Epigenetic clock for skin and blood cells applied to Hutchinson Gilford Progeria Syndrome and ex vivo studies

DNA methylation data sets

We analyzed both novel and existing DNA methylation data sets that were generated on the Illumina Infinium platform (Table 1). DNA was extracted from human fibroblasts, keratinocytes, buccal cells, endothelial cells, blood, and saliva. We analyzed data from two Illumina platforms (Infinium 450K and the EPIC array, also known as the 850K array) to ensure that the resulting estimator would apply to the latest Illumina platform (the EPIC array).

The DNAm age estimator for skin and blood

To ensure an unbiased validation of the test data, we used only the training data to define and construct the DNAm age estimator. As detailed in Methods, a transformed version of chronological age was regressed on methylation states of CpGs using elastic net regression [25], which automatically selected 391 CpGs (Methods). We refer to the 391 CpGs as (epigenetic) clock CpGs since their weighted average (formed by the regression coefficients) amounts to a highly accurate epigenetic aging clock.

The following description will demonstrate that the resulting age estimator (referred to as skin & blood clock) performs remarkably well across a wide spectrum of cells that are most frequently used in ex vivo studies. The new skin & blood clock even outperforms the pan-tissue clock (Horvath 2013) [6] in all metrics of accuracy (age correlation, median error) in fibroblasts, microvascular endothelial cells, buccal epithelial cells, keratinocytes, and dermis/epidermis samples (Figure 1 and Supplementary Figure 1). As indicated by its name, the new skin & blood clock is also a highly accurate age estimator of blood methylation data, where it provides more accurate age estimates than the widely used estimators by Horvath (2013) [6] and Hannum (2013) [9] (Figure 2A,D,G and Supplementary Figure 2). Further, it outperforms the Horvath and Hannum DNAm age estimators when applied to lymphoblastoid cell lines (Figure 2B,E,H), i.e. B cells that have been immortalized using EBV transformation. Interestingly, the DNAm age of blood is highly correlated with the DNAm age estimate of the lymphoblastoid cell line collected from the same donor at the same time (r=0.83, Figure 2C). The skin & blood clock accurately estimates age in two different saliva DNA methylation data sets (age correlations r=0.9 and r=0.95) and outperforms the pan-tissue DNAm age estimator in these data (Supplementary Figure 3). The skin & blood clock also applies to cord blood samples as can be seen from the fact that it accurately estimates gestational age in three different data (with correlations ranging from r=0.15 to r=0.66, Supplementary Figure 4).

Skin & blood clock applied to brain, liver, bone, and other body parts

The skin & blood clock leads to DNAm age estimates that strongly correlate with chronological age in a host of different cell types and tissues including sorted neurons and glial cells (Supplementary Figure 5), brain samples (Supplementary Figure 6), liver samples (Supplementary Figure 7), and trabecular bone samples (Supplementary Figure 8). In the following, we provide more details on the individual studies.

In sorted neurons, DNAm age correlates strongly (r=0.83) with chronological age but the DNAm age estimator is substantially lower than chronological age (mean DNAm Age=6 years in a group of people whose mean chronological age is 31 years, Supplementary Figure 5A). Interestingly, the DNAm age of glial cells is significantly higher than that of neurons from the same individual (p=0.00024, Supplementary Figure 5B,C). Chronological age is also highly correlated with DNAm age estimates in different brain regions (r between 0.67 and 0.96, Supplementary Figure 6) but the age estimates tend to be systematically lower than chronological age. In particular, the cerebellum ages substantially slower than other brain regions echoing previous results for the pan-tissue clock [17].

DNAm age of liver tissue is highly correlated with chronological age (r=0.86, Supplementary Figure 7) and the corresponding measure of epigenetic age acceleration correlates with body mass index (r=0.27, p=0.027, Supplementary Figure 7B). These results echo earlier results obtained from the pan-tissue DNAm age estimator [13].

An analysis of 30 different body parts from a 112 year old woman reveals a) that most body parts have roughly the same age and b) that the cerebellum is substantially younger (Supplementary Figure 9) which is consistent with previous results [17].

Epigenetic age of fibroblasts from Hutchinson Gilford Progeria fibroblasts

Previously, the use of the pan-tissue clock revealed epigenetic age acceleration in segmental progeroid syndromes such as Down syndrome and Werner syndrome [15,24], but not in syndrome X, whose patients exhibit dramatically delayed development (seemingly eternal toddler-like state) [26]. The status of the epigenetic aging rate in regards to HGPS and Atypical Werner Syndrome (AWS) is less clear. These two conditions can be caused by different progeroid mutations of the LMNA gene (Figure 3). It is not yet known whether HGPS patients, who generally appear normal at birth but exhibit a “failure-to-thrive” syndrome, exhibit any epigenetic age acceleration. HGPS is associated with many clinical manifestations of accelerated aging including loss of subcutaneous fat, joint contractures, and a striking aged appearance during the first to third years of life [27]. Virtually all HGPS patients die of myocardial infarction at a median age of 14.6 years [28]. Classic HGPS is caused by a recurrent heterozygous pathological variant, c.1824C>T in exon 11 of the LMNA gene, which activates a cryptic splice site and causes a 50-amino acid in-frame deletion (Δ50) [29]. The resulting abnormal protein, termed progerin, lacks the proteolytic site for an essential but transient post-translational modification by the ZMPSTE24 metalloprotease. This causes retention of the C-terminal farnesylated moiety, resulting in aberrant nuclear structure and function [29]. Non-classical HGPS mutations at the exon 11 and intron 11 boundary, including c.1968+1G>A [30] and c.1968+2T>C [31], can also activate the cryptic splice site, leading to the accumulation of progerin and an infantile-onset HGPS phenotype. Biallelic ZMPSTE24 mutations also cause accumulations of farnesylated lamin A and a varying degree of progeroid phenotypes, depending on the residual enzymatic activity of ZMPSTE24 [32,33]. In rare instances, a homozygous amino acid substitution of lamin A can present with a phenotype similar to HGPS or mandibuloacral dysplasia, as described in cases with [p.Met540Thr; p.Met540Thr] [34] and [p.Thr528Met; p.Met540Thr] [35].

A small subset of cases of Atypical Werner syndrome (AWS) (those with some features of Werner syndrome, without mutations in WRN or altered expressions of the WRN protein) may be caused by accumulations of low levels of progerin [36,37]. Pathological lamin A variants found in some patients with AWS include c.1968G>A and c.1968+5G>A [36]. While there is a general genotype-phenotype correlation between the amount of progerin and the severity of the disease, the amounts and structures of progerin can vary among those who carry the same LMNA splice mutation, and the severity of the disease can vary among patients within the same family [36,37]. The median age of death of classic HGPS is ~14.6 years [38], while the range in AWS patients with low levels of progerin is 37 to 60 years [36].

The original pan-tissue DNAm age estimator did not detect any age acceleration in HGPS individuals (Supplementary Table 1). By contrast, the application of the novel skin & blood clock showed that while DNAm age is highly correlated with chronological age in normal fibroblasts, those from HGPS cases exhibited accelerated epigenetic aging (Figure 4). The correlation between age and DNAm age in HGPS children (<10 years old) is substantially lower (r=0.71) than that of control children (r=0.71, Figure 4B). The epigenetic age acceleration effects become particularly pronounced after adjusting for differences in cell population doubling levels and when the analysis was restricted to children who are younger than 10 years old (p=0.00021, Table 2, Table 3). There is a non-significant trend of increased DNAm age in Atypical Werner Syndrome cases with low levels of progerin. It is perhaps not unexpected that AWS, which presents with a lower progerin concentration (Figure 3) is not significantly associated with greater magnitude of epigenetic age acceleration.

Although non-classic HGPS patients often present at later ages, they can nevertheless be diagnosed at ages that are slightly younger than patients with classic HGPS [27]. It should indeed be noted that the cases examined in this study (see Methods for mutation details), have exceptionally early manifestations – as early as birth or younger than 5 months of age. Interestingly, their DNA methylation age acceleration is comparable and consistent with that of classic HGPS, which as mentioned, is an early onset progeria condition (Figure 4D).

Detailed results for the lines of skin fibroblasts provided by The Progeria Research Foundation are presented in Table 2 and Supplementary Table 2. The skin & blood clock provides marginally significant evidence (p=0.062) that fibroblasts from boys with classic HGPS are epigenetically older than those from girls with classic HGPS, but this gender effect is not apparent when classic and non-classic HGPS samples are pooled for analyses (Supplementary Figure 10).

It is to be further noted that the epigenetic age acceleration of HGPS fibroblasts revealed by the skin & blood clock escaped detection when measurements were carried out with the pan-tissue clock; indeed, the opposite appears to be the case (Supplementary Figure 11C, Supplementary Table 2). Evidently, the ability to detect such epigenetic age changes in fibroblasts is dependent upon the choice of the DNAm age estimator that is used.

Ex vivo studies of anti-aging interventions

While it may appear obvious that the skin & blood clock is superior in terms of compatibility with fibroblasts, it was still necessary to verify and validate this deduction by applying this clock to non-progeria fibroblasts and other cell types. To this end, fibroblasts derived from non-progeria neonatal foreskins are ideal as they pose minimal to no confounding factors that could alter their age. While the skin & blood clock correctly estimated the neonatal fibroblast cells to be of ages close to zero years, the pan-tissue age estimator leads to age estimates that are greater than 10 years (Figure 5A and B). Analyses of other skin cell types namely, keratinocytes and microvascular endothelial cells derived from neonatal foreskins also demonstrated greater accuracy of the skin & blood clock over the other age estimators. This conclusion continues to hold true even when the analyses were extended to isogenic skin cells derived from adult tissues (Supplementary Figure 1).

Having established the robustness of the skin & blood clock in measuring age of cells isolated from human tissues, we proceeded to test the applicability of the clock on human cells cultured ex vivo. As observed previously using the pan-tissue age estimator, the skin & blood clock revealed that human fibroblasts cultured ex vivo undergo epigenetic aging. However, unlike the former which over-estimates the DNAm age of fibroblasts (Figure 5B), the skin & blood clock correctly estimated the age of the neonatal cells to within 6 months (Figure 5A). Proliferation of human fibroblasts in culture, measured as population doubling, was observed to correlate with continual increase in DNAm age. Similar correlation was also seen with neonatal foreskin keratinocytes (Figure 5A and B). Although this association was revealed by both clocks, the resolution and accuracy of the skin & blood clock are clearly much greater and better. The new clock also out-performs the pan-tissue clock when applied to non-blood or skin cells; namely, the human coronary artery endothelial cells, whose increase in epigenetic age in function of population doubling was readily detected by the former but not the latter. Cells whose proliferative capacity was extended by hTERT beyond their otherwise natural limits (senescence), continued to age epigenetically, underscoring the correlation between cellular proliferation and DNA methylation aging.

By its ability to quantitatively track aging of human cells ex vivo, the skin & blood clock lends itself to be used in the development of an ex vivo human cell aging assay that can be used for testing and screening compounds with anti-aging or pro-aging effects.

Analysis of blood samples from human cohort studies

Epigenetic clocks are associated with a host of different age-related phenotypes and conditions (reviewed in [11,12]). In the following, we report the results of several post-hoc analyses that demonstrate that the new skin & blood clock satisfies most of the properties observed for other blood based DNAm age estimators.

Similar to what has been observed with previous age estimators, epigenetic age acceleration in blood (according to the skin & blood clock) is highly predictive of time to all-cause mortality (p=9.6E-7) according to a univariate Cox regression model fixed effects meta-analysis across multiple epidemiological cohort studies (Supplementary Figure 12).

Blood samples from individuals with Down syndrome exhibit positive epigenetic age acceleration compared to controls (p=0.034, Supplementary Figure 13) consistent with previous findings [15].

Similar to the previous epigenetic aging clock analyses of blood [21], cross sectional studies of n=3700 blood samples from postmenopausal women from the Women's Health Initiative revealed relationships to lifestyle factors and dietary variables (Supplementary Table 3). Slow epigenetic age acceleration in blood was associated with higher education (p=6E-5), physical exercise (p=4E-3), fish consumption (p=2E-4), poultry consumption (p=3E-4), high mean carotenoid levels (p=8E-6), beta cryptoxhanthin (p=2E-7), beta carotene levels (p=4E-4), and HDL levels (p=5E-4, Supplementary Table 3). Conversely, faster epigenetic aging in blood is associated with higher C-reactive protein levels (p=1E-3), body mass index (p=0.01), triglyceride levels (p=3E-3), and insulin (p=2E-3). However, it is worth emphasizing that the respective correlation coefficients were weak (|r|<0.11, Supplementary Table 3). Physical exercise was associated with a slower epigenetic aging effect in African American women (p=4E-3) and perhaps in Caucasian women (p=0.07) but not in Hispanic women (p=0.74). Current smoking status was only associated with increased age acceleration in Caucasian women (p=0.04).

Epigenetic age acceleration is highly conserved across a 9 year follow up time period (r=0.71, Supplementary Figure 14). In other words, if an individual exhibits positive epigenetic age acceleration at age 40 then he/she will probably continue to exhibit positive epigenetic age acceleration at age 49.

Collectively these characteristics demonstrate that although the new clock is highly and uniquely accurate for skin cells, it has not acquired this at the cost of losing any of the features shared amongst existing age estimators in being also highly accurate with blood DNA methylation data, buccal cells, saliva. As such, this clock could become useful for forensic applications.

Relationship to other DNAm age estimators

The skin & blood clock (based on 391 CpGs) shares 45 CpGs (out of 71 CpGs) with the blood-based clock from Hannum (2013) and 60 CpGs (out of 353 CpGs) with the pan tissue clock from Horvath (2013) as detailed in Supplementary Table 4. Despite this significant overlap, epigenetic age acceleration of the skin & blood clock exhibits only moderate correlations with corresponding epigenetic age acceleration measures by Horvath (2013) and Hannum (2013) (r=0.5 and r=0.59, p<1.E-110) in the blood samples from the Women's Health Initiative (BA23 study).

Leukocyte telomere length and blood cell counts

We find a weak a weak negative correlation between leukocyte telomere length and epigenetic age acceleration of blood (r=-0.087, p=0.0088 in n=905 samples from the Framingham Heart Study, r =-0.081, p=0.0011 in n=1639 samples from the Jackson Heart Study, and r=-0.117, p=0.00079 in the 818 samples from the Women's Health Initiative). This weak negative relationship between telomere length and epigenetic age acceleration is similar to that of the Hannum-based DNAm age estimator and other blood based biomarkers [11,39,40].

Epigenetic age acceleration measured by the skin & blood clock is only weakly correlated with, or affected by blood cell type counts, as is evident from the analyses of postmenopausal women from the Women's Health Initiative (Supplementary Figure 15). The strongest correlations are observed with exhausted CD8+ T cells (r=0.22), naive CD8+ T cells (r=-0.21), and naive CD4+T cells (r=-0.19, Supplementary Figure 15B-D). These correlations suggest that individuals with positive epigenetic age accelerations exhibit an adaptive immune system that is older than expected.

Definition of DNAm age using a penalized regression model

A penalized regression model (implemented in the R package glmnet [46]) was used to regress a calibrated version of chronological age on the CpG probes in the training set. We restricted the analysis to CpGs that were present both on the Illumina 450K and EPIC platforms and were in one of the following subsets: 1) most significant CpGs with high positive/negative correlation with chronological age in different cell types or 2) 500 CpGs with the least significant correlation with age. The alpha parameter of glmnet was chosen as 0.5 (elastic net regression) and the lambda value was chosen using cross-validation on the training data. DNAm age was defined as predicted age.

Fibroblasts from The Progeria Research Foundation

Human primary dermal fibroblast cell lines were obtained from The Progeria Research Foundation (PRF) Cell and Tissue Bank (www.progeriaresearch.org). The fibroblast cell lines originated from cases with classic mutations, non-classic mutations and parental controls as detailed in Table 2. The following citations provide additional details on cases carrying the specific variants: LMNA c.1968+1G>A heterozygote [30], LMNA c.1968+2T>C heterozygote [31], LMNA p.Met540Thr homozygotes [34] and compound heterozygotes of ZMPSTE24 p.Pro248Leu and p.Trp450* [32]. Additional details can be found in the Supplement.

Isolation and culture of cells for ex vivo experiments

Informed consent was obtained prior to collection of human skin samples with approved by the Oxford Research Ethics Committee; reference 10/H0605/1. Human skin samples were acquired under ethical approval. Primary human skin keratinocytes, fibroblasts and microvascular endothelial cells were isolated from neonatal foreskin and adult facial/neck skin. Keratinocytes were cultured in CnT media (CellnTec) on fibronectin/collagen-coated plates, fibroblasts were cultured in DMEM (Sigma) supplemented with 10% foetal calf serum and CD31-selected microvascular endothelial cells were cultured in Endothelial Cell Growth Medium MV (PromoCell, C-22020) on gelatin-coated plates. Human Coronary Artery Endothelial Cells (HCAEC) from a male, aged 26 years, were obtained from Sigma and grown in MesoEndo Cell Growth Medium (Sigma). HCAEC were immortalized with pBABE-neo-hTERT (Addgene, cat. 1774) and after selection, were cultured in parallel with uninfected control until control reached senescence. All cells were maintained in a 37 °C, 5% CO2 humidified environment. At each passaging step, cells were counted, population doubling calculated and 10,000 were seeded into a fresh 10cm plate. Remaining cells were used for DNA extraction. Population doubling was calculated with the following formula: [log(number of cells harvested) – log(number of cells seeded)] x 3.32. Cumulative population doubling was obtained by addition of population doubling of each passage. Additional details can be found in the Supplement.

Sample preparation

DNA was extracted from cells using the Zymo Quick DNA mini-prep plus kit (D4069) according to the manufacturer’s instructions and DNA methylation levels were measured on Illumina 450 or Illumina 850 EPIC arrays according to the manufacturer’s instructions.

Blood methylation data from different epidemiological cohorts

A number of validation studies were used to test associations between DNAm age and various aging-related traits including time to all-cause mortality. Details on these studies can be found in the Supplement and in [20,40].

Competing interests

The Regents of the University of California is the sole owner of a provisional patent application directed at this invention for which SH is a named inventor.

Ethics approval

This study was reviewed by the UCLA institutional review board (IRB#14-000140, IRB#15-001479). Collection of human skin samples were carried out under the ethical approval issues by the Oxford Research Ethics Committee; reference 10/H0605/1 (United Kingdom)