Conserved DNA methylation patterns in healthy blood cells and extensive changes in leukemia measured by a new quantitative technique

Abstract

Genome wide analysis of DNA methylation provides important information in a variety of diseases, including cancer. Here, we describe a simple method, Digital Restriction Enzyme Analysis of Methylation (DREAM), based on next generation sequencing analysis of methylation-specific signatures created by sequential digestion of genomic DNA with SmaI and XmaI enzymes. DREAM provides information on 150,000 unique CpG sites, of which 39,000 are in CpG islands and 30,000 are at transcription start sites of 13,000 RefSeq genes. We analyzed DNA methylation in healthy white blood cells and found methylation patterns to be remarkably uniform. Inter individual differences > 30% were observed only at 227 of 28,331 (0.8%) of autosomal CpG sites. Similarly, > 30% differences were observed at only 59 sites when we comparing the cord and adult blood. These conserved methylation patterns contrasted with extensive changes affecting 18-40% of CpG sites in a patient with acute myeloid leukemia and in two leukemia cell lines. The method is cost effective, quantitative (r ( 2) = 0.93 when compared with bisulfite pyrosequencing) and reproducible (r ( 2) = 0.997). Using 100-fold coverage, DREAM can detect differences in methylation greater than 10% or 30% with a false positive rate below 0.05 or 0.001, respectively. DREAM can be useful in quantifying epigenetic effects of environment and nutrition, correlating developmental epigenetic variation with phenotypes, understanding epigenetics of cancer and chronic diseases, measuring the effects of drugs on DNA methylation or deriving new biological insights into mammalian genomes.

Figure 1. Digital Restriction Enzyme Analysis of Methylation (DREAM). (A) Schematic outline of the principle. SmaI restriction endonuclease can cut only unmethylated CCCGGG sites creating 5′-GGG signatures (u). Remaining methylated CC^meCGGG sites are then cut by XmaI restriction endonuclease creating 5′-CCGGG signatures (m). Sequencing adapters are ligated to SmaI/XmaI fragments and the libraries are subjected to next generation sequencing. Methylation levels at unique SmaI/XmaI sites are calculated based on the numbers of methylated and total signatures. (B) Numbers of unique CpG sites captured by DREAM in healthy human white blood cells based on the minimum sequencing depth. Black, 4 HiSeq lanes; red, single HiSeq lane, 4 samples; blue, ¼ of HiSeq lane, 4 samples. Solid lines show means; colored areas between broken lines show mean ± SEM (C) Good correlation between methylation levels at unique CpG sites (n = 29,574) covered by 100+ reads in DREAM libraries generated from 5 µg or 500 ng of gDNA from the same sample of healthy leukocytes. Pearson r² = 0.9971. (D) False positive rate (FPR) calculation based on methylation differences between 93,420 replicate measurements of methylation at CCCGGG sites covered by 100+ reads. Horizontal broken lines show FPR of 1% and 5%. Sites with low methylation (0–20%) or high methylation (80–100%) showed the lowest FPR.

Figure 2. Validation of DREAM results. (A) Bisulfite sequencing vs. DREAM. Methylation at CCCGGG sites. Linear regression r² = 0.847, p < 0.0001. (B) Bisulfite pyrosequencing vs. DREAM. Linear regression r² = 0.930, p < 0.0001. (C) Reduced representation bisulfite sequencing, K562 cell line vs. DREAM. Minimum coverage 50+ reads. Linear regression r² = 0.953, p < 0.0001. Broken lines show linear regression. (D) Methyl 450K Bead Array, K562 cell line vs. DREAM. Minimum coverage 50+ reads. Linear regression r² = 0.854, p < 0.0001.

Figure 3. DNA methylation in normal leukocytes. (A) CpG islands (CGI, green) are largely unmethylated. The majority of CpG sites outside of CpG islands (NCGI, orange) is methylated. (B) Methylation of CGI and NCGI sites drops at the vicinity of transcription start sites (TSS). (C) Small inter individual differences in DNA methylation in healthy leukocytes. (D) Small methylation differences between cord blood and adult blood. Grey, differences < 30%; green, CGI sites; orange NCGI sites.

Figure 4. Methylation changes in leukemia. (A) Acute myeloid leukemia bone marrow. (B) HEL erythroleukemia cell line. (C) K562 leukemia cell line. Hypermethylation of CpG islands (CGI, green), hypomethylation outside CpG islands (NCGI, orange). Grey, CpG sites with methylation changes vs healthy leukocytes < 30%. (D) Numbers of genes with CpG sites affected by methylation changes over 30% compared with normal leukocytes.

Figure 5. Methylation of repetitive elements. (A) Proportions of uniquely mapped reads and reads mapped to repetitive elements. (B) Repetitive elements showing high levels of methylation in normal blood were hypomethylated in leukemia. (C) Repeats with low methylation levels in normal blood became hypermethylated in leukemia. CB, cord blood; WBC, adult blood white blood cells; AML, bone marrow from an acute myeloid leukemia patient; HEL, human erythroleukemia cell line; K562, human myeloid leukemia cell line.