed to follow a logarithmic model of age (R2 = 0.89; Figure 4d) rather than a linear association (R2 = 0.86). These results provided evidence for accelerated epigenetic aging of DBA/2 mice.

Figure 4. Epigenetic aging is accelerated in DBA/2 mice as compared to C57BL/6 mice.

Figure 4.

Open in a new tab

(a–c) Age-related DNA methylation (DNAm) determined by pyrosequencing assay for three candidate CpGs on 33 of DBA/2 blood samples (14 mice from the University of Ulm and 19 mice from the University of Groningen; red). For comparison we provided measurements of the C57BL/6 mice (only from validation sets; blue). (d) Epigenetic age-predictions using the three CpG multivariable model for the C57BL/6 mice (blue; linear regression) and DBA/2 mice (red, logarithmic regression). Age-predictions in DBA/2 mice rather followed a logarithmic regression (R = Pearson correlation); (e) Based on the DNAm measurements in DBA/2 we adjusted the multivariate regression model for age-predictions of this mouse strain as described in the text (DBA/2 predictor).

Either way, epigenetic age-predictions were overall significantly overestimated in the shorter-lived DBA/2 mice, suggesting that age-predictors need to be adjusted for different inbreed mice strains. To this end, we have retrained a multivariate model for DBA/2 mice:

Predicted ageDBA/2 (in weeks) = 87.54294–1.22221 α + 0.991558 β + 0.355444 γ

This adjusted model facilitated relatively precise age-predictions for DBA/2 mice (R2 = 0.95; MAD = 7.1 weeks; MAE = 5.3 weeks; Figure 4e).

Discussion

Generation of confined epigenetic signatures is always a tradeoff between integrating more CpGs for higher precision and higher costs for analysis (Wagner, 2017). It was somewhat unexpected that with only three CpGs our signature facilitated similar precision of epigenetic age-predictions as the previously published signatures based on more than 90 CpGs. This can be attributed to the higher precision of DNAm measurements at individual CpGs by bisulfite pyrosequencing, which is one of the most precise methods for determining DNAm at single CpG resolution (BLUEPRINT consortium, 2016). Particularly in RRBS data not all CpG sites are covered in all samples and a limited number of reads notoriously entails lower precision of DNAm levels at these genomic locations. Thus, genome wide deep sequencing approaches facilitate generation of robust large epigenetic age-predictors, while site specific analysis may compensate by higher precision of DNAm measurement at individual CpGs.

The ultimate goal of epigenetic age-predictors for mice is not to develop near perfect age predictors, but to provide a surrogate for biological aging that facilitates assessment of interventions on aging. In fact, using deep sequencing approaches (RRBS or WGBS) several groups already indicated that relevant parameters that affect aging of the organism - such as diet, genetic background, and drugs - do also impact on epigenetic aging (Cole et al., 2017; Hahn et al., 2017; Maegawa et al., 2017). It is yet unclear if epigenetic aging signatures can be specifically trained to either correlate with chronological age or biological age. For humans, recent studies indicate that this might be possible (Levine et al., 2018) and we have previously demonstrated that even individual age-associated CpGs can be indicative for life expectancy (Zhang et al., 2017). Further studies will be necessary to gain better understanding how epigenetic age predictions are related to the real state of biological aging, and how it is related to alternative approaches to quantify biological aging, such as telomere length (Belsky et al., 2018).

Our three CpG model has been trained for blood samples – a specimen that is commonly used in biochemical analysis and the small required volume can be taken without sacrificing the mice. However, epigenetic aging may occur at different rates in different tissues. It is difficult to address this question in humans because it is difficult to collect samples of various tissues in large aging cohorts, whereas this is feasible in mice. We demonstrate that age-associated DNAm changes occur in multiple tissues in our three CpGs albeit they were initially identified in blood (Petkovich et al., 2017). Furthermore, DNAm levels may vary between different hematopoietic subsets (Frobel et al., 2018; Houseman et al., 2014). In the future, sorted subsets should be analyzed to determine how the three CpG signature is affected by blood counts.

The results of our three CpG signature suggest that epigenetic aging is accelerated in DBA/2 mice. Notably, in elderly DBA/2 mice the epigenetic age predictions revealed higher ‘errors’ from chronological age, which might be attributed to the fact that the variation of lifespan is higher in DBA/2 than C57BL/6 mice (de Haan et al., 1998; Goodrick, 1975). It will be important to validate the association of the epigenetic age-predictions with biological age by additional correlative studies, including life expectancy in mice.

Taken together, we describe an easily applicable but quite precise approach to determine epigenetic age of mice. We believe that our assay will be instrumental to gain additional insight into mechanisms that regulate age-associated DNAm and for longevity intervention studies in mice.

Materials and methods

Mouse strains and blood collection

Blood samples of C57BL/6J mice of the training set and of the validation set one were taken at the University of Ulm by submandibular bleeding (100–200 μl) of living mice or postmortem from the vena cava. C57BL/6J samples of the validation set two were taken at the University of Groningen from the cheek. DBA/2J samples were taken at the University of Ulm (n = 14) and Groningen (n = 19). All mice were accommodated under pathogen-free conditions. Experiments were approved by the Institutional Animal Care of the Ulm University as well as by Regierungspräsidium Tübingen and by the Institutional Animal Care and Use Committee of the University of Groningen (IACUC-RUG), respectively. To analyze age-associated changes in different tissues we used three young (9.6 weeks old) and three old mice (56.9 weeks old) C57BL/6J mice (JaxMice) in accordance with relevant Spanish and European guidelines after approval by the Biodonostia Animal Care Committee. These mice were sacrificed and dissected immediately. 25 mg of tissue (10 mg in the case of spleen) or 200 µl of blood were used for DNA extraction.

Genomic DNA isolation and bisulfite conversion

Genomic DNA was isolated from 50 µl blood using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany). Kidney and liver DNA extractions were digested with Ribonuclease A (100 mg/ml, Sigma R4875). DNA concentration was quantified by Nanodrop 2000 Spectrophotometers (Thermo Scientific, Wilmington, USA). 200 ng of genomic DNA was subsequently bisulfite-converted with the EZ DNA Methylation Kit (Zymo Research, Irvine, USA).

Pyrosequencing

Bisulfite converted DNA was subjected to PCR amplification. Primers were purchased at Metabion and the sequences are provided in Supplementary file 3. 20 µg PCR product was immobilized to 5 µl Streptavidin Sepharose High Performance Bead (GE Healthcare, Piscataway, NJ, USA), and then annealed to 1 µl sequencing primer (5 μM) for 2 min at 80°C. Amplicons were sequenced on PyroMark Q96 ID System (Qiagen, Hilden, Germany) and analyzed with PyroMark Q CpG software (Qiagen).

Alternative approaches to select CpGs for multivariable models

We used a penalized regression model from the R package glmnet on the training dataset to establish a predictor of mouse age based on CpG methylation. The alpha parameter of glmnet was set to 1 (lasso regression) and the lambda parameter was chosen by cross-fold validation of the training dataset (10-fold cross validation). Alternatively, we trained our multivariable model with preselected CpGs based on location in three different amplicons, high Pearson correlation (R) of DNAm with chronological age, and combination of hyper- and hypomethylated sites.

Statistical analysis

Linear regressions, MAD and MAE were calculated with Excel. Statistical significance of the deviations between predicted and chronological age was estimated by Mann–Whitney U test or Student´s t-test as indicated.

Acknowledgements

This work was supported by the Else Kröner-Fresenius-Stiftung (2014_A193; to WW), by the German Research Foundation (DFG; WA 1706/8-1 to WW; and GRK 1789 CEMMA, GRK 2254 HEIST and SFBs 1074, 1149 and 1275 to HG), by the German Ministry of Education and Research (BMBF; 01KU1402B to WW; and SyStarR to HG), and by the NIH (R01HL134617 and R01DK104814 to HG). The Groningen samples were obtained from the Mouse Clinic for Cancer and Ageing (http://www.mccanet.nl), which is supported by a grant from the Netherlands Organization for Scientific Research (NWO). The funding bodies were not involved in study design, data analysis, or writing of the manuscript.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Wolfgang Wagner, Email: wwagner@ukaachen.de.

Vadim N Gladyshev, Brigham and Women's Hospital, Harvard Medical School, United States.

Jessica K Tyler, Weill Cornell Medicine, United States.

Funding Information

This paper was supported by the following grants:

  • Else Kröner-Fresenius-Stiftung 2014_A193 to Wolfgang Wagner.

  • Deutsche Forschungsgemeinschaft WA 1706/8-1 to Wolfgang Wagner.

  • Bundesministerium für Bildung und Forschung 01KU1402B to Wolfgang Wagner.

  • NIH Clinical Center R01HL134617 to Hartmut Geiger.

  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek to Gerald de Haan.

  • Deutsche Forschungsgemeinschaft GRK 1789 CEMMA to Hartmut Geiger.

  • Deutsche Forschungsgemeinschaft GRK 2254 HEIST to Hartmut Geiger.

  • Deutsche Forschungsgemeinschaft SFBs 1074 to Hartmut Geiger.

  • Deutsche Forschungsgemeinschaft SFBs 1149 to Hartmut Geiger.

  • Deutsche Forschungsgemeinschaft SFBs 1275 to Hartmut Geiger.

  • NIH Clinical Center R01DK104814 to Hartmut Geiger.

  • Bundesministerium für Bildung und Forschung SyStarR to Hartmut Geiger.

Additional information

Competing interests

No competing interests declared.

cofounder of Cygenia GmbH that can provide service for Epigenetic-Aging-Signatures (http://www.cygenia.com), but the method is fully described in this manuscript.

Author contributions

Formal analysis, Writing—original draft, Performed pyrosequencing.

Formal analysis, Performed pyrosequencing.

Tested alternative aging models.

Resources.

Resources.

Resources.

Resources, Supervision, Project administration.

Resources, Supervision, Project administration.

Conceptualization, Resources, Supervision, Funding acquisition, Project administration.

Conceptualization, Formal analysis, Supervision, Funding acquisition, Writing—original draft, Project administration.

Ethics

Animal experimentation: Experiments were approved by the Institutional Animal Care of the Ulm University as well as by Regierungspräsidium Tübingen and by the Institutional Animal Care and Use Committee of the University of Groningen (IACUC-RUG), respectively. To analyze age-associated changes in different tissues we used 3 young (67 days old) and 3 old (398 days old) C57BL/6J mice (JaxMice) in accordance with relevant Spanish and European guidelines after approval by the Biodonostia Animal Care Committee.

Additional files

Source data 1. Pyrosequencing raw data of mouse epigenetic aging predictor.
elife-37462-data1.xlsx (16.4KB, xlsx)
DOI: 10.7554/eLife.37462.008
Supplementary file 1. Age-associated DNAm in nine genomic regions of the training set.
elife-37462-supp1.docx (25.9KB, docx)
DOI: 10.7554/eLife.37462.009
Supplementary file 2. Multivariable model based on 15 CpGs.
elife-37462-supp2.docx (18.4KB, docx)
DOI: 10.7554/eLife.37462.010
Supplementary file 3. Primers for pyrosequencing.
elife-37462-supp3.docx (13.4KB, docx)
DOI: 10.7554/eLife.37462.011
Transparent reporting form
elife-37462-transrepform.pdf (312.8KB, pdf)
DOI: 10.7554/eLife.37462.012

Data availability

Raw data of pyrosequencing is provided as supplemental EXCEL table (Source data 1).

References

  1. Alisch RS, Barwick BG, Chopra P, Myrick LK, Satten GA, Conneely KN, Warren ST. Age-associated DNA methylation in pediatric populations. Genome Research. 2012;22:623–632. doi: 10.1101/gr.125187.111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Belsky DW, Moffitt TE, Cohen AA, Corcoran DL, Levine ME, Prinz JA, Schaefer J, Sugden K, Williams B, Poulton R, Caspi A. Eleven telomere, epigenetic clock and biomarker-composite quantifications of biological aging: do they measure the same thing? American Journal of Epidemiology. 2018;187:1220–1230. doi: 10.1093/aje/kwx346. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Blueprint consortium Quantitative comparison of DNA methylation assays for biomarker development and clinical applications. Nature Biotechnology. 2016;34:726–737. doi: 10.1038/nbt.3605. [DOI] [PubMed] [Google Scholar]
  4. Bocklandt S, Lin W, Sehl ME, Sánchez FJ, Sinsheimer JS, Horvath S, Vilain E. Epigenetic predictor of age. PLoS One. 2011;6:e14821. doi: 10.1371/journal.pone.0014821. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cole JJ, Robertson NA, Rather MI, Thomson JP, McBryan T, Sproul D, Wang T, Brock C, Clark W, Ideker T, Meehan RR, Miller RA, Brown-Borg HM, Adams PD. Diverse interventions that extend mouse lifespan suppress shared age-associated epigenetic changes at critical gene regulatory regions. Genome Biology. 2017;18:58. doi: 10.1186/s13059-017-1185-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. de Haan G, Gelman R, Watson A, Yunis E, Van Zant G. A putative gene causes variability in lifespan among genotypically identical mice. Nature Genetics. 1998;19:114–116. doi: 10.1038/465. [DOI] [PubMed] [Google Scholar]
  7. Frobel J, Božić T, Lenz M, Uciechowski P, Han Y, Herwartz R, Strathmann K, Isfort S, Panse J, Esser A, Birkhofer C, Gerstenmaier U, Kraus T, Rink L, Koschmieder S, Wagner W. Leukocyte counts based on DNA methylation at individual cytosines. Clinical Chemistry. 2018;64:566–575. doi: 10.1373/clinchem.2017.279935. [DOI] [PubMed] [Google Scholar]
  8. Goodrick CL. Life-span and the inheritance of longevity of inbred mice. Journal of Gerontology. 1975;30:257–263. doi: 10.1093/geronj/30.3.257. [DOI] [PubMed] [Google Scholar]
  9. Hahn O, Grönke S, Stubbs TM, Ficz G, Hendrich O, Krueger F, Andrews S, Zhang Q, Wakelam MJ, Beyer A, Reik W, Partridge L. Dietary restriction protects from age-associated DNA methylation and induces epigenetic reprogramming of lipid metabolism. Genome Biology. 2017;18:56. doi: 10.1186/s13059-017-1187-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Hannum G, Guinney J, Zhao L, Zhang L, Hughes G, Sadda S, Klotzle B, Bibikova M, Fan JB, Gao Y, Deconde R, Chen M, Rajapakse I, Friend S, Ideker T, Zhang K. Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular Cell. 2013;49:359–367. doi: 10.1016/j.molcel.2012.10.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14:R115. doi: 10.1186/gb-2013-14-10-r115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Houseman EA, Molitor J, Marsit CJ. Reference-free cell mixture adjustments in analysis of DNA methylation data. Bioinformatics. 2014;30:1431–1439. doi: 10.1093/bioinformatics/btu029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Koch CM, Wagner W. Epigenetic-aging-signature to determine age in different tissues. Aging. 2011;3:1018–1027. doi: 10.18632/aging.100395. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Levine ME, Lu AT, Quach A, Chen BH, Assimes TL, Bandinelli S, Hou L, Baccarelli AA, Stewart JD, Li Y, Whitsel EA, Wilson JG, Reiner AP, Aviv A, Lohman K, Liu Y, Ferrucci L, Horvath S. An epigenetic biomarker of aging for lifespan and healthspan. Aging. 2018;10:573–591. doi: 10.18632/aging.101414. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Lin Q, Weidner CI, Costa IG, Marioni RE, Ferreira MR, Deary IJ, Wagner W. DNA methylation levels at individual age-associated CpG sites can be indicative for life expectancy. Aging. 2016;8:394–401. doi: 10.18632/aging.100908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Maegawa S, Lu Y, Tahara T, Lee JT, Madzo J, Liang S, Jelinek J, Colman RJ, Issa JJ. Caloric restriction delays age-related methylation drift. Nature Communications. 2017;8:539. doi: 10.1038/s41467-017-00607-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Marioni RE, Shah S, McRae AF, Chen BH, Colicino E, Harris SE, Gibson J, Henders AK, Redmond P, Cox SR, Pattie A, Corley J, Murphy L, Martin NG, Montgomery GW, Feinberg AP, Fallin MD, Multhaup ML, Jaffe AE, Joehanes R, Schwartz J, Just AC, Lunetta KL, Murabito JM, Starr JM, Horvath S, Baccarelli AA, Levy D, Visscher PM, Wray NR, Deary IJ. DNA methylation age of blood predicts all-cause mortality in later life. Genome Biology. 2015;16:25. doi: 10.1186/s13059-015-0584-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Petkovich DA, Podolskiy DI, Lobanov AV, Lee S-G, Miller RA, Gladyshev VN. Using DNA methylation profiling to evaluate biological age and longevity interventions. Cell Metabolism. 2017;25:954–960. doi: 10.1016/j.cmet.2017.03.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Stubbs TM, Bonder MJ, Stark AK, Krueger F, von Meyenn F, Stegle O, Reik W, BI Ageing Clock Team Multi-tissue DNA methylation age predictor in mouse. Genome Biology. 2017;18:68. doi: 10.1186/s13059-017-1203-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Wagner W. Epigenetic aging clocks in mice and men. Genome Biology. 2017;18:107. doi: 10.1186/s13059-017-1245-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Wang T, Tsui B, Kreisberg JF, Robertson NA, Gross AM, Yu MK, Carter H, Brown-Borg HM, Adams PD, Ideker T. Epigenetic aging signatures in mice livers are slowed by dwarfism, calorie restriction and rapamycin treatment. Genome Biology. 2017;18:57. doi: 10.1186/s13059-017-1186-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Weidner CI, Lin Q, Koch CM, Eisele L, Beier F, Ziegler P, Bauerschlag DO, Jöckel KH, Erbel R, Mühleisen TW, Zenke M, Brümmendorf TH, Wagner W. Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biology. 2014;15:R24. doi: 10.1186/gb-2014-15-2-r24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Zhang Y, Hapala J, Brenner H, Wagner W. Individual CpG sites that are associated with age and life expectancy become hypomethylated upon aging. Clinical Epigenetics. 2017;9:9. doi: 10.1186/s13148-017-0315-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision letter

Editor: Vadim N Gladyshev1
Reviewed by: Morgan Levine2

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Epigenetic Age-Predictor for Mice based on Three CpG Sites" for consideration by eLife. Your article has been reviewed by three peer reviewers and the evaluation has been overseen by a Reviewing Editor and Jessica Tyler as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission. We consider this work more appropriate in the category of a Tools and Resources paper rather than as a Research Article.

Summary:

The submitted study presents a new methylation clock for mouse blood based on analysis of only three CpG sites. This seems to be a useful and practical tool, as previous studies produced accurate methylation clocks based on ~100 sites. The training set included 24 blood samples obtained from C57BL/6 mice representing 12 age groups ranging from 11 to 117 week old. Although the number of samples is not large, fair representation of age groups seems to make the clock usable across ages. The authors focused on 9 genomic regions enriched in age-associated sites obtained from previous studies. Three individual sites with the highest correlation with the training set were then selected, and the clock was built based on a simple multivariate linear model. The validation set included 21 mice from the same site (University of Ulm) and 19 mice from a different site (University of Groningen). Precision of the clock was equal to MAE = 3.6 weeks for the training set and MAE = 5 and MAE = 5.9 weeks for validation sets. This is comparable with the available clocks produced by other methods. The new clock was applied to 25 samples of DBA/2 mice having a shorter lifespan. The clock showed a higher age for these short-lived mice compared to C57BL/6 mice of the same chronological age.

Essential revisions:

1) Methods used for site selection are not completely clear. The authors explain their selection by choosing the sites with maximal correlation with age. However, first, it's not clear why exactly three sites were chosen. Second, selection of sites with maximal individual correlation doesn't guarantee that the multivariate model based on these sites would result in highest precision. To make site selection more convincing, you may apply a machine learning approach (linear model with L1 regularization) to the whole set of sites (all sites from 9 genomic regions) and vary the regularization parameter to obtain models with different numbers of sites. Then, these models can be applied to the validation set 1, and precision can be calculated. In this case, you could show how precision changes with the number of sites (the number of remaining sites in the model on the x axis and precision (R2 or MAE) on the y axis). This will tell how much precision you lose when proceeding to the model with fewer sites. Based on this plot, you could select the model with the optimal number of sites (minimal number of sites that provides precision, which doesn't significantly increase with the addition of additional sites). And then apply it to the validation set 2 to get the unbiased estimate of precision. This approach could make the analysis much more convincing and also explain the choice of the number of sites.

2) R2 is shown for every training and validation set as a metric of quality. However, in the text it is explained as Spearman correlation. This complicates interpretation of the results as usually the ratio of explained variance is denoted by R2, which is equal to the square of Pearson correlation, but not to the Spearman correlation. Please, either change the symbol you use (for example, correlation coefficient is usually denoted as ρ), or explain the R2 in the text (for example, specify that this is Spearman correlation squared).

3) You didn't specify the number of age groups used for the development of the clock. From the figure, it seems 12 age groups were used. We recommend adding this information to the text as it supports the analysis (12 age groups is a broad range that makes the results more convincing).

4) Comparison of age prediction for C57BL/6 and DBA/2 mice is questionable. DBA/2 samples represent a narrow range of ages, which includes almost no young mice (based on the figure it appears that only 4 samples represent mice <75 weeks old). This reduces quality of the analysis, as nonlinear behavior is often observed in the old ages, which can partly explain the difference between the ages predicted for C57BL/6 and DBA/2 mice. Development of the clock for DBA/2 samples is even more dependent on the age range. Therefore, quality of the clock built for DBA/2 does not look reliable. Additional samples of young DBA/2 mice could improve quality of the findings. Alternatively, this drawback should be clearly noted in the text and text revised accordingly.

5) In the Abstract, you state "DBA/2J mice revealed accelerated epigenetic aging as compared to C57BL6 mice" In fact, Figure 2 appears to show that the DBA/2 mice are about "40 weeks older" at every age – there is barely any age-associated divergence of the predicted aged for DBA/2 and C57BL/6. In other words, it does not seem as if the DBA/2 are aging faster. Rather, they appear to be born older and remain so throughout life. This is perhaps best explained by a need for re-calibrating the clock in different strains of mice. Figure 4E appears to confirm this. So, we agree with the authors conclusion that "age-predictors should be adjusted for different inbred mice strains" but do not agree that "DBA/2J mice revealed accelerated epigenetic aging as compared to C57BL6 mice."

6) You didn't specify if both training and validation sets or only validation sets of C57BL/6 mice were used when the predicted age was compared between this strain and DBA/2. To make the analysis unbiased from the construction of the clock, only validation sets should be used there. Based on the figure, it seems this was indeed the case, but anyway it should be specified in the text as this is important from the methodological point of view.

7) There is far too much emphasis placed on age prediction. Ultimately, the residual or difference between chronological and epigenetic age is of the most interest. The goal is not to develop near perfect age predictors. In humans, the clocks with the strongest age predictions typically do not contribute the most to differential risk of aging-related conditions, which should be the goal. This point can be addressed by revision of the text.

8) Is this new measure specific to blood? Were any experiments done to validate it in any other tissue or in sorted cells? It is possible that changes in methylation may be confounded by blood cell composition, etc. If it is not possible to address it experimentally within the revision timeframe, it should at the very least be discussed as a limitation.

9) Epigenetic measures in mice will only be useful if (a) they track difference in lifespan/healthspan both between and within strains, and (b) they show response to intervention. We feel like the utility of this clock is being over-sold prior to the necessary validations being shown.

10) "These results provide further evidence that age-associated DNAm is generally rather related to biological age […]" is an overstatement. If the DBA/2J mice are not aging epigenetically, there is little data in this manuscript to support the idea that the epigenetic clock is a measure of biological age as opposed to chronological age. The clock that the authors have reported here was calibrated vs. chronological age and appears to function well as such. However, there is no direct evidence that it reports on biological age.

eLife. 2018 Aug 24;7:e37462. doi: 10.7554/eLife.37462.015

Author response

Essential revisions:

1) Methods used for site selection are not completely clear. The authors explain their selection by choosing the sites with maximal correlation with age. However, first, it's not clear why exactly three sites were chosen. Second, selection of sites with maximal individual correlation doesn't guarantee that the multivariate model based on these sites would result in highest precision. To make site selection more convincing, you may apply a machine learning approach (linear model with L1 regularization) to the whole set of sites (all sites from 9 genomic regions) and vary the regularization parameter to obtain models with different numbers of sites. Then, these models can be applied to the validation set 1, and precision can be calculated. In this case, you could show how precision changes with the number of sites (the number of remaining sites in the model on the x axis and precision (R2 or MAE) on the y axis). This will tell how much precision you lose when proceeding to the model with fewer sites. Based on this plot, you could select the model with the optimal number of sites (minimal number of sites that provides precision, which doesn't significantly increase with the addition of additional sites). And then apply it to the validation set 2 to get the unbiased estimate of precision. This approach could make the analysis much more convincing and also explain the choice of the number of sites.

We agree that the suggested machine learning approach is very well suited for selection of CpGs in genome wide DNA methylation data. However, for our bisulfite pyrosequencing assays additional points need to be taken into account: not all genomic regions are suitable to design reliable pyrosequencing assays, precision of pyrosequencing declines towards the end of the reads, and the costs for pyrosequencing runs can be significantly reduced by shorter reads in the sequencing step. Following the reviewers’ advice, we have now done the suggested machine learning approach to identify the best set of CpGs from the 71 CpGs that were covered by the nine amplicons (all sites from 9 genomic regions). These experiments were done by Julia Franzen who is now coauthor on the manuscript. In fact, taking 15 CpGs into account the precision was better in cross-validation with data of the training set and this is now described in the text. On the other hand, the training set is relatively small for this approach and the precision should be further validated in the future. In the validation set, we did not do pyrosequencing of all amplicons and we sequenced only a shorter read length to save some of the costs – thus, we cannot validate the precision of this 15 CpG signature in our independent datasets. Either way, using machine learning to derive more focused signatures with three CpGs the approach selected those three CpGs that revealed the highest Pearson correlations of all 71 CpGs – but all of these would be in the same amplicon. We reasoned that combination of hyper- and hypomethylated CpGs and shorter reads would be advantageous. Thus, we feel that our selection was plausible and this is now described in the text.

2) R2 is shown for every training and validation set as a metric of quality. However, in the text it is explained as Spearman correlation. This complicates interpretation of the results as usually the ratio of explained variance is denoted by R2, which is equal to the square of Pearson correlation, but not to the Spearman correlation. Please, either change the symbol you use (for example, correlation coefficient is usually denoted as ρ), or explain the R2 in the text (for example, specify that this is Spearman correlation squared).

Thank you for notifying us on this mistake. We have now corrected this indicating that R is Pearson correlation.

3) You didn't specify the number of age groups used for the development of the clock. From the figure, it seems 12 age groups were used. We recommend adding this information to the text as it supports the analysis (12 age groups is a broad range that makes the results more convincing).

Indeed, the training set comprised mice of 12 different ages, covering a broad range of ages. As suggested, this is now stated in the manuscript.

Results section: “analyzed in 24 blood samples of female C57BL/6 mice that covered a broad range of 12 different age groups (11 to 117 weeks old).”

4) Comparison of age prediction for C57BL/6 and DBA/2 mice is questionable. DBA/2 samples represent a narrow range of ages, which includes almost no young mice (based on the figure it appears that only 4 samples represent mice <75 weeks old). This reduces quality of the analysis, as nonlinear behavior is often observed in the old ages, which can partly explain the difference between the ages predicted for C57BL/6 and DBA/2 mice. Development of the clock for DBA/2 samples is even more dependent on the age range. Therefore, quality of the clock built for DBA/2 does not look reliable. Additional samples of young DBA/2 mice could improve quality of the findings. Alternatively, this drawback should be clearly noted in the text and text revised accordingly.

This point was very well taken. Following the reviewer’s suggestion, we have now analyzed eight additional samples of young DBA/2 mice. Figures and regression models were adjusted accordingly. In fact, the results clearly support the notion that DNAm followed a non-linear behavior particularly in young DBA/2 mice. The additional results are now in line with our previous assumption that DBA/2 mice reveal faster epigenetic aging. Furthermore, correlation and predictions of the DBA/2 model became much better with the additional measurements (MAE improved from 8.3 weeks to 5.3 weeks).

5) In the Abstract, you state "DBA/2J mice revealed accelerated epigenetic aging as compared to C57BL6 mice" In fact, Figure 2 appears to show that the DBA/2 mice are about "40 weeks older" at every age – there is barely any age-associated divergence of the predicted aged for DBA/2 and C57BL/6. In other words, it does not seem as if the DBA/2 are aging faster. Rather, they appear to be born older and remain so throughout life. This is perhaps best explained by a need for re-calibrating the clock in different strains of mice. Figure 4Eappears to confirm this. So, we agree with the authors conclusion that "age-predictors should be adjusted for different inbred mice strains" but do not agree that "DBA/2J mice revealed accelerated epigenetic aging as compared to C57BL6 mice."

Based on our previous data this comment was very well taken. As mentioned above, we have now measured additional young DBA/2 mice and these results support the notion that there is accelerated epigenetic aging in the young DBA/2 mice. It now became evident that the epigenetic age predictions followed rather a logarithmic function (Figure 4D), as previously reported for human pediatric age predictions by Alisch et al., (Genome Research, 2012).

6) You didn't specify if both training and validation sets or only validations sets of C57BL/6 mice were used when the predicted age was compared between this strain and DBA/2. To make the analysis unbiased from the construction of the clock, only validation sets should be used there. Based on the figure, it seems this was indeed the case, but anyway it should be specified in the text as this is important from the methodological point of view.

In fact, we only used the results of the validation sets of C57BL/6 mice for this comparison and this is now clarified in the figure legend of Figure 4 as suggested.

“For comparison we provided measurements of the C57BL/6 mice (only from validation sets; blue).”

7) There is far too much emphasis placed on age prediction. Ultimately, the residual or difference between chronological and epigenetic age is of the most interest. The goal is not to develop near perfect age predictors. In humans, the clocks with the strongest age predictions typically do not contribute the most to differential risk of aging-related conditions, which should be the goal. This point can be addressed by revision of the text.

This point was well taken and we have now better clarified in the discussion that the residual of chronological and predicted age is of the highest relevance to identifying conditions that impact on epigenetic aging.

For example, in the Discussion section: “The ultimate goal of epigenetic age-predictors for mice is not to develop near perfect age predictors, but to provide a surrogate for biological aging that facilitates assessment of interventions on aging. In fact, using deep sequencing approaches (RRBS or WGBS) several groups already indicated that relevant parameters that affect aging of the organism – such as diet, genetic background, and drugs – do also impact on epigenetic aging (Cole et al., 2017; Hahn et al., 2017; Maegawa et al., 2017).”

8) Is this new measure specific to blood? Were any experiments done to validate it in any other tissue or in sorted cells? It is possible that changes in methylation may be confounded by blood cell composition, etc. If it is not possible to address it experimentally within the revision timeframe, it should at the very least be discussed as a limitation.

Following the reviewer’s advice, we have teamed up with Dr. Ander Izeta and Laura Yndriago, who are now coauthors on the manuscript. They provided 60 samples of young and old mice of 10 different tissues. These samples were analyzed with our 3 CpG signature (new Figure 3). Several tissues revealed significant age-associated changes, similar to blood. On the other hand, the DNAm levels varied between different tissues, indicating that the signature needs to be retrained to test applicability for these tissues. According to the reviewers’ advice, we have also discussed that epigenetic age predictions might be influenced by blood counts and that sorted subsets should be analyzed in the future.

Discussion section: “Furthermore, DNAm levels may vary between different hematopoietic subsets (Frobel et al., 2017; Houseman, Molitor and Marsit, 2014). In the future, sorted subsets should be analyzed to determine how the three CpG signature is affected by blood counts.”

9) Epigenetic measures in mice will only be useful if (a) they track difference in lifespan/healthspan both between and within strains, and (b) they show response to intervention. We feel like the utility of this clock is being over-sold prior to the necessary validations being shown.

We agree with this critical comment and indicated that additional studies are required to unequivocally demonstrate association with lifespan. At least, the additional results with young DBA/2 mice now provide evidence that our epigenetic measures track differences in lifespan between strains. Furthermore, we have preliminary results of another study with dietary interventions which indicate that there might be association with life-expectancy, but these results cannot be integrated into this manuscript since this is ongoing work with another group. In the revised manuscript we have much better discussed the relevance of tracking the difference in lifespan/healthspan.

10) "These results provide further evidence that age-associated DNAm is generally rather related to biological age […]" is an overstatement. If the DBA/2J mice are not aging epigenetically, there is little data in this manuscript to support the idea that the epigenetic clock is a measure of biological age as opposed to chronological age. The clock that the authors have reported here was calibrated vs chronological age and appears to function well as such. However, there is no direct evidence that it reports on biological age.

As indicated above our additional results support the notion that DBA/2 mice are aging epigenetically faster. Furthermore, we have added the following passage to refer to this important issue, which is still under debate.

Discussion section: “It is yet unclear if epigenetic aging signatures can be specifically trained to either correlate with chronological age or biological age. For humans, recent studies indicate that this might be possible (Levine et al., 2018) and we have previously demonstrated that even individual age-associated CpGs can be indicative for life expectancy (Zhang et al., 2017). Further studies will be necessary to gain better understanding how epigenetic age predictions are related to the real state of biological aging, and how it is related to alternative approaches to quantify biological aging, such as telomere length (Belsky et al., 2018).”

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    Source data 1. Pyrosequencing raw data of mouse epigenetic aging predictor.
    elife-37462-data1.xlsx (16.4KB, xlsx)
    DOI: 10.7554/eLife.37462.008
    Supplementary file 1. Age-associated DNAm in nine genomic regions of the training set.
    elife-37462-supp1.docx (25.9KB, docx)
    DOI: 10.7554/eLife.37462.009
    Supplementary file 2. Multivariable model based on 15 CpGs.
    elife-37462-supp2.docx (18.4KB, docx)
    DOI: 10.7554/eLife.37462.010
    Supplementary file 3. Primers for pyrosequencing.
    elife-37462-supp3.docx (13.4KB, docx)
    DOI: 10.7554/eLife.37462.011
    Transparent reporting form
    elife-37462-transrepform.pdf (312.8KB, pdf)
    DOI: 10.7554/eLife.37462.012

    Data Availability Statement

    Raw data of pyrosequencing is provided as supplemental EXCEL table (Source data 1).

    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES