https://onlinelibrary.wiley.com/doi/full/10.1111/acel.70437
chatGPT:
Here’s a focused review of the uploaded paper,
Summary
This paper uses the European MARK-AGE cohort to build a blood-based biological age model from a very broad biomarker panel. The study included about 3300 adults aged 35–74 and initially measured 362 molecular, cellular, genetic, and clinical chemistry biomarkers. The authors then derived sex-specific formulas based on 10 biomarkers in women and 10 in men to estimate a “bioage score,” and defined age difference as biological age minus chronological age.
The selected markers are a mixed panel. In women, the final model included several DNA methylation markers at ELOVL2 and FHL2, plus DHEAS, a glycan marker, ferritin, a glycan ratio, and α-tocopherol. In men, the panel again included ELOVL2 and FHL2 methylation markers, DHEAS, a glycan marker, plus α2-macroglobulin, lycopene, and PSA. The resulting bioage scores correlated strongly with chronological age, with Pearson r ≈ 0.90 in females and ≈ 0.87 in males.
The authors then ask a more interesting question: among people of the same chronological age, what traits track age advancement or age delay? They report that people who are biologically younger tend to have a more favorable profile for HDL, 25-hydroxy-vitamin D, and a CD3+CD4+/CD45+ T-cell ratio. By contrast, glucose and HbA1c were associated with chronological age but not with age difference. In males, insulin and HOMA also tracked age difference; in females, insulin tracked both chronological age and age difference, while HOMA was less clear.
They also perform a few “known-groups” checks. Down syndrome subjects appeared biologically older by about 5.1 years in females and 3.9 years in males. Female smokers showed higher biological age with greater cumulative smoking exposure, while this was not significant in males. Postmenopausal women taking hormones were estimated to be biologically younger by about 2.1 years than non-users over age 50. They did not find a significant difference between offspring of long-lived families and their spouses.
What seems novel
The main novelty is not just that the paper builds another biological-age clock. The more original part is the attempt to separate markers that track chronological age itself from markers that track the biological-age residual. That is, it asks: what distinguishes two people who are the same age in years, but different in apparent biological aging? That “age difference versus chronological age” framing is the paper’s strongest idea.
A second novelty is the use of a mixed biomarker panel rather than a pure epigenetic clock. The model combines methylation loci with endocrine, lipid, glycomic, and other blood-based measures, separately for women and men. That gives it a broader physiological flavor than a methylation-only clock.
A third novelty is the paper’s conceptual suggestion that some correlates may be drivers or determinants of pace of aging, while others may be bystanders. The paper is careful not to overclaim, but it explicitly proposes this interpretation for the contrast between HDL/Vitamin D/CD4-related measures and glucose/HbA1c.
Critique
The biggest limitation is that this is a cross-sectional study. The authors themselves acknowledge that MARK-AGE is not longitudinal and say that proper validation would require later follow-up on physical, cognitive, clinical, morbidity, and mortality outcomes. So the paper can show association, but it cannot establish whether HDL, vitamin D, or immune ratios are actually causing slower aging rather than merely co-varying with healthier people.
A second issue is that the model is still fundamentally trained to reproduce chronological age. That is useful, but it is not the same as training directly on hard outcomes such as mortality, frailty, disability, or disease incidence. A clock that predicts age well is not automatically a clock that best captures biologically meaningful aging. This paper tries to bridge that gap using residual associations, but it does not fully close it.
Third, some of the highlighted associations have small effect sizes. The authors explicitly say that the correlations with HDL, vitamin D, and CD3+CD4+/CD45+ are statistically significant but that the slopes and R² values are rather small, and they do not claim these are major standalone determinants. That is an important restraint, and it also means the findings are more hypothesis-generating than decisive.
Fourth, several findings are vulnerable to confounding. Vitamin D is influenced by supplementation, sunlight exposure, adiposity, illness, and season. HDL is a marker with complex biology and is not a simple causal “good aging” factor. The HRT result is especially hard to interpret because the study could not distinguish clearly between contraceptive use and menopausal HRT based on the questionnaire alone, and it lacked duration data. The authors explicitly note this limitation.
Fifth, there is some potential for selection bias because subjects with incomplete data or outlier biomarkers were excluded. The paper argues that this likely did not materially distort age distribution, but exclusion-based modeling always raises generalizability questions.
Sixth, some of the male-specific markers in the clock, especially PSA, may raise questions about biological interpretation. PSA can track age statistically, but that does not necessarily make it a clean mechanistic aging marker. More broadly, several selected biomarkers may be very good predictors without necessarily being central biological drivers. That is a common tension in biomarker clocks, and this paper does not fully resolve it.
Bottom line
My view is that this is a good, thoughtful, and fairly careful paper, but more as a hypothesis-generating biomarker study than as a definitive biological-age clock. Its strongest contribution is the distinction between markers associated with chronological aging and markers associated with age acceleration/deceleration within age-matched individuals. Its weakest point is that, because the study is cross-sectional and effect sizes are modest, it cannot tell you whether the highlighted correlates are genuine drivers of aging or just correlated features of healthier people.
Overall verdict: interesting and worthwhile, with a genuinely useful conceptual angle, but not yet a decisive advance in proving what drives human biological aging.
If you would like, I can also turn this into a table with columns for summary / novelty / strengths / weaknesses / verdict.