The mean age of the participants (N = 3,813) was 70.0 years, and the majority were female (57.7%, Table 1). Participants self-identified as non-Hispanic White (67.2%), non-Hispanic Black (16.3%), Hispanic (13.3%), or some other group (3.1%). Over half of the participants had at least a high school degree (59.1%) and 24.3% had at least a college degree. A total of 44.2% were non-smokers, 44.7% were former smokers, and 11.1% were current smokers. The mean BMI was 28.9 kg/m2. Nearly half of the participants were using lipid-lowering medications (49.0%), and 66.4% of the participants reported fasting at the time of the blood draw.
Table 1
Descriptive characteristics of the Health and Retirement Study participants (N = 3,813)
Characteristic | Mean (SD) or N (%) |
Age (years) | 70.0 (9.3) |
Female sex | 2201 (57.7%) |
Race/ethnicity (self-reported) | |
Non-Hispanic White | 2564 (67.2%) |
Non-Hispanic Black | 623 (16.3%) |
Hispanic | 509 (13.3%) |
All other groups | 117 (3.1%) |
Educational attainment | |
Less than high school degree | 633 (16.6%) |
High school degree or equivalent | 2254 (59.1%) |
College degree and above | 926 (24.3%) |
BMI (kg/m2) | 28.9 (6.3) |
Smoking status | |
Never smoker | 1685 (44.2%) |
Former smoker | 1704 (44.7%) |
Current smoker | 424 (11.1%) |
Lipid-lowering medication use | 1869 (49.0%) |
Fasting at time of blood draw | 2531 (66.4%) |
Blood lipids |
Total cholesterol (mg/dL) (n = 3,811) | 187.6 (41.6) |
HDL-C (mg/dL)a | 57.0 (18.8) |
LDL-C (mg/dL) (n = 3,731) | 101.7 (35.0) |
Triglycerides (mg/dL) (n = 3,806)a | 144.8 (84.0) |
Epigenetic age acceleration | |
HorvathAA (years) | 0.017 (6.3) |
HannumAA (years) | -0.018 (5.1) |
PhenoAA (years) | -0.036 (6.8) |
GrimAA (years) | 0.009 (4.7) |
DunedinPACE | 1.038 (0.1) |
SD, standard deviation; TC, total cholesterol; HDL-C, high-density lipoprotein; LDL-C, low-density lipoprotein; TG, triglycerides; HorvathAA, HorvathAge acceleration; HannumAA, HannumAge acceleration; PhenoAA, LevineAge (PhenoAge) acceleration; GrimAA, GrimAge acceleration; BMI, body mass index
aHDL-C and triglycerides were natural-log transformed prior to analysis
Correlations among the epigenetic clocks ranged from 0.64 to 0.77 (Supplemental Table 1A), and correlations among DNAmAA measures ranged from − 0.03 to 0.60, with HorvathAA having particularly weak correlations with more recently developed DNAmAA measures (Supplemental Table 1B). All correlations had P-values < 0.05 except for HorvathAA and DunedinPACE. Older age was associated with lower levels of all lipid measures (P < 0.05) except for HDL-C (Supplemental Table 2). Being a female was associated with higher levels of TC, HDL-C, and LDL-C compared to males. Finally, compared with those who did not have a high school degree, having a high school degree or equivalent was associated with higher HDL-C, and having a college degree and above was associated with higher HDL-C and lower TG.
Higher DNAmAA was associated with lower levels of TC, HDL-C, and LDL-C, and higher levels of TG in Model 1 (P < 0.05; Table 2) for all DNAmAA measures except for HorvathAA. After further adjustment for BMI, smoking status, and educational attainment (Model 2), GrimAA and DunedinPACE remained associated with all lipid measures. All other associations remained significant although most had slightly attenuated effect estimates, except for PhenoAA with TC and HannumAA with LDL-C and TG, which were no longer significant. Among all significant associations in Model 2, DunedinPACE had the most significant associations with all blood lipids, and GrimAA had the second most significant associations. HannumAA and PhenoAA shared similar but smaller estimates. Specifically, a one-unit increase in DunedinPACE was associated with decreases of 38.74 mg/dL, 20.55 mg/dL, and 27.87 mg/dL in TC, HDL-C, and LDL-C, respectively, and an increase of 73.17 mg/dL in TG. Additionally, a one-year increase in GrimAA was associated with decreases of 0.66, 0.51, and 0.62 mg/dL in TC, HDL-C, and LDL-C, respectively, and an increase of 1.75 mg/dL in TG. Because HDL-C and TG were natural log-transformed prior to analyses, these effect size estimates, and those reported subsequently, correspond to a person with the mean blood lipid level in the sample (i.e., HDL-C at 57.0 mg/dL or TG at 144.8 mg/dL). Most of the associations remained significant in non-Hispanic White and non-Hispanic Black samples, with similar effect estimates. Few associations were detected in Hispanics (Supplemental Table 3).
Table 2
Associations between epigenetic age acceleration and blood lipids
| Model 1a | Model 2b |
| βc | SE | P | βc | SE | P |
TC (n = 3,811) | | | | | |
HorvathAA | 0.141 | 0.109 | 0.200 | 0.181 | 0.110 | 0.107 |
HannumAA | -0.322 | 0.113 | 0.007 | -0.261 | 0.112 | 0.024 |
PhenoAA | -0.254 | 0.105 | 0.019 | -0.195 | 0.103 | 0.065 |
GrimAA | -0.552 | 0.149 | 0.001 | -0.656 | 0.191 | 0.001 |
DunedinPACE | -41.90 | 4.051 | 8.35×10− 14 | -38.74 | 4.417 | 3.95×10− 11 |
ln(HDL-C) (n = 3,813) | | | | | |
HorvathAA | -0.0001 | 0.001 | 0.892 | 0.001 | 0.001 | 0.533 |
HannumAA | -0.004 | 0.001 | 0.002 | -0.002 | 0.001 | 0.050 |
PhenoAA | -0.004 | 0.001 | 8.49×10− 5 | -0.002 | 0.001 | 0.017 |
GrimAA | -0.011 | 0.001 | 3.84×10− 10 | -0.009 | 0.002 | 2.44×10− 5 |
DunedinPACE | -0.633 | 0.041 | 3.56×10− 20 | -0.447 | 0.044 | 4.08×10− 13 |
LDL-C (n = 3,731) | | | | | |
HorvathAA | 0.069 | 0.089 | 0.442 | 0.084 | 0.089 | 0.352 |
HannumAA | -0.220 | 0.106 | 0.043 | -0.202 | 0.106 | 0.064 |
PhenoAA | -0.239 | 0.091 | 0.011 | -0.220 | 0.090 | 0.018 |
GrimAA | -0.501 | 0.142 | 0.001 | -0.619 | 0.166 | 0.001 |
DunedinPACE | -26.83 | 4.471 | 2.50×10− 7 | -27.87 | 4.882 | 9.69×10− 7 |
ln(TG) (n = 3,806) | | | | | |
HorvathAA | 0.002 | 0.002 | 0.218 | 0.001 | 0.002 | 0.378 |
HannumAA | 0.004 | 0.002 | 0.042 | 0.002 | 0.002 | 0.241 |
PhenoAA | 0.006 | 0.002 | 4.76×10− 4 | 0.004 | 0.002 | 0.012 |
GrimAA | 0.014 | 0.002 | 2.90×10− 7 | 0.012 | 0.003 | 3.89×10− 5 |
DunedinPACE | 0.633 | 0.064 | 4.19×10− 13 | 0.409 | 0.073 | 1.35×10− 6 |
TC, total cholesterol; HDL-C, high-density lipoprotein; LDL-C, low-density lipoprotein; TG, triglycerides; HorvathAA, HorvathAge acceleration; HannumAA, HannumAge acceleration; PhenoAA, PhenoAge acceleration; GrimAA, GrimAge acceleration
aModel 1: blood lipid level = epigenetic age acceleration + age at methylation measurement + sex + race/ethnicity + fasting status + lipid-lowering medication use
bModel 2: Model 1 + body mass index + smoking status + educational attainment (less than high school degree, high school degree or equivalent, college degree and above)
cβ corresponds to the change in blood lipid level associated with a 1-unit increase in the measure of epigenetic age acceleration.
We further stratified the sample by fasting status (Supplemental Table 4) and lipid-lowering medication use (Supplemental Table 5), respectively, in Model 2 to see whether the associations between epigenetic age acceleration and lipid levels vary across groups. Associations between DNAmAA and blood lipids were detected in participants who both fasted and did not fast, except for LDL-C, which was only detected in participants who fasted. We also detected associations mainly in participants who did not use lipid-lowering medication at the time of the blood draw. Overall, DunedinPACE had the most consistent associations among the DNAmAA measures, as it was associated with all lipids across all groups except for LDL-C in non-fasting participants.
Next, to understand whether associations between DNAmAA and blood lipid levels are modified by demographic factors, we introduced multiplicative terms into Model 2 (Table 3, Fig. 1, Supplemental Fig. 1). Six interactions between DunedinPACE and demographic factors on blood lipids were detected (Fig. 1). For TC, a one-unit increase in DunedinPACE was associated with a decrease of 29.99 mg/dL in younger participants (aged 62 years, 25th percentile) vs. a 49.76 mg/dL in older participants (aged 77 years, 75th percentile) (Fig. 1.A). For TG, a one-unit increase in DunedinPACE was associated with an increase of 42.29 vs. 97.10 mg/dL in younger vs. older participants (Fig. 1.B). In addition, the inverse associations between DunedinPACE and both TC and HDL-C were stronger in females compared with males, as a one-unit increase in DunedinPACE was associated with a decrease of 50.51 vs. 23.65 mg/dL in TC (Fig. 1.C) and a decrease of 23.32 vs. 16.67 mg/dL in HDL-C in females vs. males (Fig. 1.D). Finally, for those with a college degree vs. without, a one-unit increase in DunedinPACE was associated with a decrease of 19.29 vs. 25.05 mg/dL in HDL-C (Fig. 1.E) and an increase of 55.81 vs. 142.45 mg/dL in TG (Fig. 1.F). The significant interactions between GrimAA and PhenoAA and demographic factors on blood lipids showed similar trends (Supplemental Fig. 1.A-C) However, interactions between HannumAA and educational attainment on TG and HDL-C showed different patterns of association than DunedinPACE (Supplemental Fig. 1.D-E).
Table 3
Interactions between epigenetic age acceleration and demographic factors on blood lipid levels
Multiplicative term | Lipid | DNAmAA | βDNAmAA | PDNAmAA | βDemo | PDemo | βInteraction | PInteraction |
DNAmAA×Agea | TC | GrimAA | -0.755 | 2.33×10− 4 | -0.727 | 2.51×10− 13 | -0.051 | 0.004 |
| | DunedinPACE | -40.57 | 3.25×10− 12 | 0.760 | 0.190 | -1.318 | 0.018 |
| ln(TG) | GrimAA | 0.011 | 4.84×10− 5 | -0.004 | 0.002 | -4.94×10− 4 | 0.037 |
| | DunedinPACE | 0.392 | 2.05×10− 6 | 0.008 | 0.173 | -0.012 | 0.033 |
DNAmAA×Sexb | TC | DunedinPACE | -23.65 | 0.002 | 44.10 | 5.62×10− 5 | -26.86 | 0.007 |
| ln(HDL-C) | PhenoAA | -0.0005 | 0.737 | 0.221 | 6.10×10− 23 | -0.003 | 0.037 |
| | GrimAA | -0.005 | 0.033 | 0.197 | 1.14×10− 19 | -0.008 | 0.007 |
| | DunedinPACE | -0.346 | 8.76×10− 8 | 0.393 | 2.34×10− 7 | -0.180 | 0.005 |
DNAmAA×High school degreec | ln(HDL-C) | HannumAA | -0.009 | 0.003 | 0.072 | 1.14×10− 4 | 0.007 | 0.018 |
DNAmAA×College degreed | ln(HDL-C) | DunedinPACE | -0.413 | 7.81×10− 11 | 0.209 | 0.018 | -0.166 | 0.049 |
| ln(TG) | HannumAA | 0.005 | 0.057 | -0.083 | 0.002 | -0.008 | 0.036 |
| | DunedinPACE | 0.326 | 2.46×10− 4 | -0.422 | 0.004 | 0.359 | 0.011 |
TC, total cholesterol; HDL-C, high-density lipoprotein; TG, triglycerides; DNAmAA: epigenetic age acceleration; HannumAA, HannumAge acceleration; PhenoAA, LevineAge (PhenoAge) acceleration; GrimAA, GrimAge acceleration; Demo, demographic factors |
Model: blood lipid level ~ epigenetic age acceleration + age at methylation measurement + sex + race/ethnicity + fasting status + lipid-lowering medication use + body mass index + smoking status + high school degree or equivalent + college degree and above + epigenetic age acceleration × demographic factor
We only tested for an interaction when both the DNAmAA and the demographic factor were associated with blood lipids in Model 2.
Only significant interactions between epigenetic age acceleration and demographic factors (PInteraction<0.05) were included in this table.
DNAmAA effect sizes (βDNAmAA) correspond to the change in blood lipid level associated with a 1-unit increase in DNAmAA. Demographic factor effect sizes (βDemo) correspond to the change in blood lipid level associated with a 1-unit increase in age or with the non-reference level for sex or educational attainment. Interaction effect sizes (βInteraction) correspond to the change in effect of βDNAmAA on blood lipids for each 1-unit increase (or level) of the demographic factor.
aAge was centered in this analysis.
bReference group: male
cReference group: less than high school degree
dReference group: less than college degree
P-value < 0.05 in bold