Intraindividual variation and short-term temporal trend in DNA methylation of human blood

Abstract

Background: Between- and within-person variation in DNA methylation levels are important parameters to be considered in epigenome-wide association studies. Temporal change is one source of within-person variation in DNA methylation that has been linked to aging and disease.

Methods: We analyzed CpG-site-specific intraindividual variation and short-term temporal trend in leukocyte DNA methylation among 24 healthy Chinese women, with blood samples drawn at study entry and after 9 months. Illumina HumanMethylation450 BeadChip was used to measure methylation. Intraclass correlation coefficients (ICC) and trend estimates were summarized by genomic location and probe type.

Results: The median ICC was 0.36 across nonsex chromosomes and 0.80 on the X chromosome. There was little difference in ICC profiles by genomic region and probe type. Among CpG loci with high variability between participants, more than 99% had ICC > 0.8. Statistically significant trend was observed in 10.9% CpG loci before adjustment for cell-type composition and in 3.4% loci after adjustment.

Conclusions: For CpG loci differentially methylated across subjects, methylation levels can be reliably assessed with one blood sample. More samples per subject are needed for low-variability and unmethylated loci. Temporal changes are largely driven by changes in cell-type composition of blood samples, but temporal trend unrelated to cell types is detected in a small percentage of CpG sites.

Impact: This study shows that one measurement can reliably assess methylation of differentially methylated CpG loci. Cancer Epidemiol Biomarkers Prev; 24(3); 490-7. ©2014 AACR.

Figure 1

Scatter plot of within- vs. between-person standard error (β-scale), all CpG sites.

Figure 2

The effect of quantile normalization on ICC distribution and trend estimates, all CpG sites. (A) High, mid and low ICC distribution for untransformed and quantile-normalized methylation levels, by genomic location and probe type. (B) Proportion of CpG loci with significant temporal trend for untransformed and quantile-normalized methylation levels, by genomic location and probe type. All estimates are adjusted for cell type composition. DMR: differentially methylated region; TSS: transcription start site; UTR: untranslated region.

Figure 3

The effect of cell-type composition adjustment and quantile normalization on temporal trend in DNA methylation: scatter plots of trend estimates. (A) Untransformed vs. quantile-normalized data, no cell-type adjustment. (B) Untransformed data: without vs. with cell-type adjustment. (C) Untransformed vs. quantile-normalized data, with cell-type adjustment., (D) Untransformed data with cell-type adjustment vs. quantile-normalized data without cell-type adjustment. All plots show only CpG loci with significant trend in both models. CTA: cell type adjustment; QN: quantile normalization.