ArticlePDF Available

Abstract and Figures

Identifying reliable biomarkers of aging is a major goal in geroscience. While the first generation of epigenetic biomarkers of aging were developed using chronological age as a surrogate for biological age, we hypothesized that incorporation of composite clinical measures of phenotypic age that capture differences in lifespan and healthspan may identify novel CpGs and facilitate the development of a more powerful epigenetic biomarker of aging. Using an innovative two-step process, we develop a new epigenetic biomarker of aging, DNAm PhenoAge, that strongly outperforms previous measures in regards to predictions for a variety of aging outcomes, including all-cause mortality, cancers, healthspan, physical functioning, and Alzheimer's disease. While this biomarker was developed using data from whole blood, it correlates strongly with age in every tissue and cell tested. Based on an in-depth transcriptional analysis in sorted cells, we find that increased epigenetic, relative to chronological age, is associated with increased activation of pro-inflammatory and interferon pathways, and decreased activation of transcriptional/translational machinery, DNA damage response, and mitochondrial signatures. Overall, this single epigenetic biomarker of aging is able to capture risks for an array of diverse outcomes across multiple tissues and cells, and provide insight into important pathways in aging.
Content may be subject to copyright.
www.agingus.com573AGING  AGING2018,Vol.10,No.4
Research Paper
Carolina,ChapelHill,NC 27599,USA
Received:March20,2018Accepted:April8,2018 Published:April17,2018
healthspanmayidentifynovelCpGsandfacilitatethedevelopmentofamore powerful epigenetic biomarker of
One of the major goals of geroscience research is to
define ‘biomarkers of aging’[1, 2], which can be
thought of as individual-level measures of aging that
capture inter-individual differences in the timing of
disease onset, functional decline, and death over the life
course. While chronological age is arguably the
strongest risk factor for aging-related death and disease,
it is important to distinguish chronological time from
biological aging. Individuals of the same chronological
age may exhibit greatly different susceptibilities to age-
related diseases and death, which is likely reflective of
differences in their underlying biological aging
processes. Such biomarkers of aging will be crucial to
enable evaluation of interventions aimed at promoting
healthier aging, by providing a measurable outcome,
which unlike incidence of death and/or disease, does not
require extremely long follow-up observation.
One potential biomarker that has gained significant
interest in recent years is DNA methylation (DNAm).
Chronological time has been shown to elicit predictable
hypo- and hyper-methylation changes at many regions
across the genome [3-7], and as a result, the first
generation of DNAm based biomarkers of aging were
developed to predict chronological chronological age
[8-13]. The blood-based algorithm by Hannum [10] and
the multi-tissue algorithm by Horvath [11] produce age
estimates (DNAm age) that correlate with chronological
age well above r=0.90 for full age range samples.
Nevertheless, while the current epigenetic age
estimators exhibit statistically significant associations
with many age-related diseases and conditions [14-27],
the effect sizes are typically small to moderate. One
explanation is that using chronological age as the
reference, by definition, may exclude CpGs whose
methylation patterns don’t display strong time-
dependent changes, but instead signal the departure of
biological age from chronological age. Thus, it is
important to not only capture CpGs that display changes
with chronological time, but also those that account for
differences in risk and physiological status among
individuals of the same chronological age.
Previous work by us and others have shown that
“phenotypic aging measures”, derived from clinical
biomarkers [28-32], strongly predict differences in the
risk of all-cause mortality, cause-specific mortality,
physical functioning, cognitive performance measures,
and facial aging among same-aged individuals. What’s
more, in representative population data, some of these
measures have been shown to be better indicators of
remaining life expectancy than chronological age [28],
suggesting that they may be approximating individual-
level differences in biological aging rates. As a result,
we hypothesize that a more powerful epigenetic
biomarker of aging could be developed by replacing
prediction of chronological age with prediction of a
surrogate measure of "phenotypic age" that, in and of
itself, differentiates morbidity and mortality risk among
same-age individuals.
Overview of the statistical model and analysis
Our development of the new epigenetic biomarker of
aging proceeded along three main steps (Fig. 1). In step
1, a novel measure of ‘phenotypic age’ was developed
using clinical data from the third National Health and
Nutrition Examination Survey (NHANES). Details on
the phenotypic age estimator can be found in Table 1
and in Supplement 1. In step 2, DNAm from whole
blood was used to predict phenotypic age, such that:
 
 
The coefficient values of this model can be found in
Supplement 2 (Table S6). Predicted estimates from this
model represent a person’s epigenetic age, which we
refer to as ‘DNAm PhenoAge’. Using multiple indepen-
dent datasets, we then tested whether DNAm PhenoAge
was associated with a number of aging-related outcomes.
We also tested whether it differed as a function of social,
behavioral, and demographic characteristics, and whether
it was applicable to tissues other than whole blood.
Finally, in step 3, we examine the underlying biology of
the 513 CpGs in the DNAm PhenoAge measure by
examining differential expression, GO and pathway
enrichment, chromosomal locations, and heritability.
Estimating phenotypic age from clinical biomarkers
For step 1, NHANES III was used to generate a
measure of phenotypic age. NHANES III is a
nationally-representative sample, with over twenty-
three years of mortality follow-up, from which our
analytical sample included 9,926 adults with complete
biomarker data. A Cox penalized regression model—
where the hazard of mortality was regressed on forty-
two clinical markers and chronological age—was used
to select variables for inclusion in our phenotypic age
score. The forty-two biomarkers considered represent
those that were available in both NHANES III and IV.
Based on 10-fold cross-validation, ten variables
(including chronological age) were selected for the
phenotypic age predictor (Table 1, Table S1). These
nine biomarkers and chronological age were then
combined in a phenotypic age estimate (in units of
years) as detailed in Methods.
Validation data for phenotypic age came from
NHANES IV, and included up to 12 years of mortality
follow-up for n=6,209 national representative US
adults. In this population, phenotypic age is correlated
with chronological age at r=0.94. Results from all-cause
and cause-specific (competing risk) mortality pre-
dictions, adjusting for chronological age (Table 2),
show that a one year increase in phenotypic age is
associated with a 9% increase in the risk of all-cause
mortality (HR=1.09, p=3.8E-49), a 9% increase in the
risk of mortality from aging-related diseases (HR=1.09,
p=4.5E-34), a 10% increase in the risk of CVD
mortality (HR=1.10, p=5.1E-17), a 7% increase in the
risk of cancer mortality (HR=1.07, p=7.9E-10), a 20%
increase in the risk of diabetes mortality (HR=1.20,
p=1.9E-11), and a 9% increase in the risk of chronic
lower respiratory disease mortality (HR=1.09, p=6.3E-
4). Further, phenotypic age is highly associated with
comorbidity count (p=3.9E-21) and physical
functioning measures (p=2.1E-10, Supplement 1: Fig.
An epigenetic biomarker of aging (DNAm
For step 2 (Fig. 1), data from n=456 participants at two
time-points in the Invecchiare in Chianti (InCHIANTI)
study was used to relate blood DNAm levels to
phenotypic age. InCHIANTI was used as training data
for the new epigenetic biomarker because the study
assessed all clinical measures needed to estimate
phenotypic age, contained data on DNAm, and had a
large age range population (21-100 years). A total of
20,169 CpGs were considered when generating the new
DNAm measure. They represented those CpGs
available on all three chips (27k, 450k, EPIC), so as to
facilitate usability across platforms. Elastic net regres-
sion, with 10-fold cross-validation, produced a model in
which phenotypic age is predicted by DNAm levels at
513 of the 20,169 CpGs. The linear combination of the
weighted 513 CpGs yields a DNAm based estimator of
phenotypic age that we refer to as ‘DNAm PhenoAge’
(mean=58.9, s.d.=18.2, range=9.1-106.1), in contrast to
the previously published Hannum and Horvath ‘DNAm
Age’ measures.
While our new clock was trained on cross-sectional data
in InCHIANTI, we capitalized on the repeated time-
points to test whether changes in DNAm PhenoAge are
related to changes in phenotypic age. As expected,
between 1998 and 2007, mean change in DNAm
PhenoAge was 8.51 years, whereas mean change in
clinical phenotypic age was 8.88 years. Moreover,
participants’ clinical phenotypic age (adjusting for
chronological age) at the two time-points was correlated
at r=0.50, whereas participants’ DNAm PhenoAge
(adjusting for chronological age) at the two time-points
was correlated at r=0.68 (Supplement 1: Fig. S2). We
also find that the change in phenotypic age between
Variable Units Weight
Albumin Liver g/L -0.0336
Creatinine Kidney umol/L 0.0095
Glucose, serum Metabolic mmol/L 0.1953
C-reactive protein (log) Inflammation mg/dL 0.0954
Lymphocyte percent Immune % -0.0120
Mean (red) cell volume Immune fL 0.0268
Red cell distribution width Immune % 0.3306
Alkaline phosphatase Liver U/L 0.0019
White blood cell count Immune 1000 cells/uL 0.0554
Age Years 0.0804
Mortality Cause Cases HR P-Value
All-Cause 1052 1.09 3.8E-49
Aging-Related 661 1.09 4.5E-34
CVD 272 1.10 5.1E-17
Cancer 265 1.07 7.9E-10
Alzheimer's 30 1.04 2.6E-1
Diabetes 41 1.20 1.9E-11
Chronic lower respiratory
diseases 53 1.09 6.3E-4
1998 and 2007 is highly correlated with the change in
DNAm PhenoAge between these two time-points
(r=0.74, p=3.2E-80, Supplement 1: Fig. S2).
DNAm PhenoAge strongly relates to all-cause
In step 2 (Fig. 1), the epigenetic biomarker, DNAm
PhenoAge, was calculated in five independent large-
scale samples—two samples from Women’s Health
Initiative (WHI) (n=2,016; and n=2,191), the
Framingham Heart Study (FHS) (n=2,553), the
Normative Aging Study (NAS) (n=657), and the
Jackson Heart Study (JHS) (n=1,747). The first four
studies used the Illumina 450K array while the JHS
employed the latest Illumina EPIC array platform. In
these studies, DNAm PhenoAge correlated with
chronological age at r=0.66 in WHI (Sample 1), r=0.69
in WHI (Sample 2), r=0.78 in FHS, r=0.62 in the NAS,
and r=0.89 in JHS. The five validation samples were
then used to assess the effects of DNAm PhenoAge on
mortality in comparison to the Horvath and Hannum
DNAm Age measures. DNAm PhenoAge was
significantly associated with subsequent mortality risk
in all studies (independent of chronological age), such
that, a one year increase in DNAm PhenoAge is
associated with a 4.5% increase in the risk of all-cause
mortality (Meta(FE)=1.045, Meta p=7.9E-47, Fig. 2).
To better conceptualize what this increase represents,
we compared the predicted life expectancy and
mortality risk for person’s representing the top 5%
(fastest agers), the average, and the bottom 5% (slowest
agers). Results suggest that those in the top 5% of
fastest agers have a mortality hazard of death that is
about 1.62 times that of the average person, i.e. the
hazard of death is 62% higher than that of an average
person. Further, contrasting the 5% fastest agers with
the 5% slowest agers, we find that the hazard of death
of the fastest agers is 2.58 times higher than that of the
bottom 5% slowest agers (HR=1.04511.0/1.045-10.5).
Additionally, both observed and predicted Kaplan-
Meier survival estimates showed that faster agers had
much lower life expectancy and survival rates compared
to average and/or slow agers (Fig. 2).
As shown in Fig. 2, the DNAm age based measures
from Hannum and Horvath also related to all-cause
mortality, consistent with what has been reported
previously [15, 19, 23, 33, 34]. To directly compare the
three epigenetic measures, we contrasted their accuracy
in predicting 10-year and 20-year mortality risk, using
receiver operating characteristics (ROC) curves. DNAm
PhenoAge (adjusted for age) predicts both 10-year
mortality and 20-year mortality significantly better than
the Horvath and Hannum DNAmAge measures (Sup-
plement 1: Table S2). When examining a model that
includes all three measures (Supplement 1: Table S3),
we find that only DNAm PhenoAge is positively
associated with mortality (HR=1.04, p=1.33E-8),
whereas Horvath DNAm Age is now negatively
associated (HR=0.98, p=2.72E-2), and Hannum DNAm
Age has no association (HR=1.01, p=4.66E-1).
DNAm PhenoAge strongly relates to aging-related
Given that aging is believed to also influence disease
incidence/prevalence, we examined whether DNAm
PhenoAge relates to diverse age-related morbidity
outcomes. We observe strong associations between
DNAm PhenoAge and a variety of other aging out-
comes using the same five validation samples (Table 3).
For instance, independent of chronological age, higher
DNAm PhenoAge is associated with an increase in a
person’s number of coexisting morbidities (β=0.008 to
0.031; Meta P-value=1.95E-20), a decrease in
likelihood of being disease-free (β=-0.002 to -0.039;
Meta P-value=2.10E-10), an increase in physical
functioning problems (β=-0.016 to -0.473; Meta P-
value=2.05E-13), an increase in the risk of coronary
heart disease (CHD) risk (β=0.016 to 0.073; Meta P-
DNAm PhenoAge and smoking
Cigarette exposure has been shown to have an
epigenetic fingerprint[35-37], which has been reflected
in previous DNAm risk predictors[38]. Similarly, we
find that DNAm PhenoAge significantly differs
between never (n=1,097), current (n=209), and former
smokers (n=710) (p=0.0033) (Supplement 1, Fig. S3A);
however, conversely, we do not find a robust
association between pack-years and DNAm PhenoAge
(Supplement 1, Fig. S3B-D). Given the association
between DNAm PhenoAge and smoking, we re-
evaluated the morbidity and mortality associations
(fully-adjusted) in our four samples, stratifying by
smoking status (Supplement 1: Fig. S4 and Table S4).
We find that DNAm PhenoAge is associated with
mortality among both smokers (adjusted for pack-
years) (Meta(FE)=1.050, Meta p=7.9E-31), and non-
smokers (Meta(FE)=1.033, Meta p=1.2E-10). DNAm
PhenoAge relates to the number of coexisting
morbidities, physical functioning status, disease free
status, and CHD for both smokers and non-smokers
(Supplement 1: Table S4). In previous work we showed
that Horvath DNAm age of blood predicts lung cancer
risk in the first WHI sample [20]. Using the same data,
we find that a one year increase in DNAm PhenoAge
(adjusting for chronological age, race/ethnicity, pack-
years, and smoking status) is associated with a 5%
increase in the risk of lung cancer incidence and/or
mortality (HR=1.05, p=0.031). Further, when restricting
the model to current smokers only, we find that the
effect of DNAm PhenoAge on future lung cancer
incidence and/or mortality is even stronger (HR=1.10,
DNAm PhenoAge in other tissues
One advantage of developing biological aging estimates
based on molecular markers (like DNAm), rather than
clinical risk measures (e.g. those in the phenotypic age
variable), is that they may lend themselves to measuring
tissue/cell specific aging. Although DNAm PhenoAge
was developed using samples from whole blood, our
empirical results show that it strongly correlates with
chronological age in a host of different tissues and cell
types (Fig. 3). For instance, when examining all tissues
concurrently, the correlation between DNAm PhenoAge
and chronological age was 0.71. Age correlations in
brain tissue ranged from 0.54 to 0.92, while correlations
were also found in breast (r=0.47), buccal cells (r=0.88),
dermal fibroblasts (r=0.87), epidermis (r=0.84), colon
(r=0.88), heart (r=0.66), kidney (r=0.64), liver (r=0.80),
lung (r=055), and saliva (r=0.81).
Alzheimer's disease and brain samples
Based on the accuracy of the age prediction in other
tissues/cells, we examined whether aging in a given
tissue was associated with tissue-associated outcomes.
For instance, using data from approximately 700 post-
mortem samples from the Religious Order Study (ROS)
and the Memory and Aging Project (MAP) [39, 40] we
tested the association between pathologically diagnosed
Alzheimer’s disease and DNAm PhenoAge in dorsolate-
ral prefrontal cortex (DLPFX). Results suggest (Fig. 4)
that those who are diagnosed with Alzheimer’s disease
(AD), based on postmortem autopsy, have DLPFX that
appear more than one year older than same aged
individuals who are not diagnosed with AD postmortem
(p=4.6E-4). Further, age adjusted DNAm PhenoAge was
found to be positively associated with neuropathological
hallmarks of Alzheimer’s disease, such as amyloid load
(r=0.094, p=0.012), neuritic plaques (r=0.11, p=0.0032),
and neurofibrillary tangles (r=0.10, p=0.0073).
Comorbidity Disease Free CHD Risk Physical Functioning
Sample Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value
DNAm PhenoAge
WHI BA23 White 0.008 2.38E-01 -0.002 3.82E-01 0.016 5.36E-02 -0.396 1.04E-04
WHI BA23 Black 0.013 6.15E-02 -0.006 2.40E-02 0.021 2.02E-02 -0.423 4.50E-04
WHI BA23 Hispanic 0.024 1.64E-02 -0.004 3.67E-01 0.033 5.07E-02 -0.329 7.37E-02
WHI EMPC White 0.031 2.95E-07 -0.026 1.63E-02 0.023 1.89E-01 -0.361 3.81E-05
WHI EMPC Black 0.014 7.67E-02 -0.023 6.98E-02 0.048 2.27E-02 -0.473 3.75E-04
WHI EMPC Hispanic 0.003 7.83E-01 0.002 9.28E-01 0.073 1.98E-01 -0.377 6.54E-02
FHS 0.022 3.93E-07 -0.034 1.59E-03 0.028 5.47E-06 -0.016 4.60E-01
NAS 0.023 7.59E-06 -0.062 2.00E-04 0.030 2.27E-02 NA NA
JHS 0.018 1.86E-08 -0.039 5.92E-05 0.033 4.73E-02 NA NA
Meta P-value (Stouffer) 1.95E-20 2.14E-10 3.35E-11 2.05E-13
DNAmAge Hannum
WHI BA23 White 0.007 3.90E-01 -0.003 3.48E-01 0.013 2.36E-01 -0.399 2.90E-03
WHI BA23 Black 0.022 2.72E-02 -0.007 6.03E-02 0.015 2.67E-01 -0.345 4.29E-02
WHI BA23 Hispanic 0.010 4.33E-01 -0.010 6.24E-02 0.011 6.10E-01 -0.599 1.16E-02
WHI EMPC White 0.025 1.53E-03 -0.020 1.55E-01 0.022 3.30E-01 -0.284 1.43E-02
WHI EMPC Black 0.022 6.34E-02 -0.008 6.62E-01 0.055 6.12E-02 -0.323 9.56E-02
WHI EMPC Hispanic -0.012 4.17E-01 0.035 2.09E-01 -0.012 8.85E-01 -0.345 2.54E-01
FHS 0.019 5.94E-04 -0.030 2.55E-02 0.022 1.55E-02 0.040 1.32E-01
NAS 0.009 2.19E-01 -0.026 2.26E-01 0.025 1.83E-01 NA NA
JHS 0.020 2.09E-05 -0.036 9.91E-03 0.086 1.64E-04 NA NA
Meta P-value (Stouffer) 1.50E-08 1.64E-04 1.40E-05 2.03E-05
DNAmAge Horvath
WHI BA23 White 0.007 3.49E-01 -0.004 1.69E-01 0.001 9.12E-01 -0.440 5.10E-04
WHI BA23 Black 0.018 3.96E-02 -0.006 6.25E-02 0.009 4.07E-01 -0.305 4.52E-02
WHI BA23 Hispanic 0.012 3.65E-01 -0.007 1.86E-01 -0.001 9.78E-01 -0.204 4.12E-01
WHI EMPC White 0.031 1.99E-04 -0.043 5.56E-03 0.000 9.88E-01 -0.288 1.74E-02
WHI EMPC Black 0.016 1.93E-01 -0.003 8.56E-01 0.033 2.87E-01 -0.144 4.68E-01
WHI EMPC Hispanic -0.025 8.99E-02 -0.016 5.70E-01 -0.064 4.63E-01 -0.012 9.70E-01
FHS 0.011 5.82E-02 -0.021 8.34E-02 0.007 5.19E-01 0.027 3.16E-01
NAS 0.011 7.90E-02 -0.039 4.53E-02 0.006 7.14E-01 NA NA
JHS 0.014 2.03E-03 -0.040 1.78E-03 0.049 3.93E-02 NA NA
Meta P-value (Stouffer) 3.26E-06 6.36E-07 1.49E-01 1.43E-03
Lifestyle and demographic variables
In evaluating the relationship between DNAm
PhenoAge in blood and additional characteristics we
observe significant differences between racial/ethnic
groups (p=5.1E-5), with non-Hispanic blacks having the
highest DNAm PhenoAge on average, and non-
Hispanic whites having the lowest (Supplement 1: Fig.
S5). We also find evidence of social gradients in DNAm
PhenoAge, such that those with higher education
(p=6E-9) and higher income (p=9E-5) appear younger
(Figure 5). DNAm PhenoAge relates to exercise and
dietary habits, such that increased exercise (p=7E-5)
and markers of fruit/vegetable consumption (such as
carotenoids, p=2E-27) are associated with lower DNAm
PhenoAge (Figure 5, Supplement 1: Fig. S6). Cross-
sectional studies in the WHI also revealed that
DNAmPhenoAge acceleration is positively correlated
with C-reactive protein (r=0.18, p=5E-22, Figure 5),
insulin (r=0.15, p=2E-20), glucose (r=0.10, p=2E-10),
triglycerides (r=0.09, p=5E-9), waist to hip ratio
(r=0.15, p=5E-22) but it is negatively correlated with
HDL cholesterol (r=-0.09, p=7E-9).
DNAm PhenoAge and Immunosenescence
To test the hypothesis that DNAm PhenoAge captures
aspects of age-related decline of the immune system, we
correlated DNAm PhenoAge with estimated blood cell
count (Supplement 1, Fig. S7). After adjusting for age,
we find that DNAm PhenoAgeAccel is negatively
correlated with naïve CD8+ T cells (r=-0.35, p=9.2E-
65), naïve CD4+ T cells (r=-0.29, p=4.2E-42), CD4+
helper T cells (r=-0.34, p=3.6E-58), and B cells (r=-
0.18, p=8.4E-17). Further, DNAm PhenoAgeAccel is
positively correlated with the proportion of granulocytes
(r=0.32, p=2.3E-51), exhausted CD8+ (defined as
CD28-CD45RA-) T cells (r=0.20, p=1.9E-20), and
plasma blast cells (r=0.26, p=6.7E-34). These results are
consistent with age related changes in blood cells [41]
and suggest that DNAm PhenoAge may capture aspects
of immuno-senescence in blood. However, three lines
of evidence suggest that DNAm PhenoAge is not
simply a measure of immunosenescence. First, another
measure of immunosenescence, leukocyte telomere
length, is only weakly correlated with DNAm
PhenoAgeAccel (r=-0.13 p=0.00019 in the WHI; r=-
0.087, P=7.6E-3 in Framingham Heart study; JHS
p=7.83E-7, Supplement 1, Fig. S8). Second, the strong
association between DNAm PhenoAge and mortality
does not simply reflect changes in blood cell
composition, as can be seen from the fact that in
Supplement 1, Fig. S9 the robust association remains
even after adjusting for estimates of seven blood cell
count measures (Meta(FE)=1.036, Meta p=5.6E-21).
Third, DNAmPhenoAge correlates with chronological
age in non-blood tissue.
DNA sequence characteristics of the 513 CpGs in
DNAm PhenoAge
Of the 513 CpGs in DNAm PhenoAge, we find that, 41
CpGs were also in the Horvath DNAm age measure
(Supplement 2: Table S6). This represents a 4.88-fold
increase over what would be expected by chance
(p=8.97E-15). Of the 41 overlapping CpGs, the average
absolute value for their age correlations was r=0.40, and
31 had age correlations with absolute values in the top
20% of what is found among the 513 CpGs in the
DNAm PhenoAge score. We also observed 6 CpGs that
overlapped between the Hannum DNAm Age score and
the DNAm PhenoAge score—five of which were also
found in the Horvath DNAm Age measure. All six
CpGs had extremely high age correlations (half positive,
half negative), with absolute values between r=0.49 and
r=0.76. The five CpGs that are found in all three
epigenetic aging measures were: cg05442902 (P2RXL1),
cg06493994 (SCGN), cg09809672 (EDARADD),
cg19722847 (IPO8), and cg22736354 (NHLRC1).
Several additional DNAm biomarkers have been
described in the literature [12, 13]. A direct comparison
of 6 DNAm biomarkers (including DNAm PhenoAge)
reveals that DNAm PhenoAge stands out in terms of its
predictive accuracy for lifespan, its relationship with
smoking status, its relationship with leukocyte telomere
length, naïve CD8+ T cells and CD4+ T cells
(Supplement 1: Table S5).
Next, we conducted a functional enrichment analysis of
the chromosomal locations of the 513 CpGs and found
that 149 CpGs whose age correlation exceeded 0.2
tended to be located in CpG islands (p=0.0045, Supple-
ment 1: Fig. S10) and were significantly enriched with
polycomb group protein targets (p=8.7E-5, Supplement
1: Fig. S10), which in line with results of epigenome
wide studies of aging effects [4, 5, 42].
Transcriptional and genetic studies of DNAm
Using the genome-wide data from FHS and WHI, we
estimated the heritability of DNAm PhenoAge. The heri-
tability estimated by the SOLAR polygenic model for
those of European ancestry in the FHS was =0.33,
while the heritability estimated for those of European
ancestry in WHI, using GCTA-GREML analysis [43]
was =0.51.
Using the monocyte data mentioned above, as well as
PBMC expression data on 2,188 persons from the FHS,
we conducted a transcriptional analysis to identify dif-
ferential expression associated with DNAm PhenoAge-
Accel (Supplement 3: Table S7). Overall, we find that
genes show similar associations with chronological age
and DNAm PhenoAgeAccel. DNAm PhenoAgeAccel
represents aging differences among same-aged
individuals and is adjusted so as to exhibit a correlation
of r=0.0 with chronological age. Thus, this observation
suggests that genes whose transcription increases with
age are upregulated among epigenetically older
compared to epigenetically younger persons of the same
chronological age (Supplement 1: Fig. S11); the same
applies for genes that show decreases with chrono-
logical age being downregulated in epi-genetically older
versus younger persons of the same age.
Using the transcriptional data from monocytes
described above (adjusting for array, sex, race/ethnicity,
age, and imputed cell counts), we tested for GO
enrichment among genes that are positively associated
with DNAm PhenoAge and those that are negatively
associated with DNAm PhenoAge (Supplement 4:
Table S8). Among those with positive aging
associations (over-expression among epigenetically
older individuals), we observed enrichment for a
number of pro-inflammatory signaling pathways. These
pathways included, but are not limited to: multiple toll-
like receptor signaling pathways (7,9,3,2), regulation of
inflammatory response, JAK-STAT cascade, response
to lipopoly-saccharide, tumor necrosis factor-mediated
signaling pathway, and positive regulation of NF-
kappaB transcription factor activity. Additionally,
positively associated genes were also enriched for a
number anti-viral response pathways—type I interferon
signaling, defense response to virus, interferon-gamma-
mediated signaling pathway, cellular response to
interferon-alpha, etc. Other interesting GO terms
enriched among positively associated genes included:
response to nutrient, JAK-STAT cascade involved in
growth hormone signaling pathway, multicellular
organism growth, and regulation of DNA methylation.
When testing for enrichment among genes that were
negatively associated with DNAm PhenoAgeAccel
(decreased expression among epigenetically older
persons) we observed that many were implicated in
processes involving transcriptional and translational
machinery, as well as damage recognition and repair.
These included: translational initiation; regulation of
translational initiation; ribosomal large subunit assemb-
ly; ribosomal small subunit assembly; translational
elongation; transcription initiation from RNA
polymerase I promoter; transcription-coupled nucleo-
tide-excision repair; nucleotide-excision repair, DNA
incision, 5'-to lesion; nucleotide-excision repair, DNA
damage recognition; DNA damage response, detection
of DNA damage; and regulation of DNA damage
Using a novel two-step method, we were successful in
developing a DNAm based biomarker of aging that is
highly predictive of nearly every morbidity and mortali-
ty outcome we tested. Training an epigenetic predictor
of phenotypic age instead of chronological age led to
substantial improvement in mortality/healthspan
predictions over the first generation of DNAm based
biomarkers of chronological age from Hannum[10],
Horvath[11] and other published DNAm biomarkers. In
doing so, this is the first study to conclusively demons-
trate that DNAm biomarkers of aging are highly
predictive of cardiovascular disease and coronary heart
disease. DNAm PhenoAge also tracks chronological
age and relates to disease risk in samples other than
whole blood. Finally, we find that an individual’s
DNAm PhenoAge, relative to his/her chronological age,
is moderately heritable and is associated with activation
of pro-inflammatory, interferon, DNAm damage repair,
transcriptional/ translational signaling, and various
markers of immuno-senescence: a decline of naïve T
cells and shortened leukocyte telomere length
(Supplementary Information).
The ability of our measure to predict multifactorial
aging conditions is consistent with the fundamental
underpinnings of Geroscience research [1, 44], which
posits that aging mechanisms give rise to multiple
pathologies and thus, differences in the rate of aging
will have implications for a wide array of diseases and
conditions. Further, these results answer a fundamental
biological question of whether differences in multi-
system dysregulation (estimated using clinical
phenotypic age measures), healthspan, and lifespan are
reflected at the epigenetic level, in the form of
differential DNAm at specific CpG sites.
The improvement over previous epigenetic biomarkers,
likely comes down to the types of CpGs selected for the
various measures. Only 41 of the 513 CpGs in DNAm
PhenoAge were shared with the Horvath clock, while
only five CpGs were shared between all three clocks
(DNAm PhenoAge, Horvath, and Hannum). In general,
these CpGs did not tend to be drivers of the DNAm
PhenoAge score, and instead represented those with
large age correlations. This may explain the
improvements of DNAm PhenoAge over previous
epigenetic biomarkers of aging. While the previous
DNAm age estimators selected CpGs to optimize
prediction of chronological age, the CpGs in DNAm
PhenoAge were optimized to predict a multi-system
proxy of physiological dysregulation (phenotypic age).
In doing so, we were able to not only capture CpGs that
exhibited strong correlations with age, but also those
that captured variations in risk of death and disease
among same aged individuals. In general, the CpGs
with the highest weights in the new clock did not
correlate with chronological age (Supplement 1: Fig.
S12), but instead were related to the difference between
phenotypic and chronological age—i.e. divergence in
the rate of aging.
While DNAm PhenoAge greatly outperformed all
previous DNAm biomarkers of aging (Supplement 1:
Table S5), the utility of DNAm PhenoAge for
estimating risk does not imply that it should replace
clinical biomarkers when it comes to informing medical
and health-related decisions. In fact, but perhaps not
surprisingly, the phenotypic age measure used to select
CpGs is a better predictor of morbidity and mortality
outcomes than DNAm PhenoAge. While the addition of
error in performing a two-step process, rather than
training a DNAm predictor directly on mortality may
contribute, we don’t believe this accounts for the
difference in predictive performance. In fact, a recent
DNAm measure by Zhang et al. [38] was trained to
directly predict mortality risk, yet it appears to be a
weaker predictor than both our DNAm PhenoAge
measure and our clinical phenotypic age measure
(Supplement 1: Table S9). The first generation of
DNAm age estimators only exhibit weak associations
with clinical measures of physiological dysregulation
[24, 45]. Physiological dysregulation, which is more
closely related to our clinical age measure “phenotypic
age” than to chronological age, is not only the result of
exogenous/endogenous stress factors (such as obesity,
infections) but also a result of age related molecular
alterations, one example of which are modifications to
the epigenome. Over time, dysregulation within organ
systems leads to pathogenesis of disease (age-related
molecular changes Æ physiological dysregulation Æ
morbidity Æ mortality)[46]. However, stochasticity and
variability exist at each of these transitions. Therefore,
measures of physiological dysregulation, will be better
predictors of transition to the next stage in the aging
trajectory (i.e. morbidity and mortality) than will
measures of age related molecular alterations, like
DNAm PhenoAge. Similarly, quantification of disease
pathogenesis (cancer stage, Alzheimer’s stage) is likely
a better predictor of mortality risk than clinical
phenotypic aging measures. As a result, clinical pheno-
typic aging measures may be preferable to epigenetic
measures when the goal is risk prediction, and samples
come from blood.
That being said, when the aim is to study the
mechanisms of the aging process, DNAm measures
have advantages over clinical measures. First, they may
better capture “pre-clinical aging” and thus may be
more suited for differentiating aging in children, young
adults, or extremely healthy individuals, for whom
measures like CRP, albumin, creatinine, glucose, etc.
are still fairly homogenous. Second, as demonstrated,
these molecular measures can capture cell and/or tissue
specific aging rates and therefore may also lend
themselves to in vitro studies of aging, studies for which
blood is not available, studies using postmortem
samples, and/or studies comparing aging rates between
tissues/cells. While the fundamental drivers of aging are
believed to be shared across cells/tissues, that is not to
say that all the cells and tissues within an individual will
age at the same rate. In fact, it is more likely that
individuals will vary in their patterning of aging rates
across tissues, and that this will have implications for
death and disease risk. Relatedly, it is not known how
predictions based on DNAm PhenoAge measures from
non-blood samples will compare to phenotypic age
predictions. It may be the case that various outcomes
will be more tightly related to aging in specific
cells/tissues, rather than blood. Finally, examination of
DNAm based aging rates facilitates the direct study of
the proposed mechanisms of aging, of which
“epigenetic alterations” is one of the seven hypo-
thesized “pillars of aging” [1].
While more work needs to be done to model the biology
linking DNAm PhenoAge and aging outcomes, we
began to explore this using differential expression,
functional enrichment, and heritability estimates.
Overall, we found that CpGs that had larger increases
with aging tended to be located in CpG islands and
enriched with polycomb group protein targets,
consistent with what has been reported in previous
epigenome wide studies of aging effects [4-7, 42].
While typically DNAm of CpG islands and/or
polycomb recruitment is linked to transcriptional
silencing [47], for the most part, we did not observe
associations between DNAm and expression for co-
locating CpG-gene pairs—this was also true when only
considering CpGs located in islands. These findings
may suggest that the genes annotated to the CpGs in our
score are not part of the link between changes in DNAm
and aging. Nevertheless, we also recognize that these
null results could stem from the fact that 1) associations
were only tested in monocytes, 2) DNAm and
expression represents what is present globally for each
sample, rather than on a cell-by-cell basis, and 3)
stronger associations between DNAm and gene
expression levels may only exist early in life.
Nevertheless, we do identify potentially promising
transcriptional pathways when considering DNAm
PhenoAge as a whole. For instance, we observe that
higher DNAm PhenoAge is associated with increases in
the activation of proinflammatory pathways, such as
NF-kappaB; increased interferon (IFN) signaling;
decreases in ribosomal–related and translational
machinery pathways; and decreases in damage
recognition and repair pathways. These findings are
consistent with previous work describing aging
associated changes, comprising increases in dysregulat-
ed inflammatory activation, increased DNA damage,
and loss of translational fidelity. For instance, there
exists a large body of literature highlighting the
importance of an increased low-grade pro-inflammatory
status as a driver of the aging process, termed inflamm-
aging [41, 48, 49]. IFN signaling pathways have been
shown to be markers of DNA damage and mediators of
cellular senescence[50]. Additionally, it has been shown
that breakdown of the transcriptional and translational
machinery may play a central role in the aging process
[51, 52]. For instance, the ribosome is believed to be a
key regulator of proteostasis, and in turn, aging [51, 53].
Relatedly, loss of integrity in DNA damage repair
pathways is considered another hallmark of the aging
process [54-56].
In general, many of these pathways will have implica-
tions for adaptation to exogenous and endogenous
stressors. Factors related to stress resistance and
response have repeatedly been shown to be drivers of
differences in lifespan and aging [49, 57-61]. This may
partially account for our findings related to smoking. In
general, it is not surprising that a biomarker of aging
and mortality risk relates to smoking, given that life
expectancies of smokers are on average ten years shorter
than never smokers, and smoking history is associated
with a drastic increase in the risk of a number of age-
related conditions. However, perhaps more interesting-
ly, we find that the effects of DNAm PhenoAge on
mortality appear to be higher for smokers than non-
smokers, which could suggest that DNAm PhenoAge
represent differences in innate resilience/ vulnerability to
pro-aging stressors, such as cigarette smoke.
Interestingly, we observed moderately high heritability
estimates for DNAm PhenoAge. For instance, we
estimated that genetic differences accounted for one-
third to one-half of the variance in DNAm PhenoAge,
relative to chronological age. In moving forward, it will
be useful to identify the genetic architecture underlying
differences in epigenetic aging. Finally, we reported
that individuals’ DNAm PhenoAges—relative to their
chronological ages—remained fairly stable over a nine-
year period. However, it is unclear whether it is
attributable to genetic influences, or the fact that social
and behavioral characteristics tend to also remain stable
for most individuals.
If the goal is to utilize accurate quantifiable measures of
the rate of aging, such as DNAm PhenoAge, to assess
the efficacy of aging interventions, more work will be
needed to evaluate the dynamics of DNAmPhenoAge
following various treatments. For instance, it remains to
be seen whether interventions can reverse
DNAmPhenoAge in the short term. Along these lines, it
will be essential to determine causality—does DNAm
drive the aging process, or is it simply a surrogate
marker of organismal senescence? If the former is true,
DNAm PhenoAge could provide insight into promising
targets for therapies aimed at lifespan, and more
importantly, healthspan extension.
Overall, DNAm PhenoAge is an attractive composite
biomarker that captures organismal age and the func-
tional state of many organ systems and tissues, above
and beyond what is explained by chronological time.
Our validation studies in multiple large and independent
cohorts demonstrate that DNAm PhenoAge is a highly
robust predictor of both morbidity and mortality
outcomes, and represents a promising biomarker of
aging, which may prove to be beneficial to both basic
science and translational research.
Using the NHANES training data, we applied a Cox
penalized regression model—where the hazard of
aging-related mortality (mortality from diseases of the
heart, malignant neoplasms, chronic lower respiratory
disease, cerebrovascular disease, Alzheimer’s disease,
Diabetes mellitus, nephritis, nephrotic syndrome, and
nephrosis) was regressed on forty-two clinical markers
and chronological age to select variables for inclusion in
our phenotypic age score. Ten-fold cross-validation was
employed to select the parameter value, lambda, for the
penalized regression. In order to develop a sparse
parsimonious phenotypic age estimator (fewer
biomarker variables preferred to produce robust results)
we selected a lambda of 0.0192, which represented a
one standard deviation increase over the lambda with
minimum mean-squared error during cross-validation
(Supplement 1, Fig. S13). Of the forty-two biomarkers
included in the penalized Cox regression model, this
resulted in ten variables (including chronological age)
that were selected for the phenotypic age predictor.
These nine biomarkers and chronological age were then
included in a parametric proportional hazards model
based on the Gompertz distribution. Based on this
model, we estimated the 10-year (120 months)
mortality risk of the j-the individual. Next, the
mortality score was converted into units of years
(Supplement 1). The resulting phenotypic age estimate
was regressed on DNA methylation data using an
elastic net regression analysis. The penalization
parameter was chosen to minimize the cross validated
mean square error rate (Supplement 1, Fig. S14), which
resulted in 513 CpGs.
Estimation of blood cell counts based on DNAm
We estimate blood cell counts using two different
software tools. First, Houseman's estimation method
[62] was used to estimate the proportions of CD8+ T
cells, CD4+ T, natural killer, B cells, and granulocytes
(mainly neutrophils). Second, the Horvath method,
implemented in the advanced analysis option of the
epigenetic clock software [11, 18], was used to estimate
the percentage of exhausted CD8+ T cells (defined as
CD28-CD45RA-), the number (count) of naïve CD8+ T
cells (defined as CD45RA+CCR7+) and plasmablasts.
We and others have shown that the estimated blood cell
counts have moderately high correlations with
corresponding flow cytometric measures [62, 63].
Additional descriptions of methods and materials can be
found in Supplement 1.
Ethics approval
This study was reviewed by the UCLA institutional
review board (IRB#13-000671, IRB#15-000697,
IRB#16-001841, IRB#15-000682).
Availability of data and materials
The WHI data are available at dbGaP under the
accession numbers phs000200.v10.p3. The FHS data
are available at dbGaP under the accession numbers
phs000342 and phs000724. The Normative Aging data
are available from dbGAP phs000853.v1.p1. The
Jackson Heart Study data are available from
https: //
ML and SH developed the DNAmPhenoAge estimator
and wrote the article. SH, ML, LF conceived of the
study. ML, SH, AL, AQ carried out the statistical
analysis. The remaining authors contributed data and
participated in the interpretation of the results.
We would like to acknowledge The WHI Investigators
listed below:
Program Office: (National Heart, Lung, and Blood
Institute, Bethesda, Maryland) Jacques Rossouw, Shari
Ludlam, Dale Burwen, Joan McGowan, Leslie Ford,
and Nancy Geller.
Clinical Coordinating Center: (Fred Hutchinson Cancer
Research Center, Seattle, WA) Garnet Anderson, Ross
Prentice, Andrea LaCroix, and Charles Kooperberg.
Investigators and Academic Centers: (Brigham and
Women's Hospital, Harvard Medical School, Boston,
MA) JoAnn E. Manson; (MedStar Health Research
Institute/Howard University, Washington, DC) Barbara
V. Howard; (Stanford Prevention Research Center,
Stanford, CA).
Marcia L. Stefanick; (The Ohio State University,
Columbus, OH) Rebecca Jackson; (University of
Arizona, Tucson/Phoenix, AZ) Cynthia A. Thomson;
(University at Buffalo, Buffalo, NY).
Jean Wactawski-Wende; (University of Florida,
Gainesville/Jacksonville, FL) Marian Limacher;
(University of Iowa, Iowa City/Davenport, IA) Robert
Wallace; (University of Pittsburgh, Pittsburgh, PA)
Lewis Kuller; (Wake Forest University School of
Medicine, Winston-Salem, NC) Sally Shumaker.
The Framingham Heart Study is funded by National
Institutes of Health contract N01-HC-25195 and
HHSN268201500001I. The laboratory work for this
investigation was funded by the Division of Intramural
Research, National Heart, Lung, and Blood Institute,
National Institutes of Health. The analytical component
of this project was funded by the Division of Intramural
Research, National Heart, Lung, and Blood Institute,
and the Center for Information Technology, National
Institutes of Health, Bethesda, MD. The Framingham
Heart Study is conducted and supported by the National
Heart, Lung, and Blood Institute (NHLBI) in
collaboration with Boston University. This manuscript
was not prepared in collaboration with investigators of
the Framingham Heart Study and does not necessarily
reflect the opinions or views of the Framingham Heart
Study, Boston University, or the NHLBI.
The United States Department of Veterans Affairs (VA)
Normative Aging Study (NAS) is supported by the
Cooperative Studies Program/ERIC and is a research
component of the Massachusetts Veterans Epidemio-
logy Research and Information Center (MAVERIC),
Boston Massachusetts.
The MESA Epigenomics and Transcriptomics Studies
were funded by R01HL101250, R01 DK103531-01,
R01 DK103531, R01 AG054474, and R01 HL135009-
01 to Wake Forest University Health Sciences.
We thank the Jackson Heart Study (JHS) participants
and staff for their contributions to this work. The JHS is
supported by contracts HHSN268201300046C,
HHSN268201300047C, HHSN268201300048C,
HHSN268201300049C, HHSN268201300050C from
the National Heart, Lung, and Blood Institute and the
National Institute on Minority Health and Health
Disparities. Dr. Wilson is supported by U54GM115428
from the National Institute of General Medical
The Regents of the University of California is the sole
owner of a provisional patent application directed at this
invention for which MEL, SH are named inventors.
This study was supported by NIH/NIA U34AG051425-
01 (Horvath) and NIH/NIA K99AG052604 (Levine).
The WHI epigenetic studies were supported by
NIH/NHLBI 60442456 BAA23 (Assimes, Absher,
Horvath) and by the National Institute of Environmental
Health Sciences R01-ES020836 WHI-EMPC (Whitsel,
Baccarelli, Hou). The WHI program is funded by the
National Heart, Lung, and Blood Institute, National
Institutes of Health, U.S. Department of Health and
Human Services through contracts
HHSN268201100046C, HHSN268201100001C,
HHSN268201100002C, HHSN268201100003C,
HHSN268201100004C, and HHSN271201100004C.
The authors thank the WHI investigators and staff for
their dedication, and the study participants for making
the program possible. A full listing of WHI
investigators can be found at:
The InCHIANTI study baseline (1998-2000) was
supported as a "targeted project" (ICS110.1/RF97.71)
by the Italian Ministry of Health and in part by the U.S.
National Institute on Aging (Contracts: 263 MD 9164
and 263 MD 821336).
Funding for the DNA methylation in JHS provided by
NHLBI R01HL116446.
The funding bodies played no role in the design, the
collection, analysis, or interpretation of the data.
1. KennedyBK,BergerSL,BrunetA,CampisiJ,Cuervo
2. BurchJB,AugustineAD,FriedenLA,HadleyE,
3. FragaMF,EstellerM.Epigeneticsandaging:the
4. RakyanVK,DownTA,MaslauS,AndrewT,YangTP,
5. TeschendorffAE,MenonU,GentryMaharajA,Ramus
6. JungM,PfeiferGP.AgingandDNAmethylation.BMC
8. BocklandtS,LinW,SehlME,SánchezFJ,Sinsheimer
9. GaragnaniP,BacaliniMG,PirazziniC,GoriD,Giuliani
10. HannumG,GuinneyJ,ZhaoL,ZhangL,HughesG,
11. HorvathS.DNAmethylationageofhumantissues
12. WeidnerCI,LinQ,KochCM,EiseleL,BeierF,Ziegler
13. LinQ,WeidnerCI,CostaIG,MarioniRE,FerreiraMR,
14. HorvathS,ErhartW,BroschM,AmmerpohlO,von
16. MarioniRE,ShahS,McRaeAF,RitchieSJ,Muniz
17. HorvathS,GaragnaniP,BacaliniMG,PirazziniC,
18. HorvathS,LevineAJ.HIV1InfectionAcceleratesAge
19. HorvathS,PirazziniC,BacaliniMG,GentiliniD,Di
20. LevineME,HosgoodHD,ChenB,AbsherD,AssimesT,
21. LevineME,LuAT,BennettDA,HorvathS.Epigenetic
22. LevineME,LuAT,ChenBH,HernandezDG,Singleton
23. ChenBH,MarioniRE,ColicinoE,PetersMJ,Ward
24. QuachA,LevineME,TanakaT,LuAT,ChenBH,
25. DuguePA,BassettJK,JooJE,JungCH,MingWongE,
26. SimpkinAJ,HoweLD,TillingK,GauntTR,LyttletonO,
27. MaierhoferA,FlunkertJ,OshimaJ,MartinGM,Haaf
28. LevineME.Modelingtherateofsenescence:can
29. BelskyDW,CaspiA,HoutsR,CohenHJ,CorcoranDL,
30. LiS,WongEM,JooJE,JungCH,ChungJ,ApicellaC,
31. SebastianiP,ThyagarajanB,SunF,SchupfN,
32. FerrucciL,HesdorfferC,BandinelliS,SimonsickEM.
33. ChristiansenL,LenartA,TanQ,VaupelJW,AvivA,
34. PernaL,ZhangY,MonsU,HolleczekB,SaumKU,
35. BreitlingLP,YangR,KornB,BurwinkelB,BrennerH.
36. JoehanesR,JustAC,MarioniRE,PillingLC,Reynolds
37. LeeKW,PausovaZ.CigarettesmokingandDNA
38. ZhangY,WilsonR,HeissJ,BreitlingLP,SaumKU,
39. BennettDA,SchneiderJA,ArvanitakisZ,WilsonRS.
40. BennettDA,SchneiderJA,BuchmanAS,BarnesLL,
41. FranceschiC,BonafèM,ValensinS,OlivieriF,DeLuca
42. HorvathS,ZhangY,LangfelderP,KahnRS,BoksMP,
43. YangJ,LeeSH,GoddardME,VisscherPM.GCTA:a
44. SierraF,KohanskiR.GeroscienceandthetransNIH
45. BelskyDW,MoffittTE,CohenAA,CorcoranDL,Levine
46. KubbenN,MisteliT.Sharedmolecularandcellular
47. MerloA,HermanJG,MaoL,LeeDJ,GabrielsonE,
48. FinchCE.CHAPTER1‐InflammationandOxidationin
49. FinchCE.EvolutioninhealthandmedicineSackler
50. MoiseevaO,MalletteFA,MukhopadhyayUK,Moores
51. GonskikhY,PolacekN.Alterationsofthetranslation
apparatus during aging and stress response. Mech
52. WoltersS,SchumacherB.Genomemaintenanceand
53. SteffenKK,DillinA.ARibosomalPerspectiveon
54. GarinisGA,UittenboogaardLM,StachelscheidH,
55. SchumacherB,GarinisGA,HoeijmakersJH.Ageto
56. McMurrayCT,VijgJ.Editorialoverview:Molecular
57. LevineME,CrimminsEM.AGeneticNetwork
58. ZhouKI,PincusZ,SlackFJ.Longevityandstressin
59. LinYJ,SeroudeL,BenzerS.Extendedlifespanand
60. YuBP,ChungHY.Stressresistancebycaloric
61. MartinGM,AustadSN,JohnsonTE.Geneticanalysis
62. HousemanEA,AccomandoWP,KoestlerDC,
63. HorvathS,GurvenM,LevineME,TrumbleBC,Kaplan
Please browse the Full Text version of this manuscript
to see Supplementary Methods, Tables, and Figures
presented in Supplements 1-4.
... 11 This approach is somewhat analogous to older tools like telomere length, which captured the aging process by measuring the length of telomere "caps" on the ends of chromosomes. 9 Although DNA methylation is a normal biological process integral to the regulation of gene transcription and cellular activity, 15 the development of epigenetic clocks such as those by Horvath, 16 Levine,17 and Lu 18 highlight the ability of DNA methylation to correlate with the aging process in a regular and predictable manner across populations. Epigenetic clocks derived from DNAm measures are also robustly predictive of health and mortality across the lifespan, as individuals with an estimated biological age greater than their chronological age (termed "age acceleration") are at increased risk for disease and death. ...
... Epigenetic clocks derived from DNAm measures are also robustly predictive of health and mortality across the lifespan, as individuals with an estimated biological age greater than their chronological age (termed "age acceleration") are at increased risk for disease and death. [16][17][18][19] There is robust evidence that toxic environmental exposures such as Pb [20][21][22][23] and Hg 24,25 have diverse impacts on the epigenome. 2,13,14 Estimating the effects of Pb and Hg on epigenetic aging provides an opportunity to understand their impacts on biological aging and laterlife mortality prior to the observance of chronic disease and mortality in the population, which can take decades to manifest. ...
... 37 Horvath Age is highly correlated with chronological age in multiple tissue types 16 but is a poorer marker of physiologic dysfunction than PhenoAge and GrimAge. 11,19 PhenoAge and GrimAge were developed using CpGs present on both the Ilumina 450K and 850K platforms in combination with other blood-based biomarkers, 17,18 and both are robustly predictive of physiologic decline and all-cause mortality. 11,[17][18][19] All three metrics of DNAm age are strongly correlated with chronological age (Supplemental Figure 1) and each other (Supplemental We employed two different modeling strategies to investigate the effects of Pb, Hg, Mn, and Cu on DNAm age acceleration. ...
Full-text available
Although the effects of lead, mercury, manganese, and copper on individual disease processes are well understood, estimating the health effects of long-term exposure to these metals at the low concentrations often observed in the general population is difficult. In addition, the health effects of joint exposure to multiple metals are difficult to estimate. Biological aging refers to the integrative progression of multiple physiologic and molecular changes that make individuals more at risk of disease. Biomarkers of biological aging may be useful to estimate the population-level effects of metal exposure prior to the development of disease in the population. We used data from 290 participants in the Detroit Neighborhood Health Study to estimate the effect of serum lead, mercury, manganese, and copper on three DNA methylation-based biomarkers of biological aging (Horvath Age, PhenoAge, and GrimAge). We used mixed models and Bayesian kernel machine regression and controlled for participant sex, race, ethnicity, cigarette use, income, educational attainment, and block group poverty. We observed consistently positive estimates of effect between lead and GrimAge acceleration and mercury and PhenoAge acceleration. In contrast, we observed consistently negative associations between manganese and PhenoAge acceleration and mercury and Horvath Age acceleration. We also observed curvilinear relationships between copper and both PhenoAge and GrimAge acceleration. Increasing total exposure to the observed mixture of metals was associated with increased PhenoAge and GrimAge acceleration and decreased Horvath Age acceleration. These findings indicate that an increase in serum lead or mercury from the 25th to 75th percentile is associated with an approximate 0.25-year increase in two epigenetic markers of all-cause mortality in a population of adults in Detroit, Michigan. While few findings were statistically significant, their consistency and novelty warrant interest.
... "Second-generation" epigenetic clocks, like PhenoAge (18) and GrimAge (19), integrate data from nine clinical biomarkers (e. g., white blood cell count, C-reactive protein, lymphocyte percentage, albumin, creatinine, etc.) and 513 CpG associated with mortality. The CpG component site of GrimAge is a surrogate for disease-related and health-related proteins and smoking history, it shows a high association with all-cause mortality and age-related health conditions and has a good ability to predict both morbidity rates and mortality (20). ...
... Recent genome-wide association studies based on a metaanalysis of European ancestry of 34,710 participants from 28 cohorts, identify 137 loci for DNA biomarkers related to aging 1 ( Table 1). From this study, we obtained summary genetic association estimates for epigenetic age acceleration measures of HannumAge (16), Intrinsic HorvathAge (17), PhenoAge (18), and GrimAge (19). The analysis included 28 European ancestry studies with 57.3% female participants. ...
Full-text available
Observational data from China, the United States, France, and Italy suggest that chronological age is an adverse COVID-19 outcome risk factor, with older patients having a higher severity and mortality rate than younger patients. Most studies have gotten the same view. However, the role of aging in COVID-19 adverse effects is unclear. To more accurately assess the effect of aging on adverse COVID-19, we conducted this bidirectional Mendelian randomization (MR) study. Epigenetic clocks and telomere length were used as biological indicators of aging. Data on epigenetic age (PhenoAge, GrimAge, Intrinsic HorvathAge, and HannumAge) were derived from an analysis of biological aging based on genome-wide association studies (GWAS) data. The telomere length data are derived from GWAS and the susceptibility and severity data are derived from the COVID-19 Host Genetics Initiative (HGI). Firstly, epigenetic age and telomere length were used as exposures, and following a screen for appropriate instrumental variables, we used random-effects inverse variance weighting (IVW) for the main analysis, and combined it with other analysis methods (e.g., MR Egger, Weighted median, simple mode, Weighted mode) and multiple sensitivity analysis (heterogeneity analysis, horizontal multiplicity analysis, “leave-one-out” analysis). For reducing false-positive rates, Bonferroni corrected significance thresholds were used. A reverse Mendelian randomization analysis was subsequently performed with COVID-19 susceptibility and severity as the exposure. The results of the MR analysis showed no significant differences in susceptibility to aging and COVID-19. It might suggest that aging is not a risk factor for COVID-19 infection (P-values are in the range of 0.05–0.94). According to the results of our analysis, we found that aging was not a risk factor for the increased severity of COVID-19 (P > 0.05). However, severe COVID-19 can cause telomere lengths to become shorter (beta = −0.01; se = 0.01; P = 0.02779). In addition to this, severe COVID-19 infection can slow the acceleration of the epigenetic clock “GrimAge” (beta = −0.24, se = 0.07, P = 0.00122), which may be related to the closely correlation of rs35081325 and COVID-19 severity. Our study provides partial evidence for the causal effects of aging on the susceptibility and severity of COVID-19.
... We and others have shown that DNA methylation (DNAm) is also substantially altered as a direct function of cell division [6][7][8][9]. Further, the epigenome has been shown to undergo dramatic changes with aging and is implicated in establishing, driving and maintaining many cancers [10][11][12][13]. Coincidently, the DNA methylation changes observed in aging, cancer, and proliferation share some notable patterns. ...
... The following PCs were selected from each module: Yellow [PC1, PC2, PC3, PC10, PC17] and Tan [PC1, PC2, PC3, PC4, PC5, PC7, PC10, PC11]. Further details on PC-trained measures can be found in our previous reports [5,11]. A package is under construction for easy calculation of CellDRIFT from external datasets. ...
Full-text available
Aging is the leading risk factor for cancer. While it’s been proposed that the age-related accumulation of somatic mutations drives this relationship, it is likely not the full story. Here, we show that both aging and cancer share a common epigenetic replication signature, which we modeled from DNA methylation data in extensively passaged immortalized human cells in vitro and tested on clinical tissues. This epigenetic signature of replication – termed CellDRIFT – increased with age across multiple tissues, distinguished tumor from normal tissue, and was escalated in normal breast tissue from cancer patients. Additionally, within-person tissue differences were correlated with both predicted lifetime tissue-specific stem cell divisions and tissue-specific cancer risk. Overall, our findings suggest that age-related replication drives epigenetic changes in cells, pushing them towards a more tumorigenic state. One sentence summary Cellular replication leaves an epigenetic fingerprint that may partially underly the age-associated increase in cancer risk.
... Epigenetic clocks are mathematical algorithms which can, with great precision, predict the chronological age of subjects by combining measured DNAm levels at multiple CpG sites of their genome (Horvath & Raj, 2018). There is now clear evidence of the association between alterations in the predictions made by epigenetic clocks and lifestyle factors, disease-including cancer-or outright mortality (Fransquet et al., 2019;Levine et al., 2018;Lu et al., 2019;Marioni et al., 2015;Oblak et al., 2021), indicating that these algorithms capture a combination of chronological and biological age, the latter being a measure of the "healthiness" of an individual in terms of their risk of developing age-associated adverse outcomes (Jylhävä et al., 2017). For instance, supercentenarian or long-lived subjects, who display reduced incidence, or delayed onset, of diseases (Andersen et al., 2012) also manifest younger epigenetic ages (Armstrong et al., 2017;Horvath et al., 2015). ...
... With regards to the alterations-acceleration or decelerationobserved in the clocks, various associations have been investigated, particularly the aforementioned links with environmental factors and disease (Fransquet et al., 2019;Levine et al., 2018;Lu et al., 2019;Marioni et al., 2015;Oblak et al., 2021), to ascertain whether the DNAm alterations which occur during these processes could drive the changes observed in the clocks. For instance, a recent study investigating DNAm age acceleration in the non-tumoral breast tissue of breast cancer patients demonstrated that the observed alteration in the epigenetic clock could be explained by a subset of the clock CpGs that suffer from the well-known cancerassociated hypermethylation of Polycomb associated loci (Rozenblit et al., 2022). ...
Full-text available
Aging and cancer are clearly associated processes, at both the epidemiological and molecular level. Epigenetic mechanisms are good candidates to explain the molecular links between the two phenomena, but recent reports have also revealed considerable differences, particularly regarding the loss of DNA methylation in the two processes. The large‐scale generation and availability of genome‐wide epigenetic data now permits systematic studies to be undertaken which may help clarify the similarities and differences between aging and cancer epigenetic alterations. In addition, the development of epigenetic clocks provides a new dimension in which to investigate diseases at the molecular level. Here, we examine current and future questions about the roles of DNA methylation mechanisms as causal factors in the processes of aging and cancer so that we may better understand if and how aging‐associated epigenetic alterations lead to tumorigenesis. It seems certain that comprehending the molecular mechanisms underlying epigenetic clocks, especially with regard to somatic stem cell aging, combined with applying single‐cell epigenetic‐age profiling technologies to aging and cancer cohorts, and the integration of existing and upcoming epigenetic evidence within the genetic damage models of aging will prove to be crucial to improving understanding of these two interrelated phenomena. Aging and cancer are interrelated processes which share common epigenetic alterations. Nonetheless, there are differences in some of the DNA methylation changes which occur in both phenomena. The recent development of epigenetic clocks will help dissect the common and specific epigenetic characteristics of aging and cancer, although the mechanisms underlying epigenetic clocks are yet to be clarified, particularly in relation to somatic stem cell epigenetic aging.
... PD is not associated with accelerated epigenetic aging Over the past decade, recognition has been growing that the epigenome, especially DNA methylation, can be used to closely predict chronological age 41,[81][82][83] . Based on this, researchers have established epigenetic clocks that use loci with the most predictable DNA methylation levels to provide estimates of epigenetic (or biological) age 81 . ...
Full-text available
Evidence for epigenetic regulation playing a role in Parkinson’s disease (PD) is growing, particularly for DNA methylation. Approximately 90% of PD cases are due to a complex interaction between age, genes, and environmental factors, and epigenetic marks are thought to mediate the relationship between aging, genetics, the environment, and disease risk. To date, there are a small number of published genome-wide studies of DNA methylation in PD, but none accounted for cell type or sex in their analyses. Given the heterogeneity of bulk brain tissue samples and known sex differences in PD risk, progression, and severity, these are critical variables to account for. In this genome-wide analysis of DNA methylation in an enriched neuronal population from PD postmortem parietal cortex, we report sex-specific PD-associated methylation changes in PARK7 (DJ-1), SLC17A6 (VGLUT2), PTPRN2 (IA-2β), NR4A2 (NURR1), and other genes involved in developmental pathways, neurotransmitter packaging and release, and axon and neuron projection guidance.
... DNA Methylation Phenotypic Age (DNAm PhenoAge) was used to calculate BA due to its higher reported accuracy. 9 DNAm PhenoAge was run in R using the methyl package at 10 After calculation, CpG sites that were statistically different between nondiabetic and T2D samples were matched to gene symbols, and Pathway analysis was completed using Reactome. ...
Full-text available
Background: Biological age (BA) closely depicts age-related changes at a cellular level. Type 2 Diabetes mellitus (T2D) accelerates BA when calculated using clinical biomarkers. However, there is a large spread of individual BA within these groups and it is unclear what clinical biomarkers correlate with different speeds of aging and whether pharmacological treatment of diabetes alter BA. We hypothesized that accelerated BA would be seen at the DNA methylation (DNAm) level, the gold standard to determine BA, and biomarkers and treatments would correlate the rate of BA in T2D. Methods: Publicly available DNAm samples were obtained from the GEO NCBI database and the NHANES 2017-2018 and ACCORD Cohorts were used for our analysis. We used the DNA Methylation Phenotypic Age algorithm and the Klemera and Doubal (KDM) methods to calculate BA with DNA methylation and clinical biomarkers, respectively. Results: DNAm showed increased BA in whole blood and pancreatic islets in T2D in aging-related pathways, such as DNA damage and inflammation. Using the NHANES and ACCORD Trial cohorts, we found that avoidance of fried and fatty foods, and vigorous activity correlated with decreased BA in T2D. Cardiovascular, glycemic, and inflammatory biomarkers associated with the rate of aging in DM. Intensive blood pressure and T2D treatment associated with a greater deceleration in the speed of aging as compared to the standard. Conclusions: Overall, we show that certain tissues age faster in people with T2D and this strongly associates with blood glucose control, inflammation and cardiovascular health. Effective treatment of the disease can decelerate aging and decrease BA suggesting the latter as a novel and integrated index to evaluate and follow people with T2D. Funding: This study was supported by Institutional Startup Funds to C.A.M. (Joslin Diabetes Center) and NIH grants P30 DK036836 Joslin Diabetes Research Center (Bioinformatic Core).
Biomarkers defining biological age are typically laborious or expensive to assess. Instead, in the current study, we identified parameters based on standard laboratory blood tests across metabolic, cardiovascular, inflammatory, and kidney functioning that had been assessed in the Berlin Aging Study (BASE) (n = 384) and Berlin Aging Study II (BASE-II) (n = 1517). We calculated biological age using those 12 parameters that individually predicted mortality hazards over 26 years in BASE. In BASE, older biological age was associated with more physician-observed morbidity and higher mortality hazards, over and above the effects of chronological age, sex, and education. Similarly, in BASE-II, biological age was associated with physician-observed morbidity and subjective health, over and above the effects of chronological age, sex, and education as well as alternative biomarkers including telomere length, DNA methylation age, skin age, and subjective age but not PhenoAge. We discuss the importance of biological age as one indicator of aging.
Social determinants of health (SDoH) are defined as the conditions in which people are born, grow, live, work, and age. The distribution of these conditions is influenced by underlying structural factors and may be linked to adverse pregnancy outcomes through epigenetic modifications of gestational tissues. A promising modification is epigenetic gestational age (eGA), which captures 'biological' age at birth. Measuring eGA in placenta, an organ critical for foetal development, may provide information about how SDoH 'get under the skin' during pregnancy to influence birth outcomes and ethnic/racial disparities. We examined relationships of placental eGA with sociodemographic factors, smoking, and two key clinical outcomes: Apgar scores and NICU length of stay. Using the Robust Placental Clock, we estimated eGA for placental samples from the Extremely Low Gestational Age Newborns cohort (N = 408). Regression modelling revealed smoking during pregnancy was associated with placental eGA acceleration (i.e., eGA higher than chronologic gestational age). This association differed by maternal race: among infants born to mothers racialized as Black, we observed greater eGA acceleration (+0.89 week, 95% CI: 0.38, 1.40) as compared to those racialized as white (+0.27 week, 95% CI: -0.06, 0.59). Placental eGA acceleration was also correlated with shorter NICU lengths of stay, but only among infants born to mothers racialized as Black (-0.08 d/week-eGA, 95% CI: -0.12, -0.05). Together, these observed associations suggest that interpretations of epigenetic gestational aging may be tissue-specific.
The acceleration of biological aging is a risk factor for Alzheimer's disease (AD). Here, we performed weighted gene co‐expression network analysis (WGCNA) to identify modules and dysregulated genesinvolved in biological aging in AD. We performed WGCNA to identify modules associated with biological clocks and hub genes of the module with the highest module significance. In addition, we performed differential expression analysis and association analysis with AD biomarkers. WGCNA identified five modules associated with biological clocks, with the module designated as “purple” showing the strongest association. Functional enrichment analysis revealed that the purple module was related to cell migration and death. Ten genes were identified as hub genes in purple modules, of which CX3CR1 was downregulated in AD and low levels of CX3CR1 expression were associated with AD biomarkers. Network analysis identified genes associated with biological clocks, which suggests the genetic architecture underlying biological aging in AD. Examine links between Alzheimer's disease (AD) peripheral transcriptome and biological aging changes. Weighted gene co‐expression network analysis (WGCNA) found five modules related to biological aging. Among the hub genes of the module, CX3CR1 was downregulated in AD. The CX3CR1 expression level was associated with cognitive performance and brain atrophy. Examine links between Alzheimer's disease (AD) peripheral transcriptome and biological aging changes. Weighted gene co‐expression network analysis (WGCNA) found five modules related to biological aging. Among the hub genes of the module, CX3CR1 was downregulated in AD. The CX3CR1 expression level was associated with cognitive performance and brain atrophy.