Epigenetic clocks have been widely evaluated as cancer biomarkers, but findings have been inconsistent across tumor types and clinical endpoints. This inconsistency may reflect a fundamental misspecification: standard clock analyses treat clock CpGs as scalar aging readouts, assuming that tumors shift along the same methylation-aging axis as normal tissue. We tested this assumption across nine TCGA (The Cancer Genome Atlas) cancer types using Horvath clock CpGs. Tumors did not show a consistent mean shift in Horvath age acceleration relative to normal tissue. Instead, they showed an order-of-magnitude increase in age-acceleration variance. Scalar clock summaries also failed as survival biomarkers, producing zero nominal associations after adjustment for age, sex, and stage.
We therefore analyzed Horvath clock CpGs as a coordinated methylation system. PCA-derived tumor methylation programs captured 34.9%-50.8% of clock-CpG variance across cancers, and this structure persisted after adjustment for tumor purity, proliferative history, and global methylation instability. This indicates that tumor-associated clock-CpG variation is not random methylation disorder, but is organized along major axes of covariation. In fully adjusted Cox models, tumor methylation programs produced 15 nominal survival associations, four of which remained significant after FDR correction. Survival-associated programs were linked to transcriptional pathway activity, including UPR, EMT, IFN-{gamma} response, and stemness, in directions consistent with their survival effects. These programs also persisted after controlling for tumor-normal differentially methylated regions, indicating that they were not explained by average cancer-associated methylation shifts. External validation in GSE72308 showed that two of three TCGA-nominated BRCA survival associations reproduced in the same protective direction.
Gande, P. et al. · CC-BY 4.0