Genetic Evidence for Causal Effects of Socioeconomic, Lifestyle, and Cardiometabolic Factors on Epigenetic-Age Acceleration

Abstract

GrimAge acceleration (GrimAgeAccel) and PhenoAge acceleration (PhenoAgeAccel) are DNA methylation-based markers of accelerated biological aging, standing out in predicting mortality and age-related cardiometabolic morbidities. Causal risk factors for GrimAgeAccel and PhenoAgeAccel are unclear. In this study, we performed two-sample univariable and multivariable Mendelian randomization (MR) to investigate causal associations of 19 modifiable socioeconomic, lifestyle, and cardiometabolic factors with GrimAgeAccel and PhenoAgeAccel. Instrument variants representing 19 modifiable factors were extracted from genome-wide association studies (GWASs) with up to 1 million Europeans. Summary statistics for GrimAgeAccel and PhenoAgeAccel were derived from a GWAS of 34,710 Europeans. We identified 12 and eight factors causally associated with GrimAgeAccel and PhenoAgeAccel, respectively. Smoking was the strongest risk factor (β [SE]: 1.299 [0.107] year) for GrimAgeAccel, followed by higher alcohol intake, higher waist circumference, daytime napping, higher body fat percentage, higher body mass index, higher C-reactive protein, higher triglycerides, childhood obesity, and type 2 diabetes; whereas education was the strongest protective factor (β [SE]: -1.143 [0.121] year), followed by household income. Furthermore, higher waist circumference (β [SE]: 0.850 [0.269] year) and education (β [SE]: -0.718 [0.151] year) were the leading causal risk and protective factors for PhenoAgeAccel, respectively. Sensitivity analyses strengthened the robustness of these causal associations. Multivariable MR analyses further demonstrated independent effects of the strongest risk and protective factors on GrimAgeAccel and PhenoAgeAccel, respectively. In conclusion, our findings provide novel quantitative evidence on modifiable causal risk factors for accelerated epigenetic aging, suggesting promising intervention targets against age-related morbidity and improving healthy longevity.

Full text

1083
Journals of Gerontology: Biological Sciences
cite as: J Gerontol A Biol Sci Med Sci, 2023, Vol. 78, No. 7, 1083–1091
https://doi.org/10.1093/gerona/glad078
Advance Access publication March 4, 2023
Editor’s choice
© The Author(s) 2023. Published by Oxford University Press on behalf of The Gerontological Society of America.
All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
Original Article
Genetic Evidence for Causal Effects of Socioeconomic,
Lifestyle, and Cardiometabolic Factors on Epigenetic-Age
Acceleration
Lijie Kong, MD,1,2,†, Chaojie Ye, MD,1,2,† Yiying Wang, MD,1,2 Tianzhichao Hou, MD,1,2
JieZheng, PhD,1,2, ZhiyunZhao, MD, PhD,1,2, MianLi, MD, PhD,1,2 YuXu, MD, PhD,1,2,
JieliLu, MD, PhD,1,2, YuhongChen, MD, PhD,1,2 MinXu, MD, PhD,1,2, WeiqingWang, MD,
PhD,1,2, GuangNing, MD, PhD,1,2, YufangBi, MD, PhD,1,2 and TiangeWang, MD, PhD1,2,*,
1Department of Endocrine and Metabolic Diseases, Shanghai Institute of Endocrine and Metabolic Diseases, Ruijin Hospital, Shanghai
Jiao Tong University School of Medicine, Shanghai, China. 2Shanghai National Clinical Research Center for Metabolic Diseases, Key
Laboratory for Endocrine and Metabolic Diseases of the National Health Commission of the PR China, Shanghai Key Laboratory for
Endocrine Tumor, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
*Address correspondence to: Tiange Wang, MD, PhD, Shanghai National Clinical Research Center for Endocrine and Metabolic Diseases,
Shanghai Institute of Endocrine and Metabolic Diseases, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, 197 Rujin 2nd Road,
Shanghai, China. E-mail: tiange.wang@shsmu.edu.cn
†These authors contributed equally as first authors.
Received: September 25, 2022; Editorial Decision Date: February 22, 2023
Decision Editor: DavidLeCouteur , MBBS, FRACP, PhD
Abstract
GrimAge acceleration (GrimAgeAccel) and PhenoAge acceleration (PhenoAgeAccel) are DNA methylation-based markers of accelerated
biological aging, standing out in predicting mortality and age-related cardiometabolic morbidities. Causal risk factors for GrimAgeAccel
and PhenoAgeAccel are unclear. In this study, we performed 2-sample univariable and multivariable Mendelian randomization (MR) to
investigate causal associations of 19 modiable socioeconomic, lifestyle, and cardiometabolic factors with GrimAgeAccel and PhenoAgeAccel.
Instrument variants representing 19 modiable factors were extracted from genome-wide association studies (GWASs) with up to 1 million
Europeans. Summary statistics for GrimAgeAccel and PhenoAgeAccel were derived from a GWAS of 34710 Europeans. We identied 12 and
8 factors causally associated with GrimAgeAccel and PhenoAgeAccel, respectively. Smoking was the strongest risk factor (β [standard error
{SE}]: 1.299 [0.107] year) for GrimAgeAccel, followed by higher alcohol intake, higher waist circumference, daytime napping, higher body fat
percentage, higher body mass index, higher C-reactive protein, higher triglycerides, childhood obesity, and type 2 diabetes; whereas education
was the strongest protective factor (β [SE]: −1.143 [0.121] year), followed by household income. Furthermore, higher waist circumference (β
[SE]: 0.850 [0.269] year) and education (β [SE]: −0.718 [0.151] year) were the leading causal risk and protective factors for PhenoAgeAccel,
respectively. Sensitivity analyses strengthened the robustness of these causal associations. Multivariable MR analyses further demonstrated
independent effects of the strongest risk and protective factors on GrimAgeAccel and PhenoAgeAccel, respectively. In conclusion, our ndings
provide novel quantitative evidence on modiable causal risk factors for accelerated epigenetic aging, suggesting promising intervention targets
against age-related morbidity and improving healthy longevity.
Keywords: Causal risk factors, DNA methylation, Epigenetic aging, Mendelian randomization
Biological
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

Aging involves the gradual accumulation of a decline in multiple bio-
logical functions over time, leading to increased risks of developing
age-related diseases and mortality (1,2). Although chronological
aging is uniform and unchangeable, the rate of biological aging is
variable and modiable depending on individual genetics, envir-
onmental exposures, and health-related behaviors (3). Of several
potential types of biological age predictors (eg, epigenetic clock,
leukocyte telomere length, and transcriptomic predictors), the epi-
genetic clock composed of DNA methylation at multiple cytosine–
phosphate–guanine sites is currently the best one, as it correlates
well with age and predicts mortality across populations (4). The
epigenetic-age acceleration is the difference between chronological
age and epigenetic age and represents accelerated biological aging
(5). The second-generation epigenetic-age acceleration indicators,
namely GrimAge acceleration (GrimAgeAccel) and PhenoAge ac-
celeration (PhenoAgeAccel), have evolved to incorporate aging-
related traits, and stand out in terms of predicting mortality and
age-related cardiometabolic morbidities (6–9).
Limited observational evidence has suggested that certain
socioeconomic status (SES), lifestyle behaviors, and cardiometabolic
traits may be related to GrimAgeAccel and PhenoAgeAccel
(6,7,10,11). Whether and to what extent modiable risk factors inu-
ence GrimAgeAccel and PhenoAgeAccel, if causally established, could
shed light on potential contributors to the aging process and elucidate
promising targets for preventing age-related diseases and improving
healthy longevity (12,13). Thus far, such causal evidence isscarce.
To ll this knowledge gap, we applied Mendelian randomiza-
tion (MR) to evaluate the causal associations of 19 common modi-
able socioeconomic, lifestyle, and cardiometabolic factors with
GrimAgeAccel and PhenoAgeAccel. The MR method uses genetic vari-
ants that are robustly associated with exposure as instrumental vari-
ables (IVs) to estimate the causal effect of exposure on an outcome.
Since genetic variants are randomly allocated at conception, the MR
study is less susceptible to confounding and reverse causality than con-
ventional observational studies (14). We performed both univariable
and multivariable MR analyses to discern if modiable factors have
independent causal effects on GrimAgeAccel and PhenoAgeAccel.
Method
StudyDesign
The MR study design is shown in Figure 1. We performed
2-sample univariable and multivariable MR strictly following
the Strengthening the Reporting of Observational Studies in
Epidemiology using Mendelian Randomization (STROBE-MR)
guidelines (Supplementary Table 1) (15). To obtain unbiased esti-
mates of the causal effects, the MR analysis should adhere to three
fundamental assumptions (16): rst, the IVs are truly associated
with the exposures; second, the IVs are independent of confounders
of the relationship between exposures and outcomes (GrimAgeAccel
and PhenoAgeAccel); third, the IVs inuence the outcomes only
through the exposures, but not any direct or indirect pathways. All
data used in this MR study are publicly available. Ethical approval
and informed consent had been obtained in all original studies.
Selection Rationale and Data Sources of Genetic
Instruments
We selected 19 common modiable factors, including SES (educa-
tion and household income), lifestyle behaviors (smoking initiation,
alcohol intake, coffee consumption, daytime napping, sleep dur-
ation, and moderate-to-vigorous physical activity [MVPA]), and
cardiometabolic traits (body mass index [BMI], waist circumfer-
ence, body fat percentage [BF%], childhood obesity, type 2 diabetes,
low-density lipoprotein [LDL] cholesterol, high-density lipopro-
tein [HDL] cholesterol, triglycerides, systolic blood pressure [SBP],
diastolic blood pressure [DBP], and C-reactive protein [CRP]).
Denitions of the 19 modiable factors are shown in Supplementary
Table 2.
We extracted genetic variants for each of the modiable factors
from the most authoritative and appropriate genome-wide asso-
ciation studies (GWASs) of European ancestry (17–29), ensuring
minimum sample overlap with the GWAS of GrimAgeAccel and
PhenoAgeAccel (Supplementary Table 3). We included single nucleo-
tide polymorphisms (SNPs) robustly associated with the 19 modi-
able factors at the genome-wide signicance (p<5×10−8). To select
independent genetic variants, a stringent condition (linkage disequi-
librium threshold of r2 < 0.01) was set to minimize the inuence of
linkage disequilibrium which might bias the results of randomized
allele allocation. Where SNPs for the exposures were not available in
the GWAS summary statistics of GrimAgeAccel or PhenoAgeAccel,
we used proxies of SNPs with r2 > 0.8 as substitutes, by using the
LDproxy search on the online platform LDlink (https://ldlink.nci.
nih.gov/) (30).
Data Source for Epigenetic-Age Acceleration
The genetic associations with GrimAgeAccel and PhenoAgeAccel
were extracted from a recent GWAS meta-analysis (summary stat-
istics available at https://datashare.is.ed.ac.uk/handle/10283/3645),
which included 34710 European participants from 28 cohorts (31).
GrimAgeAccel and PhenoAgeAccel are 2-generation epigenetic-age
acceleration indicators, expressing the biological aging rate in years,
of which GrimAgeAccel is more strongly associated with mortality
than PhenoAgeAccel (11). Detailed denitions of GrimAgeAccel
and PhenoAgeAccel and data preparation in GWAS are shown in
Supplementary Table 2.
Figure 1. Study design and assumptions of the MR analysis. Assumption
1 indicates that the genetic variants proposed as instrumental variables
should be robustly associated with the exposures; assumption 2 indicates
that the used instrumental variables should not be associated with potential
confounders of the relationship between exposures and outcomes; and
assumption 3 indicates that the selected instrumental variables should
influence the outcomes only through the exposures, not via alternative
pathways. GrimAgeAccel = epigenetic-age acceleration obtained using
the GrimAge clock; HDL = high-density lipoprotein; LDL = low-density
lipoprotein; MR = Mendelian randomization; PhenoAgeAccel = epigenetic-
age acceleration obtained using the PhenoAge clock; PRESSO= pleiotropy
residual sum and outlier.
1084 Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

Table 1. MR Results of the Associations Between 19 Modifiable Factors and GrimAgeAccel
Modiable Factor
No. of
SNPs F-Statistic
Main Analysis Sensitivity Analyses
IVW Weighted Median MR-Egger MR-PRESSO
β (SE), year p Value q Value β (SE), years p Value q Value Β (SE), years p Value q Value
No. of
outliers β (SE), years p Value q Value
Socioeconomic status
Education 751 50 −1.143 (0.121) 3.31E−21 3.14E−20 −1.129 (0.172) 5.47E−11 5.20E−10 −0.829 (0.452) 6.70E−02 2.93E−01 4 −1.116 (0.116) 6.54E-21 6.21E-20
Household income 51 41 −0.774 (0.263) 3.24E−03 8.79E−03 −0.710 (0.378) 6.03E−02 1.43E−01 −1.298 (1.322) 3.31E−01 6.29E−01 0 −0.764 (0.255) 4.14E-03 1.12E-02
Lifestyle behavior
Smoking initiation 299 46 1.299 (0.107) 4.90E−34 9.31E−33 1.257 (0.157) 1.42E−15 2.70E−14 1.321 (0.446) 3.00E−03 2.85E−02 0 1.276 (0.103) 7.63E-29 1.45E-27
Alcohol intake 80 66 0.899 (0.361) 1.30E−02 2.47E−02 0.380 (0.567) 5.02E−01 5.96E−01 0.481 (0.738) 5.16E−01 7.56E−01 0 0.967 (0.357) 8.00E-03 1.69E-02
Coffee consumption 12 176 0.003 (0.005) 5.89E−01 6.22E−01 −0.003 (0.006) 6.15E−01 6.87E−01 −0.008 (0.010) 4.64E−01 7.56E−01 0 0.003 (0.005) 6.00E-01 6.33E-01
Sleep
Daytime napping 115 47 0.805 (0.355) 2.30E−02 3.64E−02 0.524 (0.514) 3.08E−01 4.40E−01 0.411 (1.235) 7.40E−01 8.27E−01 0 0.813 (0.346) 2.00E-02 3.17E-02
Sleep duration 77 40 −0.301 (0.266) 2.57E−01 3.49E−01 −0.403 (0.400) 3.13E−01 4.40E−01 −0.077 (1.017) 9.40E−01 9.40E−01 0 −0.275 (0.261) 2.94E-01 3.72E-01
MVPA 6 40 −0.186 (0.903) 8.37E−01 8.37E−01 −0.036 (1.198) 9.76E−01 9.76E−01 −9.279 (5.557) 1.70E−01 5.38E−01 0 −0.186 (0.763) 8.17E-01 8.17E-01
Cardiometabolic trait
Adiposity
BMI 941 59 0.592 (0.079) 9.16E−14 5.80E−13 0.490 (0.124) 7.92E−05 3.76E−04 0.790 (0.231) 1.00E−03 1.90E−02 2 0.595 (0.078) 6.30E-14 3.99E-13
Waist circumference 44 53 0.815 (0.184) 9.26E−06 2.93E−05 0.776 (0.272) 4.00E−03 1.52E−02 0.913 (0.504) 7.70E−02 2.93E−01 0 0.815 (0.184) 6.31E-05 2.00E-04
BF% 641 50 0.748 (0.120) 4.12E−10 1.96E−09 0.740 (0.182) 4.87E−05 3.08E−04 0.843 (0.400) 3.60E−02 2.28E−01 3 0.796 (0.115) 1.19E-11 5.65E-11
Childhood obesity 5 43 0.200 (0.075) 7.00E−03 1.48E−02 0.142 (0.098) 1.46E−01 2.77E−01 0.789 (0.554) 2.49E−01 5.91E−01 0 0.204 (0.057) 1.60E-02 2.76E-02
Type 2 diabetes 232 62 0.095 (0.041) 2.00E−02 3.45E−02 0.084 (0.071) 2.36E−01 4.08E−01 0.043 (0.092) 6.44E−01 7.65E−01 0 0.100 (0.040) 1.30E-02 2.47E-02
Lipids
LDL cholesterol 145 146 −0.094 (0.096) 3.27E−01 3.88E−01 −0.064 (0.163) 6.95E−01 7.34E−01 0.016 (0.142) 9.08E−01 9.40E−01 1 −0.069 (0.094) 4.63E-01 5.39E-01
HDL cholesterol 223 138 −0.181 (0.095) 5.60E−02 8.18E−02 −0.267 (0.138) 5.30E−02 1.43E−01 −0.078 (0.162) 6.32E−01 7.65E−01 2 −0.175 (0.090) 5.40E-02 7.89E-02
Triglycerides 173 116 0.249 (0.091) 6.00E−03 1.43E−02 0.239 (0.144) 9.60E−02 2.03E−01 0.152 (0.152) 3.19E−01 6.29E−01 0 0.254 (0.091) 6.00E-03 1.43E-02
Blood pressure
SBP 222 52 0.013 (0.012) 2.79E−01 3.53E−01 0.015 (0.018) 4.21E−01 5.33E−01 0.029 (0.049) 5.57E−01 7.56E−01 0 0.013 (0.012) 2.82E-01 3.72E-01
DBP 264 50 0.011 (0.019) 5.65E−01 6.22E−01 0.027 (0.027) 3.24E−01 4.40E−01 0.083 (0.066) 2.13E−01 5.78E−01 0 0.013 (0.018) 4.82E-01 5.39E-01
CRP 299 170 0.345 (0.073) 2.31E−06 8.78E−06 0.245 (0.107) 2.20E−02 6.97E−02 0.060 (0.100) 5.53E−01 7.56E−01 1 0.348 (0.071) 1.64E-06 6.23E-06
Notes: GrimAgeAccel represents epigenetic-age acceleration obtained using the GrimAge clock. The q value represents the false discovery rate (FDR) − adjusted p value. β represents the associations of each modiable
factor with GrimAgeAccel, and the unit for each risk factor is: 1-SD increase in years of schooling; 1-SD increase in household income; Ever smoked regularly compared to never smoked; 1-SD increase in log-transformed
alcoholic drinks per week; 1%-change in coffee consumption; 1-unit increase in napping category (never, sometimes, usually); 1-h/d increase in sleep duration; 1-SD increase in MET-min/wk of MVPA; 1-SD increase in BMI;
1-SD increase in waist circumference; 1-SD increase in BF%; 1-unit increase in log-transformed odds of childhood obesity; 1-unit increase in log-transformed odds of type 2 diabetes; 1-SD increase in LDL cholesterol; 1-SD
increase in HDL cholesterol; 1-SD increase in triglycerides; 1-mmHg increase in SBP; 1-mmHg increase in DBP; 1-SD increase in serum CRP levels. BF%=body fat percentage; BMI=body mass index; CRP=C-reactive
protein; DBP=diastolic blood pressure; HDL=high-density lipoprotein; IVW=Inverse-variance weighted; LDL=low-density lipoprotein; MVPA=moderate-to-vigorous physical activity; MR=Mendelian randomization;
MR-PRESSO=Mendelian randomization pleiotropy residual sum and outlier; No=number; SBP=systolic blood pressure; SD=standard deviation; SNP=single nucleotide polymorphism.
Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7 1085
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

Statistical Analyses
In the main analysis, we used the inverse-variance weighted (IVW)
method to determine MR causal estimates (β coefcients with
standard errors [SEs]) for associations of each modiable factor
with GrimAgeAccel and PhenoAgeAccel. The IVW combines the
Wald ratio estimates of every single SNP in the set of IVs into one
causal estimate using the random-effects meta-analysis approach
(32). To evaluate the robustness of the IVW estimates under different
assumptions and to detect possible pleiotropy, we performed 3 sen-
sitivity analyses, including the MR weighted median, the MR-Egger,
and the MR pleiotropy residual sum and outlier (MR-PRESSO)
methods (Supplementary Table 4). The weighted median method
selects the median MR estimate as the causal estimate and provides
a consistent causal estimate if over 50% of the weight in the analysis
was derived from valid IVs (33). The MR-Egger method which al-
lows the intercept to be freely estimated as an indicator of pleiotropy
was used to identify and adjust for the potential directional pleio-
tropic bias but has limited precision (34). The MR-PRESSO method
detects outlying SNPs that are potentially horizontally pleiotropic
and evaluates whether the exclusion of outlying SNPs inuences the
causal estimates under the assumption that the largest group of can-
didate instruments with similar estimates is the group of valid IVs
(35). We evaluated the heterogeneity for the IVW estimates using
the Cochran’s Q test (36) and identied the horizontal pleiotropy
based on the p value for the intercept in the MR-Egger model (34).
The F-statistics were calculated to measure the strength of IVs,
with larger F-statistics indicating more valid IVs. Afalse discovery
rate (FDR) method was used to correct results for multiple testing,
and FDR q values were provided. In this study, strong causal evi-
dence was dened as an association supported by the main analysis
(FDR q value < 0.05) and at least 1 sensitivity analysis. Suggestive
causal evidence was dened as a suggestive association with p < .05
and FDR q value ≥ 0.05 in the main analysis. Null causal evidence
was dened as no statistically signicant association revealed from
the main analysis (p ≥ .05). Given that the phenotype of GrimAge
is partly determined by smoking, we also conducted leave-one-out
analyses to evaluate the inuence of individual genetic variants of
smoking initiation on the association with GrimAgeAccel.
We further conducted multivariable MR analyses to assess
whether the causal effects of the strongest protective factor and the
strongest risk factor on GrimAgeAccel and PhenoAgeAccel were in-
dependent of other exposure factors (37). Taking into account the ef-
fect size and signicance of causal associations, we selected exposure
factors with β coefcients >0.5 and p < .05 in the main analysis as
covariates in the adjustmentmodels.
Moreover, we performed bidirectional MR analyses to investigate
potential reverse causality from GrimAgeAccel and PhenoAgeAccel to
these modiable socioeconomic, lifestyle, and cardiometabolic factors.
The two-sample MR analyses were conducted with the R pack-
ages “TwoSampleMR” and “MRPRESSO,” and the FDR q values
were estimated using the R package “fdrtool,” in R software (version
4.1.1; R Foundation for Statistical Computing, Vienna, Austria).
Results
Univariable MR Estimates for the Causal Effects of
19 Modifiable Factors on GrimAgeAccel
Ten out of the 19 risk factors showed strong associations with in-
creased GrimAgeAccel after FDR adjustment for multiple com-
parisons (Table 1). Smoking initiation was the strongest risk factor
(β [SE]: 1.299 [0.107] year) for GrimAgeAccel, followed by higher
alcohol intake (β [SE] per 1-standard deviation [SD] increase: 0.899
[0.361] year), higher waist circumference (β [SE] per 1-SD increase:
0.815 [0.184] year), daytime napping (β [SE]: 0.805 [0.355] year),
higher BF% (β [SE] per 1-SD increase: 0.748 [0.120] year), higher
BMI (β [SE] per 1-SD increase: 0.592 [0.079] year), higher CRP (β
[SE] per 1-SD increase: 0.345 [0.073] year), higher triglycerides (β
[SE] per 1-SD increase: 0.249 [0.091] year), childhood obesity (β
[SE]: 0.200 [0.075] year), and type 2 diabetes (β [SE]: 0.095 [0.041]
year; Figure 2A). By contrast, education in years of schooling (β [SE]
per 1-SD increase: −1.143 [0.121] year) and household income (β
[SE] per 1-SD increase: −0.774 [0.263] year) were protective fac-
tors associated with decreased GrimAgeAccel. There was little evi-
dence to support a causal association of other modiable factors
with GrimAgeAccel.
Associations of the above 12 modiable factors with
GrimAgeAccel were robust across sensitivity analyses with con-
sistent effect directions and p < .05 in at least one sensitivity analysis.
The mean F-statistics for the genetic instruments were 40 or greater,
indicating the limited potential for weak instrument bias (Table 1).
The MR-Egger intercept tests indicated potential pleiotropy for CRP
(pintercept < .05; Supplementary Table 5). Cochran’s Q test showed
possible heterogeneity for education, alcohol intake, BMI, BF%,
type 2 diabetes, LDL cholesterol, HDL cholesterol, and CRP (ph <
.05; Supplementary Table 5). With the exclusion of outlying SNPs,
the MR-PRESSO analysis showed consistent results with the IVW
analysis (Table 1). Leave-one-out analysis revealed that no single
SNP drove the MR estimate of smoking initiation on GrimAgeAccel
(Supplementary Figure 1).
For the reverse causality from GrimAgeAccel to modiable fac-
tors, genetically determined each 1-year increased GrimAgeAccel
was associated with lower CRP (β [SE]: −0.042 [0.006] SD), lower
BMI (β [SE]: −0.032 [0.006] SD), and higher alcohol intake (β [SE]:
0.018 [0.007] SD; Supplementary Table 6).
Figure 2. Overview of the findings on associations of 19 modifiable factors
with GrimAgeAccel and PhenoAgeAccel. (A) Causal associations between 19
modifiable factors and GrimAgeAccel; (B) Causal associations between 19
modifiable factors and PhenoAgeAccel. GrimAgeAccel represents epigenetic-
age acceleration obtained using the GrimAge clock; PhenoAgeAccel
represents epigenetic-age acceleration obtained using the PhenoAge
clock; β represents the effect of each modifiable factor on epigenetic-age
acceleration. Red box indicates a strong association with p value of <.05 and
FDR q value < 0.05. Blue box indicates a suggestive association with p value
of <.05 and FDR q value of ≥0.05. Gray box indicates null association with p
value of ≥.05. The IVW method was used for the main analysis. Sensitivity
analyses included the MR-PRESSO, the MR-WM, and the MR-Egger methods.
BF%=body fat percentage; BMI=body mass index; CRP=C-reactive protein;
DBP=diastolic blood pressure; FDR=false discovery rate; HDL=high-density
lipoprotein; IVW=inverse-variance weighted; LDL=low-density lipoprotein;
MR = Mendelian randomization; MVPA = moderate-to-vigorous physical
activity; PRESSO=pleiotropy residual sum and outlier; SBP=systolic blood
pressure; SD=standard deviation; WM=weighted median.
1086 Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

Table 2. MR Results of the Associations Between 19 Modifiable Factors and PhenoAgeAccel
Modiable factor
No. of
SNP F-statistic
Main analysis Sensitivity analyses
IVW Weighted median MR-Egger MR-PRESSO
β (SE), year p Value q Value β (SE), year p Value q Value β (SE), year p Value q Value
No. of
outlier β (SE), year p Value q Value
Socioeconomic factor
Education 751 50 −0.718 (0.151) 1.83E-06 1.74E-05 −0.633 (0.225) 5.00E-03 3.17E-02 −0.361 (0.565) 5.23E-01 9.03E-01 3 −0.689 (0.145) 2.52E-06 2.39E-05
Household income 51 41 −0.565 (0.367) 1.24E-01 2.06E-01 −0.626 (0.466) 1.78E-01 3.64E-01 −1.340 (1.862) 4.75E-01 9.03E-01 0 −0.620 (0.360) 9.12E-02 1.46E-01
Lifestyle behavior
Smoking initiation 299 46 0.519 (0.142) 2.43E-04 9.46E-04 0.404 (0.205) 4.90E-02 2.33E-01 1.222 (0.589) 3.90E-02 2.53E-01 1 0.545 (0.135) 7.20E-05 3.42E-04
Alcohol intake 80 66 0.669 (0.441) 1.30E-01 2.06E-01 1.134 (0.656) 8.40E-02 2.66E-01 1.693 (0.923) 7.00E-02 2.66E-01 0 0.737 (0.432) 9.20E-02 1.46E-01
Coffee consumption 12 176 0.004 (0.008) 6.00E-01 7.30E-01 0.002 (0.007) 7.73E-01 9.79E-01 -0.003 (0.015) 8.61E-01 9.27E-01 0 0.004 (0.008) 6.10E-01 7.03E-01
Sleep
Daytime napping 115 47 0.115 (0.500) 8.18E-01 8.18E-01 0.014 (0.665) 9.84E-01 9.91E-01 0.152 (1.742) 9.31E-01 9.31E-01 0 0.083 (0.488) 8.66E-01 8.66E-01
Sleep duration 77 40 −0.472 (0.357) 1.87E-01 2.73E-01 −0.526 (0.462) 2.55E-01 4.04E-01 2.771 (1.324) 4.00E-02 2.53E-01 0 −0.319 (0.357) 3.75E-01 5.09E-01
MVPA 6 40 0.368 (1.138) 7.47E-01 7.89E-01 0.262 (1.347) 8.46E-01 9.91E-01 −2.565 (6.947) 7.31E-01 9.27E-01 0 0.368 (0.649) 5.96E-01 7.03E-01
Cardiometabolic trait
Adiposity
BMI 941 59 0.586 (0.102) 1.08E-08 2.05E-07 0.680 (0.170) 6.33E-05 1.20E-03 0.510 (0.298) 8.70E-02 2.76E-01 4 0.605 (0.099) 1.53E-09 2.91E-08
Waist circumference 44 53 0.850 (0.269) 2.00E-03 6.33E-03 0.632 (0.339) 6.20E-02 2.36E-01 0.739 (0.740) 3.24E-01 7.43E-01 1 0.771 (0.249) 3.00E-03 9.50E-03
BF% 641 50 0.711 (0.152) 2.88E-06 1.82E-05 0.734 (0.227) 1.00E-03 9.50E-03 1.666 (0.509) 1.00E-03 1.90E-02 4 0.664 (0.145) 5.75E-06 3.64E-05
Childhood obesity 5 43 0.229 (0.095) 1.60E-02 3.80E-02 0.154 (0.123) 2.11E-01 3.64E-01 0.140 (0.724) 8.59E-01 9.27E-01 0 0.223 (0.071) 2.50E-02 5.28E-02
Type 2 diabetes 232 62 0.125 (0.051) 1.40E-02 3.80E-02 0.102 (0.080) 2.05E-01 3.64E-01 0.056 (0.116) 6.28E-01 9.18E-01 0 0.120 (0.049) 1.60E-02 3.80E-02
Lipids
LDL cholesterol 145 146 0.055 (0.123) 6.53E-01 7.30E-01 −0.002 (0.174) 9.91E-01 9.91E-01 0.169 (0.181) 3.52E-01 7.43E-01 1 0.059 (0.123) 6.33E-01 7.03E-01
HDL cholesterol 223 138 −0.127 (0.109) 2.43E-01 3.30E-01 0.068 (0.166) 6.84E-01 9.28E-01 0.103 (0.187) 5.83E-01 9.18E-01 0 −0.121 (0.109) 2.67E-01 3.90E-01
Triglycerides 173 116 −0.058 (0.121) 6.35E-01 7.30E-01 0.009 (0.196) 9.64E-01 9.91E-01 −0.202 (0.203) 3.20E-01 7.43E-01 0 −0.052 (0.121) 6.66E-01 7.03E-01
Blood pressure
SBP 222 52 0.032 (0.016) 5.30E-02 1.01E-01 0.033 (0.023) 1.49E-01 3.56E-01 −0.013 (0.065) 8.40E-01 9.27E-01 1 0.038 (0.016) 1.50E-02 3.80E-02
DBP 264 50 0.049 (0.024) 3.70E-02 7.81E-02 0.048 (0.034) 1.50E-01 3.56E-01 0.161 (0.084) 5.50E-02 2.61E-01 0 0.042 (0.023) 6.90E-02 1.31E-01
CRP 298 170 0.349 (0.095) 2.49E-04 9.46E-04 0.061 (0.150) 6.84E-01 9.28E-01 0.020 (0.132) 8.78E-01 9.27E-01 1 0.362 (0.093) 1.16E-04 4.41E-04
Notes: PhenoAgeAccel represents epigenetic-age acceleration obtained using the PhenoAge clock; The q value represents the false discovery rate (FDR)‐adjusted p value. β represents the associations of each modiable
factor with PhenoAgeAccel, and the unit for each risk factor is: 1-SD increase in years of educational attainment; 1-SD increase in household income; Ever smoked regularly compared to never smoked; 1-SD increase in
log-transformed alcoholic drinks per week; 1%-change in coffee consumption; 1-unit increase in napping category (never, sometimes, usually); 1-h/d increase in sleep duration; 1-SD increase in MET-min/wk of MVPA; 1-SD
increase in BMI; 1-SD increase in waist circumference; 1-SD increase in BF%; 1-unit increase in log-transformed odds of childhood obesity; 1-unit increase in log-transformed odds of type 2 diabetes; 1-SD increase in LDL
cholesterol; 1-SD increase in HDL cholesterol; 1-SD increase in triglycerides; 1-mmHg increase in SBP; 1-mmHg increase in DBP; 1-SD increase in serum CRP levels. BF%=body fat percentage; BMI=body mass index;
CRP=C-reactive protein; DBP=diastolic blood pressure; HDL=high-density lipoprotein; IVW=inverse-variance weighted; LDL=low-density lipoprotein; MVPA=moderate-to-vigorous physical activity; MR=Mendel-
ian randomization; MR-PRESSO=Mendelian randomization pleiotropy residual sum and outlier; No=number; SBP=systolic blood pressure; SD=standard deviation; SNP=single nucleotide polymorphism.
Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7 1087
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

Univariable MR Estimates for the Causal Effects of
19 Modifiable Factors on PhenoAgeAccel
Eight modiable factors showed strong associations with
PhenoAgeAccel after FDR adjustment for multiple comparisons
(Table 2). Higher waist circumference was the strongest risk factor
(β [SE] per 1-SD increase: 0.850 [0.269] year) for PhenoAgeAccel,
followed by higher BF% (β [SE] per 1-SD increase: 0.711 [0.152]
year), higher BMI (β [SE] per 1-SD increase: 0.586 [0.102] year),
smoking initiation (β [SE]: 0.519 [0.142] year), higher CRP (β [SE]
per 1-SD increase: 0.349 [0.095] year), childhood obesity (β [SE]:
0.229 [0.095] year), and type 2 diabetes (β [SE]: 0.125 [0.051] year;
Figure 2B). Whereas the genetically predicted higher education in
years of schooling was associated with decreased PhenoAgeAccel
(β [SE] per 1-SD increase: −0.718 [0.151] year). Suggestive caus-
ality was identied for the association between genetically predicted
higher DBP and increased PhenoAgeAccel (β [SE] per 1-mmHg in-
crease: 0.049 [0.024] year). No signicant association was observed
with PhenoAgeAccel for the other modiable factors.
Associations for the above eight modiable factors were ro-
bust in sensitivity analyses with consistent effect directions and p
< .05 in at least one sensitivity analysis. The mean F-statistics for
the genetic instruments were above 40, indicating the limited po-
tential for weak instrument bias (Table 2). There was no evidence
of pleiotropy for these risk factors except for CRP (pintercept < .05;
Supplementary Table 7). Potential heterogeneity was observed for
education, smoking initiation, BMI, waist circumference, BF%, type
2 diabetes, and CRP (ph < .05). One to four outliers were detected in
the MR-PRESSO analyses; however, the associations remained con-
sistent after the removal of these outliers (Table 2). For SBP, after
excluding the outlying SNP rs62523863, the MR-PRESSO analysis
revealed a potentially positive association between genetically pre-
dicted higher SBP and PhenoAgeAccel (β [SE] per 1-mmHg increase:
0.038 [0.016]year).
For the reverse causality from PhenoAgeAccel to modiable fac-
tors, genetically determined each 1-year increased PhenoAgeAccel
was associated with decreased napping (β [SE]: −0.004 [0.002]
unit) and sleep duration (β [SE]: −0.004 [0.002] h/d; Supplementary
Table8).
Multivariable MR Estimates for the Independent
Effects of the Strongest Protective and Risk Factors
on GrimAgeAccel and PhenoAgeAccel
In multivariable MR analyses, the inverse association between edu-
cation (the strongest protective factor) and GrimAgeAccel remained
signicant with adjustment for household income, smoking initi-
ation, alcohol intake, waist circumference, daytime napping, BF%,
or BMI (Figure 3A, Supplementary Table 9). Likewise, the positive
association between smoking initiation (the strongest risk factor) and
GrimAgeAccel persisted after adjusting for each of these covariates.
All covariates in multivariable MR analyses were selected based on
their considerable effects on GrimAgeAccel in terms of effect size
(β coefcients >0.5) and signicance (p < .05 in the main analysis).
The associations of education (the strongest protective factor) and
waist circumference (the strongest risk factor) with PhenoAgeAccel
were not substantially changed after adjustment for waist circum-
ference, BF%, BMI, or smoking initiation, which had considerable
causal effects on PhenoAgeAccel (β coefcients > 0.5; Figure 3B).
Discussion
The MR study delineated potential causal relationships of 19
common modiable factors with GrimAgeAccel and PhenoAgeAccel,
the robust second-generation epigenetic-age acceleration indicators.
We identied strong evidence for 12 and 8 factors causally associ-
ated with GrimAgeAccel and PhenoAgeAccel, respectively. Smoking
initiation exhibited the greatest effect on increased GrimAgeAccel
(1.299years), followed by higher alcohol intake, higher waist cir-
cumference, daytime napping, higher BF%, higher BMI, higher CRP,
higher triglycerides, childhood obesity, and type 2 diabetes; whereas
education showed the greatest effect on decreased GrimAgeAccel
(−1.143 years per 1-SD increase in years of schooling), followed
by household income. Higher waist circumference and education
were the leading causal risk and protective factors associated with
PhenoAgeAccel, respectively; BF%, BMI, smoking initiation, CRP,
childhood obesity, and type 2 diabetes were also associated with
increased PhenoAgeAccel. Multiple sensitivity analyses strength-
ened the robustness of these causal relationships. Multivariable
MR analyses further demonstrated the independent causal effects
of the strongest risk and protective factors on GrimAgeAccel and
PhenoAgeAccel, respectively.
In this study, education was the major protective factor for
both GrimAgeAccel and PhenoAgeAccel, and this causal effect was
largely independent of other related factors, such as income, lifestyle,
Figure 3. Multi variable MR assessing the effects of the strongest protective
factor and risk factor on GrimAgeAccel and PhenoAgeAccel. (A) Effect of the
strongest protective factor and risk factor on GrimAgeAccel; (B) Effect of the
strongest protective factor and risk factor on PhenoAgeAccel. GrimAgeAccel
represents epigenetic-age acceleration obtained using the GrimAge clock;
PhenoAgeAccel represents epigenetic-age acceleration obtained using
the PhenoAge clock; Causal estimates are β (95% CI) in years. Exposure
factors with β coefficients >0.5 and p value of <.05 in the main analysis as
shown in Figure 2 were selected as covariates in the adjustment models.
BF%=body fat percentage; BMI=body mass index; CI=confidence intervals;
MR=Mendelian randomization.
1088 Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

and adiposity indicators. By contrast, the protective causal effect of
income on GrimAgeAccel did not persist after adjustment for educa-
tion, suggesting that this effect was largely inuenced by education.
Education is a strong proxy for SES and a more upstream deter-
minant of health, with broad implications for a person’s life-long
lifestyle behaviors and health-promoting resources (38,39). A re-
cent study using UK Biobank data has documented that each 1-year
increase in genetically determined education was associated with
equivalently 4.2 years of age-related increases in telomere length
(40). Telomere length and epigenetic-age acceleration metrics (eg,
GrimAgeAccel and PhenoAgeAccel) point toward distinct mech-
anisms of the aging process that are marked by telomeres and the
DNA methylation-based epigenetic clocks, both of which are inde-
pendently associated with chronological age and mortality risk (41).
Our ndings, together with those of the UK Biobank, highlight the
important impact of education on biological aging rates from two
different aspects of the aging process. Therefore, public health strat-
egies aimed at reducing educational inequalities and improving edu-
cation may slow the biological aging rate and help reduce age-related
health burdens.
This MR study identied several common lifestyle behaviors,
including smoking initiation, alcohol intake, and daytime napping,
causally associated with increased GrimAgeAccel or PhenoAgeAccel
(smoking initiation only), which were consistent with previous
evidence from observational studies (6,7). Smoking methylation
proxy is a component of the GrimAge clock (7), thus it is not sur-
prising that smoking initiation exhibited a large effect on increased
GrimAgeAccel in this study. Our multivariable MR analysis further
conrmed that the effect of smoking on GrimAgeAccel was inde-
pendent of other causal lifestyle behaviors and adiposity indicators.
Various sensitivity analyses also supported the robustness of the
causal association between smoking initiation and GrimAgeAccel.
Similarly, genetically determined smoking has been associated with
shorter telomere length in the UK Biobank (40), and evidence from
the Danish Health Interview Survey suggested that the life expectancy
of a heavy smoker was a little more than seven years shorter than
that of a never smoker (42). Our study also provided strong MR evi-
dence that genetically predicted higher alcohol intake was associated
with increased GrimAgeAccel. Astudy composed of the Hannum
cohort and the Family and Community Health Studies cohort found
that the relationship between alcohol use and the rst-generation
epigenetic clocks seemed to be nonlinear (43). However, given the
dose-dependent relationship of alcohol intake with all-cause mor-
tality and cancers (44), our ndings suggest that reducing alcohol in-
take is necessary to decrease GrimAgeAccel and retard overall health
loss. Previous observational studies also reported positive or inverse
correlations of other lifestyle behaviors such as MVPA, coffee, and
sleep duration with GrimAgeAccel, PhenoAgeAccel, or mortality
(11,45,46). Nevertheless, in this study, there was little evidence
supporting that these associations were causal. The discrepancy be-
tween our ndings and previous observations may result from the
potential confounding or reverse causation in conventional obser-
vational studies, or because of insufcient power of MR analyses
due to the relatively low variance explained by genetic instruments
(eg, MVPA). Moreover, the nonlinear association patterns, as in the
case of sleep duration and mortality, might also partially explain the
inconsistent results (46). Therefore, our null ndings should be cau-
tiously interpreted.
Interestingly, of all cardiometabolic traits (ie, adiposity indica-
tors, type 2 diabetes, triglycerides, and CRP) which causally in-
creased GrimAgeAccel or PhenoAgeAccel, adiposity indicators were
the most dominant traits. Emerging observational studies have pro-
nounced the positive associations of BMI with GrimAgeAccel and
PhenoAgeAccel (11), and a meta-analysis of 87 observational studies
showed each 5kg/m2 higher BMI corresponded to about 1year of
age-related decrease in telomere length (47). Our study further pro-
vided strong evidence for causal associations of various adiposity
indicators, including waist circumference, BF%, BMI, and child-
hood obesity, with increased GrimAgeAccel and PhenoAgeAccel.
In this study, type 2 diabetes, triglycerides, and CRP showed strong
but modest effects on GrimAgeAccel or PhenoAgeAccel, which was
consistent with the ndings of the UK Biobank study on telomere
length (40). Previous cross-sectional studies have found that HDL
cholesterol, LDL cholesterol, and triglycerides were signicantly
related to epigenetic aging (6,7). The prospective Coronary Artery
Risk Development in Young Adults study of 1 118 White and
Black individuals indicated that high triglycerides and low HDL
cholesterol in early adulthood were associated with accelerated
epigenetic aging by midlife (48). In our MR study, only triglycer-
ides were causally associated with GrimAgeAccel, but neither HDL
cholesterol nor LDL cholesterol was identied as a causal factor
for GrimAgeAccel or PhenoAgeAccel. Our ndings suggested that
the signicant associations found in these observational studies
might be partially inuenced by residual confounding. Moreover,
the inammatory phenotype (eg, CRP), which tends to increase
with aging, has been recognized as an important risk factor for
age-related morbidity and mortality (49). Previous cross-sectional
studies have shown that CRP was signicantly related to epigen-
etic aging (6,7). However, one recent longitudinal study found that
although CRP was associated with epigenetic aging measures in
cross-sectional analysis, baseline CRP was not associated with epi-
genetic aging after 11years of follow-up (50). Our study extended
previous observational studies by adding evidence for a causal ef-
fect of CRP on epigeneticaging.
The GWAS for GrimAgeAccel and PhenoAgeAccel also incorp-
orated data-driven MR analyses to investigate causal relationships
between epigenetic aging and 150 UK Biobank traits and found that
adiposity, education, and CRP were causally associated with epigen-
etic aging (31). Based on this study, our hypothesis-driven MR study
further applied GWAS summary statistics from the most authorita-
tive and appropriate consortiums to enhance the precision of causal
estimates. We systematically investigated whether and to what extent
common and clinically signicant modiable factors, including SES,
lifestyle behaviors, and cardiometabolic traits, are likely to inuence
GrimAgeAccel and PhenoAgeAccel, in order to inform clinical prac-
tice and biological understanding. Importantly, we for the rst time
assessed whether these modiable factors had independent causal ef-
fects on GrimAgeAccel and PhenoAgeAccel using multivariable MR
analyses. This study brings us one step closer to understanding the
potential contributors to epigenetic aging and provides promising
intervention targets for healthy aging. Given the enormous burden
induced by age-related morbidity and mortality, strategies to re-
duce educational inequalities, promote healthy lifestyles primarily
through reducing smoking, alcohol intake, and daytime napping,
and improve cardiometabolic traits, specically adiposity, type 2
diabetes, triglycerides, and CRP, to slow epigenetic aging rate are
imminent.
In this study, we included independent and genome-wide signi-
cant SNPs as instruments for each of the modiable factors to en-
sure the rst MR assumption was fullled. Moreover, we applied
strict criteria strengthened by the FDR-corrected signicance and the
cross-validations by main and sensitivity analyses to draw robust
Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7 1089
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

causal conclusions. However, several limitations merited consider-
ation. First, we found potential pleiotropy from the MR-Egger inter-
cept test for CRP. However, we conducted the MR-PRESSO analysis,
and the association remained consistent after the removal of out-
lying SNPs. Second, we could not rule out the possibility that the
associations of certain modiable risk factors such as alcohol intake
and sleep duration with GrimAgeAccel or PhenoAgeAccel may be
nonlinear. Future studies with individual-level data are warranted
to conrm the linear or nonlinear relationships. Third, to ensure the
consistency of genetic background, this MR study was performed
only in European-ancestry participants, thus the generalization of
our results to other ethnic groups should be cautious. Forth, due
to the MR design, this study is not applicable to investigating the
nongenetically determined portion of the trait. However, based on
the MR study design as a natural experiment, the MR estimates
provided in this study are complementary to the previous observa-
tional estimates, providing additional causal information for clinical
practice.
In conclusion, this MR study provided novel quantitative evi-
dence on modiable causal risk factors for accelerated epigen-
etic aging, among which adiposity indicators, low education, and
smoking exhibited the most signicant inuence. Our ndings shed
light on the underlying contributors to biological aging and point
toward promising intervention targets to slow the biological aging
rate and promote healthy longevity.
Supplementary Material
Supplementary data are available at The Journals of Gerontology,
Series A: Biological Sciences and Medical Sciencesonline.
Funding
This work was supported by the grants from the National Natural Science
Foundation of China (82022011, 81970706, 82088102, 81970728), the
“Shanghai Municipal Education Commission–Gaofeng Clinical Medicine
Grant Support” from Shanghai Jiao Tong University School of Medicine
(20171901 Round 2), and the Innovative Research Team of High-level Local
Universities in Shanghai.
Conflict of Interest
None declared.
Acknowledgments
We gratefully acknowledge the investigators and participants of all genome-
wide association studies from which we used data.
Author Contributions
L.K.and T.W.contributed to the conception and design of the study. L.K.and
C.Y. contributed to statistical analysis. L.K. contributed to drafting of the
manuscript. T.W.guaranteed this work and take responsibility for the integrity
of the data. All authors contributed to acquisition or interpretation of data,
critical revision of the manuscript for important intellectual content, and nal
approval of the version to be published.
Ethics Approval
This study is based on publicly available summarized data. Ethical approval
and informed consent had been obtained in all original studies.
Data Availability
The genetic association data of the 19 modiable factors are available in
Supplementary Table 3. The GWAS summary statistics for GrimAgeAccel
and PhenoAgeAccel are available at https://datashare.is.ed.ac.uk/
handle/10283/3645. The analytical script of the MR analyses conducted in
this study is available via the GitHub repository of the “TwoSampleMR” R
package (https://github.com/MRCIEU/TwoSampleMR/).
References
1. Niccoli T, Partridge L. Ageing as a risk factor for disease. Curr Biol.
2012;22(17):R741–R752. doi:10.1016/j.cub.2012.07.024
2. Schumacher B, Pothof J, Vijg J, Hoeijmakers JHJ. The central role of
DNA damage in the ageing process. Nature. 2021;592(7856):695–703.
doi:10.1038/s41586-021-03307-7
3. WarnerHR. Current status of efforts to measure and modulate the bio-
logical rate of aging. J Gerontol ABiol Sci Med Sci. 2004;59(7):692–696.
doi:10.1093/gerona/59.7.b692
4. JylhäväJ, PedersenNL, HäggS. Biological age predictors. EBioMedicine.
2017;21:29–36. doi:10.1016/j.ebiom.2017.03.046
5. HorvathS, RajK. DNA methylation-based biomarkers and the epigenetic
clock theory of ageing. Nat Rev Genet. 2018;19(6):371–384. doi:10.1038/
s41576-018-0004-3
6. 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.
doi:10.18632/aging.101414
7. Lu AT, Quach A, WilsonJG, etal. DNA methylation GrimAge strongly
predicts lifespan and healthspan. Aging (Albany NY). 2019;11(2):303–
327. doi:10.18632/aging.101684
8. WangC, Ni W, YaoY, etal. DNA methylation-based biomarkers of age
acceleration and all-cause death, myocardial infarction, stroke, and cancer
in two cohorts: the NAS, and KORA F4. EBioMedicine. 2021;63:103151.
doi:10.1016/j.ebiom.2020.103151
9. LiA, Koch Z, IdekerT. Epigenetic aging: biological age prediction and
informing a mechanistic theory of aging. J Intern Med. 2022;292(5):733–
744. doi:10.1111/joim.13533
10. Quach A, Levine ME, Tanaka T, et al. Epigenetic clock analysis of
diet, exercise, education, and lifestyle factors. Aging (Albany NY).
2017;9(2):419–446. doi:10.18632/aging.101168
11. OblakL, vanderZaagJ, Higgins-ChenAT, LevineME, BoksMP. A sys-
tematic review of biological, social and environmental factors associated
with epigenetic clock acceleration. Ageing Res Rev. 2021;69:101348.
doi:10.1016/j.arr.2021.101348
12. SierraF, CaspiA, FortinskyRH, etal. Moving geroscience from the bench
to clinical care and health policy. J Am Geriatr Soc. 2021;69(9):2455–
2463. doi:10.1111/jgs.17301
13. CampisiJ, KapahiP, LithgowGJ, MelovS, NewmanJC, VerdinE. From
discoveries in ageing research to therapeutics for healthy ageing. Nature.
2019;571(7764):183–192. doi:10.1038/s41586-019-1365-2
14. Davies NM, Holmes MV, Davey Smith G. Reading Mendelian ran-
domisation studies: a guide, glossary, and checklist for clinicians. BMJ.
2018;362:k601. doi:10.1136/bmj.k601
15. Skrivankova VW, RichmondRC, WoolfBAR, et al. Strengthening the
Reporting of Observational Studies in Epidemiology using Mendelian
randomisation (STROBE-MR): explanation and elaboration. BMJ.
2021;375:n2233. doi:10.1136/bmj.n2233
16. ZhengJ, BairdD, BorgesMC, etal. Recent developments in Mendelian
randomization studies. Curr Epidemiol Rep. 2017;4(4):330–345.
doi:10.1007/s40471-017-0128-6
17. Lee JJ, WedowR, OkbayA, et al. Gene discovery and polygenic predic-
tion from a genome-wide association study of educational attainment in
1.1 million individuals. Nat Genet. 2018;50(8):1112–1121. doi:10.1038/
s41588-018-0147-3
18. LiuM, JiangY, WedowR, etal. Association studies of up to 1.2 million in-
dividuals yield new insights into the genetic etiology of tobacco and alcohol
use. Nat Genet. 2019;51(2):237–244. doi:10.1038/s41588-018-0307-5
1090 Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023

19. Zhong VW, KuangA, Danning RD, etal. A genome-wide association
study of bitter and sweet beverage consumption. Hum Mol Genet.
2019;28(14):2449–2457. doi:10.1093/hmg/ddz061
20. Dashti HS, Daghlas I, Lane JM, et al. Genetic determinants of day-
time napping and effects on cardiometabolic health. Nat Commun.
2021;12(1):900. doi:10.1038/s41467-020-20585-3
21. DashtiHS, Jones SE, WoodAR, et al. Genome-wide association study
identies genetic loci for self-reported habitual sleep duration supported
by accelerometer-derived estimates. Nat Commun. 2019;10(1):1100.
doi:10.1038/s41467-019-08917-4
22. KlimentidisYC, RaichlenDA, BeaJ, etal. Genome-wide association study
of habitual physical activity in over 377,000 UK Biobank participants
identies multiple variants including CADM2 and APOE. Int J Obes
(Lond). 2018;42(6):1161–1176. doi:10.1038/s41366-018-0120-3
23. YengoL, SidorenkoJ, Kemper KE, etal. Meta-analysis of genome-wide
association studies for height and body mass index in <700000 individ-
uals of European ancestry. Hum Mol Genet. 2018;27(20):3641–3649.
doi:10.1093/hmg/ddy271
24. Shungin D, Winkler TW, Croteau-Chonka DC, et al. New genetic
loci link adipose and insulin biology to body fat distribution. Nature.
2015;518(7538):187–196. doi:10.1038/nature14132
25. BradeldJP, TaalHR, TimpsonNJ, etal. A genome-wide association meta-
analysis identies new childhood obesity loci. Nat Genet. 2012;44(5):526–
531. doi:10.1038/ng.2247
26. MahajanA, TaliunD, ThurnerM, etal. Fine-mapping type 2 diabetes loci
to single-variant resolution using high-density imputation and islet-specic
epigenome maps. Nat Genet. 2018;50(11):1505–1513. doi:10.1038/
s41588-018-0241-6
27. Klarin D, DamrauerSM, Cho K, etal. Genetics of blood lipids among
~300,000 multi-ethnic participants of the Million Veteran Program. Nat
Genet. 2018;50(11):1514–1523. doi:10.1038/s41588-018-0222-9
28. EvangelouE, WarrenHR, Mosen-AnsorenaD, etal. Publisher correction:
genetic analysis of over 1 million people identies 535 new loci associated
with blood pressure traits. Nat Genet. 2018;50(12):1755. doi:10.1038/
s41588-018-0297-3
29. HanX, Ong JS, AnJ, HewittAW, GharahkhaniP, MacGregorS. Using
Mendelian randomization to evaluate the causal relationship between
serum C-reactive protein levels and age-related macular degeneration. Eur
J Epidemiol. 2020;35(2):139–146. doi:10.1007/s10654-019-00598-z
30. MachielaMJ, ChanockSJ. LDlink: a web-based application for exploring
population-specic haplotype structure and linking correlated alleles of
possible functional variants. Bioinformatics. 2015;31(21):3555–3557.
doi:10.1093/bioinformatics/btv402
31. McCartneyDL, Min JL, RichmondRC, etal. Genome-wide association
studies identify 137 genetic loci for DNA methylation biomarkers of
aging. Genome Biol. 2021;22(1):194. doi:10.1186/s13059-021-02398-9
32. Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G.
Mendelian randomization: using genes as instruments for making
causal inferences in epidemiology. Stat Med. 2008;27(8):1133–1163.
doi:10.1002/sim.3034
33. BowdenJ, DaveySmith G, Haycock PC, Burgess S. Consistent estima-
tion in Mendelian randomization with some invalid instruments using
a weighted median estimator. Genet Epidemiol. 2016;40(4):304–314.
doi:10.1002/gepi.21965
34. BurgessS, BowdenJ, FallT, IngelssonE, ThompsonSG. Sensitivity analyses
for robust causal inference from Mendelian randomization analyses with
multiple genetic variants. Epidemiology. 2017;28(1):30–42. doi:10.1097/
EDE.0000000000000559
35. VerbanckM, ChenCY, NealeB, DoR. Detection of widespread horizontal
pleiotropy in causal relationships inferred from Mendelian randomization
between complex traits and diseases. Nat Genet. 2018;50(5):693–698.
doi:10.1038/s41588-018-0099-7
36. Bowden J, Del Greco MF, Minelli C, Davey Smith G, Sheehan N,
ThompsonJ. A framework for the investigation of pleiotropy in two-sample
summary data Mendelian randomization. Stat Med. 2017;36(11):1783–
1802. doi:10.1002/sim.7221
37. BurgessS, ThompsonSG. Multivariable Mendelian randomization: the use
of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol.
2015;181(4):251–260. doi:10.1093/aje/kwu283
38. FigginsE, Pieruccini-FariaF, SpeechleyM, Montero-OdassoM. Potentially
modiable risk factors for slow gait in community-dwelling older adults:
a systematic review. Ageing Res Rev. 2021;66:101253. doi:10.1016/j.
arr.2020.101253
39. Wang Y, Ye C, Kong L, et al. Independent associations of educa-
tion, intelligence, and cognition with hypertension and the mediating
effects of cardiometabolic risk factors: a Mendelian randomiza-
tion study. Hypertension. 2023;80(1):192–203. doi:10.1161/
HYPERTENSIONAHA.122.20286
40. BountzioukaV, MusichaC, AllaraE, etal. Modiable traits, healthy be-
haviours, and leukocyte telomere length: a population-based study in UK
Biobank. Lancet Healthy Longev. 2022;3(5):e321–e331. doi:10.1016/
S2666-7568(22)00072-1
41. Marioni RE, HarrisSE, ShahS, etal. The epigenetic clock and telomere
length are independently associated with chronological age and mortality.
Int J Epidemiol. 2018;45(2):424–432. doi:10.1093/ije/dyw041
42. Brønnum-Hansen H, Juel K. Abstention from smoking extends life
and compresses morbidity: a population based study of health expect-
ancy among smokers and never smokers in Denmark. Tob Control.
2001;10(3):273–278. doi:10.1136/tc.10.3.273
43. BeachSR, DoganMV, LeiMK, etal. Methylomic aging as a window onto
the inuence of lifestyle: tobacco and alcohol use alter the rate of bio-
logical aging. J Am Geriatr Soc. 2015;63(12):2519–2525. doi:10.1111/
jgs.13830
44. GBD 2016 Alcohol and Drug Use Collaborators. The global burden of
disease attributable to alcohol and drug use in 195 countries and terri-
tories, 1990–2016a systematic analysis for the Global Burden of Disease
Study 2016. Lancet Psychiatry. 2018;5(12):987–1012. doi:10.1016/
s2215-0366(18)30337-7
45. Lofteld E, Cornelis MC, Caporaso N, Yu K, Sinha R, Freedman N.
Association of coffee drinking with mortality by genetic variation in
caffeine metabolism: ndings from the UK biobank. JAMA Intern Med.
2018;178(8):1086–1097. doi:10.1001/jamainternmed.2018.2425
46. WangC, BangdiwalaSI, RangarajanS, etal. Association of estimated sleep
duration and naps with mortality and cardiovascular events: a study of
116 632 people from 21 countries. Eur Heart J. 2019;40(20):1620–1629.
doi:10.1093/eurheartj/ehy695
47. GielenM, HagemanGJ, AntoniouEE, etal. Body mass index is negatively
associated with telomere length: a collaborative cross-sectional meta-
analysis of 87 observational studies. Am J Clin Nutr. 2018;108(3):453–
475. doi:10.1093/ajcn/nqy107
48. GaoT, WilkinsJT, ZhengY, etal. Plasma lipid proles in early adult-
hood are associated with epigenetic aging in the Coronary Artery Risk
Development in Young Adults (CARDIA) study. Clin Epigenetics.
2022;14(1):16. doi:10.1186/s13148-021-01222-2
49. López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G.
Hallmarks of aging: an expanding universe. Cell. 2023;186(2):243–278.
doi:10.1016/j.cell.2022.11.001
50. CribbL, HodgeAM, YuC, et al. Inammation and epigenetic aging are
largely independent markers of biological aging and mortality. J Gerontol
ABiol Sci Med Sci. 2022;77(12):2378–2386. doi:10.1093/gerona/glac147
Journals of Gerontology: BIOLOGICAL SCIENCES, 2023, Vol. 78, No. 7 1091
Downloaded from https://academic.oup.com/biomedgerontology/article/78/7/1083/7069451 by guest on 13 September 2023