Horvath · Genome biology 2013 · Pooled cross-sectional biomarker development and validation study · n=14,000 samples

DNA methylation age of human tissues and cell types.

Cited 7491 times in the scientific literature.

Level 3 - non-randomized controlled study

Level 3 by design analogy: large-scale computational development and validation study across pooled observational cohorts.

PubMed 24138928 · doi:10.1186/gb-2013-14-10-r115 · record verified 2026-08-27

What was done

Developed a multi-tissue predictor of age using 8,000 samples from 82 Illumina DNA methylation array datasets covering 51 healthy human tissues and cell types. Tested the predictor on embryonic stem cells, induced pluripotent stem cells, multiple cell passages, and chimpanzee tissues. Evaluated age acceleration using 6,000 cancer samples from 32 datasets across 20 cancer types, examining associations with somatic mutations, TP53 mutations, and steroid receptor status.

What was found

The 353-CpG epigenetic clock yielded an estimated age close to zero for embryonic and induced pluripotent stem cells, correlated with cell passage number, was heritable, and applied successfully to chimpanzee tissues. All 20 evaluated cancer types exhibited significant age acceleration, averaging 36 years. Low age acceleration in cancer was linked to high somatic mutation burdens and TP53 mutations, whereas steroid receptor mutations were associated with accelerated epigenetic age in breast cancer. No specific correlation coefficients or numerical error rates for age estimation in healthy tissues were reported in the abstract.

Why it matters

Provides a widely applicable multi-tissue biomarker of biological age across human tissues and cell types. It establishes a standardized framework for investigating epigenetic aging mechanisms, development, and oncogenesis.

Limits

The abstract omits numeric precision metrics (such as correlation coefficients or median absolute error) for healthy tissue age prediction. The study relies on retrospective, aggregated microarray datasets without prospective longitudinal follow-up reported in the abstract.

Cited by