Plasma proteomic signatures of cellular aging predict human disease

Daisy Yi Ding, Veronica Augustina Bot, Robert Palovics, Hamilton Se-Hwee Oh, Carlos Cruchaga, Jonathan M. Schott, Tony Wyss-Coray

1. 01

   Machine-learning models estimated biological age of >40 cell types from >7,000 plasma proteins in 60,542 individuals

2. 02

   20-25% of individuals showed accelerated aging confined to a single cell type; 1-3% across ten or more cell types

3. 03

   Extreme astrocyte aging tripled incident Alzheimer's risk in APOE4 homozygotes; youthful astrocytes reduced risk

4. 04

   Extremely aged vs youthful skeletal myocytes carried a 12.7-fold higher risk of incident ALS

5. 05

   In smokers, extreme respiratory epithelial aging added 58% to lung-cancer risk beyond smoking alone

6. 06

   A polycellular aging risk score stratified mortality across cohorts and proteomics platforms

Study Design

Cohort Study

Sample Size

60,542 subjects

Study Duration

15 years of follow-up

Methodology

Large-scale plasma proteomics (>7,000 proteins) with cell-type-resolved machine-learning age models; associations tested against prevalent disease and 15-year incident disease and mortality; replication across cohorts and proteomic platforms.

Limitations

Observational cohort design — associations, not causal effects, and no intervention has been shown to move these signatures with clinical benefit. The headline hazard ratios come from extreme deciles and will attenuate substantially in unselected clinical populations. Cohorts are predominantly European-ancestry; performance in other populations is unverified. Not a validated surrogate endpoint for geroprotective trials.