Aging of hematopoietic stem cells (HSCs) underlies the age-related decline of immunity and the rising risk of leukemias. Earlier mouse studies uncovered key mechanisms such as inflammaging and disruptions in RNA splicing, but how well they hold true in humans remained unclear. Existing aging clocks, built on transcriptomes of peripheral blood cells, had never been tested on stem cells across the full span of life. This work fills that gap by creating a reference map of age-related changes in human HSCs.
The researchers reanalyzed single-cell RNA sequencing data from six published datasets. After strict quality control, 186,123 cells from 28 healthy donors remained. Using the Harmony algorithm, they integrated data from different experiments, then isolated the population of HSCs and multipotent progenitors (HSC/MPP) by marker genes and AUCell scoring. To uncover hidden transcriptional programs, they applied consensus non-negative matrix factorization (cNMF). To build the aging clock, they aggregated cells into metacells (20 random cells per donor) and trained six machine learning models, choosing the best one by validation error.
Six age-related molecular programs were identified (MP1-MP6). MP1 (inflammation via NF-κB) rises monotonically with age, while MP5 (RNA splicing and proteostasis) drops sharply after birth. MP4 (aerobic respiration) is characteristic of early development. The MLPRegressor model, trained on 55 genes from MP1 and MP5, predicts chronological age with an error of 3.13 years and a correlation of 0.99. Applied to 90 patients with acute myeloid leukemia, it revealed a "transcriptomic age deviation" (TAD): low TAD (cells look younger) is associated with unfavorable genetic risk and worse survival (median 8.6 versus 13.1 months in TCGA-LAML).
This work stands apart from neighboring studies in that it is the first to build an aging clock specifically for blood stem cells, covering the prenatal period and the whole of adulthood. Neighboring studies either described individual age-related changes or built clocks for other cell types. Here, a large-scale atlas, molecular programs, and a clinical application to leukemia are brought together. This provides a quantitative axis for assessing the biological age of stem cells and opens the way to finding vulnerabilities linked to aging.
The field could move toward integrating epigenomic, proteomic, and metabolomic data at the single-cell level, which would give a fuller picture of HSC aging. It would also be promising to check how TAD changes over time in the same leukemia patients during treatment. The aging clock might become a tool for assessing the effectiveness of anti-aging interventions aimed at stem cells.
The results could influence hematology, oncology, and gerontology by offering a biomarker for stratifying leukemia patients and a quantitative measure of cellular aging.
The next step will be functional validation of the key genes identified by the model and testing the clock on larger and more diverse cohorts, including longitudinal observations.
The work directly links stem cell aging to the pathogenesis of acute myeloid leukemia, answering the question of which age-related programs sustain leukemic cells. This helps explain why age is a risk factor and points to molecular targets for therapy.
The aging clock's error was just 3 years across a range from 4 weeks after conception to 77 years, which is comparable to the accuracy of predicting age from a photograph of a face.
The idea of an "aging clock" echoes the film "In Time" (2011), where people have a countdown of life on their arm. Here the clock is hidden inside cells, and it can be "tweaked" by influencing molecular programs.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.