Content uploaded by Kyle Bourassa
Author content
All content in this area was uploaded by Kyle Bourassa on Jan 18, 2023
Content may be subject to copyright.
Available via license: CC BY 4.0
Content may be subject to copyright.
RESEARCH ARTICLE OPEN ACCESS
Association of Pace of Aging Measured by Blood-Based
DNA Methylation With Age-Related Cognitive
Impairment and Dementia
Karen Sugden, PhD, Avshalom Caspi, PhD, Maxwell L. Elliott, PhD, Kyle J. Bourassa, PhD, Kartik Chamarti,
David L. Corcoran, PhD, Ahmad R. Hariri, PhD, Renate M. Houts, PhD, Meeraj Kothari, MPH,
Stephen Kritchevsky, PhD, George A. Kuchel, MD, Jonathan S. Mill, PhD, Benjamin S. Williams, BSc,
Daniel W. Belsky, PhD, and Terrie E. Moffitt, PhD, for the Alzheimer’s Disease Neuroimaging Initiative*
Neurology®2022;99:e1402-e1413. doi:10.1212/WNL.0000000000200898
Correspondence
Dr. Sugden
karen.sugden@duke.edu
Abstract
Background and Objectives
DNA methylation algorithms are increasingly usedtoestimatebiologicalaging;however,howthese
proposed measures of whole-organism biological aging relate to aging in the brain is not known. We
used data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Framingham
Heart Study (FHS) Offspring Cohort to test the association between blood-based DNA methyl-
ation measures of biological aging and cognitive impairment and dementia in older adults.
Methods
We tested 3 “generations”of DNA methylation age algorithms (first generation: Horvath and
Hannum clocks; second generation: PhenoAge and GrimAge; and third generation: Dun-
edinPACE, Dunedin Pace of Aging Calculated from the Epigenome) against the following
measures of cognitive impairment in ADNI: clinical diagnosis of dementia and mild cognitive
impairment, scores on Alzheimer disease (AD) / Alzheimer disease and related dementias
(ADRD) screening tests (Alzheimer’s Disease Assessment Scale, Mini-Mental State Exami-
nation, and Montreal Cognitive Assessment), and scores on cognitive tests (Rey Auditory
Verbal Learning Test, Logical Memory test, and Trail Making Test). In an independent
replication in the FHS Offspring Cohort, we further tested the longitudinal association between
the DNA methylation algorithms and the risk of developing dementia.
Results
In ADNI (N= 649 individuals), the first-generation (Horvath and Hannum DNA methylation age
clocks) and the second-generation (PhenoAge and GrimAge) DNA methylation measures of aging
were not consistently associated with measures of cognitive impairment in older adults. By contrast, a
third-generation measure of biological aging, DunedinPACE, was associated with clinical diagnosis of
Alzheimer disease (beta [95% CI] = 0.28 [0.08–0.47]), poorer scores on Alzheimer disease/ADRD
screening tests (beta [Robust SE] = −0.10 [0.04] to 0.08[0.04]), and cognitive tests (beta [Robust
SE] = −0.12 [0.04] to 0.10 [0.03]). The association between faster pace of aging, as measured by
DunedinPACE, and risk of developing dementia was confirmed in a longitudinal analysis of the FHS
Offspring Cohort (N= 2,264 individuals, hazard ratio [95% CI] = 1.27 [1.07–1.49]).
MORE ONLINE
H
0
Null Hypothesis
A collection of negative,
inconclusive, or replication
studies; in partnership with
the Center for Biomedical
Research Transparency
NPub.org/Null
From the Department of Psychology and Neuroscience (K.S., A.C., M.L .E., K.C., A.R.H., R.M.H., B.S.W. , T.E.M.), and Center for Genomic an d Computational Biology (K.S., A.C ., B.S.W.,
T.E.M.), Duke University, Durha m, NC; Department of Psychi atry and Behavioral Sciences (A.C ., T.E.M.), Duke University Scho ol of Medicine, Durham, NC; So cial, Genetic, and
Developmental Psychiatry Centre ( A.C, T.E.M.), Institute of Psychi atry, Psychology, and Neuroscien ce, King's College London, UK. Cen ter for the Study of Aging and Human
Development (K.J.B.), Duke University, Durham, NC; Depar tment of Genetics (D.L.C.), University of North Carolina School of Medi cine, Chapel Hill; Butler Columbia Aging Center
(M.K., D.W.B.), Columbia University, New York, New York; Sticht Center for Healthy Aging and Alzheimer’s P revention (S.K.), Wake Forest Schoo l of Medicine, Winston-Salem, NC;
UConn Center on Aging (G.A.K.), University of Connecticut, Farmington, Connecti cut, USA; College of Medicine and Health (J.S.M.) , University of Exeter Medical School, Devon, UK; and
Department of Epidemiology (D.W.B.) , Columbia University Mailman School of Public Health, New York, New York.
Go to Neurology.org/N for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the article.
The Article Processing Charge was funded by RCUK.
*Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the
ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can
be found at links.lww.com/WNL/C175.
This Null Hypothesis article is published as part of a collaborative effort between Neurology
®
and CBMRT.
This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (CC BY), which permits unrestricted use, distribution, and reproduction in any
medium, provided the original work is properly cited.
e1402 Copyright © 2022 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the American Academy of Neurology.
Discussion
Third-generation blood-based DNA methylation measures of aging could prove valuable for measuring differences between
individuals in the rate at which they age and in their risk for cognitive decline, and for evaluating interventions to slow aging.
Aging can be conceptualized as gradual and progressive de-
terioration in biological system integrity causing morbidity and
disability.
1
These changes, in turn, are believed to increase vul-
nerability to multiple age-related diseases,
2,3
including dementias.
Advances in both basic and applied aging research could be
spurred by the availability of tools that can measure biological
aging. Medical, behavioral, and social sciences need measures of
biological aging to identify risk factors and mechanisms that ac-
celerate aging and to use in studies of social groups that are
believedtobeagingatdifferent rates.
4
Applied science needs
measures of biological aging to evaluate whether interventions
succeed in slowing aging. Multiple companies are developing drug
therapies that target aging biology, and several are being evaluated
in human trials (clinicaltrials.gov). Behavior-change science is also
working toward interventions to extend healthspan, including
increasing physical activity, hypertension control, cognitive stim-
ulation, dietary modification, and social engagement.
5-7
Whether
they are pharmaceutical or behavioral, interventions that aim to
extend healthspan need to have a measure to evaluate whether
aging has indeed been slowed. However, as of this writing, there is
no widely accepted measure of biological aging.
8,9
This article
reports the association between dementia, one of the most feared
and costly diseases of aging, and 5 leading candidate measures of
aging based on DNA methylation.
DNA methylation is an epigenetic mechanism by which
specific points of the genome (CpGs) are chemically modified
(methylated) and thereby affect gene regulation. Recent ef-
forts to develop measures of aging have focused on blood
DNA methylation because it is a biological substrate that is
sensitive to age-related changes.
10,11
Using machine learning,
these measurement efforts involve developing algorithms to
capture information about aging by using data about DNA
methylation levels of multiple CpGs across the genome.
These methylation algorithms have evolved rapidly. The first
generation of methylation algorithms was trained on chro-
nologic age in samples ranging from children to older adults.
These “clocks”identified methylation patterns that vary by
chronologic age. If a person’s score on such clocks is older
than his/her actual age, it is inferred that s/he is biologically
older. The first-generation algorithms include the “Hannum
clock”
12
and the “Horvath clock.”
13
The second generation of
methylation algorithms added measures of people’s current
physiologic status to identify methylation patterns that ac-
count for differences in current health and that predict mortality.
These second-generation algorithms include PhenoAge
14
and
GrimAge.
15
TheDunedinPACE(PaceofAgingCalculatedfrom
the Epigenome) is a third-generation algorithm that was de-
veloped by first measuring people’s rate of physiologic change
over time and then identifying the methylation patterns that
captured individual differences in their age-related decline.
Specifically, it measured age-related change in 19 biomarkers
among individuals of the same chronologic age over a 20-year
observation period
16
and then identified methylation patterns at
the end of the observation period that estimated how fast aging
occurred during the years leading up to the point of measure-
ment.
17
Thus, it is designed to capture methylation patterns
reflecting individual differences in age-related decline.
These DNA methylation algorithms have been embraced by
the research community and private companies. However, the
literature evaluating them is fragmented. Although all of these
algorithms purport to measure aging, they have surprisingly
low agreement
18,19
; articles often report promising findings
from one (or more) DNA methylation algorithms, but often
in different samples, and many algorithms show inconsistent
associations with outcomes.
11,20-23
Important validation steps
are now being taken to rigorously evaluate multiple DNA
methylation algorithms in the same study (e.g., in the Health
and Retirement Survey).
19,24
What has not been reported is
an evaluation of multiple DNA methylation algorithms in the
same study with the outcome of dementia.
Here, we leverage data from the Alzheimer’s Disease Neu-
roimaging Initiative (ADNI) to test associations of the leading
measures from the 3 generations of DNA methylation algo-
rithms with cognitive aging and dementia. We examined 3 sets
of “gold standard”measurements in cognitive-aging research.
First, we compared the DNA methylation algorithms’scores
as a function of ADNI participants’diagnoses: cognitively
normal (CN), mild cognitive impairment (MCI), or De-
mentia. Second, we evaluated the algorithms in relation to 3
instruments that are used as cognitive screening tools for
Alzheimer disease (AD) / Alzheimer disease and related
Glossary
AD = Alzheimer disease; ADAS-Cog-13 = Alzheimer’s Disease Assessment Scale–Cognitive; ADNI = Alzheimer’s Disease
Neuroimaging Initiative; ADRD = Alzheimer Disease and Related Dementias; CN = cognitively normal; FHS = Framingham
Heart Study; HR = hazard ratio; MCI = mild cognitive impairment; MMSE = Mini-Mental State Examination; MoCA =
Montreal Cognitive Assessment; SNP = Single Nucleotide Polymorphism.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1403
dementias (ADRD): the Alzheimer’s Disease Assessment
Scale–Cognitive (ADAS-Cog-13
25
), the Mini-Mental State
Examination (MMSE
26
), and the Montreal Cognitive As-
sessment (MoCA
27
). Third, we evaluated the algorithms in
relation to well-established tests of learning and memory (Rey
Auditory Verbal Learning Test
28
), episodic memory (Logical
Memory test
29
), and executive function (Trail Making
Test
30
) that are known to decline with age. We then turned to
a second independent sample, the Framingham Heart Study
(FHS) Offspring Cohort, to evaluate whether and which
DNA methylation measures of biological aging could longi-
tudinally predict risk of developing dementia.
Methods
The ADNI DNA Methylation Sample
Data were obtained from the ADNI database. The primary
goal of ADNI has been to test whether MRI, PET, other
biological markers, and clinical and neuropsychological as-
sessment can be combined to measure the progression of
MCI and early AD. Inclusion and exclusion criteria included
Hachinski Ischemic Score
31
≤4, Geriatric Depression Scale
score
32
<6, visual and auditory acuity adequate for neuro-
psychological testing, good general health with no diseases
precluding enrollment, sixth grade education or work-history
equivalent, no medical contraindications to MRI, no psycho-
active medications that affect cognitive function, medications
stable for 4 weeks before screening, and not enrolled in other
trials or studies concurrently.
33
Data were downloaded from
the ADNI data repository (adni.loni.usc.edu) on June 3, 2020.
DNA Methylation Data
DNA methylation was measured in DNA samples from whole
blood using Illumina 450k arrays and run at the University
of Minnesota and Johns Hopkins University (dbGaP
phs000724.v9.p13).
DNA methylation was measured in DNA from whole blood
using the Illumina Infinium HumanMethylationEPIC BeadChip
Array and run at AbbVie. A total of 649 ADNI participants had
methylation data. Participants varied on the number of repeat
DNA methylation measurements they had: 83 had only a
baseline measurement; 121 had 2 measurements (baseline plus 1
repeat), mean 14.5 months between measurements (SD = 7.06);
407 had 3 measurements, mean 12.1 months between mea-
surements (SD = 1.5); 29 had 4, mean 10.4 months between
measurements (SD = 3.54); and 9 had 5, mean 12.5 months
between measurements (SD = 2.25). Samples were randomized
using a modified incomplete balanced block design, whereby all
samples from a participant were placed on the same chip, with
remaining chip space occupied by age-matched and sex-matched
samples. Participants from different diagnosis groups were
placed on the same chip to avoid confounding.
DNA methylation data were subjected to QC by ADNI in-
vestigators before receipt. Samples with missing rate >1% at
p< 0.05, poor single nucleotide polymorphism (SNP)
matching to the 65 SNP control probe locations, and un-
certain sex were excluded. Filtered data were normalized using
the “NormalizeMethylumiSet”function in the Rpackage
Methylumi. Before normalization, replicate samples (test-
retest of the same sample, N= 198) were identified and
removed from the data set. Probes were removed if detection
the p-value was >0.05 in more than 10% of individuals (N=
611 probes).
Aflowchart documenting the number of samples at each stage
of data preparation is shown in eFigure 1A (links.lww.com/
WNL/C174).
Cognitive Assessments
Data about diagnosis, cognitive impairment screening tests,
and cognitive function were extracted from data tables avail-
able in the “ADNIMERGE”package in Rand then cross-
matched to participants with available DNA methylation data.
Measures are described below.
Diagnosis
Diagnosis was made by a study physician at the time of as-
sessment and categorized as CN, MCI, and AD-dementia.
Clinical Assessment
The ADAS-Cog-13 is a structured scale that evaluates mem-
ory, reasoning, language, orientation, ideational praxis, and
constructional praxis.
25
Delayed Word Recall and Number
Cancellation are included in addition to the 11 standard
ADAS-Cog Items.
34
The test is scored for errors, ranging from
0 (best performance) to 85 (worse performance). The MMSE
is a screening instrument that evaluates orientation, memory,
attention, concentration, naming, repetition, comprehension,
and ability to create a sentence and to copy 2 overlapping
pentagons.
26
The MMSE is scored as the number of correctly
completed items ranging from 0 (worse performance) to 30
(best performance). The MoCA is designed to detect indi-
viduals at the MCI stage of cognitive dysfunction.
27
The scale
ranges from 0 (worse performance) to 30 (best performance).
Cognitive Function
The Rey Auditory Verbal Learning Test is a list learning task
which assesses learning and memory. On each of 5 learning
trials, 15 unrelated nouns are presented orally at the rate of 1
word per second and immediate free recall of the words is
elicited. After a 30-minute delay filled with unrelated testing,
free recall of the original 15-word list is elicited. Both imme-
diate recall and the percent forgotten are used. The Logical
Memory tests I and II (Delayed Paragraph Recall) is from the
Wechsler
29
Memory Scale–Revised. Free recall of 1 short
story is elicited immediately after being read aloud to the
participant and again after a 30-minute delay. The total bits of
information recalled after the delay interval (maximum score
= 25) are analyzed. The Trail Making Test, Part B, consists of
25 circles, either numbered (1 through 13) or containing
letters (A through L). Participants connect the circles while
e1404 Neurology | Volume 99, Number 13 | September 27, 2022 Neurology.org/N
alternating between numbers and letters (e.g., A to 1; 1 to B; B
to 2; 2 to C). Time to complete (300 seconds maximum) is
the primary measure of interest.
The FHS Offspring DNA Methylation Sample
The FHS tracks the development of cardiovascular disease in 3
generations of families recruited in Framingham, Massachusetts,
beginning in 1948.
35
We analyzed data from the second gener-
ation of study participants, who were recruited beginning in
1971. They are known as the Offspring Cohort.
36
To be in-
cluded in the DNA methylation study, participants must have
attended the Framingham Offspring 8th follow-up visit during
2005 and 2008 and have provided a buffycoatsample.
DNA Methylation Data
Data were normalized using the “dasen”method in the wa-
teRmelon R package and subjected to downstream QC. Samples
with missing rate >1% at p<0.01,poorSNPmatchingtothe65
SNP control probe locations, and outliers by multidimensional
scaling techniques were excluded. Probes with missing rate of
>20% at p< 0.01 were also excluded. Additional information on
DNA methylation, normalization, and quality control is avail-
able in the work of Mendelson et al.
37
Aflowchart documenting the number of samples at each stage
of data preparation is shown in eFigure 1B (links.lww.com/
WNL/C174).
Dementia Diagnosis
As previously published,
38-40
participants in this cohort have
been assessed at each examination with the MMSE and flag-
ged for further examinations if (1) they were identified as
having possible cognitive impairment on the basis of screen-
ing assessments, (2) when subjective cognitive decline was
reported by the participant or a family member, (3) on referral
by a treating physician or by ancillary investigators of the FHS,
or (4) after review of outside medical records. All cases of
possible cognitive decline and dementia were reviewed to
determine the presence of dementia, as well as dementia
subtype and date of onset.
39
Dementia ascertainment in our data set extended through 2018
(dbGaP accession pht010750.v1.p13, data set vr_demsurv_
2018_a_1281s). Dementia status was determined for 2,468 par-
ticipants. Of this group, N= 2,264 were alive and free of dementia
at DNA methylation baseline. This analysis sample contributed a
maximum of 14 years of follow-up time for dementia ascertain-
ment, over which interval n = 151 (64 men and 87 women)
developed dementia at an average age of 82 years (SD = 6).
DNA Methylation Clock Estimation
In both ADNI and the FHS Offspring Cohort, we calculated 4
of the DNA methylation age (DNAmAge) clocks (Horvath,
Hannum, PhenoAge, and GrimAge) using the online calcu-
lator found at dnamage.genetics.ucla.edu/new. The “nor-
malization”and “advanced analysis in blood”options were
selected, and data were anonymized before upload. From the
results file, we extracted the corresponding DNA methylation
age calculations (DNAmAge, DNAmAgeHannum, DNAm-
PhenoAge, and DNAmGrimAge) along with the estimates of
white blood cell type abundance. DunedinPACE was calcu-
lated in Rfollowing the procedures described in the work of
Belsky et al.
17
To account for potential technical confounding
introduced during DNA methylation measurement (e.g.,
differential reaction efficiency between batches of assays),
values of the 5 algorithms were residualized for the DNA plate
number. Finally, to derive estimates of DNA methylation age
advancement, these values were further residualized for
chronologic age at the time of the DNA assessment.
Statistical Analysis
All analyses were conducted in R, except for Cox proportional
regression analyses in the FHS Offspring Cohort which were
conducted in STATA. All regression models were adjusted for
sex. To enable effect size comparisons, all age-residualized
scores were standardized to mean = 0, SD = 1 before analysis.
In ADNI, we calculated Huber-White robust standard errors
using the plm and lmtest packages in Rto account for the fact
that some individuals contribute more than 1 time point as
described in the ADNI DNA Methylation Sample. In FHS,
effect sizes are reported as hazard ratios (HRs) per SD in-
crement of the aging measures estimated from Cox pro-
portional hazard regression. To adjust means for sex, we
calculated least-squares means with proportional weights in
the lsmeans package in R. To account for technical variation,
we also tested models adjusted for white blood cell abundance
(plasmablasts, +CD8pCD28nCD45RA-T cells, na¨ıve CD8
T cells, CD4 T cells, natural killer cells, monocytes, and
granulocytes
13,41
). All analyses were performed in parallel by a
second, independent researcher to confirm reproducibility.
Standard Protocol Approvals, Registrations,
and Patient Consents
All research activities were approved by Institutional Review
Boards at the participating study sites. Participants provided
written informed consent.
Data Availability
All data used in this report are publicly available; access is granted
after application approval from the relevant study’s research re-
view committee (ADNI: adni.loni.usc.edu/data-samples/access-
data/; FHS Offspring Cohort: ncbi.nlm.nih.gov/projects/gap/
cgi-bin/study.cgi?study_id=phs000724.v9.p13).
Results
DNA Methylation Measures of Aging in ADNI
DNA methylation data were available for 649 individuals and
1,706 samples (mean [SD] age at first DNA collection = 74.77
(7.66), male = 55.6%). The mean education of the 649 in-
dividuals was 16.22 years (SD = 2.71), and the majority self-
identified as White (N= 636, 98.0%). Table 1 describes
characteristics of participants in ADNI.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1405
Table 2 contains descriptive data (mean [SD] and range)
about the 5 measures of DNA methylation aging. The Hor-
vath, Hannum, PhenoAge, and GrimAge clocks are measured
in units of chronologic years, and DunedinPACE is measured
in years of physiologic decline per 1 chronologic year. All
DNA methylation measures of aging were associated with sex;
males had older DNA methylation age on the clocks and faster
DunedinPACE. All the following analyses include sex as a
covariate. Similarly, all DNA methylation measures of aging
were correlated with chronologic age such that chronologi-
cally older individuals appeared to have older DNA
methylation age on the clocks and faster DunedinPACE
(ranging from r= 0.30 for DunedinPACE to r= 0.85 for
GrimAge). Going forward, we use measures of DNA meth-
ylation age advancement, derived by residualizing the mea-
sures described in Table 2 for participant age at the time of
DNA data collection, rendering them uncorrelated with age.
Figure 1 shows the correlations between the measures of
DNA methylation age advancement. The measures were
significantly intercorrelated; the largest correlations were
observed between the first-generation clocks and PhenoAge
(r=0.45–0.56) and between GrimAge and DunedinPACE (r
= 0.47); otherwise, correlations ranged from r=0.14 to r
=0.28.
Association Between DNA Methylation
Measures of Aging and Dementia Diagnosis
in ADNI
At each DNA data collection point, ADNI participants were
categorized into 3 diagnostic groups. Figure 2 shows the mean
values of the 5 DNA methylation measures of aging for the 3
diagnostic groups: CN, MCI, and Dementia (for comparison
purposes, DNA methylation age advancement values have
been standardized to mean = 0 and SD = 1). The 3 diagnostic
groups did not differ significantly from one another on
first-generation clocks (Horvath and Hannum) or second-
generation clocks (PhenoAge and GrimAge). By contrast, we
observed an ordered association between diagnoses of CN,
MCI, and Dementia for DunedinPACE: Individuals with a
diagnosis of MCI (beta = 0.19, 95% CI: 0.03–0.34) and, to a
greater extent, individuals with a diagnosis of Dementia
(beta = 0.28, 95% CI: 0.08–0.47) had significantly faster
Table 1 Demographic and Clinical Characteristics of the
ADNI and FHS Offspring Cohorts
ADNI
FHS Offspring
CohortIndividuals
All available
samples
N649 1,706 2,264
Age, y (SD) 74.77 (7.66) 75.44 (7.66) 66.05 (8.88)
Male sex, % 55.62 55.28 46.0
Education, y (SD) 16.22 (2.71) 16.21 (2.70) 14.32 (2.60)
White (%) 98.00 98.00 98.50
Dementia (%) 14.48 19.93 6.67
a
Abbreviations: ADNI = Alzheimer’s Disease Neuroimaging Initiative; FHS =
Framingham Heart Study.
Values indicate mean (SD) unless otherwise indicated.
a
Incident dementia (%).
Table 2 Descriptive Data of the First-Generation (Horvath and Hannum), Second-Generation (PhenoAge and GrimAge),
and Third-Generation (DunedinPACE) DNA Methylation Measures in ADNI
Mean (SD) Range Mean difference, men 2women
Correlation (r) with age
(95% CI)
Native measure Age advancement measure
Chronologic age (y) 75.44 (7.66) 55.00–95.62 1.51 ——
First generation
Horvath 64.24 (9.36) 28.41 to 111.74 3.18 0.72 (0.67 to 0.77)
a
0(−0.07 to 0.07)
Hannum 66.11 (7.95) 44.78 to 99.36 2.69 0.78 (0.73 to 0.82)
a
0(−0.07 to 0.07)
Second generation
PhenoAge 63.53 (10.31) 32.07 to 118.80 2.80 0.74 (0.70 to 0.78)
a
0(−0.06 to 0.06)
GrimAge 76.00 (7.46) 55.62 to 102.85 3.87 0.85 (0.81 to 0.89)
a
0(−0.08 to 0.08)
Third generation
DunedinPACE 1.00 (0.12) 0.55 to 1.59 0.03 0.30 (0.23 to 0.37)
a
0(−0.07 to 0.07)
The Horvath, Hannum, PhenoAge, and GrimAge clocks are measured in units of chronologic years, and DunedinPACE is measured in years of physiologic
decline per 1 chronologic year. The third column reports the mean difference between men and women, and the fourth column reports correlations (95% CIs,
adjusted for clustered data) between native DNA methylation aging measures (i.e., unresidualized for age) and chronologic age (Pearson r). The fifth column
reports correlations (95% CIs, adjusted for clustered data) between age advancement measures (i.e., residualized for age) and chronologic age (Pearson r).
a
p< 0.001.
e1406 Neurology | Volume 99, Number 13 | September 27, 2022 Neurology.org/N
DunedinPACE scores than CN individuals (see eTable 1,
links.lww.com/WNL/C174, for details).
Association Between DNA Methylation
Measures of Aging and Cognitive Function
in ADNI
At each DNA collection, ADNI participants were given 3
cognitive screening tests: the ADAS-Cog-13, the MMSE, and
the MoCA. Table 3, Panel A presents the associations be-
tween the 5 DNA methylation measures of aging and scores
on these 3 cognitive screening tests. Neither of the first-
generation DNA methylation clocks nor GrimAge was asso-
ciated with scores on the ADAS-Cog-13, MMSE, or MoCA
(beta = −0.03 to 0.03). By contrast, advanced PhenoAge and
faster DunedinPACE scores were both associated with worse
scores on ADAS-Cog-13 (beta = 0.07 to 0.08) as well as
MMSE and MoCA (beta = −0.06 to −0.10), indicating greater
cognitive impairment.
ADNI participants were also administered a battery of cog-
nitive function tests. Table 3, Panel B presents the associa-
tions between the 5 DNA methylation measures of aging and
4 measures of cognitive functioning: Rey Auditory Verbal
Learning Test (both learning and memory), Logical Memory
test (episodic memory), and Trail Making Test (executive
function). Neither the first-generation clocks (Horvath and
Hannum) nor GrimAge was consistently associated with
performance on these tests (beta = −0.05 to 0.01). By con-
trast, advanced PhenoAge and, to a greater extent, faster
DunedinPACE scores were both associated with significantly
worse learning (beta = −0.06 to −0.12), more forgetting (beta
= 0.06 to 0.10), and worse episodic memory (beta = −0.10 to
−0.11) (Figure 3 and eFigure 2, links.lww.com/WNL/C174).
Sensitivity and Secondary Analyses
Associations reported here were robust in several sensitivity
analyses (eTables 1–3, links.lww.com/WNL/C174). First,
Figure 1 Correlations Between the 5 DNA Methylation Measures of Aging in ADNI
The matrix above the diagonal plots the Pearson rstatistic (with cell color depicting magnitude from light = low to dark = high), while the matrix below the
diagonal shows the scatterplots for each comparison. The red dotted line describes the linear regression line. Correlations are adjusted for sex.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1407
associations were robust to distributional assumptions. Both
the dementia-screening tests and the cognitive function
measures had distributions that deviated from normal. As
such, we repeated all analyses comparing the results with the
“native”(i.e., original) scores, log-transformed scores, and
scores binned into quintiles. Regardless of how we handled
the distributions, the results were comparable (eTables 2 and
3). Second, after controlling for abundance estimates of dif-
ferent types of white blood cells, associations between Dun-
edinPACE and clinical diagnoses and cognitive function tests
were smaller but remained statistically significant at the alpha
= 0.05 level (eTables 1 and 3), whereas those with the
dementia-screening tests fell short of significance (eTable 2).
It is important that Pace of Aging, on which DunedinPACE
was trained in an independent sample,
16,17
includes a longi-
tudinal change in observed white blood cell abundance,
making this an overcontrol. Finally, APOE e4 is known to be
associated with dementia risk; however, it was not associated
with first-generation, second-generation, or third-generation
DNA methylation measures of aging (eTable 4).
Association Between DNA Methylation
Measures of Aging and Dementia in the FHS
Offspring Cohort: Replication and Extension
To replicate and extend the test of the association between
DunedinPACE and dementia, we turned to the FHS Off-
spring Cohort. This longitudinal analysis included N= 2,264
participants with a maximum of 14 years of follow-up time for
dementia ascertainment. Over this time interval, n = 151 (64
men and 87 women) developed dementia at an average age of
82 years (SD = 6).
Participants measured to have more advanced aging on the
clocks and faster DunedinPACE at baseline were at increased
risk of developing dementia over follow-up; the largest effect
was for DunedinPACE (hazard ratio [HR] [95% CI] = 1.39
[1.21–1.61]), followed by PhenoAge, GrimAge, and Horvath
(Table 4). As with ADNI, sensitivity analyses controlling for
white blood cell abundance estimates attenuated effect sizes;
only DunedinPACE (HR [95% CI] = 1.27 [1.07–1.49]) and
the Horvath clock (HR [95% CI] = 1.21 [1.08–1.36]) sig-
nificantly predicted risk of dementia at p< 0.05 (Table 4 and
eFigure 3, links.lww.com/WNL/C174).
Discussion
Aging increases risk for Alzheimer disease, related dementias,
and cognitive impairment.
42
Moreover, the majority of cases
occur later in life and for such individuals, unlike those with
the less common familial AD, aging represents the largest risk
factor for dementia.
43
The potential to capture the individual
dynamics that define the risk of cognitive decline attributable
to biological aging is of great interest to gerontologists and
clinicians alike. In this report, using data from ADNI and the
FHS Offspring Cohort, we compared associations between
first-generation, second-generation, and third-generation
DNA methylation measures of aging and multiple measures
of cognitive aging and dementia. When evaluated against
clinical screening test scores, measures of cognitive func-
tioning, and a clinical diagnosis of dementia, the third-
generation DunedinPACE measure was more predictive than
earlier generations of clocks. In ADNI, it was the only
Figure 2 Mean DNA Methylation Age Advancement Values in Aging in ADNI Within Each of the 3 Diagnostic Categories
Values are grouped by diagnostic category at the time of interview: CN (blue bars), MCI (gold bars), and Dementia (grey bars). The 3 diagnostic status groups
did not differ significantly from one another on either of the first-generation DNA methylation clocks (Horvath and Hannum clocks) or on the second-
generation clocks (PhenoAge and GrimAge). By contrast, individuals with MCI or Dementia had faster DunedinPACE scores than those who were CN. Bars
represent means, and whiskers represent 95% CIs. Values are standardized to mean = 0, SD = 1. CN = cognitively normal; MCI = mild cognitive impairment.
e1408 Neurology | Volume 99, Number 13 | September 27, 2022 Neurology.org/N
biological aging estimate to show consistent associations with
every measure of cognitive impairment tested in the predicted
direction of faster aging and more impairment. Moreover,
faster DunedinPACE was associated with increased risk of
developing future dementia in the FHS Offspring Cohort.
A DNA methylation algorithm that can assess biological aging
should be robustly associated with cognitive dysfunction
characteristic of AD/ADRD. First, we showed that individuals
with a diagnosis of dementia and, to a lesser extent, mild
cognitive impairment had faster DunedinPACE compared
with individuals who were CN. This pattern was not observed
for the first-generation and second-generation DNA methyl-
ation age advancement clocks. Second, individuals who
scored poorly on screening measures commonly used in
memory clinics (ADAS-Cog-13, MMSE, and MoCA) had
older DNA methylation age advancement (assessed by Phe-
noAge) and faster DunedinPACE. Third, individuals’worse
cognitive function was associated with older DNA methyla-
tion age advancement (assessed by PhenoAge) and faster
DunedinPACE. It is important to note that the cognitive
measures that we examined overlap to some extent, for ex-
ample, the Logical Memory test is used to derive AD
diagnoses. However, we think it essential to present evidence
from all the cognitive measures because different studies often
evaluate different cognitive measures, making it difficult to
compare studies and reconcile inconsistencies.
Previous studies testing associations between DNA methyl-
ation clocks and late-life cognition and dementia have yielded
equivocal and inconsistent evidence.
21,44-46
By contrast, this
study suggests that the newer generation DunedinPACE
measure is consistently associated with multiple manifesta-
tions of age-related cognitive deficits. This is consistent with
previously reported evidence that faster DunedinPACE is
associated with greater cognitive decline during midlife.
17
This consistency suggests that vulnerability to cognitive im-
pairment that is the hallmark of risk for dementia can be
captured by considering how fast a person is aging biologically
compared with their age peers. The finding that extremely fast
DunedinPACE scores occur with dementia is consistent with
the view that dementia is not part of normal aging.
Consistent with previous studies (e.g., references 17-
19,24,47,48), the 5 tested DNA methylation measures of
aging vary in the extent to which they are intercorrelated, and
Table 3 Associations Between DNA Methylation Measures of Aging and Cognitive Screening Tests and Function Tests
in ADNI
Panel A: Screening tests
DNA methylation measures of aging
ADAS-Cog-13 MMSE MoCA
Beta (Robust SE) Beta (Robust SE) Beta (Robust SE)
Horvath 0.00 (0.04) 0.01 (0.03) 0.03 (0.03)
Hannum 0.02 (0.04) −0.02 (0.04) −0.02 (0.03)
PhenoAge 0.07 (0.03)
a
−0.06 (0.03) −0.07 (0.03)
a
GrimAge 0.01 (0.03) 0.01 (0.03) −0.03 (0.03)
DunedinPACE 0.08 (0.04)
a
−0.08 (0.03)
a
−0.10 (0.04)
b
Panel B: Cognitive function tests
DNA methylation measures of aging
RAVLT immediate recall RAVLT percent forgotten Logical Memory Trail Making Test, Part B
Beta (Robust SE) Beta (Robust SE) Beta (Robust SE) Beta (Robust SE)
Horvath 0.01 (0.04) 0.01 (0.03) −0.01 (0.04) −0.02 (0.03)
Hannum −0.02 (0.04) 0.00 (0.03) −0.01 (0.04) 0.00 (0.03)
PhenoAge −0.06 (0.04) 0.06 (0.03) −0.10 (0.04)
b
0.03 (0.03)
GrimAge −0.05 (0.03) 0.03 (0.03) −0.03 (0.03) 0.00 (0.03)
DunedinPACE −0.12 (0.04)
c
0.10 (0.03)
b
−0.11 (0.04)
b
0.06 (0.04)
Abbreviations: ADAS-Cog-13 = Alzheimer’s Disease Assessment Scale–Cognitive; MMSE = Mini-Mental State Examination; MoCA; Montreal Cognitive As-
sessment; RAVLT = Rey Auditory Verbal Learning Test.
Panel A shows the results of linear regressions of cognitive screening scores (ADAS-Cog-13, higher scores = poorer performance; MMSE, lower scores = poorer
performance; MoCA, lower scores = poorer performance) on the 5 DNA methylation measures of aging. Panel B shows the results of linear regressions of
cognitive screening scores (RAVLT immediate recall, lower scores = poorer performance; RAVLT percent forgotten, higher scores = poorer performance;
Logical Memory, lower scores = poorer performance; Trail Making Test, Part B, higher scores = poorer performance) on the 5 DNA methylation measures of
aging. Both cognitive screening and function scores and DNA methylation measures were standardized to mean = 0, SD = 1 before analysis. All analyses
included sex as a covariate in the model. To account for clustering, we report Huber-White robust standard errors.
a
p< 0.05.
b
p< 0.01.
c
p< 0.001.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1409
clocks in the same generation tend to be more highly corre-
lated with one another. This suggests that although different
DNA methylation measures of aging capture some common
elements, they are also clearly distinct. First-generation clocks
were trained to predict chronologic age. This approach is
based on the assumption that differences in DNA methylation
between older and younger people represent biological pro-
cesses of aging. Second-generation clocks were trained to
predict mortality, using physiologic variables as intermediates.
This approach is based on the assumption that differences in
DNA methylation between people with higher as compared
with lower risk for mortality represent biological processes of
aging. The third-generation DunedinPACE was trained to
predict biological change between ages 26 and 45 years in a
same-age cohort. This approach is based on the assumption
that DNA methylation differences between people experi-
encing slower as compared with more rapid decline in the
function of multiple organ systems represent biological pro-
cesses of aging. The evidence presented here suggests that
progressive generations of clocks may be more sensitive
predictors of cognitive outcomes. Moreover, the association
of DunedinPACE with dementia recommends midlife pre-
vention if some patients’course toward dementia begins in
midlife.
There are caveats and limitations. First, despite robust asso-
ciations between DunedinPACE and measures of cognitive
aging, none of the currently available measures of DNA
methylation aging match clinically validated risk markers of
ADRD on strength-of-association. For example, within the
ADNI participants analyzed in the present report, individuals
with a diagnosis of dementia were 12 times more likely to
carry 2 APOE «4alleles than individuals who were CN, an
effect size of a 0.94 SD-unit difference between dementia vs
CN. To put the effect size for the DunedinPACE comparison
between dementia vs CN in perspective, it yielded a 0.28 SD-
unit difference. Second, most of the participants are White
because of the lack of ethnic diversity of the participants en-
rolled in ADNI and Framingham. Initial evidence shows that
an earlier version of a methylation Pace of Aging algorithm,
DunedinPoAm, is associated with poorer physical health
among both Black and White participants,
19
but more re-
search is needed on this front. Third, we were able to report
only cross-sectional associations between DNA methylation
measures of aging and cognitive impairment and AD in ADNI
because the number transitioning to a new diagnosis was too
small for statistical power among ADNI participants who had
methylation data. To overcome this limitation, we extended
our analysis to the larger FHS Offspring Cohort and found
that DunedinPACE was associated prospectively with future
risk of developing dementia. Fourth, this study reports initial
replicated evidence that DunedinPACE derived in midlife
signals dementia risk in late life, but life-course longitudinal
studies should evaluate potential causal pathways including
Figure 3 DunedinPACE Values by Test Score Quintile for the Rey Auditory Verbal Learning Test, Logical Memory Test, and
Trail Making Test Cognitive Assessments in Aging in ADNI
Faster DunedinPACE was associated with poorerlearning and memory (RAVLT, immediate recall [A] and percent forgotten [B]), episodic memory (LogicalMemorytest
[C]), and executive functioning (Trail Making Test, Part B [D]). Cognitive function scores (x-axis) are binned into quintiles (1–5); grey dots represent mean age
advancement value, and whiskers represent 95% CIs. The y-axis represents DunedinPACE (age-residualized, adjusted for sex, and standardized to mean=0,SD=1).
e1410 Neurology | Volume 99, Number 13 | September 27, 2022 Neurology.org/N
early-life age accelerators (e.g., low socioeconomic status, low
education, and smoking) and potential late-life mediators
(e.g., disease multimorbidity
17
). Fifth, dementia is also not a
single disease and future, adequately powered studies with
dementia subtypes should test for DunedinPACE’s specificity.
Sixth, ample evidence points to genetic loci contributing to
dementias.
49
By contrast, DNA methylation variation repre-
sents the epigenetic results of processes along pathways to-
ward dementia, suggesting DunedinPACE is best considered
a noncausal risk indicator.
As the search gains steam for geroscience-guided interven-
tions that might slow aging and prevent the onset of age-
related diseases, including Alzheimer’s disease and related
dementias, the need for reliable measures of biological aging
related to dementia is becoming more apparent. Such
measures could serve to identify people at high risk for future
dementia and could serve as surrogate measures to evaluate
interventions while waiting for the longer term outcome of
dementia. DNA methylation measures of aging have offered
promise, but their relation to cognitive aging and dementia
has been equivocal. Here, we find evidence that a third-
generation DNA methylation measure of aging, trained on
longitudinally measured biological decline, may prove useful
in dementia research.
Acknowledgment
The authors thank the Northern California Institute for
Research and Education, the Alzheimer’s Therapeutic Re-
search Institute at the University of Southern California, and
the Laboratory for Neuro Imaging at the University of
Southern California for curation, coordination, and
dissemination of ADNI data. The Framingham Heart
Study is conducted and supported by the National Heart,
Lung, and Blood Institute (NHLBI) in collaboration
with Boston University (Contract No. N01-HC-25195,
HHSN268201500001I, and 75N92019D00031). Funding
support for the Framingham Dementia Survival Informa-
tion for All Cohorts dataset was provided by NIA (Grant
No. R01-AG054076). 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 NHLBI.
Study Funding
This work was supported by the National Institute on
Aging (R01AG032282, R01AG049789, F99 AG068432-01,
R01AG061378, R01AG073207, and T32-AG000029), the
UK Medical Research Council (MR/P005918/1), the
Lafitte Foundation, the National Science Foundation
(Graduate Research Fellowship No. NSF DGE-1644868),
the Russell Sage Foundation BioSS (1810-08987), and the
North Carolina Biotechnology Center (2016-IDG-1013).
Data collection and sharing for this project was funded by
the Alzheimer’s Disease Neuroimaging Initiative (ADNI)
supported by the NIH (U01AG024904) and Department
of Defense (W81XWH-12-2-0012). ADNI is funded by the
National Institute on Aging, the National Institute of
Biomedical Imaging and Bioengineering, and through
generous contributions from the following: AbbVie, Alz-
heimer’s Association; Alzheimer’s Drug Discovery Foun-
dation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-
Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai
Inc.; Elan Pharmaceuticals, Inc.; EliLilly and Company;
EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated
company Genentech, Inc.; Fujirebio; GE Healthcare;
IXICO Ltd.; Janssen Alzheimer Immunotherapy Research
& Development, LLC.; Johnson & Johnson Pharmaceutical
Research & Development LLC.; Lumosity; Lundbeck;
Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx
Research; Neurotrack Technologies; Novartis Pharma-
ceuticals Corporation; Pfizer Inc.; Piramal Imaging; Serv-
ier; Takeda Pharmaceutical Company; and Transition
Table 4 Longitudinal Associations of DNA Methylation Measures of Aging With Dementia in the FHS
DNA methylation measures of aging
Model adjusted for sex Model adjusted for sex and estimated cell counts
HR 95% CI HR 95% CI
Horvath 1.18
a
1.06–1.32 1.21
a
1.08–1.36
Hannum 1.09 0.96–1.23 0.96 0.83–1.12
PhenoAge 1.25
a
1.08–1.44 1.15 0.98–1.36
GrimAge 1.24
a
1.07–1.44 1.05 0.86–1.27
DunedinPACE 1.39
b
1.21–1.61 1.27
a
1.07–1.49
Abbreviations: HR = hazard ratios.
This table reports effect sizes for DNA methylation m easures of aging from time-to-event analysis of dementia. The first panel shows theresults from
a model including sex and age as covariates, and the second panel shows the results from a model that includes these covariates in addition to white
blood cell abundance estimated from the DNA methylation data. Time-to-event model effect sizes are reported as HR per SD increase in the aging
measures.
a
p< 0.01.
b
p< 0.001.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1411
Therapeutics. The Canadian Institutes of Health Research
is providing funds to support ADNI clinical sites in Canada.
Private sector contributions are facilitated by the Founda-
tion for the NIH (fnih.org).
Disclosure
K. Sugden, A. Caspi, D.L. Corcoran, D.W. Belsky, and T.E.
Moffitt are listed as inventors on a Duke University and
University of Otago invention that is licensed to a commercial
entity. The other authors report no relevant disclosures. Go to
Neurology.org/N for full disclosures.
Publication History
Received by Neurology December 22, 2021. Accepted in final form
May 13, 2022. Submitted and externally peer reviewed. The handling
editor was Linda Hershey, MD, PhD, FAAN.
References
1. Kirkwood TB. Understanding the odd science of aging. Cell. 2005;120(4):437-447.
2. Kennedy BK, Berger SL, Brunet A, et al. Geroscience: linking aging to chronic disease.
Cell. 2014;159(4):709-713.
3. Lopez-Otin C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of
aging. Cell. 2013;153(6):1194-1217.
4. Moffitt TE. Behavioral and social research to accelerate the geroscience translation
agenda. Ageing Res Rev. 2020;63:101146.
5. Queen NJ, Hassan QN II, Cao L. Improvements to healthspan through environ-
mental enrichment and lifestyle interventions: where are we now? Front Neurosci.
2020;14:605.
6. Gonzalez-Freire M, Diaz-Ruiz A, Hauser D, et al. The road ahead for health and
lifespan interventions. Ageing Res Rev. 2020;59:101037.
Appendix 1 Authors
Name Location Contribution
Karen
Sugden, PhD
Psychology and
Neuroscience, Duke
University; Center for
Genomic and Computational
Biology, Duke University
Drafting/revision of the
manuscript for content,
including medical writing for
content; study concept or
design; analysis or
interpretation of data
Avshalom
Caspi, PhD
Psychology and
Neuroscience, Duke
University; Center for
Genomic and
Computational Biology,
Duke University;
Department of Psychiatry
and Behavioral Sciences,
Duke University School of
Medicine; Institute of
Psychiatry, Psychology, and
Neuroscience, King’s
College London, Social,
Genetic, and Developmental
Psychiatry Centre
Drafting/revision of the
manuscript for content,
including medical writing for
content; study concept or
design; analysis or
interpretation of data
Maxwell L.
Elliott, PhD
Psychology and
Neuroscience, Duke
University
Drafting/revision of the
manuscript for content,
including medical
writing for content; major
role in the acquisition of
data
Kyle J.
Bourassa,
PhD
Center for the Study of Aging
and Human Development,
Duke University
Drafting/revision of the
manuscript for content,
including medical writing for
content
Kartik
Chamarti
Psychology and
Neuroscience, Duke
University
Drafting/revision of the
manuscript for content,
including medical writing
for content
David L.
Corcoran,
PhD
Department of Genetics,
University of North Carolina
School of Medicine
Drafting/revision of the
manuscript for content,
including medical writing for
content; analysis or
interpretation of data
Ahmad R.
Hariri, PhD
Psychology and
Neuroscience, Duke
University
Drafting/revision of the
manuscript for content,
including medical writing for
content
Appendix 1 (continued)
Name Location Contribution
Renate M.
Houts, PhD
Psychology and
Neuroscience, Duke
University
Drafting/revision of the
manuscript for content,
including medical writing for
content; analysis or
interpretation of data
Meeraj
Kothari,
MPH
Butler Columbia Aging
Center, Columbia University
Analysis or interpretation of
data
Stephen
Kritchevsky,
PhD
Sticht Center for Healthy
Aging and Alzheimer’s
Prevention, Wake Forest
School of Medicine
Drafting/revision of the
manuscript for content,
including medical writing for
content
George A.
Kuchel, MD
UConn Center on Aging,
University of Connecticut
Health Center
Drafting/revision of the
manuscript for content,
including medical writing for
content
Jonathan S.
Mill, PhD
College of Medicine and
Health, University of Exeter
Medical School
Drafting/revision of the
manuscript for content,
including medical writing for
content
Benjamin S.
Williams,
BSc
Psychology and
Neuroscience, Duke
University; Center for
Genomic and Computational
Biology, Duke University
Drafting/revision of the
manuscript for content,
including medical writing for
content
Daniel W.
Belsky, PhD
Butler Columbia Aging
Center, Columbia University;
Department of
Epidemiology, Columbia
University Mailman School of
Public Health
Drafting/revision of the
manuscript for content,
including medical writing for
content; major role in the
acquisition of data
Terrie E.
Moffitt, PhD
Psychology and
Neuroscience, Duke
University; Center for
Genomic and Computational
Biology, Duke University;
Department of Psychiatry
and Behavioral Sciences,
Duke University School of
Medicine; Institute of
Psychiatry, Psychology, and
Neuroscience, King’s College
London, Social, Genetic, and
Developmental Psychiatry
Centre
Drafting/revision of the
manuscript for content,
including medical writing for
content; study concept or
design
Appendix 2 Coinvestigators
Coinvestigators are listed at links.lww.com/WNL/C175
e1412 Neurology | Volume 99, Number 13 | September 27, 2022 Neurology.org/N
7. O’Rourke HM, Collins L, Sidani S. Interventions to address social connectedness and
loneliness for older adults: a scoping review. BMC Geriatr. 2018;18(1):214.
8. Cohen AA, Legault V, Fulop T. What if there’s no such thing as “aging”.Mech Ageing
Dev. 2020;192:111344.
9. Crimmins EM, Thyagarajan B, Kim JK, Weir D, Faul J. Quest for a summary measure
of biological age: the health and retirement study. Geroscience. 2021;43(1):395-408.
10. Horvath S, Raj K. DNA methylation-based biomarkers and the epigenetic clock
theory of ageing. Nat Rev Genet. 2018;19(6):371-384.
11. Levine ME. Assessment of epigenetic clocks as biomarkers of aging in basic and
population research. J Gerontol A Biol Sci Med Sci. 2020;75(3):463-465.
12. Hannum G, Guinney J, Zhao L, et al. Genome-wide methylation profiles reveal
quantitative views of human aging rates. Mol Cell. 2013;49(2):359-367.
13. Horvath S. DNA methylation age of human tissues and cell types. Geno me Biol. 2013;
14(10):R115.
14. Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging for lifespan and
healthspan. Aging (Albany NY). 2018;10(4):573-591.
15. Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts
lifespan and healthspan. Aging (Albany NY). 2019;11(2):303-327.
16. Elliott ML, Caspi A, Houts RM, et al. Disparities in the pace of biological aging among
midlife adults of the same chronological age have implications for future frailty risk
and policy. Nat Aging. 2021;1(3):295-308.
17. Belsky DW, Caspi A, Corcoran DL, et al. DunedinPACE, a DNA methylation bio-
marker of the pace of aging. Elife. 2022;11:e73420.
18. Belsky DW, Caspi A, Arseneault L, et al. Quantification of the pace of biological aging
in humans through a blood test, the DunedinPoAm DNA methylation algorithm. Elife
2020;9:e54870.
19. Graf GH, Crowe CL, Kothari M, et al. Testing Black-White disparities in biological
aging in older adults in the United States: analysis of DNA-methylation and blood-
chemistry methods. Am J Epidemiol. 2021;191(4):613-625.
20. Oblak L, van der Zaag J, Higgins-Chen AT, Levine ME, Boks MP. A syst ematic review
of biological, social and environmental factors associated with epigenetic clock ac-
celeration. Ageing Res Rev. 2021;69:101348.
21. Maddock J, Castillo-Fernandez J, Wong A, et al. DNA methylation age and physical
and cognitive aging. J Gerontol A Biol Sci Med Sci. 2020;75(3):504-511.
22. McCrory C, Fiorito G, Hernandez B, et al. GrimAge outperforms other epigenetic
clocks in the prediction of age-related clinical phenotypes and all-cause mortality.
J Gerontol A Biol Sci Med Sci. 2021;76(5):741-749.
23. Fransquet PD, Wrigglesworth J, Woods RL, Ernst ME, Ryan J. The epigenetic clock as
a predictor of disease and mortality risk: a systematic review and meta-analysis. Clin
Epigenetics. 2019;11(1):62.
24. Crimmins EM, Thyagarajan B, Levine ME, Weir DR, Faul J. Associations of age, sex,
race/ethnicity, and education with 13 epigenetic clocks in a nationally representative
U.S. Sample: the health and retirement study. J Gerontol A Biol Sci Med Sci. 2021;
76(6):1117-1123.
25. Rosen WG, Mohs RC, Davis KL. A new rating scale for Alzheimer’s disease. Am J
Psychiatry 1984;141(11):1356-1364.
26. Folstein MF, Folstein SE, McHugh PR. “Mini-mental state”. A practical method for grading
the cognitive state of patients for the clinician. J Psychiatr Res. 1975;12(3):189-198.
27. Nasreddine ZS, Phillips NA, Bedirian V, et al. The Montreal Cognitive Assessment,
MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. 2005;
53(4):695-699.
28. Rey A. L’examen clinique en psychologie. Presses universitaires de France; 1964.
29. Wechsler D. Wechsler Memory Scale-Revised Manual. The Psychological Corporation;
1987.
30. Reitan RM. Validity of the Trail Making test as an indicator of organic brain damage.
Percept Mot Skills. 1958;8(3):271-276.
31. Hachinski VC, IliffLD, Zilhka E, et al. Cerebral blood flow in dementia. Arch Neurol.
1975;32(9):632-637.
32. Yesavage JA, Brink TL, Rose TL, et al. Development and validation of a geriatric
depression screening scale: a preliminary report. J Psychiatr Res. 1982;17(1):37-49.
33. Petersen RC, Aisen PS, Beckett LA, et al. Alzheimer’s disease neuroimaging initiative
(ADNI): clinical characterization. Neurology. 2010;74(3):201-209.
34. Mohs RC, Knopman D, Petersen RC, et al. Development of cognitive instruments for
use in clinical trials of antidementia drugs: additions to the Alzheimer’s Disease
Assessment Scale that broaden its scope. Alzheimer Dis Assoc Disord. 1997;11(suppl
2):S13-S21.
35. Tsao CW, Vasan RS. The Framingham Heart Study: past, present and future. Int J
Epidemiol. 2015;44(6):1763-1766.
36. Feinleib M, Kannel WB, Garrison RJ, McNamara PM, Castelli WP. The Framingham
offspring study. Design and preliminary data. Prev Med. 1975;4(4):518-525.
37. Mendelson MM, Marioni RE, Joehanes R, et al. Association of body mass index with
DNA methylation and gene expression in blood cells and relations to cardiometabolic
disease: a mendelian randomization approach. Plos Med. 2017;14(1):e1002215.
38. Satizabal C, Beiser AS, Seshadri S. Incidence of dementia over three decades in the
Framingham Heart study. N Engl J Med. 2016;375(1):93-94.
39. Seshadri S, Beiser A, Au R, et al. Operationalizing diagnostic criteria for Alzheimer’s
disease and other age-related cognitive impairment-Part 2. Alzheimers Dement 2011;
7(1):35-52.
40. Seshadri S, Wolf PA, Beiser A, et al. Lifetime risk of deme ntia and Alzheimer’s disease.
The impact of mortality on risk estimates in the Framingham Study. Neurology. 1997;
49(6):1498-1504.
41. Houseman EA, Accomando WP, Koestler DC, et al. DNA methylation arrays as
surrogate measures of cell mixture distribution. BMC Bioinformatics. 2012;13:86.
42. Qiu C, Fratiglioni L. Aging without dementia is achievable: current evidence from
epidemiological research. J Alzheimers Dis. 2018;62(3):933-942.
43. Kaeberlein M. Time for a new strategy in the war on Alzheimer’s disease. Public Policy
Aging Rep. 2019;29(4):119-122.
44. Marioni RE, Shah S, McRae AF, et al. The epigenetic clock is correlated with physical
and cognitive fitness in the Lothian Birth Cohort 1936. Int J Epidemiol. 2015;44(4):
1388-1396.
45. Vaccarino V, Huang M, Wang Z, et al. Epigenetic age acceleration and cognitive
decline: a twin study. J Gerontol A Biol Sci Med Sci. 2021;76(10):1854-1863.
46. Starnawska A, Tan Q, Lenart A, et al. Blood DNA methylation age is not associated
with cognitive functioning in middle-aged monozygotic twins. Neurobiol Aging. 2017;
50:60-63.
47. Nwanaji-Enwerem JC, Nwanaji-Enwerem U, Van Der Laan L, Galazka JM, Redeker
NS, Cardenas A. A longitudinal epigenetic aging and leukocyte analysis of simulated
space travel: the mars-500 mission. Cell Rep. 2020;33(10):108406.
48. Schmitz LL, Zhao W, RatliffSM, et al. The socioeconomic gradient in epigenetic aging
clocks: evidence from the multi-ethnic study of atherosclerosis and the health and
retirement study. medRxiv. 2021.
49. Wightman DP, Jansen IE, Savage JE, et al. A genome-wide association study with
1,126,563 individuals identifies new risk loci for Alzheimer’s disease. Nat Genet. 2021;
53(9):1276-1282.
Neurology.org/N Neurology | Volume 99, Number 13 | September 27, 2022 e1413
