Metformin Slows Molecular Aging in Human Cells as Rapamycin Shows Limited Effects
An accelerated human-cell aging model found broader effects from metformin than rapamycin, but it cannot establish benefits in people.
Summary
Testing anti-aging drugs in people takes decades, and mice do not always predict human results. Researchers tried a different approach. They kept human skin cells from women aged 53 to 68 alive but not dividing for six months, then tracked their aging with DNA methylation and gene-activity clocks. The cells aged much faster than they would in the body, without becoming senescent, and they changed in ways that matched real aging in people. When the aged cells were converted into neurons, those neurons kept the older biological age. The team then used the system to test two famous longevity drugs. Metformin broadly slowed the aging signals, while rapamycin did little. The model could give researchers a faster, human-based way to screen interventions, though it is early and based on only four donors.
Detailed Summary
Studying how human cells age over time is hard. Mice are imperfect stand-ins for human aging, and most cell models either compare young and old donors at a single moment or force aging with stressors such as oxidants, progerin, or repeated division until senescence. Traxler and colleagues asked whether simply keeping non-dividing human skin fibroblasts alive at confluency for six months could create a controllable, time-based model of cell-intrinsic aging.
The team collected skin-biopsy fibroblasts from 33 donors aged 0 to 92 as an in vivo reference. Four female donors aged 53, 55, 67, and 68 were then held at contact-inhibited confluency without passaging for six months. Cells were sampled monthly (5 to 7 time points) for RNA sequencing and EPIC methylation arrays. The cells did not become senescent. They showed no p21 or p16 staining, no SASP upregulation, and small nuclei, and they could still divide when passaged after six months. Among DNA methylation clocks, Blood&Skin performed best on the in vivo samples (R2 = 0.98, velocity 1.01x), compared with Hannum (0.40x) and PhenoAge (0.48x).
Aging clocks ran much faster in the dish. The DNA methylation clock advanced at an average of 8.2 times real-time speed. The tAge transcriptomic clock advanced at 22.5x and RNAAgeCalc at 48.3x, with more variability between donors. Over five months, cells gained about 3.6 epigenetic years (95% CI -1.0 to 8.2) and 21.5 or 10.4 transcriptomic years depending on the clock. Transcriptional aging therefore ran roughly 3 to 6 times faster than methylation aging, although the two remained coupled.
Pathway analysis compared the in vitro data with the authors' in vivo cohort (n = 20) and the Fleischer cohort (n = 143). The team found 17 overlapping downregulated and 7 overlapping upregulated KEGG aging pathways shared by all three datasets. Of the in vitro pathways, 60 of 66 downregulated and 20 of 31 upregulated were also significant in vivo. All overlapping pathways were accelerated in vitro without reversing direction. Promoter methylation correlated inversely with expression (rho = -0.86 for downregulated and -0.75 for upregulated pathways). Sex adjustment had minimal effect. Single-cell data showed age heterogeneity growing over time, mirroring in vivo findings.
Pre-aged fibroblasts converted into induced neurons just as efficiently as baseline cells, and the neurons retained the added age. DNA methylation age rose 4.2 ± 1.7 years, tAge 6.6 ± 1.5, and RNAAgeCalc 10.7 ± 4.4, and the signal persisted after replating. In the drug tests, the abstract reports that metformin broadly attenuated transcriptomic and epigenetic aging, while rapamycin showed limited efficacy. This is consistent with some preclinical evidence.
The caveats are substantial. Only four female donors were aged, and they started at different baseline epigenetic ages, likely reflecting prior time in culture. The model captures a reproducible subset of aging programs, not full in vivo complexity, and it complements rather than replaces replicative senescence models. Acceleration rates are not real-time equivalents. It is also a single cell type, with no human outcome data. Treat the drug results as hypothesis-generating, not as evidence about clinical efficacy.
Key Findings
- Fibroblasts held non-dividing for 6 months stayed non-senescent: no p21/p16 staining or SASP upregulation, and they remained able to proliferate when passaged
- The Blood&Skin DNA methylation clock tracked in vivo age almost perfectly (R2 = 0.98, velocity 1.01x), versus Hannum (0.40x) and PhenoAge (0.48x)
- In vitro aging velocities were 8.2x for DNA methylation, 22.5x for tAge, and 48.3x for RNAAgeCalc, so transcriptomic aging ran about 3-6x faster than epigenetic aging
- 17 downregulated and 7 upregulated KEGG pathways overlapped across in vitro and two in vivo cohorts (n = 20 and n = 143); 60 of 66 downregulated and 20 of 31 upregulated in vitro pathways were also significant in vivo
- Promoter methylation correlated inversely with gene expression in aging pathways (rho = -0.86 downregulated, p = 0.01; rho = -0.75 upregulated, p = 0.03)
- Neurons converted from 6-month pre-aged fibroblasts retained the added age: +4.2 ± 1.7 DNAm years, +6.6 ± 1.5 tAge years, +10.7 ± 4.4 RNAAgeCalc years, with unchanged conversion efficiency
- Metformin broadly attenuated transcriptomic and epigenetic aging, while rapamycin showed limited efficacy in this platform
Methodology
Skin-biopsy fibroblasts from 33 donors aged 0 to 92 served as the in vivo reference. Four female donors (53, 55, 67, 68 years) were held at confluency without passaging for 6 months, with RNA-seq and EPIC methylation arrays at 5 to 7 monthly time points. Aging was quantified with multiple DNAm and transcriptomic clocks, and results were compared with an in-house in vivo cohort (n = 20) and the public Fleischer dataset (n = 143). Analyses used linear mixed-effects models with donor as a random effect, gene-set enrichment, ClueGO pathway overlap with Fisher's exact tests, and Spearman correlation of methylation with expression. Single-cell RNA-seq assessed heterogeneity, and direct neuronal conversion was tested in 3 donors.
Study Limitations
Only four female donors were aged in vitro, and they started at different baseline epigenetic ages, likely reflecting differences in prior culture time. The model recapitulates a subset of in vivo aging programs, with some pathways reflecting culture- or donor-specific responses, and it is limited to one cell type. Acceleration rates differ by clock and are not real-time equivalents. Drug-testing details and conflict-of-interest disclosures were not available in the supplied text.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.
Enter your email to subscribe: