Integrative analysis of clinical and epigenetic biomarkers of mortality

Tianxiao Huan*, Steve Nguyen, Elena Colicino, Carolina Ochoa-Rosales, W. David Hill, Jennifer A. Brody, Mette Soerensen, Yan Zhang, Antoine Baldassari, Mohamed Ahmed Elhadad, Tanaka Toshiko, Yinan Zheng, Arce Domingo-Relloso, Dong Heon Lee, Jiantao Ma, Chen Yao, Chunyu Liu, Shih Jen Hwang, Roby Joehanes, Myriam FornageJan Bressler, Joyce B.J. van Meurs, Birgit Debrabant, Jonas Mengel-From, Jacob Hjelmborg, Kaare Christensen, Pantel Vokonas, Joel Schwartz, Sina A. Gahrib, Nona Sotoodehnia, Colleen M. Sitlani, Sonja Kunze, Christian Gieger, Annette Peters, Melanie Waldenberger, Ian J. Deary, Luigi Ferrucci, Yishu Qu, Philip Greenland, Donald M. Lloyd-Jones, Lifang Hou, Stefania Bandinelli, Trudy Voortman, Brenner Hermann, Andrea Baccarelli, Eric Whitsel, James S. Pankow, Daniel Levy

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

DNA methylation (DNAm) has been reported to be associated with many diseases and with mortality. We hypothesized that the integration of DNAm with clinical risk factors would improve mortality prediction. We performed an epigenome-wide association study of whole blood DNAm in relation to mortality in 15 cohorts (n = 15,013). During a mean follow-up of 10 years, there were 4314 deaths from all causes including 1235 cardiovascular disease (CVD) deaths and 868 cancer deaths. Ancestry-stratified meta-analysis of all-cause mortality identified 163 CpGs in European ancestry (EA) and 17 in African ancestry (AA) participants at p < 1 × 10 -7 , of which 41 (EA) and 16 (AA) were also associated with CVD death, and 15 (EA) and 9 (AA) with cancer death. We built DNAm-based prediction models for all-cause mortality that predicted mortality risk after adjusting for clinical risk factors. The mortality prediction model trained by integrating DNAm with clinical risk factors showed an improvement in prediction of cancer death with 5% increase in the C-index in a replication cohort, compared with the model including clinical risk factors alone. Mendelian randomization identified 15 putatively causal CpGs in relation to longevity, CVD, or cancer risk. For example, cg06885782 (in KCNQ4) was positively associated with risk for prostate cancer (Beta = 1.2, P MR = 4.1 × 10 -4 ) and negatively associated with longevity (Beta = -1.9, P MR = 0.02). Pathway analysis revealed that genes associated with mortality-related CpGs are enriched for immune- and cancer-related pathways. We identified replicable DNAm signatures of mortality and demonstrated the potential utility of CpGs as informative biomarkers for prediction of mortality risk.

Original languageEnglish
Article numbere13608
JournalAging Cell
Volume21
Issue number6
Early online date12 May 2022
DOIs
Publication statusPublished - Jun 2022

Keywords

  • cancer
  • cardiovascular disease
  • DNA methylation
  • machine learning
  • mortality

Fingerprint

Dive into the research topics of 'Integrative analysis of clinical and epigenetic biomarkers of mortality'. Together they form a unique fingerprint.

Cite this