Undulating changes in human plasma proteome across lifespan are linked to disease

Key results

1. ShownDE-SWAN, a 20-year sliding-window scan, finds three waves of plasma change cresting at ages 34, 60 and 78 with largely distinct pathways (extracellular matrix decline at 34; hormonal/blood at 60; blood and bone morphogenetic protein signaling at 78); most old-age changes escaped linear models.

   Numbers of differentially expressed proteins peaked at ages 34, 60 and 78 with the window slid in 1-year steps; peaks disappeared after jointly permuting age and sex but remained detectable across parcel widths of 5, 10, 15 and 20 years and after subcohort adjustment; overlap of the 3 crest protein sets was significant (p<0.05) but limited. · Quantification of proteomic changes across human…; Differential Expression - Sliding Window…; Figure legends

2. SuggestedProteins linked to cognitive/physical function, cardiovascular disease, Alzheimer's disease and Down syndrome preferentially overlap the age-60 and age-78 waves over the age-34 wave; a body mass index control showed no selectivity, and proteins with trans protein-QTLs ranked more randomly with age.

   Cumulative overlap counted down DE-SWAN p-value rankings at ages 34, 60 and 78 against Sun et al. pQTLs (1,927 associations, 1,104 proteins), LonGenity trait models, AddNeuroMed AD data (645 visit-1 samples) and published CVD, DS and BMI statistics; pQTL tests used 1e+5 permutations. Disease overlaps are shown graphically; significance is deferred to Supplementary Table 15. · Proteins linked to age-related diseases are…; Discussion; Relationships between the aging waves and the…; Figure legends

3. ShownLifespan trajectories are mostly non-linear: hierarchical clustering of locally smoothed (LOESS) z-scored profiles of 2,925 proteins gave 8 clusters (8 to 1,415 proteins), clusters 2, 3, 6, 7 and 8 being stepwise, logarithmic or exponential; shapes replicated in independent human cohorts.

   6 of 8 clusters were pathway-enriched (q<0.05): blood microparticle proteins declined steadily (cluster 5), while axon guidance and EPH-ephrin signaling proteins stayed flat until age 60 and then rose exponentially (cluster 6); undulation also appeared in the 4 independent cohorts (n=171) and mice (n=81). · Clustering individual protein trajectories…; Clustering of protein trajectories; Figure legends

4. ShownAge and sex effects are entangled, yet a single sex-independent 373-protein LASSO clock predicts chronological age, and individuals whose plasma looks younger than their age did better on cognitive and physical testing.

   895 of 1,379 age-associated proteins (q<0.05) also differed by sex (q<0.05). The clock, trained on 2,858 subjects and tested on 1,473, reached r=0.93-0.97 across discovery, validation and 4 independent cohorts (n=171); delta age was screened against 334 LonGenity clinical/functional variables; a 9-protein subset also predicted age well. · Linear modeling links the plasma proteome to…; Prediction of human chronological age using the…; Associations between Δage and clinical and…

5. SuggestedA 46-protein aging signature is shared by human and mouse plasma, and heterochronic parabiosis partly shifts it: young mice exposed to old blood show an older profile, old mice exposed to young blood a younger one.

   Mouse SomaScan (1,305 aptamers; 110 plasma samples spanning 1 month to 30 months, age/sex model on 81) found 172 age-changing proteins (q<0.05), 46 shifting in the same direction as in humans; parabiosis used 3- and 18-month-old mice and was summarized by principal component analysis only; mean mouse-human sequence identity was 0.85. · Linear modeling links the plasma proteome to…; Validation of the aging signature in mice; Figure legends