Article activity feed

  1. Evaluation Summary:

    This paper describes a two-sample randomization study of the impact of several measures of epigenetic aging acceleration (four different DNA methylation clocks) on risk for various common cancers (prostate, colon, lung, ovary, and breast). Data from large case-control cancer GWAS results are leveraged, as well as large cohort GWAS (UK Biobank and FinnGen) and GWAS of epigenetic aging. The most convincing finding is an an estimated effect of GrimAge on colon cancer risk (while results for other cancers are null or suggestive). This analysis is an important contribution as it addresses a question of substantial interest in cancer epidemiology.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

  2. Reviewer #1 (Public Review):

    In this manuscript, the authors performed a two-sample Mendelian randomization (MR) study to investigate the effects of epigenetic age acceleration (i.e., HannumAge, Horvath Intrinsic Age, PhenoAge, and GrimAge) on the risk of multiple cancers (i.e., breast, prostate, colorectal, ovarian, and lung cancer). Genetic instruments were selected based on a genome-wide association study (GWAS) of epigenetic age acceleration including 34,710 participants of European ancestry. Genetic association data for cancer outcomes were obtained from the UK Biobank, FinnGen, and several international cancer genetic consortia. The analyses yielded several interesting findings, including the associations of genetically-determined GrimAge acceleration with increased risk of colorectal cancer and decreased risk of prostate cancer. The authors presented the three core assumptions required for MR. The F statistic for each genetic instrument was reported to assess the validity of the relevance assumption. Other sensitivity analyses, including MR-Egger, weighted median, weighted mode, and CAUSE methods, were used to detect and correct for potential horizontal pleiotropy.


    1. The large-scale GWAS datasets used in this study had improved the power to detect the associations between epigenetic age acceleration and cancer risks.
    2. This study represents a comprehensive effort to estimate the effect of genetic loci associated with epigenetic age acceleration on the risk of common cancers.


    1. GrimAge includes data from 1,030 age-related CpGs associated with smoking pack-years and seven plasma proteins. However, only 4 SNPs were identified as genetic instruments of GrimAge acceleration, which explained 0.47% of the trait variance and thus limited the statistical power to detect the exposure-outcome associations.
    2. Although the authors had stated the independence assumption, they did not evaluate whether this assumption hold. If the genetic instruments of GrimAge acceleration are associated with confounders, then the validity of the results is questionable.
    3. Another concern is that the MR-Egger intercept test had low power to detect uncorrelated horizontal pleiotropy in the context of very few SNPs. The weighted mode method may also be misleading in this context. The evidence is not enough for making a claim that there is no horizontal pleiotropy.

    Taken together, potential violations of the independence and exclusion restriction assumptions cannot be entirely ruled out, which may mislead the causal inference. Results from this study should be interpreted with caution.

  3. Reviewer #2 (Public Review):

    This article reports Mendelian Randomization (MR) analysis of associations between several DNA methylation (DNAm) clock biomarkers of aging and several cancers. The main positive finding is that GrimAge acceleration shows evidence of causal effect on development of colon cancer. Other associations were either null or proved inconsistent in sensitivity analysis. This result contrasts with past observational studies reporting correlations between multiple DNAm clocks and multiple types of cancer. It therefore has potential to change the way the field thinks about the integration of DNAm clocks into cancer epidemiology.

    Connections between the biology of aging and cancer are many, motivating interest in whether biomarkers of aging can also inform assessments of cancer risk. However, the black-box nature of the DNAm clocks (they are developed from machine learning and the biology of their included CpG sites is mostly unknown) complicates their interpretation in clinical settings. Positive results in MR analysis would bolster confidence that clock associations with cancer are causal and motivate further inquiry to understand mechanism. This study therefore has the potential to contribute to interpretation of DNAm clocks in cancer risk assessment by informing causal interpretations of associations reported between older clock ages and risk of developing several cancers.

    The main limitation of the study is that the genetics of DNAm clocks are not well established. GWAS sample sizes for clock analyses have been relatively small. In this study, the GWAS was based on just 30,000 participants and the SNPs identified explain only small fractions of the heritability of the clocks. The authors are transparent about statistical power for their analysis and report several sensitivity checks for their results. Nevertheless, a consequence of this limitation is that null results do not rule out causal effects of the clocks on cancer risk. Moreover, positive findings are observed only for the DNAm clock with the fewest identified SNPs (just 4 SNPs explaining <1% of variation in GrimAge).

    Despite these limitations, it is my view that this study makes a substantive contribution to the emerging literature linking DNAm clocks and cancer.

  4. Reviewer #3 (Public Review):

    The authors' goal is to estimate the impact of DNA methylation aging (acceleration) on risk of various types of cancer using Mendelian randomization methods. This work reflects substantial efforts for data acquisition, formatting, and analysis. The authors use appropriate large scale data resources and a set of sophisticated analyses tools to address their hypotheses of interest. While the results consist of largely null and/or suggestive estimates of associations/effects, the hypothesis addressed is of substantial interest in the field, so the results presented are of significant interest to the community. Replication in future studies will be needed. The paper is well-written and easily understandable. I have no major concerns regarding the analysis approach, but additional details and data visualizations are needed for readers to have a complete picture of the results. There are limitations and biases associated with the MR method, but these are described and accounted for (to the extent possible) through sensitivity analyses.