Identifying blood-specific age-related DNA methylation markers on the Illumina MethylationEPIC® BeadChip

Research output: Contribution to journalArticle

Abstract

The past decade has seen rapid development in DNA methylation (DNAm) microarrays, including the Illumina HumanMethylation27 and HumanMethylation450 (450 K) chips, which have played an essential role in identifying and evaluating age-related (AR) DNAm markers in different tissues. Recently, a new array, the Illumina MethylationEPIC (EPIC) was introduced, with nearly double the number of probes as the 450 K (∼850,000 probes). In this study, we test these newly added probes for age association using a large cohort of 754 DNAm profiles from blood samples assayed on the EPIC BeadChip, for individuals aged 0–88 years old. 52 AR CpG sites (Spearman’s abs(rho) >0.6 and P-value <10−83) were identified, 21 of which were novel sites and mapped to 18 genes, nine of which (LHFPL4, SLC12A8, EGFEM1P, GPR158, TAL1, KIAA1755, LOC730668, DUSP16, and FAM65C) have never previously been reported to be associated with age. The data were subsequently split into a 527-sample training set and a 227-sample testing set to build and validate two age prediction models using elastic net regression and multivariate regression. Elastic net regression selected 425 CpG markers with a mean absolute deviation (MAD) of 2.6 years based on the testing set. To build a multivariate linear regression model, AR CpG sites with R2 > 0.5 at FDR < 0.05 were input into stepwise regression to select the best subset for age prediction. The resulting six CpG markers were linearly modelled with age and explained 81% of age-correlated variation in DNAm levels. Age estimation accuracy using bootstrap analysis was 4.5 years, with 95% confidence intervals of 4.56 to 4.57 years based on the testing set. These results suggest that EPIC BeadChip probes for age estimation fall within the range of probes found on the previous Illumina HumanMethylation platforms in terms of their age-prediction ability.
LanguageEnglish
Article number109944
Number of pages11
JournalForensic Science International
Volume303
Early online date12 Sep 2019
DOIs
Publication statusPublished - 31 Oct 2019

Fingerprint

DNA Methylation
Genetic Markers
Blood
Oligonucleotide Array Sequence Analysis
Microarrays
Set theory
Confidence Intervals
Association reactions
Tissue
Testing

Keywords

  • DNA methylation
  • forensic epigenetics
  • forensic age estimation
  • Illumina MethylationEPIC
  • age
  • CpG sites

Cite this

@article{1f27c6661ca24dc2baac091e935e4d97,
title = "Identifying blood-specific age-related DNA methylation markers on the Illumina MethylationEPIC{\circledR} BeadChip",
abstract = "The past decade has seen rapid development in DNA methylation (DNAm) microarrays, including the Illumina HumanMethylation27 and HumanMethylation450 (450 K) chips, which have played an essential role in identifying and evaluating age-related (AR) DNAm markers in different tissues. Recently, a new array, the Illumina MethylationEPIC (EPIC) was introduced, with nearly double the number of probes as the 450 K (∼850,000 probes). In this study, we test these newly added probes for age association using a large cohort of 754 DNAm profiles from blood samples assayed on the EPIC BeadChip, for individuals aged 0–88 years old. 52 AR CpG sites (Spearman’s abs(rho) >0.6 and P-value <10−83) were identified, 21 of which were novel sites and mapped to 18 genes, nine of which (LHFPL4, SLC12A8, EGFEM1P, GPR158, TAL1, KIAA1755, LOC730668, DUSP16, and FAM65C) have never previously been reported to be associated with age. The data were subsequently split into a 527-sample training set and a 227-sample testing set to build and validate two age prediction models using elastic net regression and multivariate regression. Elastic net regression selected 425 CpG markers with a mean absolute deviation (MAD) of 2.6 years based on the testing set. To build a multivariate linear regression model, AR CpG sites with R2 > 0.5 at FDR < 0.05 were input into stepwise regression to select the best subset for age prediction. The resulting six CpG markers were linearly modelled with age and explained 81{\%} of age-correlated variation in DNAm levels. Age estimation accuracy using bootstrap analysis was 4.5 years, with 95{\%} confidence intervals of 4.56 to 4.57 years based on the testing set. These results suggest that EPIC BeadChip probes for age estimation fall within the range of probes found on the previous Illumina HumanMethylation platforms in terms of their age-prediction ability.",
keywords = "DNA methylation, forensic epigenetics, forensic age estimation, Illumina MethylationEPIC, age, CpG sites",
author = "Hussain Alsaleh and Haddrill, {Penelope R.}",
year = "2019",
month = "10",
day = "31",
doi = "10.1016/j.forsciint.2019.109944",
language = "English",
volume = "303",
journal = "Forensic Science International",
issn = "0379-0738",

}

TY - JOUR

T1 - Identifying blood-specific age-related DNA methylation markers on the Illumina MethylationEPIC® BeadChip

AU - Alsaleh, Hussain

AU - Haddrill, Penelope R.

PY - 2019/10/31

Y1 - 2019/10/31

N2 - The past decade has seen rapid development in DNA methylation (DNAm) microarrays, including the Illumina HumanMethylation27 and HumanMethylation450 (450 K) chips, which have played an essential role in identifying and evaluating age-related (AR) DNAm markers in different tissues. Recently, a new array, the Illumina MethylationEPIC (EPIC) was introduced, with nearly double the number of probes as the 450 K (∼850,000 probes). In this study, we test these newly added probes for age association using a large cohort of 754 DNAm profiles from blood samples assayed on the EPIC BeadChip, for individuals aged 0–88 years old. 52 AR CpG sites (Spearman’s abs(rho) >0.6 and P-value <10−83) were identified, 21 of which were novel sites and mapped to 18 genes, nine of which (LHFPL4, SLC12A8, EGFEM1P, GPR158, TAL1, KIAA1755, LOC730668, DUSP16, and FAM65C) have never previously been reported to be associated with age. The data were subsequently split into a 527-sample training set and a 227-sample testing set to build and validate two age prediction models using elastic net regression and multivariate regression. Elastic net regression selected 425 CpG markers with a mean absolute deviation (MAD) of 2.6 years based on the testing set. To build a multivariate linear regression model, AR CpG sites with R2 > 0.5 at FDR < 0.05 were input into stepwise regression to select the best subset for age prediction. The resulting six CpG markers were linearly modelled with age and explained 81% of age-correlated variation in DNAm levels. Age estimation accuracy using bootstrap analysis was 4.5 years, with 95% confidence intervals of 4.56 to 4.57 years based on the testing set. These results suggest that EPIC BeadChip probes for age estimation fall within the range of probes found on the previous Illumina HumanMethylation platforms in terms of their age-prediction ability.

AB - The past decade has seen rapid development in DNA methylation (DNAm) microarrays, including the Illumina HumanMethylation27 and HumanMethylation450 (450 K) chips, which have played an essential role in identifying and evaluating age-related (AR) DNAm markers in different tissues. Recently, a new array, the Illumina MethylationEPIC (EPIC) was introduced, with nearly double the number of probes as the 450 K (∼850,000 probes). In this study, we test these newly added probes for age association using a large cohort of 754 DNAm profiles from blood samples assayed on the EPIC BeadChip, for individuals aged 0–88 years old. 52 AR CpG sites (Spearman’s abs(rho) >0.6 and P-value <10−83) were identified, 21 of which were novel sites and mapped to 18 genes, nine of which (LHFPL4, SLC12A8, EGFEM1P, GPR158, TAL1, KIAA1755, LOC730668, DUSP16, and FAM65C) have never previously been reported to be associated with age. The data were subsequently split into a 527-sample training set and a 227-sample testing set to build and validate two age prediction models using elastic net regression and multivariate regression. Elastic net regression selected 425 CpG markers with a mean absolute deviation (MAD) of 2.6 years based on the testing set. To build a multivariate linear regression model, AR CpG sites with R2 > 0.5 at FDR < 0.05 were input into stepwise regression to select the best subset for age prediction. The resulting six CpG markers were linearly modelled with age and explained 81% of age-correlated variation in DNAm levels. Age estimation accuracy using bootstrap analysis was 4.5 years, with 95% confidence intervals of 4.56 to 4.57 years based on the testing set. These results suggest that EPIC BeadChip probes for age estimation fall within the range of probes found on the previous Illumina HumanMethylation platforms in terms of their age-prediction ability.

KW - DNA methylation

KW - forensic epigenetics

KW - forensic age estimation

KW - Illumina MethylationEPIC

KW - age

KW - CpG sites

U2 - 10.1016/j.forsciint.2019.109944

DO - 10.1016/j.forsciint.2019.109944

M3 - Article

VL - 303

JO - Forensic Science International

T2 - Forensic Science International

JF - Forensic Science International

SN - 0379-0738

M1 - 109944

ER -