henhaobio
Sign in

bioRxivMolecular BiologyPreprint

Circulating extracellular vesicles in plasma carry accessible molecular signatures of aging in mice

Kristine A. TsantilasiD, Michael E. RiffleiD, Gennifer E. MerrihewiD, Christine C WuiD, Gregory R. KeeleiD, Aaron J. MauraisiD, Richard S. JohnsoniD, Alison Luciano, Laura L. Robinson, Gary A. ChurchilliD, Michael J. MacCossiD

PDFDOICodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Cells release membrane-bound extracellular vesicles into the bloodstream laden with proteins that may reflect their physiological state. How this circulating EV proteome changes across life remains poorly understood. Identifying molecular signatures of aging in accessible biofluids could facilitate earlier intervention and monitoring of age-related disease. Many circulating aging proteome studies rely on affinity-based platforms which suffer from poor cross-species translation, ambiguous signal attribution, and inconsistent agreement between platforms. Here, we present a characterization of the aging plasma EV proteome from a cross-sectional cohort of 86 male and female C57BL/6J mice (5-31 months). We leveraged a species-agnostic EV enrichment (Mag-Net) and mass spectrometry to detect 2,575 protein groups from 15,969 peptides. Protein abundance heterogeneity increased with age and the abundance of 272 proteins were significantly correlated with chronological age including established senescence and frailty markers. Proteins increasing with age were enriched in genome maintenance pathways, while those decreasing were associated with the extracellular matrix organization and lipid metabolism. Notably, several of the strongest age-increased proteins converged on Alzheimers and Parkinsons disease pathology. We observed sexual divergence in the aging EV proteome not previously characterized at this resolution. A proteomic clock built from this data accurately predicts chronological age, and peptide-level analysis reveals aging signals invisible at protein-level. These findings demonstrate that EV-enriched plasma proteomics can identify known aging markers, reveal novel sex-specific age-related changes, and generate predictive models of chronological age. This study provides a species-agnostic foundation for proteomic clocks that complement epigenetic approaches to monitor aging and evaluate healthspan.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (28 Jul 2026) · CC BY 4.0

Mass spectrometry of Mag-Net-enriched plasma extracellular vesicles from 86 C57BL/6J mice reveals 272 age-correlated proteins, sex-divergent trajectories, and an age clock with 2.75-month mean absolute error.

Plasma aging proteome studies mostly use affinity reagents that lose sensitivity across species, attribute signal ambiguously and disagree between platforms. Plasma's dynamic range also usually needs depletion tools that model organisms lack, so this work tests size- and charge-based extracellular vesicle (EV) capture plus mass spectrometry as a species-agnostic route to aging markers and clocks in mice.

Key results

  1. ShownAcross both sexes, 272 plasma EV proteins track chronological age, including known senescence and frailty markers. The proteins rising most with age (CATB, CSF1R, ITM2B, CATD) have prior links to Alzheimer's or Parkinson's pathology.

    Spearman ρ > 0.3 or ρ < -0.3 with padjusted < 0.05: 126 up and 146 down (from 1,536 peptides). PAI1, OPN and GRN rose in both sexes. Top increases: CATB ρ = 0.64, CSF1R ρ = 0.59, ITM2B ρ = 0.55, CATD ρ = 0.54. Top decreases: LUM ρ = -0.69, OBSL1 ρ = -0.67, FBLN1 ρ = -0.66. · Spearman correlation identifies proteins…; Spearman correlation and Reactome pathway…

  2. ShownAn Elastic Net regression clock built on EV protein abundances predicts chronological age within this cohort. A peptide-level clock performs slightly worse, while a separate peptide-level OLS analysis flags age-associated peptides from proteins that were not significant at the protein level.

    Repeated 10-fold cross-validation (5 repeats): MAE 2.75 months, r² 0.79; accuracy fell at the youngest and oldest ages. The final model kept non-zero coefficients for 277 of 2,575 proteins. Peptide model (15,969 peptides): MAE 3.03, r² 0.75. OLS found 96 age-associated peptides whose parent proteins were not significant. · Circulating plasma aging biomarkers predict…; Elastic Net Regularized Linear Regression

  3. SuggestedAging trajectories differ by sex. Females gain coagulation and proteolysis proteins (notably TFPI2), while males lose major urinary proteins (MUPs) and gain immune and ECM proteins. Known markers such as CALR and ICAM1 change with age in only one sex.

    Sex-stratified Spearman (42 females, 44 males) flagged 21 proteins; the OLS sex-by-age interaction flagged 16 (q < 0.01), and 10 were shared. TFPI2 ρ♀ = 0.840; MUP1 ρ♂ = -0.635; AMBN ρ♂ = 0.689; CALR ρ = 0.55 in females vs -0.1 in males; ICAM1 ρ = 0.42 in males vs 0.001 in females. · Sexual dimorphism in aging plasma EV proteome…; Sex-divergent biological themes in aging; Figure legends

  4. ShownVariability in protein abundance between animals rises specifically in old age: older mice have higher coefficients of variation (CV) and spread more widely in PCA, and males have lower CVs than females.

    Ordinary least squares (OLS) model of CV comparing mice aged 21 months and older with those under 21 months: significant positive β₂ (about 1.1% higher CV per log mean-abundance unit). Males: β₃ = -0.0072, p < 0.001. The age effect was not significant when comparing 5–15 vs 17–21 months. · Protein abundance is more variable in aged mice; Tables

  5. SuggestedProteins that rise with age are enriched for genome maintenance pathways; proteins that fall with age are enriched for extracellular matrix (ECM) and lipoprotein metabolism pathways.

    Reactome enrichment on the 272 proteins (p < 0.05): most of the top 10 increasing pathways involved genome maintenance. Among decreasing pathways, dermatan and chondroitin sulfate biosynthesis and degradation had the highest odds ratios, followed by lipoprotein pathways (plasma lipoprotein assembly, HDL remodeling, plus assembly and remodeling of chylomicrons). · Reactome pathway analysis reveals coherent…; Spearman correlation and Reactome pathway…

Methods, data and reproducibility

Plasma EVs were captured from 20 μL per mouse with Mag-Net (SAX beads binding by vesicle size and negative membrane charge) and profiled by data-independent acquisition mass spectrometry on an Orbitrap Astral. Age effects were tested by Spearman correlation, OLS models including sex, Reactome enrichment and Elastic Net regression.

  • Cross-sectional aging cohort of male and female C57BL/6J mice (Jackson Laboratory stock 000664), 9 age groups spanning 25–131 weeks; plasma EV proteomes · mouse (Mus musculus, C57BL/6J) · n = 86 mice (42 female, 44 male; 3–6 per sex per group)
  • Commercial pooled mouse plasma processed neat and with Mag-Net in parallel to test EV protein enrichment and depletion of high-abundance proteins · mouse (C57BL/6N)
  • External QC samples: inter-experiment QC from a commercial plasma pool and inter-batch QC pooled from mice housed with the cohort · mouse
  • Mus musculus reference proteome FASTA with yeast enolase 1 appended as internal QC · mouse · UniProt UP000000589
Methods & tools
Mag-Net EV enrichment (SAX magnetic beads), Thermo KingFisher Apex, DIA-MS on Thermo Orbitrap Astral with Vanquish Neo LC, Thermo Orbitrap Exploris 480 with EASY-nLC 1200, nf-skyline-dia-ms Nextflow pipeline, Skyline 24.04.4, EncyclopeDIA, Nanosight NS300 nanoparticle tracking analysis (NTA 3.4), ReactomePA (Bioconductor), statsmodels OLS, scikit-learn Elastic Net with RepeatedKFold cross-validation, Targeted PRM system suitability runs
Reagents & environment
R 4.4.1 and Python (statsmodels, scikit-learn, NumPy, SciPy); Skyline 24.04.4; NTA 3.4 (Build 3.4.4); UniProt mouse proteome UP000000589 downloaded February 2, 2023; Nextflow configuration files deposited on PanoramaWeb.
Reproducibility gaps
Detailed methods appear only in the Supplementary Materials. GitHub holds only inputs under 25 MB; larger files are on PanoramaWeb.
Limitations (6)
  • The cross-sectional design cannot separate biological from chronological aging in individual animals; longitudinal data are needed to test whether the clock follows interventions or health outcomes. · stated by the authors
  • Only a single inbred strain (C57BL/6J) was profiled, so the age signatures may not hold in other genetic backgrounds. · stated by the authors
  • With 3–6 animals per age and sex group, statistical power is limited, particularly for the sex-by-age interaction analyses. · stated by the authors
  • No candidate protein was checked with an orthogonal assay, and the authors say an independent replication cohort is required before any marker advances as a biomarker candidate. · stated by the authors
  • Hemolysis was judged only qualitatively and produced a visible PCA signal, but the main text does not say it was modeled or excluded, so hemolysis-related proteins could confound the age associations. · noted in this summary
  • Elastic Net hyperparameters were chosen by Bayesian optimization, but the main text describes no nested tuning or held-out test set, so the cross-validated MAE and r² may be optimistic. · noted in this summary

Read this if you study aging biomarkers or plasma proteomics in model organisms and want a species-agnostic EV-enrichment mass spectrometry workflow, sex-resolved age trajectories and a mouse proteomic age clock.

Citations
0

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISHuang '26Bhunia '25Heimler '26Lehallier '19Lim '26Liao '26
73% semantic matchSystems Biology
Qiongrong Huang, Mark A. Watson, John C. Earls+12 more·bioRxiv·Oct 24, 2026